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/lock.h>
  51 #include <sys/sockio.h>
  52 #include <sys/syslog.h>
  53 #include <sys/sysctl.h>
  54 #include <sys/domain.h>
  55 #include <sys/thread.h>
  56 #include <sys/serialize.h>
  57 #include <sys/bus.h>
  58 #include <sys/jail.h>
  59
  60 #include <sys/thread2.h>
  61 #include <sys/msgport2.h>
  62 #include <sys/mutex2.h>
  63
  64 #include <net/if.h>
  65 #include <net/if_arp.h>
  66 #include <net/if_dl.h>
  67 #include <net/if_types.h>
  68 #include <net/if_var.h>
  69 #include <net/if_ringmap.h>
  70 #include <net/ifq_var.h>
  71 #include <net/radix.h>
  72 #include <net/route.h>
  73 #include <net/if_clone.h>
  74 #include <net/netisr2.h>
  75 #include <net/netmsg2.h>
  76
  77 #include <machine/atomic.h>
  78 #include <machine/stdarg.h>
  79 #include <machine/smp.h>
  80
  81 #if defined(INET) || defined(INET6)
  82 #include <netinet/in.h>
  83 #include <netinet/in_var.h>
  84 #include <netinet/if_ether.h>
  85 #ifdef INET6
  86 #include <netinet6/in6_var.h>
  87 #include <netinet6/in6_ifattach.h>
  88 #endif /* INET6 */
  89 #endif /* INET || INET6 */
  90
  91 struct netmsg_ifaddr {
  92         struct netmsg_base base;
  93         struct ifaddr   *ifa;
  94         struct ifnet    *ifp;
  95         int             tail;
  96 };
  97
  98 struct ifsubq_stage_head {
  99         TAILQ_HEAD(, ifsubq_stage)      stg_head;
 100 } __cachealign;
 101
 102 struct if_ringmap {
 103         int             rm_cnt;
 104         int             rm_grid;
 105         int             rm_cpumap[];
 106 };
 107
 108 #define RINGMAP_FLAG_NONE               0x0
 109 #define RINGMAP_FLAG_POWEROF2           0x1
 110
 111 /*
 112  * System initialization
 113  */
 114 static void     if_attachdomain(void *);
 115 static void     if_attachdomain1(struct ifnet *);
 116 static int      ifconf(u_long, caddr_t, struct ucred *);
 117 static void     ifinit(void *);
 118 static void     ifnetinit(void *);
 119 static void     if_slowtimo(void *);
 120 static void     link_rtrequest(int, struct rtentry *);
 121 static int      if_rtdel(struct radix_node *, void *);
 122 static void     if_slowtimo_dispatch(netmsg_t);
 123
 124 /* Helper functions */
 125 static void     ifsq_watchdog_reset(struct ifsubq_watchdog *);
 126 static int      if_delmulti_serialized(struct ifnet *, struct sockaddr *);
 127 static struct ifnet_array *ifnet_array_alloc(int);
 128 static void     ifnet_array_free(struct ifnet_array *);
 129 static struct ifnet_array *ifnet_array_add(struct ifnet *,
 130                     const struct ifnet_array *);
 131 static struct ifnet_array *ifnet_array_del(struct ifnet *,
 132                     const struct ifnet_array *);
 133 static struct ifg_group *if_creategroup(const char *);
 134 static int      if_destroygroup(struct ifg_group *);
 135 static int      if_delgroup_locked(struct ifnet *, const char *);
 136 static int      if_getgroups(struct ifgroupreq *, struct ifnet *);
 137 static int      if_getgroupmembers(struct ifgroupreq *);
 138
 139 #ifdef INET6
 140 /*
 141  * XXX: declare here to avoid to include many inet6 related files..
 142  * should be more generalized?
 143  */
 144 extern void     nd6_setmtu(struct ifnet *);
 145 #endif
 146
 147 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers");
 148 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management");
 149 SYSCTL_NODE(_net_link, OID_AUTO, ringmap, CTLFLAG_RW, 0, "link ringmap");
 150
 151 static int ifsq_stage_cntmax = 16;
 152 TUNABLE_INT("net.link.stage_cntmax", &ifsq_stage_cntmax);
 153 SYSCTL_INT(_net_link, OID_AUTO, stage_cntmax, CTLFLAG_RW,
 154     &ifsq_stage_cntmax, 0, "ifq staging packet count max");
 155
 156 static int if_stats_compat = 0;
 157 SYSCTL_INT(_net_link, OID_AUTO, stats_compat, CTLFLAG_RW,
 158     &if_stats_compat, 0, "Compat the old ifnet stats");
 159
 160 static int if_ringmap_dumprdr = 0;
 161 SYSCTL_INT(_net_link_ringmap, OID_AUTO, dump_rdr, CTLFLAG_RW,
 162     &if_ringmap_dumprdr, 0, "dump redirect table");
 163
 164 SYSINIT(interfaces, SI_SUB_PROTO_IF, SI_ORDER_FIRST, ifinit, NULL);
 165 SYSINIT(ifnet, SI_SUB_PRE_DRIVERS, SI_ORDER_ANY, ifnetinit, NULL);
 166
 167 static if_com_alloc_t *if_com_alloc[256];
 168 static if_com_free_t *if_com_free[256];
 169
 170 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
 171 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
 172 MALLOC_DEFINE(M_IFNET, "ifnet", "interface structure");
 173
 174 int                     ifqmaxlen = IFQ_MAXLEN;
 175 struct ifnethead        ifnet = TAILQ_HEAD_INITIALIZER(ifnet);
 176 struct ifgrouphead      ifg_head = TAILQ_HEAD_INITIALIZER(ifg_head);
 177 static struct lock      ifgroup_lock;
 178
 179 static struct ifnet_array       ifnet_array0;
 180 static struct ifnet_array       *ifnet_array = &ifnet_array0;
 181
 182 static struct callout           if_slowtimo_timer;
 183 static struct netmsg_base       if_slowtimo_netmsg;
 184
 185 int                     if_index = 0;
 186 struct ifnet            **ifindex2ifnet = NULL;
 187 static struct mtx       ifnet_mtx = MTX_INITIALIZER("ifnet");
 188
 189 static struct ifsubq_stage_head ifsubq_stage_heads[MAXCPU];
 190
 191 #ifdef notyet
 192 #define IFQ_KTR_STRING          "ifq=%p"
 193 #define IFQ_KTR_ARGS            struct ifaltq *ifq
 194 #ifndef KTR_IFQ
 195 #define KTR_IFQ                 KTR_ALL
 196 #endif
 197 KTR_INFO_MASTER(ifq);
 198 KTR_INFO(KTR_IFQ, ifq, enqueue, 0, IFQ_KTR_STRING, IFQ_KTR_ARGS);
 199 KTR_INFO(KTR_IFQ, ifq, dequeue, 1, IFQ_KTR_STRING, IFQ_KTR_ARGS);
 200 #define logifq(name, arg)       KTR_LOG(ifq_ ## name, arg)
 201
 202 #define IF_START_KTR_STRING     "ifp=%p"
 203 #define IF_START_KTR_ARGS       struct ifnet *ifp
 204 #ifndef KTR_IF_START
 205 #define KTR_IF_START            KTR_ALL
 206 #endif
 207 KTR_INFO_MASTER(if_start);
 208 KTR_INFO(KTR_IF_START, if_start, run, 0,
 209          IF_START_KTR_STRING, IF_START_KTR_ARGS);
 210 KTR_INFO(KTR_IF_START, if_start, sched, 1,
 211          IF_START_KTR_STRING, IF_START_KTR_ARGS);
 212 KTR_INFO(KTR_IF_START, if_start, avoid, 2,
 213          IF_START_KTR_STRING, IF_START_KTR_ARGS);
 214 KTR_INFO(KTR_IF_START, if_start, contend_sched, 3,
 215          IF_START_KTR_STRING, IF_START_KTR_ARGS);
 216 KTR_INFO(KTR_IF_START, if_start, chase_sched, 4,
 217          IF_START_KTR_STRING, IF_START_KTR_ARGS);
 218 #define logifstart(name, arg)   KTR_LOG(if_start_ ## name, arg)
 219 #endif /* notyet */
 220
 221 /*
 222  * Network interface utility routines.
 223  *
 224  * Routines with ifa_ifwith* names take sockaddr *'s as
 225  * parameters.
 226  */
 227 /* ARGSUSED */
 228 static void
 229 ifinit(void *dummy)
 230 {
 231         lockinit(&ifgroup_lock, "ifgroup", 0, 0);
 232
 233         callout_init_mp(&if_slowtimo_timer);
 234         netmsg_init(&if_slowtimo_netmsg, NULL, &netisr_adone_rport,
 235             MSGF_PRIORITY, if_slowtimo_dispatch);
 236
 237         /* Start if_slowtimo */
 238         lwkt_sendmsg(netisr_cpuport(0), &if_slowtimo_netmsg.lmsg);
 239 }
 240
 241 static void
 242 ifsq_ifstart_ipifunc(void *arg)
 243 {
 244         struct ifaltq_subque *ifsq = arg;
 245         struct lwkt_msg *lmsg = ifsq_get_ifstart_lmsg(ifsq, mycpuid);
 246
 247         crit_enter();
 248         if (lmsg->ms_flags & MSGF_DONE)
 249                 lwkt_sendmsg_oncpu(netisr_cpuport(mycpuid), lmsg);
 250         crit_exit();
 251 }
 252
 253 static __inline void
 254 ifsq_stage_remove(struct ifsubq_stage_head *head, struct ifsubq_stage *stage)
 255 {
 256         KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
 257         TAILQ_REMOVE(&head->stg_head, stage, stg_link);
 258         stage->stg_flags &= ~(IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED);
 259         stage->stg_cnt = 0;
 260         stage->stg_len = 0;
 261 }
 262
 263 static __inline void
 264 ifsq_stage_insert(struct ifsubq_stage_head *head, struct ifsubq_stage *stage)
 265 {
 266         KKASSERT((stage->stg_flags &
 267             (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
 268         stage->stg_flags |= IFSQ_STAGE_FLAG_QUED;
 269         TAILQ_INSERT_TAIL(&head->stg_head, stage, stg_link);
 270 }
 271
 272 /*
 273  * Schedule ifnet.if_start on the subqueue owner CPU
 274  */
 275 static void
 276 ifsq_ifstart_schedule(struct ifaltq_subque *ifsq, int force)
 277 {
 278         int cpu;
 279
 280         if (!force && curthread->td_type == TD_TYPE_NETISR &&
 281             ifsq_stage_cntmax > 0) {
 282                 struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid);
 283
 284                 stage->stg_cnt = 0;
 285                 stage->stg_len = 0;
 286                 if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
 287                         ifsq_stage_insert(&ifsubq_stage_heads[mycpuid], stage);
 288                 stage->stg_flags |= IFSQ_STAGE_FLAG_SCHED;
 289                 return;
 290         }
 291
 292         cpu = ifsq_get_cpuid(ifsq);
 293         if (cpu != mycpuid)
 294                 lwkt_send_ipiq(globaldata_find(cpu), ifsq_ifstart_ipifunc, ifsq);
 295         else
 296                 ifsq_ifstart_ipifunc(ifsq);
 297 }
 298
 299 /*
 300  * NOTE:
 301  * This function will release ifnet.if_start subqueue interlock,
 302  * if ifnet.if_start for the subqueue does not need to be scheduled
 303  */
 304 static __inline int
 305 ifsq_ifstart_need_schedule(struct ifaltq_subque *ifsq, int running)
 306 {
 307         if (!running || ifsq_is_empty(ifsq)
 308 #ifdef ALTQ
 309             || ifsq->ifsq_altq->altq_tbr != NULL
 310 #endif
 311         ) {
 312                 ALTQ_SQ_LOCK(ifsq);
 313                 /*
 314                  * ifnet.if_start subqueue interlock is released, if:
 315                  * 1) Hardware can not take any packets, due to
 316                  *    o  interface is marked down
 317                  *    o  hardware queue is full (ifsq_is_oactive)
 318                  *    Under the second situation, hardware interrupt
 319                  *    or polling(4) will call/schedule ifnet.if_start
 320                  *    on the subqueue when hardware queue is ready
 321                  * 2) There is no packet in the subqueue.
 322                  *    Further ifq_dispatch or ifq_handoff will call/
 323                  *    schedule ifnet.if_start on the subqueue.
 324                  * 3) TBR is used and it does not allow further
 325                  *    dequeueing.
 326                  *    TBR callout will call ifnet.if_start on the
 327                  *    subqueue.
 328                  */
 329                 if (!running || !ifsq_data_ready(ifsq)) {
 330                         ifsq_clr_started(ifsq);
 331                         ALTQ_SQ_UNLOCK(ifsq);
 332                         return 0;
 333                 }
 334                 ALTQ_SQ_UNLOCK(ifsq);
 335         }
 336         return 1;
 337 }
 338
 339 static void
 340 ifsq_ifstart_dispatch(netmsg_t msg)
 341 {
 342         struct lwkt_msg *lmsg = &msg->base.lmsg;
 343         struct ifaltq_subque *ifsq = lmsg->u.ms_resultp;
 344         struct ifnet *ifp = ifsq_get_ifp(ifsq);
 345         struct globaldata *gd = mycpu;
 346         int running = 0, need_sched;
 347
 348         crit_enter_gd(gd);
 349
 350         lwkt_replymsg(lmsg, 0); /* reply ASAP */
 351
 352         if (gd->gd_cpuid != ifsq_get_cpuid(ifsq)) {
 353                 /*
 354                  * We need to chase the subqueue owner CPU change.
 355                  */
 356                 ifsq_ifstart_schedule(ifsq, 1);
 357                 crit_exit_gd(gd);
 358                 return;
 359         }
 360
 361         ifsq_serialize_hw(ifsq);
 362         if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
 363                 ifp->if_start(ifp, ifsq);
 364                 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
 365                         running = 1;
 366         }
 367         need_sched = ifsq_ifstart_need_schedule(ifsq, running);
 368         ifsq_deserialize_hw(ifsq);
 369
 370         if (need_sched) {
 371                 /*
 372                  * More data need to be transmitted, ifnet.if_start is
 373                  * scheduled on the subqueue owner CPU, and we keep going.
 374                  * NOTE: ifnet.if_start subqueue interlock is not released.
 375                  */
 376                 ifsq_ifstart_schedule(ifsq, 0);
 377         }
 378
 379         crit_exit_gd(gd);
 380 }
 381
 382 /* Device driver ifnet.if_start helper function */
 383 void
 384 ifsq_devstart(struct ifaltq_subque *ifsq)
 385 {
 386         struct ifnet *ifp = ifsq_get_ifp(ifsq);
 387         int running = 0;
 388
 389         ASSERT_ALTQ_SQ_SERIALIZED_HW(ifsq);
 390
 391         ALTQ_SQ_LOCK(ifsq);
 392         if (ifsq_is_started(ifsq) || !ifsq_data_ready(ifsq)) {
 393                 ALTQ_SQ_UNLOCK(ifsq);
 394                 return;
 395         }
 396         ifsq_set_started(ifsq);
 397         ALTQ_SQ_UNLOCK(ifsq);
 398
 399         ifp->if_start(ifp, ifsq);
 400
 401         if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
 402                 running = 1;
 403
 404         if (ifsq_ifstart_need_schedule(ifsq, running)) {
 405                 /*
 406                  * More data need to be transmitted, ifnet.if_start is
 407                  * scheduled on ifnet's CPU, and we keep going.
 408                  * NOTE: ifnet.if_start interlock is not released.
 409                  */
 410                 ifsq_ifstart_schedule(ifsq, 0);
 411         }
 412 }
 413
 414 void
 415 if_devstart(struct ifnet *ifp)
 416 {
 417         ifsq_devstart(ifq_get_subq_default(&ifp->if_snd));
 418 }
 419
 420 /* Device driver ifnet.if_start schedule helper function */
 421 void
 422 ifsq_devstart_sched(struct ifaltq_subque *ifsq)
 423 {
 424         ifsq_ifstart_schedule(ifsq, 1);
 425 }
 426
 427 void
 428 if_devstart_sched(struct ifnet *ifp)
 429 {
 430         ifsq_devstart_sched(ifq_get_subq_default(&ifp->if_snd));
 431 }
 432
 433 static void
 434 if_default_serialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
 435 {
 436         lwkt_serialize_enter(ifp->if_serializer);
 437 }
 438
 439 static void
 440 if_default_deserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
 441 {
 442         lwkt_serialize_exit(ifp->if_serializer);
 443 }
 444
 445 static int
 446 if_default_tryserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
 447 {
 448         return lwkt_serialize_try(ifp->if_serializer);
 449 }
 450
 451 #ifdef INVARIANTS
 452 static void
 453 if_default_serialize_assert(struct ifnet *ifp,
 454                             enum ifnet_serialize slz __unused,
 455                             boolean_t serialized)
 456 {
 457         if (serialized)
 458                 ASSERT_SERIALIZED(ifp->if_serializer);
 459         else
 460                 ASSERT_NOT_SERIALIZED(ifp->if_serializer);
 461 }
 462 #endif
 463
 464 /*
 465  * Attach an interface to the list of "active" interfaces.
 466  *
 467  * The serializer is optional.
 468  */
 469 void
 470 if_attach(struct ifnet *ifp, lwkt_serialize_t serializer)
 471 {
 472         unsigned socksize;
 473         int namelen, masklen;
 474         struct sockaddr_dl *sdl, *sdl_addr;
 475         struct ifaddr *ifa;
 476         struct ifaltq *ifq;
 477         struct ifnet **old_ifindex2ifnet = NULL;
 478         struct ifnet_array *old_ifnet_array;
 479         int i, q, qlen;
 480         char qlenname[64];
 481
 482         static int if_indexlim = 8;
 483
 484         if (ifp->if_serialize != NULL) {
 485                 KASSERT(ifp->if_deserialize != NULL &&
 486                         ifp->if_tryserialize != NULL &&
 487                         ifp->if_serialize_assert != NULL,
 488                         ("serialize functions are partially setup"));
 489
 490                 /*
 491                  * If the device supplies serialize functions,
 492                  * then clear if_serializer to catch any invalid
 493                  * usage of this field.
 494                  */
 495                 KASSERT(serializer == NULL,
 496                         ("both serialize functions and default serializer "
 497                          "are supplied"));
 498                 ifp->if_serializer = NULL;
 499         } else {
 500                 KASSERT(ifp->if_deserialize == NULL &&
 501                         ifp->if_tryserialize == NULL &&
 502                         ifp->if_serialize_assert == NULL,
 503                         ("serialize functions are partially setup"));
 504                 ifp->if_serialize = if_default_serialize;
 505                 ifp->if_deserialize = if_default_deserialize;
 506                 ifp->if_tryserialize = if_default_tryserialize;
 507 #ifdef INVARIANTS
 508                 ifp->if_serialize_assert = if_default_serialize_assert;
 509 #endif
 510
 511                 /*
 512                  * The serializer can be passed in from the device,
 513                  * allowing the same serializer to be used for both
 514                  * the interrupt interlock and the device queue.
 515                  * If not specified, the netif structure will use an
 516                  * embedded serializer.
 517                  */
 518                 if (serializer == NULL) {
 519                         serializer = &ifp->if_default_serializer;
 520                         lwkt_serialize_init(serializer);
 521                 }
 522                 ifp->if_serializer = serializer;
 523         }
 524
 525         /*
 526          * Make if_addrhead available on all CPUs, since they
 527          * could be accessed by any threads.
 528          */
 529         ifp->if_addrheads = kmalloc(ncpus * sizeof(struct ifaddrhead),
 530                                     M_IFADDR, M_WAITOK | M_ZERO);
 531         for (i = 0; i < ncpus; ++i)
 532                 TAILQ_INIT(&ifp->if_addrheads[i]);
 533
 534         TAILQ_INIT(&ifp->if_multiaddrs);
 535         TAILQ_INIT(&ifp->if_groups);
 536         getmicrotime(&ifp->if_lastchange);
 537         if_addgroup(ifp, IFG_ALL);
 538
 539         /*
 540          * create a Link Level name for this device
 541          */
 542         namelen = strlen(ifp->if_xname);
 543         masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + namelen;
 544         socksize = masklen + ifp->if_addrlen;
 545         if (socksize < sizeof(*sdl))
 546                 socksize = sizeof(*sdl);
 547         socksize = RT_ROUNDUP(socksize);
 548         ifa = ifa_create(sizeof(struct ifaddr) + 2 * socksize);
 549         sdl = sdl_addr = (struct sockaddr_dl *)(ifa + 1);
 550         sdl->sdl_len = socksize;
 551         sdl->sdl_family = AF_LINK;
 552         bcopy(ifp->if_xname, sdl->sdl_data, namelen);
 553         sdl->sdl_nlen = namelen;
 554         sdl->sdl_type = ifp->if_type;
 555         ifp->if_lladdr = ifa;
 556         ifa->ifa_ifp = ifp;
 557         ifa->ifa_rtrequest = link_rtrequest;
 558         ifa->ifa_addr = (struct sockaddr *)sdl;
 559         sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
 560         ifa->ifa_netmask = (struct sockaddr *)sdl;
 561         sdl->sdl_len = masklen;
 562         while (namelen != 0)
 563                 sdl->sdl_data[--namelen] = 0xff;
 564         ifa_iflink(ifa, ifp, 0 /* Insert head */);
 565
 566         /*
 567          * Make if_data available on all CPUs, since they could
 568          * be updated by hardware interrupt routing, which could
 569          * be bound to any CPU.
 570          */
 571         ifp->if_data_pcpu = kmalloc(ncpus * sizeof(struct ifdata_pcpu),
 572                                     M_DEVBUF,
 573                                     M_WAITOK | M_ZERO | M_CACHEALIGN);
 574
 575         if (ifp->if_mapsubq == NULL)
 576                 ifp->if_mapsubq = ifq_mapsubq_default;
 577
 578         ifq = &ifp->if_snd;
 579         ifq->altq_type = 0;
 580         ifq->altq_disc = NULL;
 581         ifq->altq_flags &= ALTQF_CANTCHANGE;
 582         ifq->altq_tbr = NULL;
 583         ifq->altq_ifp = ifp;
 584
 585         if (ifq->altq_subq_cnt <= 0)
 586                 ifq->altq_subq_cnt = 1;
 587         ifq->altq_subq =
 588                 kmalloc(ifq->altq_subq_cnt * sizeof(struct ifaltq_subque),
 589                         M_DEVBUF,
 590                         M_WAITOK | M_ZERO | M_CACHEALIGN);
 591
 592         if (ifq->altq_maxlen == 0) {
 593                 if_printf(ifp, "driver didn't set altq_maxlen\n");
 594                 ifq_set_maxlen(ifq, ifqmaxlen);
 595         }
 596
 597         /* Allow user to override driver's setting. */
 598         ksnprintf(qlenname, sizeof(qlenname), "net.%s.qlenmax", ifp->if_xname);
 599         qlen = -1;
 600         TUNABLE_INT_FETCH(qlenname, &qlen);
 601         if (qlen > 0) {
 602                 if_printf(ifp, "qlenmax -> %d\n", qlen);
 603                 ifq_set_maxlen(ifq, qlen);
 604         }
 605
 606         for (q = 0; q < ifq->altq_subq_cnt; ++q) {
 607                 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
 608
 609                 ALTQ_SQ_LOCK_INIT(ifsq);
 610                 ifsq->ifsq_index = q;
 611
 612                 ifsq->ifsq_altq = ifq;
 613                 ifsq->ifsq_ifp = ifp;
 614
 615                 ifsq->ifsq_maxlen = ifq->altq_maxlen;
 616                 ifsq->ifsq_maxbcnt = ifsq->ifsq_maxlen * MCLBYTES;
 617                 ifsq->ifsq_prepended = NULL;
 618                 ifsq->ifsq_started = 0;
 619                 ifsq->ifsq_hw_oactive = 0;
 620                 ifsq_set_cpuid(ifsq, 0);
 621                 if (ifp->if_serializer != NULL)
 622                         ifsq_set_hw_serialize(ifsq, ifp->if_serializer);
 623
 624                 /* XXX: netisr_ncpus */
 625                 ifsq->ifsq_stage =
 626                         kmalloc(ncpus * sizeof(struct ifsubq_stage),
 627                                 M_DEVBUF,
 628                                 M_WAITOK | M_ZERO | M_CACHEALIGN);
 629                 for (i = 0; i < ncpus; ++i)
 630                         ifsq->ifsq_stage[i].stg_subq = ifsq;
 631
 632                 /*
 633                  * Allocate one if_start message for each CPU, since
 634                  * the hardware TX ring could be assigned to any CPU.
 635                  *
 636                  * NOTE:
 637                  * If the hardware TX ring polling CPU and the hardware
 638                  * TX ring interrupt CPU are same, one if_start message
 639                  * should be enough.
 640                  */
 641                 ifsq->ifsq_ifstart_nmsg =
 642                     kmalloc(ncpus * sizeof(struct netmsg_base),
 643                     M_LWKTMSG, M_WAITOK);
 644                 for (i = 0; i < ncpus; ++i) {
 645                         netmsg_init(&ifsq->ifsq_ifstart_nmsg[i], NULL,
 646                             &netisr_adone_rport, 0, ifsq_ifstart_dispatch);
 647                         ifsq->ifsq_ifstart_nmsg[i].lmsg.u.ms_resultp = ifsq;
 648                 }
 649         }
 650         ifq_set_classic(ifq);
 651
 652         /*
 653          * Increase mbuf cluster/jcluster limits for the mbufs that
 654          * could sit on the device queues for quite some time.
 655          */
 656         if (ifp->if_nmbclusters > 0)
 657                 mcl_inclimit(ifp->if_nmbclusters);
 658         if (ifp->if_nmbjclusters > 0)
 659                 mjcl_inclimit(ifp->if_nmbjclusters);
 660
 661         /*
 662          * Install this ifp into ifindex2inet, ifnet queue and ifnet
 663          * array after it is setup.
 664          *
 665          * Protect ifindex2ifnet, ifnet queue and ifnet array changes
 666          * by ifnet lock, so that non-netisr threads could get a
 667          * consistent view.
 668          */
 669         ifnet_lock();
 670
 671         /* Don't update if_index until ifindex2ifnet is setup */
 672         ifp->if_index = if_index + 1;
 673         sdl_addr->sdl_index = ifp->if_index;
 674
 675         /*
 676          * Install this ifp into ifindex2ifnet
 677          */
 678         if (ifindex2ifnet == NULL || ifp->if_index >= if_indexlim) {
 679                 unsigned int n;
 680                 struct ifnet **q;
 681
 682                 /*
 683                  * Grow ifindex2ifnet
 684                  */
 685                 if_indexlim <<= 1;
 686                 n = if_indexlim * sizeof(*q);
 687                 q = kmalloc(n, M_IFADDR, M_WAITOK | M_ZERO);
 688                 if (ifindex2ifnet != NULL) {
 689                         bcopy(ifindex2ifnet, q, n/2);
 690                         /* Free old ifindex2ifnet after sync all netisrs */
 691                         old_ifindex2ifnet = ifindex2ifnet;
 692                 }
 693                 ifindex2ifnet = q;
 694         }
 695         ifindex2ifnet[ifp->if_index] = ifp;
 696         /*
 697          * Update if_index after this ifp is installed into ifindex2ifnet,
 698          * so that netisrs could get a consistent view of ifindex2ifnet.
 699          */
 700         cpu_sfence();
 701         if_index = ifp->if_index;
 702
 703         /*
 704          * Install this ifp into ifnet array.
 705          */
 706         /* Free old ifnet array after sync all netisrs */
 707         old_ifnet_array = ifnet_array;
 708         ifnet_array = ifnet_array_add(ifp, old_ifnet_array);
 709
 710         /*
 711          * Install this ifp into ifnet queue.
 712          */
 713         TAILQ_INSERT_TAIL(&ifnetlist, ifp, if_link);
 714
 715         ifnet_unlock();
 716
 717         /*
 718          * Sync all netisrs so that the old ifindex2ifnet and ifnet array
 719          * are no longer accessed and we can free them safely later on.
 720          */
 721         netmsg_service_sync();
 722         if (old_ifindex2ifnet != NULL)
 723                 kfree(old_ifindex2ifnet, M_IFADDR);
 724         ifnet_array_free(old_ifnet_array);
 725
 726         if (!SLIST_EMPTY(&domains))
 727                 if_attachdomain1(ifp);
 728
 729         /* Announce the interface. */
 730         EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
 731         devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
 732         rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
 733 }
 734
 735 static void
 736 if_attachdomain(void *dummy)
 737 {
 738         struct ifnet *ifp;
 739
 740         ifnet_lock();
 741         TAILQ_FOREACH(ifp, &ifnetlist, if_list)
 742                 if_attachdomain1(ifp);
 743         ifnet_unlock();
 744 }
 745 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_FIRST,
 746         if_attachdomain, NULL);
 747
 748 static void
 749 if_attachdomain1(struct ifnet *ifp)
 750 {
 751         struct domain *dp;
 752
 753         crit_enter();
 754
 755         /* address family dependent data region */
 756         bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
 757         SLIST_FOREACH(dp, &domains, dom_next)
 758                 if (dp->dom_ifattach)
 759                         ifp->if_afdata[dp->dom_family] =
 760                                 (*dp->dom_ifattach)(ifp);
 761         crit_exit();
 762 }
 763
 764 /*
 765  * Purge all addresses whose type is _not_ AF_LINK
 766  */
 767 static void
 768 if_purgeaddrs_nolink_dispatch(netmsg_t nmsg)
 769 {
 770         struct ifnet *ifp = nmsg->lmsg.u.ms_resultp;
 771         struct ifaddr_container *ifac, *next;
 772
 773         ASSERT_NETISR0;
 774
 775         /*
 776          * The ifaddr processing in the following loop will block,
 777          * however, this function is called in netisr0, in which
 778          * ifaddr list changes happen, so we don't care about the
 779          * blockness of the ifaddr processing here.
 780          */
 781         TAILQ_FOREACH_MUTABLE(ifac, &ifp->if_addrheads[mycpuid],
 782                               ifa_link, next) {
 783                 struct ifaddr *ifa = ifac->ifa;
 784
 785                 /* Ignore marker */
 786                 if (ifa->ifa_addr->sa_family == AF_UNSPEC)
 787                         continue;
 788
 789                 /* Leave link ifaddr as it is */
 790                 if (ifa->ifa_addr->sa_family == AF_LINK)
 791                         continue;
 792 #ifdef INET
 793                 /* XXX: Ugly!! ad hoc just for INET */
 794                 if (ifa->ifa_addr->sa_family == AF_INET) {
 795                         struct ifaliasreq ifr;
 796                         struct sockaddr_in saved_addr, saved_dst;
 797 #ifdef IFADDR_DEBUG_VERBOSE
 798                         int i;
 799
 800                         kprintf("purge in4 addr %p: ", ifa);
 801                         for (i = 0; i < ncpus; ++i) {
 802                                 kprintf("%d ",
 803                                     ifa->ifa_containers[i].ifa_refcnt);
 804                         }
 805                         kprintf("\n");
 806 #endif
 807
 808                         /* Save information for panic. */
 809                         memcpy(&saved_addr, ifa->ifa_addr, sizeof(saved_addr));
 810                         if (ifa->ifa_dstaddr != NULL) {
 811                                 memcpy(&saved_dst, ifa->ifa_dstaddr,
 812                                     sizeof(saved_dst));
 813                         } else {
 814                                 memset(&saved_dst, 0, sizeof(saved_dst));
 815                         }
 816
 817                         bzero(&ifr, sizeof ifr);
 818                         ifr.ifra_addr = *ifa->ifa_addr;
 819                         if (ifa->ifa_dstaddr)
 820                                 ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
 821                         if (in_control(SIOCDIFADDR, (caddr_t)&ifr, ifp,
 822                                        NULL) == 0)
 823                                 continue;
 824
 825                         /* MUST NOT HAPPEN */
 826                         panic("%s: in_control failed %x, dst %x", ifp->if_xname,
 827                             ntohl(saved_addr.sin_addr.s_addr),
 828                             ntohl(saved_dst.sin_addr.s_addr));
 829                 }
 830 #endif /* INET */
 831 #ifdef INET6
 832                 if (ifa->ifa_addr->sa_family == AF_INET6) {
 833 #ifdef IFADDR_DEBUG_VERBOSE
 834                         int i;
 835
 836                         kprintf("purge in6 addr %p: ", ifa);
 837                         for (i = 0; i < ncpus; ++i) {
 838                                 kprintf("%d ",
 839                                     ifa->ifa_containers[i].ifa_refcnt);
 840                         }
 841                         kprintf("\n");
 842 #endif
 843
 844                         in6_purgeaddr(ifa);
 845                         /* ifp_addrhead is already updated */
 846                         continue;
 847                 }
 848 #endif /* INET6 */
 849                 if_printf(ifp, "destroy ifaddr family %d\n",
 850                     ifa->ifa_addr->sa_family);
 851                 ifa_ifunlink(ifa, ifp);
 852                 ifa_destroy(ifa);
 853         }
 854
 855         netisr_replymsg(&nmsg->base, 0);
 856 }
 857
 858 void
 859 if_purgeaddrs_nolink(struct ifnet *ifp)
 860 {
 861         struct netmsg_base nmsg;
 862
 863         netmsg_init(&nmsg, NULL, &curthread->td_msgport, 0,
 864             if_purgeaddrs_nolink_dispatch);
 865         nmsg.lmsg.u.ms_resultp = ifp;
 866         netisr_domsg(&nmsg, 0);
 867 }
 868
 869 static void
 870 ifq_stage_detach_handler(netmsg_t nmsg)
 871 {
 872         struct ifaltq *ifq = nmsg->lmsg.u.ms_resultp;
 873         int q;
 874
 875         for (q = 0; q < ifq->altq_subq_cnt; ++q) {
 876                 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
 877                 struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid);
 878
 879                 if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED)
 880                         ifsq_stage_remove(&ifsubq_stage_heads[mycpuid], stage);
 881         }
 882         lwkt_replymsg(&nmsg->lmsg, 0);
 883 }
 884
 885 static void
 886 ifq_stage_detach(struct ifaltq *ifq)
 887 {
 888         struct netmsg_base base;
 889         int cpu;
 890
 891         netmsg_init(&base, NULL, &curthread->td_msgport, 0,
 892             ifq_stage_detach_handler);
 893         base.lmsg.u.ms_resultp = ifq;
 894
 895         /* XXX netisr_ncpus */
 896         for (cpu = 0; cpu < ncpus; ++cpu)
 897                 lwkt_domsg(netisr_cpuport(cpu), &base.lmsg, 0);
 898 }
 899
 900 struct netmsg_if_rtdel {
 901         struct netmsg_base      base;
 902         struct ifnet            *ifp;
 903 };
 904
 905 static void
 906 if_rtdel_dispatch(netmsg_t msg)
 907 {
 908         struct netmsg_if_rtdel *rmsg = (void *)msg;
 909         int i, cpu;
 910
 911         cpu = mycpuid;
 912         ASSERT_NETISR_NCPUS(cpu);
 913
 914         for (i = 1; i <= AF_MAX; i++) {
 915                 struct radix_node_head  *rnh;
 916
 917                 if ((rnh = rt_tables[cpu][i]) == NULL)
 918                         continue;
 919                 rnh->rnh_walktree(rnh, if_rtdel, rmsg->ifp);
 920         }
 921         netisr_forwardmsg(&msg->base, cpu + 1);
 922 }
 923
 924 /*
 925  * Detach an interface, removing it from the
 926  * list of "active" interfaces.
 927  */
 928 void
 929 if_detach(struct ifnet *ifp)
 930 {
 931         struct ifnet_array *old_ifnet_array;
 932         struct ifg_list *ifgl;
 933         struct netmsg_if_rtdel msg;
 934         struct domain *dp;
 935         int q;
 936
 937         /* Announce that the interface is gone. */
 938         EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
 939         rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
 940         devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
 941
 942         /*
 943          * Remove this ifp from ifindex2inet, ifnet queue and ifnet
 944          * array before it is whacked.
 945          *
 946          * Protect ifindex2ifnet, ifnet queue and ifnet array changes
 947          * by ifnet lock, so that non-netisr threads could get a
 948          * consistent view.
 949          */
 950         ifnet_lock();
 951
 952         /*
 953          * Remove this ifp from ifindex2ifnet and maybe decrement if_index.
 954          */
 955         ifindex2ifnet[ifp->if_index] = NULL;
 956         while (if_index > 0 && ifindex2ifnet[if_index] == NULL)
 957                 if_index--;
 958
 959         /*
 960          * Remove this ifp from ifnet queue.
 961          */
 962         TAILQ_REMOVE(&ifnetlist, ifp, if_link);
 963
 964         /*
 965          * Remove this ifp from ifnet array.
 966          */
 967         /* Free old ifnet array after sync all netisrs */
 968         old_ifnet_array = ifnet_array;
 969         ifnet_array = ifnet_array_del(ifp, old_ifnet_array);
 970
 971         ifnet_unlock();
 972
 973         ifgroup_lockmgr(LK_EXCLUSIVE);
 974         while ((ifgl = TAILQ_FIRST(&ifp->if_groups)) != NULL)
 975                 if_delgroup_locked(ifp, ifgl->ifgl_group->ifg_group);
 976         ifgroup_lockmgr(LK_RELEASE);
 977
 978         /*
 979          * Sync all netisrs so that the old ifnet array is no longer
 980          * accessed and we can free it safely later on.
 981          */
 982         netmsg_service_sync();
 983         ifnet_array_free(old_ifnet_array);
 984
 985         /*
 986          * Remove routes and flush queues.
 987          */
 988         crit_enter();
 989 #ifdef IFPOLL_ENABLE
 990         if (ifp->if_flags & IFF_NPOLLING)
 991                 ifpoll_deregister(ifp);
 992 #endif
 993         if_down(ifp);
 994
 995         /* Decrease the mbuf clusters/jclusters limits increased by us */
 996         if (ifp->if_nmbclusters > 0)
 997                 mcl_inclimit(-ifp->if_nmbclusters);
 998         if (ifp->if_nmbjclusters > 0)
 999                 mjcl_inclimit(-ifp->if_nmbjclusters);
1000
1001 #ifdef ALTQ
1002         if (ifq_is_enabled(&ifp->if_snd))
1003                 altq_disable(&ifp->if_snd);
1004         if (ifq_is_attached(&ifp->if_snd))
1005                 altq_detach(&ifp->if_snd);
1006 #endif
1007
1008         /*
1009          * Clean up all addresses.
1010          */
1011         ifp->if_lladdr = NULL;
1012
1013         if_purgeaddrs_nolink(ifp);
1014         if (!TAILQ_EMPTY(&ifp->if_addrheads[mycpuid])) {
1015                 struct ifaddr *ifa;
1016
1017                 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
1018                 KASSERT(ifa->ifa_addr->sa_family == AF_LINK,
1019                         ("non-link ifaddr is left on if_addrheads"));
1020
1021                 ifa_ifunlink(ifa, ifp);
1022                 ifa_destroy(ifa);
1023                 KASSERT(TAILQ_EMPTY(&ifp->if_addrheads[mycpuid]),
1024                         ("there are still ifaddrs left on if_addrheads"));
1025         }
1026
1027 #ifdef INET
1028         /*
1029          * Remove all IPv4 kernel structures related to ifp.
1030          */
1031         in_ifdetach(ifp);
1032 #endif
1033
1034 #ifdef INET6
1035         /*
1036          * Remove all IPv6 kernel structs related to ifp.  This should be done
1037          * before removing routing entries below, since IPv6 interface direct
1038          * routes are expected to be removed by the IPv6-specific kernel API.
1039          * Otherwise, the kernel will detect some inconsistency and bark it.
1040          */
1041         in6_ifdetach(ifp);
1042 #endif
1043
1044         /*
1045          * Delete all remaining routes using this interface
1046          */
1047         netmsg_init(&msg.base, NULL, &curthread->td_msgport, MSGF_PRIORITY,
1048             if_rtdel_dispatch);
1049         msg.ifp = ifp;
1050         netisr_domsg_global(&msg.base);
1051
1052         SLIST_FOREACH(dp, &domains, dom_next) {
1053                 if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
1054                         (*dp->dom_ifdetach)(ifp,
1055                                 ifp->if_afdata[dp->dom_family]);
1056         }
1057
1058         kfree(ifp->if_addrheads, M_IFADDR);
1059
1060         lwkt_synchronize_ipiqs("if_detach");
1061         ifq_stage_detach(&ifp->if_snd);
1062
1063         for (q = 0; q < ifp->if_snd.altq_subq_cnt; ++q) {
1064                 struct ifaltq_subque *ifsq = &ifp->if_snd.altq_subq[q];
1065
1066                 kfree(ifsq->ifsq_ifstart_nmsg, M_LWKTMSG);
1067                 kfree(ifsq->ifsq_stage, M_DEVBUF);
1068         }
1069         kfree(ifp->if_snd.altq_subq, M_DEVBUF);
1070
1071         kfree(ifp->if_data_pcpu, M_DEVBUF);
1072
1073         crit_exit();
1074 }
1075
1076 int
1077 ifgroup_lockmgr(u_int flags)
1078 {
1079         return lockmgr(&ifgroup_lock, flags);
1080 }
1081
1082 /*
1083  * Create an empty interface group.
1084  */
1085 static struct ifg_group *
1086 if_creategroup(const char *groupname)
1087 {
1088         struct ifg_group *ifg;
1089
1090         ifg = kmalloc(sizeof(*ifg), M_IFNET, M_WAITOK);
1091         strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group));
1092         ifg->ifg_refcnt = 0;
1093         ifg->ifg_carp_demoted = 0;
1094         TAILQ_INIT(&ifg->ifg_members);
1095
1096         ifgroup_lockmgr(LK_EXCLUSIVE);
1097         TAILQ_INSERT_TAIL(&ifg_head, ifg, ifg_next);
1098         ifgroup_lockmgr(LK_RELEASE);
1099
1100         EVENTHANDLER_INVOKE(group_attach_event, ifg);
1101
1102         return (ifg);
1103 }
1104
1105 /*
1106  * Destroy an empty interface group.
1107  */
1108 static int
1109 if_destroygroup(struct ifg_group *ifg)
1110 {
1111         KASSERT(ifg->ifg_refcnt == 0,
1112                 ("trying to delete a non-empty interface group"));
1113
1114         ifgroup_lockmgr(LK_EXCLUSIVE);
1115         TAILQ_REMOVE(&ifg_head, ifg, ifg_next);
1116         ifgroup_lockmgr(LK_RELEASE);
1117
1118         EVENTHANDLER_INVOKE(group_detach_event, ifg);
1119         kfree(ifg, M_IFNET);
1120
1121         return (0);
1122 }
1123
1124 /*
1125  * Add the interface to a group.
1126  * The target group will be created if it doesn't exist.
1127  */
1128 int
1129 if_addgroup(struct ifnet *ifp, const char *groupname)
1130 {
1131         struct ifg_list *ifgl;
1132         struct ifg_group *ifg;
1133         struct ifg_member *ifgm;
1134
1135         if (groupname[0] &&
1136             groupname[strlen(groupname) - 1] >= '0' &&
1137             groupname[strlen(groupname) - 1] <= '9')
1138                 return (EINVAL);
1139
1140         ifgroup_lockmgr(LK_SHARED);
1141
1142         TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1143                 if (strcmp(ifgl->ifgl_group->ifg_group, groupname) == 0) {
1144                         ifgroup_lockmgr(LK_RELEASE);
1145                         return (EEXIST);
1146                 }
1147         }
1148
1149         TAILQ_FOREACH(ifg, &ifg_head, ifg_next) {
1150                 if (strcmp(ifg->ifg_group, groupname) == 0)
1151                         break;
1152         }
1153
1154         ifgroup_lockmgr(LK_RELEASE);
1155
1156         if (ifg == NULL)
1157                 ifg = if_creategroup(groupname);
1158
1159         ifgl = kmalloc(sizeof(*ifgl), M_IFNET, M_WAITOK);
1160         ifgm = kmalloc(sizeof(*ifgm), M_IFNET, M_WAITOK);
1161         ifgl->ifgl_group = ifg;
1162         ifgm->ifgm_ifp = ifp;
1163         ifg->ifg_refcnt++;
1164
1165         ifgroup_lockmgr(LK_EXCLUSIVE);
1166         TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next);
1167         TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next);
1168         ifgroup_lockmgr(LK_RELEASE);
1169
1170         EVENTHANDLER_INVOKE(group_change_event, groupname);
1171
1172         return (0);
1173 }
1174
1175 /*
1176  * Remove the interface from a group.
1177  * The group will be destroyed if it becomes empty.
1178  *
1179  * The 'ifgroup_lock' must be hold exclusively when calling this.
1180  */
1181 static int
1182 if_delgroup_locked(struct ifnet *ifp, const char *groupname)
1183 {
1184         struct ifg_list *ifgl;
1185         struct ifg_member *ifgm;
1186
1187         KKASSERT(lockstatus(&ifgroup_lock, curthread) == LK_EXCLUSIVE);
1188
1189         TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1190                 if (strcmp(ifgl->ifgl_group->ifg_group, groupname) == 0)
1191                         break;
1192         }
1193         if (ifgl == NULL)
1194                 return (ENOENT);
1195
1196         TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
1197
1198         TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next) {
1199                 if (ifgm->ifgm_ifp == ifp)
1200                         break;
1201         }
1202
1203         if (ifgm != NULL) {
1204                 TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next);
1205
1206                 ifgroup_lockmgr(LK_RELEASE);
1207                 EVENTHANDLER_INVOKE(group_change_event, groupname);
1208                 ifgroup_lockmgr(LK_EXCLUSIVE);
1209
1210                 kfree(ifgm, M_IFNET);
1211                 ifgl->ifgl_group->ifg_refcnt--;
1212         }
1213
1214         if (ifgl->ifgl_group->ifg_refcnt == 0) {
1215                 ifgroup_lockmgr(LK_RELEASE);
1216                 if_destroygroup(ifgl->ifgl_group);
1217                 ifgroup_lockmgr(LK_EXCLUSIVE);
1218         }
1219
1220         kfree(ifgl, M_IFNET);
1221
1222         return (0);
1223 }
1224
1225 int
1226 if_delgroup(struct ifnet *ifp, const char *groupname)
1227 {
1228         int error;
1229
1230         ifgroup_lockmgr(LK_EXCLUSIVE);
1231         error = if_delgroup_locked(ifp, groupname);
1232         ifgroup_lockmgr(LK_RELEASE);
1233
1234         return (error);
1235 }
1236
1237 /*
1238  * Store all the groups that the interface belongs to in memory
1239  * pointed to by data.
1240  */
1241 static int
1242 if_getgroups(struct ifgroupreq *ifgr, struct ifnet *ifp)
1243 {
1244         struct ifg_list *ifgl;
1245         struct ifg_req *ifgrq, *p;
1246         int len, error;
1247
1248         len = 0;
1249         ifgroup_lockmgr(LK_SHARED);
1250         TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1251                 len += sizeof(struct ifg_req);
1252         ifgroup_lockmgr(LK_RELEASE);
1253
1254         if (ifgr->ifgr_len == 0) {
1255                 /*
1256                  * Caller is asking how much memory should be allocated in
1257                  * the next request in order to hold all the groups.
1258                  */
1259                 ifgr->ifgr_len = len;
1260                 return (0);
1261         } else if (ifgr->ifgr_len != len) {
1262                 return (EINVAL);
1263         }
1264
1265         ifgrq = kmalloc(len, M_TEMP, M_INTWAIT | M_NULLOK | M_ZERO);
1266         if (ifgrq == NULL)
1267                 return (ENOMEM);
1268
1269         ifgroup_lockmgr(LK_SHARED);
1270         p = ifgrq;
1271         TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1272                 if (len < sizeof(struct ifg_req)) {
1273                         ifgroup_lockmgr(LK_RELEASE);
1274                         error = EINVAL;
1275                         goto failed;
1276                 }
1277
1278                 strlcpy(p->ifgrq_group, ifgl->ifgl_group->ifg_group,
1279                         sizeof(ifgrq->ifgrq_group));
1280                 len -= sizeof(struct ifg_req);
1281                 p++;
1282         }
1283         ifgroup_lockmgr(LK_RELEASE);
1284
1285         error = copyout(ifgrq, ifgr->ifgr_groups, ifgr->ifgr_len);
1286 failed:
1287         kfree(ifgrq, M_TEMP);
1288         return error;
1289 }
1290
1291 /*
1292  * Store all the members of a group in memory pointed to by data.
1293  */
1294 static int
1295 if_getgroupmembers(struct ifgroupreq *ifgr)
1296 {
1297         struct ifg_group *ifg;
1298         struct ifg_member *ifgm;
1299         struct ifg_req *ifgrq, *p;
1300         int len, error;
1301
1302         ifgroup_lockmgr(LK_SHARED);
1303
1304         TAILQ_FOREACH(ifg, &ifg_head, ifg_next) {
1305                 if (strcmp(ifg->ifg_group, ifgr->ifgr_name) == 0)
1306                         break;
1307         }
1308         if (ifg == NULL) {
1309                 ifgroup_lockmgr(LK_RELEASE);
1310                 return (ENOENT);
1311         }
1312
1313         len = 0;
1314         TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
1315                 len += sizeof(struct ifg_req);
1316
1317         ifgroup_lockmgr(LK_RELEASE);
1318
1319         if (ifgr->ifgr_len == 0) {
1320                 ifgr->ifgr_len = len;
1321                 return (0);
1322         } else if (ifgr->ifgr_len != len) {
1323                 return (EINVAL);
1324         }
1325
1326         ifgrq = kmalloc(len, M_TEMP, M_INTWAIT | M_NULLOK | M_ZERO);
1327         if (ifgrq == NULL)
1328                 return (ENOMEM);
1329
1330         ifgroup_lockmgr(LK_SHARED);
1331         p = ifgrq;
1332         TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
1333                 if (len < sizeof(struct ifg_req)) {
1334                         ifgroup_lockmgr(LK_RELEASE);
1335                         error = EINVAL;
1336                         goto failed;
1337                 }
1338
1339                 strlcpy(p->ifgrq_member, ifgm->ifgm_ifp->if_xname,
1340                         sizeof(p->ifgrq_member));
1341                 len -= sizeof(struct ifg_req);
1342                 p++;
1343         }
1344         ifgroup_lockmgr(LK_RELEASE);
1345
1346         error = copyout(ifgrq, ifgr->ifgr_groups, ifgr->ifgr_len);
1347 failed:
1348         kfree(ifgrq, M_TEMP);
1349         return error;
1350 }
1351
1352 /*
1353  * Delete Routes for a Network Interface
1354  *
1355  * Called for each routing entry via the rnh->rnh_walktree() call above
1356  * to delete all route entries referencing a detaching network interface.
1357  *
1358  * Arguments:
1359  *      rn      pointer to node in the routing table
1360  *      arg     argument passed to rnh->rnh_walktree() - detaching interface
1361  *
1362  * Returns:
1363  *      0       successful
1364  *      errno   failed - reason indicated
1365  *
1366  */
1367 static int
1368 if_rtdel(struct radix_node *rn, void *arg)
1369 {
1370         struct rtentry  *rt = (struct rtentry *)rn;
1371         struct ifnet    *ifp = arg;
1372         int             err;
1373
1374         if (rt->rt_ifp == ifp) {
1375
1376                 /*
1377                  * Protect (sorta) against walktree recursion problems
1378                  * with cloned routes
1379                  */
1380                 if (!(rt->rt_flags & RTF_UP))
1381                         return (0);
1382
1383                 err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1384                                 rt_mask(rt), rt->rt_flags,
1385                                 NULL);
1386                 if (err) {
1387                         log(LOG_WARNING, "if_rtdel: error %d\n", err);
1388                 }
1389         }
1390
1391         return (0);
1392 }
1393
1394 static __inline boolean_t
1395 ifa_prefer(const struct ifaddr *cur_ifa, const struct ifaddr *old_ifa)
1396 {
1397         if (old_ifa == NULL)
1398                 return TRUE;
1399
1400         if ((old_ifa->ifa_ifp->if_flags & IFF_UP) == 0 &&
1401             (cur_ifa->ifa_ifp->if_flags & IFF_UP))
1402                 return TRUE;
1403         if ((old_ifa->ifa_flags & IFA_ROUTE) == 0 &&
1404             (cur_ifa->ifa_flags & IFA_ROUTE))
1405                 return TRUE;
1406         return FALSE;
1407 }
1408
1409 /*
1410  * Locate an interface based on a complete address.
1411  */
1412 struct ifaddr *
1413 ifa_ifwithaddr(struct sockaddr *addr)
1414 {
1415         const struct ifnet_array *arr;
1416         int i;
1417
1418         arr = ifnet_array_get();
1419         for (i = 0; i < arr->ifnet_count; ++i) {
1420                 struct ifnet *ifp = arr->ifnet_arr[i];
1421                 struct ifaddr_container *ifac;
1422
1423                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1424                         struct ifaddr *ifa = ifac->ifa;
1425
1426                         if (ifa->ifa_addr->sa_family != addr->sa_family)
1427                                 continue;
1428                         if (sa_equal(addr, ifa->ifa_addr))
1429                                 return (ifa);
1430                         if ((ifp->if_flags & IFF_BROADCAST) &&
1431                             ifa->ifa_broadaddr &&
1432                             /* IPv6 doesn't have broadcast */
1433                             ifa->ifa_broadaddr->sa_len != 0 &&
1434                             sa_equal(ifa->ifa_broadaddr, addr))
1435                                 return (ifa);
1436                 }
1437         }
1438         return (NULL);
1439 }
1440
1441 /*
1442  * Locate the point to point interface with a given destination address.
1443  */
1444 struct ifaddr *
1445 ifa_ifwithdstaddr(struct sockaddr *addr)
1446 {
1447         const struct ifnet_array *arr;
1448         int i;
1449
1450         arr = ifnet_array_get();
1451         for (i = 0; i < arr->ifnet_count; ++i) {
1452                 struct ifnet *ifp = arr->ifnet_arr[i];
1453                 struct ifaddr_container *ifac;
1454
1455                 if (!(ifp->if_flags & IFF_POINTOPOINT))
1456                         continue;
1457
1458                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1459                         struct ifaddr *ifa = ifac->ifa;
1460
1461                         if (ifa->ifa_addr->sa_family != addr->sa_family)
1462                                 continue;
1463                         if (ifa->ifa_dstaddr &&
1464                             sa_equal(addr, ifa->ifa_dstaddr))
1465                                 return (ifa);
1466                 }
1467         }
1468         return (NULL);
1469 }
1470
1471 /*
1472  * Find an interface on a specific network.  If many, choice
1473  * is most specific found.
1474  */
1475 struct ifaddr *
1476 ifa_ifwithnet(struct sockaddr *addr)
1477 {
1478         struct ifaddr *ifa_maybe = NULL;
1479         u_int af = addr->sa_family;
1480         char *addr_data = addr->sa_data, *cplim;
1481         const struct ifnet_array *arr;
1482         int i;
1483
1484         /*
1485          * AF_LINK addresses can be looked up directly by their index number,
1486          * so do that if we can.
1487          */
1488         if (af == AF_LINK) {
1489                 struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
1490
1491                 if (sdl->sdl_index && sdl->sdl_index <= if_index)
1492                         return (ifindex2ifnet[sdl->sdl_index]->if_lladdr);
1493         }
1494
1495         /*
1496          * Scan though each interface, looking for ones that have
1497          * addresses in this address family.
1498          */
1499         arr = ifnet_array_get();
1500         for (i = 0; i < arr->ifnet_count; ++i) {
1501                 struct ifnet *ifp = arr->ifnet_arr[i];
1502                 struct ifaddr_container *ifac;
1503
1504                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1505                         struct ifaddr *ifa = ifac->ifa;
1506                         char *cp, *cp2, *cp3;
1507
1508                         if (ifa->ifa_addr->sa_family != af)
1509 next:                           continue;
1510                         if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT) {
1511                                 /*
1512                                  * This is a bit broken as it doesn't
1513                                  * take into account that the remote end may
1514                                  * be a single node in the network we are
1515                                  * looking for.
1516                                  * The trouble is that we don't know the
1517                                  * netmask for the remote end.
1518                                  */
1519                                 if (ifa->ifa_dstaddr != NULL &&
1520                                     sa_equal(addr, ifa->ifa_dstaddr))
1521                                         return (ifa);
1522                         } else {
1523                                 /*
1524                                  * if we have a special address handler,
1525                                  * then use it instead of the generic one.
1526                                  */
1527                                 if (ifa->ifa_claim_addr) {
1528                                         if ((*ifa->ifa_claim_addr)(ifa, addr)) {
1529                                                 return (ifa);
1530                                         } else {
1531                                                 continue;
1532                                         }
1533                                 }
1534
1535                                 /*
1536                                  * Scan all the bits in the ifa's address.
1537                                  * If a bit dissagrees with what we are
1538                                  * looking for, mask it with the netmask
1539                                  * to see if it really matters.
1540                                  * (A byte at a time)
1541                                  */
1542                                 if (ifa->ifa_netmask == 0)
1543                                         continue;
1544                                 cp = addr_data;
1545                                 cp2 = ifa->ifa_addr->sa_data;
1546                                 cp3 = ifa->ifa_netmask->sa_data;
1547                                 cplim = ifa->ifa_netmask->sa_len +
1548                                         (char *)ifa->ifa_netmask;
1549                                 while (cp3 < cplim)
1550                                         if ((*cp++ ^ *cp2++) & *cp3++)
1551                                                 goto next; /* next address! */
1552                                 /*
1553                                  * If the netmask of what we just found
1554                                  * is more specific than what we had before
1555                                  * (if we had one) then remember the new one
1556                                  * before continuing to search for an even
1557                                  * better one.  If the netmasks are equal,
1558                                  * we prefer the this ifa based on the result
1559                                  * of ifa_prefer().
1560                                  */
1561                                 if (ifa_maybe == NULL ||
1562                                     rn_refines((char *)ifa->ifa_netmask,
1563                                         (char *)ifa_maybe->ifa_netmask) ||
1564                                     (sa_equal(ifa_maybe->ifa_netmask,
1565                                         ifa->ifa_netmask) &&
1566                                      ifa_prefer(ifa, ifa_maybe)))
1567                                         ifa_maybe = ifa;
1568                         }
1569                 }
1570         }
1571         return (ifa_maybe);
1572 }
1573
1574 /*
1575  * Find an interface address specific to an interface best matching
1576  * a given address.
1577  */
1578 struct ifaddr *
1579 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1580 {
1581         struct ifaddr_container *ifac;
1582         char *cp, *cp2, *cp3;
1583         char *cplim;
1584         struct ifaddr *ifa_maybe = NULL;
1585         u_int af = addr->sa_family;
1586
1587         if (af >= AF_MAX)
1588                 return (0);
1589         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1590                 struct ifaddr *ifa = ifac->ifa;
1591
1592                 if (ifa->ifa_addr->sa_family != af)
1593                         continue;
1594                 if (ifa_maybe == NULL)
1595                         ifa_maybe = ifa;
1596                 if (ifa->ifa_netmask == NULL) {
1597                         if (sa_equal(addr, ifa->ifa_addr) ||
1598                             (ifa->ifa_dstaddr != NULL &&
1599                              sa_equal(addr, ifa->ifa_dstaddr)))
1600                                 return (ifa);
1601                         continue;
1602                 }
1603                 if (ifp->if_flags & IFF_POINTOPOINT) {
1604                         if (sa_equal(addr, ifa->ifa_dstaddr))
1605                                 return (ifa);
1606                 } else {
1607                         cp = addr->sa_data;
1608                         cp2 = ifa->ifa_addr->sa_data;
1609                         cp3 = ifa->ifa_netmask->sa_data;
1610                         cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1611                         for (; cp3 < cplim; cp3++)
1612                                 if ((*cp++ ^ *cp2++) & *cp3)
1613                                         break;
1614                         if (cp3 == cplim)
1615                                 return (ifa);
1616                 }
1617         }
1618         return (ifa_maybe);
1619 }
1620
1621 /*
1622  * Default action when installing a route with a Link Level gateway.
1623  * Lookup an appropriate real ifa to point to.
1624  * This should be moved to /sys/net/link.c eventually.
1625  */
1626 static void
1627 link_rtrequest(int cmd, struct rtentry *rt)
1628 {
1629         struct ifaddr *ifa;
1630         struct sockaddr *dst;
1631         struct ifnet *ifp;
1632
1633         if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL ||
1634             (ifp = ifa->ifa_ifp) == NULL || (dst = rt_key(rt)) == NULL)
1635                 return;
1636         ifa = ifaof_ifpforaddr(dst, ifp);
1637         if (ifa != NULL) {
1638                 IFAFREE(rt->rt_ifa);
1639                 IFAREF(ifa);
1640                 rt->rt_ifa = ifa;
1641                 if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1642                         ifa->ifa_rtrequest(cmd, rt);
1643         }
1644 }
1645
1646 struct netmsg_if {
1647         struct netmsg_base      base;
1648         struct ifnet            *ifp;
1649 };
1650
1651 /*
1652  * Mark an interface down and notify protocols of the transition.
1653  */
1654 static void
1655 if_down_dispatch(netmsg_t nmsg)
1656 {
1657         struct netmsg_if *msg = (struct netmsg_if *)nmsg;
1658         struct ifnet *ifp = msg->ifp;
1659         struct ifaddr_container *ifac;
1660         struct domain *dp;
1661
1662         ASSERT_NETISR0;
1663
1664         ifp->if_flags &= ~IFF_UP;
1665         getmicrotime(&ifp->if_lastchange);
1666         rt_ifmsg(ifp);
1667
1668         /*
1669          * The ifaddr processing in the following loop will block,
1670          * however, this function is called in netisr0, in which
1671          * ifaddr list changes happen, so we don't care about the
1672          * blockness of the ifaddr processing here.
1673          */
1674         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1675                 struct ifaddr *ifa = ifac->ifa;
1676
1677                 /* Ignore marker */
1678                 if (ifa->ifa_addr->sa_family == AF_UNSPEC)
1679                         continue;
1680
1681                 kpfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1682         }
1683
1684         SLIST_FOREACH(dp, &domains, dom_next)
1685                 if (dp->dom_if_down != NULL)
1686                         dp->dom_if_down(ifp);
1687
1688         ifq_purge_all(&ifp->if_snd);
1689         netisr_replymsg(&nmsg->base, 0);
1690 }
1691
1692 /*
1693  * Mark an interface up and notify protocols of the transition.
1694  */
1695 static void
1696 if_up_dispatch(netmsg_t nmsg)
1697 {
1698         struct netmsg_if *msg = (struct netmsg_if *)nmsg;
1699         struct ifnet *ifp = msg->ifp;
1700         struct ifaddr_container *ifac;
1701         struct domain *dp;
1702
1703         ASSERT_NETISR0;
1704
1705         ifq_purge_all(&ifp->if_snd);
1706         ifp->if_flags |= IFF_UP;
1707         getmicrotime(&ifp->if_lastchange);
1708         rt_ifmsg(ifp);
1709
1710         /*
1711          * The ifaddr processing in the following loop will block,
1712          * however, this function is called in netisr0, in which
1713          * ifaddr list changes happen, so we don't care about the
1714          * blockness of the ifaddr processing here.
1715          */
1716         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1717                 struct ifaddr *ifa = ifac->ifa;
1718
1719                 /* Ignore marker */
1720                 if (ifa->ifa_addr->sa_family == AF_UNSPEC)
1721                         continue;
1722
1723                 kpfctlinput(PRC_IFUP, ifa->ifa_addr);
1724         }
1725
1726         SLIST_FOREACH(dp, &domains, dom_next)
1727                 if (dp->dom_if_up != NULL)
1728                         dp->dom_if_up(ifp);
1729
1730         netisr_replymsg(&nmsg->base, 0);
1731 }
1732
1733 /*
1734  * Mark an interface down and notify protocols of the transition.  An
1735  * interface going down is also considered to be a synchronizing event.
1736  * We must ensure that all packet processing related to the interface
1737  * has completed before we return so e.g. the caller can free the ifnet
1738  * structure that the mbufs may be referencing.
1739  *
1740  * NOTE: must be called at splnet or eqivalent.
1741  */
1742 void
1743 if_down(struct ifnet *ifp)
1744 {
1745         struct netmsg_if msg;
1746
1747         EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_DOWN);
1748         netmsg_init(&msg.base, NULL, &curthread->td_msgport, 0,
1749             if_down_dispatch);
1750         msg.ifp = ifp;
1751         netisr_domsg(&msg.base, 0);
1752         netmsg_service_sync();
1753 }
1754
1755 /*
1756  * Mark an interface up and notify protocols of
1757  * the transition.
1758  * NOTE: must be called at splnet or eqivalent.
1759  */
1760 void
1761 if_up(struct ifnet *ifp)
1762 {
1763         struct netmsg_if msg;
1764
1765         netmsg_init(&msg.base, NULL, &curthread->td_msgport, 0,
1766             if_up_dispatch);
1767         msg.ifp = ifp;
1768         netisr_domsg(&msg.base, 0);
1769         EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_UP);
1770 }
1771
1772 /*
1773  * Process a link state change.
1774  * NOTE: must be called at splsoftnet or equivalent.
1775  */
1776 void
1777 if_link_state_change(struct ifnet *ifp)
1778 {
1779         int link_state = ifp->if_link_state;
1780
1781         rt_ifmsg(ifp);
1782         devctl_notify("IFNET", ifp->if_xname,
1783             (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL);
1784
1785         EVENTHANDLER_INVOKE(ifnet_link_event, ifp, link_state);
1786 }
1787
1788 /*
1789  * Handle interface watchdog timer routines.  Called
1790  * from softclock, we decrement timers (if set) and
1791  * call the appropriate interface routine on expiration.
1792  */
1793 static void
1794 if_slowtimo_dispatch(netmsg_t nmsg)
1795 {
1796         struct globaldata *gd = mycpu;
1797         const struct ifnet_array *arr;
1798         int i;
1799
1800         ASSERT_NETISR0;
1801
1802         crit_enter_gd(gd);
1803         lwkt_replymsg(&nmsg->lmsg, 0);  /* reply ASAP */
1804         crit_exit_gd(gd);
1805
1806         arr = ifnet_array_get();
1807         for (i = 0; i < arr->ifnet_count; ++i) {
1808                 struct ifnet *ifp = arr->ifnet_arr[i];
1809
1810                 crit_enter_gd(gd);
1811
1812                 if (if_stats_compat) {
1813                         IFNET_STAT_GET(ifp, ipackets, ifp->if_ipackets);
1814                         IFNET_STAT_GET(ifp, ierrors, ifp->if_ierrors);
1815                         IFNET_STAT_GET(ifp, opackets, ifp->if_opackets);
1816                         IFNET_STAT_GET(ifp, oerrors, ifp->if_oerrors);
1817                         IFNET_STAT_GET(ifp, collisions, ifp->if_collisions);
1818                         IFNET_STAT_GET(ifp, ibytes, ifp->if_ibytes);
1819                         IFNET_STAT_GET(ifp, obytes, ifp->if_obytes);
1820                         IFNET_STAT_GET(ifp, imcasts, ifp->if_imcasts);
1821                         IFNET_STAT_GET(ifp, omcasts, ifp->if_omcasts);
1822                         IFNET_STAT_GET(ifp, iqdrops, ifp->if_iqdrops);
1823                         IFNET_STAT_GET(ifp, noproto, ifp->if_noproto);
1824                         IFNET_STAT_GET(ifp, oqdrops, ifp->if_oqdrops);
1825                 }
1826
1827                 if (ifp->if_timer == 0 || --ifp->if_timer) {
1828                         crit_exit_gd(gd);
1829                         continue;
1830                 }
1831                 if (ifp->if_watchdog) {
1832                         if (ifnet_tryserialize_all(ifp)) {
1833                                 (*ifp->if_watchdog)(ifp);
1834                                 ifnet_deserialize_all(ifp);
1835                         } else {
1836                                 /* try again next timeout */
1837                                 ++ifp->if_timer;
1838                         }
1839                 }
1840
1841                 crit_exit_gd(gd);
1842         }
1843
1844         callout_reset(&if_slowtimo_timer, hz / IFNET_SLOWHZ, if_slowtimo, NULL);
1845 }
1846
1847 static void
1848 if_slowtimo(void *arg __unused)
1849 {
1850         struct lwkt_msg *lmsg = &if_slowtimo_netmsg.lmsg;
1851
1852         KASSERT(mycpuid == 0, ("not on cpu0"));
1853         crit_enter();
1854         if (lmsg->ms_flags & MSGF_DONE)
1855                 lwkt_sendmsg_oncpu(netisr_cpuport(0), lmsg);
1856         crit_exit();
1857 }
1858
1859 /*
1860  * Map interface name to
1861  * interface structure pointer.
1862  */
1863 struct ifnet *
1864 ifunit(const char *name)
1865 {
1866         struct ifnet *ifp;
1867
1868         /*
1869          * Search all the interfaces for this name/number
1870          */
1871         KASSERT(mtx_owned(&ifnet_mtx), ("ifnet is not locked"));
1872
1873         TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
1874                 if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1875                         break;
1876         }
1877         return (ifp);
1878 }
1879
1880 struct ifnet *
1881 ifunit_netisr(const char *name)
1882 {
1883         const struct ifnet_array *arr;
1884         int i;
1885
1886         /*
1887          * Search all the interfaces for this name/number
1888          */
1889
1890         arr = ifnet_array_get();
1891         for (i = 0; i < arr->ifnet_count; ++i) {
1892                 struct ifnet *ifp = arr->ifnet_arr[i];
1893
1894                 if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1895                         return ifp;
1896         }
1897         return NULL;
1898 }
1899
1900 /*
1901  * Interface ioctls.
1902  */
1903 int
1904 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct ucred *cred)
1905 {
1906         struct ifnet *ifp;
1907         struct ifgroupreq *ifgr;
1908         struct ifreq *ifr;
1909         struct ifstat *ifs;
1910         int error, do_ifup = 0;
1911         short oif_flags;
1912         int new_flags;
1913         size_t namelen, onamelen;
1914         char new_name[IFNAMSIZ];
1915         struct ifaddr *ifa;
1916         struct sockaddr_dl *sdl;
1917
1918         switch (cmd) {
1919         case SIOCGIFCONF:
1920                 return (ifconf(cmd, data, cred));
1921         default:
1922                 break;
1923         }
1924
1925         ifr = (struct ifreq *)data;
1926
1927         switch (cmd) {
1928         case SIOCIFCREATE:
1929         case SIOCIFCREATE2:
1930                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1931                         return (error);
1932                 return (if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name),
1933                         cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL));
1934         case SIOCIFDESTROY:
1935                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1936                         return (error);
1937                 return (if_clone_destroy(ifr->ifr_name));
1938         case SIOCIFGCLONERS:
1939                 return (if_clone_list((struct if_clonereq *)data));
1940         case SIOCGIFGMEMB:
1941                 return (if_getgroupmembers((struct ifgroupreq *)data));
1942         default:
1943                 break;
1944         }
1945
1946         /*
1947          * Nominal ioctl through interface, lookup the ifp and obtain a
1948          * lock to serialize the ifconfig ioctl operation.
1949          */
1950         ifnet_lock();
1951
1952         ifp = ifunit(ifr->ifr_name);
1953         if (ifp == NULL) {
1954                 ifnet_unlock();
1955                 return (ENXIO);
1956         }
1957         error = 0;
1958
1959         switch (cmd) {
1960         case SIOCGIFINDEX:
1961                 ifr->ifr_index = ifp->if_index;
1962                 break;
1963
1964         case SIOCGIFFLAGS:
1965                 ifr->ifr_flags = ifp->if_flags;
1966                 ifr->ifr_flagshigh = ifp->if_flags >> 16;
1967                 break;
1968
1969         case SIOCGIFCAP:
1970                 ifr->ifr_reqcap = ifp->if_capabilities;
1971                 ifr->ifr_curcap = ifp->if_capenable;
1972                 break;
1973
1974         case SIOCGIFMETRIC:
1975                 ifr->ifr_metric = ifp->if_metric;
1976                 break;
1977
1978         case SIOCGIFMTU:
1979                 ifr->ifr_mtu = ifp->if_mtu;
1980                 break;
1981
1982         case SIOCGIFTSOLEN:
1983                 ifr->ifr_tsolen = ifp->if_tsolen;
1984                 break;
1985
1986         case SIOCGIFDATA:
1987                 error = copyout((caddr_t)&ifp->if_data, ifr->ifr_data,
1988                                 sizeof(ifp->if_data));
1989                 break;
1990
1991         case SIOCGIFPHYS:
1992                 ifr->ifr_phys = ifp->if_physical;
1993                 break;
1994
1995         case SIOCGIFPOLLCPU:
1996                 ifr->ifr_pollcpu = -1;
1997                 break;
1998
1999         case SIOCSIFPOLLCPU:
2000                 break;
2001
2002         case SIOCSIFFLAGS:
2003                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2004                 if (error)
2005                         break;
2006                 new_flags = (ifr->ifr_flags & 0xffff) |
2007                     (ifr->ifr_flagshigh << 16);
2008                 if (ifp->if_flags & IFF_SMART) {
2009                         /* Smart drivers twiddle their own routes */
2010                 } else if (ifp->if_flags & IFF_UP &&
2011                     (new_flags & IFF_UP) == 0) {
2012                         if_down(ifp);
2013                 } else if (new_flags & IFF_UP &&
2014                     (ifp->if_flags & IFF_UP) == 0) {
2015                         do_ifup = 1;
2016                 }
2017
2018 #ifdef IFPOLL_ENABLE
2019                 if ((new_flags ^ ifp->if_flags) & IFF_NPOLLING) {
2020                         if (new_flags & IFF_NPOLLING)
2021                                 ifpoll_register(ifp);
2022                         else
2023                                 ifpoll_deregister(ifp);
2024                 }
2025 #endif
2026
2027                 ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
2028                         (new_flags &~ IFF_CANTCHANGE);
2029                 if (new_flags & IFF_PPROMISC) {
2030                         /* Permanently promiscuous mode requested */
2031                         ifp->if_flags |= IFF_PROMISC;
2032                 } else if (ifp->if_pcount == 0) {
2033                         ifp->if_flags &= ~IFF_PROMISC;
2034                 }
2035                 if (ifp->if_ioctl) {
2036                         ifnet_serialize_all(ifp);
2037                         ifp->if_ioctl(ifp, cmd, data, cred);
2038                         ifnet_deserialize_all(ifp);
2039                 }
2040                 if (do_ifup)
2041                         if_up(ifp);
2042                 getmicrotime(&ifp->if_lastchange);
2043                 break;
2044
2045         case SIOCSIFCAP:
2046                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2047                 if (error)
2048                         break;
2049                 if (ifr->ifr_reqcap & ~ifp->if_capabilities) {
2050                         error = EINVAL;
2051                         break;
2052                 }
2053                 ifnet_serialize_all(ifp);
2054                 ifp->if_ioctl(ifp, cmd, data, cred);
2055                 ifnet_deserialize_all(ifp);
2056                 break;
2057
2058         case SIOCSIFNAME:
2059                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2060                 if (error)
2061                         break;
2062                 error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
2063                 if (error)
2064                         break;
2065                 if (new_name[0] == '\0') {
2066                         error = EINVAL;
2067                         break;
2068                 }
2069                 if (ifunit(new_name) != NULL) {
2070                         error = EEXIST;
2071                         break;
2072                 }
2073
2074                 EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
2075
2076                 /* Announce the departure of the interface. */
2077                 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
2078
2079                 strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
2080                 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
2081                 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
2082                 namelen = strlen(new_name);
2083                 onamelen = sdl->sdl_nlen;
2084                 /*
2085                  * Move the address if needed.  This is safe because we
2086                  * allocate space for a name of length IFNAMSIZ when we
2087                  * create this in if_attach().
2088                  */
2089                 if (namelen != onamelen) {
2090                         bcopy(sdl->sdl_data + onamelen,
2091                             sdl->sdl_data + namelen, sdl->sdl_alen);
2092                 }
2093                 bcopy(new_name, sdl->sdl_data, namelen);
2094                 sdl->sdl_nlen = namelen;
2095                 sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
2096                 bzero(sdl->sdl_data, onamelen);
2097                 while (namelen != 0)
2098                         sdl->sdl_data[--namelen] = 0xff;
2099
2100                 EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
2101
2102                 /* Announce the return of the interface. */
2103                 rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
2104                 break;
2105
2106         case SIOCSIFMETRIC:
2107                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2108                 if (error)
2109                         break;
2110                 ifp->if_metric = ifr->ifr_metric;
2111                 getmicrotime(&ifp->if_lastchange);
2112                 break;
2113
2114         case SIOCSIFPHYS:
2115                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2116                 if (error)
2117                         break;
2118                 if (ifp->if_ioctl == NULL) {
2119                         error = EOPNOTSUPP;
2120                         break;
2121                 }
2122                 ifnet_serialize_all(ifp);
2123                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2124                 ifnet_deserialize_all(ifp);
2125                 if (error == 0)
2126                         getmicrotime(&ifp->if_lastchange);
2127                 break;
2128
2129         case SIOCSIFMTU:
2130         {
2131                 u_long oldmtu = ifp->if_mtu;
2132
2133                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2134                 if (error)
2135                         break;
2136                 if (ifp->if_ioctl == NULL) {
2137                         error = EOPNOTSUPP;
2138                         break;
2139                 }
2140                 if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU) {
2141                         error = EINVAL;
2142                         break;
2143                 }
2144                 ifnet_serialize_all(ifp);
2145                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2146                 ifnet_deserialize_all(ifp);
2147                 if (error == 0) {
2148                         getmicrotime(&ifp->if_lastchange);
2149                         rt_ifmsg(ifp);
2150                 }
2151                 /*
2152                  * If the link MTU changed, do network layer specific procedure.
2153                  */
2154                 if (ifp->if_mtu != oldmtu) {
2155 #ifdef INET6
2156                         nd6_setmtu(ifp);
2157 #endif
2158                 }
2159                 break;
2160         }
2161
2162         case SIOCSIFTSOLEN:
2163                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2164                 if (error)
2165                         break;
2166
2167                 /* XXX need driver supplied upper limit */
2168                 if (ifr->ifr_tsolen <= 0) {
2169                         error = EINVAL;
2170                         break;
2171                 }
2172                 ifp->if_tsolen = ifr->ifr_tsolen;
2173                 break;
2174
2175         case SIOCADDMULTI:
2176         case SIOCDELMULTI:
2177                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2178                 if (error)
2179                         break;
2180
2181                 /* Don't allow group membership on non-multicast interfaces. */
2182                 if ((ifp->if_flags & IFF_MULTICAST) == 0) {
2183                         error = EOPNOTSUPP;
2184                         break;
2185                 }
2186
2187                 /* Don't let users screw up protocols' entries. */
2188                 if (ifr->ifr_addr.sa_family != AF_LINK) {
2189                         error = EINVAL;
2190                         break;
2191                 }
2192
2193                 if (cmd == SIOCADDMULTI) {
2194                         struct ifmultiaddr *ifma;
2195                         error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
2196                 } else {
2197                         error = if_delmulti(ifp, &ifr->ifr_addr);
2198                 }
2199                 if (error == 0)
2200                         getmicrotime(&ifp->if_lastchange);
2201                 break;
2202
2203         case SIOCSIFPHYADDR:
2204         case SIOCDIFPHYADDR:
2205 #ifdef INET6
2206         case SIOCSIFPHYADDR_IN6:
2207 #endif
2208         case SIOCSLIFPHYADDR:
2209         case SIOCSIFMEDIA:
2210         case SIOCSIFGENERIC:
2211                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2212                 if (error)
2213                         break;
2214                 if (ifp->if_ioctl == NULL) {
2215                         error = EOPNOTSUPP;
2216                         break;
2217                 }
2218                 ifnet_serialize_all(ifp);
2219                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2220                 ifnet_deserialize_all(ifp);
2221                 if (error == 0)
2222                         getmicrotime(&ifp->if_lastchange);
2223                 break;
2224
2225         case SIOCGIFSTATUS:
2226                 ifs = (struct ifstat *)data;
2227                 ifs->ascii[0] = '\0';
2228                 /* fall through */
2229         case SIOCGIFPSRCADDR:
2230         case SIOCGIFPDSTADDR:
2231         case SIOCGLIFPHYADDR:
2232         case SIOCGIFMEDIA:
2233         case SIOCGIFGENERIC:
2234                 if (ifp->if_ioctl == NULL) {
2235                         error = EOPNOTSUPP;
2236                         break;
2237                 }
2238                 ifnet_serialize_all(ifp);
2239                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2240                 ifnet_deserialize_all(ifp);
2241                 break;
2242
2243         case SIOCSIFLLADDR:
2244                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2245                 if (error)
2246                         break;
2247                 error = if_setlladdr(ifp, ifr->ifr_addr.sa_data,
2248                                      ifr->ifr_addr.sa_len);
2249                 EVENTHANDLER_INVOKE(iflladdr_event, ifp);
2250                 break;
2251
2252         case SIOCAIFGROUP:
2253                 ifgr = (struct ifgroupreq *)ifr;
2254                 if ((error = priv_check_cred(cred, PRIV_NET_ADDIFGROUP, 0)))
2255                         return (error);
2256                 if ((error = if_addgroup(ifp, ifgr->ifgr_group)))
2257                         return (error);
2258                 break;
2259
2260         case SIOCDIFGROUP:
2261                 ifgr = (struct ifgroupreq *)ifr;
2262                 if ((error = priv_check_cred(cred, PRIV_NET_DELIFGROUP, 0)))
2263                         return (error);
2264                 if ((error = if_delgroup(ifp, ifgr->ifgr_group)))
2265                         return (error);
2266                 break;
2267
2268         case SIOCGIFGROUP:
2269                 ifgr = (struct ifgroupreq *)ifr;
2270                 if ((error = if_getgroups(ifgr, ifp)))
2271                         return (error);
2272                 break;
2273
2274         default:
2275                 oif_flags = ifp->if_flags;
2276                 if (so->so_proto == 0) {
2277                         error = EOPNOTSUPP;
2278                         break;
2279                 }
2280                 error = so_pru_control_direct(so, cmd, data, ifp);
2281
2282                 /*
2283                  * If the socket control method returns EOPNOTSUPP, pass the
2284                  * request directly to the interface.
2285                  *
2286                  * Exclude the SIOCSIF{ADDR,BRDADDR,DSTADDR,NETMASK} ioctls,
2287                  * because drivers may trust these ioctls to come from an
2288                  * already privileged layer and thus do not perform credentials
2289                  * checks or input validation.
2290                  */
2291                 if (error == EOPNOTSUPP &&
2292                     ifp->if_ioctl != NULL &&
2293                     cmd != SIOCSIFADDR &&
2294                     cmd != SIOCSIFBRDADDR &&
2295                     cmd != SIOCSIFDSTADDR &&
2296                     cmd != SIOCSIFNETMASK) {
2297                         ifnet_serialize_all(ifp);
2298                         error = ifp->if_ioctl(ifp, cmd, data, cred);
2299                         ifnet_deserialize_all(ifp);
2300                 }
2301
2302                 if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
2303 #ifdef INET6
2304                         DELAY(100);/* XXX: temporary workaround for fxp issue*/
2305                         if (ifp->if_flags & IFF_UP) {
2306                                 crit_enter();
2307                                 in6_if_up(ifp);
2308                                 crit_exit();
2309                         }
2310 #endif
2311                 }
2312                 break;
2313         }
2314
2315         ifnet_unlock();
2316         return (error);
2317 }
2318
2319 /*
2320  * Set/clear promiscuous mode on interface ifp based on the truth value
2321  * of pswitch.  The calls are reference counted so that only the first
2322  * "on" request actually has an effect, as does the final "off" request.
2323  * Results are undefined if the "off" and "on" requests are not matched.
2324  */
2325 int
2326 ifpromisc(struct ifnet *ifp, int pswitch)
2327 {
2328         struct ifreq ifr;
2329         int error;
2330         int oldflags;
2331
2332         oldflags = ifp->if_flags;
2333         if (ifp->if_flags & IFF_PPROMISC) {
2334                 /* Do nothing if device is in permanently promiscuous mode */
2335                 ifp->if_pcount += pswitch ? 1 : -1;
2336                 return (0);
2337         }
2338         if (pswitch) {
2339                 /*
2340                  * If the device is not configured up, we cannot put it in
2341                  * promiscuous mode.
2342                  */
2343                 if ((ifp->if_flags & IFF_UP) == 0)
2344                         return (ENETDOWN);
2345                 if (ifp->if_pcount++ != 0)
2346                         return (0);
2347                 ifp->if_flags |= IFF_PROMISC;
2348                 log(LOG_INFO, "%s: promiscuous mode enabled\n",
2349                     ifp->if_xname);
2350         } else {
2351                 if (--ifp->if_pcount > 0)
2352                         return (0);
2353                 ifp->if_flags &= ~IFF_PROMISC;
2354                 log(LOG_INFO, "%s: promiscuous mode disabled\n",
2355                     ifp->if_xname);
2356         }
2357         ifr.ifr_flags = ifp->if_flags;
2358         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2359         ifnet_serialize_all(ifp);
2360         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr, NULL);
2361         ifnet_deserialize_all(ifp);
2362         if (error == 0)
2363                 rt_ifmsg(ifp);
2364         else
2365                 ifp->if_flags = oldflags;
2366         return error;
2367 }
2368
2369 /*
2370  * Return interface configuration
2371  * of system.  List may be used
2372  * in later ioctl's (above) to get
2373  * other information.
2374  */
2375 static int
2376 ifconf(u_long cmd, caddr_t data, struct ucred *cred)
2377 {
2378         struct ifconf *ifc = (struct ifconf *)data;
2379         struct ifnet *ifp;
2380         struct sockaddr *sa;
2381         struct ifreq ifr, *ifrp;
2382         int space = ifc->ifc_len, error = 0;
2383
2384         ifrp = ifc->ifc_req;
2385
2386         ifnet_lock();
2387         TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
2388                 struct ifaddr_container *ifac, *ifac_mark;
2389                 struct ifaddr_marker mark;
2390                 struct ifaddrhead *head;
2391                 int addrs;
2392
2393                 if (space <= sizeof ifr)
2394                         break;
2395
2396                 /*
2397                  * Zero the stack declared structure first to prevent
2398                  * memory disclosure.
2399                  */
2400                 bzero(&ifr, sizeof(ifr));
2401                 if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
2402                     >= sizeof(ifr.ifr_name)) {
2403                         error = ENAMETOOLONG;
2404                         break;
2405                 }
2406
2407                 /*
2408                  * Add a marker, since copyout() could block and during that
2409                  * period the list could be changed.  Inserting the marker to
2410                  * the header of the list will not cause trouble for the code
2411                  * assuming that the first element of the list is AF_LINK; the
2412                  * marker will be moved to the next position w/o blocking.
2413                  */
2414                 ifa_marker_init(&mark, ifp);
2415                 ifac_mark = &mark.ifac;
2416                 head = &ifp->if_addrheads[mycpuid];
2417
2418                 addrs = 0;
2419                 TAILQ_INSERT_HEAD(head, ifac_mark, ifa_link);
2420                 while ((ifac = TAILQ_NEXT(ifac_mark, ifa_link)) != NULL) {
2421                         struct ifaddr *ifa = ifac->ifa;
2422
2423                         TAILQ_REMOVE(head, ifac_mark, ifa_link);
2424                         TAILQ_INSERT_AFTER(head, ifac, ifac_mark, ifa_link);
2425
2426                         /* Ignore marker */
2427                         if (ifa->ifa_addr->sa_family == AF_UNSPEC)
2428                                 continue;
2429
2430                         if (space <= sizeof ifr)
2431                                 break;
2432                         sa = ifa->ifa_addr;
2433                         if (cred->cr_prison && prison_if(cred, sa))
2434                                 continue;
2435                         addrs++;
2436                         /*
2437                          * Keep a reference on this ifaddr, so that it will
2438                          * not be destroyed when its address is copied to
2439                          * the userland, which could block.
2440                          */
2441                         IFAREF(ifa);
2442                         if (sa->sa_len <= sizeof(*sa)) {
2443                                 ifr.ifr_addr = *sa;
2444                                 error = copyout(&ifr, ifrp, sizeof ifr);
2445                                 ifrp++;
2446                         } else {
2447                                 if (space < (sizeof ifr) + sa->sa_len -
2448                                             sizeof(*sa)) {
2449                                         IFAFREE(ifa);
2450                                         break;
2451                                 }
2452                                 space -= sa->sa_len - sizeof(*sa);
2453                                 error = copyout(&ifr, ifrp,
2454                                                 sizeof ifr.ifr_name);
2455                                 if (error == 0)
2456                                         error = copyout(sa, &ifrp->ifr_addr,
2457                                                         sa->sa_len);
2458                                 ifrp = (struct ifreq *)
2459                                         (sa->sa_len + (caddr_t)&ifrp->ifr_addr);
2460                         }
2461                         IFAFREE(ifa);
2462                         if (error)
2463                                 break;
2464                         space -= sizeof ifr;
2465                 }
2466                 TAILQ_REMOVE(head, ifac_mark, ifa_link);
2467                 if (error)
2468                         break;
2469                 if (!addrs) {
2470                         bzero(&ifr.ifr_addr, sizeof ifr.ifr_addr);
2471                         error = copyout(&ifr, ifrp, sizeof ifr);
2472                         if (error)
2473                                 break;
2474                         space -= sizeof ifr;
2475                         ifrp++;
2476                 }
2477         }
2478         ifnet_unlock();
2479
2480         ifc->ifc_len -= space;
2481         return (error);
2482 }
2483
2484 /*
2485  * Just like if_promisc(), but for all-multicast-reception mode.
2486  */
2487 int
2488 if_allmulti(struct ifnet *ifp, int onswitch)
2489 {
2490         int error = 0;
2491         struct ifreq ifr;
2492
2493         crit_enter();
2494
2495         if (onswitch) {
2496                 if (ifp->if_amcount++ == 0) {
2497                         ifp->if_flags |= IFF_ALLMULTI;
2498                         ifr.ifr_flags = ifp->if_flags;
2499                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2500                         ifnet_serialize_all(ifp);
2501                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2502                                               NULL);
2503                         ifnet_deserialize_all(ifp);
2504                 }
2505         } else {
2506                 if (ifp->if_amcount > 1) {
2507                         ifp->if_amcount--;
2508                 } else {
2509                         ifp->if_amcount = 0;
2510                         ifp->if_flags &= ~IFF_ALLMULTI;
2511                         ifr.ifr_flags = ifp->if_flags;
2512                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2513                         ifnet_serialize_all(ifp);
2514                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2515                                               NULL);
2516                         ifnet_deserialize_all(ifp);
2517                 }
2518         }
2519
2520         crit_exit();
2521
2522         if (error == 0)
2523                 rt_ifmsg(ifp);
2524         return error;
2525 }
2526
2527 /*
2528  * Add a multicast listenership to the interface in question.
2529  * The link layer provides a routine which converts
2530  */
2531 int
2532 if_addmulti_serialized(struct ifnet *ifp, struct sockaddr *sa,
2533     struct ifmultiaddr **retifma)
2534 {
2535         struct sockaddr *llsa, *dupsa;
2536         int error;
2537         struct ifmultiaddr *ifma;
2538
2539         ASSERT_IFNET_SERIALIZED_ALL(ifp);
2540
2541         /*
2542          * If the matching multicast address already exists
2543          * then don't add a new one, just add a reference
2544          */
2545         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2546                 if (sa_equal(sa, ifma->ifma_addr)) {
2547                         ifma->ifma_refcount++;
2548                         if (retifma)
2549                                 *retifma = ifma;
2550                         return 0;
2551                 }
2552         }
2553
2554         /*
2555          * Give the link layer a chance to accept/reject it, and also
2556          * find out which AF_LINK address this maps to, if it isn't one
2557          * already.
2558          */
2559         if (ifp->if_resolvemulti) {
2560                 error = ifp->if_resolvemulti(ifp, &llsa, sa);
2561                 if (error)
2562                         return error;
2563         } else {
2564                 llsa = NULL;
2565         }
2566
2567         ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_INTWAIT);
2568         dupsa = kmalloc(sa->sa_len, M_IFMADDR, M_INTWAIT);
2569         bcopy(sa, dupsa, sa->sa_len);
2570
2571         ifma->ifma_addr = dupsa;
2572         ifma->ifma_lladdr = llsa;
2573         ifma->ifma_ifp = ifp;
2574         ifma->ifma_refcount = 1;
2575         ifma->ifma_protospec = NULL;
2576         rt_newmaddrmsg(RTM_NEWMADDR, ifma);
2577
2578         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2579         if (retifma)
2580                 *retifma = ifma;
2581
2582         if (llsa != NULL) {
2583                 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2584                         if (sa_equal(ifma->ifma_addr, llsa))
2585                                 break;
2586                 }
2587                 if (ifma) {
2588                         ifma->ifma_refcount++;
2589                 } else {
2590                         ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_INTWAIT);
2591                         dupsa = kmalloc(llsa->sa_len, M_IFMADDR, M_INTWAIT);
2592                         bcopy(llsa, dupsa, llsa->sa_len);
2593                         ifma->ifma_addr = dupsa;
2594                         ifma->ifma_ifp = ifp;
2595                         ifma->ifma_refcount = 1;
2596                         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2597                 }
2598         }
2599         /*
2600          * We are certain we have added something, so call down to the
2601          * interface to let them know about it.
2602          */
2603         if (ifp->if_ioctl)
2604                 ifp->if_ioctl(ifp, SIOCADDMULTI, 0, NULL);
2605
2606         return 0;
2607 }
2608
2609 int
2610 if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
2611     struct ifmultiaddr **retifma)
2612 {
2613         int error;
2614
2615         ifnet_serialize_all(ifp);
2616         error = if_addmulti_serialized(ifp, sa, retifma);
2617         ifnet_deserialize_all(ifp);
2618
2619         return error;
2620 }
2621
2622 /*
2623  * Remove a reference to a multicast address on this interface.  Yell
2624  * if the request does not match an existing membership.
2625  */
2626 static int
2627 if_delmulti_serialized(struct ifnet *ifp, struct sockaddr *sa)
2628 {
2629         struct ifmultiaddr *ifma;
2630
2631         ASSERT_IFNET_SERIALIZED_ALL(ifp);
2632
2633         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2634                 if (sa_equal(sa, ifma->ifma_addr))
2635                         break;
2636         if (ifma == NULL)
2637                 return ENOENT;
2638
2639         if (ifma->ifma_refcount > 1) {
2640                 ifma->ifma_refcount--;
2641                 return 0;
2642         }
2643
2644         rt_newmaddrmsg(RTM_DELMADDR, ifma);
2645         sa = ifma->ifma_lladdr;
2646         TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2647         /*
2648          * Make sure the interface driver is notified
2649          * in the case of a link layer mcast group being left.
2650          */
2651         if (ifma->ifma_addr->sa_family == AF_LINK && sa == NULL)
2652                 ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2653         kfree(ifma->ifma_addr, M_IFMADDR);
2654         kfree(ifma, M_IFMADDR);
2655         if (sa == NULL)
2656                 return 0;
2657
2658         /*
2659          * Now look for the link-layer address which corresponds to
2660          * this network address.  It had been squirreled away in
2661          * ifma->ifma_lladdr for this purpose (so we don't have
2662          * to call ifp->if_resolvemulti() again), and we saved that
2663          * value in sa above.  If some nasty deleted the
2664          * link-layer address out from underneath us, we can deal because
2665          * the address we stored was is not the same as the one which was
2666          * in the record for the link-layer address.  (So we don't complain
2667          * in that case.)
2668          */
2669         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2670                 if (sa_equal(sa, ifma->ifma_addr))
2671                         break;
2672         if (ifma == NULL)
2673                 return 0;
2674
2675         if (ifma->ifma_refcount > 1) {
2676                 ifma->ifma_refcount--;
2677                 return 0;
2678         }
2679
2680         TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2681         ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2682         kfree(ifma->ifma_addr, M_IFMADDR);
2683         kfree(sa, M_IFMADDR);
2684         kfree(ifma, M_IFMADDR);
2685
2686         return 0;
2687 }
2688
2689 int
2690 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
2691 {
2692         int error;
2693
2694         ifnet_serialize_all(ifp);
2695         error = if_delmulti_serialized(ifp, sa);
2696         ifnet_deserialize_all(ifp);
2697
2698         return error;
2699 }
2700
2701 /*
2702  * Delete all multicast group membership for an interface.
2703  * Should be used to quickly flush all multicast filters.
2704  */
2705 void
2706 if_delallmulti_serialized(struct ifnet *ifp)
2707 {
2708         struct ifmultiaddr *ifma, mark;
2709         struct sockaddr sa;
2710
2711         ASSERT_IFNET_SERIALIZED_ALL(ifp);
2712
2713         bzero(&sa, sizeof(sa));
2714         sa.sa_family = AF_UNSPEC;
2715         sa.sa_len = sizeof(sa);
2716
2717         bzero(&mark, sizeof(mark));
2718         mark.ifma_addr = &sa;
2719
2720         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, &mark, ifma_link);
2721         while ((ifma = TAILQ_NEXT(&mark, ifma_link)) != NULL) {
2722                 TAILQ_REMOVE(&ifp->if_multiaddrs, &mark, ifma_link);
2723                 TAILQ_INSERT_AFTER(&ifp->if_multiaddrs, ifma, &mark,
2724                     ifma_link);
2725
2726                 if (ifma->ifma_addr->sa_family == AF_UNSPEC)
2727                         continue;
2728
2729                 if_delmulti_serialized(ifp, ifma->ifma_addr);
2730         }
2731         TAILQ_REMOVE(&ifp->if_multiaddrs, &mark, ifma_link);
2732 }
2733
2734
2735 /*
2736  * Set the link layer address on an interface.
2737  *
2738  * At this time we only support certain types of interfaces,
2739  * and we don't allow the length of the address to change.
2740  */
2741 int
2742 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
2743 {
2744         struct sockaddr_dl *sdl;
2745         struct ifreq ifr;
2746
2747         sdl = IF_LLSOCKADDR(ifp);
2748         if (sdl == NULL)
2749                 return (EINVAL);
2750         if (len != sdl->sdl_alen)       /* don't allow length to change */
2751                 return (EINVAL);
2752         switch (ifp->if_type) {
2753         case IFT_ETHER:                 /* these types use struct arpcom */
2754         case IFT_XETHER:
2755         case IFT_L2VLAN:
2756         case IFT_IEEE8023ADLAG:
2757                 bcopy(lladdr, ((struct arpcom *)ifp->if_softc)->ac_enaddr, len);
2758                 bcopy(lladdr, LLADDR(sdl), len);
2759                 break;
2760         default:
2761                 return (ENODEV);
2762         }
2763         /*
2764          * If the interface is already up, we need
2765          * to re-init it in order to reprogram its
2766          * address filter.
2767          */
2768         ifnet_serialize_all(ifp);
2769         if ((ifp->if_flags & IFF_UP) != 0) {
2770 #ifdef INET
2771                 struct ifaddr_container *ifac;
2772 #endif
2773
2774                 ifp->if_flags &= ~IFF_UP;
2775                 ifr.ifr_flags = ifp->if_flags;
2776                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2777                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2778                               NULL);
2779                 ifp->if_flags |= IFF_UP;
2780                 ifr.ifr_flags = ifp->if_flags;
2781                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2782                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2783                                  NULL);
2784 #ifdef INET
2785                 /*
2786                  * Also send gratuitous ARPs to notify other nodes about
2787                  * the address change.
2788                  */
2789                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2790                         struct ifaddr *ifa = ifac->ifa;
2791
2792                         if (ifa->ifa_addr != NULL &&
2793                             ifa->ifa_addr->sa_family == AF_INET)
2794                                 arp_gratuitous(ifp, ifa);
2795                 }
2796 #endif
2797         }
2798         ifnet_deserialize_all(ifp);
2799         return (0);
2800 }
2801
2802
2803 /*
2804  * Locate an interface based on a complete address.
2805  */
2806 struct ifnet *
2807 if_bylla(const void *lla, unsigned char lla_len)
2808 {
2809         const struct ifnet_array *arr;
2810         struct ifnet *ifp;
2811         struct sockaddr_dl *sdl;
2812         int i;
2813
2814         arr = ifnet_array_get();
2815         for (i = 0; i < arr->ifnet_count; ++i) {
2816                 ifp = arr->ifnet_arr[i];
2817                 if (ifp->if_addrlen != lla_len)
2818                         continue;
2819
2820                 sdl = IF_LLSOCKADDR(ifp);
2821                 if (memcmp(lla, LLADDR(sdl), lla_len) == 0)
2822                         return (ifp);
2823         }
2824         return (NULL);
2825 }
2826
2827 struct ifmultiaddr *
2828 ifmaof_ifpforaddr(struct sockaddr *sa, struct ifnet *ifp)
2829 {
2830         struct ifmultiaddr *ifma;
2831
2832         /* TODO: need ifnet_serialize_main */
2833         ifnet_serialize_all(ifp);
2834         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2835                 if (sa_equal(ifma->ifma_addr, sa))
2836                         break;
2837         ifnet_deserialize_all(ifp);
2838
2839         return ifma;
2840 }
2841
2842 /*
2843  * This function locates the first real ethernet MAC from a network
2844  * card and loads it into node, returning 0 on success or ENOENT if
2845  * no suitable interfaces were found.  It is used by the uuid code to
2846  * generate a unique 6-byte number.
2847  */
2848 int
2849 if_getanyethermac(uint16_t *node, int minlen)
2850 {
2851         struct ifnet *ifp;
2852         struct sockaddr_dl *sdl;
2853
2854         ifnet_lock();
2855         TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
2856                 if (ifp->if_type != IFT_ETHER)
2857                         continue;
2858                 sdl = IF_LLSOCKADDR(ifp);
2859                 if (sdl->sdl_alen < minlen)
2860                         continue;
2861                 bcopy(((struct arpcom *)ifp->if_softc)->ac_enaddr, node,
2862                       minlen);
2863                 ifnet_unlock();
2864                 return(0);
2865         }
2866         ifnet_unlock();
2867         return (ENOENT);
2868 }
2869
2870 /*
2871  * The name argument must be a pointer to storage which will last as
2872  * long as the interface does.  For physical devices, the result of
2873  * device_get_name(dev) is a good choice and for pseudo-devices a
2874  * static string works well.
2875  */
2876 void
2877 if_initname(struct ifnet *ifp, const char *name, int unit)
2878 {
2879         ifp->if_dname = name;
2880         ifp->if_dunit = unit;
2881         if (unit != IF_DUNIT_NONE)
2882                 ksnprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
2883         else
2884                 strlcpy(ifp->if_xname, name, IFNAMSIZ);
2885 }
2886
2887 int
2888 if_printf(struct ifnet *ifp, const char *fmt, ...)
2889 {
2890         __va_list ap;
2891         int retval;
2892
2893         retval = kprintf("%s: ", ifp->if_xname);
2894         __va_start(ap, fmt);
2895         retval += kvprintf(fmt, ap);
2896         __va_end(ap);
2897         return (retval);
2898 }
2899
2900 struct ifnet *
2901 if_alloc(uint8_t type)
2902 {
2903         struct ifnet *ifp;
2904         size_t size;
2905
2906         /*
2907          * XXX temporary hack until arpcom is setup in if_l2com
2908          */
2909         if (type == IFT_ETHER)
2910                 size = sizeof(struct arpcom);
2911         else
2912                 size = sizeof(struct ifnet);
2913
2914         ifp = kmalloc(size, M_IFNET, M_WAITOK|M_ZERO);
2915
2916         ifp->if_type = type;
2917
2918         if (if_com_alloc[type] != NULL) {
2919                 ifp->if_l2com = if_com_alloc[type](type, ifp);
2920                 if (ifp->if_l2com == NULL) {
2921                         kfree(ifp, M_IFNET);
2922                         return (NULL);
2923                 }
2924         }
2925         return (ifp);
2926 }
2927
2928 void
2929 if_free(struct ifnet *ifp)
2930 {
2931         kfree(ifp, M_IFNET);
2932 }
2933
2934 void
2935 ifq_set_classic(struct ifaltq *ifq)
2936 {
2937         ifq_set_methods(ifq, ifq->altq_ifp->if_mapsubq,
2938             ifsq_classic_enqueue, ifsq_classic_dequeue, ifsq_classic_request);
2939 }
2940
2941 void
2942 ifq_set_methods(struct ifaltq *ifq, altq_mapsubq_t mapsubq,
2943     ifsq_enqueue_t enqueue, ifsq_dequeue_t dequeue, ifsq_request_t request)
2944 {
2945         int q;
2946
2947         KASSERT(mapsubq != NULL, ("mapsubq is not specified"));
2948         KASSERT(enqueue != NULL, ("enqueue is not specified"));
2949         KASSERT(dequeue != NULL, ("dequeue is not specified"));
2950         KASSERT(request != NULL, ("request is not specified"));
2951
2952         ifq->altq_mapsubq = mapsubq;
2953         for (q = 0; q < ifq->altq_subq_cnt; ++q) {
2954                 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
2955
2956                 ifsq->ifsq_enqueue = enqueue;
2957                 ifsq->ifsq_dequeue = dequeue;
2958                 ifsq->ifsq_request = request;
2959         }
2960 }
2961
2962 static void
2963 ifsq_norm_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
2964 {
2965
2966         classq_add(&ifsq->ifsq_norm, m);
2967         ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
2968 }
2969
2970 static void
2971 ifsq_prio_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
2972 {
2973
2974         classq_add(&ifsq->ifsq_prio, m);
2975         ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
2976         ALTQ_SQ_PRIO_CNTR_INC(ifsq, m->m_pkthdr.len);
2977 }
2978
2979 static struct mbuf *
2980 ifsq_norm_dequeue(struct ifaltq_subque *ifsq)
2981 {
2982         struct mbuf *m;
2983
2984         m = classq_get(&ifsq->ifsq_norm);
2985         if (m != NULL)
2986                 ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
2987         return (m);
2988 }
2989
2990 static struct mbuf *
2991 ifsq_prio_dequeue(struct ifaltq_subque *ifsq)
2992 {
2993         struct mbuf *m;
2994
2995         m = classq_get(&ifsq->ifsq_prio);
2996         if (m != NULL) {
2997                 ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
2998                 ALTQ_SQ_PRIO_CNTR_DEC(ifsq, m->m_pkthdr.len);
2999         }
3000         return (m);
3001 }
3002
3003 int
3004 ifsq_classic_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m,
3005     struct altq_pktattr *pa __unused)
3006 {
3007
3008         M_ASSERTPKTHDR(m);
3009 again:
3010         if (ifsq->ifsq_len >= ifsq->ifsq_maxlen ||
3011             ifsq->ifsq_bcnt >= ifsq->ifsq_maxbcnt) {
3012                 struct mbuf *m_drop;
3013
3014                 if (m->m_flags & M_PRIO) {
3015                         m_drop = NULL;
3016                         if (ifsq->ifsq_prio_len < (ifsq->ifsq_maxlen >> 1) &&
3017                             ifsq->ifsq_prio_bcnt < (ifsq->ifsq_maxbcnt >> 1)) {
3018                                 /* Try dropping some from normal queue. */
3019                                 m_drop = ifsq_norm_dequeue(ifsq);
3020                         }
3021                         if (m_drop == NULL)
3022                                 m_drop = ifsq_prio_dequeue(ifsq);
3023                 } else {
3024                         m_drop = ifsq_norm_dequeue(ifsq);
3025                 }
3026                 if (m_drop != NULL) {
3027                         IFNET_STAT_INC(ifsq->ifsq_ifp, oqdrops, 1);
3028                         m_freem(m_drop);
3029                         goto again;
3030                 }
3031                 /*
3032                  * No old packets could be dropped!
3033                  * NOTE: Caller increases oqdrops.
3034                  */
3035                 m_freem(m);
3036                 return (ENOBUFS);
3037         } else {
3038                 if (m->m_flags & M_PRIO)
3039                         ifsq_prio_enqueue(ifsq, m);
3040                 else
3041                         ifsq_norm_enqueue(ifsq, m);
3042                 return (0);
3043         }
3044 }
3045
3046 struct mbuf *
3047 ifsq_classic_dequeue(struct ifaltq_subque *ifsq, int op)
3048 {
3049         struct mbuf *m;
3050
3051         switch (op) {
3052         case ALTDQ_POLL:
3053                 m = classq_head(&ifsq->ifsq_prio);
3054                 if (m == NULL)
3055                         m = classq_head(&ifsq->ifsq_norm);
3056                 break;
3057
3058         case ALTDQ_REMOVE:
3059                 m = ifsq_prio_dequeue(ifsq);
3060                 if (m == NULL)
3061                         m = ifsq_norm_dequeue(ifsq);
3062                 break;
3063
3064         default:
3065                 panic("unsupported ALTQ dequeue op: %d", op);
3066         }
3067         return m;
3068 }
3069
3070 int
3071 ifsq_classic_request(struct ifaltq_subque *ifsq, int req, void *arg)
3072 {
3073         switch (req) {
3074         case ALTRQ_PURGE:
3075                 for (;;) {
3076                         struct mbuf *m;
3077
3078                         m = ifsq_classic_dequeue(ifsq, ALTDQ_REMOVE);
3079                         if (m == NULL)
3080                                 break;
3081                         m_freem(m);
3082                 }
3083                 break;
3084
3085         default:
3086                 panic("unsupported ALTQ request: %d", req);
3087         }
3088         return 0;
3089 }
3090
3091 static void
3092 ifsq_ifstart_try(struct ifaltq_subque *ifsq, int force_sched)
3093 {
3094         struct ifnet *ifp = ifsq_get_ifp(ifsq);
3095         int running = 0, need_sched;
3096
3097         /*
3098          * Try to do direct ifnet.if_start on the subqueue first, if there is
3099          * contention on the subqueue hardware serializer, ifnet.if_start on
3100          * the subqueue will be scheduled on the subqueue owner CPU.
3101          */
3102         if (!ifsq_tryserialize_hw(ifsq)) {
3103                 /*
3104                  * Subqueue hardware serializer contention happened,
3105                  * ifnet.if_start on the subqueue is scheduled on
3106                  * the subqueue owner CPU, and we keep going.
3107                  */
3108                 ifsq_ifstart_schedule(ifsq, 1);
3109                 return;
3110         }
3111
3112         if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
3113                 ifp->if_start(ifp, ifsq);
3114                 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
3115                         running = 1;
3116         }
3117         need_sched = ifsq_ifstart_need_schedule(ifsq, running);
3118
3119         ifsq_deserialize_hw(ifsq);
3120
3121         if (need_sched) {
3122                 /*
3123                  * More data need to be transmitted, ifnet.if_start on the
3124                  * subqueue is scheduled on the subqueue owner CPU, and we
3125                  * keep going.
3126                  * NOTE: ifnet.if_start subqueue interlock is not released.
3127                  */
3128                 ifsq_ifstart_schedule(ifsq, force_sched);
3129         }
3130 }
3131
3132 /*
3133  * Subqeue packets staging mechanism:
3134  *
3135  * The packets enqueued into the subqueue are staged to a certain amount
3136  * before the ifnet.if_start on the subqueue is called.  In this way, the
3137  * driver could avoid writing to hardware registers upon every packet,
3138  * instead, hardware registers could be written when certain amount of
3139  * packets are put onto hardware TX ring.  The measurement on several modern
3140  * NICs (emx(4), igb(4), bnx(4), bge(4), jme(4)) shows that the hardware
3141  * registers writing aggregation could save ~20% CPU time when 18bytes UDP
3142  * datagrams are transmitted at 1.48Mpps.  The performance improvement by
3143  * hardware registers writing aggeregation is also mentioned by Luigi Rizzo's
3144  * netmap paper (http://info.iet.unipi.it/~luigi/netmap/).
3145  *
3146  * Subqueue packets staging is performed for two entry points into drivers'
3147  * transmission function:
3148  * - Direct ifnet.if_start calling on the subqueue, i.e. ifsq_ifstart_try()
3149  * - ifnet.if_start scheduling on the subqueue, i.e. ifsq_ifstart_schedule()
3150  *
3151  * Subqueue packets staging will be stopped upon any of the following
3152  * conditions:
3153  * - If the count of packets enqueued on the current CPU is great than or
3154  *   equal to ifsq_stage_cntmax. (XXX this should be per-interface)
3155  * - If the total length of packets enqueued on the current CPU is great
3156  *   than or equal to the hardware's MTU - max_protohdr.  max_protohdr is
3157  *   cut from the hardware's MTU mainly bacause a full TCP segment's size
3158  *   is usually less than hardware's MTU.
3159  * - ifsq_ifstart_schedule() is not pending on the current CPU and
3160  *   ifnet.if_start subqueue interlock (ifaltq_subq.ifsq_started) is not
3161  *   released.
3162  * - The if_start_rollup(), which is registered as low priority netisr
3163  *   rollup function, is called; probably because no more work is pending
3164  *   for netisr.
3165  *
3166  * NOTE:
3167  * Currently subqueue packet staging is only performed in netisr threads.
3168  */
3169 int
3170 ifq_dispatch(struct ifnet *ifp, struct mbuf *m, struct altq_pktattr *pa)
3171 {
3172         struct ifaltq *ifq = &ifp->if_snd;
3173         struct ifaltq_subque *ifsq;
3174         int error, start = 0, len, mcast = 0, avoid_start = 0;
3175         struct ifsubq_stage_head *head = NULL;
3176         struct ifsubq_stage *stage = NULL;
3177         struct globaldata *gd = mycpu;
3178         struct thread *td = gd->gd_curthread;
3179
3180         crit_enter_quick(td);
3181
3182         ifsq = ifq_map_subq(ifq, gd->gd_cpuid);
3183         ASSERT_ALTQ_SQ_NOT_SERIALIZED_HW(ifsq);
3184
3185         len = m->m_pkthdr.len;
3186         if (m->m_flags & M_MCAST)
3187                 mcast = 1;
3188
3189         if (td->td_type == TD_TYPE_NETISR) {
3190                 head = &ifsubq_stage_heads[mycpuid];
3191                 stage = ifsq_get_stage(ifsq, mycpuid);
3192
3193                 stage->stg_cnt++;
3194                 stage->stg_len += len;
3195                 if (stage->stg_cnt < ifsq_stage_cntmax &&
3196                     stage->stg_len < (ifp->if_mtu - max_protohdr))
3197                         avoid_start = 1;
3198         }
3199
3200         ALTQ_SQ_LOCK(ifsq);
3201         error = ifsq_enqueue_locked(ifsq, m, pa);
3202         if (error) {
3203                 IFNET_STAT_INC(ifp, oqdrops, 1);
3204                 if (!ifsq_data_ready(ifsq)) {
3205                         ALTQ_SQ_UNLOCK(ifsq);
3206                         crit_exit_quick(td);
3207                         return error;
3208                 }
3209                 avoid_start = 0;
3210         }
3211         if (!ifsq_is_started(ifsq)) {
3212                 if (avoid_start) {
3213                         ALTQ_SQ_UNLOCK(ifsq);
3214
3215                         KKASSERT(!error);
3216                         if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
3217                                 ifsq_stage_insert(head, stage);
3218
3219                         IFNET_STAT_INC(ifp, obytes, len);
3220                         if (mcast)
3221                                 IFNET_STAT_INC(ifp, omcasts, 1);
3222                         crit_exit_quick(td);
3223                         return error;
3224                 }
3225
3226                 /*
3227                  * Hold the subqueue interlock of ifnet.if_start
3228                  */
3229                 ifsq_set_started(ifsq);
3230                 start = 1;
3231         }
3232         ALTQ_SQ_UNLOCK(ifsq);
3233
3234         if (!error) {
3235                 IFNET_STAT_INC(ifp, obytes, len);
3236                 if (mcast)
3237                         IFNET_STAT_INC(ifp, omcasts, 1);
3238         }
3239
3240         if (stage != NULL) {
3241                 if (!start && (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)) {
3242                         KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
3243                         if (!avoid_start) {
3244                                 ifsq_stage_remove(head, stage);
3245                                 ifsq_ifstart_schedule(ifsq, 1);
3246                         }
3247                         crit_exit_quick(td);
3248                         return error;
3249                 }
3250
3251                 if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED) {
3252                         ifsq_stage_remove(head, stage);
3253                 } else {
3254                         stage->stg_cnt = 0;
3255                         stage->stg_len = 0;
3256                 }
3257         }
3258
3259         if (!start) {
3260                 crit_exit_quick(td);
3261                 return error;
3262         }
3263
3264         ifsq_ifstart_try(ifsq, 0);
3265
3266         crit_exit_quick(td);
3267         return error;
3268 }
3269
3270 void *
3271 ifa_create(int size)
3272 {
3273         struct ifaddr *ifa;
3274         int i;
3275
3276         KASSERT(size >= sizeof(*ifa), ("ifaddr size too small"));
3277
3278         ifa = kmalloc(size, M_IFADDR, M_INTWAIT | M_ZERO);
3279
3280         /*
3281          * Make ifa_container availabel on all CPUs, since they
3282          * could be accessed by any threads.
3283          */
3284         ifa->ifa_containers =
3285                 kmalloc(ncpus * sizeof(struct ifaddr_container),
3286                         M_IFADDR,
3287                         M_INTWAIT | M_ZERO | M_CACHEALIGN);
3288
3289         ifa->ifa_ncnt = ncpus;
3290         for (i = 0; i < ncpus; ++i) {
3291                 struct ifaddr_container *ifac = &ifa->ifa_containers[i];
3292
3293                 ifac->ifa_magic = IFA_CONTAINER_MAGIC;
3294                 ifac->ifa = ifa;
3295                 ifac->ifa_refcnt = 1;
3296         }
3297 #ifdef IFADDR_DEBUG
3298         kprintf("alloc ifa %p %d\n", ifa, size);
3299 #endif
3300         return ifa;
3301 }
3302
3303 void
3304 ifac_free(struct ifaddr_container *ifac, int cpu_id)
3305 {
3306         struct ifaddr *ifa = ifac->ifa;
3307
3308         KKASSERT(ifac->ifa_magic == IFA_CONTAINER_MAGIC);
3309         KKASSERT(ifac->ifa_refcnt == 0);
3310         KASSERT(ifac->ifa_listmask == 0,
3311                 ("ifa is still on %#x lists", ifac->ifa_listmask));
3312
3313         ifac->ifa_magic = IFA_CONTAINER_DEAD;
3314
3315 #ifdef IFADDR_DEBUG_VERBOSE
3316         kprintf("try free ifa %p cpu_id %d\n", ifac->ifa, cpu_id);
3317 #endif
3318
3319         KASSERT(ifa->ifa_ncnt > 0 && ifa->ifa_ncnt <= ncpus,
3320                 ("invalid # of ifac, %d", ifa->ifa_ncnt));
3321         if (atomic_fetchadd_int(&ifa->ifa_ncnt, -1) == 1) {
3322 #ifdef IFADDR_DEBUG
3323                 kprintf("free ifa %p\n", ifa);
3324 #endif
3325                 kfree(ifa->ifa_containers, M_IFADDR);
3326                 kfree(ifa, M_IFADDR);
3327         }
3328 }
3329
3330 static void
3331 ifa_iflink_dispatch(netmsg_t nmsg)
3332 {
3333         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3334         struct ifaddr *ifa = msg->ifa;
3335         struct ifnet *ifp = msg->ifp;
3336         int cpu = mycpuid;
3337         struct ifaddr_container *ifac;
3338
3339         crit_enter();
3340
3341         ifac = &ifa->ifa_containers[cpu];
3342         ASSERT_IFAC_VALID(ifac);
3343         KASSERT((ifac->ifa_listmask & IFA_LIST_IFADDRHEAD) == 0,
3344                 ("ifaddr is on if_addrheads"));
3345
3346         ifac->ifa_listmask |= IFA_LIST_IFADDRHEAD;
3347         if (msg->tail)
3348                 TAILQ_INSERT_TAIL(&ifp->if_addrheads[cpu], ifac, ifa_link);
3349         else
3350                 TAILQ_INSERT_HEAD(&ifp->if_addrheads[cpu], ifac, ifa_link);
3351
3352         crit_exit();
3353
3354         netisr_forwardmsg_all(&nmsg->base, cpu + 1);
3355 }
3356
3357 void
3358 ifa_iflink(struct ifaddr *ifa, struct ifnet *ifp, int tail)
3359 {
3360         struct netmsg_ifaddr msg;
3361
3362         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3363                     0, ifa_iflink_dispatch);
3364         msg.ifa = ifa;
3365         msg.ifp = ifp;
3366         msg.tail = tail;
3367
3368         netisr_domsg(&msg.base, 0);
3369 }
3370
3371 static void
3372 ifa_ifunlink_dispatch(netmsg_t nmsg)
3373 {
3374         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3375         struct ifaddr *ifa = msg->ifa;
3376         struct ifnet *ifp = msg->ifp;
3377         int cpu = mycpuid;
3378         struct ifaddr_container *ifac;
3379
3380         crit_enter();
3381
3382         ifac = &ifa->ifa_containers[cpu];
3383         ASSERT_IFAC_VALID(ifac);
3384         KASSERT(ifac->ifa_listmask & IFA_LIST_IFADDRHEAD,
3385                 ("ifaddr is not on if_addrhead"));
3386
3387         TAILQ_REMOVE(&ifp->if_addrheads[cpu], ifac, ifa_link);
3388         ifac->ifa_listmask &= ~IFA_LIST_IFADDRHEAD;
3389
3390         crit_exit();
3391
3392         netisr_forwardmsg_all(&nmsg->base, cpu + 1);
3393 }
3394
3395 void
3396 ifa_ifunlink(struct ifaddr *ifa, struct ifnet *ifp)
3397 {
3398         struct netmsg_ifaddr msg;
3399
3400         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3401                     0, ifa_ifunlink_dispatch);
3402         msg.ifa = ifa;
3403         msg.ifp = ifp;
3404
3405         netisr_domsg(&msg.base, 0);
3406 }
3407
3408 static void
3409 ifa_destroy_dispatch(netmsg_t nmsg)
3410 {
3411         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3412
3413         IFAFREE(msg->ifa);
3414         netisr_forwardmsg_all(&nmsg->base, mycpuid + 1);
3415 }
3416
3417 void
3418 ifa_destroy(struct ifaddr *ifa)
3419 {
3420         struct netmsg_ifaddr msg;
3421
3422         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3423                     0, ifa_destroy_dispatch);
3424         msg.ifa = ifa;
3425
3426         netisr_domsg(&msg.base, 0);
3427 }
3428
3429 static void
3430 if_start_rollup(void)
3431 {
3432         struct ifsubq_stage_head *head = &ifsubq_stage_heads[mycpuid];
3433         struct ifsubq_stage *stage;
3434
3435         crit_enter();
3436
3437         while ((stage = TAILQ_FIRST(&head->stg_head)) != NULL) {
3438                 struct ifaltq_subque *ifsq = stage->stg_subq;
3439                 int is_sched = 0;
3440
3441                 if (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)
3442                         is_sched = 1;
3443                 ifsq_stage_remove(head, stage);
3444
3445                 if (is_sched) {
3446                         ifsq_ifstart_schedule(ifsq, 1);
3447                 } else {
3448                         int start = 0;
3449
3450                         ALTQ_SQ_LOCK(ifsq);
3451                         if (!ifsq_is_started(ifsq)) {
3452                                 /*
3453                                  * Hold the subqueue interlock of
3454                                  * ifnet.if_start
3455                                  */
3456                                 ifsq_set_started(ifsq);
3457                                 start = 1;
3458                         }
3459                         ALTQ_SQ_UNLOCK(ifsq);
3460
3461                         if (start)
3462                                 ifsq_ifstart_try(ifsq, 1);
3463                 }
3464                 KKASSERT((stage->stg_flags &
3465                     (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
3466         }
3467
3468         crit_exit();
3469 }
3470
3471 static void
3472 ifnetinit(void *dummy __unused)
3473 {
3474         int i;
3475
3476         /* XXX netisr_ncpus */
3477         for (i = 0; i < ncpus; ++i)
3478                 TAILQ_INIT(&ifsubq_stage_heads[i].stg_head);
3479         netisr_register_rollup(if_start_rollup, NETISR_ROLLUP_PRIO_IFSTART);
3480 }
3481
3482 void
3483 if_register_com_alloc(u_char type,
3484     if_com_alloc_t *a, if_com_free_t *f)
3485 {
3486
3487         KASSERT(if_com_alloc[type] == NULL,
3488             ("if_register_com_alloc: %d already registered", type));
3489         KASSERT(if_com_free[type] == NULL,
3490             ("if_register_com_alloc: %d free already registered", type));
3491
3492         if_com_alloc[type] = a;
3493         if_com_free[type] = f;
3494 }
3495
3496 void
3497 if_deregister_com_alloc(u_char type)
3498 {
3499
3500         KASSERT(if_com_alloc[type] != NULL,
3501             ("if_deregister_com_alloc: %d not registered", type));
3502         KASSERT(if_com_free[type] != NULL,
3503             ("if_deregister_com_alloc: %d free not registered", type));
3504         if_com_alloc[type] = NULL;
3505         if_com_free[type] = NULL;
3506 }
3507
3508 void
3509 ifq_set_maxlen(struct ifaltq *ifq, int len)
3510 {
3511         ifq->altq_maxlen = len + (ncpus * ifsq_stage_cntmax);
3512 }
3513
3514 int
3515 ifq_mapsubq_default(struct ifaltq *ifq __unused, int cpuid __unused)
3516 {
3517         return ALTQ_SUBQ_INDEX_DEFAULT;
3518 }
3519
3520 int
3521 ifq_mapsubq_modulo(struct ifaltq *ifq, int cpuid)
3522 {
3523
3524         return (cpuid % ifq->altq_subq_mappriv);
3525 }
3526
3527 static void
3528 ifsq_watchdog(void *arg)
3529 {
3530         struct ifsubq_watchdog *wd = arg;
3531         struct ifnet *ifp;
3532
3533         if (__predict_true(wd->wd_timer == 0 || --wd->wd_timer))
3534                 goto done;
3535
3536         ifp = ifsq_get_ifp(wd->wd_subq);
3537         if (ifnet_tryserialize_all(ifp)) {
3538                 wd->wd_watchdog(wd->wd_subq);
3539                 ifnet_deserialize_all(ifp);
3540         } else {
3541                 /* try again next timeout */
3542                 wd->wd_timer = 1;
3543         }
3544 done:
3545         ifsq_watchdog_reset(wd);
3546 }
3547
3548 static void
3549 ifsq_watchdog_reset(struct ifsubq_watchdog *wd)
3550 {
3551         callout_reset_bycpu(&wd->wd_callout, hz, ifsq_watchdog, wd,
3552             ifsq_get_cpuid(wd->wd_subq));
3553 }
3554
3555 void
3556 ifsq_watchdog_init(struct ifsubq_watchdog *wd, struct ifaltq_subque *ifsq,
3557     ifsq_watchdog_t watchdog)
3558 {
3559         callout_init_mp(&wd->wd_callout);
3560         wd->wd_timer = 0;
3561         wd->wd_subq = ifsq;
3562         wd->wd_watchdog = watchdog;
3563 }
3564
3565 void
3566 ifsq_watchdog_start(struct ifsubq_watchdog *wd)
3567 {
3568         wd->wd_timer = 0;
3569         ifsq_watchdog_reset(wd);
3570 }
3571
3572 void
3573 ifsq_watchdog_stop(struct ifsubq_watchdog *wd)
3574 {
3575         wd->wd_timer = 0;
3576         callout_stop(&wd->wd_callout);
3577 }
3578
3579 void
3580 ifnet_lock(void)
3581 {
3582         KASSERT(curthread->td_type != TD_TYPE_NETISR,
3583             ("try holding ifnet lock in netisr"));
3584         mtx_lock(&ifnet_mtx);
3585 }
3586
3587 void
3588 ifnet_unlock(void)
3589 {
3590         KASSERT(curthread->td_type != TD_TYPE_NETISR,
3591             ("try holding ifnet lock in netisr"));
3592         mtx_unlock(&ifnet_mtx);
3593 }
3594
3595 static struct ifnet_array *
3596 ifnet_array_alloc(int count)
3597 {
3598         struct ifnet_array *arr;
3599
3600         arr = kmalloc(__offsetof(struct ifnet_array, ifnet_arr[count]),
3601             M_IFNET, M_WAITOK);
3602         arr->ifnet_count = count;
3603
3604         return arr;
3605 }
3606
3607 static void
3608 ifnet_array_free(struct ifnet_array *arr)
3609 {
3610         if (arr == &ifnet_array0)
3611                 return;
3612         kfree(arr, M_IFNET);
3613 }
3614
3615 static struct ifnet_array *
3616 ifnet_array_add(struct ifnet *ifp, const struct ifnet_array *old_arr)
3617 {
3618         struct ifnet_array *arr;
3619         int count, i;
3620
3621         KASSERT(old_arr->ifnet_count >= 0,
3622             ("invalid ifnet array count %d", old_arr->ifnet_count));
3623         count = old_arr->ifnet_count + 1;
3624         arr = ifnet_array_alloc(count);
3625
3626         /*
3627          * Save the old ifnet array and append this ifp to the end of
3628          * the new ifnet array.
3629          */
3630         for (i = 0; i < old_arr->ifnet_count; ++i) {
3631                 KASSERT(old_arr->ifnet_arr[i] != ifp,
3632                     ("%s is already in ifnet array", ifp->if_xname));
3633                 arr->ifnet_arr[i] = old_arr->ifnet_arr[i];
3634         }
3635         KASSERT(i == count - 1,
3636             ("add %s, ifnet array index mismatch, should be %d, but got %d",
3637              ifp->if_xname, count - 1, i));
3638         arr->ifnet_arr[i] = ifp;
3639
3640         return arr;
3641 }
3642
3643 static struct ifnet_array *
3644 ifnet_array_del(struct ifnet *ifp, const struct ifnet_array *old_arr)
3645 {
3646         struct ifnet_array *arr;
3647         int count, i, idx, found = 0;
3648
3649         KASSERT(old_arr->ifnet_count > 0,
3650             ("invalid ifnet array count %d", old_arr->ifnet_count));
3651         count = old_arr->ifnet_count - 1;
3652         arr = ifnet_array_alloc(count);
3653
3654         /*
3655          * Save the old ifnet array, but skip this ifp.
3656          */
3657         idx = 0;
3658         for (i = 0; i < old_arr->ifnet_count; ++i) {
3659                 if (old_arr->ifnet_arr[i] == ifp) {
3660                         KASSERT(!found,
3661                             ("dup %s is in ifnet array", ifp->if_xname));
3662                         found = 1;
3663                         continue;
3664                 }
3665                 KASSERT(idx < count,
3666                     ("invalid ifnet array index %d, count %d", idx, count));
3667                 arr->ifnet_arr[idx] = old_arr->ifnet_arr[i];
3668                 ++idx;
3669         }
3670         KASSERT(found, ("%s is not in ifnet array", ifp->if_xname));
3671         KASSERT(idx == count,
3672             ("del %s, ifnet array count mismatch, should be %d, but got %d ",
3673              ifp->if_xname, count, idx));
3674
3675         return arr;
3676 }
3677
3678 const struct ifnet_array *
3679 ifnet_array_get(void)
3680 {
3681         const struct ifnet_array *ret;
3682
3683         KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
3684         ret = ifnet_array;
3685         /* Make sure 'ret' is really used. */
3686         cpu_ccfence();
3687         return (ret);
3688 }
3689
3690 int
3691 ifnet_array_isempty(void)
3692 {
3693         KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
3694         if (ifnet_array->ifnet_count == 0)
3695                 return 1;
3696         else
3697                 return 0;
3698 }
3699
3700 void
3701 ifa_marker_init(struct ifaddr_marker *mark, struct ifnet *ifp)
3702 {
3703         struct ifaddr *ifa;
3704
3705         memset(mark, 0, sizeof(*mark));
3706         ifa = &mark->ifa;
3707
3708         mark->ifac.ifa = ifa;
3709
3710         ifa->ifa_addr = &mark->addr;
3711         ifa->ifa_dstaddr = &mark->dstaddr;
3712         ifa->ifa_netmask = &mark->netmask;
3713         ifa->ifa_ifp = ifp;
3714 }
3715
3716 static int
3717 if_ringcnt_fixup(int ring_cnt, int ring_cntmax)
3718 {
3719
3720         KASSERT(ring_cntmax > 0, ("invalid ring count max %d", ring_cntmax));
3721
3722         if (ring_cnt <= 0 || ring_cnt > ring_cntmax)
3723                 ring_cnt = ring_cntmax;
3724         if (ring_cnt > netisr_ncpus)
3725                 ring_cnt = netisr_ncpus;
3726         return (ring_cnt);
3727 }
3728
3729 static void
3730 if_ringmap_set_grid(device_t dev, struct if_ringmap *rm, int grid)
3731 {
3732         int i, offset;
3733
3734         KASSERT(grid > 0, ("invalid if_ringmap grid %d", grid));
3735         KASSERT(grid >= rm->rm_cnt, ("invalid if_ringmap grid %d, count %d",
3736             grid, rm->rm_cnt));
3737         rm->rm_grid = grid;
3738
3739         offset = (rm->rm_grid * device_get_unit(dev)) % netisr_ncpus;
3740         for (i = 0; i < rm->rm_cnt; ++i) {
3741                 rm->rm_cpumap[i] = offset + i;
3742                 KASSERT(rm->rm_cpumap[i] < netisr_ncpus,
3743                     ("invalid cpumap[%d] = %d, offset %d", i,
3744                      rm->rm_cpumap[i], offset));
3745         }
3746 }
3747
3748 static struct if_ringmap *
3749 if_ringmap_alloc_flags(device_t dev, int ring_cnt, int ring_cntmax,
3750     uint32_t flags)
3751 {
3752         struct if_ringmap *rm;
3753         int i, grid = 0, prev_grid;
3754
3755         ring_cnt = if_ringcnt_fixup(ring_cnt, ring_cntmax);
3756         rm = kmalloc(__offsetof(struct if_ringmap, rm_cpumap[ring_cnt]),
3757             M_DEVBUF, M_WAITOK | M_ZERO);
3758
3759         rm->rm_cnt = ring_cnt;
3760         if (flags & RINGMAP_FLAG_POWEROF2)
3761                 rm->rm_cnt = 1 << (fls(rm->rm_cnt) - 1);
3762
3763         prev_grid = netisr_ncpus;
3764         for (i = 0; i < netisr_ncpus; ++i) {
3765                 if (netisr_ncpus % (i + 1) != 0)
3766                         continue;
3767
3768                 grid = netisr_ncpus / (i + 1);
3769                 if (rm->rm_cnt > grid) {
3770                         grid = prev_grid;
3771                         break;
3772                 }
3773
3774                 if (rm->rm_cnt > netisr_ncpus / (i + 2))
3775                         break;
3776                 prev_grid = grid;
3777         }
3778         if_ringmap_set_grid(dev, rm, grid);
3779
3780         return (rm);
3781 }
3782
3783 struct if_ringmap *
3784 if_ringmap_alloc(device_t dev, int ring_cnt, int ring_cntmax)
3785 {
3786
3787         return (if_ringmap_alloc_flags(dev, ring_cnt, ring_cntmax,
3788             RINGMAP_FLAG_NONE));
3789 }
3790
3791 struct if_ringmap *
3792 if_ringmap_alloc2(device_t dev, int ring_cnt, int ring_cntmax)
3793 {
3794
3795         return (if_ringmap_alloc_flags(dev, ring_cnt, ring_cntmax,
3796             RINGMAP_FLAG_POWEROF2));
3797 }
3798
3799 void
3800 if_ringmap_free(struct if_ringmap *rm)
3801 {
3802
3803         kfree(rm, M_DEVBUF);
3804 }
3805
3806 /*
3807  * Align the two ringmaps.
3808  *
3809  * e.g. 8 netisrs, rm0 contains 4 rings, rm1 contains 2 rings.
3810  *
3811  * Before:
3812  *
3813  * CPU      0  1  2  3   4  5  6  7
3814  * NIC_RX               n0 n1 n2 n3
3815  * NIC_TX        N0 N1
3816  *
3817  * After:
3818  *
3819  * CPU      0  1  2  3   4  5  6  7
3820  * NIC_RX               n0 n1 n2 n3
3821  * NIC_TX               N0 N1
3822  */
3823 void
3824 if_ringmap_align(device_t dev, struct if_ringmap *rm0, struct if_ringmap *rm1)
3825 {
3826
3827         if (rm0->rm_grid > rm1->rm_grid)
3828                 if_ringmap_set_grid(dev, rm1, rm0->rm_grid);
3829         else if (rm0->rm_grid < rm1->rm_grid)
3830                 if_ringmap_set_grid(dev, rm0, rm1->rm_grid);
3831 }
3832
3833 void
3834 if_ringmap_match(device_t dev, struct if_ringmap *rm0, struct if_ringmap *rm1)
3835 {
3836         int subset_grid, cnt, divisor, mod, offset, i;
3837         struct if_ringmap *subset_rm, *rm;
3838         int old_rm0_grid, old_rm1_grid;
3839
3840         if (rm0->rm_grid == rm1->rm_grid)
3841                 return;
3842
3843         /* Save grid for later use */
3844         old_rm0_grid = rm0->rm_grid;
3845         old_rm1_grid = rm1->rm_grid;
3846
3847         if_ringmap_align(dev, rm0, rm1);
3848
3849         /*
3850          * Re-shuffle rings to get more even distribution.
3851          *
3852          * e.g. 12 netisrs, rm0 contains 4 rings, rm1 contains 2 rings.
3853          *
3854          * CPU       0  1  2  3   4  5  6  7   8  9 10 11
3855          *
3856          * NIC_RX   a0 a1 a2 a3  b0 b1 b2 b3  c0 c1 c2 c3
3857          * NIC_TX   A0 A1        B0 B1        C0 C1
3858          *
3859          * NIC_RX   d0 d1 d2 d3  e0 e1 e2 e3  f0 f1 f2 f3
3860          * NIC_TX         D0 D1        E0 E1        F0 F1
3861          */
3862
3863         if (rm0->rm_cnt >= (2 * old_rm1_grid)) {
3864                 cnt = rm0->rm_cnt;
3865                 subset_grid = old_rm1_grid;
3866                 subset_rm = rm1;
3867                 rm = rm0;
3868         } else if (rm1->rm_cnt > (2 * old_rm0_grid)) {
3869                 cnt = rm1->rm_cnt;
3870                 subset_grid = old_rm0_grid;
3871                 subset_rm = rm0;
3872                 rm = rm1;
3873         } else {
3874                 /* No space to shuffle. */
3875                 return;
3876         }
3877
3878         mod = cnt / subset_grid;
3879         KKASSERT(mod >= 2);
3880         divisor = netisr_ncpus / rm->rm_grid;
3881         offset = ((device_get_unit(dev) / divisor) % mod) * subset_grid;
3882
3883         for (i = 0; i < subset_rm->rm_cnt; ++i) {
3884                 subset_rm->rm_cpumap[i] += offset;
3885                 KASSERT(subset_rm->rm_cpumap[i] < netisr_ncpus,
3886                     ("match: invalid cpumap[%d] = %d, offset %d",
3887                      i, subset_rm->rm_cpumap[i], offset));
3888         }
3889 #ifdef INVARIANTS
3890         for (i = 0; i < subset_rm->rm_cnt; ++i) {
3891                 int j;
3892
3893                 for (j = 0; j < rm->rm_cnt; ++j) {
3894                         if (rm->rm_cpumap[j] == subset_rm->rm_cpumap[i])
3895                                 break;
3896                 }
3897                 KASSERT(j < rm->rm_cnt,
3898                     ("subset cpumap[%d] = %d not found in superset",
3899                      i, subset_rm->rm_cpumap[i]));
3900         }
3901 #endif
3902 }
3903
3904 int
3905 if_ringmap_count(const struct if_ringmap *rm)
3906 {
3907
3908         return (rm->rm_cnt);
3909 }
3910
3911 int
3912 if_ringmap_cpumap(const struct if_ringmap *rm, int ring)
3913 {
3914
3915         KASSERT(ring >= 0 && ring < rm->rm_cnt, ("invalid ring %d", ring));
3916         return (rm->rm_cpumap[ring]);
3917 }
3918
3919 void
3920 if_ringmap_rdrtable(const struct if_ringmap *rm, int table[], int table_nent)
3921 {
3922         int i, grid_idx, grid_cnt, patch_off, patch_cnt, ncopy;
3923
3924         KASSERT(table_nent > 0 && (table_nent & NETISR_CPUMASK) == 0,
3925             ("invalid redirect table entries %d", table_nent));
3926
3927         grid_idx = 0;
3928         for (i = 0; i < NETISR_CPUMAX; ++i) {
3929                 table[i] = grid_idx++ % rm->rm_cnt;
3930
3931                 if (grid_idx == rm->rm_grid)
3932                         grid_idx = 0;
3933         }
3934
3935         /*
3936          * Make the ring distributed more evenly for the remainder
3937          * of each grid.
3938          *
3939          * e.g. 12 netisrs, rm contains 8 rings.
3940          *
3941          * Redirect table before:
3942          *
3943          *  0  1  2  3  4  5  6  7  0  1  2  3  0  1  2  3
3944          *  4  5  6  7  0  1  2  3  0  1  2  3  4  5  6  7
3945          *  0  1  2  3  0  1  2  3  4  5  6  7  0  1  2  3
3946          *  ....
3947          *
3948          * Redirect table after being patched (pX, patched entries):
3949          *
3950          *  0  1  2  3  4  5  6  7 p0 p1 p2 p3  0  1  2  3
3951          *  4  5  6  7 p4 p5 p6 p7  0  1  2  3  4  5  6  7
3952          * p0 p1 p2 p3  0  1  2  3  4  5  6  7 p4 p5 p6 p7
3953          *  ....
3954          */
3955         patch_cnt = rm->rm_grid % rm->rm_cnt;
3956         if (patch_cnt == 0)
3957                 goto done;
3958         patch_off = rm->rm_grid - (rm->rm_grid % rm->rm_cnt);
3959
3960         grid_cnt = roundup(NETISR_CPUMAX, rm->rm_grid) / rm->rm_grid;
3961         grid_idx = 0;
3962         for (i = 0; i < grid_cnt; ++i) {
3963                 int j;
3964
3965                 for (j = 0; j < patch_cnt; ++j) {
3966                         int fix_idx;
3967
3968                         fix_idx = (i * rm->rm_grid) + patch_off + j;
3969                         if (fix_idx >= NETISR_CPUMAX)
3970                                 goto done;
3971                         table[fix_idx] = grid_idx++ % rm->rm_cnt;
3972                 }
3973         }
3974 done:
3975         /*
3976          * If the device supports larger redirect table, duplicate
3977          * the first NETISR_CPUMAX entries to the rest of the table,
3978          * so that it matches upper layer's expectation:
3979          * (hash & NETISR_CPUMASK) % netisr_ncpus
3980          */
3981         ncopy = table_nent / NETISR_CPUMAX;
3982         for (i = 1; i < ncopy; ++i) {
3983                 memcpy(&table[i * NETISR_CPUMAX], table,
3984                     NETISR_CPUMAX * sizeof(table[0]));
3985         }
3986         if (if_ringmap_dumprdr) {
3987                 for (i = 0; i < table_nent; ++i) {
3988                         if (i != 0 && i % 16 == 0)
3989                                 kprintf("\n");
3990                         kprintf("%03d ", table[i]);
3991                 }
3992                 kprintf("\n");
3993         }
3994 }
3995
3996 int
3997 if_ringmap_cpumap_sysctl(SYSCTL_HANDLER_ARGS)
3998 {
3999         struct if_ringmap *rm = arg1;
4000         int i, error = 0;
4001
4002         for (i = 0; i < rm->rm_cnt; ++i) {
4003                 int cpu = rm->rm_cpumap[i];
4004
4005                 error = SYSCTL_OUT(req, &cpu, sizeof(cpu));
4006                 if (error)
4007                         break;
4008         }
4009         return (error);
4010 }