1 /* 2 * Copyright (c) 1980, 1986, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)if.c 8.3 (Berkeley) 1/4/94 30 * $FreeBSD: src/sys/net/if.c,v 1.185 2004/03/13 02:35:03 brooks Exp $ 31 */ 32 33 #include "opt_inet6.h" 34 #include "opt_inet.h" 35 #include "opt_ifpoll.h" 36 37 #include <sys/param.h> 38 #include <sys/malloc.h> 39 #include <sys/mbuf.h> 40 #include <sys/systm.h> 41 #include <sys/proc.h> 42 #include <sys/priv.h> 43 #include <sys/protosw.h> 44 #include <sys/socket.h> 45 #include <sys/socketvar.h> 46 #include <sys/socketops.h> 47 #include <sys/kernel.h> 48 #include <sys/ktr.h> 49 #include <sys/mutex.h> 50 #include <sys/lock.h> 51 #include <sys/sockio.h> 52 #include <sys/syslog.h> 53 #include <sys/sysctl.h> 54 #include <sys/domain.h> 55 #include <sys/thread.h> 56 #include <sys/serialize.h> 57 #include <sys/bus.h> 58 #include <sys/jail.h> 59 60 #include <sys/thread2.h> 61 #include <sys/msgport2.h> 62 #include <sys/mutex2.h> 63 64 #include <net/if.h> 65 #include <net/if_arp.h> 66 #include <net/if_dl.h> 67 #include <net/if_types.h> 68 #include <net/if_var.h> 69 #include <net/if_ringmap.h> 70 #include <net/ifq_var.h> 71 #include <net/radix.h> 72 #include <net/route.h> 73 #include <net/if_clone.h> 74 #include <net/netisr2.h> 75 #include <net/netmsg2.h> 76 77 #include <machine/atomic.h> 78 #include <machine/stdarg.h> 79 #include <machine/smp.h> 80 81 #if defined(INET) || defined(INET6) 82 #include <netinet/in.h> 83 #include <netinet/in_var.h> 84 #include <netinet/if_ether.h> 85 #ifdef INET6 86 #include <netinet6/in6_var.h> 87 #include <netinet6/in6_ifattach.h> 88 #endif /* INET6 */ 89 #endif /* INET || INET6 */ 90 91 struct netmsg_ifaddr { 92 struct netmsg_base base; 93 struct ifaddr *ifa; 94 struct ifnet *ifp; 95 int tail; 96 }; 97 98 struct ifsubq_stage_head { 99 TAILQ_HEAD(, ifsubq_stage) stg_head; 100 } __cachealign; 101 102 struct if_ringmap { 103 int rm_cnt; 104 int rm_grid; 105 int rm_cpumap[]; 106 }; 107 108 #define RINGMAP_FLAG_NONE 0x0 109 #define RINGMAP_FLAG_POWEROF2 0x1 110 111 /* 112 * System initialization 113 */ 114 static void if_attachdomain(void *); 115 static void if_attachdomain1(struct ifnet *); 116 static int ifconf(u_long, caddr_t, struct ucred *); 117 static void ifinit(void *); 118 static void ifnetinit(void *); 119 static void if_slowtimo(void *); 120 static void link_rtrequest(int, struct rtentry *); 121 static int if_rtdel(struct radix_node *, void *); 122 static void if_slowtimo_dispatch(netmsg_t); 123 124 /* Helper functions */ 125 static void ifsq_watchdog_reset(struct ifsubq_watchdog *); 126 static int if_delmulti_serialized(struct ifnet *, struct sockaddr *); 127 static struct ifnet_array *ifnet_array_alloc(int); 128 static void ifnet_array_free(struct ifnet_array *); 129 static struct ifnet_array *ifnet_array_add(struct ifnet *, 130 const struct ifnet_array *); 131 static struct ifnet_array *ifnet_array_del(struct ifnet *, 132 const struct ifnet_array *); 133 static struct ifg_group *if_creategroup(const char *); 134 static int if_destroygroup(struct ifg_group *); 135 static int if_delgroup_locked(struct ifnet *, const char *); 136 static int if_getgroups(struct ifgroupreq *, struct ifnet *); 137 static int if_getgroupmembers(struct ifgroupreq *); 138 139 #ifdef INET6 140 /* 141 * XXX: declare here to avoid to include many inet6 related files.. 142 * should be more generalized? 143 */ 144 extern void nd6_setmtu(struct ifnet *); 145 #endif 146 147 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers"); 148 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management"); 149 SYSCTL_NODE(_net_link, OID_AUTO, ringmap, CTLFLAG_RW, 0, "link ringmap"); 150 151 static int ifsq_stage_cntmax = 16; 152 TUNABLE_INT("net.link.stage_cntmax", &ifsq_stage_cntmax); 153 SYSCTL_INT(_net_link, OID_AUTO, stage_cntmax, CTLFLAG_RW, 154 &ifsq_stage_cntmax, 0, "ifq staging packet count max"); 155 156 static int if_stats_compat = 0; 157 SYSCTL_INT(_net_link, OID_AUTO, stats_compat, CTLFLAG_RW, 158 &if_stats_compat, 0, "Compat the old ifnet stats"); 159 160 static int if_ringmap_dumprdr = 0; 161 SYSCTL_INT(_net_link_ringmap, OID_AUTO, dump_rdr, CTLFLAG_RW, 162 &if_ringmap_dumprdr, 0, "dump redirect table"); 163 164 SYSINIT(interfaces, SI_SUB_PROTO_IF, SI_ORDER_FIRST, ifinit, NULL); 165 SYSINIT(ifnet, SI_SUB_PRE_DRIVERS, SI_ORDER_ANY, ifnetinit, NULL); 166 167 static if_com_alloc_t *if_com_alloc[256]; 168 static if_com_free_t *if_com_free[256]; 169 170 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address"); 171 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address"); 172 MALLOC_DEFINE(M_IFNET, "ifnet", "interface structure"); 173 174 int ifqmaxlen = IFQ_MAXLEN; 175 struct ifnethead ifnet = TAILQ_HEAD_INITIALIZER(ifnet); 176 struct ifgrouphead ifg_head = TAILQ_HEAD_INITIALIZER(ifg_head); 177 static struct lock ifgroup_lock; 178 179 static struct ifnet_array ifnet_array0; 180 static struct ifnet_array *ifnet_array = &ifnet_array0; 181 182 static struct callout if_slowtimo_timer; 183 static struct netmsg_base if_slowtimo_netmsg; 184 185 int if_index = 0; 186 struct ifnet **ifindex2ifnet = NULL; 187 static struct mtx ifnet_mtx = MTX_INITIALIZER("ifnet"); 188 189 static struct ifsubq_stage_head ifsubq_stage_heads[MAXCPU]; 190 191 #ifdef notyet 192 #define IFQ_KTR_STRING "ifq=%p" 193 #define IFQ_KTR_ARGS struct ifaltq *ifq 194 #ifndef KTR_IFQ 195 #define KTR_IFQ KTR_ALL 196 #endif 197 KTR_INFO_MASTER(ifq); 198 KTR_INFO(KTR_IFQ, ifq, enqueue, 0, IFQ_KTR_STRING, IFQ_KTR_ARGS); 199 KTR_INFO(KTR_IFQ, ifq, dequeue, 1, IFQ_KTR_STRING, IFQ_KTR_ARGS); 200 #define logifq(name, arg) KTR_LOG(ifq_ ## name, arg) 201 202 #define IF_START_KTR_STRING "ifp=%p" 203 #define IF_START_KTR_ARGS struct ifnet *ifp 204 #ifndef KTR_IF_START 205 #define KTR_IF_START KTR_ALL 206 #endif 207 KTR_INFO_MASTER(if_start); 208 KTR_INFO(KTR_IF_START, if_start, run, 0, 209 IF_START_KTR_STRING, IF_START_KTR_ARGS); 210 KTR_INFO(KTR_IF_START, if_start, sched, 1, 211 IF_START_KTR_STRING, IF_START_KTR_ARGS); 212 KTR_INFO(KTR_IF_START, if_start, avoid, 2, 213 IF_START_KTR_STRING, IF_START_KTR_ARGS); 214 KTR_INFO(KTR_IF_START, if_start, contend_sched, 3, 215 IF_START_KTR_STRING, IF_START_KTR_ARGS); 216 KTR_INFO(KTR_IF_START, if_start, chase_sched, 4, 217 IF_START_KTR_STRING, IF_START_KTR_ARGS); 218 #define logifstart(name, arg) KTR_LOG(if_start_ ## name, arg) 219 #endif /* notyet */ 220 221 /* 222 * Network interface utility routines. 223 * 224 * Routines with ifa_ifwith* names take sockaddr *'s as 225 * parameters. 226 */ 227 /* ARGSUSED */ 228 static void 229 ifinit(void *dummy) 230 { 231 lockinit(&ifgroup_lock, "ifgroup", 0, 0); 232 233 callout_init_mp(&if_slowtimo_timer); 234 netmsg_init(&if_slowtimo_netmsg, NULL, &netisr_adone_rport, 235 MSGF_PRIORITY, if_slowtimo_dispatch); 236 237 /* Start if_slowtimo */ 238 lwkt_sendmsg(netisr_cpuport(0), &if_slowtimo_netmsg.lmsg); 239 } 240 241 static void 242 ifsq_ifstart_ipifunc(void *arg) 243 { 244 struct ifaltq_subque *ifsq = arg; 245 struct lwkt_msg *lmsg = ifsq_get_ifstart_lmsg(ifsq, mycpuid); 246 247 crit_enter(); 248 if (lmsg->ms_flags & MSGF_DONE) 249 lwkt_sendmsg_oncpu(netisr_cpuport(mycpuid), lmsg); 250 crit_exit(); 251 } 252 253 static __inline void 254 ifsq_stage_remove(struct ifsubq_stage_head *head, struct ifsubq_stage *stage) 255 { 256 KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED); 257 TAILQ_REMOVE(&head->stg_head, stage, stg_link); 258 stage->stg_flags &= ~(IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED); 259 stage->stg_cnt = 0; 260 stage->stg_len = 0; 261 } 262 263 static __inline void 264 ifsq_stage_insert(struct ifsubq_stage_head *head, struct ifsubq_stage *stage) 265 { 266 KKASSERT((stage->stg_flags & 267 (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0); 268 stage->stg_flags |= IFSQ_STAGE_FLAG_QUED; 269 TAILQ_INSERT_TAIL(&head->stg_head, stage, stg_link); 270 } 271 272 /* 273 * Schedule ifnet.if_start on the subqueue owner CPU 274 */ 275 static void 276 ifsq_ifstart_schedule(struct ifaltq_subque *ifsq, int force) 277 { 278 int cpu; 279 280 if (!force && curthread->td_type == TD_TYPE_NETISR && 281 ifsq_stage_cntmax > 0) { 282 struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid); 283 284 stage->stg_cnt = 0; 285 stage->stg_len = 0; 286 if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0) 287 ifsq_stage_insert(&ifsubq_stage_heads[mycpuid], stage); 288 stage->stg_flags |= IFSQ_STAGE_FLAG_SCHED; 289 return; 290 } 291 292 cpu = ifsq_get_cpuid(ifsq); 293 if (cpu != mycpuid) 294 lwkt_send_ipiq(globaldata_find(cpu), ifsq_ifstart_ipifunc, ifsq); 295 else 296 ifsq_ifstart_ipifunc(ifsq); 297 } 298 299 /* 300 * NOTE: 301 * This function will release ifnet.if_start subqueue interlock, 302 * if ifnet.if_start for the subqueue does not need to be scheduled 303 */ 304 static __inline int 305 ifsq_ifstart_need_schedule(struct ifaltq_subque *ifsq, int running) 306 { 307 if (!running || ifsq_is_empty(ifsq) 308 #ifdef ALTQ 309 || ifsq->ifsq_altq->altq_tbr != NULL 310 #endif 311 ) { 312 ALTQ_SQ_LOCK(ifsq); 313 /* 314 * ifnet.if_start subqueue interlock is released, if: 315 * 1) Hardware can not take any packets, due to 316 * o interface is marked down 317 * o hardware queue is full (ifsq_is_oactive) 318 * Under the second situation, hardware interrupt 319 * or polling(4) will call/schedule ifnet.if_start 320 * on the subqueue when hardware queue is ready 321 * 2) There is no packet in the subqueue. 322 * Further ifq_dispatch or ifq_handoff will call/ 323 * schedule ifnet.if_start on the subqueue. 324 * 3) TBR is used and it does not allow further 325 * dequeueing. 326 * TBR callout will call ifnet.if_start on the 327 * subqueue. 328 */ 329 if (!running || !ifsq_data_ready(ifsq)) { 330 ifsq_clr_started(ifsq); 331 ALTQ_SQ_UNLOCK(ifsq); 332 return 0; 333 } 334 ALTQ_SQ_UNLOCK(ifsq); 335 } 336 return 1; 337 } 338 339 static void 340 ifsq_ifstart_dispatch(netmsg_t msg) 341 { 342 struct lwkt_msg *lmsg = &msg->base.lmsg; 343 struct ifaltq_subque *ifsq = lmsg->u.ms_resultp; 344 struct ifnet *ifp = ifsq_get_ifp(ifsq); 345 struct globaldata *gd = mycpu; 346 int running = 0, need_sched; 347 348 crit_enter_gd(gd); 349 350 lwkt_replymsg(lmsg, 0); /* reply ASAP */ 351 352 if (gd->gd_cpuid != ifsq_get_cpuid(ifsq)) { 353 /* 354 * We need to chase the subqueue owner CPU change. 355 */ 356 ifsq_ifstart_schedule(ifsq, 1); 357 crit_exit_gd(gd); 358 return; 359 } 360 361 ifsq_serialize_hw(ifsq); 362 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) { 363 ifp->if_start(ifp, ifsq); 364 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) 365 running = 1; 366 } 367 need_sched = ifsq_ifstart_need_schedule(ifsq, running); 368 ifsq_deserialize_hw(ifsq); 369 370 if (need_sched) { 371 /* 372 * More data need to be transmitted, ifnet.if_start is 373 * scheduled on the subqueue owner CPU, and we keep going. 374 * NOTE: ifnet.if_start subqueue interlock is not released. 375 */ 376 ifsq_ifstart_schedule(ifsq, 0); 377 } 378 379 crit_exit_gd(gd); 380 } 381 382 /* Device driver ifnet.if_start helper function */ 383 void 384 ifsq_devstart(struct ifaltq_subque *ifsq) 385 { 386 struct ifnet *ifp = ifsq_get_ifp(ifsq); 387 int running = 0; 388 389 ASSERT_ALTQ_SQ_SERIALIZED_HW(ifsq); 390 391 ALTQ_SQ_LOCK(ifsq); 392 if (ifsq_is_started(ifsq) || !ifsq_data_ready(ifsq)) { 393 ALTQ_SQ_UNLOCK(ifsq); 394 return; 395 } 396 ifsq_set_started(ifsq); 397 ALTQ_SQ_UNLOCK(ifsq); 398 399 ifp->if_start(ifp, ifsq); 400 401 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) 402 running = 1; 403 404 if (ifsq_ifstart_need_schedule(ifsq, running)) { 405 /* 406 * More data need to be transmitted, ifnet.if_start is 407 * scheduled on ifnet's CPU, and we keep going. 408 * NOTE: ifnet.if_start interlock is not released. 409 */ 410 ifsq_ifstart_schedule(ifsq, 0); 411 } 412 } 413 414 void 415 if_devstart(struct ifnet *ifp) 416 { 417 ifsq_devstart(ifq_get_subq_default(&ifp->if_snd)); 418 } 419 420 /* Device driver ifnet.if_start schedule helper function */ 421 void 422 ifsq_devstart_sched(struct ifaltq_subque *ifsq) 423 { 424 ifsq_ifstart_schedule(ifsq, 1); 425 } 426 427 void 428 if_devstart_sched(struct ifnet *ifp) 429 { 430 ifsq_devstart_sched(ifq_get_subq_default(&ifp->if_snd)); 431 } 432 433 static void 434 if_default_serialize(struct ifnet *ifp, enum ifnet_serialize slz __unused) 435 { 436 lwkt_serialize_enter(ifp->if_serializer); 437 } 438 439 static void 440 if_default_deserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused) 441 { 442 lwkt_serialize_exit(ifp->if_serializer); 443 } 444 445 static int 446 if_default_tryserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused) 447 { 448 return lwkt_serialize_try(ifp->if_serializer); 449 } 450 451 #ifdef INVARIANTS 452 static void 453 if_default_serialize_assert(struct ifnet *ifp, 454 enum ifnet_serialize slz __unused, 455 boolean_t serialized) 456 { 457 if (serialized) 458 ASSERT_SERIALIZED(ifp->if_serializer); 459 else 460 ASSERT_NOT_SERIALIZED(ifp->if_serializer); 461 } 462 #endif 463 464 /* 465 * Attach an interface to the list of "active" interfaces. 466 * 467 * The serializer is optional. 468 */ 469 void 470 if_attach(struct ifnet *ifp, lwkt_serialize_t serializer) 471 { 472 unsigned socksize; 473 int namelen, masklen; 474 struct sockaddr_dl *sdl, *sdl_addr; 475 struct ifaddr *ifa; 476 struct ifaltq *ifq; 477 struct ifnet **old_ifindex2ifnet = NULL; 478 struct ifnet_array *old_ifnet_array; 479 int i, q, qlen; 480 char qlenname[64]; 481 482 static int if_indexlim = 8; 483 484 if (ifp->if_serialize != NULL) { 485 KASSERT(ifp->if_deserialize != NULL && 486 ifp->if_tryserialize != NULL && 487 ifp->if_serialize_assert != NULL, 488 ("serialize functions are partially setup")); 489 490 /* 491 * If the device supplies serialize functions, 492 * then clear if_serializer to catch any invalid 493 * usage of this field. 494 */ 495 KASSERT(serializer == NULL, 496 ("both serialize functions and default serializer " 497 "are supplied")); 498 ifp->if_serializer = NULL; 499 } else { 500 KASSERT(ifp->if_deserialize == NULL && 501 ifp->if_tryserialize == NULL && 502 ifp->if_serialize_assert == NULL, 503 ("serialize functions are partially setup")); 504 ifp->if_serialize = if_default_serialize; 505 ifp->if_deserialize = if_default_deserialize; 506 ifp->if_tryserialize = if_default_tryserialize; 507 #ifdef INVARIANTS 508 ifp->if_serialize_assert = if_default_serialize_assert; 509 #endif 510 511 /* 512 * The serializer can be passed in from the device, 513 * allowing the same serializer to be used for both 514 * the interrupt interlock and the device queue. 515 * If not specified, the netif structure will use an 516 * embedded serializer. 517 */ 518 if (serializer == NULL) { 519 serializer = &ifp->if_default_serializer; 520 lwkt_serialize_init(serializer); 521 } 522 ifp->if_serializer = serializer; 523 } 524 525 /* 526 * Make if_addrhead available on all CPUs, since they 527 * could be accessed by any threads. 528 */ 529 ifp->if_addrheads = kmalloc(ncpus * sizeof(struct ifaddrhead), 530 M_IFADDR, M_WAITOK | M_ZERO); 531 for (i = 0; i < ncpus; ++i) 532 TAILQ_INIT(&ifp->if_addrheads[i]); 533 534 TAILQ_INIT(&ifp->if_multiaddrs); 535 TAILQ_INIT(&ifp->if_groups); 536 getmicrotime(&ifp->if_lastchange); 537 if_addgroup(ifp, IFG_ALL); 538 539 /* 540 * create a Link Level name for this device 541 */ 542 namelen = strlen(ifp->if_xname); 543 masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + namelen; 544 socksize = masklen + ifp->if_addrlen; 545 if (socksize < sizeof(*sdl)) 546 socksize = sizeof(*sdl); 547 socksize = RT_ROUNDUP(socksize); 548 ifa = ifa_create(sizeof(struct ifaddr) + 2 * socksize); 549 sdl = sdl_addr = (struct sockaddr_dl *)(ifa + 1); 550 sdl->sdl_len = socksize; 551 sdl->sdl_family = AF_LINK; 552 bcopy(ifp->if_xname, sdl->sdl_data, namelen); 553 sdl->sdl_nlen = namelen; 554 sdl->sdl_type = ifp->if_type; 555 ifp->if_lladdr = ifa; 556 ifa->ifa_ifp = ifp; 557 ifa->ifa_rtrequest = link_rtrequest; 558 ifa->ifa_addr = (struct sockaddr *)sdl; 559 sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl); 560 ifa->ifa_netmask = (struct sockaddr *)sdl; 561 sdl->sdl_len = masklen; 562 while (namelen != 0) 563 sdl->sdl_data[--namelen] = 0xff; 564 ifa_iflink(ifa, ifp, 0 /* Insert head */); 565 566 /* 567 * Make if_data available on all CPUs, since they could 568 * be updated by hardware interrupt routing, which could 569 * be bound to any CPU. 570 */ 571 ifp->if_data_pcpu = kmalloc(ncpus * sizeof(struct ifdata_pcpu), 572 M_DEVBUF, 573 M_WAITOK | M_ZERO | M_CACHEALIGN); 574 575 if (ifp->if_mapsubq == NULL) 576 ifp->if_mapsubq = ifq_mapsubq_default; 577 578 ifq = &ifp->if_snd; 579 ifq->altq_type = 0; 580 ifq->altq_disc = NULL; 581 ifq->altq_flags &= ALTQF_CANTCHANGE; 582 ifq->altq_tbr = NULL; 583 ifq->altq_ifp = ifp; 584 585 if (ifq->altq_subq_cnt <= 0) 586 ifq->altq_subq_cnt = 1; 587 ifq->altq_subq = 588 kmalloc(ifq->altq_subq_cnt * sizeof(struct ifaltq_subque), 589 M_DEVBUF, 590 M_WAITOK | M_ZERO | M_CACHEALIGN); 591 592 if (ifq->altq_maxlen == 0) { 593 if_printf(ifp, "driver didn't set altq_maxlen\n"); 594 ifq_set_maxlen(ifq, ifqmaxlen); 595 } 596 597 /* Allow user to override driver's setting. */ 598 ksnprintf(qlenname, sizeof(qlenname), "net.%s.qlenmax", ifp->if_xname); 599 qlen = -1; 600 TUNABLE_INT_FETCH(qlenname, &qlen); 601 if (qlen > 0) { 602 if_printf(ifp, "qlenmax -> %d\n", qlen); 603 ifq_set_maxlen(ifq, qlen); 604 } 605 606 for (q = 0; q < ifq->altq_subq_cnt; ++q) { 607 struct ifaltq_subque *ifsq = &ifq->altq_subq[q]; 608 609 ALTQ_SQ_LOCK_INIT(ifsq); 610 ifsq->ifsq_index = q; 611 612 ifsq->ifsq_altq = ifq; 613 ifsq->ifsq_ifp = ifp; 614 615 ifsq->ifsq_maxlen = ifq->altq_maxlen; 616 ifsq->ifsq_maxbcnt = ifsq->ifsq_maxlen * MCLBYTES; 617 ifsq->ifsq_prepended = NULL; 618 ifsq->ifsq_started = 0; 619 ifsq->ifsq_hw_oactive = 0; 620 ifsq_set_cpuid(ifsq, 0); 621 if (ifp->if_serializer != NULL) 622 ifsq_set_hw_serialize(ifsq, ifp->if_serializer); 623 624 /* XXX: netisr_ncpus */ 625 ifsq->ifsq_stage = 626 kmalloc(ncpus * sizeof(struct ifsubq_stage), 627 M_DEVBUF, 628 M_WAITOK | M_ZERO | M_CACHEALIGN); 629 for (i = 0; i < ncpus; ++i) 630 ifsq->ifsq_stage[i].stg_subq = ifsq; 631 632 /* 633 * Allocate one if_start message for each CPU, since 634 * the hardware TX ring could be assigned to any CPU. 635 * 636 * NOTE: 637 * If the hardware TX ring polling CPU and the hardware 638 * TX ring interrupt CPU are same, one if_start message 639 * should be enough. 640 */ 641 ifsq->ifsq_ifstart_nmsg = 642 kmalloc(ncpus * sizeof(struct netmsg_base), 643 M_LWKTMSG, M_WAITOK); 644 for (i = 0; i < ncpus; ++i) { 645 netmsg_init(&ifsq->ifsq_ifstart_nmsg[i], NULL, 646 &netisr_adone_rport, 0, ifsq_ifstart_dispatch); 647 ifsq->ifsq_ifstart_nmsg[i].lmsg.u.ms_resultp = ifsq; 648 } 649 } 650 ifq_set_classic(ifq); 651 652 /* 653 * Increase mbuf cluster/jcluster limits for the mbufs that 654 * could sit on the device queues for quite some time. 655 */ 656 if (ifp->if_nmbclusters > 0) 657 mcl_inclimit(ifp->if_nmbclusters); 658 if (ifp->if_nmbjclusters > 0) 659 mjcl_inclimit(ifp->if_nmbjclusters); 660 661 /* 662 * Install this ifp into ifindex2inet, ifnet queue and ifnet 663 * array after it is setup. 664 * 665 * Protect ifindex2ifnet, ifnet queue and ifnet array changes 666 * by ifnet lock, so that non-netisr threads could get a 667 * consistent view. 668 */ 669 ifnet_lock(); 670 671 /* Don't update if_index until ifindex2ifnet is setup */ 672 ifp->if_index = if_index + 1; 673 sdl_addr->sdl_index = ifp->if_index; 674 675 /* 676 * Install this ifp into ifindex2ifnet 677 */ 678 if (ifindex2ifnet == NULL || ifp->if_index >= if_indexlim) { 679 unsigned int n; 680 struct ifnet **q; 681 682 /* 683 * Grow ifindex2ifnet 684 */ 685 if_indexlim <<= 1; 686 n = if_indexlim * sizeof(*q); 687 q = kmalloc(n, M_IFADDR, M_WAITOK | M_ZERO); 688 if (ifindex2ifnet != NULL) { 689 bcopy(ifindex2ifnet, q, n/2); 690 /* Free old ifindex2ifnet after sync all netisrs */ 691 old_ifindex2ifnet = ifindex2ifnet; 692 } 693 ifindex2ifnet = q; 694 } 695 ifindex2ifnet[ifp->if_index] = ifp; 696 /* 697 * Update if_index after this ifp is installed into ifindex2ifnet, 698 * so that netisrs could get a consistent view of ifindex2ifnet. 699 */ 700 cpu_sfence(); 701 if_index = ifp->if_index; 702 703 /* 704 * Install this ifp into ifnet array. 705 */ 706 /* Free old ifnet array after sync all netisrs */ 707 old_ifnet_array = ifnet_array; 708 ifnet_array = ifnet_array_add(ifp, old_ifnet_array); 709 710 /* 711 * Install this ifp into ifnet queue. 712 */ 713 TAILQ_INSERT_TAIL(&ifnetlist, ifp, if_link); 714 715 ifnet_unlock(); 716 717 /* 718 * Sync all netisrs so that the old ifindex2ifnet and ifnet array 719 * are no longer accessed and we can free them safely later on. 720 */ 721 netmsg_service_sync(); 722 if (old_ifindex2ifnet != NULL) 723 kfree(old_ifindex2ifnet, M_IFADDR); 724 ifnet_array_free(old_ifnet_array); 725 726 if (!SLIST_EMPTY(&domains)) 727 if_attachdomain1(ifp); 728 729 /* Announce the interface. */ 730 EVENTHANDLER_INVOKE(ifnet_attach_event, ifp); 731 devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL); 732 rt_ifannouncemsg(ifp, IFAN_ARRIVAL); 733 } 734 735 static void 736 if_attachdomain(void *dummy) 737 { 738 struct ifnet *ifp; 739 740 ifnet_lock(); 741 TAILQ_FOREACH(ifp, &ifnetlist, if_list) 742 if_attachdomain1(ifp); 743 ifnet_unlock(); 744 } 745 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_FIRST, 746 if_attachdomain, NULL); 747 748 static void 749 if_attachdomain1(struct ifnet *ifp) 750 { 751 struct domain *dp; 752 753 crit_enter(); 754 755 /* address family dependent data region */ 756 bzero(ifp->if_afdata, sizeof(ifp->if_afdata)); 757 SLIST_FOREACH(dp, &domains, dom_next) 758 if (dp->dom_ifattach) 759 ifp->if_afdata[dp->dom_family] = 760 (*dp->dom_ifattach)(ifp); 761 crit_exit(); 762 } 763 764 /* 765 * Purge all addresses whose type is _not_ AF_LINK 766 */ 767 static void 768 if_purgeaddrs_nolink_dispatch(netmsg_t nmsg) 769 { 770 struct ifnet *ifp = nmsg->lmsg.u.ms_resultp; 771 struct ifaddr_container *ifac, *next; 772 773 ASSERT_NETISR0; 774 775 /* 776 * The ifaddr processing in the following loop will block, 777 * however, this function is called in netisr0, in which 778 * ifaddr list changes happen, so we don't care about the 779 * blockness of the ifaddr processing here. 780 */ 781 TAILQ_FOREACH_MUTABLE(ifac, &ifp->if_addrheads[mycpuid], 782 ifa_link, next) { 783 struct ifaddr *ifa = ifac->ifa; 784 785 /* Ignore marker */ 786 if (ifa->ifa_addr->sa_family == AF_UNSPEC) 787 continue; 788 789 /* Leave link ifaddr as it is */ 790 if (ifa->ifa_addr->sa_family == AF_LINK) 791 continue; 792 #ifdef INET 793 /* XXX: Ugly!! ad hoc just for INET */ 794 if (ifa->ifa_addr->sa_family == AF_INET) { 795 struct ifaliasreq ifr; 796 struct sockaddr_in saved_addr, saved_dst; 797 #ifdef IFADDR_DEBUG_VERBOSE 798 int i; 799 800 kprintf("purge in4 addr %p: ", ifa); 801 for (i = 0; i < ncpus; ++i) { 802 kprintf("%d ", 803 ifa->ifa_containers[i].ifa_refcnt); 804 } 805 kprintf("\n"); 806 #endif 807 808 /* Save information for panic. */ 809 memcpy(&saved_addr, ifa->ifa_addr, sizeof(saved_addr)); 810 if (ifa->ifa_dstaddr != NULL) { 811 memcpy(&saved_dst, ifa->ifa_dstaddr, 812 sizeof(saved_dst)); 813 } else { 814 memset(&saved_dst, 0, sizeof(saved_dst)); 815 } 816 817 bzero(&ifr, sizeof ifr); 818 ifr.ifra_addr = *ifa->ifa_addr; 819 if (ifa->ifa_dstaddr) 820 ifr.ifra_broadaddr = *ifa->ifa_dstaddr; 821 if (in_control(SIOCDIFADDR, (caddr_t)&ifr, ifp, 822 NULL) == 0) 823 continue; 824 825 /* MUST NOT HAPPEN */ 826 panic("%s: in_control failed %x, dst %x", ifp->if_xname, 827 ntohl(saved_addr.sin_addr.s_addr), 828 ntohl(saved_dst.sin_addr.s_addr)); 829 } 830 #endif /* INET */ 831 #ifdef INET6 832 if (ifa->ifa_addr->sa_family == AF_INET6) { 833 #ifdef IFADDR_DEBUG_VERBOSE 834 int i; 835 836 kprintf("purge in6 addr %p: ", ifa); 837 for (i = 0; i < ncpus; ++i) { 838 kprintf("%d ", 839 ifa->ifa_containers[i].ifa_refcnt); 840 } 841 kprintf("\n"); 842 #endif 843 844 in6_purgeaddr(ifa); 845 /* ifp_addrhead is already updated */ 846 continue; 847 } 848 #endif /* INET6 */ 849 if_printf(ifp, "destroy ifaddr family %d\n", 850 ifa->ifa_addr->sa_family); 851 ifa_ifunlink(ifa, ifp); 852 ifa_destroy(ifa); 853 } 854 855 netisr_replymsg(&nmsg->base, 0); 856 } 857 858 void 859 if_purgeaddrs_nolink(struct ifnet *ifp) 860 { 861 struct netmsg_base nmsg; 862 863 netmsg_init(&nmsg, NULL, &curthread->td_msgport, 0, 864 if_purgeaddrs_nolink_dispatch); 865 nmsg.lmsg.u.ms_resultp = ifp; 866 netisr_domsg(&nmsg, 0); 867 } 868 869 static void 870 ifq_stage_detach_handler(netmsg_t nmsg) 871 { 872 struct ifaltq *ifq = nmsg->lmsg.u.ms_resultp; 873 int q; 874 875 for (q = 0; q < ifq->altq_subq_cnt; ++q) { 876 struct ifaltq_subque *ifsq = &ifq->altq_subq[q]; 877 struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid); 878 879 if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED) 880 ifsq_stage_remove(&ifsubq_stage_heads[mycpuid], stage); 881 } 882 lwkt_replymsg(&nmsg->lmsg, 0); 883 } 884 885 static void 886 ifq_stage_detach(struct ifaltq *ifq) 887 { 888 struct netmsg_base base; 889 int cpu; 890 891 netmsg_init(&base, NULL, &curthread->td_msgport, 0, 892 ifq_stage_detach_handler); 893 base.lmsg.u.ms_resultp = ifq; 894 895 /* XXX netisr_ncpus */ 896 for (cpu = 0; cpu < ncpus; ++cpu) 897 lwkt_domsg(netisr_cpuport(cpu), &base.lmsg, 0); 898 } 899 900 struct netmsg_if_rtdel { 901 struct netmsg_base base; 902 struct ifnet *ifp; 903 }; 904 905 static void 906 if_rtdel_dispatch(netmsg_t msg) 907 { 908 struct netmsg_if_rtdel *rmsg = (void *)msg; 909 int i, cpu; 910 911 cpu = mycpuid; 912 ASSERT_NETISR_NCPUS(cpu); 913 914 for (i = 1; i <= AF_MAX; i++) { 915 struct radix_node_head *rnh; 916 917 if ((rnh = rt_tables[cpu][i]) == NULL) 918 continue; 919 rnh->rnh_walktree(rnh, if_rtdel, rmsg->ifp); 920 } 921 netisr_forwardmsg(&msg->base, cpu + 1); 922 } 923 924 /* 925 * Detach an interface, removing it from the 926 * list of "active" interfaces. 927 */ 928 void 929 if_detach(struct ifnet *ifp) 930 { 931 struct ifnet_array *old_ifnet_array; 932 struct ifg_list *ifgl; 933 struct netmsg_if_rtdel msg; 934 struct domain *dp; 935 int q; 936 937 /* Announce that the interface is gone. */ 938 EVENTHANDLER_INVOKE(ifnet_detach_event, ifp); 939 rt_ifannouncemsg(ifp, IFAN_DEPARTURE); 940 devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL); 941 942 /* 943 * Remove this ifp from ifindex2inet, ifnet queue and ifnet 944 * array before it is whacked. 945 * 946 * Protect ifindex2ifnet, ifnet queue and ifnet array changes 947 * by ifnet lock, so that non-netisr threads could get a 948 * consistent view. 949 */ 950 ifnet_lock(); 951 952 /* 953 * Remove this ifp from ifindex2ifnet and maybe decrement if_index. 954 */ 955 ifindex2ifnet[ifp->if_index] = NULL; 956 while (if_index > 0 && ifindex2ifnet[if_index] == NULL) 957 if_index--; 958 959 /* 960 * Remove this ifp from ifnet queue. 961 */ 962 TAILQ_REMOVE(&ifnetlist, ifp, if_link); 963 964 /* 965 * Remove this ifp from ifnet array. 966 */ 967 /* Free old ifnet array after sync all netisrs */ 968 old_ifnet_array = ifnet_array; 969 ifnet_array = ifnet_array_del(ifp, old_ifnet_array); 970 971 ifnet_unlock(); 972 973 ifgroup_lockmgr(LK_EXCLUSIVE); 974 while ((ifgl = TAILQ_FIRST(&ifp->if_groups)) != NULL) 975 if_delgroup_locked(ifp, ifgl->ifgl_group->ifg_group); 976 ifgroup_lockmgr(LK_RELEASE); 977 978 /* 979 * Sync all netisrs so that the old ifnet array is no longer 980 * accessed and we can free it safely later on. 981 */ 982 netmsg_service_sync(); 983 ifnet_array_free(old_ifnet_array); 984 985 /* 986 * Remove routes and flush queues. 987 */ 988 crit_enter(); 989 #ifdef IFPOLL_ENABLE 990 if (ifp->if_flags & IFF_NPOLLING) 991 ifpoll_deregister(ifp); 992 #endif 993 if_down(ifp); 994 995 /* Decrease the mbuf clusters/jclusters limits increased by us */ 996 if (ifp->if_nmbclusters > 0) 997 mcl_inclimit(-ifp->if_nmbclusters); 998 if (ifp->if_nmbjclusters > 0) 999 mjcl_inclimit(-ifp->if_nmbjclusters); 1000 1001 #ifdef ALTQ 1002 if (ifq_is_enabled(&ifp->if_snd)) 1003 altq_disable(&ifp->if_snd); 1004 if (ifq_is_attached(&ifp->if_snd)) 1005 altq_detach(&ifp->if_snd); 1006 #endif 1007 1008 /* 1009 * Clean up all addresses. 1010 */ 1011 ifp->if_lladdr = NULL; 1012 1013 if_purgeaddrs_nolink(ifp); 1014 if (!TAILQ_EMPTY(&ifp->if_addrheads[mycpuid])) { 1015 struct ifaddr *ifa; 1016 1017 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa; 1018 KASSERT(ifa->ifa_addr->sa_family == AF_LINK, 1019 ("non-link ifaddr is left on if_addrheads")); 1020 1021 ifa_ifunlink(ifa, ifp); 1022 ifa_destroy(ifa); 1023 KASSERT(TAILQ_EMPTY(&ifp->if_addrheads[mycpuid]), 1024 ("there are still ifaddrs left on if_addrheads")); 1025 } 1026 1027 #ifdef INET 1028 /* 1029 * Remove all IPv4 kernel structures related to ifp. 1030 */ 1031 in_ifdetach(ifp); 1032 #endif 1033 1034 #ifdef INET6 1035 /* 1036 * Remove all IPv6 kernel structs related to ifp. This should be done 1037 * before removing routing entries below, since IPv6 interface direct 1038 * routes are expected to be removed by the IPv6-specific kernel API. 1039 * Otherwise, the kernel will detect some inconsistency and bark it. 1040 */ 1041 in6_ifdetach(ifp); 1042 #endif 1043 1044 /* 1045 * Delete all remaining routes using this interface 1046 */ 1047 netmsg_init(&msg.base, NULL, &curthread->td_msgport, MSGF_PRIORITY, 1048 if_rtdel_dispatch); 1049 msg.ifp = ifp; 1050 netisr_domsg_global(&msg.base); 1051 1052 SLIST_FOREACH(dp, &domains, dom_next) { 1053 if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family]) 1054 (*dp->dom_ifdetach)(ifp, 1055 ifp->if_afdata[dp->dom_family]); 1056 } 1057 1058 kfree(ifp->if_addrheads, M_IFADDR); 1059 1060 lwkt_synchronize_ipiqs("if_detach"); 1061 ifq_stage_detach(&ifp->if_snd); 1062 1063 for (q = 0; q < ifp->if_snd.altq_subq_cnt; ++q) { 1064 struct ifaltq_subque *ifsq = &ifp->if_snd.altq_subq[q]; 1065 1066 kfree(ifsq->ifsq_ifstart_nmsg, M_LWKTMSG); 1067 kfree(ifsq->ifsq_stage, M_DEVBUF); 1068 } 1069 kfree(ifp->if_snd.altq_subq, M_DEVBUF); 1070 1071 kfree(ifp->if_data_pcpu, M_DEVBUF); 1072 1073 crit_exit(); 1074 } 1075 1076 int 1077 ifgroup_lockmgr(u_int flags) 1078 { 1079 return lockmgr(&ifgroup_lock, flags); 1080 } 1081 1082 /* 1083 * Create an empty interface group. 1084 */ 1085 static struct ifg_group * 1086 if_creategroup(const char *groupname) 1087 { 1088 struct ifg_group *ifg; 1089 1090 ifg = kmalloc(sizeof(*ifg), M_IFNET, M_WAITOK); 1091 strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group)); 1092 ifg->ifg_refcnt = 0; 1093 ifg->ifg_carp_demoted = 0; 1094 TAILQ_INIT(&ifg->ifg_members); 1095 1096 ifgroup_lockmgr(LK_EXCLUSIVE); 1097 TAILQ_INSERT_TAIL(&ifg_head, ifg, ifg_next); 1098 ifgroup_lockmgr(LK_RELEASE); 1099 1100 EVENTHANDLER_INVOKE(group_attach_event, ifg); 1101 1102 return (ifg); 1103 } 1104 1105 /* 1106 * Destroy an empty interface group. 1107 */ 1108 static int 1109 if_destroygroup(struct ifg_group *ifg) 1110 { 1111 KASSERT(ifg->ifg_refcnt == 0, 1112 ("trying to delete a non-empty interface group")); 1113 1114 ifgroup_lockmgr(LK_EXCLUSIVE); 1115 TAILQ_REMOVE(&ifg_head, ifg, ifg_next); 1116 ifgroup_lockmgr(LK_RELEASE); 1117 1118 EVENTHANDLER_INVOKE(group_detach_event, ifg); 1119 kfree(ifg, M_IFNET); 1120 1121 return (0); 1122 } 1123 1124 /* 1125 * Add the interface to a group. 1126 * The target group will be created if it doesn't exist. 1127 */ 1128 int 1129 if_addgroup(struct ifnet *ifp, const char *groupname) 1130 { 1131 struct ifg_list *ifgl; 1132 struct ifg_group *ifg; 1133 struct ifg_member *ifgm; 1134 1135 if (groupname[0] && 1136 groupname[strlen(groupname) - 1] >= '0' && 1137 groupname[strlen(groupname) - 1] <= '9') 1138 return (EINVAL); 1139 1140 ifgroup_lockmgr(LK_SHARED); 1141 1142 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) { 1143 if (strcmp(ifgl->ifgl_group->ifg_group, groupname) == 0) { 1144 ifgroup_lockmgr(LK_RELEASE); 1145 return (EEXIST); 1146 } 1147 } 1148 1149 TAILQ_FOREACH(ifg, &ifg_head, ifg_next) { 1150 if (strcmp(ifg->ifg_group, groupname) == 0) 1151 break; 1152 } 1153 1154 ifgroup_lockmgr(LK_RELEASE); 1155 1156 if (ifg == NULL) 1157 ifg = if_creategroup(groupname); 1158 1159 ifgl = kmalloc(sizeof(*ifgl), M_IFNET, M_WAITOK); 1160 ifgm = kmalloc(sizeof(*ifgm), M_IFNET, M_WAITOK); 1161 ifgl->ifgl_group = ifg; 1162 ifgm->ifgm_ifp = ifp; 1163 ifg->ifg_refcnt++; 1164 1165 ifgroup_lockmgr(LK_EXCLUSIVE); 1166 TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next); 1167 TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next); 1168 ifgroup_lockmgr(LK_RELEASE); 1169 1170 EVENTHANDLER_INVOKE(group_change_event, groupname); 1171 1172 return (0); 1173 } 1174 1175 /* 1176 * Remove the interface from a group. 1177 * The group will be destroyed if it becomes empty. 1178 * 1179 * The 'ifgroup_lock' must be hold exclusively when calling this. 1180 */ 1181 static int 1182 if_delgroup_locked(struct ifnet *ifp, const char *groupname) 1183 { 1184 struct ifg_list *ifgl; 1185 struct ifg_member *ifgm; 1186 1187 KKASSERT(lockstatus(&ifgroup_lock, curthread) == LK_EXCLUSIVE); 1188 1189 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) { 1190 if (strcmp(ifgl->ifgl_group->ifg_group, groupname) == 0) 1191 break; 1192 } 1193 if (ifgl == NULL) 1194 return (ENOENT); 1195 1196 TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next); 1197 1198 TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next) { 1199 if (ifgm->ifgm_ifp == ifp) 1200 break; 1201 } 1202 1203 if (ifgm != NULL) { 1204 TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next); 1205 1206 ifgroup_lockmgr(LK_RELEASE); 1207 EVENTHANDLER_INVOKE(group_change_event, groupname); 1208 ifgroup_lockmgr(LK_EXCLUSIVE); 1209 1210 kfree(ifgm, M_IFNET); 1211 ifgl->ifgl_group->ifg_refcnt--; 1212 } 1213 1214 if (ifgl->ifgl_group->ifg_refcnt == 0) { 1215 ifgroup_lockmgr(LK_RELEASE); 1216 if_destroygroup(ifgl->ifgl_group); 1217 ifgroup_lockmgr(LK_EXCLUSIVE); 1218 } 1219 1220 kfree(ifgl, M_IFNET); 1221 1222 return (0); 1223 } 1224 1225 int 1226 if_delgroup(struct ifnet *ifp, const char *groupname) 1227 { 1228 int error; 1229 1230 ifgroup_lockmgr(LK_EXCLUSIVE); 1231 error = if_delgroup_locked(ifp, groupname); 1232 ifgroup_lockmgr(LK_RELEASE); 1233 1234 return (error); 1235 } 1236 1237 /* 1238 * Store all the groups that the interface belongs to in memory 1239 * pointed to by data. 1240 */ 1241 static int 1242 if_getgroups(struct ifgroupreq *ifgr, struct ifnet *ifp) 1243 { 1244 struct ifg_list *ifgl; 1245 struct ifg_req *ifgrq, *p; 1246 int len, error; 1247 1248 len = 0; 1249 ifgroup_lockmgr(LK_SHARED); 1250 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) 1251 len += sizeof(struct ifg_req); 1252 ifgroup_lockmgr(LK_RELEASE); 1253 1254 if (ifgr->ifgr_len == 0) { 1255 /* 1256 * Caller is asking how much memory should be allocated in 1257 * the next request in order to hold all the groups. 1258 */ 1259 ifgr->ifgr_len = len; 1260 return (0); 1261 } else if (ifgr->ifgr_len != len) { 1262 return (EINVAL); 1263 } 1264 1265 ifgrq = kmalloc(len, M_TEMP, M_INTWAIT | M_NULLOK | M_ZERO); 1266 if (ifgrq == NULL) 1267 return (ENOMEM); 1268 1269 ifgroup_lockmgr(LK_SHARED); 1270 p = ifgrq; 1271 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) { 1272 if (len < sizeof(struct ifg_req)) { 1273 ifgroup_lockmgr(LK_RELEASE); 1274 return (EINVAL); 1275 } 1276 1277 strlcpy(p->ifgrq_group, ifgl->ifgl_group->ifg_group, 1278 sizeof(ifgrq->ifgrq_group)); 1279 len -= sizeof(struct ifg_req); 1280 p++; 1281 } 1282 ifgroup_lockmgr(LK_RELEASE); 1283 1284 error = copyout(ifgrq, ifgr->ifgr_groups, ifgr->ifgr_len); 1285 kfree(ifgrq, M_TEMP); 1286 if (error) 1287 return (error); 1288 1289 return (0); 1290 } 1291 1292 /* 1293 * Store all the members of a group in memory pointed to by data. 1294 */ 1295 static int 1296 if_getgroupmembers(struct ifgroupreq *ifgr) 1297 { 1298 struct ifg_group *ifg; 1299 struct ifg_member *ifgm; 1300 struct ifg_req *ifgrq, *p; 1301 int len, error; 1302 1303 ifgroup_lockmgr(LK_SHARED); 1304 1305 TAILQ_FOREACH(ifg, &ifg_head, ifg_next) { 1306 if (strcmp(ifg->ifg_group, ifgr->ifgr_name) == 0) 1307 break; 1308 } 1309 if (ifg == NULL) { 1310 ifgroup_lockmgr(LK_RELEASE); 1311 return (ENOENT); 1312 } 1313 1314 len = 0; 1315 TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) 1316 len += sizeof(struct ifg_req); 1317 1318 ifgroup_lockmgr(LK_RELEASE); 1319 1320 if (ifgr->ifgr_len == 0) { 1321 ifgr->ifgr_len = len; 1322 return (0); 1323 } else if (ifgr->ifgr_len != len) { 1324 return (EINVAL); 1325 } 1326 1327 ifgrq = kmalloc(len, M_TEMP, M_INTWAIT | M_NULLOK | M_ZERO); 1328 if (ifgrq == NULL) 1329 return (ENOMEM); 1330 1331 ifgroup_lockmgr(LK_SHARED); 1332 p = ifgrq; 1333 TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) { 1334 if (len < sizeof(struct ifg_req)) { 1335 ifgroup_lockmgr(LK_RELEASE); 1336 return (EINVAL); 1337 } 1338 1339 strlcpy(p->ifgrq_member, ifgm->ifgm_ifp->if_xname, 1340 sizeof(p->ifgrq_member)); 1341 len -= sizeof(struct ifg_req); 1342 p++; 1343 } 1344 ifgroup_lockmgr(LK_RELEASE); 1345 1346 error = copyout(ifgrq, ifgr->ifgr_groups, ifgr->ifgr_len); 1347 kfree(ifgrq, M_TEMP); 1348 if (error) 1349 return (error); 1350 1351 return (0); 1352 } 1353 1354 /* 1355 * Delete Routes for a Network Interface 1356 * 1357 * Called for each routing entry via the rnh->rnh_walktree() call above 1358 * to delete all route entries referencing a detaching network interface. 1359 * 1360 * Arguments: 1361 * rn pointer to node in the routing table 1362 * arg argument passed to rnh->rnh_walktree() - detaching interface 1363 * 1364 * Returns: 1365 * 0 successful 1366 * errno failed - reason indicated 1367 * 1368 */ 1369 static int 1370 if_rtdel(struct radix_node *rn, void *arg) 1371 { 1372 struct rtentry *rt = (struct rtentry *)rn; 1373 struct ifnet *ifp = arg; 1374 int err; 1375 1376 if (rt->rt_ifp == ifp) { 1377 1378 /* 1379 * Protect (sorta) against walktree recursion problems 1380 * with cloned routes 1381 */ 1382 if (!(rt->rt_flags & RTF_UP)) 1383 return (0); 1384 1385 err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway, 1386 rt_mask(rt), rt->rt_flags, 1387 NULL); 1388 if (err) { 1389 log(LOG_WARNING, "if_rtdel: error %d\n", err); 1390 } 1391 } 1392 1393 return (0); 1394 } 1395 1396 static __inline boolean_t 1397 ifa_prefer(const struct ifaddr *cur_ifa, const struct ifaddr *old_ifa) 1398 { 1399 if (old_ifa == NULL) 1400 return TRUE; 1401 1402 if ((old_ifa->ifa_ifp->if_flags & IFF_UP) == 0 && 1403 (cur_ifa->ifa_ifp->if_flags & IFF_UP)) 1404 return TRUE; 1405 if ((old_ifa->ifa_flags & IFA_ROUTE) == 0 && 1406 (cur_ifa->ifa_flags & IFA_ROUTE)) 1407 return TRUE; 1408 return FALSE; 1409 } 1410 1411 /* 1412 * Locate an interface based on a complete address. 1413 */ 1414 struct ifaddr * 1415 ifa_ifwithaddr(struct sockaddr *addr) 1416 { 1417 const struct ifnet_array *arr; 1418 int i; 1419 1420 arr = ifnet_array_get(); 1421 for (i = 0; i < arr->ifnet_count; ++i) { 1422 struct ifnet *ifp = arr->ifnet_arr[i]; 1423 struct ifaddr_container *ifac; 1424 1425 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) { 1426 struct ifaddr *ifa = ifac->ifa; 1427 1428 if (ifa->ifa_addr->sa_family != addr->sa_family) 1429 continue; 1430 if (sa_equal(addr, ifa->ifa_addr)) 1431 return (ifa); 1432 if ((ifp->if_flags & IFF_BROADCAST) && 1433 ifa->ifa_broadaddr && 1434 /* IPv6 doesn't have broadcast */ 1435 ifa->ifa_broadaddr->sa_len != 0 && 1436 sa_equal(ifa->ifa_broadaddr, addr)) 1437 return (ifa); 1438 } 1439 } 1440 return (NULL); 1441 } 1442 1443 /* 1444 * Locate the point to point interface with a given destination address. 1445 */ 1446 struct ifaddr * 1447 ifa_ifwithdstaddr(struct sockaddr *addr) 1448 { 1449 const struct ifnet_array *arr; 1450 int i; 1451 1452 arr = ifnet_array_get(); 1453 for (i = 0; i < arr->ifnet_count; ++i) { 1454 struct ifnet *ifp = arr->ifnet_arr[i]; 1455 struct ifaddr_container *ifac; 1456 1457 if (!(ifp->if_flags & IFF_POINTOPOINT)) 1458 continue; 1459 1460 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) { 1461 struct ifaddr *ifa = ifac->ifa; 1462 1463 if (ifa->ifa_addr->sa_family != addr->sa_family) 1464 continue; 1465 if (ifa->ifa_dstaddr && 1466 sa_equal(addr, ifa->ifa_dstaddr)) 1467 return (ifa); 1468 } 1469 } 1470 return (NULL); 1471 } 1472 1473 /* 1474 * Find an interface on a specific network. If many, choice 1475 * is most specific found. 1476 */ 1477 struct ifaddr * 1478 ifa_ifwithnet(struct sockaddr *addr) 1479 { 1480 struct ifaddr *ifa_maybe = NULL; 1481 u_int af = addr->sa_family; 1482 char *addr_data = addr->sa_data, *cplim; 1483 const struct ifnet_array *arr; 1484 int i; 1485 1486 /* 1487 * AF_LINK addresses can be looked up directly by their index number, 1488 * so do that if we can. 1489 */ 1490 if (af == AF_LINK) { 1491 struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr; 1492 1493 if (sdl->sdl_index && sdl->sdl_index <= if_index) 1494 return (ifindex2ifnet[sdl->sdl_index]->if_lladdr); 1495 } 1496 1497 /* 1498 * Scan though each interface, looking for ones that have 1499 * addresses in this address family. 1500 */ 1501 arr = ifnet_array_get(); 1502 for (i = 0; i < arr->ifnet_count; ++i) { 1503 struct ifnet *ifp = arr->ifnet_arr[i]; 1504 struct ifaddr_container *ifac; 1505 1506 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) { 1507 struct ifaddr *ifa = ifac->ifa; 1508 char *cp, *cp2, *cp3; 1509 1510 if (ifa->ifa_addr->sa_family != af) 1511 next: continue; 1512 if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT) { 1513 /* 1514 * This is a bit broken as it doesn't 1515 * take into account that the remote end may 1516 * be a single node in the network we are 1517 * looking for. 1518 * The trouble is that we don't know the 1519 * netmask for the remote end. 1520 */ 1521 if (ifa->ifa_dstaddr != NULL && 1522 sa_equal(addr, ifa->ifa_dstaddr)) 1523 return (ifa); 1524 } else { 1525 /* 1526 * if we have a special address handler, 1527 * then use it instead of the generic one. 1528 */ 1529 if (ifa->ifa_claim_addr) { 1530 if ((*ifa->ifa_claim_addr)(ifa, addr)) { 1531 return (ifa); 1532 } else { 1533 continue; 1534 } 1535 } 1536 1537 /* 1538 * Scan all the bits in the ifa's address. 1539 * If a bit dissagrees with what we are 1540 * looking for, mask it with the netmask 1541 * to see if it really matters. 1542 * (A byte at a time) 1543 */ 1544 if (ifa->ifa_netmask == 0) 1545 continue; 1546 cp = addr_data; 1547 cp2 = ifa->ifa_addr->sa_data; 1548 cp3 = ifa->ifa_netmask->sa_data; 1549 cplim = ifa->ifa_netmask->sa_len + 1550 (char *)ifa->ifa_netmask; 1551 while (cp3 < cplim) 1552 if ((*cp++ ^ *cp2++) & *cp3++) 1553 goto next; /* next address! */ 1554 /* 1555 * If the netmask of what we just found 1556 * is more specific than what we had before 1557 * (if we had one) then remember the new one 1558 * before continuing to search for an even 1559 * better one. If the netmasks are equal, 1560 * we prefer the this ifa based on the result 1561 * of ifa_prefer(). 1562 */ 1563 if (ifa_maybe == NULL || 1564 rn_refines((char *)ifa->ifa_netmask, 1565 (char *)ifa_maybe->ifa_netmask) || 1566 (sa_equal(ifa_maybe->ifa_netmask, 1567 ifa->ifa_netmask) && 1568 ifa_prefer(ifa, ifa_maybe))) 1569 ifa_maybe = ifa; 1570 } 1571 } 1572 } 1573 return (ifa_maybe); 1574 } 1575 1576 /* 1577 * Find an interface address specific to an interface best matching 1578 * a given address. 1579 */ 1580 struct ifaddr * 1581 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp) 1582 { 1583 struct ifaddr_container *ifac; 1584 char *cp, *cp2, *cp3; 1585 char *cplim; 1586 struct ifaddr *ifa_maybe = NULL; 1587 u_int af = addr->sa_family; 1588 1589 if (af >= AF_MAX) 1590 return (0); 1591 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) { 1592 struct ifaddr *ifa = ifac->ifa; 1593 1594 if (ifa->ifa_addr->sa_family != af) 1595 continue; 1596 if (ifa_maybe == NULL) 1597 ifa_maybe = ifa; 1598 if (ifa->ifa_netmask == NULL) { 1599 if (sa_equal(addr, ifa->ifa_addr) || 1600 (ifa->ifa_dstaddr != NULL && 1601 sa_equal(addr, ifa->ifa_dstaddr))) 1602 return (ifa); 1603 continue; 1604 } 1605 if (ifp->if_flags & IFF_POINTOPOINT) { 1606 if (sa_equal(addr, ifa->ifa_dstaddr)) 1607 return (ifa); 1608 } else { 1609 cp = addr->sa_data; 1610 cp2 = ifa->ifa_addr->sa_data; 1611 cp3 = ifa->ifa_netmask->sa_data; 1612 cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask; 1613 for (; cp3 < cplim; cp3++) 1614 if ((*cp++ ^ *cp2++) & *cp3) 1615 break; 1616 if (cp3 == cplim) 1617 return (ifa); 1618 } 1619 } 1620 return (ifa_maybe); 1621 } 1622 1623 /* 1624 * Default action when installing a route with a Link Level gateway. 1625 * Lookup an appropriate real ifa to point to. 1626 * This should be moved to /sys/net/link.c eventually. 1627 */ 1628 static void 1629 link_rtrequest(int cmd, struct rtentry *rt) 1630 { 1631 struct ifaddr *ifa; 1632 struct sockaddr *dst; 1633 struct ifnet *ifp; 1634 1635 if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL || 1636 (ifp = ifa->ifa_ifp) == NULL || (dst = rt_key(rt)) == NULL) 1637 return; 1638 ifa = ifaof_ifpforaddr(dst, ifp); 1639 if (ifa != NULL) { 1640 IFAFREE(rt->rt_ifa); 1641 IFAREF(ifa); 1642 rt->rt_ifa = ifa; 1643 if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest) 1644 ifa->ifa_rtrequest(cmd, rt); 1645 } 1646 } 1647 1648 struct netmsg_if { 1649 struct netmsg_base base; 1650 struct ifnet *ifp; 1651 }; 1652 1653 /* 1654 * Mark an interface down and notify protocols of the transition. 1655 */ 1656 static void 1657 if_down_dispatch(netmsg_t nmsg) 1658 { 1659 struct netmsg_if *msg = (struct netmsg_if *)nmsg; 1660 struct ifnet *ifp = msg->ifp; 1661 struct ifaddr_container *ifac; 1662 struct domain *dp; 1663 1664 ASSERT_NETISR0; 1665 1666 ifp->if_flags &= ~IFF_UP; 1667 getmicrotime(&ifp->if_lastchange); 1668 rt_ifmsg(ifp); 1669 1670 /* 1671 * The ifaddr processing in the following loop will block, 1672 * however, this function is called in netisr0, in which 1673 * ifaddr list changes happen, so we don't care about the 1674 * blockness of the ifaddr processing here. 1675 */ 1676 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) { 1677 struct ifaddr *ifa = ifac->ifa; 1678 1679 /* Ignore marker */ 1680 if (ifa->ifa_addr->sa_family == AF_UNSPEC) 1681 continue; 1682 1683 kpfctlinput(PRC_IFDOWN, ifa->ifa_addr); 1684 } 1685 1686 SLIST_FOREACH(dp, &domains, dom_next) 1687 if (dp->dom_if_down != NULL) 1688 dp->dom_if_down(ifp); 1689 1690 ifq_purge_all(&ifp->if_snd); 1691 netisr_replymsg(&nmsg->base, 0); 1692 } 1693 1694 /* 1695 * Mark an interface up and notify protocols of the transition. 1696 */ 1697 static void 1698 if_up_dispatch(netmsg_t nmsg) 1699 { 1700 struct netmsg_if *msg = (struct netmsg_if *)nmsg; 1701 struct ifnet *ifp = msg->ifp; 1702 struct ifaddr_container *ifac; 1703 struct domain *dp; 1704 1705 ASSERT_NETISR0; 1706 1707 ifq_purge_all(&ifp->if_snd); 1708 ifp->if_flags |= IFF_UP; 1709 getmicrotime(&ifp->if_lastchange); 1710 rt_ifmsg(ifp); 1711 1712 /* 1713 * The ifaddr processing in the following loop will block, 1714 * however, this function is called in netisr0, in which 1715 * ifaddr list changes happen, so we don't care about the 1716 * blockness of the ifaddr processing here. 1717 */ 1718 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) { 1719 struct ifaddr *ifa = ifac->ifa; 1720 1721 /* Ignore marker */ 1722 if (ifa->ifa_addr->sa_family == AF_UNSPEC) 1723 continue; 1724 1725 kpfctlinput(PRC_IFUP, ifa->ifa_addr); 1726 } 1727 1728 SLIST_FOREACH(dp, &domains, dom_next) 1729 if (dp->dom_if_up != NULL) 1730 dp->dom_if_up(ifp); 1731 1732 netisr_replymsg(&nmsg->base, 0); 1733 } 1734 1735 /* 1736 * Mark an interface down and notify protocols of the transition. An 1737 * interface going down is also considered to be a synchronizing event. 1738 * We must ensure that all packet processing related to the interface 1739 * has completed before we return so e.g. the caller can free the ifnet 1740 * structure that the mbufs may be referencing. 1741 * 1742 * NOTE: must be called at splnet or eqivalent. 1743 */ 1744 void 1745 if_down(struct ifnet *ifp) 1746 { 1747 struct netmsg_if msg; 1748 1749 EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_DOWN); 1750 netmsg_init(&msg.base, NULL, &curthread->td_msgport, 0, 1751 if_down_dispatch); 1752 msg.ifp = ifp; 1753 netisr_domsg(&msg.base, 0); 1754 netmsg_service_sync(); 1755 } 1756 1757 /* 1758 * Mark an interface up and notify protocols of 1759 * the transition. 1760 * NOTE: must be called at splnet or eqivalent. 1761 */ 1762 void 1763 if_up(struct ifnet *ifp) 1764 { 1765 struct netmsg_if msg; 1766 1767 netmsg_init(&msg.base, NULL, &curthread->td_msgport, 0, 1768 if_up_dispatch); 1769 msg.ifp = ifp; 1770 netisr_domsg(&msg.base, 0); 1771 EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_UP); 1772 } 1773 1774 /* 1775 * Process a link state change. 1776 * NOTE: must be called at splsoftnet or equivalent. 1777 */ 1778 void 1779 if_link_state_change(struct ifnet *ifp) 1780 { 1781 int link_state = ifp->if_link_state; 1782 1783 rt_ifmsg(ifp); 1784 devctl_notify("IFNET", ifp->if_xname, 1785 (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL); 1786 1787 EVENTHANDLER_INVOKE(ifnet_link_event, ifp, link_state); 1788 } 1789 1790 /* 1791 * Handle interface watchdog timer routines. Called 1792 * from softclock, we decrement timers (if set) and 1793 * call the appropriate interface routine on expiration. 1794 */ 1795 static void 1796 if_slowtimo_dispatch(netmsg_t nmsg) 1797 { 1798 struct globaldata *gd = mycpu; 1799 const struct ifnet_array *arr; 1800 int i; 1801 1802 ASSERT_NETISR0; 1803 1804 crit_enter_gd(gd); 1805 lwkt_replymsg(&nmsg->lmsg, 0); /* reply ASAP */ 1806 crit_exit_gd(gd); 1807 1808 arr = ifnet_array_get(); 1809 for (i = 0; i < arr->ifnet_count; ++i) { 1810 struct ifnet *ifp = arr->ifnet_arr[i]; 1811 1812 crit_enter_gd(gd); 1813 1814 if (if_stats_compat) { 1815 IFNET_STAT_GET(ifp, ipackets, ifp->if_ipackets); 1816 IFNET_STAT_GET(ifp, ierrors, ifp->if_ierrors); 1817 IFNET_STAT_GET(ifp, opackets, ifp->if_opackets); 1818 IFNET_STAT_GET(ifp, oerrors, ifp->if_oerrors); 1819 IFNET_STAT_GET(ifp, collisions, ifp->if_collisions); 1820 IFNET_STAT_GET(ifp, ibytes, ifp->if_ibytes); 1821 IFNET_STAT_GET(ifp, obytes, ifp->if_obytes); 1822 IFNET_STAT_GET(ifp, imcasts, ifp->if_imcasts); 1823 IFNET_STAT_GET(ifp, omcasts, ifp->if_omcasts); 1824 IFNET_STAT_GET(ifp, iqdrops, ifp->if_iqdrops); 1825 IFNET_STAT_GET(ifp, noproto, ifp->if_noproto); 1826 IFNET_STAT_GET(ifp, oqdrops, ifp->if_oqdrops); 1827 } 1828 1829 if (ifp->if_timer == 0 || --ifp->if_timer) { 1830 crit_exit_gd(gd); 1831 continue; 1832 } 1833 if (ifp->if_watchdog) { 1834 if (ifnet_tryserialize_all(ifp)) { 1835 (*ifp->if_watchdog)(ifp); 1836 ifnet_deserialize_all(ifp); 1837 } else { 1838 /* try again next timeout */ 1839 ++ifp->if_timer; 1840 } 1841 } 1842 1843 crit_exit_gd(gd); 1844 } 1845 1846 callout_reset(&if_slowtimo_timer, hz / IFNET_SLOWHZ, if_slowtimo, NULL); 1847 } 1848 1849 static void 1850 if_slowtimo(void *arg __unused) 1851 { 1852 struct lwkt_msg *lmsg = &if_slowtimo_netmsg.lmsg; 1853 1854 KASSERT(mycpuid == 0, ("not on cpu0")); 1855 crit_enter(); 1856 if (lmsg->ms_flags & MSGF_DONE) 1857 lwkt_sendmsg_oncpu(netisr_cpuport(0), lmsg); 1858 crit_exit(); 1859 } 1860 1861 /* 1862 * Map interface name to 1863 * interface structure pointer. 1864 */ 1865 struct ifnet * 1866 ifunit(const char *name) 1867 { 1868 struct ifnet *ifp; 1869 1870 /* 1871 * Search all the interfaces for this name/number 1872 */ 1873 KASSERT(mtx_owned(&ifnet_mtx), ("ifnet is not locked")); 1874 1875 TAILQ_FOREACH(ifp, &ifnetlist, if_link) { 1876 if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0) 1877 break; 1878 } 1879 return (ifp); 1880 } 1881 1882 struct ifnet * 1883 ifunit_netisr(const char *name) 1884 { 1885 const struct ifnet_array *arr; 1886 int i; 1887 1888 /* 1889 * Search all the interfaces for this name/number 1890 */ 1891 1892 arr = ifnet_array_get(); 1893 for (i = 0; i < arr->ifnet_count; ++i) { 1894 struct ifnet *ifp = arr->ifnet_arr[i]; 1895 1896 if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0) 1897 return ifp; 1898 } 1899 return NULL; 1900 } 1901 1902 /* 1903 * Interface ioctls. 1904 */ 1905 int 1906 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct ucred *cred) 1907 { 1908 struct ifnet *ifp; 1909 struct ifgroupreq *ifgr; 1910 struct ifreq *ifr; 1911 struct ifstat *ifs; 1912 int error, do_ifup = 0; 1913 short oif_flags; 1914 int new_flags; 1915 size_t namelen, onamelen; 1916 char new_name[IFNAMSIZ]; 1917 struct ifaddr *ifa; 1918 struct sockaddr_dl *sdl; 1919 1920 switch (cmd) { 1921 case SIOCGIFCONF: 1922 return (ifconf(cmd, data, cred)); 1923 default: 1924 break; 1925 } 1926 1927 ifr = (struct ifreq *)data; 1928 1929 switch (cmd) { 1930 case SIOCIFCREATE: 1931 case SIOCIFCREATE2: 1932 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0) 1933 return (error); 1934 return (if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name), 1935 cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL)); 1936 case SIOCIFDESTROY: 1937 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0) 1938 return (error); 1939 return (if_clone_destroy(ifr->ifr_name)); 1940 case SIOCIFGCLONERS: 1941 return (if_clone_list((struct if_clonereq *)data)); 1942 case SIOCGIFGMEMB: 1943 return (if_getgroupmembers((struct ifgroupreq *)data)); 1944 default: 1945 break; 1946 } 1947 1948 /* 1949 * Nominal ioctl through interface, lookup the ifp and obtain a 1950 * lock to serialize the ifconfig ioctl operation. 1951 */ 1952 ifnet_lock(); 1953 1954 ifp = ifunit(ifr->ifr_name); 1955 if (ifp == NULL) { 1956 ifnet_unlock(); 1957 return (ENXIO); 1958 } 1959 error = 0; 1960 1961 switch (cmd) { 1962 case SIOCGIFINDEX: 1963 ifr->ifr_index = ifp->if_index; 1964 break; 1965 1966 case SIOCGIFFLAGS: 1967 ifr->ifr_flags = ifp->if_flags; 1968 ifr->ifr_flagshigh = ifp->if_flags >> 16; 1969 break; 1970 1971 case SIOCGIFCAP: 1972 ifr->ifr_reqcap = ifp->if_capabilities; 1973 ifr->ifr_curcap = ifp->if_capenable; 1974 break; 1975 1976 case SIOCGIFMETRIC: 1977 ifr->ifr_metric = ifp->if_metric; 1978 break; 1979 1980 case SIOCGIFMTU: 1981 ifr->ifr_mtu = ifp->if_mtu; 1982 break; 1983 1984 case SIOCGIFTSOLEN: 1985 ifr->ifr_tsolen = ifp->if_tsolen; 1986 break; 1987 1988 case SIOCGIFDATA: 1989 error = copyout((caddr_t)&ifp->if_data, ifr->ifr_data, 1990 sizeof(ifp->if_data)); 1991 break; 1992 1993 case SIOCGIFPHYS: 1994 ifr->ifr_phys = ifp->if_physical; 1995 break; 1996 1997 case SIOCGIFPOLLCPU: 1998 ifr->ifr_pollcpu = -1; 1999 break; 2000 2001 case SIOCSIFPOLLCPU: 2002 break; 2003 2004 case SIOCSIFFLAGS: 2005 error = priv_check_cred(cred, PRIV_ROOT, 0); 2006 if (error) 2007 break; 2008 new_flags = (ifr->ifr_flags & 0xffff) | 2009 (ifr->ifr_flagshigh << 16); 2010 if (ifp->if_flags & IFF_SMART) { 2011 /* Smart drivers twiddle their own routes */ 2012 } else if (ifp->if_flags & IFF_UP && 2013 (new_flags & IFF_UP) == 0) { 2014 if_down(ifp); 2015 } else if (new_flags & IFF_UP && 2016 (ifp->if_flags & IFF_UP) == 0) { 2017 do_ifup = 1; 2018 } 2019 2020 #ifdef IFPOLL_ENABLE 2021 if ((new_flags ^ ifp->if_flags) & IFF_NPOLLING) { 2022 if (new_flags & IFF_NPOLLING) 2023 ifpoll_register(ifp); 2024 else 2025 ifpoll_deregister(ifp); 2026 } 2027 #endif 2028 2029 ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) | 2030 (new_flags &~ IFF_CANTCHANGE); 2031 if (new_flags & IFF_PPROMISC) { 2032 /* Permanently promiscuous mode requested */ 2033 ifp->if_flags |= IFF_PROMISC; 2034 } else if (ifp->if_pcount == 0) { 2035 ifp->if_flags &= ~IFF_PROMISC; 2036 } 2037 if (ifp->if_ioctl) { 2038 ifnet_serialize_all(ifp); 2039 ifp->if_ioctl(ifp, cmd, data, cred); 2040 ifnet_deserialize_all(ifp); 2041 } 2042 if (do_ifup) 2043 if_up(ifp); 2044 getmicrotime(&ifp->if_lastchange); 2045 break; 2046 2047 case SIOCSIFCAP: 2048 error = priv_check_cred(cred, PRIV_ROOT, 0); 2049 if (error) 2050 break; 2051 if (ifr->ifr_reqcap & ~ifp->if_capabilities) { 2052 error = EINVAL; 2053 break; 2054 } 2055 ifnet_serialize_all(ifp); 2056 ifp->if_ioctl(ifp, cmd, data, cred); 2057 ifnet_deserialize_all(ifp); 2058 break; 2059 2060 case SIOCSIFNAME: 2061 error = priv_check_cred(cred, PRIV_ROOT, 0); 2062 if (error) 2063 break; 2064 error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL); 2065 if (error) 2066 break; 2067 if (new_name[0] == '\0') { 2068 error = EINVAL; 2069 break; 2070 } 2071 if (ifunit(new_name) != NULL) { 2072 error = EEXIST; 2073 break; 2074 } 2075 2076 EVENTHANDLER_INVOKE(ifnet_detach_event, ifp); 2077 2078 /* Announce the departure of the interface. */ 2079 rt_ifannouncemsg(ifp, IFAN_DEPARTURE); 2080 2081 strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname)); 2082 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa; 2083 sdl = (struct sockaddr_dl *)ifa->ifa_addr; 2084 namelen = strlen(new_name); 2085 onamelen = sdl->sdl_nlen; 2086 /* 2087 * Move the address if needed. This is safe because we 2088 * allocate space for a name of length IFNAMSIZ when we 2089 * create this in if_attach(). 2090 */ 2091 if (namelen != onamelen) { 2092 bcopy(sdl->sdl_data + onamelen, 2093 sdl->sdl_data + namelen, sdl->sdl_alen); 2094 } 2095 bcopy(new_name, sdl->sdl_data, namelen); 2096 sdl->sdl_nlen = namelen; 2097 sdl = (struct sockaddr_dl *)ifa->ifa_netmask; 2098 bzero(sdl->sdl_data, onamelen); 2099 while (namelen != 0) 2100 sdl->sdl_data[--namelen] = 0xff; 2101 2102 EVENTHANDLER_INVOKE(ifnet_attach_event, ifp); 2103 2104 /* Announce the return of the interface. */ 2105 rt_ifannouncemsg(ifp, IFAN_ARRIVAL); 2106 break; 2107 2108 case SIOCSIFMETRIC: 2109 error = priv_check_cred(cred, PRIV_ROOT, 0); 2110 if (error) 2111 break; 2112 ifp->if_metric = ifr->ifr_metric; 2113 getmicrotime(&ifp->if_lastchange); 2114 break; 2115 2116 case SIOCSIFPHYS: 2117 error = priv_check_cred(cred, PRIV_ROOT, 0); 2118 if (error) 2119 break; 2120 if (ifp->if_ioctl == NULL) { 2121 error = EOPNOTSUPP; 2122 break; 2123 } 2124 ifnet_serialize_all(ifp); 2125 error = ifp->if_ioctl(ifp, cmd, data, cred); 2126 ifnet_deserialize_all(ifp); 2127 if (error == 0) 2128 getmicrotime(&ifp->if_lastchange); 2129 break; 2130 2131 case SIOCSIFMTU: 2132 { 2133 u_long oldmtu = ifp->if_mtu; 2134 2135 error = priv_check_cred(cred, PRIV_ROOT, 0); 2136 if (error) 2137 break; 2138 if (ifp->if_ioctl == NULL) { 2139 error = EOPNOTSUPP; 2140 break; 2141 } 2142 if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU) { 2143 error = EINVAL; 2144 break; 2145 } 2146 ifnet_serialize_all(ifp); 2147 error = ifp->if_ioctl(ifp, cmd, data, cred); 2148 ifnet_deserialize_all(ifp); 2149 if (error == 0) { 2150 getmicrotime(&ifp->if_lastchange); 2151 rt_ifmsg(ifp); 2152 } 2153 /* 2154 * If the link MTU changed, do network layer specific procedure. 2155 */ 2156 if (ifp->if_mtu != oldmtu) { 2157 #ifdef INET6 2158 nd6_setmtu(ifp); 2159 #endif 2160 } 2161 break; 2162 } 2163 2164 case SIOCSIFTSOLEN: 2165 error = priv_check_cred(cred, PRIV_ROOT, 0); 2166 if (error) 2167 break; 2168 2169 /* XXX need driver supplied upper limit */ 2170 if (ifr->ifr_tsolen <= 0) { 2171 error = EINVAL; 2172 break; 2173 } 2174 ifp->if_tsolen = ifr->ifr_tsolen; 2175 break; 2176 2177 case SIOCADDMULTI: 2178 case SIOCDELMULTI: 2179 error = priv_check_cred(cred, PRIV_ROOT, 0); 2180 if (error) 2181 break; 2182 2183 /* Don't allow group membership on non-multicast interfaces. */ 2184 if ((ifp->if_flags & IFF_MULTICAST) == 0) { 2185 error = EOPNOTSUPP; 2186 break; 2187 } 2188 2189 /* Don't let users screw up protocols' entries. */ 2190 if (ifr->ifr_addr.sa_family != AF_LINK) { 2191 error = EINVAL; 2192 break; 2193 } 2194 2195 if (cmd == SIOCADDMULTI) { 2196 struct ifmultiaddr *ifma; 2197 error = if_addmulti(ifp, &ifr->ifr_addr, &ifma); 2198 } else { 2199 error = if_delmulti(ifp, &ifr->ifr_addr); 2200 } 2201 if (error == 0) 2202 getmicrotime(&ifp->if_lastchange); 2203 break; 2204 2205 case SIOCSIFPHYADDR: 2206 case SIOCDIFPHYADDR: 2207 #ifdef INET6 2208 case SIOCSIFPHYADDR_IN6: 2209 #endif 2210 case SIOCSLIFPHYADDR: 2211 case SIOCSIFMEDIA: 2212 case SIOCSIFGENERIC: 2213 error = priv_check_cred(cred, PRIV_ROOT, 0); 2214 if (error) 2215 break; 2216 if (ifp->if_ioctl == NULL) { 2217 error = EOPNOTSUPP; 2218 break; 2219 } 2220 ifnet_serialize_all(ifp); 2221 error = ifp->if_ioctl(ifp, cmd, data, cred); 2222 ifnet_deserialize_all(ifp); 2223 if (error == 0) 2224 getmicrotime(&ifp->if_lastchange); 2225 break; 2226 2227 case SIOCGIFSTATUS: 2228 ifs = (struct ifstat *)data; 2229 ifs->ascii[0] = '\0'; 2230 /* fall through */ 2231 case SIOCGIFPSRCADDR: 2232 case SIOCGIFPDSTADDR: 2233 case SIOCGLIFPHYADDR: 2234 case SIOCGIFMEDIA: 2235 case SIOCGIFGENERIC: 2236 if (ifp->if_ioctl == NULL) { 2237 error = EOPNOTSUPP; 2238 break; 2239 } 2240 ifnet_serialize_all(ifp); 2241 error = ifp->if_ioctl(ifp, cmd, data, cred); 2242 ifnet_deserialize_all(ifp); 2243 break; 2244 2245 case SIOCSIFLLADDR: 2246 error = priv_check_cred(cred, PRIV_ROOT, 0); 2247 if (error) 2248 break; 2249 error = if_setlladdr(ifp, ifr->ifr_addr.sa_data, 2250 ifr->ifr_addr.sa_len); 2251 EVENTHANDLER_INVOKE(iflladdr_event, ifp); 2252 break; 2253 2254 case SIOCAIFGROUP: 2255 ifgr = (struct ifgroupreq *)ifr; 2256 if ((error = priv_check_cred(cred, PRIV_NET_ADDIFGROUP, 0))) 2257 return (error); 2258 if ((error = if_addgroup(ifp, ifgr->ifgr_group))) 2259 return (error); 2260 break; 2261 2262 case SIOCDIFGROUP: 2263 ifgr = (struct ifgroupreq *)ifr; 2264 if ((error = priv_check_cred(cred, PRIV_NET_DELIFGROUP, 0))) 2265 return (error); 2266 if ((error = if_delgroup(ifp, ifgr->ifgr_group))) 2267 return (error); 2268 break; 2269 2270 case SIOCGIFGROUP: 2271 ifgr = (struct ifgroupreq *)ifr; 2272 if ((error = if_getgroups(ifgr, ifp))) 2273 return (error); 2274 break; 2275 2276 default: 2277 oif_flags = ifp->if_flags; 2278 if (so->so_proto == 0) { 2279 error = EOPNOTSUPP; 2280 break; 2281 } 2282 error = so_pru_control_direct(so, cmd, data, ifp); 2283 2284 /* 2285 * If the socket control method returns EOPNOTSUPP, pass the 2286 * request directly to the interface. 2287 * 2288 * Exclude the SIOCSIF{ADDR,BRDADDR,DSTADDR,NETMASK} ioctls, 2289 * because drivers may trust these ioctls to come from an 2290 * already privileged layer and thus do not perform credentials 2291 * checks or input validation. 2292 */ 2293 if (error == EOPNOTSUPP && 2294 ifp->if_ioctl != NULL && 2295 cmd != SIOCSIFADDR && 2296 cmd != SIOCSIFBRDADDR && 2297 cmd != SIOCSIFDSTADDR && 2298 cmd != SIOCSIFNETMASK) { 2299 ifnet_serialize_all(ifp); 2300 error = ifp->if_ioctl(ifp, cmd, data, cred); 2301 ifnet_deserialize_all(ifp); 2302 } 2303 2304 if ((oif_flags ^ ifp->if_flags) & IFF_UP) { 2305 #ifdef INET6 2306 DELAY(100);/* XXX: temporary workaround for fxp issue*/ 2307 if (ifp->if_flags & IFF_UP) { 2308 crit_enter(); 2309 in6_if_up(ifp); 2310 crit_exit(); 2311 } 2312 #endif 2313 } 2314 break; 2315 } 2316 2317 ifnet_unlock(); 2318 return (error); 2319 } 2320 2321 /* 2322 * Set/clear promiscuous mode on interface ifp based on the truth value 2323 * of pswitch. The calls are reference counted so that only the first 2324 * "on" request actually has an effect, as does the final "off" request. 2325 * Results are undefined if the "off" and "on" requests are not matched. 2326 */ 2327 int 2328 ifpromisc(struct ifnet *ifp, int pswitch) 2329 { 2330 struct ifreq ifr; 2331 int error; 2332 int oldflags; 2333 2334 oldflags = ifp->if_flags; 2335 if (ifp->if_flags & IFF_PPROMISC) { 2336 /* Do nothing if device is in permanently promiscuous mode */ 2337 ifp->if_pcount += pswitch ? 1 : -1; 2338 return (0); 2339 } 2340 if (pswitch) { 2341 /* 2342 * If the device is not configured up, we cannot put it in 2343 * promiscuous mode. 2344 */ 2345 if ((ifp->if_flags & IFF_UP) == 0) 2346 return (ENETDOWN); 2347 if (ifp->if_pcount++ != 0) 2348 return (0); 2349 ifp->if_flags |= IFF_PROMISC; 2350 log(LOG_INFO, "%s: promiscuous mode enabled\n", 2351 ifp->if_xname); 2352 } else { 2353 if (--ifp->if_pcount > 0) 2354 return (0); 2355 ifp->if_flags &= ~IFF_PROMISC; 2356 log(LOG_INFO, "%s: promiscuous mode disabled\n", 2357 ifp->if_xname); 2358 } 2359 ifr.ifr_flags = ifp->if_flags; 2360 ifr.ifr_flagshigh = ifp->if_flags >> 16; 2361 ifnet_serialize_all(ifp); 2362 error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr, NULL); 2363 ifnet_deserialize_all(ifp); 2364 if (error == 0) 2365 rt_ifmsg(ifp); 2366 else 2367 ifp->if_flags = oldflags; 2368 return error; 2369 } 2370 2371 /* 2372 * Return interface configuration 2373 * of system. List may be used 2374 * in later ioctl's (above) to get 2375 * other information. 2376 */ 2377 static int 2378 ifconf(u_long cmd, caddr_t data, struct ucred *cred) 2379 { 2380 struct ifconf *ifc = (struct ifconf *)data; 2381 struct ifnet *ifp; 2382 struct sockaddr *sa; 2383 struct ifreq ifr, *ifrp; 2384 int space = ifc->ifc_len, error = 0; 2385 2386 ifrp = ifc->ifc_req; 2387 2388 ifnet_lock(); 2389 TAILQ_FOREACH(ifp, &ifnetlist, if_link) { 2390 struct ifaddr_container *ifac, *ifac_mark; 2391 struct ifaddr_marker mark; 2392 struct ifaddrhead *head; 2393 int addrs; 2394 2395 if (space <= sizeof ifr) 2396 break; 2397 2398 /* 2399 * Zero the stack declared structure first to prevent 2400 * memory disclosure. 2401 */ 2402 bzero(&ifr, sizeof(ifr)); 2403 if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name)) 2404 >= sizeof(ifr.ifr_name)) { 2405 error = ENAMETOOLONG; 2406 break; 2407 } 2408 2409 /* 2410 * Add a marker, since copyout() could block and during that 2411 * period the list could be changed. Inserting the marker to 2412 * the header of the list will not cause trouble for the code 2413 * assuming that the first element of the list is AF_LINK; the 2414 * marker will be moved to the next position w/o blocking. 2415 */ 2416 ifa_marker_init(&mark, ifp); 2417 ifac_mark = &mark.ifac; 2418 head = &ifp->if_addrheads[mycpuid]; 2419 2420 addrs = 0; 2421 TAILQ_INSERT_HEAD(head, ifac_mark, ifa_link); 2422 while ((ifac = TAILQ_NEXT(ifac_mark, ifa_link)) != NULL) { 2423 struct ifaddr *ifa = ifac->ifa; 2424 2425 TAILQ_REMOVE(head, ifac_mark, ifa_link); 2426 TAILQ_INSERT_AFTER(head, ifac, ifac_mark, ifa_link); 2427 2428 /* Ignore marker */ 2429 if (ifa->ifa_addr->sa_family == AF_UNSPEC) 2430 continue; 2431 2432 if (space <= sizeof ifr) 2433 break; 2434 sa = ifa->ifa_addr; 2435 if (cred->cr_prison && 2436 prison_if(cred, sa)) 2437 continue; 2438 addrs++; 2439 /* 2440 * Keep a reference on this ifaddr, so that it will 2441 * not be destroyed when its address is copied to 2442 * the userland, which could block. 2443 */ 2444 IFAREF(ifa); 2445 if (sa->sa_len <= sizeof(*sa)) { 2446 ifr.ifr_addr = *sa; 2447 error = copyout(&ifr, ifrp, sizeof ifr); 2448 ifrp++; 2449 } else { 2450 if (space < (sizeof ifr) + sa->sa_len - 2451 sizeof(*sa)) { 2452 IFAFREE(ifa); 2453 break; 2454 } 2455 space -= sa->sa_len - sizeof(*sa); 2456 error = copyout(&ifr, ifrp, 2457 sizeof ifr.ifr_name); 2458 if (error == 0) 2459 error = copyout(sa, &ifrp->ifr_addr, 2460 sa->sa_len); 2461 ifrp = (struct ifreq *) 2462 (sa->sa_len + (caddr_t)&ifrp->ifr_addr); 2463 } 2464 IFAFREE(ifa); 2465 if (error) 2466 break; 2467 space -= sizeof ifr; 2468 } 2469 TAILQ_REMOVE(head, ifac_mark, ifa_link); 2470 if (error) 2471 break; 2472 if (!addrs) { 2473 bzero(&ifr.ifr_addr, sizeof ifr.ifr_addr); 2474 error = copyout(&ifr, ifrp, sizeof ifr); 2475 if (error) 2476 break; 2477 space -= sizeof ifr; 2478 ifrp++; 2479 } 2480 } 2481 ifnet_unlock(); 2482 2483 ifc->ifc_len -= space; 2484 return (error); 2485 } 2486 2487 /* 2488 * Just like if_promisc(), but for all-multicast-reception mode. 2489 */ 2490 int 2491 if_allmulti(struct ifnet *ifp, int onswitch) 2492 { 2493 int error = 0; 2494 struct ifreq ifr; 2495 2496 crit_enter(); 2497 2498 if (onswitch) { 2499 if (ifp->if_amcount++ == 0) { 2500 ifp->if_flags |= IFF_ALLMULTI; 2501 ifr.ifr_flags = ifp->if_flags; 2502 ifr.ifr_flagshigh = ifp->if_flags >> 16; 2503 ifnet_serialize_all(ifp); 2504 error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr, 2505 NULL); 2506 ifnet_deserialize_all(ifp); 2507 } 2508 } else { 2509 if (ifp->if_amcount > 1) { 2510 ifp->if_amcount--; 2511 } else { 2512 ifp->if_amcount = 0; 2513 ifp->if_flags &= ~IFF_ALLMULTI; 2514 ifr.ifr_flags = ifp->if_flags; 2515 ifr.ifr_flagshigh = ifp->if_flags >> 16; 2516 ifnet_serialize_all(ifp); 2517 error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr, 2518 NULL); 2519 ifnet_deserialize_all(ifp); 2520 } 2521 } 2522 2523 crit_exit(); 2524 2525 if (error == 0) 2526 rt_ifmsg(ifp); 2527 return error; 2528 } 2529 2530 /* 2531 * Add a multicast listenership to the interface in question. 2532 * The link layer provides a routine which converts 2533 */ 2534 int 2535 if_addmulti_serialized(struct ifnet *ifp, struct sockaddr *sa, 2536 struct ifmultiaddr **retifma) 2537 { 2538 struct sockaddr *llsa, *dupsa; 2539 int error; 2540 struct ifmultiaddr *ifma; 2541 2542 ASSERT_IFNET_SERIALIZED_ALL(ifp); 2543 2544 /* 2545 * If the matching multicast address already exists 2546 * then don't add a new one, just add a reference 2547 */ 2548 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 2549 if (sa_equal(sa, ifma->ifma_addr)) { 2550 ifma->ifma_refcount++; 2551 if (retifma) 2552 *retifma = ifma; 2553 return 0; 2554 } 2555 } 2556 2557 /* 2558 * Give the link layer a chance to accept/reject it, and also 2559 * find out which AF_LINK address this maps to, if it isn't one 2560 * already. 2561 */ 2562 if (ifp->if_resolvemulti) { 2563 error = ifp->if_resolvemulti(ifp, &llsa, sa); 2564 if (error) 2565 return error; 2566 } else { 2567 llsa = NULL; 2568 } 2569 2570 ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_INTWAIT); 2571 dupsa = kmalloc(sa->sa_len, M_IFMADDR, M_INTWAIT); 2572 bcopy(sa, dupsa, sa->sa_len); 2573 2574 ifma->ifma_addr = dupsa; 2575 ifma->ifma_lladdr = llsa; 2576 ifma->ifma_ifp = ifp; 2577 ifma->ifma_refcount = 1; 2578 ifma->ifma_protospec = NULL; 2579 rt_newmaddrmsg(RTM_NEWMADDR, ifma); 2580 2581 TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link); 2582 if (retifma) 2583 *retifma = ifma; 2584 2585 if (llsa != NULL) { 2586 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 2587 if (sa_equal(ifma->ifma_addr, llsa)) 2588 break; 2589 } 2590 if (ifma) { 2591 ifma->ifma_refcount++; 2592 } else { 2593 ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_INTWAIT); 2594 dupsa = kmalloc(llsa->sa_len, M_IFMADDR, M_INTWAIT); 2595 bcopy(llsa, dupsa, llsa->sa_len); 2596 ifma->ifma_addr = dupsa; 2597 ifma->ifma_ifp = ifp; 2598 ifma->ifma_refcount = 1; 2599 TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link); 2600 } 2601 } 2602 /* 2603 * We are certain we have added something, so call down to the 2604 * interface to let them know about it. 2605 */ 2606 if (ifp->if_ioctl) 2607 ifp->if_ioctl(ifp, SIOCADDMULTI, 0, NULL); 2608 2609 return 0; 2610 } 2611 2612 int 2613 if_addmulti(struct ifnet *ifp, struct sockaddr *sa, 2614 struct ifmultiaddr **retifma) 2615 { 2616 int error; 2617 2618 ifnet_serialize_all(ifp); 2619 error = if_addmulti_serialized(ifp, sa, retifma); 2620 ifnet_deserialize_all(ifp); 2621 2622 return error; 2623 } 2624 2625 /* 2626 * Remove a reference to a multicast address on this interface. Yell 2627 * if the request does not match an existing membership. 2628 */ 2629 static int 2630 if_delmulti_serialized(struct ifnet *ifp, struct sockaddr *sa) 2631 { 2632 struct ifmultiaddr *ifma; 2633 2634 ASSERT_IFNET_SERIALIZED_ALL(ifp); 2635 2636 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) 2637 if (sa_equal(sa, ifma->ifma_addr)) 2638 break; 2639 if (ifma == NULL) 2640 return ENOENT; 2641 2642 if (ifma->ifma_refcount > 1) { 2643 ifma->ifma_refcount--; 2644 return 0; 2645 } 2646 2647 rt_newmaddrmsg(RTM_DELMADDR, ifma); 2648 sa = ifma->ifma_lladdr; 2649 TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link); 2650 /* 2651 * Make sure the interface driver is notified 2652 * in the case of a link layer mcast group being left. 2653 */ 2654 if (ifma->ifma_addr->sa_family == AF_LINK && sa == NULL) 2655 ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL); 2656 kfree(ifma->ifma_addr, M_IFMADDR); 2657 kfree(ifma, M_IFMADDR); 2658 if (sa == NULL) 2659 return 0; 2660 2661 /* 2662 * Now look for the link-layer address which corresponds to 2663 * this network address. It had been squirreled away in 2664 * ifma->ifma_lladdr for this purpose (so we don't have 2665 * to call ifp->if_resolvemulti() again), and we saved that 2666 * value in sa above. If some nasty deleted the 2667 * link-layer address out from underneath us, we can deal because 2668 * the address we stored was is not the same as the one which was 2669 * in the record for the link-layer address. (So we don't complain 2670 * in that case.) 2671 */ 2672 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) 2673 if (sa_equal(sa, ifma->ifma_addr)) 2674 break; 2675 if (ifma == NULL) 2676 return 0; 2677 2678 if (ifma->ifma_refcount > 1) { 2679 ifma->ifma_refcount--; 2680 return 0; 2681 } 2682 2683 TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link); 2684 ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL); 2685 kfree(ifma->ifma_addr, M_IFMADDR); 2686 kfree(sa, M_IFMADDR); 2687 kfree(ifma, M_IFMADDR); 2688 2689 return 0; 2690 } 2691 2692 int 2693 if_delmulti(struct ifnet *ifp, struct sockaddr *sa) 2694 { 2695 int error; 2696 2697 ifnet_serialize_all(ifp); 2698 error = if_delmulti_serialized(ifp, sa); 2699 ifnet_deserialize_all(ifp); 2700 2701 return error; 2702 } 2703 2704 /* 2705 * Delete all multicast group membership for an interface. 2706 * Should be used to quickly flush all multicast filters. 2707 */ 2708 void 2709 if_delallmulti_serialized(struct ifnet *ifp) 2710 { 2711 struct ifmultiaddr *ifma, mark; 2712 struct sockaddr sa; 2713 2714 ASSERT_IFNET_SERIALIZED_ALL(ifp); 2715 2716 bzero(&sa, sizeof(sa)); 2717 sa.sa_family = AF_UNSPEC; 2718 sa.sa_len = sizeof(sa); 2719 2720 bzero(&mark, sizeof(mark)); 2721 mark.ifma_addr = &sa; 2722 2723 TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, &mark, ifma_link); 2724 while ((ifma = TAILQ_NEXT(&mark, ifma_link)) != NULL) { 2725 TAILQ_REMOVE(&ifp->if_multiaddrs, &mark, ifma_link); 2726 TAILQ_INSERT_AFTER(&ifp->if_multiaddrs, ifma, &mark, 2727 ifma_link); 2728 2729 if (ifma->ifma_addr->sa_family == AF_UNSPEC) 2730 continue; 2731 2732 if_delmulti_serialized(ifp, ifma->ifma_addr); 2733 } 2734 TAILQ_REMOVE(&ifp->if_multiaddrs, &mark, ifma_link); 2735 } 2736 2737 2738 /* 2739 * Set the link layer address on an interface. 2740 * 2741 * At this time we only support certain types of interfaces, 2742 * and we don't allow the length of the address to change. 2743 */ 2744 int 2745 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len) 2746 { 2747 struct sockaddr_dl *sdl; 2748 struct ifreq ifr; 2749 2750 sdl = IF_LLSOCKADDR(ifp); 2751 if (sdl == NULL) 2752 return (EINVAL); 2753 if (len != sdl->sdl_alen) /* don't allow length to change */ 2754 return (EINVAL); 2755 switch (ifp->if_type) { 2756 case IFT_ETHER: /* these types use struct arpcom */ 2757 case IFT_XETHER: 2758 case IFT_L2VLAN: 2759 case IFT_IEEE8023ADLAG: 2760 bcopy(lladdr, ((struct arpcom *)ifp->if_softc)->ac_enaddr, len); 2761 bcopy(lladdr, LLADDR(sdl), len); 2762 break; 2763 default: 2764 return (ENODEV); 2765 } 2766 /* 2767 * If the interface is already up, we need 2768 * to re-init it in order to reprogram its 2769 * address filter. 2770 */ 2771 ifnet_serialize_all(ifp); 2772 if ((ifp->if_flags & IFF_UP) != 0) { 2773 #ifdef INET 2774 struct ifaddr_container *ifac; 2775 #endif 2776 2777 ifp->if_flags &= ~IFF_UP; 2778 ifr.ifr_flags = ifp->if_flags; 2779 ifr.ifr_flagshigh = ifp->if_flags >> 16; 2780 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr, 2781 NULL); 2782 ifp->if_flags |= IFF_UP; 2783 ifr.ifr_flags = ifp->if_flags; 2784 ifr.ifr_flagshigh = ifp->if_flags >> 16; 2785 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr, 2786 NULL); 2787 #ifdef INET 2788 /* 2789 * Also send gratuitous ARPs to notify other nodes about 2790 * the address change. 2791 */ 2792 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) { 2793 struct ifaddr *ifa = ifac->ifa; 2794 2795 if (ifa->ifa_addr != NULL && 2796 ifa->ifa_addr->sa_family == AF_INET) 2797 arp_gratuitous(ifp, ifa); 2798 } 2799 #endif 2800 } 2801 ifnet_deserialize_all(ifp); 2802 return (0); 2803 } 2804 2805 2806 /* 2807 * Locate an interface based on a complete address. 2808 */ 2809 struct ifnet * 2810 if_bylla(const void *lla, unsigned char lla_len) 2811 { 2812 const struct ifnet_array *arr; 2813 struct ifnet *ifp; 2814 struct sockaddr_dl *sdl; 2815 int i; 2816 2817 arr = ifnet_array_get(); 2818 for (i = 0; i < arr->ifnet_count; ++i) { 2819 ifp = arr->ifnet_arr[i]; 2820 if (ifp->if_addrlen != lla_len) 2821 continue; 2822 2823 sdl = IF_LLSOCKADDR(ifp); 2824 if (memcmp(lla, LLADDR(sdl), lla_len) == 0) 2825 return (ifp); 2826 } 2827 return (NULL); 2828 } 2829 2830 struct ifmultiaddr * 2831 ifmaof_ifpforaddr(struct sockaddr *sa, struct ifnet *ifp) 2832 { 2833 struct ifmultiaddr *ifma; 2834 2835 /* TODO: need ifnet_serialize_main */ 2836 ifnet_serialize_all(ifp); 2837 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) 2838 if (sa_equal(ifma->ifma_addr, sa)) 2839 break; 2840 ifnet_deserialize_all(ifp); 2841 2842 return ifma; 2843 } 2844 2845 /* 2846 * This function locates the first real ethernet MAC from a network 2847 * card and loads it into node, returning 0 on success or ENOENT if 2848 * no suitable interfaces were found. It is used by the uuid code to 2849 * generate a unique 6-byte number. 2850 */ 2851 int 2852 if_getanyethermac(uint16_t *node, int minlen) 2853 { 2854 struct ifnet *ifp; 2855 struct sockaddr_dl *sdl; 2856 2857 ifnet_lock(); 2858 TAILQ_FOREACH(ifp, &ifnetlist, if_link) { 2859 if (ifp->if_type != IFT_ETHER) 2860 continue; 2861 sdl = IF_LLSOCKADDR(ifp); 2862 if (sdl->sdl_alen < minlen) 2863 continue; 2864 bcopy(((struct arpcom *)ifp->if_softc)->ac_enaddr, node, 2865 minlen); 2866 ifnet_unlock(); 2867 return(0); 2868 } 2869 ifnet_unlock(); 2870 return (ENOENT); 2871 } 2872 2873 /* 2874 * The name argument must be a pointer to storage which will last as 2875 * long as the interface does. For physical devices, the result of 2876 * device_get_name(dev) is a good choice and for pseudo-devices a 2877 * static string works well. 2878 */ 2879 void 2880 if_initname(struct ifnet *ifp, const char *name, int unit) 2881 { 2882 ifp->if_dname = name; 2883 ifp->if_dunit = unit; 2884 if (unit != IF_DUNIT_NONE) 2885 ksnprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit); 2886 else 2887 strlcpy(ifp->if_xname, name, IFNAMSIZ); 2888 } 2889 2890 int 2891 if_printf(struct ifnet *ifp, const char *fmt, ...) 2892 { 2893 __va_list ap; 2894 int retval; 2895 2896 retval = kprintf("%s: ", ifp->if_xname); 2897 __va_start(ap, fmt); 2898 retval += kvprintf(fmt, ap); 2899 __va_end(ap); 2900 return (retval); 2901 } 2902 2903 struct ifnet * 2904 if_alloc(uint8_t type) 2905 { 2906 struct ifnet *ifp; 2907 size_t size; 2908 2909 /* 2910 * XXX temporary hack until arpcom is setup in if_l2com 2911 */ 2912 if (type == IFT_ETHER) 2913 size = sizeof(struct arpcom); 2914 else 2915 size = sizeof(struct ifnet); 2916 2917 ifp = kmalloc(size, M_IFNET, M_WAITOK|M_ZERO); 2918 2919 ifp->if_type = type; 2920 2921 if (if_com_alloc[type] != NULL) { 2922 ifp->if_l2com = if_com_alloc[type](type, ifp); 2923 if (ifp->if_l2com == NULL) { 2924 kfree(ifp, M_IFNET); 2925 return (NULL); 2926 } 2927 } 2928 return (ifp); 2929 } 2930 2931 void 2932 if_free(struct ifnet *ifp) 2933 { 2934 kfree(ifp, M_IFNET); 2935 } 2936 2937 void 2938 ifq_set_classic(struct ifaltq *ifq) 2939 { 2940 ifq_set_methods(ifq, ifq->altq_ifp->if_mapsubq, 2941 ifsq_classic_enqueue, ifsq_classic_dequeue, ifsq_classic_request); 2942 } 2943 2944 void 2945 ifq_set_methods(struct ifaltq *ifq, altq_mapsubq_t mapsubq, 2946 ifsq_enqueue_t enqueue, ifsq_dequeue_t dequeue, ifsq_request_t request) 2947 { 2948 int q; 2949 2950 KASSERT(mapsubq != NULL, ("mapsubq is not specified")); 2951 KASSERT(enqueue != NULL, ("enqueue is not specified")); 2952 KASSERT(dequeue != NULL, ("dequeue is not specified")); 2953 KASSERT(request != NULL, ("request is not specified")); 2954 2955 ifq->altq_mapsubq = mapsubq; 2956 for (q = 0; q < ifq->altq_subq_cnt; ++q) { 2957 struct ifaltq_subque *ifsq = &ifq->altq_subq[q]; 2958 2959 ifsq->ifsq_enqueue = enqueue; 2960 ifsq->ifsq_dequeue = dequeue; 2961 ifsq->ifsq_request = request; 2962 } 2963 } 2964 2965 static void 2966 ifsq_norm_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m) 2967 { 2968 2969 classq_add(&ifsq->ifsq_norm, m); 2970 ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len); 2971 } 2972 2973 static void 2974 ifsq_prio_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m) 2975 { 2976 2977 classq_add(&ifsq->ifsq_prio, m); 2978 ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len); 2979 ALTQ_SQ_PRIO_CNTR_INC(ifsq, m->m_pkthdr.len); 2980 } 2981 2982 static struct mbuf * 2983 ifsq_norm_dequeue(struct ifaltq_subque *ifsq) 2984 { 2985 struct mbuf *m; 2986 2987 m = classq_get(&ifsq->ifsq_norm); 2988 if (m != NULL) 2989 ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len); 2990 return (m); 2991 } 2992 2993 static struct mbuf * 2994 ifsq_prio_dequeue(struct ifaltq_subque *ifsq) 2995 { 2996 struct mbuf *m; 2997 2998 m = classq_get(&ifsq->ifsq_prio); 2999 if (m != NULL) { 3000 ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len); 3001 ALTQ_SQ_PRIO_CNTR_DEC(ifsq, m->m_pkthdr.len); 3002 } 3003 return (m); 3004 } 3005 3006 int 3007 ifsq_classic_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m, 3008 struct altq_pktattr *pa __unused) 3009 { 3010 3011 M_ASSERTPKTHDR(m); 3012 again: 3013 if (ifsq->ifsq_len >= ifsq->ifsq_maxlen || 3014 ifsq->ifsq_bcnt >= ifsq->ifsq_maxbcnt) { 3015 struct mbuf *m_drop; 3016 3017 if (m->m_flags & M_PRIO) { 3018 m_drop = NULL; 3019 if (ifsq->ifsq_prio_len < (ifsq->ifsq_maxlen >> 1) && 3020 ifsq->ifsq_prio_bcnt < (ifsq->ifsq_maxbcnt >> 1)) { 3021 /* Try dropping some from normal queue. */ 3022 m_drop = ifsq_norm_dequeue(ifsq); 3023 } 3024 if (m_drop == NULL) 3025 m_drop = ifsq_prio_dequeue(ifsq); 3026 } else { 3027 m_drop = ifsq_norm_dequeue(ifsq); 3028 } 3029 if (m_drop != NULL) { 3030 IFNET_STAT_INC(ifsq->ifsq_ifp, oqdrops, 1); 3031 m_freem(m_drop); 3032 goto again; 3033 } 3034 /* 3035 * No old packets could be dropped! 3036 * NOTE: Caller increases oqdrops. 3037 */ 3038 m_freem(m); 3039 return (ENOBUFS); 3040 } else { 3041 if (m->m_flags & M_PRIO) 3042 ifsq_prio_enqueue(ifsq, m); 3043 else 3044 ifsq_norm_enqueue(ifsq, m); 3045 return (0); 3046 } 3047 } 3048 3049 struct mbuf * 3050 ifsq_classic_dequeue(struct ifaltq_subque *ifsq, int op) 3051 { 3052 struct mbuf *m; 3053 3054 switch (op) { 3055 case ALTDQ_POLL: 3056 m = classq_head(&ifsq->ifsq_prio); 3057 if (m == NULL) 3058 m = classq_head(&ifsq->ifsq_norm); 3059 break; 3060 3061 case ALTDQ_REMOVE: 3062 m = ifsq_prio_dequeue(ifsq); 3063 if (m == NULL) 3064 m = ifsq_norm_dequeue(ifsq); 3065 break; 3066 3067 default: 3068 panic("unsupported ALTQ dequeue op: %d", op); 3069 } 3070 return m; 3071 } 3072 3073 int 3074 ifsq_classic_request(struct ifaltq_subque *ifsq, int req, void *arg) 3075 { 3076 switch (req) { 3077 case ALTRQ_PURGE: 3078 for (;;) { 3079 struct mbuf *m; 3080 3081 m = ifsq_classic_dequeue(ifsq, ALTDQ_REMOVE); 3082 if (m == NULL) 3083 break; 3084 m_freem(m); 3085 } 3086 break; 3087 3088 default: 3089 panic("unsupported ALTQ request: %d", req); 3090 } 3091 return 0; 3092 } 3093 3094 static void 3095 ifsq_ifstart_try(struct ifaltq_subque *ifsq, int force_sched) 3096 { 3097 struct ifnet *ifp = ifsq_get_ifp(ifsq); 3098 int running = 0, need_sched; 3099 3100 /* 3101 * Try to do direct ifnet.if_start on the subqueue first, if there is 3102 * contention on the subqueue hardware serializer, ifnet.if_start on 3103 * the subqueue will be scheduled on the subqueue owner CPU. 3104 */ 3105 if (!ifsq_tryserialize_hw(ifsq)) { 3106 /* 3107 * Subqueue hardware serializer contention happened, 3108 * ifnet.if_start on the subqueue is scheduled on 3109 * the subqueue owner CPU, and we keep going. 3110 */ 3111 ifsq_ifstart_schedule(ifsq, 1); 3112 return; 3113 } 3114 3115 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) { 3116 ifp->if_start(ifp, ifsq); 3117 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) 3118 running = 1; 3119 } 3120 need_sched = ifsq_ifstart_need_schedule(ifsq, running); 3121 3122 ifsq_deserialize_hw(ifsq); 3123 3124 if (need_sched) { 3125 /* 3126 * More data need to be transmitted, ifnet.if_start on the 3127 * subqueue is scheduled on the subqueue owner CPU, and we 3128 * keep going. 3129 * NOTE: ifnet.if_start subqueue interlock is not released. 3130 */ 3131 ifsq_ifstart_schedule(ifsq, force_sched); 3132 } 3133 } 3134 3135 /* 3136 * Subqeue packets staging mechanism: 3137 * 3138 * The packets enqueued into the subqueue are staged to a certain amount 3139 * before the ifnet.if_start on the subqueue is called. In this way, the 3140 * driver could avoid writing to hardware registers upon every packet, 3141 * instead, hardware registers could be written when certain amount of 3142 * packets are put onto hardware TX ring. The measurement on several modern 3143 * NICs (emx(4), igb(4), bnx(4), bge(4), jme(4)) shows that the hardware 3144 * registers writing aggregation could save ~20% CPU time when 18bytes UDP 3145 * datagrams are transmitted at 1.48Mpps. The performance improvement by 3146 * hardware registers writing aggeregation is also mentioned by Luigi Rizzo's 3147 * netmap paper (http://info.iet.unipi.it/~luigi/netmap/). 3148 * 3149 * Subqueue packets staging is performed for two entry points into drivers' 3150 * transmission function: 3151 * - Direct ifnet.if_start calling on the subqueue, i.e. ifsq_ifstart_try() 3152 * - ifnet.if_start scheduling on the subqueue, i.e. ifsq_ifstart_schedule() 3153 * 3154 * Subqueue packets staging will be stopped upon any of the following 3155 * conditions: 3156 * - If the count of packets enqueued on the current CPU is great than or 3157 * equal to ifsq_stage_cntmax. (XXX this should be per-interface) 3158 * - If the total length of packets enqueued on the current CPU is great 3159 * than or equal to the hardware's MTU - max_protohdr. max_protohdr is 3160 * cut from the hardware's MTU mainly bacause a full TCP segment's size 3161 * is usually less than hardware's MTU. 3162 * - ifsq_ifstart_schedule() is not pending on the current CPU and 3163 * ifnet.if_start subqueue interlock (ifaltq_subq.ifsq_started) is not 3164 * released. 3165 * - The if_start_rollup(), which is registered as low priority netisr 3166 * rollup function, is called; probably because no more work is pending 3167 * for netisr. 3168 * 3169 * NOTE: 3170 * Currently subqueue packet staging is only performed in netisr threads. 3171 */ 3172 int 3173 ifq_dispatch(struct ifnet *ifp, struct mbuf *m, struct altq_pktattr *pa) 3174 { 3175 struct ifaltq *ifq = &ifp->if_snd; 3176 struct ifaltq_subque *ifsq; 3177 int error, start = 0, len, mcast = 0, avoid_start = 0; 3178 struct ifsubq_stage_head *head = NULL; 3179 struct ifsubq_stage *stage = NULL; 3180 struct globaldata *gd = mycpu; 3181 struct thread *td = gd->gd_curthread; 3182 3183 crit_enter_quick(td); 3184 3185 ifsq = ifq_map_subq(ifq, gd->gd_cpuid); 3186 ASSERT_ALTQ_SQ_NOT_SERIALIZED_HW(ifsq); 3187 3188 len = m->m_pkthdr.len; 3189 if (m->m_flags & M_MCAST) 3190 mcast = 1; 3191 3192 if (td->td_type == TD_TYPE_NETISR) { 3193 head = &ifsubq_stage_heads[mycpuid]; 3194 stage = ifsq_get_stage(ifsq, mycpuid); 3195 3196 stage->stg_cnt++; 3197 stage->stg_len += len; 3198 if (stage->stg_cnt < ifsq_stage_cntmax && 3199 stage->stg_len < (ifp->if_mtu - max_protohdr)) 3200 avoid_start = 1; 3201 } 3202 3203 ALTQ_SQ_LOCK(ifsq); 3204 error = ifsq_enqueue_locked(ifsq, m, pa); 3205 if (error) { 3206 IFNET_STAT_INC(ifp, oqdrops, 1); 3207 if (!ifsq_data_ready(ifsq)) { 3208 ALTQ_SQ_UNLOCK(ifsq); 3209 crit_exit_quick(td); 3210 return error; 3211 } 3212 avoid_start = 0; 3213 } 3214 if (!ifsq_is_started(ifsq)) { 3215 if (avoid_start) { 3216 ALTQ_SQ_UNLOCK(ifsq); 3217 3218 KKASSERT(!error); 3219 if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0) 3220 ifsq_stage_insert(head, stage); 3221 3222 IFNET_STAT_INC(ifp, obytes, len); 3223 if (mcast) 3224 IFNET_STAT_INC(ifp, omcasts, 1); 3225 crit_exit_quick(td); 3226 return error; 3227 } 3228 3229 /* 3230 * Hold the subqueue interlock of ifnet.if_start 3231 */ 3232 ifsq_set_started(ifsq); 3233 start = 1; 3234 } 3235 ALTQ_SQ_UNLOCK(ifsq); 3236 3237 if (!error) { 3238 IFNET_STAT_INC(ifp, obytes, len); 3239 if (mcast) 3240 IFNET_STAT_INC(ifp, omcasts, 1); 3241 } 3242 3243 if (stage != NULL) { 3244 if (!start && (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)) { 3245 KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED); 3246 if (!avoid_start) { 3247 ifsq_stage_remove(head, stage); 3248 ifsq_ifstart_schedule(ifsq, 1); 3249 } 3250 crit_exit_quick(td); 3251 return error; 3252 } 3253 3254 if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED) { 3255 ifsq_stage_remove(head, stage); 3256 } else { 3257 stage->stg_cnt = 0; 3258 stage->stg_len = 0; 3259 } 3260 } 3261 3262 if (!start) { 3263 crit_exit_quick(td); 3264 return error; 3265 } 3266 3267 ifsq_ifstart_try(ifsq, 0); 3268 3269 crit_exit_quick(td); 3270 return error; 3271 } 3272 3273 void * 3274 ifa_create(int size) 3275 { 3276 struct ifaddr *ifa; 3277 int i; 3278 3279 KASSERT(size >= sizeof(*ifa), ("ifaddr size too small")); 3280 3281 ifa = kmalloc(size, M_IFADDR, M_INTWAIT | M_ZERO); 3282 3283 /* 3284 * Make ifa_container availabel on all CPUs, since they 3285 * could be accessed by any threads. 3286 */ 3287 ifa->ifa_containers = 3288 kmalloc(ncpus * sizeof(struct ifaddr_container), 3289 M_IFADDR, 3290 M_INTWAIT | M_ZERO | M_CACHEALIGN); 3291 3292 ifa->ifa_ncnt = ncpus; 3293 for (i = 0; i < ncpus; ++i) { 3294 struct ifaddr_container *ifac = &ifa->ifa_containers[i]; 3295 3296 ifac->ifa_magic = IFA_CONTAINER_MAGIC; 3297 ifac->ifa = ifa; 3298 ifac->ifa_refcnt = 1; 3299 } 3300 #ifdef IFADDR_DEBUG 3301 kprintf("alloc ifa %p %d\n", ifa, size); 3302 #endif 3303 return ifa; 3304 } 3305 3306 void 3307 ifac_free(struct ifaddr_container *ifac, int cpu_id) 3308 { 3309 struct ifaddr *ifa = ifac->ifa; 3310 3311 KKASSERT(ifac->ifa_magic == IFA_CONTAINER_MAGIC); 3312 KKASSERT(ifac->ifa_refcnt == 0); 3313 KASSERT(ifac->ifa_listmask == 0, 3314 ("ifa is still on %#x lists", ifac->ifa_listmask)); 3315 3316 ifac->ifa_magic = IFA_CONTAINER_DEAD; 3317 3318 #ifdef IFADDR_DEBUG_VERBOSE 3319 kprintf("try free ifa %p cpu_id %d\n", ifac->ifa, cpu_id); 3320 #endif 3321 3322 KASSERT(ifa->ifa_ncnt > 0 && ifa->ifa_ncnt <= ncpus, 3323 ("invalid # of ifac, %d", ifa->ifa_ncnt)); 3324 if (atomic_fetchadd_int(&ifa->ifa_ncnt, -1) == 1) { 3325 #ifdef IFADDR_DEBUG 3326 kprintf("free ifa %p\n", ifa); 3327 #endif 3328 kfree(ifa->ifa_containers, M_IFADDR); 3329 kfree(ifa, M_IFADDR); 3330 } 3331 } 3332 3333 static void 3334 ifa_iflink_dispatch(netmsg_t nmsg) 3335 { 3336 struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg; 3337 struct ifaddr *ifa = msg->ifa; 3338 struct ifnet *ifp = msg->ifp; 3339 int cpu = mycpuid; 3340 struct ifaddr_container *ifac; 3341 3342 crit_enter(); 3343 3344 ifac = &ifa->ifa_containers[cpu]; 3345 ASSERT_IFAC_VALID(ifac); 3346 KASSERT((ifac->ifa_listmask & IFA_LIST_IFADDRHEAD) == 0, 3347 ("ifaddr is on if_addrheads")); 3348 3349 ifac->ifa_listmask |= IFA_LIST_IFADDRHEAD; 3350 if (msg->tail) 3351 TAILQ_INSERT_TAIL(&ifp->if_addrheads[cpu], ifac, ifa_link); 3352 else 3353 TAILQ_INSERT_HEAD(&ifp->if_addrheads[cpu], ifac, ifa_link); 3354 3355 crit_exit(); 3356 3357 netisr_forwardmsg_all(&nmsg->base, cpu + 1); 3358 } 3359 3360 void 3361 ifa_iflink(struct ifaddr *ifa, struct ifnet *ifp, int tail) 3362 { 3363 struct netmsg_ifaddr msg; 3364 3365 netmsg_init(&msg.base, NULL, &curthread->td_msgport, 3366 0, ifa_iflink_dispatch); 3367 msg.ifa = ifa; 3368 msg.ifp = ifp; 3369 msg.tail = tail; 3370 3371 netisr_domsg(&msg.base, 0); 3372 } 3373 3374 static void 3375 ifa_ifunlink_dispatch(netmsg_t nmsg) 3376 { 3377 struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg; 3378 struct ifaddr *ifa = msg->ifa; 3379 struct ifnet *ifp = msg->ifp; 3380 int cpu = mycpuid; 3381 struct ifaddr_container *ifac; 3382 3383 crit_enter(); 3384 3385 ifac = &ifa->ifa_containers[cpu]; 3386 ASSERT_IFAC_VALID(ifac); 3387 KASSERT(ifac->ifa_listmask & IFA_LIST_IFADDRHEAD, 3388 ("ifaddr is not on if_addrhead")); 3389 3390 TAILQ_REMOVE(&ifp->if_addrheads[cpu], ifac, ifa_link); 3391 ifac->ifa_listmask &= ~IFA_LIST_IFADDRHEAD; 3392 3393 crit_exit(); 3394 3395 netisr_forwardmsg_all(&nmsg->base, cpu + 1); 3396 } 3397 3398 void 3399 ifa_ifunlink(struct ifaddr *ifa, struct ifnet *ifp) 3400 { 3401 struct netmsg_ifaddr msg; 3402 3403 netmsg_init(&msg.base, NULL, &curthread->td_msgport, 3404 0, ifa_ifunlink_dispatch); 3405 msg.ifa = ifa; 3406 msg.ifp = ifp; 3407 3408 netisr_domsg(&msg.base, 0); 3409 } 3410 3411 static void 3412 ifa_destroy_dispatch(netmsg_t nmsg) 3413 { 3414 struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg; 3415 3416 IFAFREE(msg->ifa); 3417 netisr_forwardmsg_all(&nmsg->base, mycpuid + 1); 3418 } 3419 3420 void 3421 ifa_destroy(struct ifaddr *ifa) 3422 { 3423 struct netmsg_ifaddr msg; 3424 3425 netmsg_init(&msg.base, NULL, &curthread->td_msgport, 3426 0, ifa_destroy_dispatch); 3427 msg.ifa = ifa; 3428 3429 netisr_domsg(&msg.base, 0); 3430 } 3431 3432 static void 3433 if_start_rollup(void) 3434 { 3435 struct ifsubq_stage_head *head = &ifsubq_stage_heads[mycpuid]; 3436 struct ifsubq_stage *stage; 3437 3438 crit_enter(); 3439 3440 while ((stage = TAILQ_FIRST(&head->stg_head)) != NULL) { 3441 struct ifaltq_subque *ifsq = stage->stg_subq; 3442 int is_sched = 0; 3443 3444 if (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED) 3445 is_sched = 1; 3446 ifsq_stage_remove(head, stage); 3447 3448 if (is_sched) { 3449 ifsq_ifstart_schedule(ifsq, 1); 3450 } else { 3451 int start = 0; 3452 3453 ALTQ_SQ_LOCK(ifsq); 3454 if (!ifsq_is_started(ifsq)) { 3455 /* 3456 * Hold the subqueue interlock of 3457 * ifnet.if_start 3458 */ 3459 ifsq_set_started(ifsq); 3460 start = 1; 3461 } 3462 ALTQ_SQ_UNLOCK(ifsq); 3463 3464 if (start) 3465 ifsq_ifstart_try(ifsq, 1); 3466 } 3467 KKASSERT((stage->stg_flags & 3468 (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0); 3469 } 3470 3471 crit_exit(); 3472 } 3473 3474 static void 3475 ifnetinit(void *dummy __unused) 3476 { 3477 int i; 3478 3479 /* XXX netisr_ncpus */ 3480 for (i = 0; i < ncpus; ++i) 3481 TAILQ_INIT(&ifsubq_stage_heads[i].stg_head); 3482 netisr_register_rollup(if_start_rollup, NETISR_ROLLUP_PRIO_IFSTART); 3483 } 3484 3485 void 3486 if_register_com_alloc(u_char type, 3487 if_com_alloc_t *a, if_com_free_t *f) 3488 { 3489 3490 KASSERT(if_com_alloc[type] == NULL, 3491 ("if_register_com_alloc: %d already registered", type)); 3492 KASSERT(if_com_free[type] == NULL, 3493 ("if_register_com_alloc: %d free already registered", type)); 3494 3495 if_com_alloc[type] = a; 3496 if_com_free[type] = f; 3497 } 3498 3499 void 3500 if_deregister_com_alloc(u_char type) 3501 { 3502 3503 KASSERT(if_com_alloc[type] != NULL, 3504 ("if_deregister_com_alloc: %d not registered", type)); 3505 KASSERT(if_com_free[type] != NULL, 3506 ("if_deregister_com_alloc: %d free not registered", type)); 3507 if_com_alloc[type] = NULL; 3508 if_com_free[type] = NULL; 3509 } 3510 3511 void 3512 ifq_set_maxlen(struct ifaltq *ifq, int len) 3513 { 3514 ifq->altq_maxlen = len + (ncpus * ifsq_stage_cntmax); 3515 } 3516 3517 int 3518 ifq_mapsubq_default(struct ifaltq *ifq __unused, int cpuid __unused) 3519 { 3520 return ALTQ_SUBQ_INDEX_DEFAULT; 3521 } 3522 3523 int 3524 ifq_mapsubq_modulo(struct ifaltq *ifq, int cpuid) 3525 { 3526 3527 return (cpuid % ifq->altq_subq_mappriv); 3528 } 3529 3530 static void 3531 ifsq_watchdog(void *arg) 3532 { 3533 struct ifsubq_watchdog *wd = arg; 3534 struct ifnet *ifp; 3535 3536 if (__predict_true(wd->wd_timer == 0 || --wd->wd_timer)) 3537 goto done; 3538 3539 ifp = ifsq_get_ifp(wd->wd_subq); 3540 if (ifnet_tryserialize_all(ifp)) { 3541 wd->wd_watchdog(wd->wd_subq); 3542 ifnet_deserialize_all(ifp); 3543 } else { 3544 /* try again next timeout */ 3545 wd->wd_timer = 1; 3546 } 3547 done: 3548 ifsq_watchdog_reset(wd); 3549 } 3550 3551 static void 3552 ifsq_watchdog_reset(struct ifsubq_watchdog *wd) 3553 { 3554 callout_reset_bycpu(&wd->wd_callout, hz, ifsq_watchdog, wd, 3555 ifsq_get_cpuid(wd->wd_subq)); 3556 } 3557 3558 void 3559 ifsq_watchdog_init(struct ifsubq_watchdog *wd, struct ifaltq_subque *ifsq, 3560 ifsq_watchdog_t watchdog) 3561 { 3562 callout_init_mp(&wd->wd_callout); 3563 wd->wd_timer = 0; 3564 wd->wd_subq = ifsq; 3565 wd->wd_watchdog = watchdog; 3566 } 3567 3568 void 3569 ifsq_watchdog_start(struct ifsubq_watchdog *wd) 3570 { 3571 wd->wd_timer = 0; 3572 ifsq_watchdog_reset(wd); 3573 } 3574 3575 void 3576 ifsq_watchdog_stop(struct ifsubq_watchdog *wd) 3577 { 3578 wd->wd_timer = 0; 3579 callout_stop(&wd->wd_callout); 3580 } 3581 3582 void 3583 ifnet_lock(void) 3584 { 3585 KASSERT(curthread->td_type != TD_TYPE_NETISR, 3586 ("try holding ifnet lock in netisr")); 3587 mtx_lock(&ifnet_mtx); 3588 } 3589 3590 void 3591 ifnet_unlock(void) 3592 { 3593 KASSERT(curthread->td_type != TD_TYPE_NETISR, 3594 ("try holding ifnet lock in netisr")); 3595 mtx_unlock(&ifnet_mtx); 3596 } 3597 3598 static struct ifnet_array * 3599 ifnet_array_alloc(int count) 3600 { 3601 struct ifnet_array *arr; 3602 3603 arr = kmalloc(__offsetof(struct ifnet_array, ifnet_arr[count]), 3604 M_IFNET, M_WAITOK); 3605 arr->ifnet_count = count; 3606 3607 return arr; 3608 } 3609 3610 static void 3611 ifnet_array_free(struct ifnet_array *arr) 3612 { 3613 if (arr == &ifnet_array0) 3614 return; 3615 kfree(arr, M_IFNET); 3616 } 3617 3618 static struct ifnet_array * 3619 ifnet_array_add(struct ifnet *ifp, const struct ifnet_array *old_arr) 3620 { 3621 struct ifnet_array *arr; 3622 int count, i; 3623 3624 KASSERT(old_arr->ifnet_count >= 0, 3625 ("invalid ifnet array count %d", old_arr->ifnet_count)); 3626 count = old_arr->ifnet_count + 1; 3627 arr = ifnet_array_alloc(count); 3628 3629 /* 3630 * Save the old ifnet array and append this ifp to the end of 3631 * the new ifnet array. 3632 */ 3633 for (i = 0; i < old_arr->ifnet_count; ++i) { 3634 KASSERT(old_arr->ifnet_arr[i] != ifp, 3635 ("%s is already in ifnet array", ifp->if_xname)); 3636 arr->ifnet_arr[i] = old_arr->ifnet_arr[i]; 3637 } 3638 KASSERT(i == count - 1, 3639 ("add %s, ifnet array index mismatch, should be %d, but got %d", 3640 ifp->if_xname, count - 1, i)); 3641 arr->ifnet_arr[i] = ifp; 3642 3643 return arr; 3644 } 3645 3646 static struct ifnet_array * 3647 ifnet_array_del(struct ifnet *ifp, const struct ifnet_array *old_arr) 3648 { 3649 struct ifnet_array *arr; 3650 int count, i, idx, found = 0; 3651 3652 KASSERT(old_arr->ifnet_count > 0, 3653 ("invalid ifnet array count %d", old_arr->ifnet_count)); 3654 count = old_arr->ifnet_count - 1; 3655 arr = ifnet_array_alloc(count); 3656 3657 /* 3658 * Save the old ifnet array, but skip this ifp. 3659 */ 3660 idx = 0; 3661 for (i = 0; i < old_arr->ifnet_count; ++i) { 3662 if (old_arr->ifnet_arr[i] == ifp) { 3663 KASSERT(!found, 3664 ("dup %s is in ifnet array", ifp->if_xname)); 3665 found = 1; 3666 continue; 3667 } 3668 KASSERT(idx < count, 3669 ("invalid ifnet array index %d, count %d", idx, count)); 3670 arr->ifnet_arr[idx] = old_arr->ifnet_arr[i]; 3671 ++idx; 3672 } 3673 KASSERT(found, ("%s is not in ifnet array", ifp->if_xname)); 3674 KASSERT(idx == count, 3675 ("del %s, ifnet array count mismatch, should be %d, but got %d ", 3676 ifp->if_xname, count, idx)); 3677 3678 return arr; 3679 } 3680 3681 const struct ifnet_array * 3682 ifnet_array_get(void) 3683 { 3684 const struct ifnet_array *ret; 3685 3686 KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr")); 3687 ret = ifnet_array; 3688 /* Make sure 'ret' is really used. */ 3689 cpu_ccfence(); 3690 return (ret); 3691 } 3692 3693 int 3694 ifnet_array_isempty(void) 3695 { 3696 KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr")); 3697 if (ifnet_array->ifnet_count == 0) 3698 return 1; 3699 else 3700 return 0; 3701 } 3702 3703 void 3704 ifa_marker_init(struct ifaddr_marker *mark, struct ifnet *ifp) 3705 { 3706 struct ifaddr *ifa; 3707 3708 memset(mark, 0, sizeof(*mark)); 3709 ifa = &mark->ifa; 3710 3711 mark->ifac.ifa = ifa; 3712 3713 ifa->ifa_addr = &mark->addr; 3714 ifa->ifa_dstaddr = &mark->dstaddr; 3715 ifa->ifa_netmask = &mark->netmask; 3716 ifa->ifa_ifp = ifp; 3717 } 3718 3719 static int 3720 if_ringcnt_fixup(int ring_cnt, int ring_cntmax) 3721 { 3722 3723 KASSERT(ring_cntmax > 0, ("invalid ring count max %d", ring_cntmax)); 3724 3725 if (ring_cnt <= 0 || ring_cnt > ring_cntmax) 3726 ring_cnt = ring_cntmax; 3727 if (ring_cnt > netisr_ncpus) 3728 ring_cnt = netisr_ncpus; 3729 return (ring_cnt); 3730 } 3731 3732 static void 3733 if_ringmap_set_grid(device_t dev, struct if_ringmap *rm, int grid) 3734 { 3735 int i, offset; 3736 3737 KASSERT(grid > 0, ("invalid if_ringmap grid %d", grid)); 3738 KASSERT(grid >= rm->rm_cnt, ("invalid if_ringmap grid %d, count %d", 3739 grid, rm->rm_cnt)); 3740 rm->rm_grid = grid; 3741 3742 offset = (rm->rm_grid * device_get_unit(dev)) % netisr_ncpus; 3743 for (i = 0; i < rm->rm_cnt; ++i) { 3744 rm->rm_cpumap[i] = offset + i; 3745 KASSERT(rm->rm_cpumap[i] < netisr_ncpus, 3746 ("invalid cpumap[%d] = %d, offset %d", i, 3747 rm->rm_cpumap[i], offset)); 3748 } 3749 } 3750 3751 static struct if_ringmap * 3752 if_ringmap_alloc_flags(device_t dev, int ring_cnt, int ring_cntmax, 3753 uint32_t flags) 3754 { 3755 struct if_ringmap *rm; 3756 int i, grid = 0, prev_grid; 3757 3758 ring_cnt = if_ringcnt_fixup(ring_cnt, ring_cntmax); 3759 rm = kmalloc(__offsetof(struct if_ringmap, rm_cpumap[ring_cnt]), 3760 M_DEVBUF, M_WAITOK | M_ZERO); 3761 3762 rm->rm_cnt = ring_cnt; 3763 if (flags & RINGMAP_FLAG_POWEROF2) 3764 rm->rm_cnt = 1 << (fls(rm->rm_cnt) - 1); 3765 3766 prev_grid = netisr_ncpus; 3767 for (i = 0; i < netisr_ncpus; ++i) { 3768 if (netisr_ncpus % (i + 1) != 0) 3769 continue; 3770 3771 grid = netisr_ncpus / (i + 1); 3772 if (rm->rm_cnt > grid) { 3773 grid = prev_grid; 3774 break; 3775 } 3776 3777 if (rm->rm_cnt > netisr_ncpus / (i + 2)) 3778 break; 3779 prev_grid = grid; 3780 } 3781 if_ringmap_set_grid(dev, rm, grid); 3782 3783 return (rm); 3784 } 3785 3786 struct if_ringmap * 3787 if_ringmap_alloc(device_t dev, int ring_cnt, int ring_cntmax) 3788 { 3789 3790 return (if_ringmap_alloc_flags(dev, ring_cnt, ring_cntmax, 3791 RINGMAP_FLAG_NONE)); 3792 } 3793 3794 struct if_ringmap * 3795 if_ringmap_alloc2(device_t dev, int ring_cnt, int ring_cntmax) 3796 { 3797 3798 return (if_ringmap_alloc_flags(dev, ring_cnt, ring_cntmax, 3799 RINGMAP_FLAG_POWEROF2)); 3800 } 3801 3802 void 3803 if_ringmap_free(struct if_ringmap *rm) 3804 { 3805 3806 kfree(rm, M_DEVBUF); 3807 } 3808 3809 /* 3810 * Align the two ringmaps. 3811 * 3812 * e.g. 8 netisrs, rm0 contains 4 rings, rm1 contains 2 rings. 3813 * 3814 * Before: 3815 * 3816 * CPU 0 1 2 3 4 5 6 7 3817 * NIC_RX n0 n1 n2 n3 3818 * NIC_TX N0 N1 3819 * 3820 * After: 3821 * 3822 * CPU 0 1 2 3 4 5 6 7 3823 * NIC_RX n0 n1 n2 n3 3824 * NIC_TX N0 N1 3825 */ 3826 void 3827 if_ringmap_align(device_t dev, struct if_ringmap *rm0, struct if_ringmap *rm1) 3828 { 3829 3830 if (rm0->rm_grid > rm1->rm_grid) 3831 if_ringmap_set_grid(dev, rm1, rm0->rm_grid); 3832 else if (rm0->rm_grid < rm1->rm_grid) 3833 if_ringmap_set_grid(dev, rm0, rm1->rm_grid); 3834 } 3835 3836 void 3837 if_ringmap_match(device_t dev, struct if_ringmap *rm0, struct if_ringmap *rm1) 3838 { 3839 int subset_grid, cnt, divisor, mod, offset, i; 3840 struct if_ringmap *subset_rm, *rm; 3841 int old_rm0_grid, old_rm1_grid; 3842 3843 if (rm0->rm_grid == rm1->rm_grid) 3844 return; 3845 3846 /* Save grid for later use */ 3847 old_rm0_grid = rm0->rm_grid; 3848 old_rm1_grid = rm1->rm_grid; 3849 3850 if_ringmap_align(dev, rm0, rm1); 3851 3852 /* 3853 * Re-shuffle rings to get more even distribution. 3854 * 3855 * e.g. 12 netisrs, rm0 contains 4 rings, rm1 contains 2 rings. 3856 * 3857 * CPU 0 1 2 3 4 5 6 7 8 9 10 11 3858 * 3859 * NIC_RX a0 a1 a2 a3 b0 b1 b2 b3 c0 c1 c2 c3 3860 * NIC_TX A0 A1 B0 B1 C0 C1 3861 * 3862 * NIC_RX d0 d1 d2 d3 e0 e1 e2 e3 f0 f1 f2 f3 3863 * NIC_TX D0 D1 E0 E1 F0 F1 3864 */ 3865 3866 if (rm0->rm_cnt >= (2 * old_rm1_grid)) { 3867 cnt = rm0->rm_cnt; 3868 subset_grid = old_rm1_grid; 3869 subset_rm = rm1; 3870 rm = rm0; 3871 } else if (rm1->rm_cnt > (2 * old_rm0_grid)) { 3872 cnt = rm1->rm_cnt; 3873 subset_grid = old_rm0_grid; 3874 subset_rm = rm0; 3875 rm = rm1; 3876 } else { 3877 /* No space to shuffle. */ 3878 return; 3879 } 3880 3881 mod = cnt / subset_grid; 3882 KKASSERT(mod >= 2); 3883 divisor = netisr_ncpus / rm->rm_grid; 3884 offset = ((device_get_unit(dev) / divisor) % mod) * subset_grid; 3885 3886 for (i = 0; i < subset_rm->rm_cnt; ++i) { 3887 subset_rm->rm_cpumap[i] += offset; 3888 KASSERT(subset_rm->rm_cpumap[i] < netisr_ncpus, 3889 ("match: invalid cpumap[%d] = %d, offset %d", 3890 i, subset_rm->rm_cpumap[i], offset)); 3891 } 3892 #ifdef INVARIANTS 3893 for (i = 0; i < subset_rm->rm_cnt; ++i) { 3894 int j; 3895 3896 for (j = 0; j < rm->rm_cnt; ++j) { 3897 if (rm->rm_cpumap[j] == subset_rm->rm_cpumap[i]) 3898 break; 3899 } 3900 KASSERT(j < rm->rm_cnt, 3901 ("subset cpumap[%d] = %d not found in superset", 3902 i, subset_rm->rm_cpumap[i])); 3903 } 3904 #endif 3905 } 3906 3907 int 3908 if_ringmap_count(const struct if_ringmap *rm) 3909 { 3910 3911 return (rm->rm_cnt); 3912 } 3913 3914 int 3915 if_ringmap_cpumap(const struct if_ringmap *rm, int ring) 3916 { 3917 3918 KASSERT(ring >= 0 && ring < rm->rm_cnt, ("invalid ring %d", ring)); 3919 return (rm->rm_cpumap[ring]); 3920 } 3921 3922 void 3923 if_ringmap_rdrtable(const struct if_ringmap *rm, int table[], int table_nent) 3924 { 3925 int i, grid_idx, grid_cnt, patch_off, patch_cnt, ncopy; 3926 3927 KASSERT(table_nent > 0 && (table_nent & NETISR_CPUMASK) == 0, 3928 ("invalid redirect table entries %d", table_nent)); 3929 3930 grid_idx = 0; 3931 for (i = 0; i < NETISR_CPUMAX; ++i) { 3932 table[i] = grid_idx++ % rm->rm_cnt; 3933 3934 if (grid_idx == rm->rm_grid) 3935 grid_idx = 0; 3936 } 3937 3938 /* 3939 * Make the ring distributed more evenly for the remainder 3940 * of each grid. 3941 * 3942 * e.g. 12 netisrs, rm contains 8 rings. 3943 * 3944 * Redirect table before: 3945 * 3946 * 0 1 2 3 4 5 6 7 0 1 2 3 0 1 2 3 3947 * 4 5 6 7 0 1 2 3 0 1 2 3 4 5 6 7 3948 * 0 1 2 3 0 1 2 3 4 5 6 7 0 1 2 3 3949 * .... 3950 * 3951 * Redirect table after being patched (pX, patched entries): 3952 * 3953 * 0 1 2 3 4 5 6 7 p0 p1 p2 p3 0 1 2 3 3954 * 4 5 6 7 p4 p5 p6 p7 0 1 2 3 4 5 6 7 3955 * p0 p1 p2 p3 0 1 2 3 4 5 6 7 p4 p5 p6 p7 3956 * .... 3957 */ 3958 patch_cnt = rm->rm_grid % rm->rm_cnt; 3959 if (patch_cnt == 0) 3960 goto done; 3961 patch_off = rm->rm_grid - (rm->rm_grid % rm->rm_cnt); 3962 3963 grid_cnt = roundup(NETISR_CPUMAX, rm->rm_grid) / rm->rm_grid; 3964 grid_idx = 0; 3965 for (i = 0; i < grid_cnt; ++i) { 3966 int j; 3967 3968 for (j = 0; j < patch_cnt; ++j) { 3969 int fix_idx; 3970 3971 fix_idx = (i * rm->rm_grid) + patch_off + j; 3972 if (fix_idx >= NETISR_CPUMAX) 3973 goto done; 3974 table[fix_idx] = grid_idx++ % rm->rm_cnt; 3975 } 3976 } 3977 done: 3978 /* 3979 * If the device supports larger redirect table, duplicate 3980 * the first NETISR_CPUMAX entries to the rest of the table, 3981 * so that it matches upper layer's expectation: 3982 * (hash & NETISR_CPUMASK) % netisr_ncpus 3983 */ 3984 ncopy = table_nent / NETISR_CPUMAX; 3985 for (i = 1; i < ncopy; ++i) { 3986 memcpy(&table[i * NETISR_CPUMAX], table, 3987 NETISR_CPUMAX * sizeof(table[0])); 3988 } 3989 if (if_ringmap_dumprdr) { 3990 for (i = 0; i < table_nent; ++i) { 3991 if (i != 0 && i % 16 == 0) 3992 kprintf("\n"); 3993 kprintf("%03d ", table[i]); 3994 } 3995 kprintf("\n"); 3996 } 3997 } 3998 3999 int 4000 if_ringmap_cpumap_sysctl(SYSCTL_HANDLER_ARGS) 4001 { 4002 struct if_ringmap *rm = arg1; 4003 int i, error = 0; 4004 4005 for (i = 0; i < rm->rm_cnt; ++i) { 4006 int cpu = rm->rm_cpumap[i]; 4007 4008 error = SYSCTL_OUT(req, &cpu, sizeof(cpu)); 4009 if (error) 4010 break; 4011 } 4012 return (error); 4013 } 4014