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