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                         return (EINVAL);
1275                 }
1276
1277                 strlcpy(p->ifgrq_group, ifgl->ifgl_group->ifg_group,
1278                         sizeof(ifgrq->ifgrq_group));
1279                 len -= sizeof(struct ifg_req);
1280                 p++;
1281         }
1282         ifgroup_lockmgr(LK_RELEASE);
1283
1284         error = copyout(ifgrq, ifgr->ifgr_groups, ifgr->ifgr_len);
1285         kfree(ifgrq, M_TEMP);
1286         if (error)
1287                 return (error);
1288
1289         return (0);
1290 }
1291
1292 /*
1293  * Store all the members of a group in memory pointed to by data.
1294  */
1295 static int
1296 if_getgroupmembers(struct ifgroupreq *ifgr)
1297 {
1298         struct ifg_group *ifg;
1299         struct ifg_member *ifgm;
1300         struct ifg_req *ifgrq, *p;
1301         int len, error;
1302
1303         ifgroup_lockmgr(LK_SHARED);
1304
1305         TAILQ_FOREACH(ifg, &ifg_head, ifg_next) {
1306                 if (strcmp(ifg->ifg_group, ifgr->ifgr_name) == 0)
1307                         break;
1308         }
1309         if (ifg == NULL) {
1310                 ifgroup_lockmgr(LK_RELEASE);
1311                 return (ENOENT);
1312         }
1313
1314         len = 0;
1315         TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
1316                 len += sizeof(struct ifg_req);
1317
1318         ifgroup_lockmgr(LK_RELEASE);
1319
1320         if (ifgr->ifgr_len == 0) {
1321                 ifgr->ifgr_len = len;
1322                 return (0);
1323         } else if (ifgr->ifgr_len != len) {
1324                 return (EINVAL);
1325         }
1326
1327         ifgrq = kmalloc(len, M_TEMP, M_INTWAIT | M_NULLOK | M_ZERO);
1328         if (ifgrq == NULL)
1329                 return (ENOMEM);
1330
1331         ifgroup_lockmgr(LK_SHARED);
1332         p = ifgrq;
1333         TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
1334                 if (len < sizeof(struct ifg_req)) {
1335                         ifgroup_lockmgr(LK_RELEASE);
1336                         return (EINVAL);
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         kfree(ifgrq, M_TEMP);
1348         if (error)
1349                 return (error);
1350
1351         return (0);
1352 }
1353
1354 /*
1355  * Delete Routes for a Network Interface
1356  *
1357  * Called for each routing entry via the rnh->rnh_walktree() call above
1358  * to delete all route entries referencing a detaching network interface.
1359  *
1360  * Arguments:
1361  *      rn      pointer to node in the routing table
1362  *      arg     argument passed to rnh->rnh_walktree() - detaching interface
1363  *
1364  * Returns:
1365  *      0       successful
1366  *      errno   failed - reason indicated
1367  *
1368  */
1369 static int
1370 if_rtdel(struct radix_node *rn, void *arg)
1371 {
1372         struct rtentry  *rt = (struct rtentry *)rn;
1373         struct ifnet    *ifp = arg;
1374         int             err;
1375
1376         if (rt->rt_ifp == ifp) {
1377
1378                 /*
1379                  * Protect (sorta) against walktree recursion problems
1380                  * with cloned routes
1381                  */
1382                 if (!(rt->rt_flags & RTF_UP))
1383                         return (0);
1384
1385                 err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1386                                 rt_mask(rt), rt->rt_flags,
1387                                 NULL);
1388                 if (err) {
1389                         log(LOG_WARNING, "if_rtdel: error %d\n", err);
1390                 }
1391         }
1392
1393         return (0);
1394 }
1395
1396 static __inline boolean_t
1397 ifa_prefer(const struct ifaddr *cur_ifa, const struct ifaddr *old_ifa)
1398 {
1399         if (old_ifa == NULL)
1400                 return TRUE;
1401
1402         if ((old_ifa->ifa_ifp->if_flags & IFF_UP) == 0 &&
1403             (cur_ifa->ifa_ifp->if_flags & IFF_UP))
1404                 return TRUE;
1405         if ((old_ifa->ifa_flags & IFA_ROUTE) == 0 &&
1406             (cur_ifa->ifa_flags & IFA_ROUTE))
1407                 return TRUE;
1408         return FALSE;
1409 }
1410
1411 /*
1412  * Locate an interface based on a complete address.
1413  */
1414 struct ifaddr *
1415 ifa_ifwithaddr(struct sockaddr *addr)
1416 {
1417         const struct ifnet_array *arr;
1418         int i;
1419
1420         arr = ifnet_array_get();
1421         for (i = 0; i < arr->ifnet_count; ++i) {
1422                 struct ifnet *ifp = arr->ifnet_arr[i];
1423                 struct ifaddr_container *ifac;
1424
1425                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1426                         struct ifaddr *ifa = ifac->ifa;
1427
1428                         if (ifa->ifa_addr->sa_family != addr->sa_family)
1429                                 continue;
1430                         if (sa_equal(addr, ifa->ifa_addr))
1431                                 return (ifa);
1432                         if ((ifp->if_flags & IFF_BROADCAST) &&
1433                             ifa->ifa_broadaddr &&
1434                             /* IPv6 doesn't have broadcast */
1435                             ifa->ifa_broadaddr->sa_len != 0 &&
1436                             sa_equal(ifa->ifa_broadaddr, addr))
1437                                 return (ifa);
1438                 }
1439         }
1440         return (NULL);
1441 }
1442
1443 /*
1444  * Locate the point to point interface with a given destination address.
1445  */
1446 struct ifaddr *
1447 ifa_ifwithdstaddr(struct sockaddr *addr)
1448 {
1449         const struct ifnet_array *arr;
1450         int i;
1451
1452         arr = ifnet_array_get();
1453         for (i = 0; i < arr->ifnet_count; ++i) {
1454                 struct ifnet *ifp = arr->ifnet_arr[i];
1455                 struct ifaddr_container *ifac;
1456
1457                 if (!(ifp->if_flags & IFF_POINTOPOINT))
1458                         continue;
1459
1460                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1461                         struct ifaddr *ifa = ifac->ifa;
1462
1463                         if (ifa->ifa_addr->sa_family != addr->sa_family)
1464                                 continue;
1465                         if (ifa->ifa_dstaddr &&
1466                             sa_equal(addr, ifa->ifa_dstaddr))
1467                                 return (ifa);
1468                 }
1469         }
1470         return (NULL);
1471 }
1472
1473 /*
1474  * Find an interface on a specific network.  If many, choice
1475  * is most specific found.
1476  */
1477 struct ifaddr *
1478 ifa_ifwithnet(struct sockaddr *addr)
1479 {
1480         struct ifaddr *ifa_maybe = NULL;
1481         u_int af = addr->sa_family;
1482         char *addr_data = addr->sa_data, *cplim;
1483         const struct ifnet_array *arr;
1484         int i;
1485
1486         /*
1487          * AF_LINK addresses can be looked up directly by their index number,
1488          * so do that if we can.
1489          */
1490         if (af == AF_LINK) {
1491                 struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
1492
1493                 if (sdl->sdl_index && sdl->sdl_index <= if_index)
1494                         return (ifindex2ifnet[sdl->sdl_index]->if_lladdr);
1495         }
1496
1497         /*
1498          * Scan though each interface, looking for ones that have
1499          * addresses in this address family.
1500          */
1501         arr = ifnet_array_get();
1502         for (i = 0; i < arr->ifnet_count; ++i) {
1503                 struct ifnet *ifp = arr->ifnet_arr[i];
1504                 struct ifaddr_container *ifac;
1505
1506                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1507                         struct ifaddr *ifa = ifac->ifa;
1508                         char *cp, *cp2, *cp3;
1509
1510                         if (ifa->ifa_addr->sa_family != af)
1511 next:                           continue;
1512                         if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT) {
1513                                 /*
1514                                  * This is a bit broken as it doesn't
1515                                  * take into account that the remote end may
1516                                  * be a single node in the network we are
1517                                  * looking for.
1518                                  * The trouble is that we don't know the
1519                                  * netmask for the remote end.
1520                                  */
1521                                 if (ifa->ifa_dstaddr != NULL &&
1522                                     sa_equal(addr, ifa->ifa_dstaddr))
1523                                         return (ifa);
1524                         } else {
1525                                 /*
1526                                  * if we have a special address handler,
1527                                  * then use it instead of the generic one.
1528                                  */
1529                                 if (ifa->ifa_claim_addr) {
1530                                         if ((*ifa->ifa_claim_addr)(ifa, addr)) {
1531                                                 return (ifa);
1532                                         } else {
1533                                                 continue;
1534                                         }
1535                                 }
1536
1537                                 /*
1538                                  * Scan all the bits in the ifa's address.
1539                                  * If a bit dissagrees with what we are
1540                                  * looking for, mask it with the netmask
1541                                  * to see if it really matters.
1542                                  * (A byte at a time)
1543                                  */
1544                                 if (ifa->ifa_netmask == 0)
1545                                         continue;
1546                                 cp = addr_data;
1547                                 cp2 = ifa->ifa_addr->sa_data;
1548                                 cp3 = ifa->ifa_netmask->sa_data;
1549                                 cplim = ifa->ifa_netmask->sa_len +
1550                                         (char *)ifa->ifa_netmask;
1551                                 while (cp3 < cplim)
1552                                         if ((*cp++ ^ *cp2++) & *cp3++)
1553                                                 goto next; /* next address! */
1554                                 /*
1555                                  * If the netmask of what we just found
1556                                  * is more specific than what we had before
1557                                  * (if we had one) then remember the new one
1558                                  * before continuing to search for an even
1559                                  * better one.  If the netmasks are equal,
1560                                  * we prefer the this ifa based on the result
1561                                  * of ifa_prefer().
1562                                  */
1563                                 if (ifa_maybe == NULL ||
1564                                     rn_refines((char *)ifa->ifa_netmask,
1565                                         (char *)ifa_maybe->ifa_netmask) ||
1566                                     (sa_equal(ifa_maybe->ifa_netmask,
1567                                         ifa->ifa_netmask) &&
1568                                      ifa_prefer(ifa, ifa_maybe)))
1569                                         ifa_maybe = ifa;
1570                         }
1571                 }
1572         }
1573         return (ifa_maybe);
1574 }
1575
1576 /*
1577  * Find an interface address specific to an interface best matching
1578  * a given address.
1579  */
1580 struct ifaddr *
1581 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1582 {
1583         struct ifaddr_container *ifac;
1584         char *cp, *cp2, *cp3;
1585         char *cplim;
1586         struct ifaddr *ifa_maybe = NULL;
1587         u_int af = addr->sa_family;
1588
1589         if (af >= AF_MAX)
1590                 return (0);
1591         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1592                 struct ifaddr *ifa = ifac->ifa;
1593
1594                 if (ifa->ifa_addr->sa_family != af)
1595                         continue;
1596                 if (ifa_maybe == NULL)
1597                         ifa_maybe = ifa;
1598                 if (ifa->ifa_netmask == NULL) {
1599                         if (sa_equal(addr, ifa->ifa_addr) ||
1600                             (ifa->ifa_dstaddr != NULL &&
1601                              sa_equal(addr, ifa->ifa_dstaddr)))
1602                                 return (ifa);
1603                         continue;
1604                 }
1605                 if (ifp->if_flags & IFF_POINTOPOINT) {
1606                         if (sa_equal(addr, ifa->ifa_dstaddr))
1607                                 return (ifa);
1608                 } else {
1609                         cp = addr->sa_data;
1610                         cp2 = ifa->ifa_addr->sa_data;
1611                         cp3 = ifa->ifa_netmask->sa_data;
1612                         cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1613                         for (; cp3 < cplim; cp3++)
1614                                 if ((*cp++ ^ *cp2++) & *cp3)
1615                                         break;
1616                         if (cp3 == cplim)
1617                                 return (ifa);
1618                 }
1619         }
1620         return (ifa_maybe);
1621 }
1622
1623 /*
1624  * Default action when installing a route with a Link Level gateway.
1625  * Lookup an appropriate real ifa to point to.
1626  * This should be moved to /sys/net/link.c eventually.
1627  */
1628 static void
1629 link_rtrequest(int cmd, struct rtentry *rt)
1630 {
1631         struct ifaddr *ifa;
1632         struct sockaddr *dst;
1633         struct ifnet *ifp;
1634
1635         if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL ||
1636             (ifp = ifa->ifa_ifp) == NULL || (dst = rt_key(rt)) == NULL)
1637                 return;
1638         ifa = ifaof_ifpforaddr(dst, ifp);
1639         if (ifa != NULL) {
1640                 IFAFREE(rt->rt_ifa);
1641                 IFAREF(ifa);
1642                 rt->rt_ifa = ifa;
1643                 if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1644                         ifa->ifa_rtrequest(cmd, rt);
1645         }
1646 }
1647
1648 struct netmsg_ifroute {
1649         struct netmsg_base      base;
1650         struct ifnet            *ifp;
1651         int                     flag;
1652         int                     fam;
1653 };
1654
1655 /*
1656  * Mark an interface down and notify protocols of the transition.
1657  */
1658 static void
1659 if_unroute_dispatch(netmsg_t nmsg)
1660 {
1661         struct netmsg_ifroute *msg = (struct netmsg_ifroute *)nmsg;
1662         struct ifnet *ifp = msg->ifp;
1663         int flag = msg->flag, fam = msg->fam;
1664         struct ifaddr_container *ifac;
1665         struct domain *dp;
1666
1667         ASSERT_NETISR0;
1668
1669         ifp->if_flags &= ~flag;
1670         getmicrotime(&ifp->if_lastchange);
1671         rt_ifmsg(ifp);
1672
1673         /*
1674          * The ifaddr processing in the following loop will block,
1675          * however, this function is called in netisr0, in which
1676          * ifaddr list changes happen, so we don't care about the
1677          * blockness of the ifaddr processing here.
1678          */
1679         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1680                 struct ifaddr *ifa = ifac->ifa;
1681
1682                 /* Ignore marker */
1683                 if (ifa->ifa_addr->sa_family == AF_UNSPEC)
1684                         continue;
1685
1686                 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1687                         kpfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1688         }
1689
1690         SLIST_FOREACH(dp, &domains, dom_next)
1691                 if (dp->dom_if_down != NULL)
1692                         dp->dom_if_down(ifp);
1693
1694         ifq_purge_all(&ifp->if_snd);
1695         netisr_replymsg(&nmsg->base, 0);
1696 }
1697
1698 static void
1699 if_unroute(struct ifnet *ifp, int flag, int fam)
1700 {
1701         struct netmsg_ifroute msg;
1702
1703         netmsg_init(&msg.base, NULL, &curthread->td_msgport, 0,
1704             if_unroute_dispatch);
1705         msg.ifp = ifp;
1706         msg.flag = flag;
1707         msg.fam = fam;
1708         netisr_domsg(&msg.base, 0);
1709 }
1710
1711 /*
1712  * Mark an interface up and notify protocols of the transition.
1713  */
1714 static void
1715 if_route_dispatch(netmsg_t nmsg)
1716 {
1717         struct netmsg_ifroute *msg = (struct netmsg_ifroute *)nmsg;
1718         struct ifnet *ifp = msg->ifp;
1719         int flag = msg->flag, fam = msg->fam;
1720         struct ifaddr_container *ifac;
1721         struct domain *dp;
1722
1723         ASSERT_NETISR0;
1724
1725         ifq_purge_all(&ifp->if_snd);
1726         ifp->if_flags |= flag;
1727         getmicrotime(&ifp->if_lastchange);
1728         rt_ifmsg(ifp);
1729
1730         /*
1731          * The ifaddr processing in the following loop will block,
1732          * however, this function is called in netisr0, in which
1733          * ifaddr list changes happen, so we don't care about the
1734          * blockness of the ifaddr processing here.
1735          */
1736         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1737                 struct ifaddr *ifa = ifac->ifa;
1738
1739                 /* Ignore marker */
1740                 if (ifa->ifa_addr->sa_family == AF_UNSPEC)
1741                         continue;
1742
1743                 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1744                         kpfctlinput(PRC_IFUP, ifa->ifa_addr);
1745         }
1746
1747         SLIST_FOREACH(dp, &domains, dom_next)
1748                 if (dp->dom_if_up != NULL)
1749                         dp->dom_if_up(ifp);
1750
1751         netisr_replymsg(&nmsg->base, 0);
1752 }
1753
1754 static void
1755 if_route(struct ifnet *ifp, int flag, int fam)
1756 {
1757         struct netmsg_ifroute msg;
1758
1759         netmsg_init(&msg.base, NULL, &curthread->td_msgport, 0,
1760             if_route_dispatch);
1761         msg.ifp = ifp;
1762         msg.flag = flag;
1763         msg.fam = fam;
1764         netisr_domsg(&msg.base, 0);
1765 }
1766
1767 /*
1768  * Mark an interface down and notify protocols of the transition.  An
1769  * interface going down is also considered to be a synchronizing event.
1770  * We must ensure that all packet processing related to the interface
1771  * has completed before we return so e.g. the caller can free the ifnet
1772  * structure that the mbufs may be referencing.
1773  *
1774  * NOTE: must be called at splnet or eqivalent.
1775  */
1776 void
1777 if_down(struct ifnet *ifp)
1778 {
1779         EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_DOWN);
1780         if_unroute(ifp, IFF_UP, AF_UNSPEC);
1781         netmsg_service_sync();
1782 }
1783
1784 /*
1785  * Mark an interface up and notify protocols of
1786  * the transition.
1787  * NOTE: must be called at splnet or eqivalent.
1788  */
1789 void
1790 if_up(struct ifnet *ifp)
1791 {
1792         if_route(ifp, IFF_UP, AF_UNSPEC);
1793         EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_UP);
1794 }
1795
1796 /*
1797  * Process a link state change.
1798  * NOTE: must be called at splsoftnet or equivalent.
1799  */
1800 void
1801 if_link_state_change(struct ifnet *ifp)
1802 {
1803         int link_state = ifp->if_link_state;
1804
1805         rt_ifmsg(ifp);
1806         devctl_notify("IFNET", ifp->if_xname,
1807             (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL);
1808
1809         EVENTHANDLER_INVOKE(ifnet_link_event, ifp, link_state);
1810 }
1811
1812 /*
1813  * Handle interface watchdog timer routines.  Called
1814  * from softclock, we decrement timers (if set) and
1815  * call the appropriate interface routine on expiration.
1816  */
1817 static void
1818 if_slowtimo_dispatch(netmsg_t nmsg)
1819 {
1820         struct globaldata *gd = mycpu;
1821         const struct ifnet_array *arr;
1822         int i;
1823
1824         ASSERT_NETISR0;
1825
1826         crit_enter_gd(gd);
1827         lwkt_replymsg(&nmsg->lmsg, 0);  /* reply ASAP */
1828         crit_exit_gd(gd);
1829
1830         arr = ifnet_array_get();
1831         for (i = 0; i < arr->ifnet_count; ++i) {
1832                 struct ifnet *ifp = arr->ifnet_arr[i];
1833
1834                 crit_enter_gd(gd);
1835
1836                 if (if_stats_compat) {
1837                         IFNET_STAT_GET(ifp, ipackets, ifp->if_ipackets);
1838                         IFNET_STAT_GET(ifp, ierrors, ifp->if_ierrors);
1839                         IFNET_STAT_GET(ifp, opackets, ifp->if_opackets);
1840                         IFNET_STAT_GET(ifp, oerrors, ifp->if_oerrors);
1841                         IFNET_STAT_GET(ifp, collisions, ifp->if_collisions);
1842                         IFNET_STAT_GET(ifp, ibytes, ifp->if_ibytes);
1843                         IFNET_STAT_GET(ifp, obytes, ifp->if_obytes);
1844                         IFNET_STAT_GET(ifp, imcasts, ifp->if_imcasts);
1845                         IFNET_STAT_GET(ifp, omcasts, ifp->if_omcasts);
1846                         IFNET_STAT_GET(ifp, iqdrops, ifp->if_iqdrops);
1847                         IFNET_STAT_GET(ifp, noproto, ifp->if_noproto);
1848                         IFNET_STAT_GET(ifp, oqdrops, ifp->if_oqdrops);
1849                 }
1850
1851                 if (ifp->if_timer == 0 || --ifp->if_timer) {
1852                         crit_exit_gd(gd);
1853                         continue;
1854                 }
1855                 if (ifp->if_watchdog) {
1856                         if (ifnet_tryserialize_all(ifp)) {
1857                                 (*ifp->if_watchdog)(ifp);
1858                                 ifnet_deserialize_all(ifp);
1859                         } else {
1860                                 /* try again next timeout */
1861                                 ++ifp->if_timer;
1862                         }
1863                 }
1864
1865                 crit_exit_gd(gd);
1866         }
1867
1868         callout_reset(&if_slowtimo_timer, hz / IFNET_SLOWHZ, if_slowtimo, NULL);
1869 }
1870
1871 static void
1872 if_slowtimo(void *arg __unused)
1873 {
1874         struct lwkt_msg *lmsg = &if_slowtimo_netmsg.lmsg;
1875
1876         KASSERT(mycpuid == 0, ("not on cpu0"));
1877         crit_enter();
1878         if (lmsg->ms_flags & MSGF_DONE)
1879                 lwkt_sendmsg_oncpu(netisr_cpuport(0), lmsg);
1880         crit_exit();
1881 }
1882
1883 /*
1884  * Map interface name to
1885  * interface structure pointer.
1886  */
1887 struct ifnet *
1888 ifunit(const char *name)
1889 {
1890         struct ifnet *ifp;
1891
1892         /*
1893          * Search all the interfaces for this name/number
1894          */
1895         KASSERT(mtx_owned(&ifnet_mtx), ("ifnet is not locked"));
1896
1897         TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
1898                 if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1899                         break;
1900         }
1901         return (ifp);
1902 }
1903
1904 struct ifnet *
1905 ifunit_netisr(const char *name)
1906 {
1907         const struct ifnet_array *arr;
1908         int i;
1909
1910         /*
1911          * Search all the interfaces for this name/number
1912          */
1913
1914         arr = ifnet_array_get();
1915         for (i = 0; i < arr->ifnet_count; ++i) {
1916                 struct ifnet *ifp = arr->ifnet_arr[i];
1917
1918                 if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1919                         return ifp;
1920         }
1921         return NULL;
1922 }
1923
1924 /*
1925  * Interface ioctls.
1926  */
1927 int
1928 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct ucred *cred)
1929 {
1930         struct ifnet *ifp;
1931         struct ifgroupreq *ifgr;
1932         struct ifreq *ifr;
1933         struct ifstat *ifs;
1934         int error, do_ifup = 0;
1935         short oif_flags;
1936         int new_flags;
1937         size_t namelen, onamelen;
1938         char new_name[IFNAMSIZ];
1939         struct ifaddr *ifa;
1940         struct sockaddr_dl *sdl;
1941
1942         switch (cmd) {
1943         case SIOCGIFCONF:
1944                 return (ifconf(cmd, data, cred));
1945         default:
1946                 break;
1947         }
1948
1949         ifr = (struct ifreq *)data;
1950
1951         switch (cmd) {
1952         case SIOCIFCREATE:
1953         case SIOCIFCREATE2:
1954                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1955                         return (error);
1956                 return (if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name),
1957                         cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL));
1958         case SIOCIFDESTROY:
1959                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1960                         return (error);
1961                 return (if_clone_destroy(ifr->ifr_name));
1962         case SIOCIFGCLONERS:
1963                 return (if_clone_list((struct if_clonereq *)data));
1964         case SIOCGIFGMEMB:
1965                 return (if_getgroupmembers((struct ifgroupreq *)data));
1966         default:
1967                 break;
1968         }
1969
1970         /*
1971          * Nominal ioctl through interface, lookup the ifp and obtain a
1972          * lock to serialize the ifconfig ioctl operation.
1973          */
1974         ifnet_lock();
1975
1976         ifp = ifunit(ifr->ifr_name);
1977         if (ifp == NULL) {
1978                 ifnet_unlock();
1979                 return (ENXIO);
1980         }
1981         error = 0;
1982
1983         switch (cmd) {
1984         case SIOCGIFINDEX:
1985                 ifr->ifr_index = ifp->if_index;
1986                 break;
1987
1988         case SIOCGIFFLAGS:
1989                 ifr->ifr_flags = ifp->if_flags;
1990                 ifr->ifr_flagshigh = ifp->if_flags >> 16;
1991                 break;
1992
1993         case SIOCGIFCAP:
1994                 ifr->ifr_reqcap = ifp->if_capabilities;
1995                 ifr->ifr_curcap = ifp->if_capenable;
1996                 break;
1997
1998         case SIOCGIFMETRIC:
1999                 ifr->ifr_metric = ifp->if_metric;
2000                 break;
2001
2002         case SIOCGIFMTU:
2003                 ifr->ifr_mtu = ifp->if_mtu;
2004                 break;
2005
2006         case SIOCGIFTSOLEN:
2007                 ifr->ifr_tsolen = ifp->if_tsolen;
2008                 break;
2009
2010         case SIOCGIFDATA:
2011                 error = copyout((caddr_t)&ifp->if_data, ifr->ifr_data,
2012                                 sizeof(ifp->if_data));
2013                 break;
2014
2015         case SIOCGIFPHYS:
2016                 ifr->ifr_phys = ifp->if_physical;
2017                 break;
2018
2019         case SIOCGIFPOLLCPU:
2020                 ifr->ifr_pollcpu = -1;
2021                 break;
2022
2023         case SIOCSIFPOLLCPU:
2024                 break;
2025
2026         case SIOCSIFFLAGS:
2027                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2028                 if (error)
2029                         break;
2030                 new_flags = (ifr->ifr_flags & 0xffff) |
2031                     (ifr->ifr_flagshigh << 16);
2032                 if (ifp->if_flags & IFF_SMART) {
2033                         /* Smart drivers twiddle their own routes */
2034                 } else if (ifp->if_flags & IFF_UP &&
2035                     (new_flags & IFF_UP) == 0) {
2036                         if_down(ifp);
2037                 } else if (new_flags & IFF_UP &&
2038                     (ifp->if_flags & IFF_UP) == 0) {
2039                         do_ifup = 1;
2040                 }
2041
2042 #ifdef IFPOLL_ENABLE
2043                 if ((new_flags ^ ifp->if_flags) & IFF_NPOLLING) {
2044                         if (new_flags & IFF_NPOLLING)
2045                                 ifpoll_register(ifp);
2046                         else
2047                                 ifpoll_deregister(ifp);
2048                 }
2049 #endif
2050
2051                 ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
2052                         (new_flags &~ IFF_CANTCHANGE);
2053                 if (new_flags & IFF_PPROMISC) {
2054                         /* Permanently promiscuous mode requested */
2055                         ifp->if_flags |= IFF_PROMISC;
2056                 } else if (ifp->if_pcount == 0) {
2057                         ifp->if_flags &= ~IFF_PROMISC;
2058                 }
2059                 if (ifp->if_ioctl) {
2060                         ifnet_serialize_all(ifp);
2061                         ifp->if_ioctl(ifp, cmd, data, cred);
2062                         ifnet_deserialize_all(ifp);
2063                 }
2064                 if (do_ifup)
2065                         if_up(ifp);
2066                 getmicrotime(&ifp->if_lastchange);
2067                 break;
2068
2069         case SIOCSIFCAP:
2070                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2071                 if (error)
2072                         break;
2073                 if (ifr->ifr_reqcap & ~ifp->if_capabilities) {
2074                         error = EINVAL;
2075                         break;
2076                 }
2077                 ifnet_serialize_all(ifp);
2078                 ifp->if_ioctl(ifp, cmd, data, cred);
2079                 ifnet_deserialize_all(ifp);
2080                 break;
2081
2082         case SIOCSIFNAME:
2083                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2084                 if (error)
2085                         break;
2086                 error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
2087                 if (error)
2088                         break;
2089                 if (new_name[0] == '\0') {
2090                         error = EINVAL;
2091                         break;
2092                 }
2093                 if (ifunit(new_name) != NULL) {
2094                         error = EEXIST;
2095                         break;
2096                 }
2097
2098                 EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
2099
2100                 /* Announce the departure of the interface. */
2101                 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
2102
2103                 strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
2104                 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
2105                 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
2106                 namelen = strlen(new_name);
2107                 onamelen = sdl->sdl_nlen;
2108                 /*
2109                  * Move the address if needed.  This is safe because we
2110                  * allocate space for a name of length IFNAMSIZ when we
2111                  * create this in if_attach().
2112                  */
2113                 if (namelen != onamelen) {
2114                         bcopy(sdl->sdl_data + onamelen,
2115                             sdl->sdl_data + namelen, sdl->sdl_alen);
2116                 }
2117                 bcopy(new_name, sdl->sdl_data, namelen);
2118                 sdl->sdl_nlen = namelen;
2119                 sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
2120                 bzero(sdl->sdl_data, onamelen);
2121                 while (namelen != 0)
2122                         sdl->sdl_data[--namelen] = 0xff;
2123
2124                 EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
2125
2126                 /* Announce the return of the interface. */
2127                 rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
2128                 break;
2129
2130         case SIOCSIFMETRIC:
2131                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2132                 if (error)
2133                         break;
2134                 ifp->if_metric = ifr->ifr_metric;
2135                 getmicrotime(&ifp->if_lastchange);
2136                 break;
2137
2138         case SIOCSIFPHYS:
2139                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2140                 if (error)
2141                         break;
2142                 if (ifp->if_ioctl == NULL) {
2143                         error = EOPNOTSUPP;
2144                         break;
2145                 }
2146                 ifnet_serialize_all(ifp);
2147                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2148                 ifnet_deserialize_all(ifp);
2149                 if (error == 0)
2150                         getmicrotime(&ifp->if_lastchange);
2151                 break;
2152
2153         case SIOCSIFMTU:
2154         {
2155                 u_long oldmtu = ifp->if_mtu;
2156
2157                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2158                 if (error)
2159                         break;
2160                 if (ifp->if_ioctl == NULL) {
2161                         error = EOPNOTSUPP;
2162                         break;
2163                 }
2164                 if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU) {
2165                         error = EINVAL;
2166                         break;
2167                 }
2168                 ifnet_serialize_all(ifp);
2169                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2170                 ifnet_deserialize_all(ifp);
2171                 if (error == 0) {
2172                         getmicrotime(&ifp->if_lastchange);
2173                         rt_ifmsg(ifp);
2174                 }
2175                 /*
2176                  * If the link MTU changed, do network layer specific procedure.
2177                  */
2178                 if (ifp->if_mtu != oldmtu) {
2179 #ifdef INET6
2180                         nd6_setmtu(ifp);
2181 #endif
2182                 }
2183                 break;
2184         }
2185
2186         case SIOCSIFTSOLEN:
2187                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2188                 if (error)
2189                         break;
2190
2191                 /* XXX need driver supplied upper limit */
2192                 if (ifr->ifr_tsolen <= 0) {
2193                         error = EINVAL;
2194                         break;
2195                 }
2196                 ifp->if_tsolen = ifr->ifr_tsolen;
2197                 break;
2198
2199         case SIOCADDMULTI:
2200         case SIOCDELMULTI:
2201                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2202                 if (error)
2203                         break;
2204
2205                 /* Don't allow group membership on non-multicast interfaces. */
2206                 if ((ifp->if_flags & IFF_MULTICAST) == 0) {
2207                         error = EOPNOTSUPP;
2208                         break;
2209                 }
2210
2211                 /* Don't let users screw up protocols' entries. */
2212                 if (ifr->ifr_addr.sa_family != AF_LINK) {
2213                         error = EINVAL;
2214                         break;
2215                 }
2216
2217                 if (cmd == SIOCADDMULTI) {
2218                         struct ifmultiaddr *ifma;
2219                         error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
2220                 } else {
2221                         error = if_delmulti(ifp, &ifr->ifr_addr);
2222                 }
2223                 if (error == 0)
2224                         getmicrotime(&ifp->if_lastchange);
2225                 break;
2226
2227         case SIOCSIFPHYADDR:
2228         case SIOCDIFPHYADDR:
2229 #ifdef INET6
2230         case SIOCSIFPHYADDR_IN6:
2231 #endif
2232         case SIOCSLIFPHYADDR:
2233         case SIOCSIFMEDIA:
2234         case SIOCSIFGENERIC:
2235                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2236                 if (error)
2237                         break;
2238                 if (ifp->if_ioctl == NULL) {
2239                         error = EOPNOTSUPP;
2240                         break;
2241                 }
2242                 ifnet_serialize_all(ifp);
2243                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2244                 ifnet_deserialize_all(ifp);
2245                 if (error == 0)
2246                         getmicrotime(&ifp->if_lastchange);
2247                 break;
2248
2249         case SIOCGIFSTATUS:
2250                 ifs = (struct ifstat *)data;
2251                 ifs->ascii[0] = '\0';
2252                 /* fall through */
2253         case SIOCGIFPSRCADDR:
2254         case SIOCGIFPDSTADDR:
2255         case SIOCGLIFPHYADDR:
2256         case SIOCGIFMEDIA:
2257         case SIOCGIFGENERIC:
2258                 if (ifp->if_ioctl == NULL) {
2259                         error = EOPNOTSUPP;
2260                         break;
2261                 }
2262                 ifnet_serialize_all(ifp);
2263                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2264                 ifnet_deserialize_all(ifp);
2265                 break;
2266
2267         case SIOCSIFLLADDR:
2268                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2269                 if (error)
2270                         break;
2271                 error = if_setlladdr(ifp, ifr->ifr_addr.sa_data,
2272                                      ifr->ifr_addr.sa_len);
2273                 EVENTHANDLER_INVOKE(iflladdr_event, ifp);
2274                 break;
2275
2276         case SIOCAIFGROUP:
2277                 ifgr = (struct ifgroupreq *)ifr;
2278                 if ((error = priv_check_cred(cred, PRIV_NET_ADDIFGROUP, 0)))
2279                         return (error);
2280                 if ((error = if_addgroup(ifp, ifgr->ifgr_group)))
2281                         return (error);
2282                 break;
2283
2284         case SIOCDIFGROUP:
2285                 ifgr = (struct ifgroupreq *)ifr;
2286                 if ((error = priv_check_cred(cred, PRIV_NET_DELIFGROUP, 0)))
2287                         return (error);
2288                 if ((error = if_delgroup(ifp, ifgr->ifgr_group)))
2289                         return (error);
2290                 break;
2291
2292         case SIOCGIFGROUP:
2293                 ifgr = (struct ifgroupreq *)ifr;
2294                 if ((error = if_getgroups(ifgr, ifp)))
2295                         return (error);
2296                 break;
2297
2298         default:
2299                 oif_flags = ifp->if_flags;
2300                 if (so->so_proto == 0) {
2301                         error = EOPNOTSUPP;
2302                         break;
2303                 }
2304                 error = so_pru_control_direct(so, cmd, data, ifp);
2305
2306                 /*
2307                  * If the socket control method returns EOPNOTSUPP, pass the
2308                  * request directly to the interface.
2309                  *
2310                  * Exclude the SIOCSIF{ADDR,BRDADDR,DSTADDR,NETMASK} ioctls,
2311                  * because drivers may trust these ioctls to come from an
2312                  * already privileged layer and thus do not perform credentials
2313                  * checks or input validation.
2314                  */
2315                 if (error == EOPNOTSUPP &&
2316                     ifp->if_ioctl != NULL &&
2317                     cmd != SIOCSIFADDR &&
2318                     cmd != SIOCSIFBRDADDR &&
2319                     cmd != SIOCSIFDSTADDR &&
2320                     cmd != SIOCSIFNETMASK) {
2321                         ifnet_serialize_all(ifp);
2322                         error = ifp->if_ioctl(ifp, cmd, data, cred);
2323                         ifnet_deserialize_all(ifp);
2324                 }
2325
2326                 if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
2327 #ifdef INET6
2328                         DELAY(100);/* XXX: temporary workaround for fxp issue*/
2329                         if (ifp->if_flags & IFF_UP) {
2330                                 crit_enter();
2331                                 in6_if_up(ifp);
2332                                 crit_exit();
2333                         }
2334 #endif
2335                 }
2336                 break;
2337         }
2338
2339         ifnet_unlock();
2340         return (error);
2341 }
2342
2343 /*
2344  * Set/clear promiscuous mode on interface ifp based on the truth value
2345  * of pswitch.  The calls are reference counted so that only the first
2346  * "on" request actually has an effect, as does the final "off" request.
2347  * Results are undefined if the "off" and "on" requests are not matched.
2348  */
2349 int
2350 ifpromisc(struct ifnet *ifp, int pswitch)
2351 {
2352         struct ifreq ifr;
2353         int error;
2354         int oldflags;
2355
2356         oldflags = ifp->if_flags;
2357         if (ifp->if_flags & IFF_PPROMISC) {
2358                 /* Do nothing if device is in permanently promiscuous mode */
2359                 ifp->if_pcount += pswitch ? 1 : -1;
2360                 return (0);
2361         }
2362         if (pswitch) {
2363                 /*
2364                  * If the device is not configured up, we cannot put it in
2365                  * promiscuous mode.
2366                  */
2367                 if ((ifp->if_flags & IFF_UP) == 0)
2368                         return (ENETDOWN);
2369                 if (ifp->if_pcount++ != 0)
2370                         return (0);
2371                 ifp->if_flags |= IFF_PROMISC;
2372                 log(LOG_INFO, "%s: promiscuous mode enabled\n",
2373                     ifp->if_xname);
2374         } else {
2375                 if (--ifp->if_pcount > 0)
2376                         return (0);
2377                 ifp->if_flags &= ~IFF_PROMISC;
2378                 log(LOG_INFO, "%s: promiscuous mode disabled\n",
2379                     ifp->if_xname);
2380         }
2381         ifr.ifr_flags = ifp->if_flags;
2382         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2383         ifnet_serialize_all(ifp);
2384         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr, NULL);
2385         ifnet_deserialize_all(ifp);
2386         if (error == 0)
2387                 rt_ifmsg(ifp);
2388         else
2389                 ifp->if_flags = oldflags;
2390         return error;
2391 }
2392
2393 /*
2394  * Return interface configuration
2395  * of system.  List may be used
2396  * in later ioctl's (above) to get
2397  * other information.
2398  */
2399 static int
2400 ifconf(u_long cmd, caddr_t data, struct ucred *cred)
2401 {
2402         struct ifconf *ifc = (struct ifconf *)data;
2403         struct ifnet *ifp;
2404         struct sockaddr *sa;
2405         struct ifreq ifr, *ifrp;
2406         int space = ifc->ifc_len, error = 0;
2407
2408         ifrp = ifc->ifc_req;
2409
2410         ifnet_lock();
2411         TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
2412                 struct ifaddr_container *ifac, *ifac_mark;
2413                 struct ifaddr_marker mark;
2414                 struct ifaddrhead *head;
2415                 int addrs;
2416
2417                 if (space <= sizeof ifr)
2418                         break;
2419
2420                 /*
2421                  * Zero the stack declared structure first to prevent
2422                  * memory disclosure.
2423                  */
2424                 bzero(&ifr, sizeof(ifr));
2425                 if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
2426                     >= sizeof(ifr.ifr_name)) {
2427                         error = ENAMETOOLONG;
2428                         break;
2429                 }
2430
2431                 /*
2432                  * Add a marker, since copyout() could block and during that
2433                  * period the list could be changed.  Inserting the marker to
2434                  * the header of the list will not cause trouble for the code
2435                  * assuming that the first element of the list is AF_LINK; the
2436                  * marker will be moved to the next position w/o blocking.
2437                  */
2438                 ifa_marker_init(&mark, ifp);
2439                 ifac_mark = &mark.ifac;
2440                 head = &ifp->if_addrheads[mycpuid];
2441
2442                 addrs = 0;
2443                 TAILQ_INSERT_HEAD(head, ifac_mark, ifa_link);
2444                 while ((ifac = TAILQ_NEXT(ifac_mark, ifa_link)) != NULL) {
2445                         struct ifaddr *ifa = ifac->ifa;
2446
2447                         TAILQ_REMOVE(head, ifac_mark, ifa_link);
2448                         TAILQ_INSERT_AFTER(head, ifac, ifac_mark, ifa_link);
2449
2450                         /* Ignore marker */
2451                         if (ifa->ifa_addr->sa_family == AF_UNSPEC)
2452                                 continue;
2453
2454                         if (space <= sizeof ifr)
2455                                 break;
2456                         sa = ifa->ifa_addr;
2457                         if (cred->cr_prison &&
2458                             prison_if(cred, sa))
2459                                 continue;
2460                         addrs++;
2461                         /*
2462                          * Keep a reference on this ifaddr, so that it will
2463                          * not be destroyed when its address is copied to
2464                          * the userland, which could block.
2465                          */
2466                         IFAREF(ifa);
2467                         if (sa->sa_len <= sizeof(*sa)) {
2468                                 ifr.ifr_addr = *sa;
2469                                 error = copyout(&ifr, ifrp, sizeof ifr);
2470                                 ifrp++;
2471                         } else {
2472                                 if (space < (sizeof ifr) + sa->sa_len -
2473                                             sizeof(*sa)) {
2474                                         IFAFREE(ifa);
2475                                         break;
2476                                 }
2477                                 space -= sa->sa_len - sizeof(*sa);
2478                                 error = copyout(&ifr, ifrp,
2479                                                 sizeof ifr.ifr_name);
2480                                 if (error == 0)
2481                                         error = copyout(sa, &ifrp->ifr_addr,
2482                                                         sa->sa_len);
2483                                 ifrp = (struct ifreq *)
2484                                         (sa->sa_len + (caddr_t)&ifrp->ifr_addr);
2485                         }
2486                         IFAFREE(ifa);
2487                         if (error)
2488                                 break;
2489                         space -= sizeof ifr;
2490                 }
2491                 TAILQ_REMOVE(head, ifac_mark, ifa_link);
2492                 if (error)
2493                         break;
2494                 if (!addrs) {
2495                         bzero(&ifr.ifr_addr, sizeof ifr.ifr_addr);
2496                         error = copyout(&ifr, ifrp, sizeof ifr);
2497                         if (error)
2498                                 break;
2499                         space -= sizeof ifr;
2500                         ifrp++;
2501                 }
2502         }
2503         ifnet_unlock();
2504
2505         ifc->ifc_len -= space;
2506         return (error);
2507 }
2508
2509 /*
2510  * Just like if_promisc(), but for all-multicast-reception mode.
2511  */
2512 int
2513 if_allmulti(struct ifnet *ifp, int onswitch)
2514 {
2515         int error = 0;
2516         struct ifreq ifr;
2517
2518         crit_enter();
2519
2520         if (onswitch) {
2521                 if (ifp->if_amcount++ == 0) {
2522                         ifp->if_flags |= IFF_ALLMULTI;
2523                         ifr.ifr_flags = ifp->if_flags;
2524                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2525                         ifnet_serialize_all(ifp);
2526                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2527                                               NULL);
2528                         ifnet_deserialize_all(ifp);
2529                 }
2530         } else {
2531                 if (ifp->if_amcount > 1) {
2532                         ifp->if_amcount--;
2533                 } else {
2534                         ifp->if_amcount = 0;
2535                         ifp->if_flags &= ~IFF_ALLMULTI;
2536                         ifr.ifr_flags = ifp->if_flags;
2537                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2538                         ifnet_serialize_all(ifp);
2539                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2540                                               NULL);
2541                         ifnet_deserialize_all(ifp);
2542                 }
2543         }
2544
2545         crit_exit();
2546
2547         if (error == 0)
2548                 rt_ifmsg(ifp);
2549         return error;
2550 }
2551
2552 /*
2553  * Add a multicast listenership to the interface in question.
2554  * The link layer provides a routine which converts
2555  */
2556 int
2557 if_addmulti_serialized(struct ifnet *ifp, struct sockaddr *sa,
2558     struct ifmultiaddr **retifma)
2559 {
2560         struct sockaddr *llsa, *dupsa;
2561         int error;
2562         struct ifmultiaddr *ifma;
2563
2564         ASSERT_IFNET_SERIALIZED_ALL(ifp);
2565
2566         /*
2567          * If the matching multicast address already exists
2568          * then don't add a new one, just add a reference
2569          */
2570         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2571                 if (sa_equal(sa, ifma->ifma_addr)) {
2572                         ifma->ifma_refcount++;
2573                         if (retifma)
2574                                 *retifma = ifma;
2575                         return 0;
2576                 }
2577         }
2578
2579         /*
2580          * Give the link layer a chance to accept/reject it, and also
2581          * find out which AF_LINK address this maps to, if it isn't one
2582          * already.
2583          */
2584         if (ifp->if_resolvemulti) {
2585                 error = ifp->if_resolvemulti(ifp, &llsa, sa);
2586                 if (error)
2587                         return error;
2588         } else {
2589                 llsa = NULL;
2590         }
2591
2592         ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_INTWAIT);
2593         dupsa = kmalloc(sa->sa_len, M_IFMADDR, M_INTWAIT);
2594         bcopy(sa, dupsa, sa->sa_len);
2595
2596         ifma->ifma_addr = dupsa;
2597         ifma->ifma_lladdr = llsa;
2598         ifma->ifma_ifp = ifp;
2599         ifma->ifma_refcount = 1;
2600         ifma->ifma_protospec = NULL;
2601         rt_newmaddrmsg(RTM_NEWMADDR, ifma);
2602
2603         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2604         if (retifma)
2605                 *retifma = ifma;
2606
2607         if (llsa != NULL) {
2608                 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2609                         if (sa_equal(ifma->ifma_addr, llsa))
2610                                 break;
2611                 }
2612                 if (ifma) {
2613                         ifma->ifma_refcount++;
2614                 } else {
2615                         ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_INTWAIT);
2616                         dupsa = kmalloc(llsa->sa_len, M_IFMADDR, M_INTWAIT);
2617                         bcopy(llsa, dupsa, llsa->sa_len);
2618                         ifma->ifma_addr = dupsa;
2619                         ifma->ifma_ifp = ifp;
2620                         ifma->ifma_refcount = 1;
2621                         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2622                 }
2623         }
2624         /*
2625          * We are certain we have added something, so call down to the
2626          * interface to let them know about it.
2627          */
2628         if (ifp->if_ioctl)
2629                 ifp->if_ioctl(ifp, SIOCADDMULTI, 0, NULL);
2630
2631         return 0;
2632 }
2633
2634 int
2635 if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
2636     struct ifmultiaddr **retifma)
2637 {
2638         int error;
2639
2640         ifnet_serialize_all(ifp);
2641         error = if_addmulti_serialized(ifp, sa, retifma);
2642         ifnet_deserialize_all(ifp);
2643
2644         return error;
2645 }
2646
2647 /*
2648  * Remove a reference to a multicast address on this interface.  Yell
2649  * if the request does not match an existing membership.
2650  */
2651 static int
2652 if_delmulti_serialized(struct ifnet *ifp, struct sockaddr *sa)
2653 {
2654         struct ifmultiaddr *ifma;
2655
2656         ASSERT_IFNET_SERIALIZED_ALL(ifp);
2657
2658         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2659                 if (sa_equal(sa, ifma->ifma_addr))
2660                         break;
2661         if (ifma == NULL)
2662                 return ENOENT;
2663
2664         if (ifma->ifma_refcount > 1) {
2665                 ifma->ifma_refcount--;
2666                 return 0;
2667         }
2668
2669         rt_newmaddrmsg(RTM_DELMADDR, ifma);
2670         sa = ifma->ifma_lladdr;
2671         TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2672         /*
2673          * Make sure the interface driver is notified
2674          * in the case of a link layer mcast group being left.
2675          */
2676         if (ifma->ifma_addr->sa_family == AF_LINK && sa == NULL)
2677                 ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2678         kfree(ifma->ifma_addr, M_IFMADDR);
2679         kfree(ifma, M_IFMADDR);
2680         if (sa == NULL)
2681                 return 0;
2682
2683         /*
2684          * Now look for the link-layer address which corresponds to
2685          * this network address.  It had been squirreled away in
2686          * ifma->ifma_lladdr for this purpose (so we don't have
2687          * to call ifp->if_resolvemulti() again), and we saved that
2688          * value in sa above.  If some nasty deleted the
2689          * link-layer address out from underneath us, we can deal because
2690          * the address we stored was is not the same as the one which was
2691          * in the record for the link-layer address.  (So we don't complain
2692          * in that case.)
2693          */
2694         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2695                 if (sa_equal(sa, ifma->ifma_addr))
2696                         break;
2697         if (ifma == NULL)
2698                 return 0;
2699
2700         if (ifma->ifma_refcount > 1) {
2701                 ifma->ifma_refcount--;
2702                 return 0;
2703         }
2704
2705         TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2706         ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2707         kfree(ifma->ifma_addr, M_IFMADDR);
2708         kfree(sa, M_IFMADDR);
2709         kfree(ifma, M_IFMADDR);
2710
2711         return 0;
2712 }
2713
2714 int
2715 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
2716 {
2717         int error;
2718
2719         ifnet_serialize_all(ifp);
2720         error = if_delmulti_serialized(ifp, sa);
2721         ifnet_deserialize_all(ifp);
2722
2723         return error;
2724 }
2725
2726 /*
2727  * Delete all multicast group membership for an interface.
2728  * Should be used to quickly flush all multicast filters.
2729  */
2730 void
2731 if_delallmulti_serialized(struct ifnet *ifp)
2732 {
2733         struct ifmultiaddr *ifma, mark;
2734         struct sockaddr sa;
2735
2736         ASSERT_IFNET_SERIALIZED_ALL(ifp);
2737
2738         bzero(&sa, sizeof(sa));
2739         sa.sa_family = AF_UNSPEC;
2740         sa.sa_len = sizeof(sa);
2741
2742         bzero(&mark, sizeof(mark));
2743         mark.ifma_addr = &sa;
2744
2745         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, &mark, ifma_link);
2746         while ((ifma = TAILQ_NEXT(&mark, ifma_link)) != NULL) {
2747                 TAILQ_REMOVE(&ifp->if_multiaddrs, &mark, ifma_link);
2748                 TAILQ_INSERT_AFTER(&ifp->if_multiaddrs, ifma, &mark,
2749                     ifma_link);
2750
2751                 if (ifma->ifma_addr->sa_family == AF_UNSPEC)
2752                         continue;
2753
2754                 if_delmulti_serialized(ifp, ifma->ifma_addr);
2755         }
2756         TAILQ_REMOVE(&ifp->if_multiaddrs, &mark, ifma_link);
2757 }
2758
2759
2760 /*
2761  * Set the link layer address on an interface.
2762  *
2763  * At this time we only support certain types of interfaces,
2764  * and we don't allow the length of the address to change.
2765  */
2766 int
2767 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
2768 {
2769         struct sockaddr_dl *sdl;
2770         struct ifreq ifr;
2771
2772         sdl = IF_LLSOCKADDR(ifp);
2773         if (sdl == NULL)
2774                 return (EINVAL);
2775         if (len != sdl->sdl_alen)       /* don't allow length to change */
2776                 return (EINVAL);
2777         switch (ifp->if_type) {
2778         case IFT_ETHER:                 /* these types use struct arpcom */
2779         case IFT_XETHER:
2780         case IFT_L2VLAN:
2781         case IFT_IEEE8023ADLAG:
2782                 bcopy(lladdr, ((struct arpcom *)ifp->if_softc)->ac_enaddr, len);
2783                 bcopy(lladdr, LLADDR(sdl), len);
2784                 break;
2785         default:
2786                 return (ENODEV);
2787         }
2788         /*
2789          * If the interface is already up, we need
2790          * to re-init it in order to reprogram its
2791          * address filter.
2792          */
2793         ifnet_serialize_all(ifp);
2794         if ((ifp->if_flags & IFF_UP) != 0) {
2795 #ifdef INET
2796                 struct ifaddr_container *ifac;
2797 #endif
2798
2799                 ifp->if_flags &= ~IFF_UP;
2800                 ifr.ifr_flags = ifp->if_flags;
2801                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2802                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2803                               NULL);
2804                 ifp->if_flags |= IFF_UP;
2805                 ifr.ifr_flags = ifp->if_flags;
2806                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2807                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2808                                  NULL);
2809 #ifdef INET
2810                 /*
2811                  * Also send gratuitous ARPs to notify other nodes about
2812                  * the address change.
2813                  */
2814                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2815                         struct ifaddr *ifa = ifac->ifa;
2816
2817                         if (ifa->ifa_addr != NULL &&
2818                             ifa->ifa_addr->sa_family == AF_INET)
2819                                 arp_gratuitous(ifp, ifa);
2820                 }
2821 #endif
2822         }
2823         ifnet_deserialize_all(ifp);
2824         return (0);
2825 }
2826
2827
2828 /*
2829  * Locate an interface based on a complete address.
2830  */
2831 struct ifnet *
2832 if_bylla(const void *lla, unsigned char lla_len)
2833 {
2834         const struct ifnet_array *arr;
2835         struct ifnet *ifp;
2836         struct sockaddr_dl *sdl;
2837         int i;
2838
2839         arr = ifnet_array_get();
2840         for (i = 0; i < arr->ifnet_count; ++i) {
2841                 ifp = arr->ifnet_arr[i];
2842                 if (ifp->if_addrlen != lla_len)
2843                         continue;
2844
2845                 sdl = IF_LLSOCKADDR(ifp);
2846                 if (memcmp(lla, LLADDR(sdl), lla_len) == 0)
2847                         return (ifp);
2848         }
2849         return (NULL);
2850 }
2851
2852 struct ifmultiaddr *
2853 ifmaof_ifpforaddr(struct sockaddr *sa, struct ifnet *ifp)
2854 {
2855         struct ifmultiaddr *ifma;
2856
2857         /* TODO: need ifnet_serialize_main */
2858         ifnet_serialize_all(ifp);
2859         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2860                 if (sa_equal(ifma->ifma_addr, sa))
2861                         break;
2862         ifnet_deserialize_all(ifp);
2863
2864         return ifma;
2865 }
2866
2867 /*
2868  * This function locates the first real ethernet MAC from a network
2869  * card and loads it into node, returning 0 on success or ENOENT if
2870  * no suitable interfaces were found.  It is used by the uuid code to
2871  * generate a unique 6-byte number.
2872  */
2873 int
2874 if_getanyethermac(uint16_t *node, int minlen)
2875 {
2876         struct ifnet *ifp;
2877         struct sockaddr_dl *sdl;
2878
2879         ifnet_lock();
2880         TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
2881                 if (ifp->if_type != IFT_ETHER)
2882                         continue;
2883                 sdl = IF_LLSOCKADDR(ifp);
2884                 if (sdl->sdl_alen < minlen)
2885                         continue;
2886                 bcopy(((struct arpcom *)ifp->if_softc)->ac_enaddr, node,
2887                       minlen);
2888                 ifnet_unlock();
2889                 return(0);
2890         }
2891         ifnet_unlock();
2892         return (ENOENT);
2893 }
2894
2895 /*
2896  * The name argument must be a pointer to storage which will last as
2897  * long as the interface does.  For physical devices, the result of
2898  * device_get_name(dev) is a good choice and for pseudo-devices a
2899  * static string works well.
2900  */
2901 void
2902 if_initname(struct ifnet *ifp, const char *name, int unit)
2903 {
2904         ifp->if_dname = name;
2905         ifp->if_dunit = unit;
2906         if (unit != IF_DUNIT_NONE)
2907                 ksnprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
2908         else
2909                 strlcpy(ifp->if_xname, name, IFNAMSIZ);
2910 }
2911
2912 int
2913 if_printf(struct ifnet *ifp, const char *fmt, ...)
2914 {
2915         __va_list ap;
2916         int retval;
2917
2918         retval = kprintf("%s: ", ifp->if_xname);
2919         __va_start(ap, fmt);
2920         retval += kvprintf(fmt, ap);
2921         __va_end(ap);
2922         return (retval);
2923 }
2924
2925 struct ifnet *
2926 if_alloc(uint8_t type)
2927 {
2928         struct ifnet *ifp;
2929         size_t size;
2930
2931         /*
2932          * XXX temporary hack until arpcom is setup in if_l2com
2933          */
2934         if (type == IFT_ETHER)
2935                 size = sizeof(struct arpcom);
2936         else
2937                 size = sizeof(struct ifnet);
2938
2939         ifp = kmalloc(size, M_IFNET, M_WAITOK|M_ZERO);
2940
2941         ifp->if_type = type;
2942
2943         if (if_com_alloc[type] != NULL) {
2944                 ifp->if_l2com = if_com_alloc[type](type, ifp);
2945                 if (ifp->if_l2com == NULL) {
2946                         kfree(ifp, M_IFNET);
2947                         return (NULL);
2948                 }
2949         }
2950         return (ifp);
2951 }
2952
2953 void
2954 if_free(struct ifnet *ifp)
2955 {
2956         kfree(ifp, M_IFNET);
2957 }
2958
2959 void
2960 ifq_set_classic(struct ifaltq *ifq)
2961 {
2962         ifq_set_methods(ifq, ifq->altq_ifp->if_mapsubq,
2963             ifsq_classic_enqueue, ifsq_classic_dequeue, ifsq_classic_request);
2964 }
2965
2966 void
2967 ifq_set_methods(struct ifaltq *ifq, altq_mapsubq_t mapsubq,
2968     ifsq_enqueue_t enqueue, ifsq_dequeue_t dequeue, ifsq_request_t request)
2969 {
2970         int q;
2971
2972         KASSERT(mapsubq != NULL, ("mapsubq is not specified"));
2973         KASSERT(enqueue != NULL, ("enqueue is not specified"));
2974         KASSERT(dequeue != NULL, ("dequeue is not specified"));
2975         KASSERT(request != NULL, ("request is not specified"));
2976
2977         ifq->altq_mapsubq = mapsubq;
2978         for (q = 0; q < ifq->altq_subq_cnt; ++q) {
2979                 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
2980
2981                 ifsq->ifsq_enqueue = enqueue;
2982                 ifsq->ifsq_dequeue = dequeue;
2983                 ifsq->ifsq_request = request;
2984         }
2985 }
2986
2987 static void
2988 ifsq_norm_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
2989 {
2990
2991         classq_add(&ifsq->ifsq_norm, m);
2992         ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
2993 }
2994
2995 static void
2996 ifsq_prio_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
2997 {
2998
2999         classq_add(&ifsq->ifsq_prio, m);
3000         ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
3001         ALTQ_SQ_PRIO_CNTR_INC(ifsq, m->m_pkthdr.len);
3002 }
3003
3004 static struct mbuf *
3005 ifsq_norm_dequeue(struct ifaltq_subque *ifsq)
3006 {
3007         struct mbuf *m;
3008
3009         m = classq_get(&ifsq->ifsq_norm);
3010         if (m != NULL)
3011                 ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
3012         return (m);
3013 }
3014
3015 static struct mbuf *
3016 ifsq_prio_dequeue(struct ifaltq_subque *ifsq)
3017 {
3018         struct mbuf *m;
3019
3020         m = classq_get(&ifsq->ifsq_prio);
3021         if (m != NULL) {
3022                 ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
3023                 ALTQ_SQ_PRIO_CNTR_DEC(ifsq, m->m_pkthdr.len);
3024         }
3025         return (m);
3026 }
3027
3028 int
3029 ifsq_classic_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m,
3030     struct altq_pktattr *pa __unused)
3031 {
3032
3033         M_ASSERTPKTHDR(m);
3034 again:
3035         if (ifsq->ifsq_len >= ifsq->ifsq_maxlen ||
3036             ifsq->ifsq_bcnt >= ifsq->ifsq_maxbcnt) {
3037                 struct mbuf *m_drop;
3038
3039                 if (m->m_flags & M_PRIO) {
3040                         m_drop = NULL;
3041                         if (ifsq->ifsq_prio_len < (ifsq->ifsq_maxlen >> 1) &&
3042                             ifsq->ifsq_prio_bcnt < (ifsq->ifsq_maxbcnt >> 1)) {
3043                                 /* Try dropping some from normal queue. */
3044                                 m_drop = ifsq_norm_dequeue(ifsq);
3045                         }
3046                         if (m_drop == NULL)
3047                                 m_drop = ifsq_prio_dequeue(ifsq);
3048                 } else {
3049                         m_drop = ifsq_norm_dequeue(ifsq);
3050                 }
3051                 if (m_drop != NULL) {
3052                         IFNET_STAT_INC(ifsq->ifsq_ifp, oqdrops, 1);
3053                         m_freem(m_drop);
3054                         goto again;
3055                 }
3056                 /*
3057                  * No old packets could be dropped!
3058                  * NOTE: Caller increases oqdrops.
3059                  */
3060                 m_freem(m);
3061                 return (ENOBUFS);
3062         } else {
3063                 if (m->m_flags & M_PRIO)
3064                         ifsq_prio_enqueue(ifsq, m);
3065                 else
3066                         ifsq_norm_enqueue(ifsq, m);
3067                 return (0);
3068         }
3069 }
3070
3071 struct mbuf *
3072 ifsq_classic_dequeue(struct ifaltq_subque *ifsq, int op)
3073 {
3074         struct mbuf *m;
3075
3076         switch (op) {
3077         case ALTDQ_POLL:
3078                 m = classq_head(&ifsq->ifsq_prio);
3079                 if (m == NULL)
3080                         m = classq_head(&ifsq->ifsq_norm);
3081                 break;
3082
3083         case ALTDQ_REMOVE:
3084                 m = ifsq_prio_dequeue(ifsq);
3085                 if (m == NULL)
3086                         m = ifsq_norm_dequeue(ifsq);
3087                 break;
3088
3089         default:
3090                 panic("unsupported ALTQ dequeue op: %d", op);
3091         }
3092         return m;
3093 }
3094
3095 int
3096 ifsq_classic_request(struct ifaltq_subque *ifsq, int req, void *arg)
3097 {
3098         switch (req) {
3099         case ALTRQ_PURGE:
3100                 for (;;) {
3101                         struct mbuf *m;
3102
3103                         m = ifsq_classic_dequeue(ifsq, ALTDQ_REMOVE);
3104                         if (m == NULL)
3105                                 break;
3106                         m_freem(m);
3107                 }
3108                 break;
3109
3110         default:
3111                 panic("unsupported ALTQ request: %d", req);
3112         }
3113         return 0;
3114 }
3115
3116 static void
3117 ifsq_ifstart_try(struct ifaltq_subque *ifsq, int force_sched)
3118 {
3119         struct ifnet *ifp = ifsq_get_ifp(ifsq);
3120         int running = 0, need_sched;
3121
3122         /*
3123          * Try to do direct ifnet.if_start on the subqueue first, if there is
3124          * contention on the subqueue hardware serializer, ifnet.if_start on
3125          * the subqueue will be scheduled on the subqueue owner CPU.
3126          */
3127         if (!ifsq_tryserialize_hw(ifsq)) {
3128                 /*
3129                  * Subqueue hardware serializer contention happened,
3130                  * ifnet.if_start on the subqueue is scheduled on
3131                  * the subqueue owner CPU, and we keep going.
3132                  */
3133                 ifsq_ifstart_schedule(ifsq, 1);
3134                 return;
3135         }
3136
3137         if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
3138                 ifp->if_start(ifp, ifsq);
3139                 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
3140                         running = 1;
3141         }
3142         need_sched = ifsq_ifstart_need_schedule(ifsq, running);
3143
3144         ifsq_deserialize_hw(ifsq);
3145
3146         if (need_sched) {
3147                 /*
3148                  * More data need to be transmitted, ifnet.if_start on the
3149                  * subqueue is scheduled on the subqueue owner CPU, and we
3150                  * keep going.
3151                  * NOTE: ifnet.if_start subqueue interlock is not released.
3152                  */
3153                 ifsq_ifstart_schedule(ifsq, force_sched);
3154         }
3155 }
3156
3157 /*
3158  * Subqeue packets staging mechanism:
3159  *
3160  * The packets enqueued into the subqueue are staged to a certain amount
3161  * before the ifnet.if_start on the subqueue is called.  In this way, the
3162  * driver could avoid writing to hardware registers upon every packet,
3163  * instead, hardware registers could be written when certain amount of
3164  * packets are put onto hardware TX ring.  The measurement on several modern
3165  * NICs (emx(4), igb(4), bnx(4), bge(4), jme(4)) shows that the hardware
3166  * registers writing aggregation could save ~20% CPU time when 18bytes UDP
3167  * datagrams are transmitted at 1.48Mpps.  The performance improvement by
3168  * hardware registers writing aggeregation is also mentioned by Luigi Rizzo's
3169  * netmap paper (http://info.iet.unipi.it/~luigi/netmap/).
3170  *
3171  * Subqueue packets staging is performed for two entry points into drivers'
3172  * transmission function:
3173  * - Direct ifnet.if_start calling on the subqueue, i.e. ifsq_ifstart_try()
3174  * - ifnet.if_start scheduling on the subqueue, i.e. ifsq_ifstart_schedule()
3175  *
3176  * Subqueue packets staging will be stopped upon any of the following
3177  * conditions:
3178  * - If the count of packets enqueued on the current CPU is great than or
3179  *   equal to ifsq_stage_cntmax. (XXX this should be per-interface)
3180  * - If the total length of packets enqueued on the current CPU is great
3181  *   than or equal to the hardware's MTU - max_protohdr.  max_protohdr is
3182  *   cut from the hardware's MTU mainly bacause a full TCP segment's size
3183  *   is usually less than hardware's MTU.
3184  * - ifsq_ifstart_schedule() is not pending on the current CPU and
3185  *   ifnet.if_start subqueue interlock (ifaltq_subq.ifsq_started) is not
3186  *   released.
3187  * - The if_start_rollup(), which is registered as low priority netisr
3188  *   rollup function, is called; probably because no more work is pending
3189  *   for netisr.
3190  *
3191  * NOTE:
3192  * Currently subqueue packet staging is only performed in netisr threads.
3193  */
3194 int
3195 ifq_dispatch(struct ifnet *ifp, struct mbuf *m, struct altq_pktattr *pa)
3196 {
3197         struct ifaltq *ifq = &ifp->if_snd;
3198         struct ifaltq_subque *ifsq;
3199         int error, start = 0, len, mcast = 0, avoid_start = 0;
3200         struct ifsubq_stage_head *head = NULL;
3201         struct ifsubq_stage *stage = NULL;
3202         struct globaldata *gd = mycpu;
3203         struct thread *td = gd->gd_curthread;
3204
3205         crit_enter_quick(td);
3206
3207         ifsq = ifq_map_subq(ifq, gd->gd_cpuid);
3208         ASSERT_ALTQ_SQ_NOT_SERIALIZED_HW(ifsq);
3209
3210         len = m->m_pkthdr.len;
3211         if (m->m_flags & M_MCAST)
3212                 mcast = 1;
3213
3214         if (td->td_type == TD_TYPE_NETISR) {
3215                 head = &ifsubq_stage_heads[mycpuid];
3216                 stage = ifsq_get_stage(ifsq, mycpuid);
3217
3218                 stage->stg_cnt++;
3219                 stage->stg_len += len;
3220                 if (stage->stg_cnt < ifsq_stage_cntmax &&
3221                     stage->stg_len < (ifp->if_mtu - max_protohdr))
3222                         avoid_start = 1;
3223         }
3224
3225         ALTQ_SQ_LOCK(ifsq);
3226         error = ifsq_enqueue_locked(ifsq, m, pa);
3227         if (error) {
3228                 IFNET_STAT_INC(ifp, oqdrops, 1);
3229                 if (!ifsq_data_ready(ifsq)) {
3230                         ALTQ_SQ_UNLOCK(ifsq);
3231                         crit_exit_quick(td);
3232                         return error;
3233                 }
3234                 avoid_start = 0;
3235         }
3236         if (!ifsq_is_started(ifsq)) {
3237                 if (avoid_start) {
3238                         ALTQ_SQ_UNLOCK(ifsq);
3239
3240                         KKASSERT(!error);
3241                         if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
3242                                 ifsq_stage_insert(head, stage);
3243
3244                         IFNET_STAT_INC(ifp, obytes, len);
3245                         if (mcast)
3246                                 IFNET_STAT_INC(ifp, omcasts, 1);
3247                         crit_exit_quick(td);
3248                         return error;
3249                 }
3250
3251                 /*
3252                  * Hold the subqueue interlock of ifnet.if_start
3253                  */
3254                 ifsq_set_started(ifsq);
3255                 start = 1;
3256         }
3257         ALTQ_SQ_UNLOCK(ifsq);
3258
3259         if (!error) {
3260                 IFNET_STAT_INC(ifp, obytes, len);
3261                 if (mcast)
3262                         IFNET_STAT_INC(ifp, omcasts, 1);
3263         }
3264
3265         if (stage != NULL) {
3266                 if (!start && (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)) {
3267                         KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
3268                         if (!avoid_start) {
3269                                 ifsq_stage_remove(head, stage);
3270                                 ifsq_ifstart_schedule(ifsq, 1);
3271                         }
3272                         crit_exit_quick(td);
3273                         return error;
3274                 }
3275
3276                 if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED) {
3277                         ifsq_stage_remove(head, stage);
3278                 } else {
3279                         stage->stg_cnt = 0;
3280                         stage->stg_len = 0;
3281                 }
3282         }
3283
3284         if (!start) {
3285                 crit_exit_quick(td);
3286                 return error;
3287         }
3288
3289         ifsq_ifstart_try(ifsq, 0);
3290
3291         crit_exit_quick(td);
3292         return error;
3293 }
3294
3295 void *
3296 ifa_create(int size)
3297 {
3298         struct ifaddr *ifa;
3299         int i;
3300
3301         KASSERT(size >= sizeof(*ifa), ("ifaddr size too small"));
3302
3303         ifa = kmalloc(size, M_IFADDR, M_INTWAIT | M_ZERO);
3304
3305         /*
3306          * Make ifa_container availabel on all CPUs, since they
3307          * could be accessed by any threads.
3308          */
3309         ifa->ifa_containers =
3310                 kmalloc(ncpus * sizeof(struct ifaddr_container),
3311                         M_IFADDR,
3312                         M_INTWAIT | M_ZERO | M_CACHEALIGN);
3313
3314         ifa->ifa_ncnt = ncpus;
3315         for (i = 0; i < ncpus; ++i) {
3316                 struct ifaddr_container *ifac = &ifa->ifa_containers[i];
3317
3318                 ifac->ifa_magic = IFA_CONTAINER_MAGIC;
3319                 ifac->ifa = ifa;
3320                 ifac->ifa_refcnt = 1;
3321         }
3322 #ifdef IFADDR_DEBUG
3323         kprintf("alloc ifa %p %d\n", ifa, size);
3324 #endif
3325         return ifa;
3326 }
3327
3328 void
3329 ifac_free(struct ifaddr_container *ifac, int cpu_id)
3330 {
3331         struct ifaddr *ifa = ifac->ifa;
3332
3333         KKASSERT(ifac->ifa_magic == IFA_CONTAINER_MAGIC);
3334         KKASSERT(ifac->ifa_refcnt == 0);
3335         KASSERT(ifac->ifa_listmask == 0,
3336                 ("ifa is still on %#x lists", ifac->ifa_listmask));
3337
3338         ifac->ifa_magic = IFA_CONTAINER_DEAD;
3339
3340 #ifdef IFADDR_DEBUG_VERBOSE
3341         kprintf("try free ifa %p cpu_id %d\n", ifac->ifa, cpu_id);
3342 #endif
3343
3344         KASSERT(ifa->ifa_ncnt > 0 && ifa->ifa_ncnt <= ncpus,
3345                 ("invalid # of ifac, %d", ifa->ifa_ncnt));
3346         if (atomic_fetchadd_int(&ifa->ifa_ncnt, -1) == 1) {
3347 #ifdef IFADDR_DEBUG
3348                 kprintf("free ifa %p\n", ifa);
3349 #endif
3350                 kfree(ifa->ifa_containers, M_IFADDR);
3351                 kfree(ifa, M_IFADDR);
3352         }
3353 }
3354
3355 static void
3356 ifa_iflink_dispatch(netmsg_t nmsg)
3357 {
3358         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3359         struct ifaddr *ifa = msg->ifa;
3360         struct ifnet *ifp = msg->ifp;
3361         int cpu = mycpuid;
3362         struct ifaddr_container *ifac;
3363
3364         crit_enter();
3365
3366         ifac = &ifa->ifa_containers[cpu];
3367         ASSERT_IFAC_VALID(ifac);
3368         KASSERT((ifac->ifa_listmask & IFA_LIST_IFADDRHEAD) == 0,
3369                 ("ifaddr is on if_addrheads"));
3370
3371         ifac->ifa_listmask |= IFA_LIST_IFADDRHEAD;
3372         if (msg->tail)
3373                 TAILQ_INSERT_TAIL(&ifp->if_addrheads[cpu], ifac, ifa_link);
3374         else
3375                 TAILQ_INSERT_HEAD(&ifp->if_addrheads[cpu], ifac, ifa_link);
3376
3377         crit_exit();
3378
3379         netisr_forwardmsg_all(&nmsg->base, cpu + 1);
3380 }
3381
3382 void
3383 ifa_iflink(struct ifaddr *ifa, struct ifnet *ifp, int tail)
3384 {
3385         struct netmsg_ifaddr msg;
3386
3387         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3388                     0, ifa_iflink_dispatch);
3389         msg.ifa = ifa;
3390         msg.ifp = ifp;
3391         msg.tail = tail;
3392
3393         netisr_domsg(&msg.base, 0);
3394 }
3395
3396 static void
3397 ifa_ifunlink_dispatch(netmsg_t nmsg)
3398 {
3399         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3400         struct ifaddr *ifa = msg->ifa;
3401         struct ifnet *ifp = msg->ifp;
3402         int cpu = mycpuid;
3403         struct ifaddr_container *ifac;
3404
3405         crit_enter();
3406
3407         ifac = &ifa->ifa_containers[cpu];
3408         ASSERT_IFAC_VALID(ifac);
3409         KASSERT(ifac->ifa_listmask & IFA_LIST_IFADDRHEAD,
3410                 ("ifaddr is not on if_addrhead"));
3411
3412         TAILQ_REMOVE(&ifp->if_addrheads[cpu], ifac, ifa_link);
3413         ifac->ifa_listmask &= ~IFA_LIST_IFADDRHEAD;
3414
3415         crit_exit();
3416
3417         netisr_forwardmsg_all(&nmsg->base, cpu + 1);
3418 }
3419
3420 void
3421 ifa_ifunlink(struct ifaddr *ifa, struct ifnet *ifp)
3422 {
3423         struct netmsg_ifaddr msg;
3424
3425         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3426                     0, ifa_ifunlink_dispatch);
3427         msg.ifa = ifa;
3428         msg.ifp = ifp;
3429
3430         netisr_domsg(&msg.base, 0);
3431 }
3432
3433 static void
3434 ifa_destroy_dispatch(netmsg_t nmsg)
3435 {
3436         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3437
3438         IFAFREE(msg->ifa);
3439         netisr_forwardmsg_all(&nmsg->base, mycpuid + 1);
3440 }
3441
3442 void
3443 ifa_destroy(struct ifaddr *ifa)
3444 {
3445         struct netmsg_ifaddr msg;
3446
3447         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3448                     0, ifa_destroy_dispatch);
3449         msg.ifa = ifa;
3450
3451         netisr_domsg(&msg.base, 0);
3452 }
3453
3454 static void
3455 if_start_rollup(void)
3456 {
3457         struct ifsubq_stage_head *head = &ifsubq_stage_heads[mycpuid];
3458         struct ifsubq_stage *stage;
3459
3460         crit_enter();
3461
3462         while ((stage = TAILQ_FIRST(&head->stg_head)) != NULL) {
3463                 struct ifaltq_subque *ifsq = stage->stg_subq;
3464                 int is_sched = 0;
3465
3466                 if (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)
3467                         is_sched = 1;
3468                 ifsq_stage_remove(head, stage);
3469
3470                 if (is_sched) {
3471                         ifsq_ifstart_schedule(ifsq, 1);
3472                 } else {
3473                         int start = 0;
3474
3475                         ALTQ_SQ_LOCK(ifsq);
3476                         if (!ifsq_is_started(ifsq)) {
3477                                 /*
3478                                  * Hold the subqueue interlock of
3479                                  * ifnet.if_start
3480                                  */
3481                                 ifsq_set_started(ifsq);
3482                                 start = 1;
3483                         }
3484                         ALTQ_SQ_UNLOCK(ifsq);
3485
3486                         if (start)
3487                                 ifsq_ifstart_try(ifsq, 1);
3488                 }
3489                 KKASSERT((stage->stg_flags &
3490                     (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
3491         }
3492
3493         crit_exit();
3494 }
3495
3496 static void
3497 ifnetinit(void *dummy __unused)
3498 {
3499         int i;
3500
3501         /* XXX netisr_ncpus */
3502         for (i = 0; i < ncpus; ++i)
3503                 TAILQ_INIT(&ifsubq_stage_heads[i].stg_head);
3504         netisr_register_rollup(if_start_rollup, NETISR_ROLLUP_PRIO_IFSTART);
3505 }
3506
3507 void
3508 if_register_com_alloc(u_char type,
3509     if_com_alloc_t *a, if_com_free_t *f)
3510 {
3511
3512         KASSERT(if_com_alloc[type] == NULL,
3513             ("if_register_com_alloc: %d already registered", type));
3514         KASSERT(if_com_free[type] == NULL,
3515             ("if_register_com_alloc: %d free already registered", type));
3516
3517         if_com_alloc[type] = a;
3518         if_com_free[type] = f;
3519 }
3520
3521 void
3522 if_deregister_com_alloc(u_char type)
3523 {
3524
3525         KASSERT(if_com_alloc[type] != NULL,
3526             ("if_deregister_com_alloc: %d not registered", type));
3527         KASSERT(if_com_free[type] != NULL,
3528             ("if_deregister_com_alloc: %d free not registered", type));
3529         if_com_alloc[type] = NULL;
3530         if_com_free[type] = NULL;
3531 }
3532
3533 void
3534 ifq_set_maxlen(struct ifaltq *ifq, int len)
3535 {
3536         ifq->altq_maxlen = len + (ncpus * ifsq_stage_cntmax);
3537 }
3538
3539 int
3540 ifq_mapsubq_default(struct ifaltq *ifq __unused, int cpuid __unused)
3541 {
3542         return ALTQ_SUBQ_INDEX_DEFAULT;
3543 }
3544
3545 int
3546 ifq_mapsubq_modulo(struct ifaltq *ifq, int cpuid)
3547 {
3548
3549         return (cpuid % ifq->altq_subq_mappriv);
3550 }
3551
3552 static void
3553 ifsq_watchdog(void *arg)
3554 {
3555         struct ifsubq_watchdog *wd = arg;
3556         struct ifnet *ifp;
3557
3558         if (__predict_true(wd->wd_timer == 0 || --wd->wd_timer))
3559                 goto done;
3560
3561         ifp = ifsq_get_ifp(wd->wd_subq);
3562         if (ifnet_tryserialize_all(ifp)) {
3563                 wd->wd_watchdog(wd->wd_subq);
3564                 ifnet_deserialize_all(ifp);
3565         } else {
3566                 /* try again next timeout */
3567                 wd->wd_timer = 1;
3568         }
3569 done:
3570         ifsq_watchdog_reset(wd);
3571 }
3572
3573 static void
3574 ifsq_watchdog_reset(struct ifsubq_watchdog *wd)
3575 {
3576         callout_reset_bycpu(&wd->wd_callout, hz, ifsq_watchdog, wd,
3577             ifsq_get_cpuid(wd->wd_subq));
3578 }
3579
3580 void
3581 ifsq_watchdog_init(struct ifsubq_watchdog *wd, struct ifaltq_subque *ifsq,
3582     ifsq_watchdog_t watchdog)
3583 {
3584         callout_init_mp(&wd->wd_callout);
3585         wd->wd_timer = 0;
3586         wd->wd_subq = ifsq;
3587         wd->wd_watchdog = watchdog;
3588 }
3589
3590 void
3591 ifsq_watchdog_start(struct ifsubq_watchdog *wd)
3592 {
3593         wd->wd_timer = 0;
3594         ifsq_watchdog_reset(wd);
3595 }
3596
3597 void
3598 ifsq_watchdog_stop(struct ifsubq_watchdog *wd)
3599 {
3600         wd->wd_timer = 0;
3601         callout_stop(&wd->wd_callout);
3602 }
3603
3604 void
3605 ifnet_lock(void)
3606 {
3607         KASSERT(curthread->td_type != TD_TYPE_NETISR,
3608             ("try holding ifnet lock in netisr"));
3609         mtx_lock(&ifnet_mtx);
3610 }
3611
3612 void
3613 ifnet_unlock(void)
3614 {
3615         KASSERT(curthread->td_type != TD_TYPE_NETISR,
3616             ("try holding ifnet lock in netisr"));
3617         mtx_unlock(&ifnet_mtx);
3618 }
3619
3620 static struct ifnet_array *
3621 ifnet_array_alloc(int count)
3622 {
3623         struct ifnet_array *arr;
3624
3625         arr = kmalloc(__offsetof(struct ifnet_array, ifnet_arr[count]),
3626             M_IFNET, M_WAITOK);
3627         arr->ifnet_count = count;
3628
3629         return arr;
3630 }
3631
3632 static void
3633 ifnet_array_free(struct ifnet_array *arr)
3634 {
3635         if (arr == &ifnet_array0)
3636                 return;
3637         kfree(arr, M_IFNET);
3638 }
3639
3640 static struct ifnet_array *
3641 ifnet_array_add(struct ifnet *ifp, const struct ifnet_array *old_arr)
3642 {
3643         struct ifnet_array *arr;
3644         int count, i;
3645
3646         KASSERT(old_arr->ifnet_count >= 0,
3647             ("invalid ifnet array count %d", old_arr->ifnet_count));
3648         count = old_arr->ifnet_count + 1;
3649         arr = ifnet_array_alloc(count);
3650
3651         /*
3652          * Save the old ifnet array and append this ifp to the end of
3653          * the new ifnet array.
3654          */
3655         for (i = 0; i < old_arr->ifnet_count; ++i) {
3656                 KASSERT(old_arr->ifnet_arr[i] != ifp,
3657                     ("%s is already in ifnet array", ifp->if_xname));
3658                 arr->ifnet_arr[i] = old_arr->ifnet_arr[i];
3659         }
3660         KASSERT(i == count - 1,
3661             ("add %s, ifnet array index mismatch, should be %d, but got %d",
3662              ifp->if_xname, count - 1, i));
3663         arr->ifnet_arr[i] = ifp;
3664
3665         return arr;
3666 }
3667
3668 static struct ifnet_array *
3669 ifnet_array_del(struct ifnet *ifp, const struct ifnet_array *old_arr)
3670 {
3671         struct ifnet_array *arr;
3672         int count, i, idx, found = 0;
3673
3674         KASSERT(old_arr->ifnet_count > 0,
3675             ("invalid ifnet array count %d", old_arr->ifnet_count));
3676         count = old_arr->ifnet_count - 1;
3677         arr = ifnet_array_alloc(count);
3678
3679         /*
3680          * Save the old ifnet array, but skip this ifp.
3681          */
3682         idx = 0;
3683         for (i = 0; i < old_arr->ifnet_count; ++i) {
3684                 if (old_arr->ifnet_arr[i] == ifp) {
3685                         KASSERT(!found,
3686                             ("dup %s is in ifnet array", ifp->if_xname));
3687                         found = 1;
3688                         continue;
3689                 }
3690                 KASSERT(idx < count,
3691                     ("invalid ifnet array index %d, count %d", idx, count));
3692                 arr->ifnet_arr[idx] = old_arr->ifnet_arr[i];
3693                 ++idx;
3694         }
3695         KASSERT(found, ("%s is not in ifnet array", ifp->if_xname));
3696         KASSERT(idx == count,
3697             ("del %s, ifnet array count mismatch, should be %d, but got %d ",
3698              ifp->if_xname, count, idx));
3699
3700         return arr;
3701 }
3702
3703 const struct ifnet_array *
3704 ifnet_array_get(void)
3705 {
3706         const struct ifnet_array *ret;
3707
3708         KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
3709         ret = ifnet_array;
3710         /* Make sure 'ret' is really used. */
3711         cpu_ccfence();
3712         return (ret);
3713 }
3714
3715 int
3716 ifnet_array_isempty(void)
3717 {
3718         KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
3719         if (ifnet_array->ifnet_count == 0)
3720                 return 1;
3721         else
3722                 return 0;
3723 }
3724
3725 void
3726 ifa_marker_init(struct ifaddr_marker *mark, struct ifnet *ifp)
3727 {
3728         struct ifaddr *ifa;
3729
3730         memset(mark, 0, sizeof(*mark));
3731         ifa = &mark->ifa;
3732
3733         mark->ifac.ifa = ifa;
3734
3735         ifa->ifa_addr = &mark->addr;
3736         ifa->ifa_dstaddr = &mark->dstaddr;
3737         ifa->ifa_netmask = &mark->netmask;
3738         ifa->ifa_ifp = ifp;
3739 }
3740
3741 static int
3742 if_ringcnt_fixup(int ring_cnt, int ring_cntmax)
3743 {
3744
3745         KASSERT(ring_cntmax > 0, ("invalid ring count max %d", ring_cntmax));
3746
3747         if (ring_cnt <= 0 || ring_cnt > ring_cntmax)
3748                 ring_cnt = ring_cntmax;
3749         if (ring_cnt > netisr_ncpus)
3750                 ring_cnt = netisr_ncpus;
3751         return (ring_cnt);
3752 }
3753
3754 static void
3755 if_ringmap_set_grid(device_t dev, struct if_ringmap *rm, int grid)
3756 {
3757         int i, offset;
3758
3759         KASSERT(grid > 0, ("invalid if_ringmap grid %d", grid));
3760         KASSERT(grid >= rm->rm_cnt, ("invalid if_ringmap grid %d, count %d",
3761             grid, rm->rm_cnt));
3762         rm->rm_grid = grid;
3763
3764         offset = (rm->rm_grid * device_get_unit(dev)) % netisr_ncpus;
3765         for (i = 0; i < rm->rm_cnt; ++i) {
3766                 rm->rm_cpumap[i] = offset + i;
3767                 KASSERT(rm->rm_cpumap[i] < netisr_ncpus,
3768                     ("invalid cpumap[%d] = %d, offset %d", i,
3769                      rm->rm_cpumap[i], offset));
3770         }
3771 }
3772
3773 static struct if_ringmap *
3774 if_ringmap_alloc_flags(device_t dev, int ring_cnt, int ring_cntmax,
3775     uint32_t flags)
3776 {
3777         struct if_ringmap *rm;
3778         int i, grid = 0, prev_grid;
3779
3780         ring_cnt = if_ringcnt_fixup(ring_cnt, ring_cntmax);
3781         rm = kmalloc(__offsetof(struct if_ringmap, rm_cpumap[ring_cnt]),
3782             M_DEVBUF, M_WAITOK | M_ZERO);
3783
3784         rm->rm_cnt = ring_cnt;
3785         if (flags & RINGMAP_FLAG_POWEROF2)
3786                 rm->rm_cnt = 1 << (fls(rm->rm_cnt) - 1);
3787
3788         prev_grid = netisr_ncpus;
3789         for (i = 0; i < netisr_ncpus; ++i) {
3790                 if (netisr_ncpus % (i + 1) != 0)
3791                         continue;
3792
3793                 grid = netisr_ncpus / (i + 1);
3794                 if (rm->rm_cnt > grid) {
3795                         grid = prev_grid;
3796                         break;
3797                 }
3798
3799                 if (rm->rm_cnt > netisr_ncpus / (i + 2))
3800                         break;
3801                 prev_grid = grid;
3802         }
3803         if_ringmap_set_grid(dev, rm, grid);
3804
3805         return (rm);
3806 }
3807
3808 struct if_ringmap *
3809 if_ringmap_alloc(device_t dev, int ring_cnt, int ring_cntmax)
3810 {
3811
3812         return (if_ringmap_alloc_flags(dev, ring_cnt, ring_cntmax,
3813             RINGMAP_FLAG_NONE));
3814 }
3815
3816 struct if_ringmap *
3817 if_ringmap_alloc2(device_t dev, int ring_cnt, int ring_cntmax)
3818 {
3819
3820         return (if_ringmap_alloc_flags(dev, ring_cnt, ring_cntmax,
3821             RINGMAP_FLAG_POWEROF2));
3822 }
3823
3824 void
3825 if_ringmap_free(struct if_ringmap *rm)
3826 {
3827
3828         kfree(rm, M_DEVBUF);
3829 }
3830
3831 /*
3832  * Align the two ringmaps.
3833  *
3834  * e.g. 8 netisrs, rm0 contains 4 rings, rm1 contains 2 rings.
3835  *
3836  * Before:
3837  *
3838  * CPU      0  1  2  3   4  5  6  7
3839  * NIC_RX               n0 n1 n2 n3
3840  * NIC_TX        N0 N1
3841  *
3842  * After:
3843  *
3844  * CPU      0  1  2  3   4  5  6  7
3845  * NIC_RX               n0 n1 n2 n3
3846  * NIC_TX               N0 N1
3847  */
3848 void
3849 if_ringmap_align(device_t dev, struct if_ringmap *rm0, struct if_ringmap *rm1)
3850 {
3851
3852         if (rm0->rm_grid > rm1->rm_grid)
3853                 if_ringmap_set_grid(dev, rm1, rm0->rm_grid);
3854         else if (rm0->rm_grid < rm1->rm_grid)
3855                 if_ringmap_set_grid(dev, rm0, rm1->rm_grid);
3856 }
3857
3858 void
3859 if_ringmap_match(device_t dev, struct if_ringmap *rm0, struct if_ringmap *rm1)
3860 {
3861         int subset_grid, cnt, divisor, mod, offset, i;
3862         struct if_ringmap *subset_rm, *rm;
3863         int old_rm0_grid, old_rm1_grid;
3864
3865         if (rm0->rm_grid == rm1->rm_grid)
3866                 return;
3867
3868         /* Save grid for later use */
3869         old_rm0_grid = rm0->rm_grid;
3870         old_rm1_grid = rm1->rm_grid;
3871
3872         if_ringmap_align(dev, rm0, rm1);
3873
3874         /*
3875          * Re-shuffle rings to get more even distribution.
3876          *
3877          * e.g. 12 netisrs, rm0 contains 4 rings, rm1 contains 2 rings.
3878          *
3879          * CPU       0  1  2  3   4  5  6  7   8  9 10 11
3880          *
3881          * NIC_RX   a0 a1 a2 a3  b0 b1 b2 b3  c0 c1 c2 c3
3882          * NIC_TX   A0 A1        B0 B1        C0 C1
3883          *
3884          * NIC_RX   d0 d1 d2 d3  e0 e1 e2 e3  f0 f1 f2 f3
3885          * NIC_TX         D0 D1        E0 E1        F0 F1
3886          */
3887
3888         if (rm0->rm_cnt >= (2 * old_rm1_grid)) {
3889                 cnt = rm0->rm_cnt;
3890                 subset_grid = old_rm1_grid;
3891                 subset_rm = rm1;
3892                 rm = rm0;
3893         } else if (rm1->rm_cnt > (2 * old_rm0_grid)) {
3894                 cnt = rm1->rm_cnt;
3895                 subset_grid = old_rm0_grid;
3896                 subset_rm = rm0;
3897                 rm = rm1;
3898         } else {
3899                 /* No space to shuffle. */
3900                 return;
3901         }
3902
3903         mod = cnt / subset_grid;
3904         KKASSERT(mod >= 2);
3905         divisor = netisr_ncpus / rm->rm_grid;
3906         offset = ((device_get_unit(dev) / divisor) % mod) * subset_grid;
3907
3908         for (i = 0; i < subset_rm->rm_cnt; ++i) {
3909                 subset_rm->rm_cpumap[i] += offset;
3910                 KASSERT(subset_rm->rm_cpumap[i] < netisr_ncpus,
3911                     ("match: invalid cpumap[%d] = %d, offset %d",
3912                      i, subset_rm->rm_cpumap[i], offset));
3913         }
3914 #ifdef INVARIANTS
3915         for (i = 0; i < subset_rm->rm_cnt; ++i) {
3916                 int j;
3917
3918                 for (j = 0; j < rm->rm_cnt; ++j) {
3919                         if (rm->rm_cpumap[j] == subset_rm->rm_cpumap[i])
3920                                 break;
3921                 }
3922                 KASSERT(j < rm->rm_cnt,
3923                     ("subset cpumap[%d] = %d not found in superset",
3924                      i, subset_rm->rm_cpumap[i]));
3925         }
3926 #endif
3927 }
3928
3929 int
3930 if_ringmap_count(const struct if_ringmap *rm)
3931 {
3932
3933         return (rm->rm_cnt);
3934 }
3935
3936 int
3937 if_ringmap_cpumap(const struct if_ringmap *rm, int ring)
3938 {
3939
3940         KASSERT(ring >= 0 && ring < rm->rm_cnt, ("invalid ring %d", ring));
3941         return (rm->rm_cpumap[ring]);
3942 }
3943
3944 void
3945 if_ringmap_rdrtable(const struct if_ringmap *rm, int table[], int table_nent)
3946 {
3947         int i, grid_idx, grid_cnt, patch_off, patch_cnt, ncopy;
3948
3949         KASSERT(table_nent > 0 && (table_nent & NETISR_CPUMASK) == 0,
3950             ("invalid redirect table entries %d", table_nent));
3951
3952         grid_idx = 0;
3953         for (i = 0; i < NETISR_CPUMAX; ++i) {
3954                 table[i] = grid_idx++ % rm->rm_cnt;
3955
3956                 if (grid_idx == rm->rm_grid)
3957                         grid_idx = 0;
3958         }
3959
3960         /*
3961          * Make the ring distributed more evenly for the remainder
3962          * of each grid.
3963          *
3964          * e.g. 12 netisrs, rm contains 8 rings.
3965          *
3966          * Redirect table before:
3967          *
3968          *  0  1  2  3  4  5  6  7  0  1  2  3  0  1  2  3
3969          *  4  5  6  7  0  1  2  3  0  1  2  3  4  5  6  7
3970          *  0  1  2  3  0  1  2  3  4  5  6  7  0  1  2  3
3971          *  ....
3972          *
3973          * Redirect table after being patched (pX, patched entries):
3974          *
3975          *  0  1  2  3  4  5  6  7 p0 p1 p2 p3  0  1  2  3
3976          *  4  5  6  7 p4 p5 p6 p7  0  1  2  3  4  5  6  7
3977          * p0 p1 p2 p3  0  1  2  3  4  5  6  7 p4 p5 p6 p7
3978          *  ....
3979          */
3980         patch_cnt = rm->rm_grid % rm->rm_cnt;
3981         if (patch_cnt == 0)
3982                 goto done;
3983         patch_off = rm->rm_grid - (rm->rm_grid % rm->rm_cnt);
3984
3985         grid_cnt = roundup(NETISR_CPUMAX, rm->rm_grid) / rm->rm_grid;
3986         grid_idx = 0;
3987         for (i = 0; i < grid_cnt; ++i) {
3988                 int j;
3989
3990                 for (j = 0; j < patch_cnt; ++j) {
3991                         int fix_idx;
3992
3993                         fix_idx = (i * rm->rm_grid) + patch_off + j;
3994                         if (fix_idx >= NETISR_CPUMAX)
3995                                 goto done;
3996                         table[fix_idx] = grid_idx++ % rm->rm_cnt;
3997                 }
3998         }
3999 done:
4000         /*
4001          * If the device supports larger redirect table, duplicate
4002          * the first NETISR_CPUMAX entries to the rest of the table,
4003          * so that it matches upper layer's expectation:
4004          * (hash & NETISR_CPUMASK) % netisr_ncpus
4005          */
4006         ncopy = table_nent / NETISR_CPUMAX;
4007         for (i = 1; i < ncopy; ++i) {
4008                 memcpy(&table[i * NETISR_CPUMAX], table,
4009                     NETISR_CPUMAX * sizeof(table[0]));
4010         }
4011         if (if_ringmap_dumprdr) {
4012                 for (i = 0; i < table_nent; ++i) {
4013                         if (i != 0 && i % 16 == 0)
4014                                 kprintf("\n");
4015                         kprintf("%03d ", table[i]);
4016                 }
4017                 kprintf("\n");
4018         }
4019 }
4020
4021 int
4022 if_ringmap_cpumap_sysctl(SYSCTL_HANDLER_ARGS)
4023 {
4024         struct if_ringmap *rm = arg1;
4025         int i, error = 0;
4026
4027         for (i = 0; i < rm->rm_cnt; ++i) {
4028                 int cpu = rm->rm_cpumap[i];
4029
4030                 error = SYSCTL_OUT(req, &cpu, sizeof(cpu));
4031                 if (error)
4032                         break;
4033         }
4034         return (error);
4035 }