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          * Make if_addrhead available on all CPUs, since they
 530          * could be accessed by any threads.
 531          */
 532         ifp->if_addrheads = kmalloc(ncpus * sizeof(struct ifaddrhead),
 533                                     M_IFADDR, M_WAITOK | M_ZERO);
 534         for (i = 0; i < ncpus; ++i)
 535                 TAILQ_INIT(&ifp->if_addrheads[i]);
 536
 537         TAILQ_INIT(&ifp->if_multiaddrs);
 538         TAILQ_INIT(&ifp->if_groups);
 539         getmicrotime(&ifp->if_lastchange);
 540
 541         /*
 542          * create a Link Level name for this device
 543          */
 544         namelen = strlen(ifp->if_xname);
 545         masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + namelen;
 546         socksize = masklen + ifp->if_addrlen;
 547         if (socksize < sizeof(*sdl))
 548                 socksize = sizeof(*sdl);
 549         socksize = RT_ROUNDUP(socksize);
 550         ifa = ifa_create(sizeof(struct ifaddr) + 2 * socksize);
 551         sdl = sdl_addr = (struct sockaddr_dl *)(ifa + 1);
 552         sdl->sdl_len = socksize;
 553         sdl->sdl_family = AF_LINK;
 554         bcopy(ifp->if_xname, sdl->sdl_data, namelen);
 555         sdl->sdl_nlen = namelen;
 556         sdl->sdl_type = ifp->if_type;
 557         ifp->if_lladdr = ifa;
 558         ifa->ifa_ifp = ifp;
 559         ifa->ifa_rtrequest = link_rtrequest;
 560         ifa->ifa_addr = (struct sockaddr *)sdl;
 561         sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
 562         ifa->ifa_netmask = (struct sockaddr *)sdl;
 563         sdl->sdl_len = masklen;
 564         while (namelen != 0)
 565                 sdl->sdl_data[--namelen] = 0xff;
 566         ifa_iflink(ifa, ifp, 0 /* Insert head */);
 567
 568         /*
 569          * Make if_data available on all CPUs, since they could
 570          * be updated by hardware interrupt routing, which could
 571          * be bound to any CPU.
 572          */
 573         ifp->if_data_pcpu = kmalloc_cachealign(
 574             ncpus * sizeof(struct ifdata_pcpu), M_DEVBUF, M_WAITOK | M_ZERO);
 575
 576         if (ifp->if_mapsubq == NULL)
 577                 ifp->if_mapsubq = ifq_mapsubq_default;
 578
 579         ifq = &ifp->if_snd;
 580         ifq->altq_type = 0;
 581         ifq->altq_disc = NULL;
 582         ifq->altq_flags &= ALTQF_CANTCHANGE;
 583         ifq->altq_tbr = NULL;
 584         ifq->altq_ifp = ifp;
 585
 586         if (ifq->altq_subq_cnt <= 0)
 587                 ifq->altq_subq_cnt = 1;
 588         ifq->altq_subq = kmalloc_cachealign(
 589             ifq->altq_subq_cnt * sizeof(struct ifaltq_subque),
 590             M_DEVBUF, M_WAITOK | M_ZERO);
 591
 592         if (ifq->altq_maxlen == 0) {
 593                 if_printf(ifp, "driver didn't set altq_maxlen\n");
 594                 ifq_set_maxlen(ifq, ifqmaxlen);
 595         }
 596
 597         for (q = 0; q < ifq->altq_subq_cnt; ++q) {
 598                 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
 599
 600                 ALTQ_SQ_LOCK_INIT(ifsq);
 601                 ifsq->ifsq_index = q;
 602
 603                 ifsq->ifsq_altq = ifq;
 604                 ifsq->ifsq_ifp = ifp;
 605
 606                 ifsq->ifsq_maxlen = ifq->altq_maxlen;
 607                 ifsq->ifsq_maxbcnt = ifsq->ifsq_maxlen * MCLBYTES;
 608                 ifsq->ifsq_prepended = NULL;
 609                 ifsq->ifsq_started = 0;
 610                 ifsq->ifsq_hw_oactive = 0;
 611                 ifsq_set_cpuid(ifsq, 0);
 612                 if (ifp->if_serializer != NULL)
 613                         ifsq_set_hw_serialize(ifsq, ifp->if_serializer);
 614
 615                 /* XXX: netisr_ncpus */
 616                 ifsq->ifsq_stage =
 617                     kmalloc_cachealign(ncpus * sizeof(struct ifsubq_stage),
 618                     M_DEVBUF, M_WAITOK | M_ZERO);
 619                 for (i = 0; i < ncpus; ++i)
 620                         ifsq->ifsq_stage[i].stg_subq = ifsq;
 621
 622                 /*
 623                  * Allocate one if_start message for each CPU, since
 624                  * the hardware TX ring could be assigned to any CPU.
 625                  *
 626                  * NOTE:
 627                  * If the hardware TX ring polling CPU and the hardware
 628                  * TX ring interrupt CPU are same, one if_start message
 629                  * should be enough.
 630                  */
 631                 ifsq->ifsq_ifstart_nmsg =
 632                     kmalloc(ncpus * sizeof(struct netmsg_base),
 633                     M_LWKTMSG, M_WAITOK);
 634                 for (i = 0; i < ncpus; ++i) {
 635                         netmsg_init(&ifsq->ifsq_ifstart_nmsg[i], NULL,
 636                             &netisr_adone_rport, 0, ifsq_ifstart_dispatch);
 637                         ifsq->ifsq_ifstart_nmsg[i].lmsg.u.ms_resultp = ifsq;
 638                 }
 639         }
 640         ifq_set_classic(ifq);
 641
 642         /*
 643          * Increase mbuf cluster/jcluster limits for the mbufs that
 644          * could sit on the device queues for quite some time.
 645          */
 646         if (ifp->if_nmbclusters > 0)
 647                 mcl_inclimit(ifp->if_nmbclusters);
 648         if (ifp->if_nmbjclusters > 0)
 649                 mjcl_inclimit(ifp->if_nmbjclusters);
 650
 651         /*
 652          * Install this ifp into ifindex2inet, ifnet queue and ifnet
 653          * array after it is setup.
 654          *
 655          * Protect ifindex2ifnet, ifnet queue and ifnet array changes
 656          * by ifnet lock, so that non-netisr threads could get a
 657          * consistent view.
 658          */
 659         ifnet_lock();
 660
 661         /* Don't update if_index until ifindex2ifnet is setup */
 662         ifp->if_index = if_index + 1;
 663         sdl_addr->sdl_index = ifp->if_index;
 664
 665         /*
 666          * Install this ifp into ifindex2ifnet
 667          */
 668         if (ifindex2ifnet == NULL || ifp->if_index >= if_indexlim) {
 669                 unsigned int n;
 670                 struct ifnet **q;
 671
 672                 /*
 673                  * Grow ifindex2ifnet
 674                  */
 675                 if_indexlim <<= 1;
 676                 n = if_indexlim * sizeof(*q);
 677                 q = kmalloc(n, M_IFADDR, M_WAITOK | M_ZERO);
 678                 if (ifindex2ifnet != NULL) {
 679                         bcopy(ifindex2ifnet, q, n/2);
 680                         /* Free old ifindex2ifnet after sync all netisrs */
 681                         old_ifindex2ifnet = ifindex2ifnet;
 682                 }
 683                 ifindex2ifnet = q;
 684         }
 685         ifindex2ifnet[ifp->if_index] = ifp;
 686         /*
 687          * Update if_index after this ifp is installed into ifindex2ifnet,
 688          * so that netisrs could get a consistent view of ifindex2ifnet.
 689          */
 690         cpu_sfence();
 691         if_index = ifp->if_index;
 692
 693         /*
 694          * Install this ifp into ifnet array.
 695          */
 696         /* Free old ifnet array after sync all netisrs */
 697         old_ifnet_array = ifnet_array;
 698         ifnet_array = ifnet_array_add(ifp, old_ifnet_array);
 699
 700         /*
 701          * Install this ifp into ifnet queue.
 702          */
 703         TAILQ_INSERT_TAIL(&ifnetlist, ifp, if_link);
 704
 705         ifnet_unlock();
 706
 707         /*
 708          * Sync all netisrs so that the old ifindex2ifnet and ifnet array
 709          * are no longer accessed and we can free them safely later on.
 710          */
 711         netmsg_service_sync();
 712         if (old_ifindex2ifnet != NULL)
 713                 kfree(old_ifindex2ifnet, M_IFADDR);
 714         ifnet_array_free(old_ifnet_array);
 715
 716         if (!SLIST_EMPTY(&domains))
 717                 if_attachdomain1(ifp);
 718
 719         /* Announce the interface. */
 720         EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
 721         devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
 722         rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
 723 }
 724
 725 static void
 726 if_attachdomain(void *dummy)
 727 {
 728         struct ifnet *ifp;
 729
 730         ifnet_lock();
 731         TAILQ_FOREACH(ifp, &ifnetlist, if_list)
 732                 if_attachdomain1(ifp);
 733         ifnet_unlock();
 734 }
 735 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_FIRST,
 736         if_attachdomain, NULL);
 737
 738 static void
 739 if_attachdomain1(struct ifnet *ifp)
 740 {
 741         struct domain *dp;
 742
 743         crit_enter();
 744
 745         /* address family dependent data region */
 746         bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
 747         SLIST_FOREACH(dp, &domains, dom_next)
 748                 if (dp->dom_ifattach)
 749                         ifp->if_afdata[dp->dom_family] =
 750                                 (*dp->dom_ifattach)(ifp);
 751         crit_exit();
 752 }
 753
 754 /*
 755  * Purge all addresses whose type is _not_ AF_LINK
 756  */
 757 static void
 758 if_purgeaddrs_nolink_dispatch(netmsg_t nmsg)
 759 {
 760         struct ifnet *ifp = nmsg->lmsg.u.ms_resultp;
 761         struct ifaddr_container *ifac, *next;
 762
 763         ASSERT_NETISR0;
 764
 765         /*
 766          * The ifaddr processing in the following loop will block,
 767          * however, this function is called in netisr0, in which
 768          * ifaddr list changes happen, so we don't care about the
 769          * blockness of the ifaddr processing here.
 770          */
 771         TAILQ_FOREACH_MUTABLE(ifac, &ifp->if_addrheads[mycpuid],
 772                               ifa_link, next) {
 773                 struct ifaddr *ifa = ifac->ifa;
 774
 775                 /* Ignore marker */
 776                 if (ifa->ifa_addr->sa_family == AF_UNSPEC)
 777                         continue;
 778
 779                 /* Leave link ifaddr as it is */
 780                 if (ifa->ifa_addr->sa_family == AF_LINK)
 781                         continue;
 782 #ifdef INET
 783                 /* XXX: Ugly!! ad hoc just for INET */
 784                 if (ifa->ifa_addr->sa_family == AF_INET) {
 785                         struct ifaliasreq ifr;
 786                         struct sockaddr_in saved_addr, saved_dst;
 787 #ifdef IFADDR_DEBUG_VERBOSE
 788                         int i;
 789
 790                         kprintf("purge in4 addr %p: ", ifa);
 791                         for (i = 0; i < ncpus; ++i) {
 792                                 kprintf("%d ",
 793                                     ifa->ifa_containers[i].ifa_refcnt);
 794                         }
 795                         kprintf("\n");
 796 #endif
 797
 798                         /* Save information for panic. */
 799                         memcpy(&saved_addr, ifa->ifa_addr, sizeof(saved_addr));
 800                         if (ifa->ifa_dstaddr != NULL) {
 801                                 memcpy(&saved_dst, ifa->ifa_dstaddr,
 802                                     sizeof(saved_dst));
 803                         } else {
 804                                 memset(&saved_dst, 0, sizeof(saved_dst));
 805                         }
 806
 807                         bzero(&ifr, sizeof ifr);
 808                         ifr.ifra_addr = *ifa->ifa_addr;
 809                         if (ifa->ifa_dstaddr)
 810                                 ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
 811                         if (in_control(SIOCDIFADDR, (caddr_t)&ifr, ifp,
 812                                        NULL) == 0)
 813                                 continue;
 814
 815                         /* MUST NOT HAPPEN */
 816                         panic("%s: in_control failed %x, dst %x", ifp->if_xname,
 817                             ntohl(saved_addr.sin_addr.s_addr),
 818                             ntohl(saved_dst.sin_addr.s_addr));
 819                 }
 820 #endif /* INET */
 821 #ifdef INET6
 822                 if (ifa->ifa_addr->sa_family == AF_INET6) {
 823 #ifdef IFADDR_DEBUG_VERBOSE
 824                         int i;
 825
 826                         kprintf("purge in6 addr %p: ", ifa);
 827                         for (i = 0; i < ncpus; ++i) {
 828                                 kprintf("%d ",
 829                                     ifa->ifa_containers[i].ifa_refcnt);
 830                         }
 831                         kprintf("\n");
 832 #endif
 833
 834                         in6_purgeaddr(ifa);
 835                         /* ifp_addrhead is already updated */
 836                         continue;
 837                 }
 838 #endif /* INET6 */
 839                 if_printf(ifp, "destroy ifaddr family %d\n",
 840                     ifa->ifa_addr->sa_family);
 841                 ifa_ifunlink(ifa, ifp);
 842                 ifa_destroy(ifa);
 843         }
 844
 845         netisr_replymsg(&nmsg->base, 0);
 846 }
 847
 848 void
 849 if_purgeaddrs_nolink(struct ifnet *ifp)
 850 {
 851         struct netmsg_base nmsg;
 852
 853         netmsg_init(&nmsg, NULL, &curthread->td_msgport, 0,
 854             if_purgeaddrs_nolink_dispatch);
 855         nmsg.lmsg.u.ms_resultp = ifp;
 856         netisr_domsg(&nmsg, 0);
 857 }
 858
 859 static void
 860 ifq_stage_detach_handler(netmsg_t nmsg)
 861 {
 862         struct ifaltq *ifq = nmsg->lmsg.u.ms_resultp;
 863         int q;
 864
 865         for (q = 0; q < ifq->altq_subq_cnt; ++q) {
 866                 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
 867                 struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid);
 868
 869                 if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED)
 870                         ifsq_stage_remove(&ifsubq_stage_heads[mycpuid], stage);
 871         }
 872         lwkt_replymsg(&nmsg->lmsg, 0);
 873 }
 874
 875 static void
 876 ifq_stage_detach(struct ifaltq *ifq)
 877 {
 878         struct netmsg_base base;
 879         int cpu;
 880
 881         netmsg_init(&base, NULL, &curthread->td_msgport, 0,
 882             ifq_stage_detach_handler);
 883         base.lmsg.u.ms_resultp = ifq;
 884
 885         /* XXX netisr_ncpus */
 886         for (cpu = 0; cpu < ncpus; ++cpu)
 887                 lwkt_domsg(netisr_cpuport(cpu), &base.lmsg, 0);
 888 }
 889
 890 struct netmsg_if_rtdel {
 891         struct netmsg_base      base;
 892         struct ifnet            *ifp;
 893 };
 894
 895 static void
 896 if_rtdel_dispatch(netmsg_t msg)
 897 {
 898         struct netmsg_if_rtdel *rmsg = (void *)msg;
 899         int i, cpu;
 900
 901         cpu = mycpuid;
 902         ASSERT_NETISR_NCPUS(cpu);
 903
 904         for (i = 1; i <= AF_MAX; i++) {
 905                 struct radix_node_head  *rnh;
 906
 907                 if ((rnh = rt_tables[cpu][i]) == NULL)
 908                         continue;
 909                 rnh->rnh_walktree(rnh, if_rtdel, rmsg->ifp);
 910         }
 911         netisr_forwardmsg(&msg->base, cpu + 1);
 912 }
 913
 914 /*
 915  * Detach an interface, removing it from the
 916  * list of "active" interfaces.
 917  */
 918 void
 919 if_detach(struct ifnet *ifp)
 920 {
 921         struct ifnet_array *old_ifnet_array;
 922         struct netmsg_if_rtdel msg;
 923         struct domain *dp;
 924         int q;
 925
 926         /* Announce that the interface is gone. */
 927         EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
 928         rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
 929         devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
 930
 931         /*
 932          * Remove this ifp from ifindex2inet, ifnet queue and ifnet
 933          * array before it is whacked.
 934          *
 935          * Protect ifindex2ifnet, ifnet queue and ifnet array changes
 936          * by ifnet lock, so that non-netisr threads could get a
 937          * consistent view.
 938          */
 939         ifnet_lock();
 940
 941         /*
 942          * Remove this ifp from ifindex2ifnet and maybe decrement if_index.
 943          */
 944         ifindex2ifnet[ifp->if_index] = NULL;
 945         while (if_index > 0 && ifindex2ifnet[if_index] == NULL)
 946                 if_index--;
 947
 948         /*
 949          * Remove this ifp from ifnet queue.
 950          */
 951         TAILQ_REMOVE(&ifnetlist, ifp, if_link);
 952
 953         /*
 954          * Remove this ifp from ifnet array.
 955          */
 956         /* Free old ifnet array after sync all netisrs */
 957         old_ifnet_array = ifnet_array;
 958         ifnet_array = ifnet_array_del(ifp, old_ifnet_array);
 959
 960         ifnet_unlock();
 961
 962         /*
 963          * Sync all netisrs so that the old ifnet array is no longer
 964          * accessed and we can free it safely later on.
 965          */
 966         netmsg_service_sync();
 967         ifnet_array_free(old_ifnet_array);
 968
 969         /*
 970          * Remove routes and flush queues.
 971          */
 972         crit_enter();
 973 #ifdef IFPOLL_ENABLE
 974         if (ifp->if_flags & IFF_NPOLLING)
 975                 ifpoll_deregister(ifp);
 976 #endif
 977         if_down(ifp);
 978
 979         /* Decrease the mbuf clusters/jclusters limits increased by us */
 980         if (ifp->if_nmbclusters > 0)
 981                 mcl_inclimit(-ifp->if_nmbclusters);
 982         if (ifp->if_nmbjclusters > 0)
 983                 mjcl_inclimit(-ifp->if_nmbjclusters);
 984
 985 #ifdef ALTQ
 986         if (ifq_is_enabled(&ifp->if_snd))
 987                 altq_disable(&ifp->if_snd);
 988         if (ifq_is_attached(&ifp->if_snd))
 989                 altq_detach(&ifp->if_snd);
 990 #endif
 991
 992         /*
 993          * Clean up all addresses.
 994          */
 995         ifp->if_lladdr = NULL;
 996
 997         if_purgeaddrs_nolink(ifp);
 998         if (!TAILQ_EMPTY(&ifp->if_addrheads[mycpuid])) {
 999                 struct ifaddr *ifa;
1000
1001                 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
1002                 KASSERT(ifa->ifa_addr->sa_family == AF_LINK,
1003                         ("non-link ifaddr is left on if_addrheads"));
1004
1005                 ifa_ifunlink(ifa, ifp);
1006                 ifa_destroy(ifa);
1007                 KASSERT(TAILQ_EMPTY(&ifp->if_addrheads[mycpuid]),
1008                         ("there are still ifaddrs left on if_addrheads"));
1009         }
1010
1011 #ifdef INET
1012         /*
1013          * Remove all IPv4 kernel structures related to ifp.
1014          */
1015         in_ifdetach(ifp);
1016 #endif
1017
1018 #ifdef INET6
1019         /*
1020          * Remove all IPv6 kernel structs related to ifp.  This should be done
1021          * before removing routing entries below, since IPv6 interface direct
1022          * routes are expected to be removed by the IPv6-specific kernel API.
1023          * Otherwise, the kernel will detect some inconsistency and bark it.
1024          */
1025         in6_ifdetach(ifp);
1026 #endif
1027
1028         /*
1029          * Delete all remaining routes using this interface
1030          */
1031         netmsg_init(&msg.base, NULL, &curthread->td_msgport, MSGF_PRIORITY,
1032             if_rtdel_dispatch);
1033         msg.ifp = ifp;
1034         netisr_domsg_global(&msg.base);
1035
1036         SLIST_FOREACH(dp, &domains, dom_next)
1037                 if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
1038                         (*dp->dom_ifdetach)(ifp,
1039                                 ifp->if_afdata[dp->dom_family]);
1040
1041         kfree(ifp->if_addrheads, M_IFADDR);
1042
1043         lwkt_synchronize_ipiqs("if_detach");
1044         ifq_stage_detach(&ifp->if_snd);
1045
1046         for (q = 0; q < ifp->if_snd.altq_subq_cnt; ++q) {
1047                 struct ifaltq_subque *ifsq = &ifp->if_snd.altq_subq[q];
1048
1049                 kfree(ifsq->ifsq_ifstart_nmsg, M_LWKTMSG);
1050                 kfree(ifsq->ifsq_stage, M_DEVBUF);
1051         }
1052         kfree(ifp->if_snd.altq_subq, M_DEVBUF);
1053
1054         kfree(ifp->if_data_pcpu, M_DEVBUF);
1055
1056         crit_exit();
1057 }
1058
1059 /*
1060  * Create interface group without members
1061  */
1062 struct ifg_group *
1063 if_creategroup(const char *groupname)
1064 {
1065         struct ifg_group        *ifg = NULL;
1066
1067         if ((ifg = (struct ifg_group *)kmalloc(sizeof(struct ifg_group),
1068             M_TEMP, M_NOWAIT)) == NULL)
1069                 return (NULL);
1070
1071         strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group));
1072         ifg->ifg_refcnt = 0;
1073         ifg->ifg_carp_demoted = 0;
1074         TAILQ_INIT(&ifg->ifg_members);
1075 #if NPF > 0
1076         pfi_attach_ifgroup(ifg);
1077 #endif
1078         TAILQ_INSERT_TAIL(&ifg_head, ifg, ifg_next);
1079
1080         return (ifg);
1081 }
1082
1083 /*
1084  * Add a group to an interface
1085  */
1086 int
1087 if_addgroup(struct ifnet *ifp, const char *groupname)
1088 {
1089         struct ifg_list         *ifgl;
1090         struct ifg_group        *ifg = NULL;
1091         struct ifg_member       *ifgm;
1092
1093         if (groupname[0] && groupname[strlen(groupname) - 1] >= '0' &&
1094             groupname[strlen(groupname) - 1] <= '9')
1095                 return (EINVAL);
1096
1097         TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1098                 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
1099                         return (EEXIST);
1100
1101         if ((ifgl = kmalloc(sizeof(*ifgl), M_TEMP, M_NOWAIT)) == NULL)
1102                 return (ENOMEM);
1103
1104         if ((ifgm = kmalloc(sizeof(*ifgm), M_TEMP, M_NOWAIT)) == NULL) {
1105                 kfree(ifgl, M_TEMP);
1106                 return (ENOMEM);
1107         }
1108
1109         TAILQ_FOREACH(ifg, &ifg_head, ifg_next)
1110                 if (!strcmp(ifg->ifg_group, groupname))
1111                         break;
1112
1113         if (ifg == NULL && (ifg = if_creategroup(groupname)) == NULL) {
1114                 kfree(ifgl, M_TEMP);
1115                 kfree(ifgm, M_TEMP);
1116                 return (ENOMEM);
1117         }
1118
1119         ifg->ifg_refcnt++;
1120         ifgl->ifgl_group = ifg;
1121         ifgm->ifgm_ifp = ifp;
1122
1123         TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next);
1124         TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next);
1125
1126 #if NPF > 0
1127         pfi_group_change(groupname);
1128 #endif
1129
1130         return (0);
1131 }
1132
1133 /*
1134  * Remove a group from an interface
1135  */
1136 int
1137 if_delgroup(struct ifnet *ifp, const char *groupname)
1138 {
1139         struct ifg_list         *ifgl;
1140         struct ifg_member       *ifgm;
1141
1142         TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1143                 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
1144                         break;
1145         if (ifgl == NULL)
1146                 return (ENOENT);
1147
1148         TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
1149
1150         TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
1151                 if (ifgm->ifgm_ifp == ifp)
1152                         break;
1153
1154         if (ifgm != NULL) {
1155                 TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next);
1156                 kfree(ifgm, M_TEMP);
1157         }
1158
1159         if (--ifgl->ifgl_group->ifg_refcnt == 0) {
1160                 TAILQ_REMOVE(&ifg_head, ifgl->ifgl_group, ifg_next);
1161 #if NPF > 0
1162                 pfi_detach_ifgroup(ifgl->ifgl_group);
1163 #endif
1164                 kfree(ifgl->ifgl_group, M_TEMP);
1165         }
1166
1167         kfree(ifgl, M_TEMP);
1168
1169 #if NPF > 0
1170         pfi_group_change(groupname);
1171 #endif
1172
1173         return (0);
1174 }
1175
1176 /*
1177  * Stores all groups from an interface in memory pointed
1178  * to by data
1179  */
1180 int
1181 if_getgroup(caddr_t data, struct ifnet *ifp)
1182 {
1183         int                      len, error;
1184         struct ifg_list         *ifgl;
1185         struct ifg_req           ifgrq, *ifgp;
1186         struct ifgroupreq       *ifgr = (struct ifgroupreq *)data;
1187
1188         if (ifgr->ifgr_len == 0) {
1189                 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1190                         ifgr->ifgr_len += sizeof(struct ifg_req);
1191                 return (0);
1192         }
1193
1194         len = ifgr->ifgr_len;
1195         ifgp = ifgr->ifgr_groups;
1196         TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1197                 if (len < sizeof(ifgrq))
1198                         return (EINVAL);
1199                 bzero(&ifgrq, sizeof ifgrq);
1200                 strlcpy(ifgrq.ifgrq_group, ifgl->ifgl_group->ifg_group,
1201                     sizeof(ifgrq.ifgrq_group));
1202                 if ((error = copyout((caddr_t)&ifgrq, (caddr_t)ifgp,
1203                     sizeof(struct ifg_req))))
1204                         return (error);
1205                 len -= sizeof(ifgrq);
1206                 ifgp++;
1207         }
1208
1209         return (0);
1210 }
1211
1212 /*
1213  * Stores all members of a group in memory pointed to by data
1214  */
1215 int
1216 if_getgroupmembers(caddr_t data)
1217 {
1218         struct ifgroupreq       *ifgr = (struct ifgroupreq *)data;
1219         struct ifg_group        *ifg;
1220         struct ifg_member       *ifgm;
1221         struct ifg_req           ifgrq, *ifgp;
1222         int                      len, error;
1223
1224         TAILQ_FOREACH(ifg, &ifg_head, ifg_next)
1225                 if (!strcmp(ifg->ifg_group, ifgr->ifgr_name))
1226                         break;
1227         if (ifg == NULL)
1228                 return (ENOENT);
1229
1230         if (ifgr->ifgr_len == 0) {
1231                 TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
1232                         ifgr->ifgr_len += sizeof(ifgrq);
1233                 return (0);
1234         }
1235
1236         len = ifgr->ifgr_len;
1237         ifgp = ifgr->ifgr_groups;
1238         TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
1239                 if (len < sizeof(ifgrq))
1240                         return (EINVAL);
1241                 bzero(&ifgrq, sizeof ifgrq);
1242                 strlcpy(ifgrq.ifgrq_member, ifgm->ifgm_ifp->if_xname,
1243                     sizeof(ifgrq.ifgrq_member));
1244                 if ((error = copyout((caddr_t)&ifgrq, (caddr_t)ifgp,
1245                     sizeof(struct ifg_req))))
1246                         return (error);
1247                 len -= sizeof(ifgrq);
1248                 ifgp++;
1249         }
1250
1251         return (0);
1252 }
1253
1254 /*
1255  * Delete Routes for a Network Interface
1256  *
1257  * Called for each routing entry via the rnh->rnh_walktree() call above
1258  * to delete all route entries referencing a detaching network interface.
1259  *
1260  * Arguments:
1261  *      rn      pointer to node in the routing table
1262  *      arg     argument passed to rnh->rnh_walktree() - detaching interface
1263  *
1264  * Returns:
1265  *      0       successful
1266  *      errno   failed - reason indicated
1267  *
1268  */
1269 static int
1270 if_rtdel(struct radix_node *rn, void *arg)
1271 {
1272         struct rtentry  *rt = (struct rtentry *)rn;
1273         struct ifnet    *ifp = arg;
1274         int             err;
1275
1276         if (rt->rt_ifp == ifp) {
1277
1278                 /*
1279                  * Protect (sorta) against walktree recursion problems
1280                  * with cloned routes
1281                  */
1282                 if (!(rt->rt_flags & RTF_UP))
1283                         return (0);
1284
1285                 err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1286                                 rt_mask(rt), rt->rt_flags,
1287                                 NULL);
1288                 if (err) {
1289                         log(LOG_WARNING, "if_rtdel: error %d\n", err);
1290                 }
1291         }
1292
1293         return (0);
1294 }
1295
1296 static __inline boolean_t
1297 ifa_prefer(const struct ifaddr *cur_ifa, const struct ifaddr *old_ifa)
1298 {
1299         if (old_ifa == NULL)
1300                 return TRUE;
1301
1302         if ((old_ifa->ifa_ifp->if_flags & IFF_UP) == 0 &&
1303             (cur_ifa->ifa_ifp->if_flags & IFF_UP))
1304                 return TRUE;
1305         if ((old_ifa->ifa_flags & IFA_ROUTE) == 0 &&
1306             (cur_ifa->ifa_flags & IFA_ROUTE))
1307                 return TRUE;
1308         return FALSE;
1309 }
1310
1311 /*
1312  * Locate an interface based on a complete address.
1313  */
1314 struct ifaddr *
1315 ifa_ifwithaddr(struct sockaddr *addr)
1316 {
1317         const struct ifnet_array *arr;
1318         int i;
1319
1320         arr = ifnet_array_get();
1321         for (i = 0; i < arr->ifnet_count; ++i) {
1322                 struct ifnet *ifp = arr->ifnet_arr[i];
1323                 struct ifaddr_container *ifac;
1324
1325                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1326                         struct ifaddr *ifa = ifac->ifa;
1327
1328                         if (ifa->ifa_addr->sa_family != addr->sa_family)
1329                                 continue;
1330                         if (sa_equal(addr, ifa->ifa_addr))
1331                                 return (ifa);
1332                         if ((ifp->if_flags & IFF_BROADCAST) &&
1333                             ifa->ifa_broadaddr &&
1334                             /* IPv6 doesn't have broadcast */
1335                             ifa->ifa_broadaddr->sa_len != 0 &&
1336                             sa_equal(ifa->ifa_broadaddr, addr))
1337                                 return (ifa);
1338                 }
1339         }
1340         return (NULL);
1341 }
1342
1343 /*
1344  * Locate the point to point interface with a given destination address.
1345  */
1346 struct ifaddr *
1347 ifa_ifwithdstaddr(struct sockaddr *addr)
1348 {
1349         const struct ifnet_array *arr;
1350         int i;
1351
1352         arr = ifnet_array_get();
1353         for (i = 0; i < arr->ifnet_count; ++i) {
1354                 struct ifnet *ifp = arr->ifnet_arr[i];
1355                 struct ifaddr_container *ifac;
1356
1357                 if (!(ifp->if_flags & IFF_POINTOPOINT))
1358                         continue;
1359
1360                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1361                         struct ifaddr *ifa = ifac->ifa;
1362
1363                         if (ifa->ifa_addr->sa_family != addr->sa_family)
1364                                 continue;
1365                         if (ifa->ifa_dstaddr &&
1366                             sa_equal(addr, ifa->ifa_dstaddr))
1367                                 return (ifa);
1368                 }
1369         }
1370         return (NULL);
1371 }
1372
1373 /*
1374  * Find an interface on a specific network.  If many, choice
1375  * is most specific found.
1376  */
1377 struct ifaddr *
1378 ifa_ifwithnet(struct sockaddr *addr)
1379 {
1380         struct ifaddr *ifa_maybe = NULL;
1381         u_int af = addr->sa_family;
1382         char *addr_data = addr->sa_data, *cplim;
1383         const struct ifnet_array *arr;
1384         int i;
1385
1386         /*
1387          * AF_LINK addresses can be looked up directly by their index number,
1388          * so do that if we can.
1389          */
1390         if (af == AF_LINK) {
1391                 struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
1392
1393                 if (sdl->sdl_index && sdl->sdl_index <= if_index)
1394                         return (ifindex2ifnet[sdl->sdl_index]->if_lladdr);
1395         }
1396
1397         /*
1398          * Scan though each interface, looking for ones that have
1399          * addresses in this address family.
1400          */
1401         arr = ifnet_array_get();
1402         for (i = 0; i < arr->ifnet_count; ++i) {
1403                 struct ifnet *ifp = arr->ifnet_arr[i];
1404                 struct ifaddr_container *ifac;
1405
1406                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1407                         struct ifaddr *ifa = ifac->ifa;
1408                         char *cp, *cp2, *cp3;
1409
1410                         if (ifa->ifa_addr->sa_family != af)
1411 next:                           continue;
1412                         if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT) {
1413                                 /*
1414                                  * This is a bit broken as it doesn't
1415                                  * take into account that the remote end may
1416                                  * be a single node in the network we are
1417                                  * looking for.
1418                                  * The trouble is that we don't know the
1419                                  * netmask for the remote end.
1420                                  */
1421                                 if (ifa->ifa_dstaddr != NULL &&
1422                                     sa_equal(addr, ifa->ifa_dstaddr))
1423                                         return (ifa);
1424                         } else {
1425                                 /*
1426                                  * if we have a special address handler,
1427                                  * then use it instead of the generic one.
1428                                  */
1429                                 if (ifa->ifa_claim_addr) {
1430                                         if ((*ifa->ifa_claim_addr)(ifa, addr)) {
1431                                                 return (ifa);
1432                                         } else {
1433                                                 continue;
1434                                         }
1435                                 }
1436
1437                                 /*
1438                                  * Scan all the bits in the ifa's address.
1439                                  * If a bit dissagrees with what we are
1440                                  * looking for, mask it with the netmask
1441                                  * to see if it really matters.
1442                                  * (A byte at a time)
1443                                  */
1444                                 if (ifa->ifa_netmask == 0)
1445                                         continue;
1446                                 cp = addr_data;
1447                                 cp2 = ifa->ifa_addr->sa_data;
1448                                 cp3 = ifa->ifa_netmask->sa_data;
1449                                 cplim = ifa->ifa_netmask->sa_len +
1450                                         (char *)ifa->ifa_netmask;
1451                                 while (cp3 < cplim)
1452                                         if ((*cp++ ^ *cp2++) & *cp3++)
1453                                                 goto next; /* next address! */
1454                                 /*
1455                                  * If the netmask of what we just found
1456                                  * is more specific than what we had before
1457                                  * (if we had one) then remember the new one
1458                                  * before continuing to search for an even
1459                                  * better one.  If the netmasks are equal,
1460                                  * we prefer the this ifa based on the result
1461                                  * of ifa_prefer().
1462                                  */
1463                                 if (ifa_maybe == NULL ||
1464                                     rn_refines((char *)ifa->ifa_netmask,
1465                                         (char *)ifa_maybe->ifa_netmask) ||
1466                                     (sa_equal(ifa_maybe->ifa_netmask,
1467                                         ifa->ifa_netmask) &&
1468                                      ifa_prefer(ifa, ifa_maybe)))
1469                                         ifa_maybe = ifa;
1470                         }
1471                 }
1472         }
1473         return (ifa_maybe);
1474 }
1475
1476 /*
1477  * Find an interface address specific to an interface best matching
1478  * a given address.
1479  */
1480 struct ifaddr *
1481 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1482 {
1483         struct ifaddr_container *ifac;
1484         char *cp, *cp2, *cp3;
1485         char *cplim;
1486         struct ifaddr *ifa_maybe = NULL;
1487         u_int af = addr->sa_family;
1488
1489         if (af >= AF_MAX)
1490                 return (0);
1491         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1492                 struct ifaddr *ifa = ifac->ifa;
1493
1494                 if (ifa->ifa_addr->sa_family != af)
1495                         continue;
1496                 if (ifa_maybe == NULL)
1497                         ifa_maybe = ifa;
1498                 if (ifa->ifa_netmask == NULL) {
1499                         if (sa_equal(addr, ifa->ifa_addr) ||
1500                             (ifa->ifa_dstaddr != NULL &&
1501                              sa_equal(addr, ifa->ifa_dstaddr)))
1502                                 return (ifa);
1503                         continue;
1504                 }
1505                 if (ifp->if_flags & IFF_POINTOPOINT) {
1506                         if (sa_equal(addr, ifa->ifa_dstaddr))
1507                                 return (ifa);
1508                 } else {
1509                         cp = addr->sa_data;
1510                         cp2 = ifa->ifa_addr->sa_data;
1511                         cp3 = ifa->ifa_netmask->sa_data;
1512                         cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1513                         for (; cp3 < cplim; cp3++)
1514                                 if ((*cp++ ^ *cp2++) & *cp3)
1515                                         break;
1516                         if (cp3 == cplim)
1517                                 return (ifa);
1518                 }
1519         }
1520         return (ifa_maybe);
1521 }
1522
1523 /*
1524  * Default action when installing a route with a Link Level gateway.
1525  * Lookup an appropriate real ifa to point to.
1526  * This should be moved to /sys/net/link.c eventually.
1527  */
1528 static void
1529 link_rtrequest(int cmd, struct rtentry *rt)
1530 {
1531         struct ifaddr *ifa;
1532         struct sockaddr *dst;
1533         struct ifnet *ifp;
1534
1535         if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL ||
1536             (ifp = ifa->ifa_ifp) == NULL || (dst = rt_key(rt)) == NULL)
1537                 return;
1538         ifa = ifaof_ifpforaddr(dst, ifp);
1539         if (ifa != NULL) {
1540                 IFAFREE(rt->rt_ifa);
1541                 IFAREF(ifa);
1542                 rt->rt_ifa = ifa;
1543                 if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1544                         ifa->ifa_rtrequest(cmd, rt);
1545         }
1546 }
1547
1548 struct netmsg_ifroute {
1549         struct netmsg_base      base;
1550         struct ifnet            *ifp;
1551         int                     flag;
1552         int                     fam;
1553 };
1554
1555 /*
1556  * Mark an interface down and notify protocols of the transition.
1557  */
1558 static void
1559 if_unroute_dispatch(netmsg_t nmsg)
1560 {
1561         struct netmsg_ifroute *msg = (struct netmsg_ifroute *)nmsg;
1562         struct ifnet *ifp = msg->ifp;
1563         int flag = msg->flag, fam = msg->fam;
1564         struct ifaddr_container *ifac;
1565
1566         ASSERT_NETISR0;
1567
1568         ifp->if_flags &= ~flag;
1569         getmicrotime(&ifp->if_lastchange);
1570         /*
1571          * The ifaddr processing in the following loop will block,
1572          * however, this function is called in netisr0, in which
1573          * ifaddr list changes happen, so we don't care about the
1574          * blockness of the ifaddr processing here.
1575          */
1576         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1577                 struct ifaddr *ifa = ifac->ifa;
1578
1579                 /* Ignore marker */
1580                 if (ifa->ifa_addr->sa_family == AF_UNSPEC)
1581                         continue;
1582
1583                 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1584                         kpfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1585         }
1586         ifq_purge_all(&ifp->if_snd);
1587         rt_ifmsg(ifp);
1588
1589         netisr_replymsg(&nmsg->base, 0);
1590 }
1591
1592 void
1593 if_unroute(struct ifnet *ifp, int flag, int fam)
1594 {
1595         struct netmsg_ifroute msg;
1596
1597         netmsg_init(&msg.base, NULL, &curthread->td_msgport, 0,
1598             if_unroute_dispatch);
1599         msg.ifp = ifp;
1600         msg.flag = flag;
1601         msg.fam = fam;
1602         netisr_domsg(&msg.base, 0);
1603 }
1604
1605 /*
1606  * Mark an interface up and notify protocols of the transition.
1607  */
1608 static void
1609 if_route_dispatch(netmsg_t nmsg)
1610 {
1611         struct netmsg_ifroute *msg = (struct netmsg_ifroute *)nmsg;
1612         struct ifnet *ifp = msg->ifp;
1613         int flag = msg->flag, fam = msg->fam;
1614         struct ifaddr_container *ifac;
1615
1616         ASSERT_NETISR0;
1617
1618         ifq_purge_all(&ifp->if_snd);
1619         ifp->if_flags |= flag;
1620         getmicrotime(&ifp->if_lastchange);
1621         /*
1622          * The ifaddr processing in the following loop will block,
1623          * however, this function is called in netisr0, in which
1624          * ifaddr list changes happen, so we don't care about the
1625          * blockness of the ifaddr processing here.
1626          */
1627         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1628                 struct ifaddr *ifa = ifac->ifa;
1629
1630                 /* Ignore marker */
1631                 if (ifa->ifa_addr->sa_family == AF_UNSPEC)
1632                         continue;
1633
1634                 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1635                         kpfctlinput(PRC_IFUP, ifa->ifa_addr);
1636         }
1637         rt_ifmsg(ifp);
1638 #ifdef INET6
1639         in6_if_up(ifp);
1640 #endif
1641
1642         netisr_replymsg(&nmsg->base, 0);
1643 }
1644
1645 void
1646 if_route(struct ifnet *ifp, int flag, int fam)
1647 {
1648         struct netmsg_ifroute msg;
1649
1650         netmsg_init(&msg.base, NULL, &curthread->td_msgport, 0,
1651             if_route_dispatch);
1652         msg.ifp = ifp;
1653         msg.flag = flag;
1654         msg.fam = fam;
1655         netisr_domsg(&msg.base, 0);
1656 }
1657
1658 /*
1659  * Mark an interface down and notify protocols of the transition.  An
1660  * interface going down is also considered to be a synchronizing event.
1661  * We must ensure that all packet processing related to the interface
1662  * has completed before we return so e.g. the caller can free the ifnet
1663  * structure that the mbufs may be referencing.
1664  *
1665  * NOTE: must be called at splnet or eqivalent.
1666  */
1667 void
1668 if_down(struct ifnet *ifp)
1669 {
1670         if_unroute(ifp, IFF_UP, AF_UNSPEC);
1671         netmsg_service_sync();
1672 }
1673
1674 /*
1675  * Mark an interface up and notify protocols of
1676  * the transition.
1677  * NOTE: must be called at splnet or eqivalent.
1678  */
1679 void
1680 if_up(struct ifnet *ifp)
1681 {
1682         if_route(ifp, IFF_UP, AF_UNSPEC);
1683 }
1684
1685 /*
1686  * Process a link state change.
1687  * NOTE: must be called at splsoftnet or equivalent.
1688  */
1689 void
1690 if_link_state_change(struct ifnet *ifp)
1691 {
1692         int link_state = ifp->if_link_state;
1693
1694         rt_ifmsg(ifp);
1695         devctl_notify("IFNET", ifp->if_xname,
1696             (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL);
1697 }
1698
1699 /*
1700  * Handle interface watchdog timer routines.  Called
1701  * from softclock, we decrement timers (if set) and
1702  * call the appropriate interface routine on expiration.
1703  */
1704 static void
1705 if_slowtimo_dispatch(netmsg_t nmsg)
1706 {
1707         struct globaldata *gd = mycpu;
1708         const struct ifnet_array *arr;
1709         int i;
1710
1711         ASSERT_NETISR0;
1712
1713         crit_enter_gd(gd);
1714         lwkt_replymsg(&nmsg->lmsg, 0);  /* reply ASAP */
1715         crit_exit_gd(gd);
1716
1717         arr = ifnet_array_get();
1718         for (i = 0; i < arr->ifnet_count; ++i) {
1719                 struct ifnet *ifp = arr->ifnet_arr[i];
1720
1721                 crit_enter_gd(gd);
1722
1723                 if (if_stats_compat) {
1724                         IFNET_STAT_GET(ifp, ipackets, ifp->if_ipackets);
1725                         IFNET_STAT_GET(ifp, ierrors, ifp->if_ierrors);
1726                         IFNET_STAT_GET(ifp, opackets, ifp->if_opackets);
1727                         IFNET_STAT_GET(ifp, oerrors, ifp->if_oerrors);
1728                         IFNET_STAT_GET(ifp, collisions, ifp->if_collisions);
1729                         IFNET_STAT_GET(ifp, ibytes, ifp->if_ibytes);
1730                         IFNET_STAT_GET(ifp, obytes, ifp->if_obytes);
1731                         IFNET_STAT_GET(ifp, imcasts, ifp->if_imcasts);
1732                         IFNET_STAT_GET(ifp, omcasts, ifp->if_omcasts);
1733                         IFNET_STAT_GET(ifp, iqdrops, ifp->if_iqdrops);
1734                         IFNET_STAT_GET(ifp, noproto, ifp->if_noproto);
1735                         IFNET_STAT_GET(ifp, oqdrops, ifp->if_oqdrops);
1736                 }
1737
1738                 if (ifp->if_timer == 0 || --ifp->if_timer) {
1739                         crit_exit_gd(gd);
1740                         continue;
1741                 }
1742                 if (ifp->if_watchdog) {
1743                         if (ifnet_tryserialize_all(ifp)) {
1744                                 (*ifp->if_watchdog)(ifp);
1745                                 ifnet_deserialize_all(ifp);
1746                         } else {
1747                                 /* try again next timeout */
1748                                 ++ifp->if_timer;
1749                         }
1750                 }
1751
1752                 crit_exit_gd(gd);
1753         }
1754
1755         callout_reset(&if_slowtimo_timer, hz / IFNET_SLOWHZ, if_slowtimo, NULL);
1756 }
1757
1758 static void
1759 if_slowtimo(void *arg __unused)
1760 {
1761         struct lwkt_msg *lmsg = &if_slowtimo_netmsg.lmsg;
1762
1763         KASSERT(mycpuid == 0, ("not on cpu0"));
1764         crit_enter();
1765         if (lmsg->ms_flags & MSGF_DONE)
1766                 lwkt_sendmsg_oncpu(netisr_cpuport(0), lmsg);
1767         crit_exit();
1768 }
1769
1770 /*
1771  * Map interface name to
1772  * interface structure pointer.
1773  */
1774 struct ifnet *
1775 ifunit(const char *name)
1776 {
1777         struct ifnet *ifp;
1778
1779         /*
1780          * Search all the interfaces for this name/number
1781          */
1782         KASSERT(mtx_owned(&ifnet_mtx), ("ifnet is not locked"));
1783
1784         TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
1785                 if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1786                         break;
1787         }
1788         return (ifp);
1789 }
1790
1791 struct ifnet *
1792 ifunit_netisr(const char *name)
1793 {
1794         const struct ifnet_array *arr;
1795         int i;
1796
1797         /*
1798          * Search all the interfaces for this name/number
1799          */
1800
1801         arr = ifnet_array_get();
1802         for (i = 0; i < arr->ifnet_count; ++i) {
1803                 struct ifnet *ifp = arr->ifnet_arr[i];
1804
1805                 if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1806                         return ifp;
1807         }
1808         return NULL;
1809 }
1810
1811 /*
1812  * Interface ioctls.
1813  */
1814 int
1815 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct ucred *cred)
1816 {
1817         struct ifnet *ifp;
1818         struct ifreq *ifr;
1819         struct ifstat *ifs;
1820         int error, do_ifup = 0;
1821         short oif_flags;
1822         int new_flags;
1823         size_t namelen, onamelen;
1824         char new_name[IFNAMSIZ];
1825         struct ifaddr *ifa;
1826         struct sockaddr_dl *sdl;
1827
1828         switch (cmd) {
1829         case SIOCGIFCONF:
1830         case OSIOCGIFCONF:
1831                 return (ifconf(cmd, data, cred));
1832         default:
1833                 break;
1834         }
1835
1836         ifr = (struct ifreq *)data;
1837
1838         switch (cmd) {
1839         case SIOCIFCREATE:
1840         case SIOCIFCREATE2:
1841                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1842                         return (error);
1843                 return (if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name),
1844                         cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL));
1845         case SIOCIFDESTROY:
1846                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1847                         return (error);
1848                 return (if_clone_destroy(ifr->ifr_name));
1849         case SIOCIFGCLONERS:
1850                 return (if_clone_list((struct if_clonereq *)data));
1851         default:
1852                 break;
1853         }
1854
1855         /*
1856          * Nominal ioctl through interface, lookup the ifp and obtain a
1857          * lock to serialize the ifconfig ioctl operation.
1858          */
1859         ifnet_lock();
1860
1861         ifp = ifunit(ifr->ifr_name);
1862         if (ifp == NULL) {
1863                 ifnet_unlock();
1864                 return (ENXIO);
1865         }
1866         error = 0;
1867
1868         switch (cmd) {
1869         case SIOCGIFINDEX:
1870                 ifr->ifr_index = ifp->if_index;
1871                 break;
1872
1873         case SIOCGIFFLAGS:
1874                 ifr->ifr_flags = ifp->if_flags;
1875                 ifr->ifr_flagshigh = ifp->if_flags >> 16;
1876                 break;
1877
1878         case SIOCGIFCAP:
1879                 ifr->ifr_reqcap = ifp->if_capabilities;
1880                 ifr->ifr_curcap = ifp->if_capenable;
1881                 break;
1882
1883         case SIOCGIFMETRIC:
1884                 ifr->ifr_metric = ifp->if_metric;
1885                 break;
1886
1887         case SIOCGIFMTU:
1888                 ifr->ifr_mtu = ifp->if_mtu;
1889                 break;
1890
1891         case SIOCGIFTSOLEN:
1892                 ifr->ifr_tsolen = ifp->if_tsolen;
1893                 break;
1894
1895         case SIOCGIFDATA:
1896                 error = copyout((caddr_t)&ifp->if_data, ifr->ifr_data,
1897                                 sizeof(ifp->if_data));
1898                 break;
1899
1900         case SIOCGIFPHYS:
1901                 ifr->ifr_phys = ifp->if_physical;
1902                 break;
1903
1904         case SIOCGIFPOLLCPU:
1905                 ifr->ifr_pollcpu = -1;
1906                 break;
1907
1908         case SIOCSIFPOLLCPU:
1909                 break;
1910
1911         case SIOCSIFFLAGS:
1912                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1913                 if (error)
1914                         break;
1915                 new_flags = (ifr->ifr_flags & 0xffff) |
1916                     (ifr->ifr_flagshigh << 16);
1917                 if (ifp->if_flags & IFF_SMART) {
1918                         /* Smart drivers twiddle their own routes */
1919                 } else if (ifp->if_flags & IFF_UP &&
1920                     (new_flags & IFF_UP) == 0) {
1921                         if_down(ifp);
1922                 } else if (new_flags & IFF_UP &&
1923                     (ifp->if_flags & IFF_UP) == 0) {
1924                         do_ifup = 1;
1925                 }
1926
1927 #ifdef IFPOLL_ENABLE
1928                 if ((new_flags ^ ifp->if_flags) & IFF_NPOLLING) {
1929                         if (new_flags & IFF_NPOLLING)
1930                                 ifpoll_register(ifp);
1931                         else
1932                                 ifpoll_deregister(ifp);
1933                 }
1934 #endif
1935
1936                 ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
1937                         (new_flags &~ IFF_CANTCHANGE);
1938                 if (new_flags & IFF_PPROMISC) {
1939                         /* Permanently promiscuous mode requested */
1940                         ifp->if_flags |= IFF_PROMISC;
1941                 } else if (ifp->if_pcount == 0) {
1942                         ifp->if_flags &= ~IFF_PROMISC;
1943                 }
1944                 if (ifp->if_ioctl) {
1945                         ifnet_serialize_all(ifp);
1946                         ifp->if_ioctl(ifp, cmd, data, cred);
1947                         ifnet_deserialize_all(ifp);
1948                 }
1949                 if (do_ifup)
1950                         if_up(ifp);
1951                 getmicrotime(&ifp->if_lastchange);
1952                 break;
1953
1954         case SIOCSIFCAP:
1955                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1956                 if (error)
1957                         break;
1958                 if (ifr->ifr_reqcap & ~ifp->if_capabilities) {
1959                         error = EINVAL;
1960                         break;
1961                 }
1962                 ifnet_serialize_all(ifp);
1963                 ifp->if_ioctl(ifp, cmd, data, cred);
1964                 ifnet_deserialize_all(ifp);
1965                 break;
1966
1967         case SIOCSIFNAME:
1968                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1969                 if (error)
1970                         break;
1971                 error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
1972                 if (error)
1973                         break;
1974                 if (new_name[0] == '\0') {
1975                         error = EINVAL;
1976                         break;
1977                 }
1978                 if (ifunit(new_name) != NULL) {
1979                         error = EEXIST;
1980                         break;
1981                 }
1982
1983                 EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
1984
1985                 /* Announce the departure of the interface. */
1986                 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1987
1988                 strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
1989                 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
1990                 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1991                 namelen = strlen(new_name);
1992                 onamelen = sdl->sdl_nlen;
1993                 /*
1994                  * Move the address if needed.  This is safe because we
1995                  * allocate space for a name of length IFNAMSIZ when we
1996                  * create this in if_attach().
1997                  */
1998                 if (namelen != onamelen) {
1999                         bcopy(sdl->sdl_data + onamelen,
2000                             sdl->sdl_data + namelen, sdl->sdl_alen);
2001                 }
2002                 bcopy(new_name, sdl->sdl_data, namelen);
2003                 sdl->sdl_nlen = namelen;
2004                 sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
2005                 bzero(sdl->sdl_data, onamelen);
2006                 while (namelen != 0)
2007                         sdl->sdl_data[--namelen] = 0xff;
2008
2009                 EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
2010
2011                 /* Announce the return of the interface. */
2012                 rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
2013                 break;
2014
2015         case SIOCSIFMETRIC:
2016                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2017                 if (error)
2018                         break;
2019                 ifp->if_metric = ifr->ifr_metric;
2020                 getmicrotime(&ifp->if_lastchange);
2021                 break;
2022
2023         case SIOCSIFPHYS:
2024                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2025                 if (error)
2026                         break;
2027                 if (ifp->if_ioctl == NULL) {
2028                         error = EOPNOTSUPP;
2029                         break;
2030                 }
2031                 ifnet_serialize_all(ifp);
2032                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2033                 ifnet_deserialize_all(ifp);
2034                 if (error == 0)
2035                         getmicrotime(&ifp->if_lastchange);
2036                 break;
2037
2038         case SIOCSIFMTU:
2039         {
2040                 u_long oldmtu = ifp->if_mtu;
2041
2042                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2043                 if (error)
2044                         break;
2045                 if (ifp->if_ioctl == NULL) {
2046                         error = EOPNOTSUPP;
2047                         break;
2048                 }
2049                 if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU) {
2050                         error = EINVAL;
2051                         break;
2052                 }
2053                 ifnet_serialize_all(ifp);
2054                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2055                 ifnet_deserialize_all(ifp);
2056                 if (error == 0) {
2057                         getmicrotime(&ifp->if_lastchange);
2058                         rt_ifmsg(ifp);
2059                 }
2060                 /*
2061                  * If the link MTU changed, do network layer specific procedure.
2062                  */
2063                 if (ifp->if_mtu != oldmtu) {
2064 #ifdef INET6
2065                         nd6_setmtu(ifp);
2066 #endif
2067                 }
2068                 break;
2069         }
2070
2071         case SIOCSIFTSOLEN:
2072                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2073                 if (error)
2074                         break;
2075
2076                 /* XXX need driver supplied upper limit */
2077                 if (ifr->ifr_tsolen <= 0) {
2078                         error = EINVAL;
2079                         break;
2080                 }
2081                 ifp->if_tsolen = ifr->ifr_tsolen;
2082                 break;
2083
2084         case SIOCADDMULTI:
2085         case SIOCDELMULTI:
2086                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2087                 if (error)
2088                         break;
2089
2090                 /* Don't allow group membership on non-multicast interfaces. */
2091                 if ((ifp->if_flags & IFF_MULTICAST) == 0) {
2092                         error = EOPNOTSUPP;
2093                         break;
2094                 }
2095
2096                 /* Don't let users screw up protocols' entries. */
2097                 if (ifr->ifr_addr.sa_family != AF_LINK) {
2098                         error = EINVAL;
2099                         break;
2100                 }
2101
2102                 if (cmd == SIOCADDMULTI) {
2103                         struct ifmultiaddr *ifma;
2104                         error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
2105                 } else {
2106                         error = if_delmulti(ifp, &ifr->ifr_addr);
2107                 }
2108                 if (error == 0)
2109                         getmicrotime(&ifp->if_lastchange);
2110                 break;
2111
2112         case SIOCSIFPHYADDR:
2113         case SIOCDIFPHYADDR:
2114 #ifdef INET6
2115         case SIOCSIFPHYADDR_IN6:
2116 #endif
2117         case SIOCSLIFPHYADDR:
2118         case SIOCSIFMEDIA:
2119         case SIOCSIFGENERIC:
2120                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2121                 if (error)
2122                         break;
2123                 if (ifp->if_ioctl == 0) {
2124                         error = EOPNOTSUPP;
2125                         break;
2126                 }
2127                 ifnet_serialize_all(ifp);
2128                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2129                 ifnet_deserialize_all(ifp);
2130                 if (error == 0)
2131                         getmicrotime(&ifp->if_lastchange);
2132                 break;
2133
2134         case SIOCGIFSTATUS:
2135                 ifs = (struct ifstat *)data;
2136                 ifs->ascii[0] = '\0';
2137                 /* fall through */
2138         case SIOCGIFPSRCADDR:
2139         case SIOCGIFPDSTADDR:
2140         case SIOCGLIFPHYADDR:
2141         case SIOCGIFMEDIA:
2142         case SIOCGIFGENERIC:
2143                 if (ifp->if_ioctl == NULL) {
2144                         error = EOPNOTSUPP;
2145                         break;
2146                 }
2147                 ifnet_serialize_all(ifp);
2148                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2149                 ifnet_deserialize_all(ifp);
2150                 break;
2151
2152         case SIOCSIFLLADDR:
2153                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2154                 if (error)
2155                         break;
2156                 error = if_setlladdr(ifp, ifr->ifr_addr.sa_data,
2157                                      ifr->ifr_addr.sa_len);
2158                 EVENTHANDLER_INVOKE(iflladdr_event, ifp);
2159                 break;
2160
2161         default:
2162                 oif_flags = ifp->if_flags;
2163                 if (so->so_proto == 0) {
2164                         error = EOPNOTSUPP;
2165                         break;
2166                 }
2167                 error = so_pru_control_direct(so, cmd, data, ifp);
2168
2169                 if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
2170 #ifdef INET6
2171                         DELAY(100);/* XXX: temporary workaround for fxp issue*/
2172                         if (ifp->if_flags & IFF_UP) {
2173                                 crit_enter();
2174                                 in6_if_up(ifp);
2175                                 crit_exit();
2176                         }
2177 #endif
2178                 }
2179                 break;
2180         }
2181
2182         ifnet_unlock();
2183         return (error);
2184 }
2185
2186 /*
2187  * Set/clear promiscuous mode on interface ifp based on the truth value
2188  * of pswitch.  The calls are reference counted so that only the first
2189  * "on" request actually has an effect, as does the final "off" request.
2190  * Results are undefined if the "off" and "on" requests are not matched.
2191  */
2192 int
2193 ifpromisc(struct ifnet *ifp, int pswitch)
2194 {
2195         struct ifreq ifr;
2196         int error;
2197         int oldflags;
2198
2199         oldflags = ifp->if_flags;
2200         if (ifp->if_flags & IFF_PPROMISC) {
2201                 /* Do nothing if device is in permanently promiscuous mode */
2202                 ifp->if_pcount += pswitch ? 1 : -1;
2203                 return (0);
2204         }
2205         if (pswitch) {
2206                 /*
2207                  * If the device is not configured up, we cannot put it in
2208                  * promiscuous mode.
2209                  */
2210                 if ((ifp->if_flags & IFF_UP) == 0)
2211                         return (ENETDOWN);
2212                 if (ifp->if_pcount++ != 0)
2213                         return (0);
2214                 ifp->if_flags |= IFF_PROMISC;
2215                 log(LOG_INFO, "%s: promiscuous mode enabled\n",
2216                     ifp->if_xname);
2217         } else {
2218                 if (--ifp->if_pcount > 0)
2219                         return (0);
2220                 ifp->if_flags &= ~IFF_PROMISC;
2221                 log(LOG_INFO, "%s: promiscuous mode disabled\n",
2222                     ifp->if_xname);
2223         }
2224         ifr.ifr_flags = ifp->if_flags;
2225         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2226         ifnet_serialize_all(ifp);
2227         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr, NULL);
2228         ifnet_deserialize_all(ifp);
2229         if (error == 0)
2230                 rt_ifmsg(ifp);
2231         else
2232                 ifp->if_flags = oldflags;
2233         return error;
2234 }
2235
2236 /*
2237  * Return interface configuration
2238  * of system.  List may be used
2239  * in later ioctl's (above) to get
2240  * other information.
2241  */
2242 static int
2243 ifconf(u_long cmd, caddr_t data, struct ucred *cred)
2244 {
2245         struct ifconf *ifc = (struct ifconf *)data;
2246         struct ifnet *ifp;
2247         struct sockaddr *sa;
2248         struct ifreq ifr, *ifrp;
2249         int space = ifc->ifc_len, error = 0;
2250
2251         ifrp = ifc->ifc_req;
2252
2253         ifnet_lock();
2254         TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
2255                 struct ifaddr_container *ifac, *ifac_mark;
2256                 struct ifaddr_marker mark;
2257                 struct ifaddrhead *head;
2258                 int addrs;
2259
2260                 if (space <= sizeof ifr)
2261                         break;
2262
2263                 /*
2264                  * Zero the stack declared structure first to prevent
2265                  * memory disclosure.
2266                  */
2267                 bzero(&ifr, sizeof(ifr));
2268                 if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
2269                     >= sizeof(ifr.ifr_name)) {
2270                         error = ENAMETOOLONG;
2271                         break;
2272                 }
2273
2274                 /*
2275                  * Add a marker, since copyout() could block and during that
2276                  * period the list could be changed.  Inserting the marker to
2277                  * the header of the list will not cause trouble for the code
2278                  * assuming that the first element of the list is AF_LINK; the
2279                  * marker will be moved to the next position w/o blocking.
2280                  */
2281                 ifa_marker_init(&mark, ifp);
2282                 ifac_mark = &mark.ifac;
2283                 head = &ifp->if_addrheads[mycpuid];
2284
2285                 addrs = 0;
2286                 TAILQ_INSERT_HEAD(head, ifac_mark, ifa_link);
2287                 while ((ifac = TAILQ_NEXT(ifac_mark, ifa_link)) != NULL) {
2288                         struct ifaddr *ifa = ifac->ifa;
2289
2290                         TAILQ_REMOVE(head, ifac_mark, ifa_link);
2291                         TAILQ_INSERT_AFTER(head, ifac, ifac_mark, ifa_link);
2292
2293                         /* Ignore marker */
2294                         if (ifa->ifa_addr->sa_family == AF_UNSPEC)
2295                                 continue;
2296
2297                         if (space <= sizeof ifr)
2298                                 break;
2299                         sa = ifa->ifa_addr;
2300                         if (cred->cr_prison &&
2301                             prison_if(cred, sa))
2302                                 continue;
2303                         addrs++;
2304                         /*
2305                          * Keep a reference on this ifaddr, so that it will
2306                          * not be destroyed when its address is copied to
2307                          * the userland, which could block.
2308                          */
2309                         IFAREF(ifa);
2310                         if (sa->sa_len <= sizeof(*sa)) {
2311                                 ifr.ifr_addr = *sa;
2312                                 error = copyout(&ifr, ifrp, sizeof ifr);
2313                                 ifrp++;
2314                         } else {
2315                                 if (space < (sizeof ifr) + sa->sa_len -
2316                                             sizeof(*sa)) {
2317                                         IFAFREE(ifa);
2318                                         break;
2319                                 }
2320                                 space -= sa->sa_len - sizeof(*sa);
2321                                 error = copyout(&ifr, ifrp,
2322                                                 sizeof ifr.ifr_name);
2323                                 if (error == 0)
2324                                         error = copyout(sa, &ifrp->ifr_addr,
2325                                                         sa->sa_len);
2326                                 ifrp = (struct ifreq *)
2327                                         (sa->sa_len + (caddr_t)&ifrp->ifr_addr);
2328                         }
2329                         IFAFREE(ifa);
2330                         if (error)
2331                                 break;
2332                         space -= sizeof ifr;
2333                 }
2334                 TAILQ_REMOVE(head, ifac_mark, ifa_link);
2335                 if (error)
2336                         break;
2337                 if (!addrs) {
2338                         bzero(&ifr.ifr_addr, sizeof ifr.ifr_addr);
2339                         error = copyout(&ifr, ifrp, sizeof ifr);
2340                         if (error)
2341                                 break;
2342                         space -= sizeof ifr;
2343                         ifrp++;
2344                 }
2345         }
2346         ifnet_unlock();
2347
2348         ifc->ifc_len -= space;
2349         return (error);
2350 }
2351
2352 /*
2353  * Just like if_promisc(), but for all-multicast-reception mode.
2354  */
2355 int
2356 if_allmulti(struct ifnet *ifp, int onswitch)
2357 {
2358         int error = 0;
2359         struct ifreq ifr;
2360
2361         crit_enter();
2362
2363         if (onswitch) {
2364                 if (ifp->if_amcount++ == 0) {
2365                         ifp->if_flags |= IFF_ALLMULTI;
2366                         ifr.ifr_flags = ifp->if_flags;
2367                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2368                         ifnet_serialize_all(ifp);
2369                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2370                                               NULL);
2371                         ifnet_deserialize_all(ifp);
2372                 }
2373         } else {
2374                 if (ifp->if_amcount > 1) {
2375                         ifp->if_amcount--;
2376                 } else {
2377                         ifp->if_amcount = 0;
2378                         ifp->if_flags &= ~IFF_ALLMULTI;
2379                         ifr.ifr_flags = ifp->if_flags;
2380                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2381                         ifnet_serialize_all(ifp);
2382                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2383                                               NULL);
2384                         ifnet_deserialize_all(ifp);
2385                 }
2386         }
2387
2388         crit_exit();
2389
2390         if (error == 0)
2391                 rt_ifmsg(ifp);
2392         return error;
2393 }
2394
2395 /*
2396  * Add a multicast listenership to the interface in question.
2397  * The link layer provides a routine which converts
2398  */
2399 int
2400 if_addmulti_serialized(struct ifnet *ifp, struct sockaddr *sa,
2401     struct ifmultiaddr **retifma)
2402 {
2403         struct sockaddr *llsa, *dupsa;
2404         int error;
2405         struct ifmultiaddr *ifma;
2406
2407         ASSERT_IFNET_SERIALIZED_ALL(ifp);
2408
2409         /*
2410          * If the matching multicast address already exists
2411          * then don't add a new one, just add a reference
2412          */
2413         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2414                 if (sa_equal(sa, ifma->ifma_addr)) {
2415                         ifma->ifma_refcount++;
2416                         if (retifma)
2417                                 *retifma = ifma;
2418                         return 0;
2419                 }
2420         }
2421
2422         /*
2423          * Give the link layer a chance to accept/reject it, and also
2424          * find out which AF_LINK address this maps to, if it isn't one
2425          * already.
2426          */
2427         if (ifp->if_resolvemulti) {
2428                 error = ifp->if_resolvemulti(ifp, &llsa, sa);
2429                 if (error)
2430                         return error;
2431         } else {
2432                 llsa = NULL;
2433         }
2434
2435         ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_INTWAIT);
2436         dupsa = kmalloc(sa->sa_len, M_IFMADDR, M_INTWAIT);
2437         bcopy(sa, dupsa, sa->sa_len);
2438
2439         ifma->ifma_addr = dupsa;
2440         ifma->ifma_lladdr = llsa;
2441         ifma->ifma_ifp = ifp;
2442         ifma->ifma_refcount = 1;
2443         ifma->ifma_protospec = NULL;
2444         rt_newmaddrmsg(RTM_NEWMADDR, ifma);
2445
2446         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2447         if (retifma)
2448                 *retifma = ifma;
2449
2450         if (llsa != NULL) {
2451                 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2452                         if (sa_equal(ifma->ifma_addr, llsa))
2453                                 break;
2454                 }
2455                 if (ifma) {
2456                         ifma->ifma_refcount++;
2457                 } else {
2458                         ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_INTWAIT);
2459                         dupsa = kmalloc(llsa->sa_len, M_IFMADDR, M_INTWAIT);
2460                         bcopy(llsa, dupsa, llsa->sa_len);
2461                         ifma->ifma_addr = dupsa;
2462                         ifma->ifma_ifp = ifp;
2463                         ifma->ifma_refcount = 1;
2464                         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2465                 }
2466         }
2467         /*
2468          * We are certain we have added something, so call down to the
2469          * interface to let them know about it.
2470          */
2471         if (ifp->if_ioctl)
2472                 ifp->if_ioctl(ifp, SIOCADDMULTI, 0, NULL);
2473
2474         return 0;
2475 }
2476
2477 int
2478 if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
2479     struct ifmultiaddr **retifma)
2480 {
2481         int error;
2482
2483         ifnet_serialize_all(ifp);
2484         error = if_addmulti_serialized(ifp, sa, retifma);
2485         ifnet_deserialize_all(ifp);
2486
2487         return error;
2488 }
2489
2490 /*
2491  * Remove a reference to a multicast address on this interface.  Yell
2492  * if the request does not match an existing membership.
2493  */
2494 static int
2495 if_delmulti_serialized(struct ifnet *ifp, struct sockaddr *sa)
2496 {
2497         struct ifmultiaddr *ifma;
2498
2499         ASSERT_IFNET_SERIALIZED_ALL(ifp);
2500
2501         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2502                 if (sa_equal(sa, ifma->ifma_addr))
2503                         break;
2504         if (ifma == NULL)
2505                 return ENOENT;
2506
2507         if (ifma->ifma_refcount > 1) {
2508                 ifma->ifma_refcount--;
2509                 return 0;
2510         }
2511
2512         rt_newmaddrmsg(RTM_DELMADDR, ifma);
2513         sa = ifma->ifma_lladdr;
2514         TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2515         /*
2516          * Make sure the interface driver is notified
2517          * in the case of a link layer mcast group being left.
2518          */
2519         if (ifma->ifma_addr->sa_family == AF_LINK && sa == NULL)
2520                 ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2521         kfree(ifma->ifma_addr, M_IFMADDR);
2522         kfree(ifma, M_IFMADDR);
2523         if (sa == NULL)
2524                 return 0;
2525
2526         /*
2527          * Now look for the link-layer address which corresponds to
2528          * this network address.  It had been squirreled away in
2529          * ifma->ifma_lladdr for this purpose (so we don't have
2530          * to call ifp->if_resolvemulti() again), and we saved that
2531          * value in sa above.  If some nasty deleted the
2532          * link-layer address out from underneath us, we can deal because
2533          * the address we stored was is not the same as the one which was
2534          * in the record for the link-layer address.  (So we don't complain
2535          * in that case.)
2536          */
2537         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2538                 if (sa_equal(sa, ifma->ifma_addr))
2539                         break;
2540         if (ifma == NULL)
2541                 return 0;
2542
2543         if (ifma->ifma_refcount > 1) {
2544                 ifma->ifma_refcount--;
2545                 return 0;
2546         }
2547
2548         TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2549         ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2550         kfree(ifma->ifma_addr, M_IFMADDR);
2551         kfree(sa, M_IFMADDR);
2552         kfree(ifma, M_IFMADDR);
2553
2554         return 0;
2555 }
2556
2557 int
2558 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
2559 {
2560         int error;
2561
2562         ifnet_serialize_all(ifp);
2563         error = if_delmulti_serialized(ifp, sa);
2564         ifnet_deserialize_all(ifp);
2565
2566         return error;
2567 }
2568
2569 /*
2570  * Delete all multicast group membership for an interface.
2571  * Should be used to quickly flush all multicast filters.
2572  */
2573 void
2574 if_delallmulti_serialized(struct ifnet *ifp)
2575 {
2576         struct ifmultiaddr *ifma, mark;
2577         struct sockaddr sa;
2578
2579         ASSERT_IFNET_SERIALIZED_ALL(ifp);
2580
2581         bzero(&sa, sizeof(sa));
2582         sa.sa_family = AF_UNSPEC;
2583         sa.sa_len = sizeof(sa);
2584
2585         bzero(&mark, sizeof(mark));
2586         mark.ifma_addr = &sa;
2587
2588         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, &mark, ifma_link);
2589         while ((ifma = TAILQ_NEXT(&mark, ifma_link)) != NULL) {
2590                 TAILQ_REMOVE(&ifp->if_multiaddrs, &mark, ifma_link);
2591                 TAILQ_INSERT_AFTER(&ifp->if_multiaddrs, ifma, &mark,
2592                     ifma_link);
2593
2594                 if (ifma->ifma_addr->sa_family == AF_UNSPEC)
2595                         continue;
2596
2597                 if_delmulti_serialized(ifp, ifma->ifma_addr);
2598         }
2599         TAILQ_REMOVE(&ifp->if_multiaddrs, &mark, ifma_link);
2600 }
2601
2602
2603 /*
2604  * Set the link layer address on an interface.
2605  *
2606  * At this time we only support certain types of interfaces,
2607  * and we don't allow the length of the address to change.
2608  */
2609 int
2610 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
2611 {
2612         struct sockaddr_dl *sdl;
2613         struct ifreq ifr;
2614
2615         sdl = IF_LLSOCKADDR(ifp);
2616         if (sdl == NULL)
2617                 return (EINVAL);
2618         if (len != sdl->sdl_alen)       /* don't allow length to change */
2619                 return (EINVAL);
2620         switch (ifp->if_type) {
2621         case IFT_ETHER:                 /* these types use struct arpcom */
2622         case IFT_XETHER:
2623         case IFT_L2VLAN:
2624         case IFT_IEEE8023ADLAG:
2625                 bcopy(lladdr, ((struct arpcom *)ifp->if_softc)->ac_enaddr, len);
2626                 bcopy(lladdr, LLADDR(sdl), len);
2627                 break;
2628         default:
2629                 return (ENODEV);
2630         }
2631         /*
2632          * If the interface is already up, we need
2633          * to re-init it in order to reprogram its
2634          * address filter.
2635          */
2636         ifnet_serialize_all(ifp);
2637         if ((ifp->if_flags & IFF_UP) != 0) {
2638 #ifdef INET
2639                 struct ifaddr_container *ifac;
2640 #endif
2641
2642                 ifp->if_flags &= ~IFF_UP;
2643                 ifr.ifr_flags = ifp->if_flags;
2644                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2645                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2646                               NULL);
2647                 ifp->if_flags |= IFF_UP;
2648                 ifr.ifr_flags = ifp->if_flags;
2649                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2650                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2651                                  NULL);
2652 #ifdef INET
2653                 /*
2654                  * Also send gratuitous ARPs to notify other nodes about
2655                  * the address change.
2656                  */
2657                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2658                         struct ifaddr *ifa = ifac->ifa;
2659
2660                         if (ifa->ifa_addr != NULL &&
2661                             ifa->ifa_addr->sa_family == AF_INET)
2662                                 arp_gratuitous(ifp, ifa);
2663                 }
2664 #endif
2665         }
2666         ifnet_deserialize_all(ifp);
2667         return (0);
2668 }
2669
2670 struct ifmultiaddr *
2671 ifmaof_ifpforaddr(struct sockaddr *sa, struct ifnet *ifp)
2672 {
2673         struct ifmultiaddr *ifma;
2674
2675         /* TODO: need ifnet_serialize_main */
2676         ifnet_serialize_all(ifp);
2677         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2678                 if (sa_equal(ifma->ifma_addr, sa))
2679                         break;
2680         ifnet_deserialize_all(ifp);
2681
2682         return ifma;
2683 }
2684
2685 /*
2686  * This function locates the first real ethernet MAC from a network
2687  * card and loads it into node, returning 0 on success or ENOENT if
2688  * no suitable interfaces were found.  It is used by the uuid code to
2689  * generate a unique 6-byte number.
2690  */
2691 int
2692 if_getanyethermac(uint16_t *node, int minlen)
2693 {
2694         struct ifnet *ifp;
2695         struct sockaddr_dl *sdl;
2696
2697         ifnet_lock();
2698         TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
2699                 if (ifp->if_type != IFT_ETHER)
2700                         continue;
2701                 sdl = IF_LLSOCKADDR(ifp);
2702                 if (sdl->sdl_alen < minlen)
2703                         continue;
2704                 bcopy(((struct arpcom *)ifp->if_softc)->ac_enaddr, node,
2705                       minlen);
2706                 ifnet_unlock();
2707                 return(0);
2708         }
2709         ifnet_unlock();
2710         return (ENOENT);
2711 }
2712
2713 /*
2714  * The name argument must be a pointer to storage which will last as
2715  * long as the interface does.  For physical devices, the result of
2716  * device_get_name(dev) is a good choice and for pseudo-devices a
2717  * static string works well.
2718  */
2719 void
2720 if_initname(struct ifnet *ifp, const char *name, int unit)
2721 {
2722         ifp->if_dname = name;
2723         ifp->if_dunit = unit;
2724         if (unit != IF_DUNIT_NONE)
2725                 ksnprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
2726         else
2727                 strlcpy(ifp->if_xname, name, IFNAMSIZ);
2728 }
2729
2730 int
2731 if_printf(struct ifnet *ifp, const char *fmt, ...)
2732 {
2733         __va_list ap;
2734         int retval;
2735
2736         retval = kprintf("%s: ", ifp->if_xname);
2737         __va_start(ap, fmt);
2738         retval += kvprintf(fmt, ap);
2739         __va_end(ap);
2740         return (retval);
2741 }
2742
2743 struct ifnet *
2744 if_alloc(uint8_t type)
2745 {
2746         struct ifnet *ifp;
2747         size_t size;
2748
2749         /*
2750          * XXX temporary hack until arpcom is setup in if_l2com
2751          */
2752         if (type == IFT_ETHER)
2753                 size = sizeof(struct arpcom);
2754         else
2755                 size = sizeof(struct ifnet);
2756
2757         ifp = kmalloc(size, M_IFNET, M_WAITOK|M_ZERO);
2758
2759         ifp->if_type = type;
2760
2761         if (if_com_alloc[type] != NULL) {
2762                 ifp->if_l2com = if_com_alloc[type](type, ifp);
2763                 if (ifp->if_l2com == NULL) {
2764                         kfree(ifp, M_IFNET);
2765                         return (NULL);
2766                 }
2767         }
2768         return (ifp);
2769 }
2770
2771 void
2772 if_free(struct ifnet *ifp)
2773 {
2774         kfree(ifp, M_IFNET);
2775 }
2776
2777 void
2778 ifq_set_classic(struct ifaltq *ifq)
2779 {
2780         ifq_set_methods(ifq, ifq->altq_ifp->if_mapsubq,
2781             ifsq_classic_enqueue, ifsq_classic_dequeue, ifsq_classic_request);
2782 }
2783
2784 void
2785 ifq_set_methods(struct ifaltq *ifq, altq_mapsubq_t mapsubq,
2786     ifsq_enqueue_t enqueue, ifsq_dequeue_t dequeue, ifsq_request_t request)
2787 {
2788         int q;
2789
2790         KASSERT(mapsubq != NULL, ("mapsubq is not specified"));
2791         KASSERT(enqueue != NULL, ("enqueue is not specified"));
2792         KASSERT(dequeue != NULL, ("dequeue is not specified"));
2793         KASSERT(request != NULL, ("request is not specified"));
2794
2795         ifq->altq_mapsubq = mapsubq;
2796         for (q = 0; q < ifq->altq_subq_cnt; ++q) {
2797                 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
2798
2799                 ifsq->ifsq_enqueue = enqueue;
2800                 ifsq->ifsq_dequeue = dequeue;
2801                 ifsq->ifsq_request = request;
2802         }
2803 }
2804
2805 static void
2806 ifsq_norm_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
2807 {
2808
2809         classq_add(&ifsq->ifsq_norm, m);
2810         ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
2811 }
2812
2813 static void
2814 ifsq_prio_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
2815 {
2816
2817         classq_add(&ifsq->ifsq_prio, m);
2818         ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
2819         ALTQ_SQ_PRIO_CNTR_INC(ifsq, m->m_pkthdr.len);
2820 }
2821
2822 static struct mbuf *
2823 ifsq_norm_dequeue(struct ifaltq_subque *ifsq)
2824 {
2825         struct mbuf *m;
2826
2827         m = classq_get(&ifsq->ifsq_norm);
2828         if (m != NULL)
2829                 ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
2830         return (m);
2831 }
2832
2833 static struct mbuf *
2834 ifsq_prio_dequeue(struct ifaltq_subque *ifsq)
2835 {
2836         struct mbuf *m;
2837
2838         m = classq_get(&ifsq->ifsq_prio);
2839         if (m != NULL) {
2840                 ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
2841                 ALTQ_SQ_PRIO_CNTR_DEC(ifsq, m->m_pkthdr.len);
2842         }
2843         return (m);
2844 }
2845
2846 int
2847 ifsq_classic_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m,
2848     struct altq_pktattr *pa __unused)
2849 {
2850
2851         M_ASSERTPKTHDR(m);
2852 again:
2853         if (ifsq->ifsq_len >= ifsq->ifsq_maxlen ||
2854             ifsq->ifsq_bcnt >= ifsq->ifsq_maxbcnt) {
2855                 struct mbuf *m_drop;
2856
2857                 if (m->m_flags & M_PRIO) {
2858                         m_drop = NULL;
2859                         if (ifsq->ifsq_prio_len < (ifsq->ifsq_maxlen >> 1) &&
2860                             ifsq->ifsq_prio_bcnt < (ifsq->ifsq_maxbcnt >> 1)) {
2861                                 /* Try dropping some from normal queue. */
2862                                 m_drop = ifsq_norm_dequeue(ifsq);
2863                         }
2864                         if (m_drop == NULL)
2865                                 m_drop = ifsq_prio_dequeue(ifsq);
2866                 } else {
2867                         m_drop = ifsq_norm_dequeue(ifsq);
2868                 }
2869                 if (m_drop != NULL) {
2870                         IFNET_STAT_INC(ifsq->ifsq_ifp, oqdrops, 1);
2871                         m_freem(m_drop);
2872                         goto again;
2873                 }
2874                 /*
2875                  * No old packets could be dropped!
2876                  * NOTE: Caller increases oqdrops.
2877                  */
2878                 m_freem(m);
2879                 return (ENOBUFS);
2880         } else {
2881                 if (m->m_flags & M_PRIO)
2882                         ifsq_prio_enqueue(ifsq, m);
2883                 else
2884                         ifsq_norm_enqueue(ifsq, m);
2885                 return (0);
2886         }
2887 }
2888
2889 struct mbuf *
2890 ifsq_classic_dequeue(struct ifaltq_subque *ifsq, int op)
2891 {
2892         struct mbuf *m;
2893
2894         switch (op) {
2895         case ALTDQ_POLL:
2896                 m = classq_head(&ifsq->ifsq_prio);
2897                 if (m == NULL)
2898                         m = classq_head(&ifsq->ifsq_norm);
2899                 break;
2900
2901         case ALTDQ_REMOVE:
2902                 m = ifsq_prio_dequeue(ifsq);
2903                 if (m == NULL)
2904                         m = ifsq_norm_dequeue(ifsq);
2905                 break;
2906
2907         default:
2908                 panic("unsupported ALTQ dequeue op: %d", op);
2909         }
2910         return m;
2911 }
2912
2913 int
2914 ifsq_classic_request(struct ifaltq_subque *ifsq, int req, void *arg)
2915 {
2916         switch (req) {
2917         case ALTRQ_PURGE:
2918                 for (;;) {
2919                         struct mbuf *m;
2920
2921                         m = ifsq_classic_dequeue(ifsq, ALTDQ_REMOVE);
2922                         if (m == NULL)
2923                                 break;
2924                         m_freem(m);
2925                 }
2926                 break;
2927
2928         default:
2929                 panic("unsupported ALTQ request: %d", req);
2930         }
2931         return 0;
2932 }
2933
2934 static void
2935 ifsq_ifstart_try(struct ifaltq_subque *ifsq, int force_sched)
2936 {
2937         struct ifnet *ifp = ifsq_get_ifp(ifsq);
2938         int running = 0, need_sched;
2939
2940         /*
2941          * Try to do direct ifnet.if_start on the subqueue first, if there is
2942          * contention on the subqueue hardware serializer, ifnet.if_start on
2943          * the subqueue will be scheduled on the subqueue owner CPU.
2944          */
2945         if (!ifsq_tryserialize_hw(ifsq)) {
2946                 /*
2947                  * Subqueue hardware serializer contention happened,
2948                  * ifnet.if_start on the subqueue is scheduled on
2949                  * the subqueue owner CPU, and we keep going.
2950                  */
2951                 ifsq_ifstart_schedule(ifsq, 1);
2952                 return;
2953         }
2954
2955         if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
2956                 ifp->if_start(ifp, ifsq);
2957                 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
2958                         running = 1;
2959         }
2960         need_sched = ifsq_ifstart_need_schedule(ifsq, running);
2961
2962         ifsq_deserialize_hw(ifsq);
2963
2964         if (need_sched) {
2965                 /*
2966                  * More data need to be transmitted, ifnet.if_start on the
2967                  * subqueue is scheduled on the subqueue owner CPU, and we
2968                  * keep going.
2969                  * NOTE: ifnet.if_start subqueue interlock is not released.
2970                  */
2971                 ifsq_ifstart_schedule(ifsq, force_sched);
2972         }
2973 }
2974
2975 /*
2976  * Subqeue packets staging mechanism:
2977  *
2978  * The packets enqueued into the subqueue are staged to a certain amount
2979  * before the ifnet.if_start on the subqueue is called.  In this way, the
2980  * driver could avoid writing to hardware registers upon every packet,
2981  * instead, hardware registers could be written when certain amount of
2982  * packets are put onto hardware TX ring.  The measurement on several modern
2983  * NICs (emx(4), igb(4), bnx(4), bge(4), jme(4)) shows that the hardware
2984  * registers writing aggregation could save ~20% CPU time when 18bytes UDP
2985  * datagrams are transmitted at 1.48Mpps.  The performance improvement by
2986  * hardware registers writing aggeregation is also mentioned by Luigi Rizzo's
2987  * netmap paper (http://info.iet.unipi.it/~luigi/netmap/).
2988  *
2989  * Subqueue packets staging is performed for two entry points into drivers'
2990  * transmission function:
2991  * - Direct ifnet.if_start calling on the subqueue, i.e. ifsq_ifstart_try()
2992  * - ifnet.if_start scheduling on the subqueue, i.e. ifsq_ifstart_schedule()
2993  *
2994  * Subqueue packets staging will be stopped upon any of the following
2995  * conditions:
2996  * - If the count of packets enqueued on the current CPU is great than or
2997  *   equal to ifsq_stage_cntmax. (XXX this should be per-interface)
2998  * - If the total length of packets enqueued on the current CPU is great
2999  *   than or equal to the hardware's MTU - max_protohdr.  max_protohdr is
3000  *   cut from the hardware's MTU mainly bacause a full TCP segment's size
3001  *   is usually less than hardware's MTU.
3002  * - ifsq_ifstart_schedule() is not pending on the current CPU and
3003  *   ifnet.if_start subqueue interlock (ifaltq_subq.ifsq_started) is not
3004  *   released.
3005  * - The if_start_rollup(), which is registered as low priority netisr
3006  *   rollup function, is called; probably because no more work is pending
3007  *   for netisr.
3008  *
3009  * NOTE:
3010  * Currently subqueue packet staging is only performed in netisr threads.
3011  */
3012 int
3013 ifq_dispatch(struct ifnet *ifp, struct mbuf *m, struct altq_pktattr *pa)
3014 {
3015         struct ifaltq *ifq = &ifp->if_snd;
3016         struct ifaltq_subque *ifsq;
3017         int error, start = 0, len, mcast = 0, avoid_start = 0;
3018         struct ifsubq_stage_head *head = NULL;
3019         struct ifsubq_stage *stage = NULL;
3020         struct globaldata *gd = mycpu;
3021         struct thread *td = gd->gd_curthread;
3022
3023         crit_enter_quick(td);
3024
3025         ifsq = ifq_map_subq(ifq, gd->gd_cpuid);
3026         ASSERT_ALTQ_SQ_NOT_SERIALIZED_HW(ifsq);
3027
3028         len = m->m_pkthdr.len;
3029         if (m->m_flags & M_MCAST)
3030                 mcast = 1;
3031
3032         if (td->td_type == TD_TYPE_NETISR) {
3033                 head = &ifsubq_stage_heads[mycpuid];
3034                 stage = ifsq_get_stage(ifsq, mycpuid);
3035
3036                 stage->stg_cnt++;
3037                 stage->stg_len += len;
3038                 if (stage->stg_cnt < ifsq_stage_cntmax &&
3039                     stage->stg_len < (ifp->if_mtu - max_protohdr))
3040                         avoid_start = 1;
3041         }
3042
3043         ALTQ_SQ_LOCK(ifsq);
3044         error = ifsq_enqueue_locked(ifsq, m, pa);
3045         if (error) {
3046                 IFNET_STAT_INC(ifp, oqdrops, 1);
3047                 if (!ifsq_data_ready(ifsq)) {
3048                         ALTQ_SQ_UNLOCK(ifsq);
3049                         crit_exit_quick(td);
3050                         return error;
3051                 }
3052                 avoid_start = 0;
3053         }
3054         if (!ifsq_is_started(ifsq)) {
3055                 if (avoid_start) {
3056                         ALTQ_SQ_UNLOCK(ifsq);
3057
3058                         KKASSERT(!error);
3059                         if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
3060                                 ifsq_stage_insert(head, stage);
3061
3062                         IFNET_STAT_INC(ifp, obytes, len);
3063                         if (mcast)
3064                                 IFNET_STAT_INC(ifp, omcasts, 1);
3065                         crit_exit_quick(td);
3066                         return error;
3067                 }
3068
3069                 /*
3070                  * Hold the subqueue interlock of ifnet.if_start
3071                  */
3072                 ifsq_set_started(ifsq);
3073                 start = 1;
3074         }
3075         ALTQ_SQ_UNLOCK(ifsq);
3076
3077         if (!error) {
3078                 IFNET_STAT_INC(ifp, obytes, len);
3079                 if (mcast)
3080                         IFNET_STAT_INC(ifp, omcasts, 1);
3081         }
3082
3083         if (stage != NULL) {
3084                 if (!start && (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)) {
3085                         KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
3086                         if (!avoid_start) {
3087                                 ifsq_stage_remove(head, stage);
3088                                 ifsq_ifstart_schedule(ifsq, 1);
3089                         }
3090                         crit_exit_quick(td);
3091                         return error;
3092                 }
3093
3094                 if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED) {
3095                         ifsq_stage_remove(head, stage);
3096                 } else {
3097                         stage->stg_cnt = 0;
3098                         stage->stg_len = 0;
3099                 }
3100         }
3101
3102         if (!start) {
3103                 crit_exit_quick(td);
3104                 return error;
3105         }
3106
3107         ifsq_ifstart_try(ifsq, 0);
3108
3109         crit_exit_quick(td);
3110         return error;
3111 }
3112
3113 void *
3114 ifa_create(int size)
3115 {
3116         struct ifaddr *ifa;
3117         int i;
3118
3119         KASSERT(size >= sizeof(*ifa), ("ifaddr size too small"));
3120
3121         ifa = kmalloc(size, M_IFADDR, M_INTWAIT | M_ZERO);
3122
3123         /*
3124          * Make ifa_container availabel on all CPUs, since they
3125          * could be accessed by any threads.
3126          */
3127         ifa->ifa_containers =
3128             kmalloc_cachealign(ncpus * sizeof(struct ifaddr_container),
3129                 M_IFADDR, M_INTWAIT | M_ZERO);
3130
3131         ifa->ifa_ncnt = ncpus;
3132         for (i = 0; i < ncpus; ++i) {
3133                 struct ifaddr_container *ifac = &ifa->ifa_containers[i];
3134
3135                 ifac->ifa_magic = IFA_CONTAINER_MAGIC;
3136                 ifac->ifa = ifa;
3137                 ifac->ifa_refcnt = 1;
3138         }
3139 #ifdef IFADDR_DEBUG
3140         kprintf("alloc ifa %p %d\n", ifa, size);
3141 #endif
3142         return ifa;
3143 }
3144
3145 void
3146 ifac_free(struct ifaddr_container *ifac, int cpu_id)
3147 {
3148         struct ifaddr *ifa = ifac->ifa;
3149
3150         KKASSERT(ifac->ifa_magic == IFA_CONTAINER_MAGIC);
3151         KKASSERT(ifac->ifa_refcnt == 0);
3152         KASSERT(ifac->ifa_listmask == 0,
3153                 ("ifa is still on %#x lists", ifac->ifa_listmask));
3154
3155         ifac->ifa_magic = IFA_CONTAINER_DEAD;
3156
3157 #ifdef IFADDR_DEBUG_VERBOSE
3158         kprintf("try free ifa %p cpu_id %d\n", ifac->ifa, cpu_id);
3159 #endif
3160
3161         KASSERT(ifa->ifa_ncnt > 0 && ifa->ifa_ncnt <= ncpus,
3162                 ("invalid # of ifac, %d", ifa->ifa_ncnt));
3163         if (atomic_fetchadd_int(&ifa->ifa_ncnt, -1) == 1) {
3164 #ifdef IFADDR_DEBUG
3165                 kprintf("free ifa %p\n", ifa);
3166 #endif
3167                 kfree(ifa->ifa_containers, M_IFADDR);
3168                 kfree(ifa, M_IFADDR);
3169         }
3170 }
3171
3172 static void
3173 ifa_iflink_dispatch(netmsg_t nmsg)
3174 {
3175         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3176         struct ifaddr *ifa = msg->ifa;
3177         struct ifnet *ifp = msg->ifp;
3178         int cpu = mycpuid;
3179         struct ifaddr_container *ifac;
3180
3181         crit_enter();
3182
3183         ifac = &ifa->ifa_containers[cpu];
3184         ASSERT_IFAC_VALID(ifac);
3185         KASSERT((ifac->ifa_listmask & IFA_LIST_IFADDRHEAD) == 0,
3186                 ("ifaddr is on if_addrheads"));
3187
3188         ifac->ifa_listmask |= IFA_LIST_IFADDRHEAD;
3189         if (msg->tail)
3190                 TAILQ_INSERT_TAIL(&ifp->if_addrheads[cpu], ifac, ifa_link);
3191         else
3192                 TAILQ_INSERT_HEAD(&ifp->if_addrheads[cpu], ifac, ifa_link);
3193
3194         crit_exit();
3195
3196         netisr_forwardmsg_all(&nmsg->base, cpu + 1);
3197 }
3198
3199 void
3200 ifa_iflink(struct ifaddr *ifa, struct ifnet *ifp, int tail)
3201 {
3202         struct netmsg_ifaddr msg;
3203
3204         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3205                     0, ifa_iflink_dispatch);
3206         msg.ifa = ifa;
3207         msg.ifp = ifp;
3208         msg.tail = tail;
3209
3210         netisr_domsg(&msg.base, 0);
3211 }
3212
3213 static void
3214 ifa_ifunlink_dispatch(netmsg_t nmsg)
3215 {
3216         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3217         struct ifaddr *ifa = msg->ifa;
3218         struct ifnet *ifp = msg->ifp;
3219         int cpu = mycpuid;
3220         struct ifaddr_container *ifac;
3221
3222         crit_enter();
3223
3224         ifac = &ifa->ifa_containers[cpu];
3225         ASSERT_IFAC_VALID(ifac);
3226         KASSERT(ifac->ifa_listmask & IFA_LIST_IFADDRHEAD,
3227                 ("ifaddr is not on if_addrhead"));
3228
3229         TAILQ_REMOVE(&ifp->if_addrheads[cpu], ifac, ifa_link);
3230         ifac->ifa_listmask &= ~IFA_LIST_IFADDRHEAD;
3231
3232         crit_exit();
3233
3234         netisr_forwardmsg_all(&nmsg->base, cpu + 1);
3235 }
3236
3237 void
3238 ifa_ifunlink(struct ifaddr *ifa, struct ifnet *ifp)
3239 {
3240         struct netmsg_ifaddr msg;
3241
3242         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3243                     0, ifa_ifunlink_dispatch);
3244         msg.ifa = ifa;
3245         msg.ifp = ifp;
3246
3247         netisr_domsg(&msg.base, 0);
3248 }
3249
3250 static void
3251 ifa_destroy_dispatch(netmsg_t nmsg)
3252 {
3253         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3254
3255         IFAFREE(msg->ifa);
3256         netisr_forwardmsg_all(&nmsg->base, mycpuid + 1);
3257 }
3258
3259 void
3260 ifa_destroy(struct ifaddr *ifa)
3261 {
3262         struct netmsg_ifaddr msg;
3263
3264         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3265                     0, ifa_destroy_dispatch);
3266         msg.ifa = ifa;
3267
3268         netisr_domsg(&msg.base, 0);
3269 }
3270
3271 static void
3272 if_start_rollup(void)
3273 {
3274         struct ifsubq_stage_head *head = &ifsubq_stage_heads[mycpuid];
3275         struct ifsubq_stage *stage;
3276
3277         crit_enter();
3278
3279         while ((stage = TAILQ_FIRST(&head->stg_head)) != NULL) {
3280                 struct ifaltq_subque *ifsq = stage->stg_subq;
3281                 int is_sched = 0;
3282
3283                 if (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)
3284                         is_sched = 1;
3285                 ifsq_stage_remove(head, stage);
3286
3287                 if (is_sched) {
3288                         ifsq_ifstart_schedule(ifsq, 1);
3289                 } else {
3290                         int start = 0;
3291
3292                         ALTQ_SQ_LOCK(ifsq);
3293                         if (!ifsq_is_started(ifsq)) {
3294                                 /*
3295                                  * Hold the subqueue interlock of
3296                                  * ifnet.if_start
3297                                  */
3298                                 ifsq_set_started(ifsq);
3299                                 start = 1;
3300                         }
3301                         ALTQ_SQ_UNLOCK(ifsq);
3302
3303                         if (start)
3304                                 ifsq_ifstart_try(ifsq, 1);
3305                 }
3306                 KKASSERT((stage->stg_flags &
3307                     (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
3308         }
3309
3310         crit_exit();
3311 }
3312
3313 static void
3314 ifnetinit(void *dummy __unused)
3315 {
3316         int i;
3317
3318         /* XXX netisr_ncpus */
3319         for (i = 0; i < ncpus; ++i)
3320                 TAILQ_INIT(&ifsubq_stage_heads[i].stg_head);
3321         netisr_register_rollup(if_start_rollup, NETISR_ROLLUP_PRIO_IFSTART);
3322 }
3323
3324 void
3325 if_register_com_alloc(u_char type,
3326     if_com_alloc_t *a, if_com_free_t *f)
3327 {
3328
3329         KASSERT(if_com_alloc[type] == NULL,
3330             ("if_register_com_alloc: %d already registered", type));
3331         KASSERT(if_com_free[type] == NULL,
3332             ("if_register_com_alloc: %d free already registered", type));
3333
3334         if_com_alloc[type] = a;
3335         if_com_free[type] = f;
3336 }
3337
3338 void
3339 if_deregister_com_alloc(u_char type)
3340 {
3341
3342         KASSERT(if_com_alloc[type] != NULL,
3343             ("if_deregister_com_alloc: %d not registered", type));
3344         KASSERT(if_com_free[type] != NULL,
3345             ("if_deregister_com_alloc: %d free not registered", type));
3346         if_com_alloc[type] = NULL;
3347         if_com_free[type] = NULL;
3348 }
3349
3350 void
3351 ifq_set_maxlen(struct ifaltq *ifq, int len)
3352 {
3353         ifq->altq_maxlen = len + (ncpus * ifsq_stage_cntmax);
3354 }
3355
3356 int
3357 ifq_mapsubq_default(struct ifaltq *ifq __unused, int cpuid __unused)
3358 {
3359         return ALTQ_SUBQ_INDEX_DEFAULT;
3360 }
3361
3362 int
3363 ifq_mapsubq_modulo(struct ifaltq *ifq, int cpuid)
3364 {
3365
3366         return (cpuid % ifq->altq_subq_mappriv);
3367 }
3368
3369 static void
3370 ifsq_watchdog(void *arg)
3371 {
3372         struct ifsubq_watchdog *wd = arg;
3373         struct ifnet *ifp;
3374
3375         if (__predict_true(wd->wd_timer == 0 || --wd->wd_timer))
3376                 goto done;
3377
3378         ifp = ifsq_get_ifp(wd->wd_subq);
3379         if (ifnet_tryserialize_all(ifp)) {
3380                 wd->wd_watchdog(wd->wd_subq);
3381                 ifnet_deserialize_all(ifp);
3382         } else {
3383                 /* try again next timeout */
3384                 wd->wd_timer = 1;
3385         }
3386 done:
3387         ifsq_watchdog_reset(wd);
3388 }
3389
3390 static void
3391 ifsq_watchdog_reset(struct ifsubq_watchdog *wd)
3392 {
3393         callout_reset_bycpu(&wd->wd_callout, hz, ifsq_watchdog, wd,
3394             ifsq_get_cpuid(wd->wd_subq));
3395 }
3396
3397 void
3398 ifsq_watchdog_init(struct ifsubq_watchdog *wd, struct ifaltq_subque *ifsq,
3399     ifsq_watchdog_t watchdog)
3400 {
3401         callout_init_mp(&wd->wd_callout);
3402         wd->wd_timer = 0;
3403         wd->wd_subq = ifsq;
3404         wd->wd_watchdog = watchdog;
3405 }
3406
3407 void
3408 ifsq_watchdog_start(struct ifsubq_watchdog *wd)
3409 {
3410         wd->wd_timer = 0;
3411         ifsq_watchdog_reset(wd);
3412 }
3413
3414 void
3415 ifsq_watchdog_stop(struct ifsubq_watchdog *wd)
3416 {
3417         wd->wd_timer = 0;
3418         callout_stop(&wd->wd_callout);
3419 }
3420
3421 void
3422 ifnet_lock(void)
3423 {
3424         KASSERT(curthread->td_type != TD_TYPE_NETISR,
3425             ("try holding ifnet lock in netisr"));
3426         mtx_lock(&ifnet_mtx);
3427 }
3428
3429 void
3430 ifnet_unlock(void)
3431 {
3432         KASSERT(curthread->td_type != TD_TYPE_NETISR,
3433             ("try holding ifnet lock in netisr"));
3434         mtx_unlock(&ifnet_mtx);
3435 }
3436
3437 static struct ifnet_array *
3438 ifnet_array_alloc(int count)
3439 {
3440         struct ifnet_array *arr;
3441
3442         arr = kmalloc(__offsetof(struct ifnet_array, ifnet_arr[count]),
3443             M_IFNET, M_WAITOK);
3444         arr->ifnet_count = count;
3445
3446         return arr;
3447 }
3448
3449 static void
3450 ifnet_array_free(struct ifnet_array *arr)
3451 {
3452         if (arr == &ifnet_array0)
3453                 return;
3454         kfree(arr, M_IFNET);
3455 }
3456
3457 static struct ifnet_array *
3458 ifnet_array_add(struct ifnet *ifp, const struct ifnet_array *old_arr)
3459 {
3460         struct ifnet_array *arr;
3461         int count, i;
3462
3463         KASSERT(old_arr->ifnet_count >= 0,
3464             ("invalid ifnet array count %d", old_arr->ifnet_count));
3465         count = old_arr->ifnet_count + 1;
3466         arr = ifnet_array_alloc(count);
3467
3468         /*
3469          * Save the old ifnet array and append this ifp to the end of
3470          * the new ifnet array.
3471          */
3472         for (i = 0; i < old_arr->ifnet_count; ++i) {
3473                 KASSERT(old_arr->ifnet_arr[i] != ifp,
3474                     ("%s is already in ifnet array", ifp->if_xname));
3475                 arr->ifnet_arr[i] = old_arr->ifnet_arr[i];
3476         }
3477         KASSERT(i == count - 1,
3478             ("add %s, ifnet array index mismatch, should be %d, but got %d",
3479              ifp->if_xname, count - 1, i));
3480         arr->ifnet_arr[i] = ifp;
3481
3482         return arr;
3483 }
3484
3485 static struct ifnet_array *
3486 ifnet_array_del(struct ifnet *ifp, const struct ifnet_array *old_arr)
3487 {
3488         struct ifnet_array *arr;
3489         int count, i, idx, found = 0;
3490
3491         KASSERT(old_arr->ifnet_count > 0,
3492             ("invalid ifnet array count %d", old_arr->ifnet_count));
3493         count = old_arr->ifnet_count - 1;
3494         arr = ifnet_array_alloc(count);
3495
3496         /*
3497          * Save the old ifnet array, but skip this ifp.
3498          */
3499         idx = 0;
3500         for (i = 0; i < old_arr->ifnet_count; ++i) {
3501                 if (old_arr->ifnet_arr[i] == ifp) {
3502                         KASSERT(!found,
3503                             ("dup %s is in ifnet array", ifp->if_xname));
3504                         found = 1;
3505                         continue;
3506                 }
3507                 KASSERT(idx < count,
3508                     ("invalid ifnet array index %d, count %d", idx, count));
3509                 arr->ifnet_arr[idx] = old_arr->ifnet_arr[i];
3510                 ++idx;
3511         }
3512         KASSERT(found, ("%s is not in ifnet array", ifp->if_xname));
3513         KASSERT(idx == count,
3514             ("del %s, ifnet array count mismatch, should be %d, but got %d ",
3515              ifp->if_xname, count, idx));
3516
3517         return arr;
3518 }
3519
3520 const struct ifnet_array *
3521 ifnet_array_get(void)
3522 {
3523         const struct ifnet_array *ret;
3524
3525         KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
3526         ret = ifnet_array;
3527         /* Make sure 'ret' is really used. */
3528         cpu_ccfence();
3529         return (ret);
3530 }
3531
3532 int
3533 ifnet_array_isempty(void)
3534 {
3535         KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
3536         if (ifnet_array->ifnet_count == 0)
3537                 return 1;
3538         else
3539                 return 0;
3540 }
3541
3542 void
3543 ifa_marker_init(struct ifaddr_marker *mark, struct ifnet *ifp)
3544 {
3545         struct ifaddr *ifa;
3546
3547         memset(mark, 0, sizeof(*mark));
3548         ifa = &mark->ifa;
3549
3550         mark->ifac.ifa = ifa;
3551
3552         ifa->ifa_addr = &mark->addr;
3553         ifa->ifa_dstaddr = &mark->dstaddr;
3554         ifa->ifa_netmask = &mark->netmask;
3555         ifa->ifa_ifp = ifp;
3556 }
3557
3558 static int
3559 if_ringcnt_fixup(int ring_cnt, int ring_cntmax)
3560 {
3561
3562         KASSERT(ring_cntmax > 0, ("invalid ring count max %d", ring_cntmax));
3563
3564         if (ring_cnt <= 0 || ring_cnt > ring_cntmax)
3565                 ring_cnt = ring_cntmax;
3566         if (ring_cnt > netisr_ncpus)
3567                 ring_cnt = netisr_ncpus;
3568         return (ring_cnt);
3569 }
3570
3571 static void
3572 if_ringmap_set_grid(device_t dev, struct if_ringmap *rm, int grid)
3573 {
3574         int i, offset;
3575
3576         KASSERT(grid > 0, ("invalid if_ringmap grid %d", grid));
3577         KASSERT(grid >= rm->rm_cnt, ("invalid if_ringmap grid %d, count %d",
3578             grid, rm->rm_cnt));
3579         rm->rm_grid = grid;
3580
3581         offset = (rm->rm_grid * device_get_unit(dev)) % netisr_ncpus;
3582         for (i = 0; i < rm->rm_cnt; ++i) {
3583                 rm->rm_cpumap[i] = offset + i;
3584                 KASSERT(rm->rm_cpumap[i] < netisr_ncpus,
3585                     ("invalid cpumap[%d] = %d, offset %d", i,
3586                      rm->rm_cpumap[i], offset));
3587         }
3588 }
3589
3590 static struct if_ringmap *
3591 if_ringmap_alloc_flags(device_t dev, int ring_cnt, int ring_cntmax,
3592     uint32_t flags)
3593 {
3594         struct if_ringmap *rm;
3595         int i, grid = 0, prev_grid;
3596
3597         ring_cnt = if_ringcnt_fixup(ring_cnt, ring_cntmax);
3598         rm = kmalloc(__offsetof(struct if_ringmap, rm_cpumap[ring_cnt]),
3599             M_DEVBUF, M_WAITOK | M_ZERO);
3600
3601         rm->rm_cnt = ring_cnt;
3602         if (flags & RINGMAP_FLAG_POWEROF2)
3603                 rm->rm_cnt = 1 << (fls(rm->rm_cnt) - 1);
3604
3605         prev_grid = netisr_ncpus;
3606         for (i = 0; i < netisr_ncpus; ++i) {
3607                 if (netisr_ncpus % (i + 1) != 0)
3608                         continue;
3609
3610                 grid = netisr_ncpus / (i + 1);
3611                 if (rm->rm_cnt > grid) {
3612                         grid = prev_grid;
3613                         break;
3614                 }
3615
3616                 if (rm->rm_cnt > netisr_ncpus / (i + 2))
3617                         break;
3618                 prev_grid = grid;
3619         }
3620         if_ringmap_set_grid(dev, rm, grid);
3621
3622         return (rm);
3623 }
3624
3625 struct if_ringmap *
3626 if_ringmap_alloc(device_t dev, int ring_cnt, int ring_cntmax)
3627 {
3628
3629         return (if_ringmap_alloc_flags(dev, ring_cnt, ring_cntmax,
3630             RINGMAP_FLAG_NONE));
3631 }
3632
3633 struct if_ringmap *
3634 if_ringmap_alloc2(device_t dev, int ring_cnt, int ring_cntmax)
3635 {
3636
3637         return (if_ringmap_alloc_flags(dev, ring_cnt, ring_cntmax,
3638             RINGMAP_FLAG_POWEROF2));
3639 }
3640
3641 void
3642 if_ringmap_free(struct if_ringmap *rm)
3643 {
3644
3645         kfree(rm, M_DEVBUF);
3646 }
3647
3648 /*
3649  * Align the two ringmaps.
3650  *
3651  * e.g. 8 netisrs, rm0 contains 4 rings, rm1 contains 2 rings.
3652  *
3653  * Before:
3654  *
3655  * CPU      0  1  2  3   4  5  6  7
3656  * NIC_RX               n0 n1 n2 n3
3657  * NIC_TX        N0 N1
3658  *
3659  * After:
3660  *
3661  * CPU      0  1  2  3   4  5  6  7
3662  * NIC_RX               n0 n1 n2 n3
3663  * NIC_TX               N0 N1
3664  */
3665 void
3666 if_ringmap_align(device_t dev, struct if_ringmap *rm0, struct if_ringmap *rm1)
3667 {
3668
3669         if (rm0->rm_grid > rm1->rm_grid)
3670                 if_ringmap_set_grid(dev, rm1, rm0->rm_grid);
3671         else if (rm0->rm_grid < rm1->rm_grid)
3672                 if_ringmap_set_grid(dev, rm0, rm1->rm_grid);
3673 }
3674
3675 void
3676 if_ringmap_match(device_t dev, struct if_ringmap *rm0, struct if_ringmap *rm1)
3677 {
3678         int subset_grid, cnt, divisor, mod, offset, i;
3679         struct if_ringmap *subset_rm, *rm;
3680         int old_rm0_grid, old_rm1_grid;
3681
3682         if (rm0->rm_grid == rm1->rm_grid)
3683                 return;
3684
3685         /* Save grid for later use */
3686         old_rm0_grid = rm0->rm_grid;
3687         old_rm1_grid = rm1->rm_grid;
3688
3689         if_ringmap_align(dev, rm0, rm1);
3690
3691         /*
3692          * Re-shuffle rings to get more even distribution.
3693          *
3694          * e.g. 12 netisrs, rm0 contains 4 rings, rm1 contains 2 rings.
3695          *
3696          * CPU       0  1  2  3   4  5  6  7   8  9 10 11
3697          *
3698          * NIC_RX   a0 a1 a2 a3  b0 b1 b2 b3  c0 c1 c2 c3
3699          * NIC_TX   A0 A1        B0 B1        C0 C1
3700          *
3701          * NIC_RX   d0 d1 d2 d3  e0 e1 e2 e3  f0 f1 f2 f3
3702          * NIC_TX         D0 D1        E0 E1        F0 F1
3703          */
3704
3705         if (rm0->rm_cnt >= (2 * old_rm1_grid)) {
3706                 cnt = rm0->rm_cnt;
3707                 subset_grid = old_rm1_grid;
3708                 subset_rm = rm1;
3709                 rm = rm0;
3710         } else if (rm1->rm_cnt > (2 * old_rm0_grid)) {
3711                 cnt = rm1->rm_cnt;
3712                 subset_grid = old_rm0_grid;
3713                 subset_rm = rm0;
3714                 rm = rm1;
3715         } else {
3716                 /* No space to shuffle. */
3717                 return;
3718         }
3719
3720         mod = cnt / subset_grid;
3721         KKASSERT(mod >= 2);
3722         divisor = netisr_ncpus / rm->rm_grid;
3723         offset = ((device_get_unit(dev) / divisor) % mod) * subset_grid;
3724
3725         for (i = 0; i < subset_rm->rm_cnt; ++i) {
3726                 subset_rm->rm_cpumap[i] += offset;
3727                 KASSERT(subset_rm->rm_cpumap[i] < netisr_ncpus,
3728                     ("match: invalid cpumap[%d] = %d, offset %d",
3729                      i, subset_rm->rm_cpumap[i], offset));
3730         }
3731 #ifdef INVARIANTS
3732         for (i = 0; i < subset_rm->rm_cnt; ++i) {
3733                 int j;
3734
3735                 for (j = 0; j < rm->rm_cnt; ++j) {
3736                         if (rm->rm_cpumap[j] == subset_rm->rm_cpumap[i])
3737                                 break;
3738                 }
3739                 KASSERT(j < rm->rm_cnt,
3740                     ("subset cpumap[%d] = %d not found in superset",
3741                      i, subset_rm->rm_cpumap[i]));
3742         }
3743 #endif
3744 }
3745
3746 int
3747 if_ringmap_count(const struct if_ringmap *rm)
3748 {
3749
3750         return (rm->rm_cnt);
3751 }
3752
3753 int
3754 if_ringmap_cpumap(const struct if_ringmap *rm, int ring)
3755 {
3756
3757         KASSERT(ring >= 0 && ring < rm->rm_cnt, ("invalid ring %d", ring));
3758         return (rm->rm_cpumap[ring]);
3759 }
3760
3761 void
3762 if_ringmap_rdrtable(const struct if_ringmap *rm, int table[], int table_nent)
3763 {
3764         int i, grid_idx, grid_cnt, patch_off, patch_cnt, ncopy;
3765
3766         KASSERT(table_nent > 0 && (table_nent & NETISR_CPUMASK) == 0,
3767             ("invalid redirect table entries %d", table_nent));
3768
3769         grid_idx = 0;
3770         for (i = 0; i < NETISR_CPUMAX; ++i) {
3771                 table[i] = grid_idx++ % rm->rm_cnt;
3772
3773                 if (grid_idx == rm->rm_grid)
3774                         grid_idx = 0;
3775         }
3776
3777         /*
3778          * Make the ring distributed more evenly for the remainder
3779          * of each grid.
3780          *
3781          * e.g. 12 netisrs, rm contains 8 rings.
3782          *
3783          * Redirect table before:
3784          *
3785          *  0  1  2  3  4  5  6  7  0  1  2  3  0  1  2  3
3786          *  4  5  6  7  0  1  2  3  0  1  2  3  4  5  6  7
3787          *  0  1  2  3  0  1  2  3  4  5  6  7  0  1  2  3
3788          *  ....
3789          *
3790          * Redirect table after being patched (pX, patched entries):
3791          *
3792          *  0  1  2  3  4  5  6  7 p0 p1 p2 p3  0  1  2  3
3793          *  4  5  6  7 p4 p5 p6 p7  0  1  2  3  4  5  6  7
3794          * p0 p1 p2 p3  0  1  2  3  4  5  6  7 p4 p5 p6 p7
3795          *  ....
3796          */
3797         patch_cnt = rm->rm_grid % rm->rm_cnt;
3798         if (patch_cnt == 0)
3799                 goto done;
3800         patch_off = rm->rm_grid - (rm->rm_grid % rm->rm_cnt);
3801
3802         grid_cnt = roundup(NETISR_CPUMAX, rm->rm_grid) / rm->rm_grid;
3803         grid_idx = 0;
3804         for (i = 0; i < grid_cnt; ++i) {
3805                 int j;
3806
3807                 for (j = 0; j < patch_cnt; ++j) {
3808                         int fix_idx;
3809
3810                         fix_idx = (i * rm->rm_grid) + patch_off + j;
3811                         if (fix_idx >= NETISR_CPUMAX)
3812                                 goto done;
3813                         table[fix_idx] = grid_idx++ % rm->rm_cnt;
3814                 }
3815         }
3816 done:
3817         /*
3818          * If the device supports larger redirect table, duplicate
3819          * the first NETISR_CPUMAX entries to the rest of the table,
3820          * so that it matches upper layer's expectation:
3821          * (hash & NETISR_CPUMASK) % netisr_ncpus
3822          */
3823         ncopy = table_nent / NETISR_CPUMAX;
3824         for (i = 1; i < ncopy; ++i) {
3825                 memcpy(&table[i * NETISR_CPUMAX], table,
3826                     NETISR_CPUMAX * sizeof(table[0]));
3827         }
3828         if (if_ringmap_dumprdr) {
3829                 for (i = 0; i < table_nent; ++i) {
3830                         if (i != 0 && i % 16 == 0)
3831                                 kprintf("\n");
3832                         kprintf("%03d ", table[i]);
3833                 }
3834                 kprintf("\n");
3835         }
3836 }
3837
3838 int
3839 if_ringmap_cpumap_sysctl(SYSCTL_HANDLER_ARGS)
3840 {
3841         struct if_ringmap *rm = arg1;
3842         int i, error = 0;
3843
3844         for (i = 0; i < rm->rm_cnt; ++i) {
3845                 int cpu = rm->rm_cpumap[i];
3846
3847                 error = SYSCTL_OUT(req, &cpu, sizeof(cpu));
3848                 if (error)
3849                         break;
3850         }
3851         return (error);
3852 }