sys/net/if.c

   1 /*
   2  * Copyright (c) 1980, 1986, 1993
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer.
  10  * 2. Redistributions in binary form must reproduce the above copyright
  11  *    notice, this list of conditions and the following disclaimer in the
  12  *    documentation and/or other materials provided with the distribution.
  13  * 3. Neither the name of the University nor the names of its contributors
  14  *    may be used to endorse or promote products derived from this software
  15  *    without specific prior written permission.
  16  *
  17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  27  * SUCH DAMAGE.
  28  *
  29  *      @(#)if.c        8.3 (Berkeley) 1/4/94
  30  * $FreeBSD: src/sys/net/if.c,v 1.185 2004/03/13 02:35:03 brooks Exp $
  31  */
  32
  33 #include "opt_inet6.h"
  34 #include "opt_inet.h"
  35 #include "opt_ifpoll.h"
  36
  37 #include <sys/param.h>
  38 #include <sys/malloc.h>
  39 #include <sys/mbuf.h>
  40 #include <sys/systm.h>
  41 #include <sys/proc.h>
  42 #include <sys/priv.h>
  43 #include <sys/protosw.h>
  44 #include <sys/socket.h>
  45 #include <sys/socketvar.h>
  46 #include <sys/socketops.h>
  47 #include <sys/kernel.h>
  48 #include <sys/ktr.h>
  49 #include <sys/mutex.h>
  50 #include <sys/sockio.h>
  51 #include <sys/syslog.h>
  52 #include <sys/sysctl.h>
  53 #include <sys/domain.h>
  54 #include <sys/thread.h>
  55 #include <sys/serialize.h>
  56 #include <sys/bus.h>
  57
  58 #include <sys/thread2.h>
  59 #include <sys/msgport2.h>
  60 #include <sys/mutex2.h>
  61
  62 #include <net/if.h>
  63 #include <net/if_arp.h>
  64 #include <net/if_dl.h>
  65 #include <net/if_types.h>
  66 #include <net/if_var.h>
  67 #include <net/if_ringmap.h>
  68 #include <net/ifq_var.h>
  69 #include <net/radix.h>
  70 #include <net/route.h>
  71 #include <net/if_clone.h>
  72 #include <net/netisr2.h>
  73 #include <net/netmsg2.h>
  74
  75 #include <machine/atomic.h>
  76 #include <machine/stdarg.h>
  77 #include <machine/smp.h>
  78
  79 #if defined(INET) || defined(INET6)
  80 /*XXX*/
  81 #include <netinet/in.h>
  82 #include <netinet/in_var.h>
  83 #include <netinet/if_ether.h>
  84 #ifdef INET6
  85 #include <netinet6/in6_var.h>
  86 #include <netinet6/in6_ifattach.h>
  87 #endif
  88 #endif
  89
  90 struct netmsg_ifaddr {
  91         struct netmsg_base base;
  92         struct ifaddr   *ifa;
  93         struct ifnet    *ifp;
  94         int             tail;
  95 };
  96
  97 struct ifsubq_stage_head {
  98         TAILQ_HEAD(, ifsubq_stage)      stg_head;
  99 } __cachealign;
 100
 101 struct if_ringmap {
 102         int             rm_cnt;
 103         int             rm_grid;
 104         int             rm_cpumap[];
 105 };
 106
 107 #define RINGMAP_FLAG_NONE               0x0
 108 #define RINGMAP_FLAG_POWEROF2           0x1
 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 *);
 120 static int      if_rtdel(struct radix_node *, void *);
 121 static void     if_slowtimo_dispatch(netmsg_t);
 122
 123 /* Helper functions */
 124 static void     ifsq_watchdog_reset(struct ifsubq_watchdog *);
 125 static int      if_delmulti_serialized(struct ifnet *, struct sockaddr *);
 126 static struct ifnet_array *ifnet_array_alloc(int);
 127 static void     ifnet_array_free(struct ifnet_array *);
 128 static struct ifnet_array *ifnet_array_add(struct ifnet *,
 129                     const struct ifnet_array *);
 130 static struct ifnet_array *ifnet_array_del(struct ifnet *,
 131                     const struct ifnet_array *);
 132
 133 #ifdef INET6
 134 /*
 135  * XXX: declare here to avoid to include many inet6 related files..
 136  * should be more generalized?
 137  */
 138 extern void     nd6_setmtu(struct ifnet *);
 139 #endif
 140
 141 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers");
 142 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management");
 143 SYSCTL_NODE(_net_link, OID_AUTO, ringmap, CTLFLAG_RW, 0, "link ringmap");
 144
 145 static int ifsq_stage_cntmax = 4;
 146 TUNABLE_INT("net.link.stage_cntmax", &ifsq_stage_cntmax);
 147 SYSCTL_INT(_net_link, OID_AUTO, stage_cntmax, CTLFLAG_RW,
 148     &ifsq_stage_cntmax, 0, "ifq staging packet count max");
 149
 150 static int if_stats_compat = 0;
 151 SYSCTL_INT(_net_link, OID_AUTO, stats_compat, CTLFLAG_RW,
 152     &if_stats_compat, 0, "Compat the old ifnet stats");
 153
 154 static int if_ringmap_dumprdr = 0;
 155 SYSCTL_INT(_net_link_ringmap, OID_AUTO, dump_rdr, CTLFLAG_RW,
 156     &if_ringmap_dumprdr, 0, "dump redirect table");
 157
 158 SYSINIT(interfaces, SI_SUB_PROTO_IF, SI_ORDER_FIRST, ifinit, NULL);
 159 SYSINIT(ifnet, SI_SUB_PRE_DRIVERS, SI_ORDER_ANY, ifnetinit, NULL);
 160
 161 static  if_com_alloc_t *if_com_alloc[256];
 162 static  if_com_free_t *if_com_free[256];
 163
 164 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
 165 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
 166 MALLOC_DEFINE(M_IFNET, "ifnet", "interface structure");
 167
 168 int                     ifqmaxlen = IFQ_MAXLEN;
 169 struct ifnethead        ifnet = TAILQ_HEAD_INITIALIZER(ifnet);
 170
 171 static struct ifnet_array       ifnet_array0;
 172 static struct ifnet_array       *ifnet_array = &ifnet_array0;
 173
 174 static struct callout           if_slowtimo_timer;
 175 static struct netmsg_base       if_slowtimo_netmsg;
 176
 177 int                     if_index = 0;
 178 struct ifnet            **ifindex2ifnet = NULL;
 179 static struct mtx       ifnet_mtx = MTX_INITIALIZER("ifnet");
 180
 181 static struct ifsubq_stage_head ifsubq_stage_heads[MAXCPU];
 182
 183 #ifdef notyet
 184 #define IFQ_KTR_STRING          "ifq=%p"
 185 #define IFQ_KTR_ARGS    struct ifaltq *ifq
 186 #ifndef KTR_IFQ
 187 #define KTR_IFQ                 KTR_ALL
 188 #endif
 189 KTR_INFO_MASTER(ifq);
 190 KTR_INFO(KTR_IFQ, ifq, enqueue, 0, IFQ_KTR_STRING, IFQ_KTR_ARGS);
 191 KTR_INFO(KTR_IFQ, ifq, dequeue, 1, IFQ_KTR_STRING, IFQ_KTR_ARGS);
 192 #define logifq(name, arg)       KTR_LOG(ifq_ ## name, arg)
 193
 194 #define IF_START_KTR_STRING     "ifp=%p"
 195 #define IF_START_KTR_ARGS       struct ifnet *ifp
 196 #ifndef KTR_IF_START
 197 #define KTR_IF_START            KTR_ALL
 198 #endif
 199 KTR_INFO_MASTER(if_start);
 200 KTR_INFO(KTR_IF_START, if_start, run, 0,
 201          IF_START_KTR_STRING, IF_START_KTR_ARGS);
 202 KTR_INFO(KTR_IF_START, if_start, sched, 1,
 203          IF_START_KTR_STRING, IF_START_KTR_ARGS);
 204 KTR_INFO(KTR_IF_START, if_start, avoid, 2,
 205          IF_START_KTR_STRING, IF_START_KTR_ARGS);
 206 KTR_INFO(KTR_IF_START, if_start, contend_sched, 3,
 207          IF_START_KTR_STRING, IF_START_KTR_ARGS);
 208 KTR_INFO(KTR_IF_START, if_start, chase_sched, 4,
 209          IF_START_KTR_STRING, IF_START_KTR_ARGS);
 210 #define logifstart(name, arg)   KTR_LOG(if_start_ ## name, arg)
 211 #endif
 212
 213 TAILQ_HEAD(, ifg_group) ifg_head = TAILQ_HEAD_INITIALIZER(ifg_head);
 214
 215 /*
 216  * Network interface utility routines.
 217  *
 218  * Routines with ifa_ifwith* names take sockaddr *'s as
 219  * parameters.
 220  */
 221 /* ARGSUSED*/
 222 static void
 223 ifinit(void *dummy)
 224 {
 225         struct ifnet *ifp;
 226
 227         callout_init_mp(&if_slowtimo_timer);
 228         netmsg_init(&if_slowtimo_netmsg, NULL, &netisr_adone_rport,
 229             MSGF_PRIORITY, if_slowtimo_dispatch);
 230
 231         /* XXX is this necessary? */
 232         ifnet_lock();
 233         TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
 234                 if (ifp->if_snd.altq_maxlen == 0) {
 235                         if_printf(ifp, "XXX: driver didn't set altq_maxlen\n");
 236                         ifq_set_maxlen(&ifp->if_snd, ifqmaxlen);
 237                 }
 238         }
 239         ifnet_unlock();
 240
 241         /* Start if_slowtimo */
 242         lwkt_sendmsg(netisr_cpuport(0), &if_slowtimo_netmsg.lmsg);
 243 }
 244
 245 static void
 246 ifsq_ifstart_ipifunc(void *arg)
 247 {
 248         struct ifaltq_subque *ifsq = arg;
 249         struct lwkt_msg *lmsg = ifsq_get_ifstart_lmsg(ifsq, mycpuid);
 250
 251         crit_enter();
 252         if (lmsg->ms_flags & MSGF_DONE)
 253                 lwkt_sendmsg_oncpu(netisr_cpuport(mycpuid), lmsg);
 254         crit_exit();
 255 }
 256
 257 static __inline void
 258 ifsq_stage_remove(struct ifsubq_stage_head *head, struct ifsubq_stage *stage)
 259 {
 260         KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
 261         TAILQ_REMOVE(&head->stg_head, stage, stg_link);
 262         stage->stg_flags &= ~(IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED);
 263         stage->stg_cnt = 0;
 264         stage->stg_len = 0;
 265 }
 266
 267 static __inline void
 268 ifsq_stage_insert(struct ifsubq_stage_head *head, struct ifsubq_stage *stage)
 269 {
 270         KKASSERT((stage->stg_flags &
 271             (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
 272         stage->stg_flags |= IFSQ_STAGE_FLAG_QUED;
 273         TAILQ_INSERT_TAIL(&head->stg_head, stage, stg_link);
 274 }
 275
 276 /*
 277  * Schedule ifnet.if_start on the subqueue owner CPU
 278  */
 279 static void
 280 ifsq_ifstart_schedule(struct ifaltq_subque *ifsq, int force)
 281 {
 282         int cpu;
 283
 284         if (!force && curthread->td_type == TD_TYPE_NETISR &&
 285             ifsq_stage_cntmax > 0) {
 286                 struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid);
 287
 288                 stage->stg_cnt = 0;
 289                 stage->stg_len = 0;
 290                 if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
 291                         ifsq_stage_insert(&ifsubq_stage_heads[mycpuid], stage);
 292                 stage->stg_flags |= IFSQ_STAGE_FLAG_SCHED;
 293                 return;
 294         }
 295
 296         cpu = ifsq_get_cpuid(ifsq);
 297         if (cpu != mycpuid)
 298                 lwkt_send_ipiq(globaldata_find(cpu), ifsq_ifstart_ipifunc, ifsq);
 299         else
 300                 ifsq_ifstart_ipifunc(ifsq);
 301 }
 302
 303 /*
 304  * NOTE:
 305  * This function will release ifnet.if_start subqueue interlock,
 306  * if ifnet.if_start for the subqueue does not need to be scheduled
 307  */
 308 static __inline int
 309 ifsq_ifstart_need_schedule(struct ifaltq_subque *ifsq, int running)
 310 {
 311         if (!running || ifsq_is_empty(ifsq)
 312 #ifdef ALTQ
 313             || ifsq->ifsq_altq->altq_tbr != NULL
 314 #endif
 315         ) {
 316                 ALTQ_SQ_LOCK(ifsq);
 317                 /*
 318                  * ifnet.if_start subqueue interlock is released, if:
 319                  * 1) Hardware can not take any packets, due to
 320                  *    o  interface is marked down
 321                  *    o  hardware queue is full (ifsq_is_oactive)
 322                  *    Under the second situation, hardware interrupt
 323                  *    or polling(4) will call/schedule ifnet.if_start
 324                  *    on the subqueue when hardware queue is ready
 325                  * 2) There is no packet in the subqueue.
 326                  *    Further ifq_dispatch or ifq_handoff will call/
 327                  *    schedule ifnet.if_start on the subqueue.
 328                  * 3) TBR is used and it does not allow further
 329                  *    dequeueing.
 330                  *    TBR callout will call ifnet.if_start on the
 331                  *    subqueue.
 332                  */
 333                 if (!running || !ifsq_data_ready(ifsq)) {
 334                         ifsq_clr_started(ifsq);
 335                         ALTQ_SQ_UNLOCK(ifsq);
 336                         return 0;
 337                 }
 338                 ALTQ_SQ_UNLOCK(ifsq);
 339         }
 340         return 1;
 341 }
 342
 343 static void
 344 ifsq_ifstart_dispatch(netmsg_t msg)
 345 {
 346         struct lwkt_msg *lmsg = &msg->base.lmsg;
 347         struct ifaltq_subque *ifsq = lmsg->u.ms_resultp;
 348         struct ifnet *ifp = ifsq_get_ifp(ifsq);
 349         struct globaldata *gd = mycpu;
 350         int running = 0, need_sched;
 351
 352         crit_enter_gd(gd);
 353
 354         lwkt_replymsg(lmsg, 0); /* reply ASAP */
 355
 356         if (gd->gd_cpuid != ifsq_get_cpuid(ifsq)) {
 357                 /*
 358                  * We need to chase the subqueue owner CPU change.
 359                  */
 360                 ifsq_ifstart_schedule(ifsq, 1);
 361                 crit_exit_gd(gd);
 362                 return;
 363         }
 364
 365         ifsq_serialize_hw(ifsq);
 366         if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
 367                 ifp->if_start(ifp, ifsq);
 368                 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
 369                         running = 1;
 370         }
 371         need_sched = ifsq_ifstart_need_schedule(ifsq, running);
 372         ifsq_deserialize_hw(ifsq);
 373
 374         if (need_sched) {
 375                 /*
 376                  * More data need to be transmitted, ifnet.if_start is
 377                  * scheduled on the subqueue owner CPU, and we keep going.
 378                  * NOTE: ifnet.if_start subqueue interlock is not released.
 379                  */
 380                 ifsq_ifstart_schedule(ifsq, 0);
 381         }
 382
 383         crit_exit_gd(gd);
 384 }
 385
 386 /* Device driver ifnet.if_start helper function */
 387 void
 388 ifsq_devstart(struct ifaltq_subque *ifsq)
 389 {
 390         struct ifnet *ifp = ifsq_get_ifp(ifsq);
 391         int running = 0;
 392
 393         ASSERT_ALTQ_SQ_SERIALIZED_HW(ifsq);
 394
 395         ALTQ_SQ_LOCK(ifsq);
 396         if (ifsq_is_started(ifsq) || !ifsq_data_ready(ifsq)) {
 397                 ALTQ_SQ_UNLOCK(ifsq);
 398                 return;
 399         }
 400         ifsq_set_started(ifsq);
 401         ALTQ_SQ_UNLOCK(ifsq);
 402
 403         ifp->if_start(ifp, ifsq);
 404
 405         if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
 406                 running = 1;
 407
 408         if (ifsq_ifstart_need_schedule(ifsq, running)) {
 409                 /*
 410                  * More data need to be transmitted, ifnet.if_start is
 411                  * scheduled on ifnet's CPU, and we keep going.
 412                  * NOTE: ifnet.if_start interlock is not released.
 413                  */
 414                 ifsq_ifstart_schedule(ifsq, 0);
 415         }
 416 }
 417
 418 void
 419 if_devstart(struct ifnet *ifp)
 420 {
 421         ifsq_devstart(ifq_get_subq_default(&ifp->if_snd));
 422 }
 423
 424 /* Device driver ifnet.if_start schedule helper function */
 425 void
 426 ifsq_devstart_sched(struct ifaltq_subque *ifsq)
 427 {
 428         ifsq_ifstart_schedule(ifsq, 1);
 429 }
 430
 431 void
 432 if_devstart_sched(struct ifnet *ifp)
 433 {
 434         ifsq_devstart_sched(ifq_get_subq_default(&ifp->if_snd));
 435 }
 436
 437 static void
 438 if_default_serialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
 439 {
 440         lwkt_serialize_enter(ifp->if_serializer);
 441 }
 442
 443 static void
 444 if_default_deserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
 445 {
 446         lwkt_serialize_exit(ifp->if_serializer);
 447 }
 448
 449 static int
 450 if_default_tryserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
 451 {
 452         return lwkt_serialize_try(ifp->if_serializer);
 453 }
 454
 455 #ifdef INVARIANTS
 456 static void
 457 if_default_serialize_assert(struct ifnet *ifp,
 458                             enum ifnet_serialize slz __unused,
 459                             boolean_t serialized)
 460 {
 461         if (serialized)
 462                 ASSERT_SERIALIZED(ifp->if_serializer);
 463         else
 464                 ASSERT_NOT_SERIALIZED(ifp->if_serializer);
 465 }
 466 #endif
 467
 468 /*
 469  * Attach an interface to the list of "active" interfaces.
 470  *
 471  * The serializer is optional.
 472  */
 473 void
 474 if_attach(struct ifnet *ifp, lwkt_serialize_t serializer)
 475 {
 476         unsigned socksize;
 477         int namelen, masklen;
 478         struct sockaddr_dl *sdl, *sdl_addr;
 479         struct ifaddr *ifa;
 480         struct ifaltq *ifq;
 481         struct ifnet **old_ifindex2ifnet = NULL;
 482         struct ifnet_array *old_ifnet_array;
 483         int i, q;
 484
 485         static int if_indexlim = 8;
 486
 487         if (ifp->if_serialize != NULL) {
 488                 KASSERT(ifp->if_deserialize != NULL &&
 489                         ifp->if_tryserialize != NULL &&
 490                         ifp->if_serialize_assert != NULL,
 491                         ("serialize functions are partially setup"));
 492
 493                 /*
 494                  * If the device supplies serialize functions,
 495                  * then clear if_serializer to catch any invalid
 496                  * usage of this field.
 497                  */
 498                 KASSERT(serializer == NULL,
 499                         ("both serialize functions and default serializer "
 500                          "are supplied"));
 501                 ifp->if_serializer = NULL;
 502         } else {
 503                 KASSERT(ifp->if_deserialize == NULL &&
 504                         ifp->if_tryserialize == NULL &&
 505                         ifp->if_serialize_assert == NULL,
 506                         ("serialize functions are partially setup"));
 507                 ifp->if_serialize = if_default_serialize;
 508                 ifp->if_deserialize = if_default_deserialize;
 509                 ifp->if_tryserialize = if_default_tryserialize;
 510 #ifdef INVARIANTS
 511                 ifp->if_serialize_assert = if_default_serialize_assert;
 512 #endif
 513
 514                 /*
 515                  * The serializer can be passed in from the device,
 516                  * allowing the same serializer to be used for both
 517                  * the interrupt interlock and the device queue.
 518                  * If not specified, the netif structure will use an
 519                  * embedded serializer.
 520                  */
 521                 if (serializer == NULL) {
 522                         serializer = &ifp->if_default_serializer;
 523                         lwkt_serialize_init(serializer);
 524                 }
 525                 ifp->if_serializer = serializer;
 526         }
 527
 528         /*
 529          * XXX -
 530          * The old code would work if the interface passed a pre-existing
 531          * chain of ifaddrs to this code.  We don't trust our callers to
 532          * properly initialize the tailq, however, so we no longer allow
 533          * this unlikely case.
 534          */
 535         ifp->if_addrheads = kmalloc(ncpus * sizeof(struct ifaddrhead),
 536                                     M_IFADDR, M_WAITOK | M_ZERO);
 537         for (i = 0; i < ncpus; ++i)
 538                 TAILQ_INIT(&ifp->if_addrheads[i]);
 539
 540         TAILQ_INIT(&ifp->if_multiaddrs);
 541         TAILQ_INIT(&ifp->if_groups);
 542         getmicrotime(&ifp->if_lastchange);
 543
 544         /*
 545          * create a Link Level name for this device
 546          */
 547         namelen = strlen(ifp->if_xname);
 548         masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + namelen;
 549         socksize = masklen + ifp->if_addrlen;
 550         if (socksize < sizeof(*sdl))
 551                 socksize = sizeof(*sdl);
 552         socksize = RT_ROUNDUP(socksize);
 553         ifa = ifa_create(sizeof(struct ifaddr) + 2 * socksize);
 554         sdl = sdl_addr = (struct sockaddr_dl *)(ifa + 1);
 555         sdl->sdl_len = socksize;
 556         sdl->sdl_family = AF_LINK;
 557         bcopy(ifp->if_xname, sdl->sdl_data, namelen);
 558         sdl->sdl_nlen = namelen;
 559         sdl->sdl_type = ifp->if_type;
 560         ifp->if_lladdr = ifa;
 561         ifa->ifa_ifp = ifp;
 562         ifa->ifa_rtrequest = link_rtrequest;
 563         ifa->ifa_addr = (struct sockaddr *)sdl;
 564         sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
 565         ifa->ifa_netmask = (struct sockaddr *)sdl;
 566         sdl->sdl_len = masklen;
 567         while (namelen != 0)
 568                 sdl->sdl_data[--namelen] = 0xff;
 569         ifa_iflink(ifa, ifp, 0 /* Insert head */);
 570
 571         ifp->if_data_pcpu = kmalloc_cachealign(
 572             ncpus * sizeof(struct ifdata_pcpu), M_DEVBUF, M_WAITOK | M_ZERO);
 573
 574         if (ifp->if_mapsubq == NULL)
 575                 ifp->if_mapsubq = ifq_mapsubq_default;
 576
 577         ifq = &ifp->if_snd;
 578         ifq->altq_type = 0;
 579         ifq->altq_disc = NULL;
 580         ifq->altq_flags &= ALTQF_CANTCHANGE;
 581         ifq->altq_tbr = NULL;
 582         ifq->altq_ifp = ifp;
 583
 584         if (ifq->altq_subq_cnt <= 0)
 585                 ifq->altq_subq_cnt = 1;
 586         ifq->altq_subq = kmalloc_cachealign(
 587             ifq->altq_subq_cnt * sizeof(struct ifaltq_subque),
 588             M_DEVBUF, M_WAITOK | M_ZERO);
 589
 590         if (ifq->altq_maxlen == 0) {
 591                 if_printf(ifp, "driver didn't set altq_maxlen\n");
 592                 ifq_set_maxlen(ifq, ifqmaxlen);
 593         }
 594
 595         for (q = 0; q < ifq->altq_subq_cnt; ++q) {
 596                 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
 597
 598                 ALTQ_SQ_LOCK_INIT(ifsq);
 599                 ifsq->ifsq_index = q;
 600
 601                 ifsq->ifsq_altq = ifq;
 602                 ifsq->ifsq_ifp = ifp;
 603
 604                 ifsq->ifsq_maxlen = ifq->altq_maxlen;
 605                 ifsq->ifsq_maxbcnt = ifsq->ifsq_maxlen * MCLBYTES;
 606                 ifsq->ifsq_prepended = NULL;
 607                 ifsq->ifsq_started = 0;
 608                 ifsq->ifsq_hw_oactive = 0;
 609                 ifsq_set_cpuid(ifsq, 0);
 610                 if (ifp->if_serializer != NULL)
 611                         ifsq_set_hw_serialize(ifsq, ifp->if_serializer);
 612
 613                 ifsq->ifsq_stage =
 614                     kmalloc_cachealign(ncpus * sizeof(struct ifsubq_stage),
 615                     M_DEVBUF, M_WAITOK | M_ZERO);
 616                 for (i = 0; i < ncpus; ++i)
 617                         ifsq->ifsq_stage[i].stg_subq = ifsq;
 618
 619                 ifsq->ifsq_ifstart_nmsg =
 620                     kmalloc(ncpus * sizeof(struct netmsg_base),
 621                     M_LWKTMSG, M_WAITOK);
 622                 for (i = 0; i < ncpus; ++i) {
 623                         netmsg_init(&ifsq->ifsq_ifstart_nmsg[i], NULL,
 624                             &netisr_adone_rport, 0, ifsq_ifstart_dispatch);
 625                         ifsq->ifsq_ifstart_nmsg[i].lmsg.u.ms_resultp = ifsq;
 626                 }
 627         }
 628         ifq_set_classic(ifq);
 629
 630         /*
 631          * Increase mbuf cluster/jcluster limits for the mbufs that
 632          * could sit on the device queues for quite some time.
 633          */
 634         if (ifp->if_nmbclusters > 0)
 635                 mcl_inclimit(ifp->if_nmbclusters);
 636         if (ifp->if_nmbjclusters > 0)
 637                 mjcl_inclimit(ifp->if_nmbjclusters);
 638
 639         /*
 640          * Install this ifp into ifindex2inet, ifnet queue and ifnet
 641          * array after it is setup.
 642          *
 643          * Protect ifindex2ifnet, ifnet queue and ifnet array changes
 644          * by ifnet lock, so that non-netisr threads could get a
 645          * consistent view.
 646          */
 647         ifnet_lock();
 648
 649         /* Don't update if_index until ifindex2ifnet is setup */
 650         ifp->if_index = if_index + 1;
 651         sdl_addr->sdl_index = ifp->if_index;
 652
 653         /*
 654          * Install this ifp into ifindex2ifnet
 655          */
 656         if (ifindex2ifnet == NULL || ifp->if_index >= if_indexlim) {
 657                 unsigned int n;
 658                 struct ifnet **q;
 659
 660                 /*
 661                  * Grow ifindex2ifnet
 662                  */
 663                 if_indexlim <<= 1;
 664                 n = if_indexlim * sizeof(*q);
 665                 q = kmalloc(n, M_IFADDR, M_WAITOK | M_ZERO);
 666                 if (ifindex2ifnet != NULL) {
 667                         bcopy(ifindex2ifnet, q, n/2);
 668                         /* Free old ifindex2ifnet after sync all netisrs */
 669                         old_ifindex2ifnet = ifindex2ifnet;
 670                 }
 671                 ifindex2ifnet = q;
 672         }
 673         ifindex2ifnet[ifp->if_index] = ifp;
 674         /*
 675          * Update if_index after this ifp is installed into ifindex2ifnet,
 676          * so that netisrs could get a consistent view of ifindex2ifnet.
 677          */
 678         cpu_sfence();
 679         if_index = ifp->if_index;
 680
 681         /*
 682          * Install this ifp into ifnet array.
 683          */
 684         /* Free old ifnet array after sync all netisrs */
 685         old_ifnet_array = ifnet_array;
 686         ifnet_array = ifnet_array_add(ifp, old_ifnet_array);
 687
 688         /*
 689          * Install this ifp into ifnet queue.
 690          */
 691         TAILQ_INSERT_TAIL(&ifnetlist, ifp, if_link);
 692
 693         ifnet_unlock();
 694
 695         /*
 696          * Sync all netisrs so that the old ifindex2ifnet and ifnet array
 697          * are no longer accessed and we can free them safely later on.
 698          */
 699         netmsg_service_sync();
 700         if (old_ifindex2ifnet != NULL)
 701                 kfree(old_ifindex2ifnet, M_IFADDR);
 702         ifnet_array_free(old_ifnet_array);
 703
 704         if (!SLIST_EMPTY(&domains))
 705                 if_attachdomain1(ifp);
 706
 707         /* Announce the interface. */
 708         EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
 709         devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
 710         rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
 711 }
 712
 713 static void
 714 if_attachdomain(void *dummy)
 715 {
 716         struct ifnet *ifp;
 717
 718         ifnet_lock();
 719         TAILQ_FOREACH(ifp, &ifnetlist, if_list)
 720                 if_attachdomain1(ifp);
 721         ifnet_unlock();
 722 }
 723 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_FIRST,
 724         if_attachdomain, NULL);
 725
 726 static void
 727 if_attachdomain1(struct ifnet *ifp)
 728 {
 729         struct domain *dp;
 730
 731         crit_enter();
 732
 733         /* address family dependent data region */
 734         bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
 735         SLIST_FOREACH(dp, &domains, dom_next)
 736                 if (dp->dom_ifattach)
 737                         ifp->if_afdata[dp->dom_family] =
 738                                 (*dp->dom_ifattach)(ifp);
 739         crit_exit();
 740 }
 741
 742 /*
 743  * Purge all addresses whose type is _not_ AF_LINK
 744  */
 745 static void
 746 if_purgeaddrs_nolink_dispatch(netmsg_t nmsg)
 747 {
 748         struct lwkt_msg *lmsg = &nmsg->lmsg;
 749         struct ifnet *ifp = lmsg->u.ms_resultp;
 750         struct ifaddr_container *ifac, *next;
 751
 752         ASSERT_IN_NETISR(0);
 753
 754         /*
 755          * The ifaddr processing in the following loop will block,
 756          * however, this function is called in netisr0, in which
 757          * ifaddr list changes happen, so we don't care about the
 758          * blockness of the ifaddr processing here.
 759          */
 760         TAILQ_FOREACH_MUTABLE(ifac, &ifp->if_addrheads[mycpuid],
 761                               ifa_link, next) {
 762                 struct ifaddr *ifa = ifac->ifa;
 763
 764                 /* Ignore marker */
 765                 if (ifa->ifa_addr->sa_family == AF_UNSPEC)
 766                         continue;
 767
 768                 /* Leave link ifaddr as it is */
 769                 if (ifa->ifa_addr->sa_family == AF_LINK)
 770                         continue;
 771 #ifdef INET
 772                 /* XXX: Ugly!! ad hoc just for INET */
 773                 if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET) {
 774                         struct ifaliasreq ifr;
 775 #ifdef IFADDR_DEBUG_VERBOSE
 776                         int i;
 777
 778                         kprintf("purge in4 addr %p: ", ifa);
 779                         for (i = 0; i < ncpus; ++i)
 780                                 kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
 781                         kprintf("\n");
 782 #endif
 783
 784                         bzero(&ifr, sizeof ifr);
 785                         ifr.ifra_addr = *ifa->ifa_addr;
 786                         if (ifa->ifa_dstaddr)
 787                                 ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
 788                         if (in_control(SIOCDIFADDR, (caddr_t)&ifr, ifp,
 789                                        NULL) == 0)
 790                                 continue;
 791                 }
 792 #endif /* INET */
 793 #ifdef INET6
 794                 if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET6) {
 795 #ifdef IFADDR_DEBUG_VERBOSE
 796                         int i;
 797
 798                         kprintf("purge in6 addr %p: ", ifa);
 799                         for (i = 0; i < ncpus; ++i)
 800                                 kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
 801                         kprintf("\n");
 802 #endif
 803
 804                         in6_purgeaddr(ifa);
 805                         /* ifp_addrhead is already updated */
 806                         continue;
 807                 }
 808 #endif /* INET6 */
 809                 ifa_ifunlink(ifa, ifp);
 810                 ifa_destroy(ifa);
 811         }
 812
 813         lwkt_replymsg(lmsg, 0);
 814 }
 815
 816 void
 817 if_purgeaddrs_nolink(struct ifnet *ifp)
 818 {
 819         struct netmsg_base nmsg;
 820         struct lwkt_msg *lmsg = &nmsg.lmsg;
 821
 822         ASSERT_CANDOMSG_NETISR0(curthread);
 823
 824         netmsg_init(&nmsg, NULL, &curthread->td_msgport, 0,
 825             if_purgeaddrs_nolink_dispatch);
 826         lmsg->u.ms_resultp = ifp;
 827         lwkt_domsg(netisr_cpuport(0), lmsg, 0);
 828 }
 829
 830 static void
 831 ifq_stage_detach_handler(netmsg_t nmsg)
 832 {
 833         struct ifaltq *ifq = nmsg->lmsg.u.ms_resultp;
 834         int q;
 835
 836         for (q = 0; q < ifq->altq_subq_cnt; ++q) {
 837                 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
 838                 struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid);
 839
 840                 if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED)
 841                         ifsq_stage_remove(&ifsubq_stage_heads[mycpuid], stage);
 842         }
 843         lwkt_replymsg(&nmsg->lmsg, 0);
 844 }
 845
 846 static void
 847 ifq_stage_detach(struct ifaltq *ifq)
 848 {
 849         struct netmsg_base base;
 850         int cpu;
 851
 852         netmsg_init(&base, NULL, &curthread->td_msgport, 0,
 853             ifq_stage_detach_handler);
 854         base.lmsg.u.ms_resultp = ifq;
 855
 856         for (cpu = 0; cpu < ncpus; ++cpu)
 857                 lwkt_domsg(netisr_cpuport(cpu), &base.lmsg, 0);
 858 }
 859
 860 struct netmsg_if_rtdel {
 861         struct netmsg_base      base;
 862         struct ifnet            *ifp;
 863 };
 864
 865 static void
 866 if_rtdel_dispatch(netmsg_t msg)
 867 {
 868         struct netmsg_if_rtdel *rmsg = (void *)msg;
 869         int i, nextcpu, cpu;
 870
 871         cpu = mycpuid;
 872         for (i = 1; i <= AF_MAX; i++) {
 873                 struct radix_node_head  *rnh;
 874
 875                 if ((rnh = rt_tables[cpu][i]) == NULL)
 876                         continue;
 877                 rnh->rnh_walktree(rnh, if_rtdel, rmsg->ifp);
 878         }
 879
 880         nextcpu = cpu + 1;
 881         if (nextcpu < ncpus)
 882                 lwkt_forwardmsg(netisr_cpuport(nextcpu), &rmsg->base.lmsg);
 883         else
 884                 lwkt_replymsg(&rmsg->base.lmsg, 0);
 885 }
 886
 887 /*
 888  * Detach an interface, removing it from the
 889  * list of "active" interfaces.
 890  */
 891 void
 892 if_detach(struct ifnet *ifp)
 893 {
 894         struct ifnet_array *old_ifnet_array;
 895         struct netmsg_if_rtdel msg;
 896         struct domain *dp;
 897         int q;
 898
 899         /* Announce that the interface is gone. */
 900         EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
 901         rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
 902         devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
 903
 904         /*
 905          * Remove this ifp from ifindex2inet, ifnet queue and ifnet
 906          * array before it is whacked.
 907          *
 908          * Protect ifindex2ifnet, ifnet queue and ifnet array changes
 909          * by ifnet lock, so that non-netisr threads could get a
 910          * consistent view.
 911          */
 912         ifnet_lock();
 913
 914         /*
 915          * Remove this ifp from ifindex2ifnet and maybe decrement if_index.
 916          */
 917         ifindex2ifnet[ifp->if_index] = NULL;
 918         while (if_index > 0 && ifindex2ifnet[if_index] == NULL)
 919                 if_index--;
 920
 921         /*
 922          * Remove this ifp from ifnet queue.
 923          */
 924         TAILQ_REMOVE(&ifnetlist, ifp, if_link);
 925
 926         /*
 927          * Remove this ifp from ifnet array.
 928          */
 929         /* Free old ifnet array after sync all netisrs */
 930         old_ifnet_array = ifnet_array;
 931         ifnet_array = ifnet_array_del(ifp, old_ifnet_array);
 932
 933         ifnet_unlock();
 934
 935         /*
 936          * Sync all netisrs so that the old ifnet array is no longer
 937          * accessed and we can free it safely later on.
 938          */
 939         netmsg_service_sync();
 940         ifnet_array_free(old_ifnet_array);
 941
 942         /*
 943          * Remove routes and flush queues.
 944          */
 945         crit_enter();
 946 #ifdef IFPOLL_ENABLE
 947         if (ifp->if_flags & IFF_NPOLLING)
 948                 ifpoll_deregister(ifp);
 949 #endif
 950         if_down(ifp);
 951
 952         /* Decrease the mbuf clusters/jclusters limits increased by us */
 953         if (ifp->if_nmbclusters > 0)
 954                 mcl_inclimit(-ifp->if_nmbclusters);
 955         if (ifp->if_nmbjclusters > 0)
 956                 mjcl_inclimit(-ifp->if_nmbjclusters);
 957
 958 #ifdef ALTQ
 959         if (ifq_is_enabled(&ifp->if_snd))
 960                 altq_disable(&ifp->if_snd);
 961         if (ifq_is_attached(&ifp->if_snd))
 962                 altq_detach(&ifp->if_snd);
 963 #endif
 964
 965         /*
 966          * Clean up all addresses.
 967          */
 968         ifp->if_lladdr = NULL;
 969
 970         if_purgeaddrs_nolink(ifp);
 971         if (!TAILQ_EMPTY(&ifp->if_addrheads[mycpuid])) {
 972                 struct ifaddr *ifa;
 973
 974                 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
 975                 KASSERT(ifa->ifa_addr->sa_family == AF_LINK,
 976                         ("non-link ifaddr is left on if_addrheads"));
 977
 978                 ifa_ifunlink(ifa, ifp);
 979                 ifa_destroy(ifa);
 980                 KASSERT(TAILQ_EMPTY(&ifp->if_addrheads[mycpuid]),
 981                         ("there are still ifaddrs left on if_addrheads"));
 982         }
 983
 984 #ifdef INET
 985         /*
 986          * Remove all IPv4 kernel structures related to ifp.
 987          */
 988         in_ifdetach(ifp);
 989 #endif
 990
 991 #ifdef INET6
 992         /*
 993          * Remove all IPv6 kernel structs related to ifp.  This should be done
 994          * before removing routing entries below, since IPv6 interface direct
 995          * routes are expected to be removed by the IPv6-specific kernel API.
 996          * Otherwise, the kernel will detect some inconsistency and bark it.
 997          */
 998         in6_ifdetach(ifp);
 999 #endif
1000
1001         /*
1002          * Delete all remaining routes using this interface
1003          */
1004         netmsg_init(&msg.base, NULL, &curthread->td_msgport, MSGF_PRIORITY,
1005             if_rtdel_dispatch);
1006         msg.ifp = ifp;
1007         rt_domsg_global(&msg.base);
1008
1009         SLIST_FOREACH(dp, &domains, dom_next)
1010                 if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
1011                         (*dp->dom_ifdetach)(ifp,
1012                                 ifp->if_afdata[dp->dom_family]);
1013
1014         kfree(ifp->if_addrheads, M_IFADDR);
1015
1016         lwkt_synchronize_ipiqs("if_detach");
1017         ifq_stage_detach(&ifp->if_snd);
1018
1019         for (q = 0; q < ifp->if_snd.altq_subq_cnt; ++q) {
1020                 struct ifaltq_subque *ifsq = &ifp->if_snd.altq_subq[q];
1021
1022                 kfree(ifsq->ifsq_ifstart_nmsg, M_LWKTMSG);
1023                 kfree(ifsq->ifsq_stage, M_DEVBUF);
1024         }
1025         kfree(ifp->if_snd.altq_subq, M_DEVBUF);
1026
1027         kfree(ifp->if_data_pcpu, M_DEVBUF);
1028
1029         crit_exit();
1030 }
1031
1032 /*
1033  * Create interface group without members
1034  */
1035 struct ifg_group *
1036 if_creategroup(const char *groupname)
1037 {
1038         struct ifg_group        *ifg = NULL;
1039
1040         if ((ifg = (struct ifg_group *)kmalloc(sizeof(struct ifg_group),
1041             M_TEMP, M_NOWAIT)) == NULL)
1042                 return (NULL);
1043
1044         strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group));
1045         ifg->ifg_refcnt = 0;
1046         ifg->ifg_carp_demoted = 0;
1047         TAILQ_INIT(&ifg->ifg_members);
1048 #if NPF > 0
1049         pfi_attach_ifgroup(ifg);
1050 #endif
1051         TAILQ_INSERT_TAIL(&ifg_head, ifg, ifg_next);
1052
1053         return (ifg);
1054 }
1055
1056 /*
1057  * Add a group to an interface
1058  */
1059 int
1060 if_addgroup(struct ifnet *ifp, const char *groupname)
1061 {
1062         struct ifg_list         *ifgl;
1063         struct ifg_group        *ifg = NULL;
1064         struct ifg_member       *ifgm;
1065
1066         if (groupname[0] && groupname[strlen(groupname) - 1] >= '0' &&
1067             groupname[strlen(groupname) - 1] <= '9')
1068                 return (EINVAL);
1069
1070         TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1071                 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
1072                         return (EEXIST);
1073
1074         if ((ifgl = kmalloc(sizeof(*ifgl), M_TEMP, M_NOWAIT)) == NULL)
1075                 return (ENOMEM);
1076
1077         if ((ifgm = kmalloc(sizeof(*ifgm), M_TEMP, M_NOWAIT)) == NULL) {
1078                 kfree(ifgl, M_TEMP);
1079                 return (ENOMEM);
1080         }
1081
1082         TAILQ_FOREACH(ifg, &ifg_head, ifg_next)
1083                 if (!strcmp(ifg->ifg_group, groupname))
1084                         break;
1085
1086         if (ifg == NULL && (ifg = if_creategroup(groupname)) == NULL) {
1087                 kfree(ifgl, M_TEMP);
1088                 kfree(ifgm, M_TEMP);
1089                 return (ENOMEM);
1090         }
1091
1092         ifg->ifg_refcnt++;
1093         ifgl->ifgl_group = ifg;
1094         ifgm->ifgm_ifp = ifp;
1095
1096         TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next);
1097         TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next);
1098
1099 #if NPF > 0
1100         pfi_group_change(groupname);
1101 #endif
1102
1103         return (0);
1104 }
1105
1106 /*
1107  * Remove a group from an interface
1108  */
1109 int
1110 if_delgroup(struct ifnet *ifp, const char *groupname)
1111 {
1112         struct ifg_list         *ifgl;
1113         struct ifg_member       *ifgm;
1114
1115         TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1116                 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
1117                         break;
1118         if (ifgl == NULL)
1119                 return (ENOENT);
1120
1121         TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
1122
1123         TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
1124                 if (ifgm->ifgm_ifp == ifp)
1125                         break;
1126
1127         if (ifgm != NULL) {
1128                 TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next);
1129                 kfree(ifgm, M_TEMP);
1130         }
1131
1132         if (--ifgl->ifgl_group->ifg_refcnt == 0) {
1133                 TAILQ_REMOVE(&ifg_head, ifgl->ifgl_group, ifg_next);
1134 #if NPF > 0
1135                 pfi_detach_ifgroup(ifgl->ifgl_group);
1136 #endif
1137                 kfree(ifgl->ifgl_group, M_TEMP);
1138         }
1139
1140         kfree(ifgl, M_TEMP);
1141
1142 #if NPF > 0
1143         pfi_group_change(groupname);
1144 #endif
1145
1146         return (0);
1147 }
1148
1149 /*
1150  * Stores all groups from an interface in memory pointed
1151  * to by data
1152  */
1153 int
1154 if_getgroup(caddr_t data, struct ifnet *ifp)
1155 {
1156         int                      len, error;
1157         struct ifg_list         *ifgl;
1158         struct ifg_req           ifgrq, *ifgp;
1159         struct ifgroupreq       *ifgr = (struct ifgroupreq *)data;
1160
1161         if (ifgr->ifgr_len == 0) {
1162                 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1163                         ifgr->ifgr_len += sizeof(struct ifg_req);
1164                 return (0);
1165         }
1166
1167         len = ifgr->ifgr_len;
1168         ifgp = ifgr->ifgr_groups;
1169         TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1170                 if (len < sizeof(ifgrq))
1171                         return (EINVAL);
1172                 bzero(&ifgrq, sizeof ifgrq);
1173                 strlcpy(ifgrq.ifgrq_group, ifgl->ifgl_group->ifg_group,
1174                     sizeof(ifgrq.ifgrq_group));
1175                 if ((error = copyout((caddr_t)&ifgrq, (caddr_t)ifgp,
1176                     sizeof(struct ifg_req))))
1177                         return (error);
1178                 len -= sizeof(ifgrq);
1179                 ifgp++;
1180         }
1181
1182         return (0);
1183 }
1184
1185 /*
1186  * Stores all members of a group in memory pointed to by data
1187  */
1188 int
1189 if_getgroupmembers(caddr_t data)
1190 {
1191         struct ifgroupreq       *ifgr = (struct ifgroupreq *)data;
1192         struct ifg_group        *ifg;
1193         struct ifg_member       *ifgm;
1194         struct ifg_req           ifgrq, *ifgp;
1195         int                      len, error;
1196
1197         TAILQ_FOREACH(ifg, &ifg_head, ifg_next)
1198                 if (!strcmp(ifg->ifg_group, ifgr->ifgr_name))
1199                         break;
1200         if (ifg == NULL)
1201                 return (ENOENT);
1202
1203         if (ifgr->ifgr_len == 0) {
1204                 TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
1205                         ifgr->ifgr_len += sizeof(ifgrq);
1206                 return (0);
1207         }
1208
1209         len = ifgr->ifgr_len;
1210         ifgp = ifgr->ifgr_groups;
1211         TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
1212                 if (len < sizeof(ifgrq))
1213                         return (EINVAL);
1214                 bzero(&ifgrq, sizeof ifgrq);
1215                 strlcpy(ifgrq.ifgrq_member, ifgm->ifgm_ifp->if_xname,
1216                     sizeof(ifgrq.ifgrq_member));
1217                 if ((error = copyout((caddr_t)&ifgrq, (caddr_t)ifgp,
1218                     sizeof(struct ifg_req))))
1219                         return (error);
1220                 len -= sizeof(ifgrq);
1221                 ifgp++;
1222         }
1223
1224         return (0);
1225 }
1226
1227 /*
1228  * Delete Routes for a Network Interface
1229  *
1230  * Called for each routing entry via the rnh->rnh_walktree() call above
1231  * to delete all route entries referencing a detaching network interface.
1232  *
1233  * Arguments:
1234  *      rn      pointer to node in the routing table
1235  *      arg     argument passed to rnh->rnh_walktree() - detaching interface
1236  *
1237  * Returns:
1238  *      0       successful
1239  *      errno   failed - reason indicated
1240  *
1241  */
1242 static int
1243 if_rtdel(struct radix_node *rn, void *arg)
1244 {
1245         struct rtentry  *rt = (struct rtentry *)rn;
1246         struct ifnet    *ifp = arg;
1247         int             err;
1248
1249         if (rt->rt_ifp == ifp) {
1250
1251                 /*
1252                  * Protect (sorta) against walktree recursion problems
1253                  * with cloned routes
1254                  */
1255                 if (!(rt->rt_flags & RTF_UP))
1256                         return (0);
1257
1258                 err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1259                                 rt_mask(rt), rt->rt_flags,
1260                                 NULL);
1261                 if (err) {
1262                         log(LOG_WARNING, "if_rtdel: error %d\n", err);
1263                 }
1264         }
1265
1266         return (0);
1267 }
1268
1269 static __inline boolean_t
1270 ifa_prefer(const struct ifaddr *cur_ifa, const struct ifaddr *old_ifa)
1271 {
1272         if (old_ifa == NULL)
1273                 return TRUE;
1274
1275         if ((old_ifa->ifa_ifp->if_flags & IFF_UP) == 0 &&
1276             (cur_ifa->ifa_ifp->if_flags & IFF_UP))
1277                 return TRUE;
1278         if ((old_ifa->ifa_flags & IFA_ROUTE) == 0 &&
1279             (cur_ifa->ifa_flags & IFA_ROUTE))
1280                 return TRUE;
1281         return FALSE;
1282 }
1283
1284 /*
1285  * Locate an interface based on a complete address.
1286  */
1287 struct ifaddr *
1288 ifa_ifwithaddr(struct sockaddr *addr)
1289 {
1290         const struct ifnet_array *arr;
1291         int i;
1292
1293         arr = ifnet_array_get();
1294         for (i = 0; i < arr->ifnet_count; ++i) {
1295                 struct ifnet *ifp = arr->ifnet_arr[i];
1296                 struct ifaddr_container *ifac;
1297
1298                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1299                         struct ifaddr *ifa = ifac->ifa;
1300
1301                         if (ifa->ifa_addr->sa_family != addr->sa_family)
1302                                 continue;
1303                         if (sa_equal(addr, ifa->ifa_addr))
1304                                 return (ifa);
1305                         if ((ifp->if_flags & IFF_BROADCAST) &&
1306                             ifa->ifa_broadaddr &&
1307                             /* IPv6 doesn't have broadcast */
1308                             ifa->ifa_broadaddr->sa_len != 0 &&
1309                             sa_equal(ifa->ifa_broadaddr, addr))
1310                                 return (ifa);
1311                 }
1312         }
1313         return (NULL);
1314 }
1315
1316 /*
1317  * Locate the point to point interface with a given destination address.
1318  */
1319 struct ifaddr *
1320 ifa_ifwithdstaddr(struct sockaddr *addr)
1321 {
1322         const struct ifnet_array *arr;
1323         int i;
1324
1325         arr = ifnet_array_get();
1326         for (i = 0; i < arr->ifnet_count; ++i) {
1327                 struct ifnet *ifp = arr->ifnet_arr[i];
1328                 struct ifaddr_container *ifac;
1329
1330                 if (!(ifp->if_flags & IFF_POINTOPOINT))
1331                         continue;
1332
1333                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1334                         struct ifaddr *ifa = ifac->ifa;
1335
1336                         if (ifa->ifa_addr->sa_family != addr->sa_family)
1337                                 continue;
1338                         if (ifa->ifa_dstaddr &&
1339                             sa_equal(addr, ifa->ifa_dstaddr))
1340                                 return (ifa);
1341                 }
1342         }
1343         return (NULL);
1344 }
1345
1346 /*
1347  * Find an interface on a specific network.  If many, choice
1348  * is most specific found.
1349  */
1350 struct ifaddr *
1351 ifa_ifwithnet(struct sockaddr *addr)
1352 {
1353         struct ifaddr *ifa_maybe = NULL;
1354         u_int af = addr->sa_family;
1355         char *addr_data = addr->sa_data, *cplim;
1356         const struct ifnet_array *arr;
1357         int i;
1358
1359         /*
1360          * AF_LINK addresses can be looked up directly by their index number,
1361          * so do that if we can.
1362          */
1363         if (af == AF_LINK) {
1364                 struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
1365
1366                 if (sdl->sdl_index && sdl->sdl_index <= if_index)
1367                         return (ifindex2ifnet[sdl->sdl_index]->if_lladdr);
1368         }
1369
1370         /*
1371          * Scan though each interface, looking for ones that have
1372          * addresses in this address family.
1373          */
1374         arr = ifnet_array_get();
1375         for (i = 0; i < arr->ifnet_count; ++i) {
1376                 struct ifnet *ifp = arr->ifnet_arr[i];
1377                 struct ifaddr_container *ifac;
1378
1379                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1380                         struct ifaddr *ifa = ifac->ifa;
1381                         char *cp, *cp2, *cp3;
1382
1383                         if (ifa->ifa_addr->sa_family != af)
1384 next:                           continue;
1385                         if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT) {
1386                                 /*
1387                                  * This is a bit broken as it doesn't
1388                                  * take into account that the remote end may
1389                                  * be a single node in the network we are
1390                                  * looking for.
1391                                  * The trouble is that we don't know the
1392                                  * netmask for the remote end.
1393                                  */
1394                                 if (ifa->ifa_dstaddr != NULL &&
1395                                     sa_equal(addr, ifa->ifa_dstaddr))
1396                                         return (ifa);
1397                         } else {
1398                                 /*
1399                                  * if we have a special address handler,
1400                                  * then use it instead of the generic one.
1401                                  */
1402                                 if (ifa->ifa_claim_addr) {
1403                                         if ((*ifa->ifa_claim_addr)(ifa, addr)) {
1404                                                 return (ifa);
1405                                         } else {
1406                                                 continue;
1407                                         }
1408                                 }
1409
1410                                 /*
1411                                  * Scan all the bits in the ifa's address.
1412                                  * If a bit dissagrees with what we are
1413                                  * looking for, mask it with the netmask
1414                                  * to see if it really matters.
1415                                  * (A byte at a time)
1416                                  */
1417                                 if (ifa->ifa_netmask == 0)
1418                                         continue;
1419                                 cp = addr_data;
1420                                 cp2 = ifa->ifa_addr->sa_data;
1421                                 cp3 = ifa->ifa_netmask->sa_data;
1422                                 cplim = ifa->ifa_netmask->sa_len +
1423                                         (char *)ifa->ifa_netmask;
1424                                 while (cp3 < cplim)
1425                                         if ((*cp++ ^ *cp2++) & *cp3++)
1426                                                 goto next; /* next address! */
1427                                 /*
1428                                  * If the netmask of what we just found
1429                                  * is more specific than what we had before
1430                                  * (if we had one) then remember the new one
1431                                  * before continuing to search for an even
1432                                  * better one.  If the netmasks are equal,
1433                                  * we prefer the this ifa based on the result
1434                                  * of ifa_prefer().
1435                                  */
1436                                 if (ifa_maybe == NULL ||
1437                                     rn_refines((char *)ifa->ifa_netmask,
1438                                         (char *)ifa_maybe->ifa_netmask) ||
1439                                     (sa_equal(ifa_maybe->ifa_netmask,
1440                                         ifa->ifa_netmask) &&
1441                                      ifa_prefer(ifa, ifa_maybe)))
1442                                         ifa_maybe = ifa;
1443                         }
1444                 }
1445         }
1446         return (ifa_maybe);
1447 }
1448
1449 /*
1450  * Find an interface address specific to an interface best matching
1451  * a given address.
1452  */
1453 struct ifaddr *
1454 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1455 {
1456         struct ifaddr_container *ifac;
1457         char *cp, *cp2, *cp3;
1458         char *cplim;
1459         struct ifaddr *ifa_maybe = NULL;
1460         u_int af = addr->sa_family;
1461
1462         if (af >= AF_MAX)
1463                 return (0);
1464         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1465                 struct ifaddr *ifa = ifac->ifa;
1466
1467                 if (ifa->ifa_addr->sa_family != af)
1468                         continue;
1469                 if (ifa_maybe == NULL)
1470                         ifa_maybe = ifa;
1471                 if (ifa->ifa_netmask == NULL) {
1472                         if (sa_equal(addr, ifa->ifa_addr) ||
1473                             (ifa->ifa_dstaddr != NULL &&
1474                              sa_equal(addr, ifa->ifa_dstaddr)))
1475                                 return (ifa);
1476                         continue;
1477                 }
1478                 if (ifp->if_flags & IFF_POINTOPOINT) {
1479                         if (sa_equal(addr, ifa->ifa_dstaddr))
1480                                 return (ifa);
1481                 } else {
1482                         cp = addr->sa_data;
1483                         cp2 = ifa->ifa_addr->sa_data;
1484                         cp3 = ifa->ifa_netmask->sa_data;
1485                         cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1486                         for (; cp3 < cplim; cp3++)
1487                                 if ((*cp++ ^ *cp2++) & *cp3)
1488                                         break;
1489                         if (cp3 == cplim)
1490                                 return (ifa);
1491                 }
1492         }
1493         return (ifa_maybe);
1494 }
1495
1496 /*
1497  * Default action when installing a route with a Link Level gateway.
1498  * Lookup an appropriate real ifa to point to.
1499  * This should be moved to /sys/net/link.c eventually.
1500  */
1501 static void
1502 link_rtrequest(int cmd, struct rtentry *rt)
1503 {
1504         struct ifaddr *ifa;
1505         struct sockaddr *dst;
1506         struct ifnet *ifp;
1507
1508         if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL ||
1509             (ifp = ifa->ifa_ifp) == NULL || (dst = rt_key(rt)) == NULL)
1510                 return;
1511         ifa = ifaof_ifpforaddr(dst, ifp);
1512         if (ifa != NULL) {
1513                 IFAFREE(rt->rt_ifa);
1514                 IFAREF(ifa);
1515                 rt->rt_ifa = ifa;
1516                 if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1517                         ifa->ifa_rtrequest(cmd, rt);
1518         }
1519 }
1520
1521 struct netmsg_ifroute {
1522         struct netmsg_base      base;
1523         struct ifnet            *ifp;
1524         int                     flag;
1525         int                     fam;
1526 };
1527
1528 /*
1529  * Mark an interface down and notify protocols of the transition.
1530  */
1531 static void
1532 if_unroute_dispatch(netmsg_t nmsg)
1533 {
1534         struct netmsg_ifroute *msg = (struct netmsg_ifroute *)nmsg;
1535         struct ifnet *ifp = msg->ifp;
1536         int flag = msg->flag, fam = msg->fam;
1537         struct ifaddr_container *ifac;
1538
1539         ifp->if_flags &= ~flag;
1540         getmicrotime(&ifp->if_lastchange);
1541         /*
1542          * The ifaddr processing in the following loop will block,
1543          * however, this function is called in netisr0, in which
1544          * ifaddr list changes happen, so we don't care about the
1545          * blockness of the ifaddr processing here.
1546          */
1547         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1548                 struct ifaddr *ifa = ifac->ifa;
1549
1550                 /* Ignore marker */
1551                 if (ifa->ifa_addr->sa_family == AF_UNSPEC)
1552                         continue;
1553
1554                 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1555                         kpfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1556         }
1557         ifq_purge_all(&ifp->if_snd);
1558         rt_ifmsg(ifp);
1559
1560         lwkt_replymsg(&nmsg->lmsg, 0);
1561 }
1562
1563 void
1564 if_unroute(struct ifnet *ifp, int flag, int fam)
1565 {
1566         struct netmsg_ifroute msg;
1567
1568         ASSERT_CANDOMSG_NETISR0(curthread);
1569
1570         netmsg_init(&msg.base, NULL, &curthread->td_msgport, 0,
1571             if_unroute_dispatch);
1572         msg.ifp = ifp;
1573         msg.flag = flag;
1574         msg.fam = fam;
1575         lwkt_domsg(netisr_cpuport(0), &msg.base.lmsg, 0);
1576 }
1577
1578 /*
1579  * Mark an interface up and notify protocols of the transition.
1580  */
1581 static void
1582 if_route_dispatch(netmsg_t nmsg)
1583 {
1584         struct netmsg_ifroute *msg = (struct netmsg_ifroute *)nmsg;
1585         struct ifnet *ifp = msg->ifp;
1586         int flag = msg->flag, fam = msg->fam;
1587         struct ifaddr_container *ifac;
1588
1589         ifq_purge_all(&ifp->if_snd);
1590         ifp->if_flags |= flag;
1591         getmicrotime(&ifp->if_lastchange);
1592         /*
1593          * The ifaddr processing in the following loop will block,
1594          * however, this function is called in netisr0, in which
1595          * ifaddr list changes happen, so we don't care about the
1596          * blockness of the ifaddr processing here.
1597          */
1598         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1599                 struct ifaddr *ifa = ifac->ifa;
1600
1601                 /* Ignore marker */
1602                 if (ifa->ifa_addr->sa_family == AF_UNSPEC)
1603                         continue;
1604
1605                 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1606                         kpfctlinput(PRC_IFUP, ifa->ifa_addr);
1607         }
1608         rt_ifmsg(ifp);
1609 #ifdef INET6
1610         in6_if_up(ifp);
1611 #endif
1612
1613         lwkt_replymsg(&nmsg->lmsg, 0);
1614 }
1615
1616 void
1617 if_route(struct ifnet *ifp, int flag, int fam)
1618 {
1619         struct netmsg_ifroute msg;
1620
1621         ASSERT_CANDOMSG_NETISR0(curthread);
1622
1623         netmsg_init(&msg.base, NULL, &curthread->td_msgport, 0,
1624             if_route_dispatch);
1625         msg.ifp = ifp;
1626         msg.flag = flag;
1627         msg.fam = fam;
1628         lwkt_domsg(netisr_cpuport(0), &msg.base.lmsg, 0);
1629 }
1630
1631 /*
1632  * Mark an interface down and notify protocols of the transition.  An
1633  * interface going down is also considered to be a synchronizing event.
1634  * We must ensure that all packet processing related to the interface
1635  * has completed before we return so e.g. the caller can free the ifnet
1636  * structure that the mbufs may be referencing.
1637  *
1638  * NOTE: must be called at splnet or eqivalent.
1639  */
1640 void
1641 if_down(struct ifnet *ifp)
1642 {
1643         if_unroute(ifp, IFF_UP, AF_UNSPEC);
1644         netmsg_service_sync();
1645 }
1646
1647 /*
1648  * Mark an interface up and notify protocols of
1649  * the transition.
1650  * NOTE: must be called at splnet or eqivalent.
1651  */
1652 void
1653 if_up(struct ifnet *ifp)
1654 {
1655         if_route(ifp, IFF_UP, AF_UNSPEC);
1656 }
1657
1658 /*
1659  * Process a link state change.
1660  * NOTE: must be called at splsoftnet or equivalent.
1661  */
1662 void
1663 if_link_state_change(struct ifnet *ifp)
1664 {
1665         int link_state = ifp->if_link_state;
1666
1667         rt_ifmsg(ifp);
1668         devctl_notify("IFNET", ifp->if_xname,
1669             (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL);
1670 }
1671
1672 /*
1673  * Handle interface watchdog timer routines.  Called
1674  * from softclock, we decrement timers (if set) and
1675  * call the appropriate interface routine on expiration.
1676  */
1677 static void
1678 if_slowtimo_dispatch(netmsg_t nmsg)
1679 {
1680         struct globaldata *gd = mycpu;
1681         const struct ifnet_array *arr;
1682         int i;
1683
1684         ASSERT_IN_NETISR(0);
1685
1686         crit_enter_gd(gd);
1687         lwkt_replymsg(&nmsg->lmsg, 0);  /* reply ASAP */
1688         crit_exit_gd(gd);
1689
1690         arr = ifnet_array_get();
1691         for (i = 0; i < arr->ifnet_count; ++i) {
1692                 struct ifnet *ifp = arr->ifnet_arr[i];
1693
1694                 crit_enter_gd(gd);
1695
1696                 if (if_stats_compat) {
1697                         IFNET_STAT_GET(ifp, ipackets, ifp->if_ipackets);
1698                         IFNET_STAT_GET(ifp, ierrors, ifp->if_ierrors);
1699                         IFNET_STAT_GET(ifp, opackets, ifp->if_opackets);
1700                         IFNET_STAT_GET(ifp, oerrors, ifp->if_oerrors);
1701                         IFNET_STAT_GET(ifp, collisions, ifp->if_collisions);
1702                         IFNET_STAT_GET(ifp, ibytes, ifp->if_ibytes);
1703                         IFNET_STAT_GET(ifp, obytes, ifp->if_obytes);
1704                         IFNET_STAT_GET(ifp, imcasts, ifp->if_imcasts);
1705                         IFNET_STAT_GET(ifp, omcasts, ifp->if_omcasts);
1706                         IFNET_STAT_GET(ifp, iqdrops, ifp->if_iqdrops);
1707                         IFNET_STAT_GET(ifp, noproto, ifp->if_noproto);
1708                         IFNET_STAT_GET(ifp, oqdrops, ifp->if_oqdrops);
1709                 }
1710
1711                 if (ifp->if_timer == 0 || --ifp->if_timer) {
1712                         crit_exit_gd(gd);
1713                         continue;
1714                 }
1715                 if (ifp->if_watchdog) {
1716                         if (ifnet_tryserialize_all(ifp)) {
1717                                 (*ifp->if_watchdog)(ifp);
1718                                 ifnet_deserialize_all(ifp);
1719                         } else {
1720                                 /* try again next timeout */
1721                                 ++ifp->if_timer;
1722                         }
1723                 }
1724
1725                 crit_exit_gd(gd);
1726         }
1727
1728         callout_reset(&if_slowtimo_timer, hz / IFNET_SLOWHZ, if_slowtimo, NULL);
1729 }
1730
1731 static void
1732 if_slowtimo(void *arg __unused)
1733 {
1734         struct lwkt_msg *lmsg = &if_slowtimo_netmsg.lmsg;
1735
1736         KASSERT(mycpuid == 0, ("not on cpu0"));
1737         crit_enter();
1738         if (lmsg->ms_flags & MSGF_DONE)
1739                 lwkt_sendmsg_oncpu(netisr_cpuport(0), lmsg);
1740         crit_exit();
1741 }
1742
1743 /*
1744  * Map interface name to
1745  * interface structure pointer.
1746  */
1747 struct ifnet *
1748 ifunit(const char *name)
1749 {
1750         struct ifnet *ifp;
1751
1752         /*
1753          * Search all the interfaces for this name/number
1754          */
1755         KASSERT(mtx_owned(&ifnet_mtx), ("ifnet is not locked"));
1756
1757         TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
1758                 if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1759                         break;
1760         }
1761         return (ifp);
1762 }
1763
1764 struct ifnet *
1765 ifunit_netisr(const char *name)
1766 {
1767         const struct ifnet_array *arr;
1768         int i;
1769
1770         /*
1771          * Search all the interfaces for this name/number
1772          */
1773
1774         arr = ifnet_array_get();
1775         for (i = 0; i < arr->ifnet_count; ++i) {
1776                 struct ifnet *ifp = arr->ifnet_arr[i];
1777
1778                 if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1779                         return ifp;
1780         }
1781         return NULL;
1782 }
1783
1784 /*
1785  * Interface ioctls.
1786  */
1787 int
1788 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct ucred *cred)
1789 {
1790         struct ifnet *ifp;
1791         struct ifreq *ifr;
1792         struct ifstat *ifs;
1793         int error, do_ifup = 0;
1794         short oif_flags;
1795         int new_flags;
1796         size_t namelen, onamelen;
1797         char new_name[IFNAMSIZ];
1798         struct ifaddr *ifa;
1799         struct sockaddr_dl *sdl;
1800
1801         switch (cmd) {
1802         case SIOCGIFCONF:
1803         case OSIOCGIFCONF:
1804                 return (ifconf(cmd, data, cred));
1805         default:
1806                 break;
1807         }
1808
1809         ifr = (struct ifreq *)data;
1810
1811         switch (cmd) {
1812         case SIOCIFCREATE:
1813         case SIOCIFCREATE2:
1814                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1815                         return (error);
1816                 return (if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name),
1817                         cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL));
1818         case SIOCIFDESTROY:
1819                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1820                         return (error);
1821                 return (if_clone_destroy(ifr->ifr_name));
1822         case SIOCIFGCLONERS:
1823                 return (if_clone_list((struct if_clonereq *)data));
1824         default:
1825                 break;
1826         }
1827
1828         /*
1829          * Nominal ioctl through interface, lookup the ifp and obtain a
1830          * lock to serialize the ifconfig ioctl operation.
1831          */
1832         ifnet_lock();
1833
1834         ifp = ifunit(ifr->ifr_name);
1835         if (ifp == NULL) {
1836                 ifnet_unlock();
1837                 return (ENXIO);
1838         }
1839         error = 0;
1840
1841         switch (cmd) {
1842         case SIOCGIFINDEX:
1843                 ifr->ifr_index = ifp->if_index;
1844                 break;
1845
1846         case SIOCGIFFLAGS:
1847                 ifr->ifr_flags = ifp->if_flags;
1848                 ifr->ifr_flagshigh = ifp->if_flags >> 16;
1849                 break;
1850
1851         case SIOCGIFCAP:
1852                 ifr->ifr_reqcap = ifp->if_capabilities;
1853                 ifr->ifr_curcap = ifp->if_capenable;
1854                 break;
1855
1856         case SIOCGIFMETRIC:
1857                 ifr->ifr_metric = ifp->if_metric;
1858                 break;
1859
1860         case SIOCGIFMTU:
1861                 ifr->ifr_mtu = ifp->if_mtu;
1862                 break;
1863
1864         case SIOCGIFTSOLEN:
1865                 ifr->ifr_tsolen = ifp->if_tsolen;
1866                 break;
1867
1868         case SIOCGIFDATA:
1869                 error = copyout((caddr_t)&ifp->if_data, ifr->ifr_data,
1870                                 sizeof(ifp->if_data));
1871                 break;
1872
1873         case SIOCGIFPHYS:
1874                 ifr->ifr_phys = ifp->if_physical;
1875                 break;
1876
1877         case SIOCGIFPOLLCPU:
1878                 ifr->ifr_pollcpu = -1;
1879                 break;
1880
1881         case SIOCSIFPOLLCPU:
1882                 break;
1883
1884         case SIOCSIFFLAGS:
1885                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1886                 if (error)
1887                         break;
1888                 new_flags = (ifr->ifr_flags & 0xffff) |
1889                     (ifr->ifr_flagshigh << 16);
1890                 if (ifp->if_flags & IFF_SMART) {
1891                         /* Smart drivers twiddle their own routes */
1892                 } else if (ifp->if_flags & IFF_UP &&
1893                     (new_flags & IFF_UP) == 0) {
1894                         if_down(ifp);
1895                 } else if (new_flags & IFF_UP &&
1896                     (ifp->if_flags & IFF_UP) == 0) {
1897                         do_ifup = 1;
1898                 }
1899
1900 #ifdef IFPOLL_ENABLE
1901                 if ((new_flags ^ ifp->if_flags) & IFF_NPOLLING) {
1902                         if (new_flags & IFF_NPOLLING)
1903                                 ifpoll_register(ifp);
1904                         else
1905                                 ifpoll_deregister(ifp);
1906                 }
1907 #endif
1908
1909                 ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
1910                         (new_flags &~ IFF_CANTCHANGE);
1911                 if (new_flags & IFF_PPROMISC) {
1912                         /* Permanently promiscuous mode requested */
1913                         ifp->if_flags |= IFF_PROMISC;
1914                 } else if (ifp->if_pcount == 0) {
1915                         ifp->if_flags &= ~IFF_PROMISC;
1916                 }
1917                 if (ifp->if_ioctl) {
1918                         ifnet_serialize_all(ifp);
1919                         ifp->if_ioctl(ifp, cmd, data, cred);
1920                         ifnet_deserialize_all(ifp);
1921                 }
1922                 if (do_ifup)
1923                         if_up(ifp);
1924                 getmicrotime(&ifp->if_lastchange);
1925                 break;
1926
1927         case SIOCSIFCAP:
1928                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1929                 if (error)
1930                         break;
1931                 if (ifr->ifr_reqcap & ~ifp->if_capabilities) {
1932                         error = EINVAL;
1933                         break;
1934                 }
1935                 ifnet_serialize_all(ifp);
1936                 ifp->if_ioctl(ifp, cmd, data, cred);
1937                 ifnet_deserialize_all(ifp);
1938                 break;
1939
1940         case SIOCSIFNAME:
1941                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1942                 if (error)
1943                         break;
1944                 error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
1945                 if (error)
1946                         break;
1947                 if (new_name[0] == '\0') {
1948                         error = EINVAL;
1949                         break;
1950                 }
1951                 if (ifunit(new_name) != NULL) {
1952                         error = EEXIST;
1953                         break;
1954                 }
1955
1956                 EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
1957
1958                 /* Announce the departure of the interface. */
1959                 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1960
1961                 strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
1962                 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
1963                 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1964                 namelen = strlen(new_name);
1965                 onamelen = sdl->sdl_nlen;
1966                 /*
1967                  * Move the address if needed.  This is safe because we
1968                  * allocate space for a name of length IFNAMSIZ when we
1969                  * create this in if_attach().
1970                  */
1971                 if (namelen != onamelen) {
1972                         bcopy(sdl->sdl_data + onamelen,
1973                             sdl->sdl_data + namelen, sdl->sdl_alen);
1974                 }
1975                 bcopy(new_name, sdl->sdl_data, namelen);
1976                 sdl->sdl_nlen = namelen;
1977                 sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
1978                 bzero(sdl->sdl_data, onamelen);
1979                 while (namelen != 0)
1980                         sdl->sdl_data[--namelen] = 0xff;
1981
1982                 EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
1983
1984                 /* Announce the return of the interface. */
1985                 rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
1986                 break;
1987
1988         case SIOCSIFMETRIC:
1989                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1990                 if (error)
1991                         break;
1992                 ifp->if_metric = ifr->ifr_metric;
1993                 getmicrotime(&ifp->if_lastchange);
1994                 break;
1995
1996         case SIOCSIFPHYS:
1997                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1998                 if (error)
1999                         break;
2000                 if (ifp->if_ioctl == NULL) {
2001                         error = EOPNOTSUPP;
2002                         break;
2003                 }
2004                 ifnet_serialize_all(ifp);
2005                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2006                 ifnet_deserialize_all(ifp);
2007                 if (error == 0)
2008                         getmicrotime(&ifp->if_lastchange);
2009                 break;
2010
2011         case SIOCSIFMTU:
2012         {
2013                 u_long oldmtu = ifp->if_mtu;
2014
2015                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2016                 if (error)
2017                         break;
2018                 if (ifp->if_ioctl == NULL) {
2019                         error = EOPNOTSUPP;
2020                         break;
2021                 }
2022                 if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU) {
2023                         error = EINVAL;
2024                         break;
2025                 }
2026                 ifnet_serialize_all(ifp);
2027                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2028                 ifnet_deserialize_all(ifp);
2029                 if (error == 0) {
2030                         getmicrotime(&ifp->if_lastchange);
2031                         rt_ifmsg(ifp);
2032                 }
2033                 /*
2034                  * If the link MTU changed, do network layer specific procedure.
2035                  */
2036                 if (ifp->if_mtu != oldmtu) {
2037 #ifdef INET6
2038                         nd6_setmtu(ifp);
2039 #endif
2040                 }
2041                 break;
2042         }
2043
2044         case SIOCSIFTSOLEN:
2045                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2046                 if (error)
2047                         break;
2048
2049                 /* XXX need driver supplied upper limit */
2050                 if (ifr->ifr_tsolen <= 0) {
2051                         error = EINVAL;
2052                         break;
2053                 }
2054                 ifp->if_tsolen = ifr->ifr_tsolen;
2055                 break;
2056
2057         case SIOCADDMULTI:
2058         case SIOCDELMULTI:
2059                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2060                 if (error)
2061                         break;
2062
2063                 /* Don't allow group membership on non-multicast interfaces. */
2064                 if ((ifp->if_flags & IFF_MULTICAST) == 0) {
2065                         error = EOPNOTSUPP;
2066                         break;
2067                 }
2068
2069                 /* Don't let users screw up protocols' entries. */
2070                 if (ifr->ifr_addr.sa_family != AF_LINK) {
2071                         error = EINVAL;
2072                         break;
2073                 }
2074
2075                 if (cmd == SIOCADDMULTI) {
2076                         struct ifmultiaddr *ifma;
2077                         error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
2078                 } else {
2079                         error = if_delmulti(ifp, &ifr->ifr_addr);
2080                 }
2081                 if (error == 0)
2082                         getmicrotime(&ifp->if_lastchange);
2083                 break;
2084
2085         case SIOCSIFPHYADDR:
2086         case SIOCDIFPHYADDR:
2087 #ifdef INET6
2088         case SIOCSIFPHYADDR_IN6:
2089 #endif
2090         case SIOCSLIFPHYADDR:
2091         case SIOCSIFMEDIA:
2092         case SIOCSIFGENERIC:
2093                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2094                 if (error)
2095                         break;
2096                 if (ifp->if_ioctl == 0) {
2097                         error = EOPNOTSUPP;
2098                         break;
2099                 }
2100                 ifnet_serialize_all(ifp);
2101                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2102                 ifnet_deserialize_all(ifp);
2103                 if (error == 0)
2104                         getmicrotime(&ifp->if_lastchange);
2105                 break;
2106
2107         case SIOCGIFSTATUS:
2108                 ifs = (struct ifstat *)data;
2109                 ifs->ascii[0] = '\0';
2110                 /* fall through */
2111         case SIOCGIFPSRCADDR:
2112         case SIOCGIFPDSTADDR:
2113         case SIOCGLIFPHYADDR:
2114         case SIOCGIFMEDIA:
2115         case SIOCGIFGENERIC:
2116                 if (ifp->if_ioctl == NULL) {
2117                         error = EOPNOTSUPP;
2118                         break;
2119                 }
2120                 ifnet_serialize_all(ifp);
2121                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2122                 ifnet_deserialize_all(ifp);
2123                 break;
2124
2125         case SIOCSIFLLADDR:
2126                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2127                 if (error)
2128                         break;
2129                 error = if_setlladdr(ifp, ifr->ifr_addr.sa_data,
2130                                      ifr->ifr_addr.sa_len);
2131                 EVENTHANDLER_INVOKE(iflladdr_event, ifp);
2132                 break;
2133
2134         default:
2135                 oif_flags = ifp->if_flags;
2136                 if (so->so_proto == 0) {
2137                         error = EOPNOTSUPP;
2138                         break;
2139                 }
2140                 error = so_pru_control_direct(so, cmd, data, ifp);
2141
2142                 if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
2143 #ifdef INET6
2144                         DELAY(100);/* XXX: temporary workaround for fxp issue*/
2145                         if (ifp->if_flags & IFF_UP) {
2146                                 crit_enter();
2147                                 in6_if_up(ifp);
2148                                 crit_exit();
2149                         }
2150 #endif
2151                 }
2152                 break;
2153         }
2154
2155         ifnet_unlock();
2156         return (error);
2157 }
2158
2159 /*
2160  * Set/clear promiscuous mode on interface ifp based on the truth value
2161  * of pswitch.  The calls are reference counted so that only the first
2162  * "on" request actually has an effect, as does the final "off" request.
2163  * Results are undefined if the "off" and "on" requests are not matched.
2164  */
2165 int
2166 ifpromisc(struct ifnet *ifp, int pswitch)
2167 {
2168         struct ifreq ifr;
2169         int error;
2170         int oldflags;
2171
2172         oldflags = ifp->if_flags;
2173         if (ifp->if_flags & IFF_PPROMISC) {
2174                 /* Do nothing if device is in permanently promiscuous mode */
2175                 ifp->if_pcount += pswitch ? 1 : -1;
2176                 return (0);
2177         }
2178         if (pswitch) {
2179                 /*
2180                  * If the device is not configured up, we cannot put it in
2181                  * promiscuous mode.
2182                  */
2183                 if ((ifp->if_flags & IFF_UP) == 0)
2184                         return (ENETDOWN);
2185                 if (ifp->if_pcount++ != 0)
2186                         return (0);
2187                 ifp->if_flags |= IFF_PROMISC;
2188                 log(LOG_INFO, "%s: promiscuous mode enabled\n",
2189                     ifp->if_xname);
2190         } else {
2191                 if (--ifp->if_pcount > 0)
2192                         return (0);
2193                 ifp->if_flags &= ~IFF_PROMISC;
2194                 log(LOG_INFO, "%s: promiscuous mode disabled\n",
2195                     ifp->if_xname);
2196         }
2197         ifr.ifr_flags = ifp->if_flags;
2198         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2199         ifnet_serialize_all(ifp);
2200         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr, NULL);
2201         ifnet_deserialize_all(ifp);
2202         if (error == 0)
2203                 rt_ifmsg(ifp);
2204         else
2205                 ifp->if_flags = oldflags;
2206         return error;
2207 }
2208
2209 /*
2210  * Return interface configuration
2211  * of system.  List may be used
2212  * in later ioctl's (above) to get
2213  * other information.
2214  */
2215 static int
2216 ifconf(u_long cmd, caddr_t data, struct ucred *cred)
2217 {
2218         struct ifconf *ifc = (struct ifconf *)data;
2219         struct ifnet *ifp;
2220         struct sockaddr *sa;
2221         struct ifreq ifr, *ifrp;
2222         int space = ifc->ifc_len, error = 0;
2223
2224         ifrp = ifc->ifc_req;
2225
2226         ifnet_lock();
2227         TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
2228                 struct ifaddr_container *ifac, *ifac_mark;
2229                 struct ifaddr_marker mark;
2230                 struct ifaddrhead *head;
2231                 int addrs;
2232
2233                 if (space <= sizeof ifr)
2234                         break;
2235
2236                 /*
2237                  * Zero the stack declared structure first to prevent
2238                  * memory disclosure.
2239                  */
2240                 bzero(&ifr, sizeof(ifr));
2241                 if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
2242                     >= sizeof(ifr.ifr_name)) {
2243                         error = ENAMETOOLONG;
2244                         break;
2245                 }
2246
2247                 /*
2248                  * Add a marker, since copyout() could block and during that
2249                  * period the list could be changed.  Inserting the marker to
2250                  * the header of the list will not cause trouble for the code
2251                  * assuming that the first element of the list is AF_LINK; the
2252                  * marker will be moved to the next position w/o blocking.
2253                  */
2254                 ifa_marker_init(&mark, ifp);
2255                 ifac_mark = &mark.ifac;
2256                 head = &ifp->if_addrheads[mycpuid];
2257
2258                 addrs = 0;
2259                 TAILQ_INSERT_HEAD(head, ifac_mark, ifa_link);
2260                 while ((ifac = TAILQ_NEXT(ifac_mark, ifa_link)) != NULL) {
2261                         struct ifaddr *ifa = ifac->ifa;
2262
2263                         TAILQ_REMOVE(head, ifac_mark, ifa_link);
2264                         TAILQ_INSERT_AFTER(head, ifac, ifac_mark, ifa_link);
2265
2266                         /* Ignore marker */
2267                         if (ifa->ifa_addr->sa_family == AF_UNSPEC)
2268                                 continue;
2269
2270                         if (space <= sizeof ifr)
2271                                 break;
2272                         sa = ifa->ifa_addr;
2273                         if (cred->cr_prison &&
2274                             prison_if(cred, sa))
2275                                 continue;
2276                         addrs++;
2277                         /*
2278                          * Keep a reference on this ifaddr, so that it will
2279                          * not be destroyed when its address is copied to
2280                          * the userland, which could block.
2281                          */
2282                         IFAREF(ifa);
2283                         if (sa->sa_len <= sizeof(*sa)) {
2284                                 ifr.ifr_addr = *sa;
2285                                 error = copyout(&ifr, ifrp, sizeof ifr);
2286                                 ifrp++;
2287                         } else {
2288                                 if (space < (sizeof ifr) + sa->sa_len -
2289                                             sizeof(*sa)) {
2290                                         IFAFREE(ifa);
2291                                         break;
2292                                 }
2293                                 space -= sa->sa_len - sizeof(*sa);
2294                                 error = copyout(&ifr, ifrp,
2295                                                 sizeof ifr.ifr_name);
2296                                 if (error == 0)
2297                                         error = copyout(sa, &ifrp->ifr_addr,
2298                                                         sa->sa_len);
2299                                 ifrp = (struct ifreq *)
2300                                         (sa->sa_len + (caddr_t)&ifrp->ifr_addr);
2301                         }
2302                         IFAFREE(ifa);
2303                         if (error)
2304                                 break;
2305                         space -= sizeof ifr;
2306                 }
2307                 TAILQ_REMOVE(head, ifac_mark, ifa_link);
2308                 if (error)
2309                         break;
2310                 if (!addrs) {
2311                         bzero(&ifr.ifr_addr, sizeof ifr.ifr_addr);
2312                         error = copyout(&ifr, ifrp, sizeof ifr);
2313                         if (error)
2314                                 break;
2315                         space -= sizeof ifr;
2316                         ifrp++;
2317                 }
2318         }
2319         ifnet_unlock();
2320
2321         ifc->ifc_len -= space;
2322         return (error);
2323 }
2324
2325 /*
2326  * Just like if_promisc(), but for all-multicast-reception mode.
2327  */
2328 int
2329 if_allmulti(struct ifnet *ifp, int onswitch)
2330 {
2331         int error = 0;
2332         struct ifreq ifr;
2333
2334         crit_enter();
2335
2336         if (onswitch) {
2337                 if (ifp->if_amcount++ == 0) {
2338                         ifp->if_flags |= IFF_ALLMULTI;
2339                         ifr.ifr_flags = ifp->if_flags;
2340                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2341                         ifnet_serialize_all(ifp);
2342                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2343                                               NULL);
2344                         ifnet_deserialize_all(ifp);
2345                 }
2346         } else {
2347                 if (ifp->if_amcount > 1) {
2348                         ifp->if_amcount--;
2349                 } else {
2350                         ifp->if_amcount = 0;
2351                         ifp->if_flags &= ~IFF_ALLMULTI;
2352                         ifr.ifr_flags = ifp->if_flags;
2353                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2354                         ifnet_serialize_all(ifp);
2355                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2356                                               NULL);
2357                         ifnet_deserialize_all(ifp);
2358                 }
2359         }
2360
2361         crit_exit();
2362
2363         if (error == 0)
2364                 rt_ifmsg(ifp);
2365         return error;
2366 }
2367
2368 /*
2369  * Add a multicast listenership to the interface in question.
2370  * The link layer provides a routine which converts
2371  */
2372 int
2373 if_addmulti_serialized(struct ifnet *ifp, struct sockaddr *sa,
2374     struct ifmultiaddr **retifma)
2375 {
2376         struct sockaddr *llsa, *dupsa;
2377         int error;
2378         struct ifmultiaddr *ifma;
2379
2380         ASSERT_IFNET_SERIALIZED_ALL(ifp);
2381
2382         /*
2383          * If the matching multicast address already exists
2384          * then don't add a new one, just add a reference
2385          */
2386         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2387                 if (sa_equal(sa, ifma->ifma_addr)) {
2388                         ifma->ifma_refcount++;
2389                         if (retifma)
2390                                 *retifma = ifma;
2391                         return 0;
2392                 }
2393         }
2394
2395         /*
2396          * Give the link layer a chance to accept/reject it, and also
2397          * find out which AF_LINK address this maps to, if it isn't one
2398          * already.
2399          */
2400         if (ifp->if_resolvemulti) {
2401                 error = ifp->if_resolvemulti(ifp, &llsa, sa);
2402                 if (error)
2403                         return error;
2404         } else {
2405                 llsa = NULL;
2406         }
2407
2408         ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_INTWAIT);
2409         dupsa = kmalloc(sa->sa_len, M_IFMADDR, M_INTWAIT);
2410         bcopy(sa, dupsa, sa->sa_len);
2411
2412         ifma->ifma_addr = dupsa;
2413         ifma->ifma_lladdr = llsa;
2414         ifma->ifma_ifp = ifp;
2415         ifma->ifma_refcount = 1;
2416         ifma->ifma_protospec = NULL;
2417         rt_newmaddrmsg(RTM_NEWMADDR, ifma);
2418
2419         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2420         if (retifma)
2421                 *retifma = ifma;
2422
2423         if (llsa != NULL) {
2424                 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2425                         if (sa_equal(ifma->ifma_addr, llsa))
2426                                 break;
2427                 }
2428                 if (ifma) {
2429                         ifma->ifma_refcount++;
2430                 } else {
2431                         ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_INTWAIT);
2432                         dupsa = kmalloc(llsa->sa_len, M_IFMADDR, M_INTWAIT);
2433                         bcopy(llsa, dupsa, llsa->sa_len);
2434                         ifma->ifma_addr = dupsa;
2435                         ifma->ifma_ifp = ifp;
2436                         ifma->ifma_refcount = 1;
2437                         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2438                 }
2439         }
2440         /*
2441          * We are certain we have added something, so call down to the
2442          * interface to let them know about it.
2443          */
2444         if (ifp->if_ioctl)
2445                 ifp->if_ioctl(ifp, SIOCADDMULTI, 0, NULL);
2446
2447         return 0;
2448 }
2449
2450 int
2451 if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
2452     struct ifmultiaddr **retifma)
2453 {
2454         int error;
2455
2456         ifnet_serialize_all(ifp);
2457         error = if_addmulti_serialized(ifp, sa, retifma);
2458         ifnet_deserialize_all(ifp);
2459
2460         return error;
2461 }
2462
2463 /*
2464  * Remove a reference to a multicast address on this interface.  Yell
2465  * if the request does not match an existing membership.
2466  */
2467 static int
2468 if_delmulti_serialized(struct ifnet *ifp, struct sockaddr *sa)
2469 {
2470         struct ifmultiaddr *ifma;
2471
2472         ASSERT_IFNET_SERIALIZED_ALL(ifp);
2473
2474         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2475                 if (sa_equal(sa, ifma->ifma_addr))
2476                         break;
2477         if (ifma == NULL)
2478                 return ENOENT;
2479
2480         if (ifma->ifma_refcount > 1) {
2481                 ifma->ifma_refcount--;
2482                 return 0;
2483         }
2484
2485         rt_newmaddrmsg(RTM_DELMADDR, ifma);
2486         sa = ifma->ifma_lladdr;
2487         TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2488         /*
2489          * Make sure the interface driver is notified
2490          * in the case of a link layer mcast group being left.
2491          */
2492         if (ifma->ifma_addr->sa_family == AF_LINK && sa == NULL)
2493                 ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2494         kfree(ifma->ifma_addr, M_IFMADDR);
2495         kfree(ifma, M_IFMADDR);
2496         if (sa == NULL)
2497                 return 0;
2498
2499         /*
2500          * Now look for the link-layer address which corresponds to
2501          * this network address.  It had been squirreled away in
2502          * ifma->ifma_lladdr for this purpose (so we don't have
2503          * to call ifp->if_resolvemulti() again), and we saved that
2504          * value in sa above.  If some nasty deleted the
2505          * link-layer address out from underneath us, we can deal because
2506          * the address we stored was is not the same as the one which was
2507          * in the record for the link-layer address.  (So we don't complain
2508          * in that case.)
2509          */
2510         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2511                 if (sa_equal(sa, ifma->ifma_addr))
2512                         break;
2513         if (ifma == NULL)
2514                 return 0;
2515
2516         if (ifma->ifma_refcount > 1) {
2517                 ifma->ifma_refcount--;
2518                 return 0;
2519         }
2520
2521         TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2522         ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2523         kfree(ifma->ifma_addr, M_IFMADDR);
2524         kfree(sa, M_IFMADDR);
2525         kfree(ifma, M_IFMADDR);
2526
2527         return 0;
2528 }
2529
2530 int
2531 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
2532 {
2533         int error;
2534
2535         ifnet_serialize_all(ifp);
2536         error = if_delmulti_serialized(ifp, sa);
2537         ifnet_deserialize_all(ifp);
2538
2539         return error;
2540 }
2541
2542 /*
2543  * Delete all multicast group membership for an interface.
2544  * Should be used to quickly flush all multicast filters.
2545  */
2546 void
2547 if_delallmulti_serialized(struct ifnet *ifp)
2548 {
2549         struct ifmultiaddr *ifma, mark;
2550         struct sockaddr sa;
2551
2552         ASSERT_IFNET_SERIALIZED_ALL(ifp);
2553
2554         bzero(&sa, sizeof(sa));
2555         sa.sa_family = AF_UNSPEC;
2556         sa.sa_len = sizeof(sa);
2557
2558         bzero(&mark, sizeof(mark));
2559         mark.ifma_addr = &sa;
2560
2561         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, &mark, ifma_link);
2562         while ((ifma = TAILQ_NEXT(&mark, ifma_link)) != NULL) {
2563                 TAILQ_REMOVE(&ifp->if_multiaddrs, &mark, ifma_link);
2564                 TAILQ_INSERT_AFTER(&ifp->if_multiaddrs, ifma, &mark,
2565                     ifma_link);
2566
2567                 if (ifma->ifma_addr->sa_family == AF_UNSPEC)
2568                         continue;
2569
2570                 if_delmulti_serialized(ifp, ifma->ifma_addr);
2571         }
2572         TAILQ_REMOVE(&ifp->if_multiaddrs, &mark, ifma_link);
2573 }
2574
2575
2576 /*
2577  * Set the link layer address on an interface.
2578  *
2579  * At this time we only support certain types of interfaces,
2580  * and we don't allow the length of the address to change.
2581  */
2582 int
2583 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
2584 {
2585         struct sockaddr_dl *sdl;
2586         struct ifreq ifr;
2587
2588         sdl = IF_LLSOCKADDR(ifp);
2589         if (sdl == NULL)
2590                 return (EINVAL);
2591         if (len != sdl->sdl_alen)       /* don't allow length to change */
2592                 return (EINVAL);
2593         switch (ifp->if_type) {
2594         case IFT_ETHER:                 /* these types use struct arpcom */
2595         case IFT_XETHER:
2596         case IFT_L2VLAN:
2597         case IFT_IEEE8023ADLAG:
2598                 bcopy(lladdr, ((struct arpcom *)ifp->if_softc)->ac_enaddr, len);
2599                 bcopy(lladdr, LLADDR(sdl), len);
2600                 break;
2601         default:
2602                 return (ENODEV);
2603         }
2604         /*
2605          * If the interface is already up, we need
2606          * to re-init it in order to reprogram its
2607          * address filter.
2608          */
2609         ifnet_serialize_all(ifp);
2610         if ((ifp->if_flags & IFF_UP) != 0) {
2611 #ifdef INET
2612                 struct ifaddr_container *ifac;
2613 #endif
2614
2615                 ifp->if_flags &= ~IFF_UP;
2616                 ifr.ifr_flags = ifp->if_flags;
2617                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2618                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2619                               NULL);
2620                 ifp->if_flags |= IFF_UP;
2621                 ifr.ifr_flags = ifp->if_flags;
2622                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2623                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2624                                  NULL);
2625 #ifdef INET
2626                 /*
2627                  * Also send gratuitous ARPs to notify other nodes about
2628                  * the address change.
2629                  */
2630                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2631                         struct ifaddr *ifa = ifac->ifa;
2632
2633                         if (ifa->ifa_addr != NULL &&
2634                             ifa->ifa_addr->sa_family == AF_INET)
2635                                 arp_gratuitous(ifp, ifa);
2636                 }
2637 #endif
2638         }
2639         ifnet_deserialize_all(ifp);
2640         return (0);
2641 }
2642
2643 struct ifmultiaddr *
2644 ifmaof_ifpforaddr(struct sockaddr *sa, struct ifnet *ifp)
2645 {
2646         struct ifmultiaddr *ifma;
2647
2648         /* TODO: need ifnet_serialize_main */
2649         ifnet_serialize_all(ifp);
2650         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2651                 if (sa_equal(ifma->ifma_addr, sa))
2652                         break;
2653         ifnet_deserialize_all(ifp);
2654
2655         return ifma;
2656 }
2657
2658 /*
2659  * This function locates the first real ethernet MAC from a network
2660  * card and loads it into node, returning 0 on success or ENOENT if
2661  * no suitable interfaces were found.  It is used by the uuid code to
2662  * generate a unique 6-byte number.
2663  */
2664 int
2665 if_getanyethermac(uint16_t *node, int minlen)
2666 {
2667         struct ifnet *ifp;
2668         struct sockaddr_dl *sdl;
2669
2670         ifnet_lock();
2671         TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
2672                 if (ifp->if_type != IFT_ETHER)
2673                         continue;
2674                 sdl = IF_LLSOCKADDR(ifp);
2675                 if (sdl->sdl_alen < minlen)
2676                         continue;
2677                 bcopy(((struct arpcom *)ifp->if_softc)->ac_enaddr, node,
2678                       minlen);
2679                 ifnet_unlock();
2680                 return(0);
2681         }
2682         ifnet_unlock();
2683         return (ENOENT);
2684 }
2685
2686 /*
2687  * The name argument must be a pointer to storage which will last as
2688  * long as the interface does.  For physical devices, the result of
2689  * device_get_name(dev) is a good choice and for pseudo-devices a
2690  * static string works well.
2691  */
2692 void
2693 if_initname(struct ifnet *ifp, const char *name, int unit)
2694 {
2695         ifp->if_dname = name;
2696         ifp->if_dunit = unit;
2697         if (unit != IF_DUNIT_NONE)
2698                 ksnprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
2699         else
2700                 strlcpy(ifp->if_xname, name, IFNAMSIZ);
2701 }
2702
2703 int
2704 if_printf(struct ifnet *ifp, const char *fmt, ...)
2705 {
2706         __va_list ap;
2707         int retval;
2708
2709         retval = kprintf("%s: ", ifp->if_xname);
2710         __va_start(ap, fmt);
2711         retval += kvprintf(fmt, ap);
2712         __va_end(ap);
2713         return (retval);
2714 }
2715
2716 struct ifnet *
2717 if_alloc(uint8_t type)
2718 {
2719         struct ifnet *ifp;
2720         size_t size;
2721
2722         /*
2723          * XXX temporary hack until arpcom is setup in if_l2com
2724          */
2725         if (type == IFT_ETHER)
2726                 size = sizeof(struct arpcom);
2727         else
2728                 size = sizeof(struct ifnet);
2729
2730         ifp = kmalloc(size, M_IFNET, M_WAITOK|M_ZERO);
2731
2732         ifp->if_type = type;
2733
2734         if (if_com_alloc[type] != NULL) {
2735                 ifp->if_l2com = if_com_alloc[type](type, ifp);
2736                 if (ifp->if_l2com == NULL) {
2737                         kfree(ifp, M_IFNET);
2738                         return (NULL);
2739                 }
2740         }
2741         return (ifp);
2742 }
2743
2744 void
2745 if_free(struct ifnet *ifp)
2746 {
2747         kfree(ifp, M_IFNET);
2748 }
2749
2750 void
2751 ifq_set_classic(struct ifaltq *ifq)
2752 {
2753         ifq_set_methods(ifq, ifq->altq_ifp->if_mapsubq,
2754             ifsq_classic_enqueue, ifsq_classic_dequeue, ifsq_classic_request);
2755 }
2756
2757 void
2758 ifq_set_methods(struct ifaltq *ifq, altq_mapsubq_t mapsubq,
2759     ifsq_enqueue_t enqueue, ifsq_dequeue_t dequeue, ifsq_request_t request)
2760 {
2761         int q;
2762
2763         KASSERT(mapsubq != NULL, ("mapsubq is not specified"));
2764         KASSERT(enqueue != NULL, ("enqueue is not specified"));
2765         KASSERT(dequeue != NULL, ("dequeue is not specified"));
2766         KASSERT(request != NULL, ("request is not specified"));
2767
2768         ifq->altq_mapsubq = mapsubq;
2769         for (q = 0; q < ifq->altq_subq_cnt; ++q) {
2770                 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
2771
2772                 ifsq->ifsq_enqueue = enqueue;
2773                 ifsq->ifsq_dequeue = dequeue;
2774                 ifsq->ifsq_request = request;
2775         }
2776 }
2777
2778 static void
2779 ifsq_norm_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
2780 {
2781
2782         classq_add(&ifsq->ifsq_norm, m);
2783         ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
2784 }
2785
2786 static void
2787 ifsq_prio_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
2788 {
2789
2790         classq_add(&ifsq->ifsq_prio, m);
2791         ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
2792         ALTQ_SQ_PRIO_CNTR_INC(ifsq, m->m_pkthdr.len);
2793 }
2794
2795 static struct mbuf *
2796 ifsq_norm_dequeue(struct ifaltq_subque *ifsq)
2797 {
2798         struct mbuf *m;
2799
2800         m = classq_get(&ifsq->ifsq_norm);
2801         if (m != NULL)
2802                 ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
2803         return (m);
2804 }
2805
2806 static struct mbuf *
2807 ifsq_prio_dequeue(struct ifaltq_subque *ifsq)
2808 {
2809         struct mbuf *m;
2810
2811         m = classq_get(&ifsq->ifsq_prio);
2812         if (m != NULL) {
2813                 ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
2814                 ALTQ_SQ_PRIO_CNTR_DEC(ifsq, m->m_pkthdr.len);
2815         }
2816         return (m);
2817 }
2818
2819 int
2820 ifsq_classic_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m,
2821     struct altq_pktattr *pa __unused)
2822 {
2823
2824         M_ASSERTPKTHDR(m);
2825 again:
2826         if (ifsq->ifsq_len >= ifsq->ifsq_maxlen ||
2827             ifsq->ifsq_bcnt >= ifsq->ifsq_maxbcnt) {
2828                 struct mbuf *m_drop;
2829
2830                 if (m->m_flags & M_PRIO) {
2831                         m_drop = NULL;
2832                         if (ifsq->ifsq_prio_len < (ifsq->ifsq_maxlen >> 1) &&
2833                             ifsq->ifsq_prio_bcnt < (ifsq->ifsq_maxbcnt >> 1)) {
2834                                 /* Try dropping some from normal queue. */
2835                                 m_drop = ifsq_norm_dequeue(ifsq);
2836                         }
2837                         if (m_drop == NULL)
2838                                 m_drop = ifsq_prio_dequeue(ifsq);
2839                 } else {
2840                         m_drop = ifsq_norm_dequeue(ifsq);
2841                 }
2842                 if (m_drop != NULL) {
2843                         IFNET_STAT_INC(ifsq->ifsq_ifp, oqdrops, 1);
2844                         m_freem(m_drop);
2845                         goto again;
2846                 }
2847                 /*
2848                  * No old packets could be dropped!
2849                  * NOTE: Caller increases oqdrops.
2850                  */
2851                 m_freem(m);
2852                 return (ENOBUFS);
2853         } else {
2854                 if (m->m_flags & M_PRIO)
2855                         ifsq_prio_enqueue(ifsq, m);
2856                 else
2857                         ifsq_norm_enqueue(ifsq, m);
2858                 return (0);
2859         }
2860 }
2861
2862 struct mbuf *
2863 ifsq_classic_dequeue(struct ifaltq_subque *ifsq, int op)
2864 {
2865         struct mbuf *m;
2866
2867         switch (op) {
2868         case ALTDQ_POLL:
2869                 m = classq_head(&ifsq->ifsq_prio);
2870                 if (m == NULL)
2871                         m = classq_head(&ifsq->ifsq_norm);
2872                 break;
2873
2874         case ALTDQ_REMOVE:
2875                 m = ifsq_prio_dequeue(ifsq);
2876                 if (m == NULL)
2877                         m = ifsq_norm_dequeue(ifsq);
2878                 break;
2879
2880         default:
2881                 panic("unsupported ALTQ dequeue op: %d", op);
2882         }
2883         return m;
2884 }
2885
2886 int
2887 ifsq_classic_request(struct ifaltq_subque *ifsq, int req, void *arg)
2888 {
2889         switch (req) {
2890         case ALTRQ_PURGE:
2891                 for (;;) {
2892                         struct mbuf *m;
2893
2894                         m = ifsq_classic_dequeue(ifsq, ALTDQ_REMOVE);
2895                         if (m == NULL)
2896                                 break;
2897                         m_freem(m);
2898                 }
2899                 break;
2900
2901         default:
2902                 panic("unsupported ALTQ request: %d", req);
2903         }
2904         return 0;
2905 }
2906
2907 static void
2908 ifsq_ifstart_try(struct ifaltq_subque *ifsq, int force_sched)
2909 {
2910         struct ifnet *ifp = ifsq_get_ifp(ifsq);
2911         int running = 0, need_sched;
2912
2913         /*
2914          * Try to do direct ifnet.if_start on the subqueue first, if there is
2915          * contention on the subqueue hardware serializer, ifnet.if_start on
2916          * the subqueue will be scheduled on the subqueue owner CPU.
2917          */
2918         if (!ifsq_tryserialize_hw(ifsq)) {
2919                 /*
2920                  * Subqueue hardware serializer contention happened,
2921                  * ifnet.if_start on the subqueue is scheduled on
2922                  * the subqueue owner CPU, and we keep going.
2923                  */
2924                 ifsq_ifstart_schedule(ifsq, 1);
2925                 return;
2926         }
2927
2928         if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
2929                 ifp->if_start(ifp, ifsq);
2930                 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
2931                         running = 1;
2932         }
2933         need_sched = ifsq_ifstart_need_schedule(ifsq, running);
2934
2935         ifsq_deserialize_hw(ifsq);
2936
2937         if (need_sched) {
2938                 /*
2939                  * More data need to be transmitted, ifnet.if_start on the
2940                  * subqueue is scheduled on the subqueue owner CPU, and we
2941                  * keep going.
2942                  * NOTE: ifnet.if_start subqueue interlock is not released.
2943                  */
2944                 ifsq_ifstart_schedule(ifsq, force_sched);
2945         }
2946 }
2947
2948 /*
2949  * Subqeue packets staging mechanism:
2950  *
2951  * The packets enqueued into the subqueue are staged to a certain amount
2952  * before the ifnet.if_start on the subqueue is called.  In this way, the
2953  * driver could avoid writing to hardware registers upon every packet,
2954  * instead, hardware registers could be written when certain amount of
2955  * packets are put onto hardware TX ring.  The measurement on several modern
2956  * NICs (emx(4), igb(4), bnx(4), bge(4), jme(4)) shows that the hardware
2957  * registers writing aggregation could save ~20% CPU time when 18bytes UDP
2958  * datagrams are transmitted at 1.48Mpps.  The performance improvement by
2959  * hardware registers writing aggeregation is also mentioned by Luigi Rizzo's
2960  * netmap paper (http://info.iet.unipi.it/~luigi/netmap/).
2961  *
2962  * Subqueue packets staging is performed for two entry points into drivers'
2963  * transmission function:
2964  * - Direct ifnet.if_start calling on the subqueue, i.e. ifsq_ifstart_try()
2965  * - ifnet.if_start scheduling on the subqueue, i.e. ifsq_ifstart_schedule()
2966  *
2967  * Subqueue packets staging will be stopped upon any of the following
2968  * conditions:
2969  * - If the count of packets enqueued on the current CPU is great than or
2970  *   equal to ifsq_stage_cntmax. (XXX this should be per-interface)
2971  * - If the total length of packets enqueued on the current CPU is great
2972  *   than or equal to the hardware's MTU - max_protohdr.  max_protohdr is
2973  *   cut from the hardware's MTU mainly bacause a full TCP segment's size
2974  *   is usually less than hardware's MTU.
2975  * - ifsq_ifstart_schedule() is not pending on the current CPU and
2976  *   ifnet.if_start subqueue interlock (ifaltq_subq.ifsq_started) is not
2977  *   released.
2978  * - The if_start_rollup(), which is registered as low priority netisr
2979  *   rollup function, is called; probably because no more work is pending
2980  *   for netisr.
2981  *
2982  * NOTE:
2983  * Currently subqueue packet staging is only performed in netisr threads.
2984  */
2985 int
2986 ifq_dispatch(struct ifnet *ifp, struct mbuf *m, struct altq_pktattr *pa)
2987 {
2988         struct ifaltq *ifq = &ifp->if_snd;
2989         struct ifaltq_subque *ifsq;
2990         int error, start = 0, len, mcast = 0, avoid_start = 0;
2991         struct ifsubq_stage_head *head = NULL;
2992         struct ifsubq_stage *stage = NULL;
2993         struct globaldata *gd = mycpu;
2994         struct thread *td = gd->gd_curthread;
2995
2996         crit_enter_quick(td);
2997
2998         ifsq = ifq_map_subq(ifq, gd->gd_cpuid);
2999         ASSERT_ALTQ_SQ_NOT_SERIALIZED_HW(ifsq);
3000
3001         len = m->m_pkthdr.len;
3002         if (m->m_flags & M_MCAST)
3003                 mcast = 1;
3004
3005         if (td->td_type == TD_TYPE_NETISR) {
3006                 head = &ifsubq_stage_heads[mycpuid];
3007                 stage = ifsq_get_stage(ifsq, mycpuid);
3008
3009                 stage->stg_cnt++;
3010                 stage->stg_len += len;
3011                 if (stage->stg_cnt < ifsq_stage_cntmax &&
3012                     stage->stg_len < (ifp->if_mtu - max_protohdr))
3013                         avoid_start = 1;
3014         }
3015
3016         ALTQ_SQ_LOCK(ifsq);
3017         error = ifsq_enqueue_locked(ifsq, m, pa);
3018         if (error) {
3019                 IFNET_STAT_INC(ifp, oqdrops, 1);
3020                 if (!ifsq_data_ready(ifsq)) {
3021                         ALTQ_SQ_UNLOCK(ifsq);
3022                         crit_exit_quick(td);
3023                         return error;
3024                 }
3025                 avoid_start = 0;
3026         }
3027         if (!ifsq_is_started(ifsq)) {
3028                 if (avoid_start) {
3029                         ALTQ_SQ_UNLOCK(ifsq);
3030
3031                         KKASSERT(!error);
3032                         if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
3033                                 ifsq_stage_insert(head, stage);
3034
3035                         IFNET_STAT_INC(ifp, obytes, len);
3036                         if (mcast)
3037                                 IFNET_STAT_INC(ifp, omcasts, 1);
3038                         crit_exit_quick(td);
3039                         return error;
3040                 }
3041
3042                 /*
3043                  * Hold the subqueue interlock of ifnet.if_start
3044                  */
3045                 ifsq_set_started(ifsq);
3046                 start = 1;
3047         }
3048         ALTQ_SQ_UNLOCK(ifsq);
3049
3050         if (!error) {
3051                 IFNET_STAT_INC(ifp, obytes, len);
3052                 if (mcast)
3053                         IFNET_STAT_INC(ifp, omcasts, 1);
3054         }
3055
3056         if (stage != NULL) {
3057                 if (!start && (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)) {
3058                         KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
3059                         if (!avoid_start) {
3060                                 ifsq_stage_remove(head, stage);
3061                                 ifsq_ifstart_schedule(ifsq, 1);
3062                         }
3063                         crit_exit_quick(td);
3064                         return error;
3065                 }
3066
3067                 if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED) {
3068                         ifsq_stage_remove(head, stage);
3069                 } else {
3070                         stage->stg_cnt = 0;
3071                         stage->stg_len = 0;
3072                 }
3073         }
3074
3075         if (!start) {
3076                 crit_exit_quick(td);
3077                 return error;
3078         }
3079
3080         ifsq_ifstart_try(ifsq, 0);
3081
3082         crit_exit_quick(td);
3083         return error;
3084 }
3085
3086 void *
3087 ifa_create(int size)
3088 {
3089         struct ifaddr *ifa;
3090         int i;
3091
3092         KASSERT(size >= sizeof(*ifa), ("ifaddr size too small"));
3093
3094         ifa = kmalloc(size, M_IFADDR, M_INTWAIT | M_ZERO);
3095         ifa->ifa_containers =
3096             kmalloc_cachealign(ncpus * sizeof(struct ifaddr_container),
3097                 M_IFADDR, M_INTWAIT | M_ZERO);
3098
3099         ifa->ifa_ncnt = ncpus;
3100         for (i = 0; i < ncpus; ++i) {
3101                 struct ifaddr_container *ifac = &ifa->ifa_containers[i];
3102
3103                 ifac->ifa_magic = IFA_CONTAINER_MAGIC;
3104                 ifac->ifa = ifa;
3105                 ifac->ifa_refcnt = 1;
3106         }
3107 #ifdef IFADDR_DEBUG
3108         kprintf("alloc ifa %p %d\n", ifa, size);
3109 #endif
3110         return ifa;
3111 }
3112
3113 void
3114 ifac_free(struct ifaddr_container *ifac, int cpu_id)
3115 {
3116         struct ifaddr *ifa = ifac->ifa;
3117
3118         KKASSERT(ifac->ifa_magic == IFA_CONTAINER_MAGIC);
3119         KKASSERT(ifac->ifa_refcnt == 0);
3120         KASSERT(ifac->ifa_listmask == 0,
3121                 ("ifa is still on %#x lists", ifac->ifa_listmask));
3122
3123         ifac->ifa_magic = IFA_CONTAINER_DEAD;
3124
3125 #ifdef IFADDR_DEBUG_VERBOSE
3126         kprintf("try free ifa %p cpu_id %d\n", ifac->ifa, cpu_id);
3127 #endif
3128
3129         KASSERT(ifa->ifa_ncnt > 0 && ifa->ifa_ncnt <= ncpus,
3130                 ("invalid # of ifac, %d", ifa->ifa_ncnt));
3131         if (atomic_fetchadd_int(&ifa->ifa_ncnt, -1) == 1) {
3132 #ifdef IFADDR_DEBUG
3133                 kprintf("free ifa %p\n", ifa);
3134 #endif
3135                 kfree(ifa->ifa_containers, M_IFADDR);
3136                 kfree(ifa, M_IFADDR);
3137         }
3138 }
3139
3140 static void
3141 ifa_iflink_dispatch(netmsg_t nmsg)
3142 {
3143         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3144         struct ifaddr *ifa = msg->ifa;
3145         struct ifnet *ifp = msg->ifp;
3146         int cpu = mycpuid;
3147         struct ifaddr_container *ifac;
3148
3149         crit_enter();
3150
3151         ifac = &ifa->ifa_containers[cpu];
3152         ASSERT_IFAC_VALID(ifac);
3153         KASSERT((ifac->ifa_listmask & IFA_LIST_IFADDRHEAD) == 0,
3154                 ("ifaddr is on if_addrheads"));
3155
3156         ifac->ifa_listmask |= IFA_LIST_IFADDRHEAD;
3157         if (msg->tail)
3158                 TAILQ_INSERT_TAIL(&ifp->if_addrheads[cpu], ifac, ifa_link);
3159         else
3160                 TAILQ_INSERT_HEAD(&ifp->if_addrheads[cpu], ifac, ifa_link);
3161
3162         crit_exit();
3163
3164         netisr_forwardmsg(&nmsg->base, cpu + 1);
3165 }
3166
3167 void
3168 ifa_iflink(struct ifaddr *ifa, struct ifnet *ifp, int tail)
3169 {
3170         struct netmsg_ifaddr msg;
3171
3172         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3173                     0, ifa_iflink_dispatch);
3174         msg.ifa = ifa;
3175         msg.ifp = ifp;
3176         msg.tail = tail;
3177
3178         netisr_domsg(&msg.base, 0);
3179 }
3180
3181 static void
3182 ifa_ifunlink_dispatch(netmsg_t nmsg)
3183 {
3184         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3185         struct ifaddr *ifa = msg->ifa;
3186         struct ifnet *ifp = msg->ifp;
3187         int cpu = mycpuid;
3188         struct ifaddr_container *ifac;
3189
3190         crit_enter();
3191
3192         ifac = &ifa->ifa_containers[cpu];
3193         ASSERT_IFAC_VALID(ifac);
3194         KASSERT(ifac->ifa_listmask & IFA_LIST_IFADDRHEAD,
3195                 ("ifaddr is not on if_addrhead"));
3196
3197         TAILQ_REMOVE(&ifp->if_addrheads[cpu], ifac, ifa_link);
3198         ifac->ifa_listmask &= ~IFA_LIST_IFADDRHEAD;
3199
3200         crit_exit();
3201
3202         netisr_forwardmsg(&nmsg->base, cpu + 1);
3203 }
3204
3205 void
3206 ifa_ifunlink(struct ifaddr *ifa, struct ifnet *ifp)
3207 {
3208         struct netmsg_ifaddr msg;
3209
3210         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3211                     0, ifa_ifunlink_dispatch);
3212         msg.ifa = ifa;
3213         msg.ifp = ifp;
3214
3215         netisr_domsg(&msg.base, 0);
3216 }
3217
3218 static void
3219 ifa_destroy_dispatch(netmsg_t nmsg)
3220 {
3221         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3222
3223         IFAFREE(msg->ifa);
3224         netisr_forwardmsg(&nmsg->base, mycpuid + 1);
3225 }
3226
3227 void
3228 ifa_destroy(struct ifaddr *ifa)
3229 {
3230         struct netmsg_ifaddr msg;
3231
3232         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3233                     0, ifa_destroy_dispatch);
3234         msg.ifa = ifa;
3235
3236         netisr_domsg(&msg.base, 0);
3237 }
3238
3239 static void
3240 if_start_rollup(void)
3241 {
3242         struct ifsubq_stage_head *head = &ifsubq_stage_heads[mycpuid];
3243         struct ifsubq_stage *stage;
3244
3245         crit_enter();
3246
3247         while ((stage = TAILQ_FIRST(&head->stg_head)) != NULL) {
3248                 struct ifaltq_subque *ifsq = stage->stg_subq;
3249                 int is_sched = 0;
3250
3251                 if (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)
3252                         is_sched = 1;
3253                 ifsq_stage_remove(head, stage);
3254
3255                 if (is_sched) {
3256                         ifsq_ifstart_schedule(ifsq, 1);
3257                 } else {
3258                         int start = 0;
3259
3260                         ALTQ_SQ_LOCK(ifsq);
3261                         if (!ifsq_is_started(ifsq)) {
3262                                 /*
3263                                  * Hold the subqueue interlock of
3264                                  * ifnet.if_start
3265                                  */
3266                                 ifsq_set_started(ifsq);
3267                                 start = 1;
3268                         }
3269                         ALTQ_SQ_UNLOCK(ifsq);
3270
3271                         if (start)
3272                                 ifsq_ifstart_try(ifsq, 1);
3273                 }
3274                 KKASSERT((stage->stg_flags &
3275                     (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
3276         }
3277
3278         crit_exit();
3279 }
3280
3281 static void
3282 ifnetinit(void *dummy __unused)
3283 {
3284         int i;
3285
3286         for (i = 0; i < ncpus; ++i)
3287                 TAILQ_INIT(&ifsubq_stage_heads[i].stg_head);
3288         netisr_register_rollup(if_start_rollup, NETISR_ROLLUP_PRIO_IFSTART);
3289 }
3290
3291 void
3292 if_register_com_alloc(u_char type,
3293     if_com_alloc_t *a, if_com_free_t *f)
3294 {
3295
3296         KASSERT(if_com_alloc[type] == NULL,
3297             ("if_register_com_alloc: %d already registered", type));
3298         KASSERT(if_com_free[type] == NULL,
3299             ("if_register_com_alloc: %d free already registered", type));
3300
3301         if_com_alloc[type] = a;
3302         if_com_free[type] = f;
3303 }
3304
3305 void
3306 if_deregister_com_alloc(u_char type)
3307 {
3308
3309         KASSERT(if_com_alloc[type] != NULL,
3310             ("if_deregister_com_alloc: %d not registered", type));
3311         KASSERT(if_com_free[type] != NULL,
3312             ("if_deregister_com_alloc: %d free not registered", type));
3313         if_com_alloc[type] = NULL;
3314         if_com_free[type] = NULL;
3315 }
3316
3317 void
3318 ifq_set_maxlen(struct ifaltq *ifq, int len)
3319 {
3320         ifq->altq_maxlen = len + (ncpus * ifsq_stage_cntmax);
3321 }
3322
3323 int
3324 ifq_mapsubq_default(struct ifaltq *ifq __unused, int cpuid __unused)
3325 {
3326         return ALTQ_SUBQ_INDEX_DEFAULT;
3327 }
3328
3329 int
3330 ifq_mapsubq_mask(struct ifaltq *ifq, int cpuid)
3331 {
3332
3333         return (cpuid & ifq->altq_subq_mappriv);
3334 }
3335
3336 int
3337 ifq_mapsubq_modulo(struct ifaltq *ifq, int cpuid)
3338 {
3339
3340         return (cpuid % ifq->altq_subq_mappriv);
3341 }
3342
3343 static void
3344 ifsq_watchdog(void *arg)
3345 {
3346         struct ifsubq_watchdog *wd = arg;
3347         struct ifnet *ifp;
3348
3349         if (__predict_true(wd->wd_timer == 0 || --wd->wd_timer))
3350                 goto done;
3351
3352         ifp = ifsq_get_ifp(wd->wd_subq);
3353         if (ifnet_tryserialize_all(ifp)) {
3354                 wd->wd_watchdog(wd->wd_subq);
3355                 ifnet_deserialize_all(ifp);
3356         } else {
3357                 /* try again next timeout */
3358                 wd->wd_timer = 1;
3359         }
3360 done:
3361         ifsq_watchdog_reset(wd);
3362 }
3363
3364 static void
3365 ifsq_watchdog_reset(struct ifsubq_watchdog *wd)
3366 {
3367         callout_reset_bycpu(&wd->wd_callout, hz, ifsq_watchdog, wd,
3368             ifsq_get_cpuid(wd->wd_subq));
3369 }
3370
3371 void
3372 ifsq_watchdog_init(struct ifsubq_watchdog *wd, struct ifaltq_subque *ifsq,
3373     ifsq_watchdog_t watchdog)
3374 {
3375         callout_init_mp(&wd->wd_callout);
3376         wd->wd_timer = 0;
3377         wd->wd_subq = ifsq;
3378         wd->wd_watchdog = watchdog;
3379 }
3380
3381 void
3382 ifsq_watchdog_start(struct ifsubq_watchdog *wd)
3383 {
3384         wd->wd_timer = 0;
3385         ifsq_watchdog_reset(wd);
3386 }
3387
3388 void
3389 ifsq_watchdog_stop(struct ifsubq_watchdog *wd)
3390 {
3391         wd->wd_timer = 0;
3392         callout_stop(&wd->wd_callout);
3393 }
3394
3395 void
3396 ifnet_lock(void)
3397 {
3398         KASSERT(curthread->td_type != TD_TYPE_NETISR,
3399             ("try holding ifnet lock in netisr"));
3400         mtx_lock(&ifnet_mtx);
3401 }
3402
3403 void
3404 ifnet_unlock(void)
3405 {
3406         KASSERT(curthread->td_type != TD_TYPE_NETISR,
3407             ("try holding ifnet lock in netisr"));
3408         mtx_unlock(&ifnet_mtx);
3409 }
3410
3411 static struct ifnet_array *
3412 ifnet_array_alloc(int count)
3413 {
3414         struct ifnet_array *arr;
3415
3416         arr = kmalloc(__offsetof(struct ifnet_array, ifnet_arr[count]),
3417             M_IFNET, M_WAITOK);
3418         arr->ifnet_count = count;
3419
3420         return arr;
3421 }
3422
3423 static void
3424 ifnet_array_free(struct ifnet_array *arr)
3425 {
3426         if (arr == &ifnet_array0)
3427                 return;
3428         kfree(arr, M_IFNET);
3429 }
3430
3431 static struct ifnet_array *
3432 ifnet_array_add(struct ifnet *ifp, const struct ifnet_array *old_arr)
3433 {
3434         struct ifnet_array *arr;
3435         int count, i;
3436
3437         KASSERT(old_arr->ifnet_count >= 0,
3438             ("invalid ifnet array count %d", old_arr->ifnet_count));
3439         count = old_arr->ifnet_count + 1;
3440         arr = ifnet_array_alloc(count);
3441
3442         /*
3443          * Save the old ifnet array and append this ifp to the end of
3444          * the new ifnet array.
3445          */
3446         for (i = 0; i < old_arr->ifnet_count; ++i) {
3447                 KASSERT(old_arr->ifnet_arr[i] != ifp,
3448                     ("%s is already in ifnet array", ifp->if_xname));
3449                 arr->ifnet_arr[i] = old_arr->ifnet_arr[i];
3450         }
3451         KASSERT(i == count - 1,
3452             ("add %s, ifnet array index mismatch, should be %d, but got %d",
3453              ifp->if_xname, count - 1, i));
3454         arr->ifnet_arr[i] = ifp;
3455
3456         return arr;
3457 }
3458
3459 static struct ifnet_array *
3460 ifnet_array_del(struct ifnet *ifp, const struct ifnet_array *old_arr)
3461 {
3462         struct ifnet_array *arr;
3463         int count, i, idx, found = 0;
3464
3465         KASSERT(old_arr->ifnet_count > 0,
3466             ("invalid ifnet array count %d", old_arr->ifnet_count));
3467         count = old_arr->ifnet_count - 1;
3468         arr = ifnet_array_alloc(count);
3469
3470         /*
3471          * Save the old ifnet array, but skip this ifp.
3472          */
3473         idx = 0;
3474         for (i = 0; i < old_arr->ifnet_count; ++i) {
3475                 if (old_arr->ifnet_arr[i] == ifp) {
3476                         KASSERT(!found,
3477                             ("dup %s is in ifnet array", ifp->if_xname));
3478                         found = 1;
3479                         continue;
3480                 }
3481                 KASSERT(idx < count,
3482                     ("invalid ifnet array index %d, count %d", idx, count));
3483                 arr->ifnet_arr[idx] = old_arr->ifnet_arr[i];
3484                 ++idx;
3485         }
3486         KASSERT(found, ("%s is not in ifnet array", ifp->if_xname));
3487         KASSERT(idx == count,
3488             ("del %s, ifnet array count mismatch, should be %d, but got %d ",
3489              ifp->if_xname, count, idx));
3490
3491         return arr;
3492 }
3493
3494 const struct ifnet_array *
3495 ifnet_array_get(void)
3496 {
3497         const struct ifnet_array *ret;
3498
3499         KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
3500         ret = ifnet_array;
3501         /* Make sure 'ret' is really used. */
3502         cpu_ccfence();
3503         return (ret);
3504 }
3505
3506 int
3507 ifnet_array_isempty(void)
3508 {
3509         KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
3510         if (ifnet_array->ifnet_count == 0)
3511                 return 1;
3512         else
3513                 return 0;
3514 }
3515
3516 void
3517 ifa_marker_init(struct ifaddr_marker *mark, struct ifnet *ifp)
3518 {
3519         struct ifaddr *ifa;
3520
3521         memset(mark, 0, sizeof(*mark));
3522         ifa = &mark->ifa;
3523
3524         mark->ifac.ifa = ifa;
3525
3526         ifa->ifa_addr = &mark->addr;
3527         ifa->ifa_dstaddr = &mark->dstaddr;
3528         ifa->ifa_netmask = &mark->netmask;
3529         ifa->ifa_ifp = ifp;
3530 }
3531
3532 static int
3533 if_ringcnt_fixup(int ring_cnt, int ring_cntmax)
3534 {
3535
3536         KASSERT(ring_cntmax > 0, ("invalid ring count max %d", ring_cntmax));
3537
3538         if (ring_cnt <= 0 || ring_cnt > ring_cntmax)
3539                 ring_cnt = ring_cntmax;
3540         if (ring_cnt > netisr_ncpus)
3541                 ring_cnt = netisr_ncpus;
3542         return (ring_cnt);
3543 }
3544
3545 static void
3546 if_ringmap_set_grid(device_t dev, struct if_ringmap *rm, int grid)
3547 {
3548         int i, offset;
3549
3550         KASSERT(grid > 0, ("invalid if_ringmap grid %d", grid));
3551         KASSERT(grid >= rm->rm_cnt, ("invalid if_ringmap grid %d, count %d",
3552             grid, rm->rm_cnt));
3553         rm->rm_grid = grid;
3554
3555         offset = (rm->rm_grid * device_get_unit(dev)) % netisr_ncpus;
3556         for (i = 0; i < rm->rm_cnt; ++i) {
3557                 rm->rm_cpumap[i] = offset + i;
3558                 KASSERT(rm->rm_cpumap[i] < netisr_ncpus,
3559                     ("invalid cpumap[%d] = %d, offset %d", i,
3560                      rm->rm_cpumap[i], offset));
3561         }
3562 }
3563
3564 static struct if_ringmap *
3565 if_ringmap_alloc_flags(device_t dev, int ring_cnt, int ring_cntmax,
3566     uint32_t flags)
3567 {
3568         struct if_ringmap *rm;
3569         int i, grid = 0, prev_grid;
3570
3571         ring_cnt = if_ringcnt_fixup(ring_cnt, ring_cntmax);
3572         rm = kmalloc(__offsetof(struct if_ringmap, rm_cpumap[ring_cnt]),
3573             M_DEVBUF, M_WAITOK | M_ZERO);
3574
3575         rm->rm_cnt = ring_cnt;
3576         if (flags & RINGMAP_FLAG_POWEROF2)
3577                 rm->rm_cnt = 1 << (fls(rm->rm_cnt) - 1);
3578
3579         prev_grid = netisr_ncpus;
3580         for (i = 0; i < netisr_ncpus; ++i) {
3581                 if (netisr_ncpus % (i + 1) != 0)
3582                         continue;
3583
3584                 grid = netisr_ncpus / (i + 1);
3585                 if (rm->rm_cnt > grid) {
3586                         grid = prev_grid;
3587                         break;
3588                 }
3589
3590                 if (rm->rm_cnt > netisr_ncpus / (i + 2))
3591                         break;
3592                 prev_grid = grid;
3593         }
3594         if_ringmap_set_grid(dev, rm, grid);
3595
3596         return (rm);
3597 }
3598
3599 struct if_ringmap *
3600 if_ringmap_alloc(device_t dev, int ring_cnt, int ring_cntmax)
3601 {
3602
3603         return (if_ringmap_alloc_flags(dev, ring_cnt, ring_cntmax,
3604             RINGMAP_FLAG_NONE));
3605 }
3606
3607 struct if_ringmap *
3608 if_ringmap_alloc2(device_t dev, int ring_cnt, int ring_cntmax)
3609 {
3610
3611         return (if_ringmap_alloc_flags(dev, ring_cnt, ring_cntmax,
3612             RINGMAP_FLAG_POWEROF2));
3613 }
3614
3615 void
3616 if_ringmap_free(struct if_ringmap *rm)
3617 {
3618
3619         kfree(rm, M_DEVBUF);
3620 }
3621
3622 /*
3623  * Align the two ringmaps.
3624  *
3625  * e.g. 8 netisrs, rm0 contains 4 rings, rm1 contains 2 rings.
3626  *
3627  * Before:
3628  *
3629  * CPU      0  1  2  3   4  5  6  7
3630  * NIC_RX               n0 n1 n2 n3
3631  * NIC_TX        N0 N1
3632  *
3633  * After:
3634  *
3635  * CPU      0  1  2  3   4  5  6  7
3636  * NIC_RX               n0 n1 n2 n3
3637  * NIC_TX               N0 N1
3638  */
3639 void
3640 if_ringmap_align(device_t dev, struct if_ringmap *rm0, struct if_ringmap *rm1)
3641 {
3642
3643         if (rm0->rm_grid > rm1->rm_grid)
3644                 if_ringmap_set_grid(dev, rm1, rm0->rm_grid);
3645         else if (rm0->rm_grid < rm1->rm_grid)
3646                 if_ringmap_set_grid(dev, rm0, rm1->rm_grid);
3647 }
3648
3649 void
3650 if_ringmap_match(device_t dev, struct if_ringmap *rm0, struct if_ringmap *rm1)
3651 {
3652         int subset_grid, cnt, divisor, mod, offset, i;
3653         struct if_ringmap *subset_rm, *rm;
3654         int old_rm0_grid, old_rm1_grid;
3655
3656         if (rm0->rm_grid == rm1->rm_grid)
3657                 return;
3658
3659         /* Save grid for later use */
3660         old_rm0_grid = rm0->rm_grid;
3661         old_rm1_grid = rm1->rm_grid;
3662
3663         if_ringmap_align(dev, rm0, rm1);
3664
3665         /*
3666          * Re-shuffle rings to get more even distribution.
3667          *
3668          * e.g. 12 netisrs, rm0 contains 4 rings, rm1 contains 2 rings.
3669          *
3670          * CPU       0  1  2  3   4  5  6  7   8  9 10 11
3671          *
3672          * NIC_RX   a0 a1 a2 a3  b0 b1 b2 b3  c0 c1 c2 c3
3673          * NIC_TX   A0 A1        B0 B1        C0 C1
3674          *
3675          * NIC_RX   d0 d1 d2 d3  e0 e1 e2 e3  f0 f1 f2 f3
3676          * NIC_TX         D0 D1        E0 E1        F0 F1
3677          */
3678
3679         if (rm0->rm_cnt >= (2 * old_rm1_grid)) {
3680                 cnt = rm0->rm_cnt;
3681                 subset_grid = old_rm1_grid;
3682                 subset_rm = rm1;
3683                 rm = rm0;
3684         } else if (rm1->rm_cnt > (2 * old_rm0_grid)) {
3685                 cnt = rm1->rm_cnt;
3686                 subset_grid = old_rm0_grid;
3687                 subset_rm = rm0;
3688                 rm = rm1;
3689         } else {
3690                 /* No space to shuffle. */
3691                 return;
3692         }
3693
3694         mod = cnt / subset_grid;
3695         KKASSERT(mod >= 2);
3696         divisor = netisr_ncpus / rm->rm_grid;
3697         offset = ((device_get_unit(dev) / divisor) % mod) * subset_grid;
3698
3699         for (i = 0; i < subset_rm->rm_cnt; ++i) {
3700                 subset_rm->rm_cpumap[i] += offset;
3701                 KASSERT(subset_rm->rm_cpumap[i] < netisr_ncpus,
3702                     ("match: invalid cpumap[%d] = %d, offset %d",
3703                      i, subset_rm->rm_cpumap[i], offset));
3704         }
3705 #ifdef INVARIANTS
3706         for (i = 0; i < subset_rm->rm_cnt; ++i) {
3707                 int j;
3708
3709                 for (j = 0; j < rm->rm_cnt; ++j) {
3710                         if (rm->rm_cpumap[j] == subset_rm->rm_cpumap[i])
3711                                 break;
3712                 }
3713                 KASSERT(j < rm->rm_cnt,
3714                     ("subset cpumap[%d] = %d not found in superset",
3715                      i, subset_rm->rm_cpumap[i]));
3716         }
3717 #endif
3718 }
3719
3720 int
3721 if_ringmap_count(const struct if_ringmap *rm)
3722 {
3723
3724         return (rm->rm_cnt);
3725 }
3726
3727 int
3728 if_ringmap_cpumap(const struct if_ringmap *rm, int ring)
3729 {
3730
3731         KASSERT(ring >= 0 && ring < rm->rm_cnt, ("invalid ring %d", ring));
3732         return (rm->rm_cpumap[ring]);
3733 }
3734
3735 void
3736 if_ringmap_rdrtable(const struct if_ringmap *rm, int table[], int table_nent)
3737 {
3738         int i, grid_idx, grid_cnt, patch_off, patch_cnt, ncopy;
3739
3740         KASSERT(table_nent > 0 && (table_nent & NETISR_CPUMASK) == 0,
3741             ("invalid redirect table entries %d", table_nent));
3742
3743         grid_idx = 0;
3744         for (i = 0; i < NETISR_CPUMAX; ++i) {
3745                 table[i] = grid_idx++ % rm->rm_cnt;
3746
3747                 if (grid_idx == rm->rm_grid)
3748                         grid_idx = 0;
3749         }
3750
3751         /*
3752          * Make the ring distributed more evenly for the remainder
3753          * of each grid.
3754          *
3755          * e.g. 12 netisrs, rm contains 8 rings.
3756          *
3757          * Redirect table before:
3758          *
3759          *  0  1  2  3  4  5  6  7  0  1  2  3  0  1  2  3
3760          *  4  5  6  7  0  1  2  3  0  1  2  3  4  5  6  7
3761          *  0  1  2  3  0  1  2  3  4  5  6  7  0  1  2  3
3762          *  ....
3763          *
3764          * Redirect table after being patched (pX, patched entries):
3765          *
3766          *  0  1  2  3  4  5  6  7 p0 p1 p2 p3  0  1  2  3
3767          *  4  5  6  7 p4 p5 p6 p7  0  1  2  3  4  5  6  7
3768          * p0 p1 p2 p3  0  1  2  3  4  5  6  7 p4 p5 p6 p7
3769          *  ....
3770          */
3771         patch_cnt = rm->rm_grid % rm->rm_cnt;
3772         if (patch_cnt == 0)
3773                 goto done;
3774         patch_off = rm->rm_grid - (rm->rm_grid % rm->rm_cnt);
3775
3776         grid_cnt = roundup(NETISR_CPUMAX, rm->rm_grid) / rm->rm_grid;
3777         grid_idx = 0;
3778         for (i = 0; i < grid_cnt; ++i) {
3779                 int j;
3780
3781                 for (j = 0; j < patch_cnt; ++j) {
3782                         int fix_idx;
3783
3784                         fix_idx = (i * rm->rm_grid) + patch_off + j;
3785                         if (fix_idx >= NETISR_CPUMAX)
3786                                 goto done;
3787                         table[fix_idx] = grid_idx++ % rm->rm_cnt;
3788                 }
3789         }
3790 done:
3791         /*
3792          * If the device supports larger redirect table, duplicate
3793          * the first NETISR_CPUMAX entries to the rest of the table,
3794          * so that it matches upper layer's expectation:
3795          * (hash & NETISR_CPUMASK) % netisr_ncpus
3796          */
3797         ncopy = table_nent / NETISR_CPUMAX;
3798         for (i = 1; i < ncopy; ++i) {
3799                 memcpy(&table[i * NETISR_CPUMAX], table,
3800                     NETISR_CPUMAX * sizeof(table[0]));
3801         }
3802         if (if_ringmap_dumprdr) {
3803                 for (i = 0; i < table_nent; ++i) {
3804                         if (i != 0 && i % 16 == 0)
3805                                 kprintf("\n");
3806                         kprintf("%03d ", table[i]);
3807                 }
3808                 kprintf("\n");
3809         }
3810 }
3811
3812 int
3813 if_ringmap_cpumap_sysctl(SYSCTL_HANDLER_ARGS)
3814 {
3815         struct if_ringmap *rm = arg1;
3816         int i, error = 0;
3817
3818         for (i = 0; i < rm->rm_cnt; ++i) {
3819                 int cpu = rm->rm_cpumap[i];
3820
3821                 error = SYSCTL_OUT(req, &cpu, sizeof(cpu));
3822                 if (error)
3823                         break;
3824         }
3825         return (error);
3826 }
3827
3828 int
3829 if_ring_count2(int ring_cnt, int ring_cntmax)
3830 {
3831
3832         ring_cnt = if_ringcnt_fixup(ring_cnt, ring_cntmax);
3833         return (1 << (fls(ring_cnt) - 1));
3834 }