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