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 = 16;
 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
 226         callout_init_mp(&if_slowtimo_timer);
 227         netmsg_init(&if_slowtimo_netmsg, NULL, &netisr_adone_rport,
 228             MSGF_PRIORITY, if_slowtimo_dispatch);
 229
 230         /* Start if_slowtimo */
 231         lwkt_sendmsg(netisr_cpuport(0), &if_slowtimo_netmsg.lmsg);
 232 }
 233
 234 static void
 235 ifsq_ifstart_ipifunc(void *arg)
 236 {
 237         struct ifaltq_subque *ifsq = arg;
 238         struct lwkt_msg *lmsg = ifsq_get_ifstart_lmsg(ifsq, mycpuid);
 239
 240         crit_enter();
 241         if (lmsg->ms_flags & MSGF_DONE)
 242                 lwkt_sendmsg_oncpu(netisr_cpuport(mycpuid), lmsg);
 243         crit_exit();
 244 }
 245
 246 static __inline void
 247 ifsq_stage_remove(struct ifsubq_stage_head *head, struct ifsubq_stage *stage)
 248 {
 249         KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
 250         TAILQ_REMOVE(&head->stg_head, stage, stg_link);
 251         stage->stg_flags &= ~(IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED);
 252         stage->stg_cnt = 0;
 253         stage->stg_len = 0;
 254 }
 255
 256 static __inline void
 257 ifsq_stage_insert(struct ifsubq_stage_head *head, struct ifsubq_stage *stage)
 258 {
 259         KKASSERT((stage->stg_flags &
 260             (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
 261         stage->stg_flags |= IFSQ_STAGE_FLAG_QUED;
 262         TAILQ_INSERT_TAIL(&head->stg_head, stage, stg_link);
 263 }
 264
 265 /*
 266  * Schedule ifnet.if_start on the subqueue owner CPU
 267  */
 268 static void
 269 ifsq_ifstart_schedule(struct ifaltq_subque *ifsq, int force)
 270 {
 271         int cpu;
 272
 273         if (!force && curthread->td_type == TD_TYPE_NETISR &&
 274             ifsq_stage_cntmax > 0) {
 275                 struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid);
 276
 277                 stage->stg_cnt = 0;
 278                 stage->stg_len = 0;
 279                 if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
 280                         ifsq_stage_insert(&ifsubq_stage_heads[mycpuid], stage);
 281                 stage->stg_flags |= IFSQ_STAGE_FLAG_SCHED;
 282                 return;
 283         }
 284
 285         cpu = ifsq_get_cpuid(ifsq);
 286         if (cpu != mycpuid)
 287                 lwkt_send_ipiq(globaldata_find(cpu), ifsq_ifstart_ipifunc, ifsq);
 288         else
 289                 ifsq_ifstart_ipifunc(ifsq);
 290 }
 291
 292 /*
 293  * NOTE:
 294  * This function will release ifnet.if_start subqueue interlock,
 295  * if ifnet.if_start for the subqueue does not need to be scheduled
 296  */
 297 static __inline int
 298 ifsq_ifstart_need_schedule(struct ifaltq_subque *ifsq, int running)
 299 {
 300         if (!running || ifsq_is_empty(ifsq)
 301 #ifdef ALTQ
 302             || ifsq->ifsq_altq->altq_tbr != NULL
 303 #endif
 304         ) {
 305                 ALTQ_SQ_LOCK(ifsq);
 306                 /*
 307                  * ifnet.if_start subqueue interlock is released, if:
 308                  * 1) Hardware can not take any packets, due to
 309                  *    o  interface is marked down
 310                  *    o  hardware queue is full (ifsq_is_oactive)
 311                  *    Under the second situation, hardware interrupt
 312                  *    or polling(4) will call/schedule ifnet.if_start
 313                  *    on the subqueue when hardware queue is ready
 314                  * 2) There is no packet in the subqueue.
 315                  *    Further ifq_dispatch or ifq_handoff will call/
 316                  *    schedule ifnet.if_start on the subqueue.
 317                  * 3) TBR is used and it does not allow further
 318                  *    dequeueing.
 319                  *    TBR callout will call ifnet.if_start on the
 320                  *    subqueue.
 321                  */
 322                 if (!running || !ifsq_data_ready(ifsq)) {
 323                         ifsq_clr_started(ifsq);
 324                         ALTQ_SQ_UNLOCK(ifsq);
 325                         return 0;
 326                 }
 327                 ALTQ_SQ_UNLOCK(ifsq);
 328         }
 329         return 1;
 330 }
 331
 332 static void
 333 ifsq_ifstart_dispatch(netmsg_t msg)
 334 {
 335         struct lwkt_msg *lmsg = &msg->base.lmsg;
 336         struct ifaltq_subque *ifsq = lmsg->u.ms_resultp;
 337         struct ifnet *ifp = ifsq_get_ifp(ifsq);
 338         struct globaldata *gd = mycpu;
 339         int running = 0, need_sched;
 340
 341         crit_enter_gd(gd);
 342
 343         lwkt_replymsg(lmsg, 0); /* reply ASAP */
 344
 345         if (gd->gd_cpuid != ifsq_get_cpuid(ifsq)) {
 346                 /*
 347                  * We need to chase the subqueue owner CPU change.
 348                  */
 349                 ifsq_ifstart_schedule(ifsq, 1);
 350                 crit_exit_gd(gd);
 351                 return;
 352         }
 353
 354         ifsq_serialize_hw(ifsq);
 355         if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
 356                 ifp->if_start(ifp, ifsq);
 357                 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
 358                         running = 1;
 359         }
 360         need_sched = ifsq_ifstart_need_schedule(ifsq, running);
 361         ifsq_deserialize_hw(ifsq);
 362
 363         if (need_sched) {
 364                 /*
 365                  * More data need to be transmitted, ifnet.if_start is
 366                  * scheduled on the subqueue owner CPU, and we keep going.
 367                  * NOTE: ifnet.if_start subqueue interlock is not released.
 368                  */
 369                 ifsq_ifstart_schedule(ifsq, 0);
 370         }
 371
 372         crit_exit_gd(gd);
 373 }
 374
 375 /* Device driver ifnet.if_start helper function */
 376 void
 377 ifsq_devstart(struct ifaltq_subque *ifsq)
 378 {
 379         struct ifnet *ifp = ifsq_get_ifp(ifsq);
 380         int running = 0;
 381
 382         ASSERT_ALTQ_SQ_SERIALIZED_HW(ifsq);
 383
 384         ALTQ_SQ_LOCK(ifsq);
 385         if (ifsq_is_started(ifsq) || !ifsq_data_ready(ifsq)) {
 386                 ALTQ_SQ_UNLOCK(ifsq);
 387                 return;
 388         }
 389         ifsq_set_started(ifsq);
 390         ALTQ_SQ_UNLOCK(ifsq);
 391
 392         ifp->if_start(ifp, ifsq);
 393
 394         if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
 395                 running = 1;
 396
 397         if (ifsq_ifstart_need_schedule(ifsq, running)) {
 398                 /*
 399                  * More data need to be transmitted, ifnet.if_start is
 400                  * scheduled on ifnet's CPU, and we keep going.
 401                  * NOTE: ifnet.if_start interlock is not released.
 402                  */
 403                 ifsq_ifstart_schedule(ifsq, 0);
 404         }
 405 }
 406
 407 void
 408 if_devstart(struct ifnet *ifp)
 409 {
 410         ifsq_devstart(ifq_get_subq_default(&ifp->if_snd));
 411 }
 412
 413 /* Device driver ifnet.if_start schedule helper function */
 414 void
 415 ifsq_devstart_sched(struct ifaltq_subque *ifsq)
 416 {
 417         ifsq_ifstart_schedule(ifsq, 1);
 418 }
 419
 420 void
 421 if_devstart_sched(struct ifnet *ifp)
 422 {
 423         ifsq_devstart_sched(ifq_get_subq_default(&ifp->if_snd));
 424 }
 425
 426 static void
 427 if_default_serialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
 428 {
 429         lwkt_serialize_enter(ifp->if_serializer);
 430 }
 431
 432 static void
 433 if_default_deserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
 434 {
 435         lwkt_serialize_exit(ifp->if_serializer);
 436 }
 437
 438 static int
 439 if_default_tryserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
 440 {
 441         return lwkt_serialize_try(ifp->if_serializer);
 442 }
 443
 444 #ifdef INVARIANTS
 445 static void
 446 if_default_serialize_assert(struct ifnet *ifp,
 447                             enum ifnet_serialize slz __unused,
 448                             boolean_t serialized)
 449 {
 450         if (serialized)
 451                 ASSERT_SERIALIZED(ifp->if_serializer);
 452         else
 453                 ASSERT_NOT_SERIALIZED(ifp->if_serializer);
 454 }
 455 #endif
 456
 457 /*
 458  * Attach an interface to the list of "active" interfaces.
 459  *
 460  * The serializer is optional.
 461  */
 462 void
 463 if_attach(struct ifnet *ifp, lwkt_serialize_t serializer)
 464 {
 465         unsigned socksize;
 466         int namelen, masklen;
 467         struct sockaddr_dl *sdl, *sdl_addr;
 468         struct ifaddr *ifa;
 469         struct ifaltq *ifq;
 470         struct ifnet **old_ifindex2ifnet = NULL;
 471         struct ifnet_array *old_ifnet_array;
 472         int i, q, qlen;
 473         char qlenname[64];
 474
 475         static int if_indexlim = 8;
 476
 477         if (ifp->if_serialize != NULL) {
 478                 KASSERT(ifp->if_deserialize != NULL &&
 479                         ifp->if_tryserialize != NULL &&
 480                         ifp->if_serialize_assert != NULL,
 481                         ("serialize functions are partially setup"));
 482
 483                 /*
 484                  * If the device supplies serialize functions,
 485                  * then clear if_serializer to catch any invalid
 486                  * usage of this field.
 487                  */
 488                 KASSERT(serializer == NULL,
 489                         ("both serialize functions and default serializer "
 490                          "are supplied"));
 491                 ifp->if_serializer = NULL;
 492         } else {
 493                 KASSERT(ifp->if_deserialize == NULL &&
 494                         ifp->if_tryserialize == NULL &&
 495                         ifp->if_serialize_assert == NULL,
 496                         ("serialize functions are partially setup"));
 497                 ifp->if_serialize = if_default_serialize;
 498                 ifp->if_deserialize = if_default_deserialize;
 499                 ifp->if_tryserialize = if_default_tryserialize;
 500 #ifdef INVARIANTS
 501                 ifp->if_serialize_assert = if_default_serialize_assert;
 502 #endif
 503
 504                 /*
 505                  * The serializer can be passed in from the device,
 506                  * allowing the same serializer to be used for both
 507                  * the interrupt interlock and the device queue.
 508                  * If not specified, the netif structure will use an
 509                  * embedded serializer.
 510                  */
 511                 if (serializer == NULL) {
 512                         serializer = &ifp->if_default_serializer;
 513                         lwkt_serialize_init(serializer);
 514                 }
 515                 ifp->if_serializer = serializer;
 516         }
 517
 518         /*
 519          * Make if_addrhead available on all CPUs, since they
 520          * could be accessed by any threads.
 521          */
 522         ifp->if_addrheads = kmalloc(ncpus * sizeof(struct ifaddrhead),
 523                                     M_IFADDR, M_WAITOK | M_ZERO);
 524         for (i = 0; i < ncpus; ++i)
 525                 TAILQ_INIT(&ifp->if_addrheads[i]);
 526
 527         TAILQ_INIT(&ifp->if_multiaddrs);
 528         TAILQ_INIT(&ifp->if_groups);
 529         getmicrotime(&ifp->if_lastchange);
 530
 531         /*
 532          * create a Link Level name for this device
 533          */
 534         namelen = strlen(ifp->if_xname);
 535         masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + namelen;
 536         socksize = masklen + ifp->if_addrlen;
 537         if (socksize < sizeof(*sdl))
 538                 socksize = sizeof(*sdl);
 539         socksize = RT_ROUNDUP(socksize);
 540         ifa = ifa_create(sizeof(struct ifaddr) + 2 * socksize);
 541         sdl = sdl_addr = (struct sockaddr_dl *)(ifa + 1);
 542         sdl->sdl_len = socksize;
 543         sdl->sdl_family = AF_LINK;
 544         bcopy(ifp->if_xname, sdl->sdl_data, namelen);
 545         sdl->sdl_nlen = namelen;
 546         sdl->sdl_type = ifp->if_type;
 547         ifp->if_lladdr = ifa;
 548         ifa->ifa_ifp = ifp;
 549         ifa->ifa_rtrequest = link_rtrequest;
 550         ifa->ifa_addr = (struct sockaddr *)sdl;
 551         sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
 552         ifa->ifa_netmask = (struct sockaddr *)sdl;
 553         sdl->sdl_len = masklen;
 554         while (namelen != 0)
 555                 sdl->sdl_data[--namelen] = 0xff;
 556         ifa_iflink(ifa, ifp, 0 /* Insert head */);
 557
 558         /*
 559          * Make if_data available on all CPUs, since they could
 560          * be updated by hardware interrupt routing, which could
 561          * be bound to any CPU.
 562          */
 563         ifp->if_data_pcpu = kmalloc_cachealign(
 564             ncpus * sizeof(struct ifdata_pcpu), M_DEVBUF, M_WAITOK | M_ZERO);
 565
 566         if (ifp->if_mapsubq == NULL)
 567                 ifp->if_mapsubq = ifq_mapsubq_default;
 568
 569         ifq = &ifp->if_snd;
 570         ifq->altq_type = 0;
 571         ifq->altq_disc = NULL;
 572         ifq->altq_flags &= ALTQF_CANTCHANGE;
 573         ifq->altq_tbr = NULL;
 574         ifq->altq_ifp = ifp;
 575
 576         if (ifq->altq_subq_cnt <= 0)
 577                 ifq->altq_subq_cnt = 1;
 578         ifq->altq_subq = kmalloc_cachealign(
 579             ifq->altq_subq_cnt * sizeof(struct ifaltq_subque),
 580             M_DEVBUF, M_WAITOK | M_ZERO);
 581
 582         if (ifq->altq_maxlen == 0) {
 583                 if_printf(ifp, "driver didn't set altq_maxlen\n");
 584                 ifq_set_maxlen(ifq, ifqmaxlen);
 585         }
 586
 587         /* Allow user to override driver's setting. */
 588         ksnprintf(qlenname, sizeof(qlenname), "net.%s.qlenmax", ifp->if_xname);
 589         qlen = -1;
 590         TUNABLE_INT_FETCH(qlenname, &qlen);
 591         if (qlen > 0) {
 592                 if_printf(ifp, "qlenmax -> %d\n", qlen);
 593                 ifq_set_maxlen(ifq, qlen);
 594         }
 595
 596         for (q = 0; q < ifq->altq_subq_cnt; ++q) {
 597                 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
 598
 599                 ALTQ_SQ_LOCK_INIT(ifsq);
 600                 ifsq->ifsq_index = q;
 601
 602                 ifsq->ifsq_altq = ifq;
 603                 ifsq->ifsq_ifp = ifp;
 604
 605                 ifsq->ifsq_maxlen = ifq->altq_maxlen;
 606                 ifsq->ifsq_maxbcnt = ifsq->ifsq_maxlen * MCLBYTES;
 607                 ifsq->ifsq_prepended = NULL;
 608                 ifsq->ifsq_started = 0;
 609                 ifsq->ifsq_hw_oactive = 0;
 610                 ifsq_set_cpuid(ifsq, 0);
 611                 if (ifp->if_serializer != NULL)
 612                         ifsq_set_hw_serialize(ifsq, ifp->if_serializer);
 613
 614                 /* XXX: netisr_ncpus */
 615                 ifsq->ifsq_stage =
 616                     kmalloc_cachealign(ncpus * sizeof(struct ifsubq_stage),
 617                     M_DEVBUF, M_WAITOK | M_ZERO);
 618                 for (i = 0; i < ncpus; ++i)
 619                         ifsq->ifsq_stage[i].stg_subq = ifsq;
 620
 621                 /*
 622                  * Allocate one if_start message for each CPU, since
 623                  * the hardware TX ring could be assigned to any CPU.
 624                  *
 625                  * NOTE:
 626                  * If the hardware TX ring polling CPU and the hardware
 627                  * TX ring interrupt CPU are same, one if_start message
 628                  * should be enough.
 629                  */
 630                 ifsq->ifsq_ifstart_nmsg =
 631                     kmalloc(ncpus * sizeof(struct netmsg_base),
 632                     M_LWKTMSG, M_WAITOK);
 633                 for (i = 0; i < ncpus; ++i) {
 634                         netmsg_init(&ifsq->ifsq_ifstart_nmsg[i], NULL,
 635                             &netisr_adone_rport, 0, ifsq_ifstart_dispatch);
 636                         ifsq->ifsq_ifstart_nmsg[i].lmsg.u.ms_resultp = ifsq;
 637                 }
 638         }
 639         ifq_set_classic(ifq);
 640
 641         /*
 642          * Increase mbuf cluster/jcluster limits for the mbufs that
 643          * could sit on the device queues for quite some time.
 644          */
 645         if (ifp->if_nmbclusters > 0)
 646                 mcl_inclimit(ifp->if_nmbclusters);
 647         if (ifp->if_nmbjclusters > 0)
 648                 mjcl_inclimit(ifp->if_nmbjclusters);
 649
 650         /*
 651          * Install this ifp into ifindex2inet, ifnet queue and ifnet
 652          * array after it is setup.
 653          *
 654          * Protect ifindex2ifnet, ifnet queue and ifnet array changes
 655          * by ifnet lock, so that non-netisr threads could get a
 656          * consistent view.
 657          */
 658         ifnet_lock();
 659
 660         /* Don't update if_index until ifindex2ifnet is setup */
 661         ifp->if_index = if_index + 1;
 662         sdl_addr->sdl_index = ifp->if_index;
 663
 664         /*
 665          * Install this ifp into ifindex2ifnet
 666          */
 667         if (ifindex2ifnet == NULL || ifp->if_index >= if_indexlim) {
 668                 unsigned int n;
 669                 struct ifnet **q;
 670
 671                 /*
 672                  * Grow ifindex2ifnet
 673                  */
 674                 if_indexlim <<= 1;
 675                 n = if_indexlim * sizeof(*q);
 676                 q = kmalloc(n, M_IFADDR, M_WAITOK | M_ZERO);
 677                 if (ifindex2ifnet != NULL) {
 678                         bcopy(ifindex2ifnet, q, n/2);
 679                         /* Free old ifindex2ifnet after sync all netisrs */
 680                         old_ifindex2ifnet = ifindex2ifnet;
 681                 }
 682                 ifindex2ifnet = q;
 683         }
 684         ifindex2ifnet[ifp->if_index] = ifp;
 685         /*
 686          * Update if_index after this ifp is installed into ifindex2ifnet,
 687          * so that netisrs could get a consistent view of ifindex2ifnet.
 688          */
 689         cpu_sfence();
 690         if_index = ifp->if_index;
 691
 692         /*
 693          * Install this ifp into ifnet array.
 694          */
 695         /* Free old ifnet array after sync all netisrs */
 696         old_ifnet_array = ifnet_array;
 697         ifnet_array = ifnet_array_add(ifp, old_ifnet_array);
 698
 699         /*
 700          * Install this ifp into ifnet queue.
 701          */
 702         TAILQ_INSERT_TAIL(&ifnetlist, ifp, if_link);
 703
 704         ifnet_unlock();
 705
 706         /*
 707          * Sync all netisrs so that the old ifindex2ifnet and ifnet array
 708          * are no longer accessed and we can free them safely later on.
 709          */
 710         netmsg_service_sync();
 711         if (old_ifindex2ifnet != NULL)
 712                 kfree(old_ifindex2ifnet, M_IFADDR);
 713         ifnet_array_free(old_ifnet_array);
 714
 715         if (!SLIST_EMPTY(&domains))
 716                 if_attachdomain1(ifp);
 717
 718         /* Announce the interface. */
 719         EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
 720         devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
 721         rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
 722 }
 723
 724 static void
 725 if_attachdomain(void *dummy)
 726 {
 727         struct ifnet *ifp;
 728
 729         ifnet_lock();
 730         TAILQ_FOREACH(ifp, &ifnetlist, if_list)
 731                 if_attachdomain1(ifp);
 732         ifnet_unlock();
 733 }
 734 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_FIRST,
 735         if_attachdomain, NULL);
 736
 737 static void
 738 if_attachdomain1(struct ifnet *ifp)
 739 {
 740         struct domain *dp;
 741
 742         crit_enter();
 743
 744         /* address family dependent data region */
 745         bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
 746         SLIST_FOREACH(dp, &domains, dom_next)
 747                 if (dp->dom_ifattach)
 748                         ifp->if_afdata[dp->dom_family] =
 749                                 (*dp->dom_ifattach)(ifp);
 750         crit_exit();
 751 }
 752
 753 /*
 754  * Purge all addresses whose type is _not_ AF_LINK
 755  */
 756 static void
 757 if_purgeaddrs_nolink_dispatch(netmsg_t nmsg)
 758 {
 759         struct ifnet *ifp = nmsg->lmsg.u.ms_resultp;
 760         struct ifaddr_container *ifac, *next;
 761
 762         ASSERT_NETISR0;
 763
 764         /*
 765          * The ifaddr processing in the following loop will block,
 766          * however, this function is called in netisr0, in which
 767          * ifaddr list changes happen, so we don't care about the
 768          * blockness of the ifaddr processing here.
 769          */
 770         TAILQ_FOREACH_MUTABLE(ifac, &ifp->if_addrheads[mycpuid],
 771                               ifa_link, next) {
 772                 struct ifaddr *ifa = ifac->ifa;
 773
 774                 /* Ignore marker */
 775                 if (ifa->ifa_addr->sa_family == AF_UNSPEC)
 776                         continue;
 777
 778                 /* Leave link ifaddr as it is */
 779                 if (ifa->ifa_addr->sa_family == AF_LINK)
 780                         continue;
 781 #ifdef INET
 782                 /* XXX: Ugly!! ad hoc just for INET */
 783                 if (ifa->ifa_addr->sa_family == AF_INET) {
 784                         struct ifaliasreq ifr;
 785                         struct sockaddr_in saved_addr, saved_dst;
 786 #ifdef IFADDR_DEBUG_VERBOSE
 787                         int i;
 788
 789                         kprintf("purge in4 addr %p: ", ifa);
 790                         for (i = 0; i < ncpus; ++i) {
 791                                 kprintf("%d ",
 792                                     ifa->ifa_containers[i].ifa_refcnt);
 793                         }
 794                         kprintf("\n");
 795 #endif
 796
 797                         /* Save information for panic. */
 798                         memcpy(&saved_addr, ifa->ifa_addr, sizeof(saved_addr));
 799                         if (ifa->ifa_dstaddr != NULL) {
 800                                 memcpy(&saved_dst, ifa->ifa_dstaddr,
 801                                     sizeof(saved_dst));
 802                         } else {
 803                                 memset(&saved_dst, 0, sizeof(saved_dst));
 804                         }
 805
 806                         bzero(&ifr, sizeof ifr);
 807                         ifr.ifra_addr = *ifa->ifa_addr;
 808                         if (ifa->ifa_dstaddr)
 809                                 ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
 810                         if (in_control(SIOCDIFADDR, (caddr_t)&ifr, ifp,
 811                                        NULL) == 0)
 812                                 continue;
 813
 814                         /* MUST NOT HAPPEN */
 815                         panic("%s: in_control failed %x, dst %x", ifp->if_xname,
 816                             ntohl(saved_addr.sin_addr.s_addr),
 817                             ntohl(saved_dst.sin_addr.s_addr));
 818                 }
 819 #endif /* INET */
 820 #ifdef INET6
 821                 if (ifa->ifa_addr->sa_family == AF_INET6) {
 822 #ifdef IFADDR_DEBUG_VERBOSE
 823                         int i;
 824
 825                         kprintf("purge in6 addr %p: ", ifa);
 826                         for (i = 0; i < ncpus; ++i) {
 827                                 kprintf("%d ",
 828                                     ifa->ifa_containers[i].ifa_refcnt);
 829                         }
 830                         kprintf("\n");
 831 #endif
 832
 833                         in6_purgeaddr(ifa);
 834                         /* ifp_addrhead is already updated */
 835                         continue;
 836                 }
 837 #endif /* INET6 */
 838                 if_printf(ifp, "destroy ifaddr family %d\n",
 839                     ifa->ifa_addr->sa_family);
 840                 ifa_ifunlink(ifa, ifp);
 841                 ifa_destroy(ifa);
 842         }
 843
 844         netisr_replymsg(&nmsg->base, 0);
 845 }
 846
 847 void
 848 if_purgeaddrs_nolink(struct ifnet *ifp)
 849 {
 850         struct netmsg_base nmsg;
 851
 852         netmsg_init(&nmsg, NULL, &curthread->td_msgport, 0,
 853             if_purgeaddrs_nolink_dispatch);
 854         nmsg.lmsg.u.ms_resultp = ifp;
 855         netisr_domsg(&nmsg, 0);
 856 }
 857
 858 static void
 859 ifq_stage_detach_handler(netmsg_t nmsg)
 860 {
 861         struct ifaltq *ifq = nmsg->lmsg.u.ms_resultp;
 862         int q;
 863
 864         for (q = 0; q < ifq->altq_subq_cnt; ++q) {
 865                 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
 866                 struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid);
 867
 868                 if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED)
 869                         ifsq_stage_remove(&ifsubq_stage_heads[mycpuid], stage);
 870         }
 871         lwkt_replymsg(&nmsg->lmsg, 0);
 872 }
 873
 874 static void
 875 ifq_stage_detach(struct ifaltq *ifq)
 876 {
 877         struct netmsg_base base;
 878         int cpu;
 879
 880         netmsg_init(&base, NULL, &curthread->td_msgport, 0,
 881             ifq_stage_detach_handler);
 882         base.lmsg.u.ms_resultp = ifq;
 883
 884         /* XXX netisr_ncpus */
 885         for (cpu = 0; cpu < ncpus; ++cpu)
 886                 lwkt_domsg(netisr_cpuport(cpu), &base.lmsg, 0);
 887 }
 888
 889 struct netmsg_if_rtdel {
 890         struct netmsg_base      base;
 891         struct ifnet            *ifp;
 892 };
 893
 894 static void
 895 if_rtdel_dispatch(netmsg_t msg)
 896 {
 897         struct netmsg_if_rtdel *rmsg = (void *)msg;
 898         int i, cpu;
 899
 900         cpu = mycpuid;
 901         ASSERT_NETISR_NCPUS(cpu);
 902
 903         for (i = 1; i <= AF_MAX; i++) {
 904                 struct radix_node_head  *rnh;
 905
 906                 if ((rnh = rt_tables[cpu][i]) == NULL)
 907                         continue;
 908                 rnh->rnh_walktree(rnh, if_rtdel, rmsg->ifp);
 909         }
 910         netisr_forwardmsg(&msg->base, cpu + 1);
 911 }
 912
 913 /*
 914  * Detach an interface, removing it from the
 915  * list of "active" interfaces.
 916  */
 917 void
 918 if_detach(struct ifnet *ifp)
 919 {
 920         struct ifnet_array *old_ifnet_array;
 921         struct netmsg_if_rtdel msg;
 922         struct domain *dp;
 923         int q;
 924
 925         /* Announce that the interface is gone. */
 926         EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
 927         rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
 928         devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
 929
 930         /*
 931          * Remove this ifp from ifindex2inet, ifnet queue and ifnet
 932          * array before it is whacked.
 933          *
 934          * Protect ifindex2ifnet, ifnet queue and ifnet array changes
 935          * by ifnet lock, so that non-netisr threads could get a
 936          * consistent view.
 937          */
 938         ifnet_lock();
 939
 940         /*
 941          * Remove this ifp from ifindex2ifnet and maybe decrement if_index.
 942          */
 943         ifindex2ifnet[ifp->if_index] = NULL;
 944         while (if_index > 0 && ifindex2ifnet[if_index] == NULL)
 945                 if_index--;
 946
 947         /*
 948          * Remove this ifp from ifnet queue.
 949          */
 950         TAILQ_REMOVE(&ifnetlist, ifp, if_link);
 951
 952         /*
 953          * Remove this ifp from ifnet array.
 954          */
 955         /* Free old ifnet array after sync all netisrs */
 956         old_ifnet_array = ifnet_array;
 957         ifnet_array = ifnet_array_del(ifp, old_ifnet_array);
 958
 959         ifnet_unlock();
 960
 961         /*
 962          * Sync all netisrs so that the old ifnet array is no longer
 963          * accessed and we can free it safely later on.
 964          */
 965         netmsg_service_sync();
 966         ifnet_array_free(old_ifnet_array);
 967
 968         /*
 969          * Remove routes and flush queues.
 970          */
 971         crit_enter();
 972 #ifdef IFPOLL_ENABLE
 973         if (ifp->if_flags & IFF_NPOLLING)
 974                 ifpoll_deregister(ifp);
 975 #endif
 976         if_down(ifp);
 977
 978         /* Decrease the mbuf clusters/jclusters limits increased by us */
 979         if (ifp->if_nmbclusters > 0)
 980                 mcl_inclimit(-ifp->if_nmbclusters);
 981         if (ifp->if_nmbjclusters > 0)
 982                 mjcl_inclimit(-ifp->if_nmbjclusters);
 983
 984 #ifdef ALTQ
 985         if (ifq_is_enabled(&ifp->if_snd))
 986                 altq_disable(&ifp->if_snd);
 987         if (ifq_is_attached(&ifp->if_snd))
 988                 altq_detach(&ifp->if_snd);
 989 #endif
 990
 991         /*
 992          * Clean up all addresses.
 993          */
 994         ifp->if_lladdr = NULL;
 995
 996         if_purgeaddrs_nolink(ifp);
 997         if (!TAILQ_EMPTY(&ifp->if_addrheads[mycpuid])) {
 998                 struct ifaddr *ifa;
 999
1000                 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
1001                 KASSERT(ifa->ifa_addr->sa_family == AF_LINK,
1002                         ("non-link ifaddr is left on if_addrheads"));
1003
1004                 ifa_ifunlink(ifa, ifp);
1005                 ifa_destroy(ifa);
1006                 KASSERT(TAILQ_EMPTY(&ifp->if_addrheads[mycpuid]),
1007                         ("there are still ifaddrs left on if_addrheads"));
1008         }
1009
1010 #ifdef INET
1011         /*
1012          * Remove all IPv4 kernel structures related to ifp.
1013          */
1014         in_ifdetach(ifp);
1015 #endif
1016
1017 #ifdef INET6
1018         /*
1019          * Remove all IPv6 kernel structs related to ifp.  This should be done
1020          * before removing routing entries below, since IPv6 interface direct
1021          * routes are expected to be removed by the IPv6-specific kernel API.
1022          * Otherwise, the kernel will detect some inconsistency and bark it.
1023          */
1024         in6_ifdetach(ifp);
1025 #endif
1026
1027         /*
1028          * Delete all remaining routes using this interface
1029          */
1030         netmsg_init(&msg.base, NULL, &curthread->td_msgport, MSGF_PRIORITY,
1031             if_rtdel_dispatch);
1032         msg.ifp = ifp;
1033         netisr_domsg_global(&msg.base);
1034
1035         SLIST_FOREACH(dp, &domains, dom_next)
1036                 if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
1037                         (*dp->dom_ifdetach)(ifp,
1038                                 ifp->if_afdata[dp->dom_family]);
1039
1040         kfree(ifp->if_addrheads, M_IFADDR);
1041
1042         lwkt_synchronize_ipiqs("if_detach");
1043         ifq_stage_detach(&ifp->if_snd);
1044
1045         for (q = 0; q < ifp->if_snd.altq_subq_cnt; ++q) {
1046                 struct ifaltq_subque *ifsq = &ifp->if_snd.altq_subq[q];
1047
1048                 kfree(ifsq->ifsq_ifstart_nmsg, M_LWKTMSG);
1049                 kfree(ifsq->ifsq_stage, M_DEVBUF);
1050         }
1051         kfree(ifp->if_snd.altq_subq, M_DEVBUF);
1052
1053         kfree(ifp->if_data_pcpu, M_DEVBUF);
1054
1055         crit_exit();
1056 }
1057
1058 /*
1059  * Create interface group without members
1060  */
1061 struct ifg_group *
1062 if_creategroup(const char *groupname)
1063 {
1064         struct ifg_group        *ifg = NULL;
1065
1066         if ((ifg = (struct ifg_group *)kmalloc(sizeof(struct ifg_group),
1067             M_TEMP, M_NOWAIT)) == NULL)
1068                 return (NULL);
1069
1070         strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group));
1071         ifg->ifg_refcnt = 0;
1072         ifg->ifg_carp_demoted = 0;
1073         TAILQ_INIT(&ifg->ifg_members);
1074 #if NPF > 0
1075         pfi_attach_ifgroup(ifg);
1076 #endif
1077         TAILQ_INSERT_TAIL(&ifg_head, ifg, ifg_next);
1078
1079         return (ifg);
1080 }
1081
1082 /*
1083  * Add a group to an interface
1084  */
1085 int
1086 if_addgroup(struct ifnet *ifp, const char *groupname)
1087 {
1088         struct ifg_list         *ifgl;
1089         struct ifg_group        *ifg = NULL;
1090         struct ifg_member       *ifgm;
1091
1092         if (groupname[0] && groupname[strlen(groupname) - 1] >= '0' &&
1093             groupname[strlen(groupname) - 1] <= '9')
1094                 return (EINVAL);
1095
1096         TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1097                 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
1098                         return (EEXIST);
1099
1100         if ((ifgl = kmalloc(sizeof(*ifgl), M_TEMP, M_NOWAIT)) == NULL)
1101                 return (ENOMEM);
1102
1103         if ((ifgm = kmalloc(sizeof(*ifgm), M_TEMP, M_NOWAIT)) == NULL) {
1104                 kfree(ifgl, M_TEMP);
1105                 return (ENOMEM);
1106         }
1107
1108         TAILQ_FOREACH(ifg, &ifg_head, ifg_next)
1109                 if (!strcmp(ifg->ifg_group, groupname))
1110                         break;
1111
1112         if (ifg == NULL && (ifg = if_creategroup(groupname)) == NULL) {
1113                 kfree(ifgl, M_TEMP);
1114                 kfree(ifgm, M_TEMP);
1115                 return (ENOMEM);
1116         }
1117
1118         ifg->ifg_refcnt++;
1119         ifgl->ifgl_group = ifg;
1120         ifgm->ifgm_ifp = ifp;
1121
1122         TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next);
1123         TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next);
1124
1125 #if NPF > 0
1126         pfi_group_change(groupname);
1127 #endif
1128
1129         return (0);
1130 }
1131
1132 /*
1133  * Remove a group from an interface
1134  */
1135 int
1136 if_delgroup(struct ifnet *ifp, const char *groupname)
1137 {
1138         struct ifg_list         *ifgl;
1139         struct ifg_member       *ifgm;
1140
1141         TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1142                 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
1143                         break;
1144         if (ifgl == NULL)
1145                 return (ENOENT);
1146
1147         TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
1148
1149         TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
1150                 if (ifgm->ifgm_ifp == ifp)
1151                         break;
1152
1153         if (ifgm != NULL) {
1154                 TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next);
1155                 kfree(ifgm, M_TEMP);
1156         }
1157
1158         if (--ifgl->ifgl_group->ifg_refcnt == 0) {
1159                 TAILQ_REMOVE(&ifg_head, ifgl->ifgl_group, ifg_next);
1160 #if NPF > 0
1161                 pfi_detach_ifgroup(ifgl->ifgl_group);
1162 #endif
1163                 kfree(ifgl->ifgl_group, M_TEMP);
1164         }
1165
1166         kfree(ifgl, M_TEMP);
1167
1168 #if NPF > 0
1169         pfi_group_change(groupname);
1170 #endif
1171
1172         return (0);
1173 }
1174
1175 /*
1176  * Stores all groups from an interface in memory pointed
1177  * to by data
1178  */
1179 int
1180 if_getgroup(caddr_t data, struct ifnet *ifp)
1181 {
1182         int                      len, error;
1183         struct ifg_list         *ifgl;
1184         struct ifg_req           ifgrq, *ifgp;
1185         struct ifgroupreq       *ifgr = (struct ifgroupreq *)data;
1186
1187         if (ifgr->ifgr_len == 0) {
1188                 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1189                         ifgr->ifgr_len += sizeof(struct ifg_req);
1190                 return (0);
1191         }
1192
1193         len = ifgr->ifgr_len;
1194         ifgp = ifgr->ifgr_groups;
1195         TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1196                 if (len < sizeof(ifgrq))
1197                         return (EINVAL);
1198                 bzero(&ifgrq, sizeof ifgrq);
1199                 strlcpy(ifgrq.ifgrq_group, ifgl->ifgl_group->ifg_group,
1200                     sizeof(ifgrq.ifgrq_group));
1201                 if ((error = copyout((caddr_t)&ifgrq, (caddr_t)ifgp,
1202                     sizeof(struct ifg_req))))
1203                         return (error);
1204                 len -= sizeof(ifgrq);
1205                 ifgp++;
1206         }
1207
1208         return (0);
1209 }
1210
1211 /*
1212  * Stores all members of a group in memory pointed to by data
1213  */
1214 int
1215 if_getgroupmembers(caddr_t data)
1216 {
1217         struct ifgroupreq       *ifgr = (struct ifgroupreq *)data;
1218         struct ifg_group        *ifg;
1219         struct ifg_member       *ifgm;
1220         struct ifg_req           ifgrq, *ifgp;
1221         int                      len, error;
1222
1223         TAILQ_FOREACH(ifg, &ifg_head, ifg_next)
1224                 if (!strcmp(ifg->ifg_group, ifgr->ifgr_name))
1225                         break;
1226         if (ifg == NULL)
1227                 return (ENOENT);
1228
1229         if (ifgr->ifgr_len == 0) {
1230                 TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
1231                         ifgr->ifgr_len += sizeof(ifgrq);
1232                 return (0);
1233         }
1234
1235         len = ifgr->ifgr_len;
1236         ifgp = ifgr->ifgr_groups;
1237         TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
1238                 if (len < sizeof(ifgrq))
1239                         return (EINVAL);
1240                 bzero(&ifgrq, sizeof ifgrq);
1241                 strlcpy(ifgrq.ifgrq_member, ifgm->ifgm_ifp->if_xname,
1242                     sizeof(ifgrq.ifgrq_member));
1243                 if ((error = copyout((caddr_t)&ifgrq, (caddr_t)ifgp,
1244                     sizeof(struct ifg_req))))
1245                         return (error);
1246                 len -= sizeof(ifgrq);
1247                 ifgp++;
1248         }
1249
1250         return (0);
1251 }
1252
1253 /*
1254  * Delete Routes for a Network Interface
1255  *
1256  * Called for each routing entry via the rnh->rnh_walktree() call above
1257  * to delete all route entries referencing a detaching network interface.
1258  *
1259  * Arguments:
1260  *      rn      pointer to node in the routing table
1261  *      arg     argument passed to rnh->rnh_walktree() - detaching interface
1262  *
1263  * Returns:
1264  *      0       successful
1265  *      errno   failed - reason indicated
1266  *
1267  */
1268 static int
1269 if_rtdel(struct radix_node *rn, void *arg)
1270 {
1271         struct rtentry  *rt = (struct rtentry *)rn;
1272         struct ifnet    *ifp = arg;
1273         int             err;
1274
1275         if (rt->rt_ifp == ifp) {
1276
1277                 /*
1278                  * Protect (sorta) against walktree recursion problems
1279                  * with cloned routes
1280                  */
1281                 if (!(rt->rt_flags & RTF_UP))
1282                         return (0);
1283
1284                 err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1285                                 rt_mask(rt), rt->rt_flags,
1286                                 NULL);
1287                 if (err) {
1288                         log(LOG_WARNING, "if_rtdel: error %d\n", err);
1289                 }
1290         }
1291
1292         return (0);
1293 }
1294
1295 static __inline boolean_t
1296 ifa_prefer(const struct ifaddr *cur_ifa, const struct ifaddr *old_ifa)
1297 {
1298         if (old_ifa == NULL)
1299                 return TRUE;
1300
1301         if ((old_ifa->ifa_ifp->if_flags & IFF_UP) == 0 &&
1302             (cur_ifa->ifa_ifp->if_flags & IFF_UP))
1303                 return TRUE;
1304         if ((old_ifa->ifa_flags & IFA_ROUTE) == 0 &&
1305             (cur_ifa->ifa_flags & IFA_ROUTE))
1306                 return TRUE;
1307         return FALSE;
1308 }
1309
1310 /*
1311  * Locate an interface based on a complete address.
1312  */
1313 struct ifaddr *
1314 ifa_ifwithaddr(struct sockaddr *addr)
1315 {
1316         const struct ifnet_array *arr;
1317         int i;
1318
1319         arr = ifnet_array_get();
1320         for (i = 0; i < arr->ifnet_count; ++i) {
1321                 struct ifnet *ifp = arr->ifnet_arr[i];
1322                 struct ifaddr_container *ifac;
1323
1324                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1325                         struct ifaddr *ifa = ifac->ifa;
1326
1327                         if (ifa->ifa_addr->sa_family != addr->sa_family)
1328                                 continue;
1329                         if (sa_equal(addr, ifa->ifa_addr))
1330                                 return (ifa);
1331                         if ((ifp->if_flags & IFF_BROADCAST) &&
1332                             ifa->ifa_broadaddr &&
1333                             /* IPv6 doesn't have broadcast */
1334                             ifa->ifa_broadaddr->sa_len != 0 &&
1335                             sa_equal(ifa->ifa_broadaddr, addr))
1336                                 return (ifa);
1337                 }
1338         }
1339         return (NULL);
1340 }
1341
1342 /*
1343  * Locate the point to point interface with a given destination address.
1344  */
1345 struct ifaddr *
1346 ifa_ifwithdstaddr(struct sockaddr *addr)
1347 {
1348         const struct ifnet_array *arr;
1349         int i;
1350
1351         arr = ifnet_array_get();
1352         for (i = 0; i < arr->ifnet_count; ++i) {
1353                 struct ifnet *ifp = arr->ifnet_arr[i];
1354                 struct ifaddr_container *ifac;
1355
1356                 if (!(ifp->if_flags & IFF_POINTOPOINT))
1357                         continue;
1358
1359                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1360                         struct ifaddr *ifa = ifac->ifa;
1361
1362                         if (ifa->ifa_addr->sa_family != addr->sa_family)
1363                                 continue;
1364                         if (ifa->ifa_dstaddr &&
1365                             sa_equal(addr, ifa->ifa_dstaddr))
1366                                 return (ifa);
1367                 }
1368         }
1369         return (NULL);
1370 }
1371
1372 /*
1373  * Find an interface on a specific network.  If many, choice
1374  * is most specific found.
1375  */
1376 struct ifaddr *
1377 ifa_ifwithnet(struct sockaddr *addr)
1378 {
1379         struct ifaddr *ifa_maybe = NULL;
1380         u_int af = addr->sa_family;
1381         char *addr_data = addr->sa_data, *cplim;
1382         const struct ifnet_array *arr;
1383         int i;
1384
1385         /*
1386          * AF_LINK addresses can be looked up directly by their index number,
1387          * so do that if we can.
1388          */
1389         if (af == AF_LINK) {
1390                 struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
1391
1392                 if (sdl->sdl_index && sdl->sdl_index <= if_index)
1393                         return (ifindex2ifnet[sdl->sdl_index]->if_lladdr);
1394         }
1395
1396         /*
1397          * Scan though each interface, looking for ones that have
1398          * addresses in this address family.
1399          */
1400         arr = ifnet_array_get();
1401         for (i = 0; i < arr->ifnet_count; ++i) {
1402                 struct ifnet *ifp = arr->ifnet_arr[i];
1403                 struct ifaddr_container *ifac;
1404
1405                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1406                         struct ifaddr *ifa = ifac->ifa;
1407                         char *cp, *cp2, *cp3;
1408
1409                         if (ifa->ifa_addr->sa_family != af)
1410 next:                           continue;
1411                         if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT) {
1412                                 /*
1413                                  * This is a bit broken as it doesn't
1414                                  * take into account that the remote end may
1415                                  * be a single node in the network we are
1416                                  * looking for.
1417                                  * The trouble is that we don't know the
1418                                  * netmask for the remote end.
1419                                  */
1420                                 if (ifa->ifa_dstaddr != NULL &&
1421                                     sa_equal(addr, ifa->ifa_dstaddr))
1422                                         return (ifa);
1423                         } else {
1424                                 /*
1425                                  * if we have a special address handler,
1426                                  * then use it instead of the generic one.
1427                                  */
1428                                 if (ifa->ifa_claim_addr) {
1429                                         if ((*ifa->ifa_claim_addr)(ifa, addr)) {
1430                                                 return (ifa);
1431                                         } else {
1432                                                 continue;
1433                                         }
1434                                 }
1435
1436                                 /*
1437                                  * Scan all the bits in the ifa's address.
1438                                  * If a bit dissagrees with what we are
1439                                  * looking for, mask it with the netmask
1440                                  * to see if it really matters.
1441                                  * (A byte at a time)
1442                                  */
1443                                 if (ifa->ifa_netmask == 0)
1444                                         continue;
1445                                 cp = addr_data;
1446                                 cp2 = ifa->ifa_addr->sa_data;
1447                                 cp3 = ifa->ifa_netmask->sa_data;
1448                                 cplim = ifa->ifa_netmask->sa_len +
1449                                         (char *)ifa->ifa_netmask;
1450                                 while (cp3 < cplim)
1451                                         if ((*cp++ ^ *cp2++) & *cp3++)
1452                                                 goto next; /* next address! */
1453                                 /*
1454                                  * If the netmask of what we just found
1455                                  * is more specific than what we had before
1456                                  * (if we had one) then remember the new one
1457                                  * before continuing to search for an even
1458                                  * better one.  If the netmasks are equal,
1459                                  * we prefer the this ifa based on the result
1460                                  * of ifa_prefer().
1461                                  */
1462                                 if (ifa_maybe == NULL ||
1463                                     rn_refines((char *)ifa->ifa_netmask,
1464                                         (char *)ifa_maybe->ifa_netmask) ||
1465                                     (sa_equal(ifa_maybe->ifa_netmask,
1466                                         ifa->ifa_netmask) &&
1467                                      ifa_prefer(ifa, ifa_maybe)))
1468                                         ifa_maybe = ifa;
1469                         }
1470                 }
1471         }
1472         return (ifa_maybe);
1473 }
1474
1475 /*
1476  * Find an interface address specific to an interface best matching
1477  * a given address.
1478  */
1479 struct ifaddr *
1480 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1481 {
1482         struct ifaddr_container *ifac;
1483         char *cp, *cp2, *cp3;
1484         char *cplim;
1485         struct ifaddr *ifa_maybe = NULL;
1486         u_int af = addr->sa_family;
1487
1488         if (af >= AF_MAX)
1489                 return (0);
1490         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1491                 struct ifaddr *ifa = ifac->ifa;
1492
1493                 if (ifa->ifa_addr->sa_family != af)
1494                         continue;
1495                 if (ifa_maybe == NULL)
1496                         ifa_maybe = ifa;
1497                 if (ifa->ifa_netmask == NULL) {
1498                         if (sa_equal(addr, ifa->ifa_addr) ||
1499                             (ifa->ifa_dstaddr != NULL &&
1500                              sa_equal(addr, ifa->ifa_dstaddr)))
1501                                 return (ifa);
1502                         continue;
1503                 }
1504                 if (ifp->if_flags & IFF_POINTOPOINT) {
1505                         if (sa_equal(addr, ifa->ifa_dstaddr))
1506                                 return (ifa);
1507                 } else {
1508                         cp = addr->sa_data;
1509                         cp2 = ifa->ifa_addr->sa_data;
1510                         cp3 = ifa->ifa_netmask->sa_data;
1511                         cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1512                         for (; cp3 < cplim; cp3++)
1513                                 if ((*cp++ ^ *cp2++) & *cp3)
1514                                         break;
1515                         if (cp3 == cplim)
1516                                 return (ifa);
1517                 }
1518         }
1519         return (ifa_maybe);
1520 }
1521
1522 /*
1523  * Default action when installing a route with a Link Level gateway.
1524  * Lookup an appropriate real ifa to point to.
1525  * This should be moved to /sys/net/link.c eventually.
1526  */
1527 static void
1528 link_rtrequest(int cmd, struct rtentry *rt)
1529 {
1530         struct ifaddr *ifa;
1531         struct sockaddr *dst;
1532         struct ifnet *ifp;
1533
1534         if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL ||
1535             (ifp = ifa->ifa_ifp) == NULL || (dst = rt_key(rt)) == NULL)
1536                 return;
1537         ifa = ifaof_ifpforaddr(dst, ifp);
1538         if (ifa != NULL) {
1539                 IFAFREE(rt->rt_ifa);
1540                 IFAREF(ifa);
1541                 rt->rt_ifa = ifa;
1542                 if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1543                         ifa->ifa_rtrequest(cmd, rt);
1544         }
1545 }
1546
1547 struct netmsg_ifroute {
1548         struct netmsg_base      base;
1549         struct ifnet            *ifp;
1550         int                     flag;
1551         int                     fam;
1552 };
1553
1554 /*
1555  * Mark an interface down and notify protocols of the transition.
1556  */
1557 static void
1558 if_unroute_dispatch(netmsg_t nmsg)
1559 {
1560         struct netmsg_ifroute *msg = (struct netmsg_ifroute *)nmsg;
1561         struct ifnet *ifp = msg->ifp;
1562         int flag = msg->flag, fam = msg->fam;
1563         struct ifaddr_container *ifac;
1564
1565         ASSERT_NETISR0;
1566
1567         ifp->if_flags &= ~flag;
1568         getmicrotime(&ifp->if_lastchange);
1569         /*
1570          * The ifaddr processing in the following loop will block,
1571          * however, this function is called in netisr0, in which
1572          * ifaddr list changes happen, so we don't care about the
1573          * blockness of the ifaddr processing here.
1574          */
1575         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1576                 struct ifaddr *ifa = ifac->ifa;
1577
1578                 /* Ignore marker */
1579                 if (ifa->ifa_addr->sa_family == AF_UNSPEC)
1580                         continue;
1581
1582                 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1583                         kpfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1584         }
1585         ifq_purge_all(&ifp->if_snd);
1586         rt_ifmsg(ifp);
1587
1588         netisr_replymsg(&nmsg->base, 0);
1589 }
1590
1591 void
1592 if_unroute(struct ifnet *ifp, int flag, int fam)
1593 {
1594         struct netmsg_ifroute msg;
1595
1596         netmsg_init(&msg.base, NULL, &curthread->td_msgport, 0,
1597             if_unroute_dispatch);
1598         msg.ifp = ifp;
1599         msg.flag = flag;
1600         msg.fam = fam;
1601         netisr_domsg(&msg.base, 0);
1602 }
1603
1604 /*
1605  * Mark an interface up and notify protocols of the transition.
1606  */
1607 static void
1608 if_route_dispatch(netmsg_t nmsg)
1609 {
1610         struct netmsg_ifroute *msg = (struct netmsg_ifroute *)nmsg;
1611         struct ifnet *ifp = msg->ifp;
1612         int flag = msg->flag, fam = msg->fam;
1613         struct ifaddr_container *ifac;
1614
1615         ASSERT_NETISR0;
1616
1617         ifq_purge_all(&ifp->if_snd);
1618         ifp->if_flags |= flag;
1619         getmicrotime(&ifp->if_lastchange);
1620         /*
1621          * The ifaddr processing in the following loop will block,
1622          * however, this function is called in netisr0, in which
1623          * ifaddr list changes happen, so we don't care about the
1624          * blockness of the ifaddr processing here.
1625          */
1626         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1627                 struct ifaddr *ifa = ifac->ifa;
1628
1629                 /* Ignore marker */
1630                 if (ifa->ifa_addr->sa_family == AF_UNSPEC)
1631                         continue;
1632
1633                 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1634                         kpfctlinput(PRC_IFUP, ifa->ifa_addr);
1635         }
1636         rt_ifmsg(ifp);
1637 #ifdef INET6
1638         in6_if_up(ifp);
1639 #endif
1640
1641         netisr_replymsg(&nmsg->base, 0);
1642 }
1643
1644 void
1645 if_route(struct ifnet *ifp, int flag, int fam)
1646 {
1647         struct netmsg_ifroute msg;
1648
1649         netmsg_init(&msg.base, NULL, &curthread->td_msgport, 0,
1650             if_route_dispatch);
1651         msg.ifp = ifp;
1652         msg.flag = flag;
1653         msg.fam = fam;
1654         netisr_domsg(&msg.base, 0);
1655 }
1656
1657 /*
1658  * Mark an interface down and notify protocols of the transition.  An
1659  * interface going down is also considered to be a synchronizing event.
1660  * We must ensure that all packet processing related to the interface
1661  * has completed before we return so e.g. the caller can free the ifnet
1662  * structure that the mbufs may be referencing.
1663  *
1664  * NOTE: must be called at splnet or eqivalent.
1665  */
1666 void
1667 if_down(struct ifnet *ifp)
1668 {
1669         if_unroute(ifp, IFF_UP, AF_UNSPEC);
1670         netmsg_service_sync();
1671 }
1672
1673 /*
1674  * Mark an interface up and notify protocols of
1675  * the transition.
1676  * NOTE: must be called at splnet or eqivalent.
1677  */
1678 void
1679 if_up(struct ifnet *ifp)
1680 {
1681         if_route(ifp, IFF_UP, AF_UNSPEC);
1682 }
1683
1684 /*
1685  * Process a link state change.
1686  * NOTE: must be called at splsoftnet or equivalent.
1687  */
1688 void
1689 if_link_state_change(struct ifnet *ifp)
1690 {
1691         int link_state = ifp->if_link_state;
1692
1693         rt_ifmsg(ifp);
1694         devctl_notify("IFNET", ifp->if_xname,
1695             (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL);
1696 }
1697
1698 /*
1699  * Handle interface watchdog timer routines.  Called
1700  * from softclock, we decrement timers (if set) and
1701  * call the appropriate interface routine on expiration.
1702  */
1703 static void
1704 if_slowtimo_dispatch(netmsg_t nmsg)
1705 {
1706         struct globaldata *gd = mycpu;
1707         const struct ifnet_array *arr;
1708         int i;
1709
1710         ASSERT_NETISR0;
1711
1712         crit_enter_gd(gd);
1713         lwkt_replymsg(&nmsg->lmsg, 0);  /* reply ASAP */
1714         crit_exit_gd(gd);
1715
1716         arr = ifnet_array_get();
1717         for (i = 0; i < arr->ifnet_count; ++i) {
1718                 struct ifnet *ifp = arr->ifnet_arr[i];
1719
1720                 crit_enter_gd(gd);
1721
1722                 if (if_stats_compat) {
1723                         IFNET_STAT_GET(ifp, ipackets, ifp->if_ipackets);
1724                         IFNET_STAT_GET(ifp, ierrors, ifp->if_ierrors);
1725                         IFNET_STAT_GET(ifp, opackets, ifp->if_opackets);
1726                         IFNET_STAT_GET(ifp, oerrors, ifp->if_oerrors);
1727                         IFNET_STAT_GET(ifp, collisions, ifp->if_collisions);
1728                         IFNET_STAT_GET(ifp, ibytes, ifp->if_ibytes);
1729                         IFNET_STAT_GET(ifp, obytes, ifp->if_obytes);
1730                         IFNET_STAT_GET(ifp, imcasts, ifp->if_imcasts);
1731                         IFNET_STAT_GET(ifp, omcasts, ifp->if_omcasts);
1732                         IFNET_STAT_GET(ifp, iqdrops, ifp->if_iqdrops);
1733                         IFNET_STAT_GET(ifp, noproto, ifp->if_noproto);
1734                         IFNET_STAT_GET(ifp, oqdrops, ifp->if_oqdrops);
1735                 }
1736
1737                 if (ifp->if_timer == 0 || --ifp->if_timer) {
1738                         crit_exit_gd(gd);
1739                         continue;
1740                 }
1741                 if (ifp->if_watchdog) {
1742                         if (ifnet_tryserialize_all(ifp)) {
1743                                 (*ifp->if_watchdog)(ifp);
1744                                 ifnet_deserialize_all(ifp);
1745                         } else {
1746                                 /* try again next timeout */
1747                                 ++ifp->if_timer;
1748                         }
1749                 }
1750
1751                 crit_exit_gd(gd);
1752         }
1753
1754         callout_reset(&if_slowtimo_timer, hz / IFNET_SLOWHZ, if_slowtimo, NULL);
1755 }
1756
1757 static void
1758 if_slowtimo(void *arg __unused)
1759 {
1760         struct lwkt_msg *lmsg = &if_slowtimo_netmsg.lmsg;
1761
1762         KASSERT(mycpuid == 0, ("not on cpu0"));
1763         crit_enter();
1764         if (lmsg->ms_flags & MSGF_DONE)
1765                 lwkt_sendmsg_oncpu(netisr_cpuport(0), lmsg);
1766         crit_exit();
1767 }
1768
1769 /*
1770  * Map interface name to
1771  * interface structure pointer.
1772  */
1773 struct ifnet *
1774 ifunit(const char *name)
1775 {
1776         struct ifnet *ifp;
1777
1778         /*
1779          * Search all the interfaces for this name/number
1780          */
1781         KASSERT(mtx_owned(&ifnet_mtx), ("ifnet is not locked"));
1782
1783         TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
1784                 if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1785                         break;
1786         }
1787         return (ifp);
1788 }
1789
1790 struct ifnet *
1791 ifunit_netisr(const char *name)
1792 {
1793         const struct ifnet_array *arr;
1794         int i;
1795
1796         /*
1797          * Search all the interfaces for this name/number
1798          */
1799
1800         arr = ifnet_array_get();
1801         for (i = 0; i < arr->ifnet_count; ++i) {
1802                 struct ifnet *ifp = arr->ifnet_arr[i];
1803
1804                 if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1805                         return ifp;
1806         }
1807         return NULL;
1808 }
1809
1810 /*
1811  * Interface ioctls.
1812  */
1813 int
1814 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct ucred *cred)
1815 {
1816         struct ifnet *ifp;
1817         struct ifreq *ifr;
1818         struct ifstat *ifs;
1819         int error, do_ifup = 0;
1820         short oif_flags;
1821         int new_flags;
1822         size_t namelen, onamelen;
1823         char new_name[IFNAMSIZ];
1824         struct ifaddr *ifa;
1825         struct sockaddr_dl *sdl;
1826
1827         switch (cmd) {
1828         case SIOCGIFCONF:
1829         case OSIOCGIFCONF:
1830                 return (ifconf(cmd, data, cred));
1831         default:
1832                 break;
1833         }
1834
1835         ifr = (struct ifreq *)data;
1836
1837         switch (cmd) {
1838         case SIOCIFCREATE:
1839         case SIOCIFCREATE2:
1840                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1841                         return (error);
1842                 return (if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name),
1843                         cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL));
1844         case SIOCIFDESTROY:
1845                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1846                         return (error);
1847                 return (if_clone_destroy(ifr->ifr_name));
1848         case SIOCIFGCLONERS:
1849                 return (if_clone_list((struct if_clonereq *)data));
1850         default:
1851                 break;
1852         }
1853
1854         /*
1855          * Nominal ioctl through interface, lookup the ifp and obtain a
1856          * lock to serialize the ifconfig ioctl operation.
1857          */
1858         ifnet_lock();
1859
1860         ifp = ifunit(ifr->ifr_name);
1861         if (ifp == NULL) {
1862                 ifnet_unlock();
1863                 return (ENXIO);
1864         }
1865         error = 0;
1866
1867         switch (cmd) {
1868         case SIOCGIFINDEX:
1869                 ifr->ifr_index = ifp->if_index;
1870                 break;
1871
1872         case SIOCGIFFLAGS:
1873                 ifr->ifr_flags = ifp->if_flags;
1874                 ifr->ifr_flagshigh = ifp->if_flags >> 16;
1875                 break;
1876
1877         case SIOCGIFCAP:
1878                 ifr->ifr_reqcap = ifp->if_capabilities;
1879                 ifr->ifr_curcap = ifp->if_capenable;
1880                 break;
1881
1882         case SIOCGIFMETRIC:
1883                 ifr->ifr_metric = ifp->if_metric;
1884                 break;
1885
1886         case SIOCGIFMTU:
1887                 ifr->ifr_mtu = ifp->if_mtu;
1888                 break;
1889
1890         case SIOCGIFTSOLEN:
1891                 ifr->ifr_tsolen = ifp->if_tsolen;
1892                 break;
1893
1894         case SIOCGIFDATA:
1895                 error = copyout((caddr_t)&ifp->if_data, ifr->ifr_data,
1896                                 sizeof(ifp->if_data));
1897                 break;
1898
1899         case SIOCGIFPHYS:
1900                 ifr->ifr_phys = ifp->if_physical;
1901                 break;
1902
1903         case SIOCGIFPOLLCPU:
1904                 ifr->ifr_pollcpu = -1;
1905                 break;
1906
1907         case SIOCSIFPOLLCPU:
1908                 break;
1909
1910         case SIOCSIFFLAGS:
1911                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1912                 if (error)
1913                         break;
1914                 new_flags = (ifr->ifr_flags & 0xffff) |
1915                     (ifr->ifr_flagshigh << 16);
1916                 if (ifp->if_flags & IFF_SMART) {
1917                         /* Smart drivers twiddle their own routes */
1918                 } else if (ifp->if_flags & IFF_UP &&
1919                     (new_flags & IFF_UP) == 0) {
1920                         if_down(ifp);
1921                 } else if (new_flags & IFF_UP &&
1922                     (ifp->if_flags & IFF_UP) == 0) {
1923                         do_ifup = 1;
1924                 }
1925
1926 #ifdef IFPOLL_ENABLE
1927                 if ((new_flags ^ ifp->if_flags) & IFF_NPOLLING) {
1928                         if (new_flags & IFF_NPOLLING)
1929                                 ifpoll_register(ifp);
1930                         else
1931                                 ifpoll_deregister(ifp);
1932                 }
1933 #endif
1934
1935                 ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
1936                         (new_flags &~ IFF_CANTCHANGE);
1937                 if (new_flags & IFF_PPROMISC) {
1938                         /* Permanently promiscuous mode requested */
1939                         ifp->if_flags |= IFF_PROMISC;
1940                 } else if (ifp->if_pcount == 0) {
1941                         ifp->if_flags &= ~IFF_PROMISC;
1942                 }
1943                 if (ifp->if_ioctl) {
1944                         ifnet_serialize_all(ifp);
1945                         ifp->if_ioctl(ifp, cmd, data, cred);
1946                         ifnet_deserialize_all(ifp);
1947                 }
1948                 if (do_ifup)
1949                         if_up(ifp);
1950                 getmicrotime(&ifp->if_lastchange);
1951                 break;
1952
1953         case SIOCSIFCAP:
1954                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1955                 if (error)
1956                         break;
1957                 if (ifr->ifr_reqcap & ~ifp->if_capabilities) {
1958                         error = EINVAL;
1959                         break;
1960                 }
1961                 ifnet_serialize_all(ifp);
1962                 ifp->if_ioctl(ifp, cmd, data, cred);
1963                 ifnet_deserialize_all(ifp);
1964                 break;
1965
1966         case SIOCSIFNAME:
1967                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1968                 if (error)
1969                         break;
1970                 error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
1971                 if (error)
1972                         break;
1973                 if (new_name[0] == '\0') {
1974                         error = EINVAL;
1975                         break;
1976                 }
1977                 if (ifunit(new_name) != NULL) {
1978                         error = EEXIST;
1979                         break;
1980                 }
1981
1982                 EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
1983
1984                 /* Announce the departure of the interface. */
1985                 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1986
1987                 strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
1988                 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
1989                 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1990                 namelen = strlen(new_name);
1991                 onamelen = sdl->sdl_nlen;
1992                 /*
1993                  * Move the address if needed.  This is safe because we
1994                  * allocate space for a name of length IFNAMSIZ when we
1995                  * create this in if_attach().
1996                  */
1997                 if (namelen != onamelen) {
1998                         bcopy(sdl->sdl_data + onamelen,
1999                             sdl->sdl_data + namelen, sdl->sdl_alen);
2000                 }
2001                 bcopy(new_name, sdl->sdl_data, namelen);
2002                 sdl->sdl_nlen = namelen;
2003                 sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
2004                 bzero(sdl->sdl_data, onamelen);
2005                 while (namelen != 0)
2006                         sdl->sdl_data[--namelen] = 0xff;
2007
2008                 EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
2009
2010                 /* Announce the return of the interface. */
2011                 rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
2012                 break;
2013
2014         case SIOCSIFMETRIC:
2015                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2016                 if (error)
2017                         break;
2018                 ifp->if_metric = ifr->ifr_metric;
2019                 getmicrotime(&ifp->if_lastchange);
2020                 break;
2021
2022         case SIOCSIFPHYS:
2023                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2024                 if (error)
2025                         break;
2026                 if (ifp->if_ioctl == NULL) {
2027                         error = EOPNOTSUPP;
2028                         break;
2029                 }
2030                 ifnet_serialize_all(ifp);
2031                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2032                 ifnet_deserialize_all(ifp);
2033                 if (error == 0)
2034                         getmicrotime(&ifp->if_lastchange);
2035                 break;
2036
2037         case SIOCSIFMTU:
2038         {
2039                 u_long oldmtu = ifp->if_mtu;
2040
2041                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2042                 if (error)
2043                         break;
2044                 if (ifp->if_ioctl == NULL) {
2045                         error = EOPNOTSUPP;
2046                         break;
2047                 }
2048                 if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU) {
2049                         error = EINVAL;
2050                         break;
2051                 }
2052                 ifnet_serialize_all(ifp);
2053                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2054                 ifnet_deserialize_all(ifp);
2055                 if (error == 0) {
2056                         getmicrotime(&ifp->if_lastchange);
2057                         rt_ifmsg(ifp);
2058                 }
2059                 /*
2060                  * If the link MTU changed, do network layer specific procedure.
2061                  */
2062                 if (ifp->if_mtu != oldmtu) {
2063 #ifdef INET6
2064                         nd6_setmtu(ifp);
2065 #endif
2066                 }
2067                 break;
2068         }
2069
2070         case SIOCSIFTSOLEN:
2071                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2072                 if (error)
2073                         break;
2074
2075                 /* XXX need driver supplied upper limit */
2076                 if (ifr->ifr_tsolen <= 0) {
2077                         error = EINVAL;
2078                         break;
2079                 }
2080                 ifp->if_tsolen = ifr->ifr_tsolen;
2081                 break;
2082
2083         case SIOCADDMULTI:
2084         case SIOCDELMULTI:
2085                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2086                 if (error)
2087                         break;
2088
2089                 /* Don't allow group membership on non-multicast interfaces. */
2090                 if ((ifp->if_flags & IFF_MULTICAST) == 0) {
2091                         error = EOPNOTSUPP;
2092                         break;
2093                 }
2094
2095                 /* Don't let users screw up protocols' entries. */
2096                 if (ifr->ifr_addr.sa_family != AF_LINK) {
2097                         error = EINVAL;
2098                         break;
2099                 }
2100
2101                 if (cmd == SIOCADDMULTI) {
2102                         struct ifmultiaddr *ifma;
2103                         error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
2104                 } else {
2105                         error = if_delmulti(ifp, &ifr->ifr_addr);
2106                 }
2107                 if (error == 0)
2108                         getmicrotime(&ifp->if_lastchange);
2109                 break;
2110
2111         case SIOCSIFPHYADDR:
2112         case SIOCDIFPHYADDR:
2113 #ifdef INET6
2114         case SIOCSIFPHYADDR_IN6:
2115 #endif
2116         case SIOCSLIFPHYADDR:
2117         case SIOCSIFMEDIA:
2118         case SIOCSIFGENERIC:
2119                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2120                 if (error)
2121                         break;
2122                 if (ifp->if_ioctl == 0) {
2123                         error = EOPNOTSUPP;
2124                         break;
2125                 }
2126                 ifnet_serialize_all(ifp);
2127                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2128                 ifnet_deserialize_all(ifp);
2129                 if (error == 0)
2130                         getmicrotime(&ifp->if_lastchange);
2131                 break;
2132
2133         case SIOCGIFSTATUS:
2134                 ifs = (struct ifstat *)data;
2135                 ifs->ascii[0] = '\0';
2136                 /* fall through */
2137         case SIOCGIFPSRCADDR:
2138         case SIOCGIFPDSTADDR:
2139         case SIOCGLIFPHYADDR:
2140         case SIOCGIFMEDIA:
2141         case SIOCGIFGENERIC:
2142                 if (ifp->if_ioctl == NULL) {
2143                         error = EOPNOTSUPP;
2144                         break;
2145                 }
2146                 ifnet_serialize_all(ifp);
2147                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2148                 ifnet_deserialize_all(ifp);
2149                 break;
2150
2151         case SIOCSIFLLADDR:
2152                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2153                 if (error)
2154                         break;
2155                 error = if_setlladdr(ifp, ifr->ifr_addr.sa_data,
2156                                      ifr->ifr_addr.sa_len);
2157                 EVENTHANDLER_INVOKE(iflladdr_event, ifp);
2158                 break;
2159
2160         default:
2161                 oif_flags = ifp->if_flags;
2162                 if (so->so_proto == 0) {
2163                         error = EOPNOTSUPP;
2164                         break;
2165                 }
2166                 error = so_pru_control_direct(so, cmd, data, ifp);
2167
2168                 if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
2169 #ifdef INET6
2170                         DELAY(100);/* XXX: temporary workaround for fxp issue*/
2171                         if (ifp->if_flags & IFF_UP) {
2172                                 crit_enter();
2173                                 in6_if_up(ifp);
2174                                 crit_exit();
2175                         }
2176 #endif
2177                 }
2178                 break;
2179         }
2180
2181         ifnet_unlock();
2182         return (error);
2183 }
2184
2185 /*
2186  * Set/clear promiscuous mode on interface ifp based on the truth value
2187  * of pswitch.  The calls are reference counted so that only the first
2188  * "on" request actually has an effect, as does the final "off" request.
2189  * Results are undefined if the "off" and "on" requests are not matched.
2190  */
2191 int
2192 ifpromisc(struct ifnet *ifp, int pswitch)
2193 {
2194         struct ifreq ifr;
2195         int error;
2196         int oldflags;
2197
2198         oldflags = ifp->if_flags;
2199         if (ifp->if_flags & IFF_PPROMISC) {
2200                 /* Do nothing if device is in permanently promiscuous mode */
2201                 ifp->if_pcount += pswitch ? 1 : -1;
2202                 return (0);
2203         }
2204         if (pswitch) {
2205                 /*
2206                  * If the device is not configured up, we cannot put it in
2207                  * promiscuous mode.
2208                  */
2209                 if ((ifp->if_flags & IFF_UP) == 0)
2210                         return (ENETDOWN);
2211                 if (ifp->if_pcount++ != 0)
2212                         return (0);
2213                 ifp->if_flags |= IFF_PROMISC;
2214                 log(LOG_INFO, "%s: promiscuous mode enabled\n",
2215                     ifp->if_xname);
2216         } else {
2217                 if (--ifp->if_pcount > 0)
2218                         return (0);
2219                 ifp->if_flags &= ~IFF_PROMISC;
2220                 log(LOG_INFO, "%s: promiscuous mode disabled\n",
2221                     ifp->if_xname);
2222         }
2223         ifr.ifr_flags = ifp->if_flags;
2224         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2225         ifnet_serialize_all(ifp);
2226         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr, NULL);
2227         ifnet_deserialize_all(ifp);
2228         if (error == 0)
2229                 rt_ifmsg(ifp);
2230         else
2231                 ifp->if_flags = oldflags;
2232         return error;
2233 }
2234
2235 /*
2236  * Return interface configuration
2237  * of system.  List may be used
2238  * in later ioctl's (above) to get
2239  * other information.
2240  */
2241 static int
2242 ifconf(u_long cmd, caddr_t data, struct ucred *cred)
2243 {
2244         struct ifconf *ifc = (struct ifconf *)data;
2245         struct ifnet *ifp;
2246         struct sockaddr *sa;
2247         struct ifreq ifr, *ifrp;
2248         int space = ifc->ifc_len, error = 0;
2249
2250         ifrp = ifc->ifc_req;
2251
2252         ifnet_lock();
2253         TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
2254                 struct ifaddr_container *ifac, *ifac_mark;
2255                 struct ifaddr_marker mark;
2256                 struct ifaddrhead *head;
2257                 int addrs;
2258
2259                 if (space <= sizeof ifr)
2260                         break;
2261
2262                 /*
2263                  * Zero the stack declared structure first to prevent
2264                  * memory disclosure.
2265                  */
2266                 bzero(&ifr, sizeof(ifr));
2267                 if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
2268                     >= sizeof(ifr.ifr_name)) {
2269                         error = ENAMETOOLONG;
2270                         break;
2271                 }
2272
2273                 /*
2274                  * Add a marker, since copyout() could block and during that
2275                  * period the list could be changed.  Inserting the marker to
2276                  * the header of the list will not cause trouble for the code
2277                  * assuming that the first element of the list is AF_LINK; the
2278                  * marker will be moved to the next position w/o blocking.
2279                  */
2280                 ifa_marker_init(&mark, ifp);
2281                 ifac_mark = &mark.ifac;
2282                 head = &ifp->if_addrheads[mycpuid];
2283
2284                 addrs = 0;
2285                 TAILQ_INSERT_HEAD(head, ifac_mark, ifa_link);
2286                 while ((ifac = TAILQ_NEXT(ifac_mark, ifa_link)) != NULL) {
2287                         struct ifaddr *ifa = ifac->ifa;
2288
2289                         TAILQ_REMOVE(head, ifac_mark, ifa_link);
2290                         TAILQ_INSERT_AFTER(head, ifac, ifac_mark, ifa_link);
2291
2292                         /* Ignore marker */
2293                         if (ifa->ifa_addr->sa_family == AF_UNSPEC)
2294                                 continue;
2295
2296                         if (space <= sizeof ifr)
2297                                 break;
2298                         sa = ifa->ifa_addr;
2299                         if (cred->cr_prison &&
2300                             prison_if(cred, sa))
2301                                 continue;
2302                         addrs++;
2303                         /*
2304                          * Keep a reference on this ifaddr, so that it will
2305                          * not be destroyed when its address is copied to
2306                          * the userland, which could block.
2307                          */
2308                         IFAREF(ifa);
2309                         if (sa->sa_len <= sizeof(*sa)) {
2310                                 ifr.ifr_addr = *sa;
2311                                 error = copyout(&ifr, ifrp, sizeof ifr);
2312                                 ifrp++;
2313                         } else {
2314                                 if (space < (sizeof ifr) + sa->sa_len -
2315                                             sizeof(*sa)) {
2316                                         IFAFREE(ifa);
2317                                         break;
2318                                 }
2319                                 space -= sa->sa_len - sizeof(*sa);
2320                                 error = copyout(&ifr, ifrp,
2321                                                 sizeof ifr.ifr_name);
2322                                 if (error == 0)
2323                                         error = copyout(sa, &ifrp->ifr_addr,
2324                                                         sa->sa_len);
2325                                 ifrp = (struct ifreq *)
2326                                         (sa->sa_len + (caddr_t)&ifrp->ifr_addr);
2327                         }
2328                         IFAFREE(ifa);
2329                         if (error)
2330                                 break;
2331                         space -= sizeof ifr;
2332                 }
2333                 TAILQ_REMOVE(head, ifac_mark, ifa_link);
2334                 if (error)
2335                         break;
2336                 if (!addrs) {
2337                         bzero(&ifr.ifr_addr, sizeof ifr.ifr_addr);
2338                         error = copyout(&ifr, ifrp, sizeof ifr);
2339                         if (error)
2340                                 break;
2341                         space -= sizeof ifr;
2342                         ifrp++;
2343                 }
2344         }
2345         ifnet_unlock();
2346
2347         ifc->ifc_len -= space;
2348         return (error);
2349 }
2350
2351 /*
2352  * Just like if_promisc(), but for all-multicast-reception mode.
2353  */
2354 int
2355 if_allmulti(struct ifnet *ifp, int onswitch)
2356 {
2357         int error = 0;
2358         struct ifreq ifr;
2359
2360         crit_enter();
2361
2362         if (onswitch) {
2363                 if (ifp->if_amcount++ == 0) {
2364                         ifp->if_flags |= IFF_ALLMULTI;
2365                         ifr.ifr_flags = ifp->if_flags;
2366                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2367                         ifnet_serialize_all(ifp);
2368                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2369                                               NULL);
2370                         ifnet_deserialize_all(ifp);
2371                 }
2372         } else {
2373                 if (ifp->if_amcount > 1) {
2374                         ifp->if_amcount--;
2375                 } else {
2376                         ifp->if_amcount = 0;
2377                         ifp->if_flags &= ~IFF_ALLMULTI;
2378                         ifr.ifr_flags = ifp->if_flags;
2379                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2380                         ifnet_serialize_all(ifp);
2381                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2382                                               NULL);
2383                         ifnet_deserialize_all(ifp);
2384                 }
2385         }
2386
2387         crit_exit();
2388
2389         if (error == 0)
2390                 rt_ifmsg(ifp);
2391         return error;
2392 }
2393
2394 /*
2395  * Add a multicast listenership to the interface in question.
2396  * The link layer provides a routine which converts
2397  */
2398 int
2399 if_addmulti_serialized(struct ifnet *ifp, struct sockaddr *sa,
2400     struct ifmultiaddr **retifma)
2401 {
2402         struct sockaddr *llsa, *dupsa;
2403         int error;
2404         struct ifmultiaddr *ifma;
2405
2406         ASSERT_IFNET_SERIALIZED_ALL(ifp);
2407
2408         /*
2409          * If the matching multicast address already exists
2410          * then don't add a new one, just add a reference
2411          */
2412         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2413                 if (sa_equal(sa, ifma->ifma_addr)) {
2414                         ifma->ifma_refcount++;
2415                         if (retifma)
2416                                 *retifma = ifma;
2417                         return 0;
2418                 }
2419         }
2420
2421         /*
2422          * Give the link layer a chance to accept/reject it, and also
2423          * find out which AF_LINK address this maps to, if it isn't one
2424          * already.
2425          */
2426         if (ifp->if_resolvemulti) {
2427                 error = ifp->if_resolvemulti(ifp, &llsa, sa);
2428                 if (error)
2429                         return error;
2430         } else {
2431                 llsa = NULL;
2432         }
2433
2434         ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_INTWAIT);
2435         dupsa = kmalloc(sa->sa_len, M_IFMADDR, M_INTWAIT);
2436         bcopy(sa, dupsa, sa->sa_len);
2437
2438         ifma->ifma_addr = dupsa;
2439         ifma->ifma_lladdr = llsa;
2440         ifma->ifma_ifp = ifp;
2441         ifma->ifma_refcount = 1;
2442         ifma->ifma_protospec = NULL;
2443         rt_newmaddrmsg(RTM_NEWMADDR, ifma);
2444
2445         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2446         if (retifma)
2447                 *retifma = ifma;
2448
2449         if (llsa != NULL) {
2450                 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2451                         if (sa_equal(ifma->ifma_addr, llsa))
2452                                 break;
2453                 }
2454                 if (ifma) {
2455                         ifma->ifma_refcount++;
2456                 } else {
2457                         ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_INTWAIT);
2458                         dupsa = kmalloc(llsa->sa_len, M_IFMADDR, M_INTWAIT);
2459                         bcopy(llsa, dupsa, llsa->sa_len);
2460                         ifma->ifma_addr = dupsa;
2461                         ifma->ifma_ifp = ifp;
2462                         ifma->ifma_refcount = 1;
2463                         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2464                 }
2465         }
2466         /*
2467          * We are certain we have added something, so call down to the
2468          * interface to let them know about it.
2469          */
2470         if (ifp->if_ioctl)
2471                 ifp->if_ioctl(ifp, SIOCADDMULTI, 0, NULL);
2472
2473         return 0;
2474 }
2475
2476 int
2477 if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
2478     struct ifmultiaddr **retifma)
2479 {
2480         int error;
2481
2482         ifnet_serialize_all(ifp);
2483         error = if_addmulti_serialized(ifp, sa, retifma);
2484         ifnet_deserialize_all(ifp);
2485
2486         return error;
2487 }
2488
2489 /*
2490  * Remove a reference to a multicast address on this interface.  Yell
2491  * if the request does not match an existing membership.
2492  */
2493 static int
2494 if_delmulti_serialized(struct ifnet *ifp, struct sockaddr *sa)
2495 {
2496         struct ifmultiaddr *ifma;
2497
2498         ASSERT_IFNET_SERIALIZED_ALL(ifp);
2499
2500         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2501                 if (sa_equal(sa, ifma->ifma_addr))
2502                         break;
2503         if (ifma == NULL)
2504                 return ENOENT;
2505
2506         if (ifma->ifma_refcount > 1) {
2507                 ifma->ifma_refcount--;
2508                 return 0;
2509         }
2510
2511         rt_newmaddrmsg(RTM_DELMADDR, ifma);
2512         sa = ifma->ifma_lladdr;
2513         TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2514         /*
2515          * Make sure the interface driver is notified
2516          * in the case of a link layer mcast group being left.
2517          */
2518         if (ifma->ifma_addr->sa_family == AF_LINK && sa == NULL)
2519                 ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2520         kfree(ifma->ifma_addr, M_IFMADDR);
2521         kfree(ifma, M_IFMADDR);
2522         if (sa == NULL)
2523                 return 0;
2524
2525         /*
2526          * Now look for the link-layer address which corresponds to
2527          * this network address.  It had been squirreled away in
2528          * ifma->ifma_lladdr for this purpose (so we don't have
2529          * to call ifp->if_resolvemulti() again), and we saved that
2530          * value in sa above.  If some nasty deleted the
2531          * link-layer address out from underneath us, we can deal because
2532          * the address we stored was is not the same as the one which was
2533          * in the record for the link-layer address.  (So we don't complain
2534          * in that case.)
2535          */
2536         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2537                 if (sa_equal(sa, ifma->ifma_addr))
2538                         break;
2539         if (ifma == NULL)
2540                 return 0;
2541
2542         if (ifma->ifma_refcount > 1) {
2543                 ifma->ifma_refcount--;
2544                 return 0;
2545         }
2546
2547         TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2548         ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2549         kfree(ifma->ifma_addr, M_IFMADDR);
2550         kfree(sa, M_IFMADDR);
2551         kfree(ifma, M_IFMADDR);
2552
2553         return 0;
2554 }
2555
2556 int
2557 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
2558 {
2559         int error;
2560
2561         ifnet_serialize_all(ifp);
2562         error = if_delmulti_serialized(ifp, sa);
2563         ifnet_deserialize_all(ifp);
2564
2565         return error;
2566 }
2567
2568 /*
2569  * Delete all multicast group membership for an interface.
2570  * Should be used to quickly flush all multicast filters.
2571  */
2572 void
2573 if_delallmulti_serialized(struct ifnet *ifp)
2574 {
2575         struct ifmultiaddr *ifma, mark;
2576         struct sockaddr sa;
2577
2578         ASSERT_IFNET_SERIALIZED_ALL(ifp);
2579
2580         bzero(&sa, sizeof(sa));
2581         sa.sa_family = AF_UNSPEC;
2582         sa.sa_len = sizeof(sa);
2583
2584         bzero(&mark, sizeof(mark));
2585         mark.ifma_addr = &sa;
2586
2587         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, &mark, ifma_link);
2588         while ((ifma = TAILQ_NEXT(&mark, ifma_link)) != NULL) {
2589                 TAILQ_REMOVE(&ifp->if_multiaddrs, &mark, ifma_link);
2590                 TAILQ_INSERT_AFTER(&ifp->if_multiaddrs, ifma, &mark,
2591                     ifma_link);
2592
2593                 if (ifma->ifma_addr->sa_family == AF_UNSPEC)
2594                         continue;
2595
2596                 if_delmulti_serialized(ifp, ifma->ifma_addr);
2597         }
2598         TAILQ_REMOVE(&ifp->if_multiaddrs, &mark, ifma_link);
2599 }
2600
2601
2602 /*
2603  * Set the link layer address on an interface.
2604  *
2605  * At this time we only support certain types of interfaces,
2606  * and we don't allow the length of the address to change.
2607  */
2608 int
2609 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
2610 {
2611         struct sockaddr_dl *sdl;
2612         struct ifreq ifr;
2613
2614         sdl = IF_LLSOCKADDR(ifp);
2615         if (sdl == NULL)
2616                 return (EINVAL);
2617         if (len != sdl->sdl_alen)       /* don't allow length to change */
2618                 return (EINVAL);
2619         switch (ifp->if_type) {
2620         case IFT_ETHER:                 /* these types use struct arpcom */
2621         case IFT_XETHER:
2622         case IFT_L2VLAN:
2623         case IFT_IEEE8023ADLAG:
2624                 bcopy(lladdr, ((struct arpcom *)ifp->if_softc)->ac_enaddr, len);
2625                 bcopy(lladdr, LLADDR(sdl), len);
2626                 break;
2627         default:
2628                 return (ENODEV);
2629         }
2630         /*
2631          * If the interface is already up, we need
2632          * to re-init it in order to reprogram its
2633          * address filter.
2634          */
2635         ifnet_serialize_all(ifp);
2636         if ((ifp->if_flags & IFF_UP) != 0) {
2637 #ifdef INET
2638                 struct ifaddr_container *ifac;
2639 #endif
2640
2641                 ifp->if_flags &= ~IFF_UP;
2642                 ifr.ifr_flags = ifp->if_flags;
2643                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2644                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2645                               NULL);
2646                 ifp->if_flags |= IFF_UP;
2647                 ifr.ifr_flags = ifp->if_flags;
2648                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2649                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2650                                  NULL);
2651 #ifdef INET
2652                 /*
2653                  * Also send gratuitous ARPs to notify other nodes about
2654                  * the address change.
2655                  */
2656                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2657                         struct ifaddr *ifa = ifac->ifa;
2658
2659                         if (ifa->ifa_addr != NULL &&
2660                             ifa->ifa_addr->sa_family == AF_INET)
2661                                 arp_gratuitous(ifp, ifa);
2662                 }
2663 #endif
2664         }
2665         ifnet_deserialize_all(ifp);
2666         return (0);
2667 }
2668
2669 struct ifmultiaddr *
2670 ifmaof_ifpforaddr(struct sockaddr *sa, struct ifnet *ifp)
2671 {
2672         struct ifmultiaddr *ifma;
2673
2674         /* TODO: need ifnet_serialize_main */
2675         ifnet_serialize_all(ifp);
2676         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2677                 if (sa_equal(ifma->ifma_addr, sa))
2678                         break;
2679         ifnet_deserialize_all(ifp);
2680
2681         return ifma;
2682 }
2683
2684 /*
2685  * This function locates the first real ethernet MAC from a network
2686  * card and loads it into node, returning 0 on success or ENOENT if
2687  * no suitable interfaces were found.  It is used by the uuid code to
2688  * generate a unique 6-byte number.
2689  */
2690 int
2691 if_getanyethermac(uint16_t *node, int minlen)
2692 {
2693         struct ifnet *ifp;
2694         struct sockaddr_dl *sdl;
2695
2696         ifnet_lock();
2697         TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
2698                 if (ifp->if_type != IFT_ETHER)
2699                         continue;
2700                 sdl = IF_LLSOCKADDR(ifp);
2701                 if (sdl->sdl_alen < minlen)
2702                         continue;
2703                 bcopy(((struct arpcom *)ifp->if_softc)->ac_enaddr, node,
2704                       minlen);
2705                 ifnet_unlock();
2706                 return(0);
2707         }
2708         ifnet_unlock();
2709         return (ENOENT);
2710 }
2711
2712 /*
2713  * The name argument must be a pointer to storage which will last as
2714  * long as the interface does.  For physical devices, the result of
2715  * device_get_name(dev) is a good choice and for pseudo-devices a
2716  * static string works well.
2717  */
2718 void
2719 if_initname(struct ifnet *ifp, const char *name, int unit)
2720 {
2721         ifp->if_dname = name;
2722         ifp->if_dunit = unit;
2723         if (unit != IF_DUNIT_NONE)
2724                 ksnprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
2725         else
2726                 strlcpy(ifp->if_xname, name, IFNAMSIZ);
2727 }
2728
2729 int
2730 if_printf(struct ifnet *ifp, const char *fmt, ...)
2731 {
2732         __va_list ap;
2733         int retval;
2734
2735         retval = kprintf("%s: ", ifp->if_xname);
2736         __va_start(ap, fmt);
2737         retval += kvprintf(fmt, ap);
2738         __va_end(ap);
2739         return (retval);
2740 }
2741
2742 struct ifnet *
2743 if_alloc(uint8_t type)
2744 {
2745         struct ifnet *ifp;
2746         size_t size;
2747
2748         /*
2749          * XXX temporary hack until arpcom is setup in if_l2com
2750          */
2751         if (type == IFT_ETHER)
2752                 size = sizeof(struct arpcom);
2753         else
2754                 size = sizeof(struct ifnet);
2755
2756         ifp = kmalloc(size, M_IFNET, M_WAITOK|M_ZERO);
2757
2758         ifp->if_type = type;
2759
2760         if (if_com_alloc[type] != NULL) {
2761                 ifp->if_l2com = if_com_alloc[type](type, ifp);
2762                 if (ifp->if_l2com == NULL) {
2763                         kfree(ifp, M_IFNET);
2764                         return (NULL);
2765                 }
2766         }
2767         return (ifp);
2768 }
2769
2770 void
2771 if_free(struct ifnet *ifp)
2772 {
2773         kfree(ifp, M_IFNET);
2774 }
2775
2776 void
2777 ifq_set_classic(struct ifaltq *ifq)
2778 {
2779         ifq_set_methods(ifq, ifq->altq_ifp->if_mapsubq,
2780             ifsq_classic_enqueue, ifsq_classic_dequeue, ifsq_classic_request);
2781 }
2782
2783 void
2784 ifq_set_methods(struct ifaltq *ifq, altq_mapsubq_t mapsubq,
2785     ifsq_enqueue_t enqueue, ifsq_dequeue_t dequeue, ifsq_request_t request)
2786 {
2787         int q;
2788
2789         KASSERT(mapsubq != NULL, ("mapsubq is not specified"));
2790         KASSERT(enqueue != NULL, ("enqueue is not specified"));
2791         KASSERT(dequeue != NULL, ("dequeue is not specified"));
2792         KASSERT(request != NULL, ("request is not specified"));
2793
2794         ifq->altq_mapsubq = mapsubq;
2795         for (q = 0; q < ifq->altq_subq_cnt; ++q) {
2796                 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
2797
2798                 ifsq->ifsq_enqueue = enqueue;
2799                 ifsq->ifsq_dequeue = dequeue;
2800                 ifsq->ifsq_request = request;
2801         }
2802 }
2803
2804 static void
2805 ifsq_norm_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
2806 {
2807
2808         classq_add(&ifsq->ifsq_norm, m);
2809         ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
2810 }
2811
2812 static void
2813 ifsq_prio_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
2814 {
2815
2816         classq_add(&ifsq->ifsq_prio, m);
2817         ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
2818         ALTQ_SQ_PRIO_CNTR_INC(ifsq, m->m_pkthdr.len);
2819 }
2820
2821 static struct mbuf *
2822 ifsq_norm_dequeue(struct ifaltq_subque *ifsq)
2823 {
2824         struct mbuf *m;
2825
2826         m = classq_get(&ifsq->ifsq_norm);
2827         if (m != NULL)
2828                 ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
2829         return (m);
2830 }
2831
2832 static struct mbuf *
2833 ifsq_prio_dequeue(struct ifaltq_subque *ifsq)
2834 {
2835         struct mbuf *m;
2836
2837         m = classq_get(&ifsq->ifsq_prio);
2838         if (m != NULL) {
2839                 ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
2840                 ALTQ_SQ_PRIO_CNTR_DEC(ifsq, m->m_pkthdr.len);
2841         }
2842         return (m);
2843 }
2844
2845 int
2846 ifsq_classic_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m,
2847     struct altq_pktattr *pa __unused)
2848 {
2849
2850         M_ASSERTPKTHDR(m);
2851 again:
2852         if (ifsq->ifsq_len >= ifsq->ifsq_maxlen ||
2853             ifsq->ifsq_bcnt >= ifsq->ifsq_maxbcnt) {
2854                 struct mbuf *m_drop;
2855
2856                 if (m->m_flags & M_PRIO) {
2857                         m_drop = NULL;
2858                         if (ifsq->ifsq_prio_len < (ifsq->ifsq_maxlen >> 1) &&
2859                             ifsq->ifsq_prio_bcnt < (ifsq->ifsq_maxbcnt >> 1)) {
2860                                 /* Try dropping some from normal queue. */
2861                                 m_drop = ifsq_norm_dequeue(ifsq);
2862                         }
2863                         if (m_drop == NULL)
2864                                 m_drop = ifsq_prio_dequeue(ifsq);
2865                 } else {
2866                         m_drop = ifsq_norm_dequeue(ifsq);
2867                 }
2868                 if (m_drop != NULL) {
2869                         IFNET_STAT_INC(ifsq->ifsq_ifp, oqdrops, 1);
2870                         m_freem(m_drop);
2871                         goto again;
2872                 }
2873                 /*
2874                  * No old packets could be dropped!
2875                  * NOTE: Caller increases oqdrops.
2876                  */
2877                 m_freem(m);
2878                 return (ENOBUFS);
2879         } else {
2880                 if (m->m_flags & M_PRIO)
2881                         ifsq_prio_enqueue(ifsq, m);
2882                 else
2883                         ifsq_norm_enqueue(ifsq, m);
2884                 return (0);
2885         }
2886 }
2887
2888 struct mbuf *
2889 ifsq_classic_dequeue(struct ifaltq_subque *ifsq, int op)
2890 {
2891         struct mbuf *m;
2892
2893         switch (op) {
2894         case ALTDQ_POLL:
2895                 m = classq_head(&ifsq->ifsq_prio);
2896                 if (m == NULL)
2897                         m = classq_head(&ifsq->ifsq_norm);
2898                 break;
2899
2900         case ALTDQ_REMOVE:
2901                 m = ifsq_prio_dequeue(ifsq);
2902                 if (m == NULL)
2903                         m = ifsq_norm_dequeue(ifsq);
2904                 break;
2905
2906         default:
2907                 panic("unsupported ALTQ dequeue op: %d", op);
2908         }
2909         return m;
2910 }
2911
2912 int
2913 ifsq_classic_request(struct ifaltq_subque *ifsq, int req, void *arg)
2914 {
2915         switch (req) {
2916         case ALTRQ_PURGE:
2917                 for (;;) {
2918                         struct mbuf *m;
2919
2920                         m = ifsq_classic_dequeue(ifsq, ALTDQ_REMOVE);
2921                         if (m == NULL)
2922                                 break;
2923                         m_freem(m);
2924                 }
2925                 break;
2926
2927         default:
2928                 panic("unsupported ALTQ request: %d", req);
2929         }
2930         return 0;
2931 }
2932
2933 static void
2934 ifsq_ifstart_try(struct ifaltq_subque *ifsq, int force_sched)
2935 {
2936         struct ifnet *ifp = ifsq_get_ifp(ifsq);
2937         int running = 0, need_sched;
2938
2939         /*
2940          * Try to do direct ifnet.if_start on the subqueue first, if there is
2941          * contention on the subqueue hardware serializer, ifnet.if_start on
2942          * the subqueue will be scheduled on the subqueue owner CPU.
2943          */
2944         if (!ifsq_tryserialize_hw(ifsq)) {
2945                 /*
2946                  * Subqueue hardware serializer contention happened,
2947                  * ifnet.if_start on the subqueue is scheduled on
2948                  * the subqueue owner CPU, and we keep going.
2949                  */
2950                 ifsq_ifstart_schedule(ifsq, 1);
2951                 return;
2952         }
2953
2954         if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
2955                 ifp->if_start(ifp, ifsq);
2956                 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
2957                         running = 1;
2958         }
2959         need_sched = ifsq_ifstart_need_schedule(ifsq, running);
2960
2961         ifsq_deserialize_hw(ifsq);
2962
2963         if (need_sched) {
2964                 /*
2965                  * More data need to be transmitted, ifnet.if_start on the
2966                  * subqueue is scheduled on the subqueue owner CPU, and we
2967                  * keep going.
2968                  * NOTE: ifnet.if_start subqueue interlock is not released.
2969                  */
2970                 ifsq_ifstart_schedule(ifsq, force_sched);
2971         }
2972 }
2973
2974 /*
2975  * Subqeue packets staging mechanism:
2976  *
2977  * The packets enqueued into the subqueue are staged to a certain amount
2978  * before the ifnet.if_start on the subqueue is called.  In this way, the
2979  * driver could avoid writing to hardware registers upon every packet,
2980  * instead, hardware registers could be written when certain amount of
2981  * packets are put onto hardware TX ring.  The measurement on several modern
2982  * NICs (emx(4), igb(4), bnx(4), bge(4), jme(4)) shows that the hardware
2983  * registers writing aggregation could save ~20% CPU time when 18bytes UDP
2984  * datagrams are transmitted at 1.48Mpps.  The performance improvement by
2985  * hardware registers writing aggeregation is also mentioned by Luigi Rizzo's
2986  * netmap paper (http://info.iet.unipi.it/~luigi/netmap/).
2987  *
2988  * Subqueue packets staging is performed for two entry points into drivers'
2989  * transmission function:
2990  * - Direct ifnet.if_start calling on the subqueue, i.e. ifsq_ifstart_try()
2991  * - ifnet.if_start scheduling on the subqueue, i.e. ifsq_ifstart_schedule()
2992  *
2993  * Subqueue packets staging will be stopped upon any of the following
2994  * conditions:
2995  * - If the count of packets enqueued on the current CPU is great than or
2996  *   equal to ifsq_stage_cntmax. (XXX this should be per-interface)
2997  * - If the total length of packets enqueued on the current CPU is great
2998  *   than or equal to the hardware's MTU - max_protohdr.  max_protohdr is
2999  *   cut from the hardware's MTU mainly bacause a full TCP segment's size
3000  *   is usually less than hardware's MTU.
3001  * - ifsq_ifstart_schedule() is not pending on the current CPU and
3002  *   ifnet.if_start subqueue interlock (ifaltq_subq.ifsq_started) is not
3003  *   released.
3004  * - The if_start_rollup(), which is registered as low priority netisr
3005  *   rollup function, is called; probably because no more work is pending
3006  *   for netisr.
3007  *
3008  * NOTE:
3009  * Currently subqueue packet staging is only performed in netisr threads.
3010  */
3011 int
3012 ifq_dispatch(struct ifnet *ifp, struct mbuf *m, struct altq_pktattr *pa)
3013 {
3014         struct ifaltq *ifq = &ifp->if_snd;
3015         struct ifaltq_subque *ifsq;
3016         int error, start = 0, len, mcast = 0, avoid_start = 0;
3017         struct ifsubq_stage_head *head = NULL;
3018         struct ifsubq_stage *stage = NULL;
3019         struct globaldata *gd = mycpu;
3020         struct thread *td = gd->gd_curthread;
3021
3022         crit_enter_quick(td);
3023
3024         ifsq = ifq_map_subq(ifq, gd->gd_cpuid);
3025         ASSERT_ALTQ_SQ_NOT_SERIALIZED_HW(ifsq);
3026
3027         len = m->m_pkthdr.len;
3028         if (m->m_flags & M_MCAST)
3029                 mcast = 1;
3030
3031         if (td->td_type == TD_TYPE_NETISR) {
3032                 head = &ifsubq_stage_heads[mycpuid];
3033                 stage = ifsq_get_stage(ifsq, mycpuid);
3034
3035                 stage->stg_cnt++;
3036                 stage->stg_len += len;
3037                 if (stage->stg_cnt < ifsq_stage_cntmax &&
3038                     stage->stg_len < (ifp->if_mtu - max_protohdr))
3039                         avoid_start = 1;
3040         }
3041
3042         ALTQ_SQ_LOCK(ifsq);
3043         error = ifsq_enqueue_locked(ifsq, m, pa);
3044         if (error) {
3045                 IFNET_STAT_INC(ifp, oqdrops, 1);
3046                 if (!ifsq_data_ready(ifsq)) {
3047                         ALTQ_SQ_UNLOCK(ifsq);
3048                         crit_exit_quick(td);
3049                         return error;
3050                 }
3051                 avoid_start = 0;
3052         }
3053         if (!ifsq_is_started(ifsq)) {
3054                 if (avoid_start) {
3055                         ALTQ_SQ_UNLOCK(ifsq);
3056
3057                         KKASSERT(!error);
3058                         if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
3059                                 ifsq_stage_insert(head, stage);
3060
3061                         IFNET_STAT_INC(ifp, obytes, len);
3062                         if (mcast)
3063                                 IFNET_STAT_INC(ifp, omcasts, 1);
3064                         crit_exit_quick(td);
3065                         return error;
3066                 }
3067
3068                 /*
3069                  * Hold the subqueue interlock of ifnet.if_start
3070                  */
3071                 ifsq_set_started(ifsq);
3072                 start = 1;
3073         }
3074         ALTQ_SQ_UNLOCK(ifsq);
3075
3076         if (!error) {
3077                 IFNET_STAT_INC(ifp, obytes, len);
3078                 if (mcast)
3079                         IFNET_STAT_INC(ifp, omcasts, 1);
3080         }
3081
3082         if (stage != NULL) {
3083                 if (!start && (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)) {
3084                         KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
3085                         if (!avoid_start) {
3086                                 ifsq_stage_remove(head, stage);
3087                                 ifsq_ifstart_schedule(ifsq, 1);
3088                         }
3089                         crit_exit_quick(td);
3090                         return error;
3091                 }
3092
3093                 if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED) {
3094                         ifsq_stage_remove(head, stage);
3095                 } else {
3096                         stage->stg_cnt = 0;
3097                         stage->stg_len = 0;
3098                 }
3099         }
3100
3101         if (!start) {
3102                 crit_exit_quick(td);
3103                 return error;
3104         }
3105
3106         ifsq_ifstart_try(ifsq, 0);
3107
3108         crit_exit_quick(td);
3109         return error;
3110 }
3111
3112 void *
3113 ifa_create(int size)
3114 {
3115         struct ifaddr *ifa;
3116         int i;
3117
3118         KASSERT(size >= sizeof(*ifa), ("ifaddr size too small"));
3119
3120         ifa = kmalloc(size, M_IFADDR, M_INTWAIT | M_ZERO);
3121
3122         /*
3123          * Make ifa_container availabel on all CPUs, since they
3124          * could be accessed by any threads.
3125          */
3126         ifa->ifa_containers =
3127             kmalloc_cachealign(ncpus * sizeof(struct ifaddr_container),
3128                 M_IFADDR, M_INTWAIT | M_ZERO);
3129
3130         ifa->ifa_ncnt = ncpus;
3131         for (i = 0; i < ncpus; ++i) {
3132                 struct ifaddr_container *ifac = &ifa->ifa_containers[i];
3133
3134                 ifac->ifa_magic = IFA_CONTAINER_MAGIC;
3135                 ifac->ifa = ifa;
3136                 ifac->ifa_refcnt = 1;
3137         }
3138 #ifdef IFADDR_DEBUG
3139         kprintf("alloc ifa %p %d\n", ifa, size);
3140 #endif
3141         return ifa;
3142 }
3143
3144 void
3145 ifac_free(struct ifaddr_container *ifac, int cpu_id)
3146 {
3147         struct ifaddr *ifa = ifac->ifa;
3148
3149         KKASSERT(ifac->ifa_magic == IFA_CONTAINER_MAGIC);
3150         KKASSERT(ifac->ifa_refcnt == 0);
3151         KASSERT(ifac->ifa_listmask == 0,
3152                 ("ifa is still on %#x lists", ifac->ifa_listmask));
3153
3154         ifac->ifa_magic = IFA_CONTAINER_DEAD;
3155
3156 #ifdef IFADDR_DEBUG_VERBOSE
3157         kprintf("try free ifa %p cpu_id %d\n", ifac->ifa, cpu_id);
3158 #endif
3159
3160         KASSERT(ifa->ifa_ncnt > 0 && ifa->ifa_ncnt <= ncpus,
3161                 ("invalid # of ifac, %d", ifa->ifa_ncnt));
3162         if (atomic_fetchadd_int(&ifa->ifa_ncnt, -1) == 1) {
3163 #ifdef IFADDR_DEBUG
3164                 kprintf("free ifa %p\n", ifa);
3165 #endif
3166                 kfree(ifa->ifa_containers, M_IFADDR);
3167                 kfree(ifa, M_IFADDR);
3168         }
3169 }
3170
3171 static void
3172 ifa_iflink_dispatch(netmsg_t nmsg)
3173 {
3174         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3175         struct ifaddr *ifa = msg->ifa;
3176         struct ifnet *ifp = msg->ifp;
3177         int cpu = mycpuid;
3178         struct ifaddr_container *ifac;
3179
3180         crit_enter();
3181
3182         ifac = &ifa->ifa_containers[cpu];
3183         ASSERT_IFAC_VALID(ifac);
3184         KASSERT((ifac->ifa_listmask & IFA_LIST_IFADDRHEAD) == 0,
3185                 ("ifaddr is on if_addrheads"));
3186
3187         ifac->ifa_listmask |= IFA_LIST_IFADDRHEAD;
3188         if (msg->tail)
3189                 TAILQ_INSERT_TAIL(&ifp->if_addrheads[cpu], ifac, ifa_link);
3190         else
3191                 TAILQ_INSERT_HEAD(&ifp->if_addrheads[cpu], ifac, ifa_link);
3192
3193         crit_exit();
3194
3195         netisr_forwardmsg_all(&nmsg->base, cpu + 1);
3196 }
3197
3198 void
3199 ifa_iflink(struct ifaddr *ifa, struct ifnet *ifp, int tail)
3200 {
3201         struct netmsg_ifaddr msg;
3202
3203         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3204                     0, ifa_iflink_dispatch);
3205         msg.ifa = ifa;
3206         msg.ifp = ifp;
3207         msg.tail = tail;
3208
3209         netisr_domsg(&msg.base, 0);
3210 }
3211
3212 static void
3213 ifa_ifunlink_dispatch(netmsg_t nmsg)
3214 {
3215         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3216         struct ifaddr *ifa = msg->ifa;
3217         struct ifnet *ifp = msg->ifp;
3218         int cpu = mycpuid;
3219         struct ifaddr_container *ifac;
3220
3221         crit_enter();
3222
3223         ifac = &ifa->ifa_containers[cpu];
3224         ASSERT_IFAC_VALID(ifac);
3225         KASSERT(ifac->ifa_listmask & IFA_LIST_IFADDRHEAD,
3226                 ("ifaddr is not on if_addrhead"));
3227
3228         TAILQ_REMOVE(&ifp->if_addrheads[cpu], ifac, ifa_link);
3229         ifac->ifa_listmask &= ~IFA_LIST_IFADDRHEAD;
3230
3231         crit_exit();
3232
3233         netisr_forwardmsg_all(&nmsg->base, cpu + 1);
3234 }
3235
3236 void
3237 ifa_ifunlink(struct ifaddr *ifa, struct ifnet *ifp)
3238 {
3239         struct netmsg_ifaddr msg;
3240
3241         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3242                     0, ifa_ifunlink_dispatch);
3243         msg.ifa = ifa;
3244         msg.ifp = ifp;
3245
3246         netisr_domsg(&msg.base, 0);
3247 }
3248
3249 static void
3250 ifa_destroy_dispatch(netmsg_t nmsg)
3251 {
3252         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3253
3254         IFAFREE(msg->ifa);
3255         netisr_forwardmsg_all(&nmsg->base, mycpuid + 1);
3256 }
3257
3258 void
3259 ifa_destroy(struct ifaddr *ifa)
3260 {
3261         struct netmsg_ifaddr msg;
3262
3263         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3264                     0, ifa_destroy_dispatch);
3265         msg.ifa = ifa;
3266
3267         netisr_domsg(&msg.base, 0);
3268 }
3269
3270 static void
3271 if_start_rollup(void)
3272 {
3273         struct ifsubq_stage_head *head = &ifsubq_stage_heads[mycpuid];
3274         struct ifsubq_stage *stage;
3275
3276         crit_enter();
3277
3278         while ((stage = TAILQ_FIRST(&head->stg_head)) != NULL) {
3279                 struct ifaltq_subque *ifsq = stage->stg_subq;
3280                 int is_sched = 0;
3281
3282                 if (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)
3283                         is_sched = 1;
3284                 ifsq_stage_remove(head, stage);
3285
3286                 if (is_sched) {
3287                         ifsq_ifstart_schedule(ifsq, 1);
3288                 } else {
3289                         int start = 0;
3290
3291                         ALTQ_SQ_LOCK(ifsq);
3292                         if (!ifsq_is_started(ifsq)) {
3293                                 /*
3294                                  * Hold the subqueue interlock of
3295                                  * ifnet.if_start
3296                                  */
3297                                 ifsq_set_started(ifsq);
3298                                 start = 1;
3299                         }
3300                         ALTQ_SQ_UNLOCK(ifsq);
3301
3302                         if (start)
3303                                 ifsq_ifstart_try(ifsq, 1);
3304                 }
3305                 KKASSERT((stage->stg_flags &
3306                     (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
3307         }
3308
3309         crit_exit();
3310 }
3311
3312 static void
3313 ifnetinit(void *dummy __unused)
3314 {
3315         int i;
3316
3317         /* XXX netisr_ncpus */
3318         for (i = 0; i < ncpus; ++i)
3319                 TAILQ_INIT(&ifsubq_stage_heads[i].stg_head);
3320         netisr_register_rollup(if_start_rollup, NETISR_ROLLUP_PRIO_IFSTART);
3321 }
3322
3323 void
3324 if_register_com_alloc(u_char type,
3325     if_com_alloc_t *a, if_com_free_t *f)
3326 {
3327
3328         KASSERT(if_com_alloc[type] == NULL,
3329             ("if_register_com_alloc: %d already registered", type));
3330         KASSERT(if_com_free[type] == NULL,
3331             ("if_register_com_alloc: %d free already registered", type));
3332
3333         if_com_alloc[type] = a;
3334         if_com_free[type] = f;
3335 }
3336
3337 void
3338 if_deregister_com_alloc(u_char type)
3339 {
3340
3341         KASSERT(if_com_alloc[type] != NULL,
3342             ("if_deregister_com_alloc: %d not registered", type));
3343         KASSERT(if_com_free[type] != NULL,
3344             ("if_deregister_com_alloc: %d free not registered", type));
3345         if_com_alloc[type] = NULL;
3346         if_com_free[type] = NULL;
3347 }
3348
3349 void
3350 ifq_set_maxlen(struct ifaltq *ifq, int len)
3351 {
3352         ifq->altq_maxlen = len + (ncpus * ifsq_stage_cntmax);
3353 }
3354
3355 int
3356 ifq_mapsubq_default(struct ifaltq *ifq __unused, int cpuid __unused)
3357 {
3358         return ALTQ_SUBQ_INDEX_DEFAULT;
3359 }
3360
3361 int
3362 ifq_mapsubq_modulo(struct ifaltq *ifq, int cpuid)
3363 {
3364
3365         return (cpuid % ifq->altq_subq_mappriv);
3366 }
3367
3368 static void
3369 ifsq_watchdog(void *arg)
3370 {
3371         struct ifsubq_watchdog *wd = arg;
3372         struct ifnet *ifp;
3373
3374         if (__predict_true(wd->wd_timer == 0 || --wd->wd_timer))
3375                 goto done;
3376
3377         ifp = ifsq_get_ifp(wd->wd_subq);
3378         if (ifnet_tryserialize_all(ifp)) {
3379                 wd->wd_watchdog(wd->wd_subq);
3380                 ifnet_deserialize_all(ifp);
3381         } else {
3382                 /* try again next timeout */
3383                 wd->wd_timer = 1;
3384         }
3385 done:
3386         ifsq_watchdog_reset(wd);
3387 }
3388
3389 static void
3390 ifsq_watchdog_reset(struct ifsubq_watchdog *wd)
3391 {
3392         callout_reset_bycpu(&wd->wd_callout, hz, ifsq_watchdog, wd,
3393             ifsq_get_cpuid(wd->wd_subq));
3394 }
3395
3396 void
3397 ifsq_watchdog_init(struct ifsubq_watchdog *wd, struct ifaltq_subque *ifsq,
3398     ifsq_watchdog_t watchdog)
3399 {
3400         callout_init_mp(&wd->wd_callout);
3401         wd->wd_timer = 0;
3402         wd->wd_subq = ifsq;
3403         wd->wd_watchdog = watchdog;
3404 }
3405
3406 void
3407 ifsq_watchdog_start(struct ifsubq_watchdog *wd)
3408 {
3409         wd->wd_timer = 0;
3410         ifsq_watchdog_reset(wd);
3411 }
3412
3413 void
3414 ifsq_watchdog_stop(struct ifsubq_watchdog *wd)
3415 {
3416         wd->wd_timer = 0;
3417         callout_stop(&wd->wd_callout);
3418 }
3419
3420 void
3421 ifnet_lock(void)
3422 {
3423         KASSERT(curthread->td_type != TD_TYPE_NETISR,
3424             ("try holding ifnet lock in netisr"));
3425         mtx_lock(&ifnet_mtx);
3426 }
3427
3428 void
3429 ifnet_unlock(void)
3430 {
3431         KASSERT(curthread->td_type != TD_TYPE_NETISR,
3432             ("try holding ifnet lock in netisr"));
3433         mtx_unlock(&ifnet_mtx);
3434 }
3435
3436 static struct ifnet_array *
3437 ifnet_array_alloc(int count)
3438 {
3439         struct ifnet_array *arr;
3440
3441         arr = kmalloc(__offsetof(struct ifnet_array, ifnet_arr[count]),
3442             M_IFNET, M_WAITOK);
3443         arr->ifnet_count = count;
3444
3445         return arr;
3446 }
3447
3448 static void
3449 ifnet_array_free(struct ifnet_array *arr)
3450 {
3451         if (arr == &ifnet_array0)
3452                 return;
3453         kfree(arr, M_IFNET);
3454 }
3455
3456 static struct ifnet_array *
3457 ifnet_array_add(struct ifnet *ifp, const struct ifnet_array *old_arr)
3458 {
3459         struct ifnet_array *arr;
3460         int count, i;
3461
3462         KASSERT(old_arr->ifnet_count >= 0,
3463             ("invalid ifnet array count %d", old_arr->ifnet_count));
3464         count = old_arr->ifnet_count + 1;
3465         arr = ifnet_array_alloc(count);
3466
3467         /*
3468          * Save the old ifnet array and append this ifp to the end of
3469          * the new ifnet array.
3470          */
3471         for (i = 0; i < old_arr->ifnet_count; ++i) {
3472                 KASSERT(old_arr->ifnet_arr[i] != ifp,
3473                     ("%s is already in ifnet array", ifp->if_xname));
3474                 arr->ifnet_arr[i] = old_arr->ifnet_arr[i];
3475         }
3476         KASSERT(i == count - 1,
3477             ("add %s, ifnet array index mismatch, should be %d, but got %d",
3478              ifp->if_xname, count - 1, i));
3479         arr->ifnet_arr[i] = ifp;
3480
3481         return arr;
3482 }
3483
3484 static struct ifnet_array *
3485 ifnet_array_del(struct ifnet *ifp, const struct ifnet_array *old_arr)
3486 {
3487         struct ifnet_array *arr;
3488         int count, i, idx, found = 0;
3489
3490         KASSERT(old_arr->ifnet_count > 0,
3491             ("invalid ifnet array count %d", old_arr->ifnet_count));
3492         count = old_arr->ifnet_count - 1;
3493         arr = ifnet_array_alloc(count);
3494
3495         /*
3496          * Save the old ifnet array, but skip this ifp.
3497          */
3498         idx = 0;
3499         for (i = 0; i < old_arr->ifnet_count; ++i) {
3500                 if (old_arr->ifnet_arr[i] == ifp) {
3501                         KASSERT(!found,
3502                             ("dup %s is in ifnet array", ifp->if_xname));
3503                         found = 1;
3504                         continue;
3505                 }
3506                 KASSERT(idx < count,
3507                     ("invalid ifnet array index %d, count %d", idx, count));
3508                 arr->ifnet_arr[idx] = old_arr->ifnet_arr[i];
3509                 ++idx;
3510         }
3511         KASSERT(found, ("%s is not in ifnet array", ifp->if_xname));
3512         KASSERT(idx == count,
3513             ("del %s, ifnet array count mismatch, should be %d, but got %d ",
3514              ifp->if_xname, count, idx));
3515
3516         return arr;
3517 }
3518
3519 const struct ifnet_array *
3520 ifnet_array_get(void)
3521 {
3522         const struct ifnet_array *ret;
3523
3524         KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
3525         ret = ifnet_array;
3526         /* Make sure 'ret' is really used. */
3527         cpu_ccfence();
3528         return (ret);
3529 }
3530
3531 int
3532 ifnet_array_isempty(void)
3533 {
3534         KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
3535         if (ifnet_array->ifnet_count == 0)
3536                 return 1;
3537         else
3538                 return 0;
3539 }
3540
3541 void
3542 ifa_marker_init(struct ifaddr_marker *mark, struct ifnet *ifp)
3543 {
3544         struct ifaddr *ifa;
3545
3546         memset(mark, 0, sizeof(*mark));
3547         ifa = &mark->ifa;
3548
3549         mark->ifac.ifa = ifa;
3550
3551         ifa->ifa_addr = &mark->addr;
3552         ifa->ifa_dstaddr = &mark->dstaddr;
3553         ifa->ifa_netmask = &mark->netmask;
3554         ifa->ifa_ifp = ifp;
3555 }
3556
3557 static int
3558 if_ringcnt_fixup(int ring_cnt, int ring_cntmax)
3559 {
3560
3561         KASSERT(ring_cntmax > 0, ("invalid ring count max %d", ring_cntmax));
3562
3563         if (ring_cnt <= 0 || ring_cnt > ring_cntmax)
3564                 ring_cnt = ring_cntmax;
3565         if (ring_cnt > netisr_ncpus)
3566                 ring_cnt = netisr_ncpus;
3567         return (ring_cnt);
3568 }
3569
3570 static void
3571 if_ringmap_set_grid(device_t dev, struct if_ringmap *rm, int grid)
3572 {
3573         int i, offset;
3574
3575         KASSERT(grid > 0, ("invalid if_ringmap grid %d", grid));
3576         KASSERT(grid >= rm->rm_cnt, ("invalid if_ringmap grid %d, count %d",
3577             grid, rm->rm_cnt));
3578         rm->rm_grid = grid;
3579
3580         offset = (rm->rm_grid * device_get_unit(dev)) % netisr_ncpus;
3581         for (i = 0; i < rm->rm_cnt; ++i) {
3582                 rm->rm_cpumap[i] = offset + i;
3583                 KASSERT(rm->rm_cpumap[i] < netisr_ncpus,
3584                     ("invalid cpumap[%d] = %d, offset %d", i,
3585                      rm->rm_cpumap[i], offset));
3586         }
3587 }
3588
3589 static struct if_ringmap *
3590 if_ringmap_alloc_flags(device_t dev, int ring_cnt, int ring_cntmax,
3591     uint32_t flags)
3592 {
3593         struct if_ringmap *rm;
3594         int i, grid = 0, prev_grid;
3595
3596         ring_cnt = if_ringcnt_fixup(ring_cnt, ring_cntmax);
3597         rm = kmalloc(__offsetof(struct if_ringmap, rm_cpumap[ring_cnt]),
3598             M_DEVBUF, M_WAITOK | M_ZERO);
3599
3600         rm->rm_cnt = ring_cnt;
3601         if (flags & RINGMAP_FLAG_POWEROF2)
3602                 rm->rm_cnt = 1 << (fls(rm->rm_cnt) - 1);
3603
3604         prev_grid = netisr_ncpus;
3605         for (i = 0; i < netisr_ncpus; ++i) {
3606                 if (netisr_ncpus % (i + 1) != 0)
3607                         continue;
3608
3609                 grid = netisr_ncpus / (i + 1);
3610                 if (rm->rm_cnt > grid) {
3611                         grid = prev_grid;
3612                         break;
3613                 }
3614
3615                 if (rm->rm_cnt > netisr_ncpus / (i + 2))
3616                         break;
3617                 prev_grid = grid;
3618         }
3619         if_ringmap_set_grid(dev, rm, grid);
3620
3621         return (rm);
3622 }
3623
3624 struct if_ringmap *
3625 if_ringmap_alloc(device_t dev, int ring_cnt, int ring_cntmax)
3626 {
3627
3628         return (if_ringmap_alloc_flags(dev, ring_cnt, ring_cntmax,
3629             RINGMAP_FLAG_NONE));
3630 }
3631
3632 struct if_ringmap *
3633 if_ringmap_alloc2(device_t dev, int ring_cnt, int ring_cntmax)
3634 {
3635
3636         return (if_ringmap_alloc_flags(dev, ring_cnt, ring_cntmax,
3637             RINGMAP_FLAG_POWEROF2));
3638 }
3639
3640 void
3641 if_ringmap_free(struct if_ringmap *rm)
3642 {
3643
3644         kfree(rm, M_DEVBUF);
3645 }
3646
3647 /*
3648  * Align the two ringmaps.
3649  *
3650  * e.g. 8 netisrs, rm0 contains 4 rings, rm1 contains 2 rings.
3651  *
3652  * Before:
3653  *
3654  * CPU      0  1  2  3   4  5  6  7
3655  * NIC_RX               n0 n1 n2 n3
3656  * NIC_TX        N0 N1
3657  *
3658  * After:
3659  *
3660  * CPU      0  1  2  3   4  5  6  7
3661  * NIC_RX               n0 n1 n2 n3
3662  * NIC_TX               N0 N1
3663  */
3664 void
3665 if_ringmap_align(device_t dev, struct if_ringmap *rm0, struct if_ringmap *rm1)
3666 {
3667
3668         if (rm0->rm_grid > rm1->rm_grid)
3669                 if_ringmap_set_grid(dev, rm1, rm0->rm_grid);
3670         else if (rm0->rm_grid < rm1->rm_grid)
3671                 if_ringmap_set_grid(dev, rm0, rm1->rm_grid);
3672 }
3673
3674 void
3675 if_ringmap_match(device_t dev, struct if_ringmap *rm0, struct if_ringmap *rm1)
3676 {
3677         int subset_grid, cnt, divisor, mod, offset, i;
3678         struct if_ringmap *subset_rm, *rm;
3679         int old_rm0_grid, old_rm1_grid;
3680
3681         if (rm0->rm_grid == rm1->rm_grid)
3682                 return;
3683
3684         /* Save grid for later use */
3685         old_rm0_grid = rm0->rm_grid;
3686         old_rm1_grid = rm1->rm_grid;
3687
3688         if_ringmap_align(dev, rm0, rm1);
3689
3690         /*
3691          * Re-shuffle rings to get more even distribution.
3692          *
3693          * e.g. 12 netisrs, rm0 contains 4 rings, rm1 contains 2 rings.
3694          *
3695          * CPU       0  1  2  3   4  5  6  7   8  9 10 11
3696          *
3697          * NIC_RX   a0 a1 a2 a3  b0 b1 b2 b3  c0 c1 c2 c3
3698          * NIC_TX   A0 A1        B0 B1        C0 C1
3699          *
3700          * NIC_RX   d0 d1 d2 d3  e0 e1 e2 e3  f0 f1 f2 f3
3701          * NIC_TX         D0 D1        E0 E1        F0 F1
3702          */
3703
3704         if (rm0->rm_cnt >= (2 * old_rm1_grid)) {
3705                 cnt = rm0->rm_cnt;
3706                 subset_grid = old_rm1_grid;
3707                 subset_rm = rm1;
3708                 rm = rm0;
3709         } else if (rm1->rm_cnt > (2 * old_rm0_grid)) {
3710                 cnt = rm1->rm_cnt;
3711                 subset_grid = old_rm0_grid;
3712                 subset_rm = rm0;
3713                 rm = rm1;
3714         } else {
3715                 /* No space to shuffle. */
3716                 return;
3717         }
3718
3719         mod = cnt / subset_grid;
3720         KKASSERT(mod >= 2);
3721         divisor = netisr_ncpus / rm->rm_grid;
3722         offset = ((device_get_unit(dev) / divisor) % mod) * subset_grid;
3723
3724         for (i = 0; i < subset_rm->rm_cnt; ++i) {
3725                 subset_rm->rm_cpumap[i] += offset;
3726                 KASSERT(subset_rm->rm_cpumap[i] < netisr_ncpus,
3727                     ("match: invalid cpumap[%d] = %d, offset %d",
3728                      i, subset_rm->rm_cpumap[i], offset));
3729         }
3730 #ifdef INVARIANTS
3731         for (i = 0; i < subset_rm->rm_cnt; ++i) {
3732                 int j;
3733
3734                 for (j = 0; j < rm->rm_cnt; ++j) {
3735                         if (rm->rm_cpumap[j] == subset_rm->rm_cpumap[i])
3736                                 break;
3737                 }
3738                 KASSERT(j < rm->rm_cnt,
3739                     ("subset cpumap[%d] = %d not found in superset",
3740                      i, subset_rm->rm_cpumap[i]));
3741         }
3742 #endif
3743 }
3744
3745 int
3746 if_ringmap_count(const struct if_ringmap *rm)
3747 {
3748
3749         return (rm->rm_cnt);
3750 }
3751
3752 int
3753 if_ringmap_cpumap(const struct if_ringmap *rm, int ring)
3754 {
3755
3756         KASSERT(ring >= 0 && ring < rm->rm_cnt, ("invalid ring %d", ring));
3757         return (rm->rm_cpumap[ring]);
3758 }
3759
3760 void
3761 if_ringmap_rdrtable(const struct if_ringmap *rm, int table[], int table_nent)
3762 {
3763         int i, grid_idx, grid_cnt, patch_off, patch_cnt, ncopy;
3764
3765         KASSERT(table_nent > 0 && (table_nent & NETISR_CPUMASK) == 0,
3766             ("invalid redirect table entries %d", table_nent));
3767
3768         grid_idx = 0;
3769         for (i = 0; i < NETISR_CPUMAX; ++i) {
3770                 table[i] = grid_idx++ % rm->rm_cnt;
3771
3772                 if (grid_idx == rm->rm_grid)
3773                         grid_idx = 0;
3774         }
3775
3776         /*
3777          * Make the ring distributed more evenly for the remainder
3778          * of each grid.
3779          *
3780          * e.g. 12 netisrs, rm contains 8 rings.
3781          *
3782          * Redirect table before:
3783          *
3784          *  0  1  2  3  4  5  6  7  0  1  2  3  0  1  2  3
3785          *  4  5  6  7  0  1  2  3  0  1  2  3  4  5  6  7
3786          *  0  1  2  3  0  1  2  3  4  5  6  7  0  1  2  3
3787          *  ....
3788          *
3789          * Redirect table after being patched (pX, patched entries):
3790          *
3791          *  0  1  2  3  4  5  6  7 p0 p1 p2 p3  0  1  2  3
3792          *  4  5  6  7 p4 p5 p6 p7  0  1  2  3  4  5  6  7
3793          * p0 p1 p2 p3  0  1  2  3  4  5  6  7 p4 p5 p6 p7
3794          *  ....
3795          */
3796         patch_cnt = rm->rm_grid % rm->rm_cnt;
3797         if (patch_cnt == 0)
3798                 goto done;
3799         patch_off = rm->rm_grid - (rm->rm_grid % rm->rm_cnt);
3800
3801         grid_cnt = roundup(NETISR_CPUMAX, rm->rm_grid) / rm->rm_grid;
3802         grid_idx = 0;
3803         for (i = 0; i < grid_cnt; ++i) {
3804                 int j;
3805
3806                 for (j = 0; j < patch_cnt; ++j) {
3807                         int fix_idx;
3808
3809                         fix_idx = (i * rm->rm_grid) + patch_off + j;
3810                         if (fix_idx >= NETISR_CPUMAX)
3811                                 goto done;
3812                         table[fix_idx] = grid_idx++ % rm->rm_cnt;
3813                 }
3814         }
3815 done:
3816         /*
3817          * If the device supports larger redirect table, duplicate
3818          * the first NETISR_CPUMAX entries to the rest of the table,
3819          * so that it matches upper layer's expectation:
3820          * (hash & NETISR_CPUMASK) % netisr_ncpus
3821          */
3822         ncopy = table_nent / NETISR_CPUMAX;
3823         for (i = 1; i < ncopy; ++i) {
3824                 memcpy(&table[i * NETISR_CPUMAX], table,
3825                     NETISR_CPUMAX * sizeof(table[0]));
3826         }
3827         if (if_ringmap_dumprdr) {
3828                 for (i = 0; i < table_nent; ++i) {
3829                         if (i != 0 && i % 16 == 0)
3830                                 kprintf("\n");
3831                         kprintf("%03d ", table[i]);
3832                 }
3833                 kprintf("\n");
3834         }
3835 }
3836
3837 int
3838 if_ringmap_cpumap_sysctl(SYSCTL_HANDLER_ARGS)
3839 {
3840         struct if_ringmap *rm = arg1;
3841         int i, error = 0;
3842
3843         for (i = 0; i < rm->rm_cnt; ++i) {
3844                 int cpu = rm->rm_cpumap[i];
3845
3846                 error = SYSCTL_OUT(req, &cpu, sizeof(cpu));
3847                 if (error)
3848                         break;
3849         }
3850         return (error);
3851 }