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_compat.h" 34 #include "opt_inet6.h" 35 #include "opt_inet.h" 36 #include "opt_ifpoll.h" 37 38 #include <sys/param.h> 39 #include <sys/malloc.h> 40 #include <sys/mbuf.h> 41 #include <sys/systm.h> 42 #include <sys/proc.h> 43 #include <sys/priv.h> 44 #include <sys/protosw.h> 45 #include <sys/socket.h> 46 #include <sys/socketvar.h> 47 #include <sys/socketops.h> 48 #include <sys/kernel.h> 49 #include <sys/ktr.h> 50 #include <sys/mutex.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 59 #include <sys/thread2.h> 60 #include <sys/msgport2.h> 61 #include <sys/mutex2.h> 62 63 #include <net/if.h> 64 #include <net/if_arp.h> 65 #include <net/if_dl.h> 66 #include <net/if_types.h> 67 #include <net/if_var.h> 68 #include <net/ifq_var.h> 69 #include <net/radix.h> 70 #include <net/route.h> 71 #include <net/if_clone.h> 72 #include <net/netisr2.h> 73 #include <net/netmsg2.h> 74 75 #include <machine/atomic.h> 76 #include <machine/stdarg.h> 77 #include <machine/smp.h> 78 79 #if defined(INET) || defined(INET6) 80 /*XXX*/ 81 #include <netinet/in.h> 82 #include <netinet/in_var.h> 83 #include <netinet/if_ether.h> 84 #ifdef INET6 85 #include <netinet6/in6_var.h> 86 #include <netinet6/in6_ifattach.h> 87 #endif 88 #endif 89 90 #if defined(COMPAT_43) 91 #include <emulation/43bsd/43bsd_socket.h> 92 #endif /* COMPAT_43 */ 93 94 struct netmsg_ifaddr { 95 struct netmsg_base base; 96 struct ifaddr *ifa; 97 struct ifnet *ifp; 98 int tail; 99 }; 100 101 struct ifsubq_stage_head { 102 TAILQ_HEAD(, ifsubq_stage) stg_head; 103 } __cachealign; 104 105 /* 106 * System initialization 107 */ 108 static void if_attachdomain(void *); 109 static void if_attachdomain1(struct ifnet *); 110 static int ifconf(u_long, caddr_t, struct ucred *); 111 static void ifinit(void *); 112 static void ifnetinit(void *); 113 static void if_slowtimo(void *); 114 static void link_rtrequest(int, struct rtentry *); 115 static int if_rtdel(struct radix_node *, void *); 116 117 /* Helper functions */ 118 static void ifsq_watchdog_reset(struct ifsubq_watchdog *); 119 120 #ifdef INET6 121 /* 122 * XXX: declare here to avoid to include many inet6 related files.. 123 * should be more generalized? 124 */ 125 extern void nd6_setmtu(struct ifnet *); 126 #endif 127 128 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers"); 129 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management"); 130 131 static int ifsq_stage_cntmax = 4; 132 TUNABLE_INT("net.link.stage_cntmax", &ifsq_stage_cntmax); 133 SYSCTL_INT(_net_link, OID_AUTO, stage_cntmax, CTLFLAG_RW, 134 &ifsq_stage_cntmax, 0, "ifq staging packet count max"); 135 136 static int if_stats_compat = 0; 137 SYSCTL_INT(_net_link, OID_AUTO, stats_compat, CTLFLAG_RW, 138 &if_stats_compat, 0, "Compat the old ifnet stats"); 139 140 SYSINIT(interfaces, SI_SUB_PROTO_IF, SI_ORDER_FIRST, ifinit, NULL) 141 /* Must be after netisr_init */ 142 SYSINIT(ifnet, SI_SUB_PRE_DRIVERS, SI_ORDER_SECOND, ifnetinit, NULL) 143 144 static if_com_alloc_t *if_com_alloc[256]; 145 static if_com_free_t *if_com_free[256]; 146 147 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address"); 148 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address"); 149 MALLOC_DEFINE(M_IFNET, "ifnet", "interface structure"); 150 151 int ifqmaxlen = IFQ_MAXLEN; 152 struct ifnethead ifnet = TAILQ_HEAD_INITIALIZER(ifnet); 153 154 struct callout if_slowtimo_timer; 155 156 int if_index = 0; 157 struct ifnet **ifindex2ifnet = NULL; 158 static struct thread ifnet_threads[MAXCPU]; 159 160 static struct ifsubq_stage_head ifsubq_stage_heads[MAXCPU]; 161 162 #ifdef notyet 163 #define IFQ_KTR_STRING "ifq=%p" 164 #define IFQ_KTR_ARGS struct ifaltq *ifq 165 #ifndef KTR_IFQ 166 #define KTR_IFQ KTR_ALL 167 #endif 168 KTR_INFO_MASTER(ifq); 169 KTR_INFO(KTR_IFQ, ifq, enqueue, 0, IFQ_KTR_STRING, IFQ_KTR_ARGS); 170 KTR_INFO(KTR_IFQ, ifq, dequeue, 1, IFQ_KTR_STRING, IFQ_KTR_ARGS); 171 #define logifq(name, arg) KTR_LOG(ifq_ ## name, arg) 172 173 #define IF_START_KTR_STRING "ifp=%p" 174 #define IF_START_KTR_ARGS struct ifnet *ifp 175 #ifndef KTR_IF_START 176 #define KTR_IF_START KTR_ALL 177 #endif 178 KTR_INFO_MASTER(if_start); 179 KTR_INFO(KTR_IF_START, if_start, run, 0, 180 IF_START_KTR_STRING, IF_START_KTR_ARGS); 181 KTR_INFO(KTR_IF_START, if_start, sched, 1, 182 IF_START_KTR_STRING, IF_START_KTR_ARGS); 183 KTR_INFO(KTR_IF_START, if_start, avoid, 2, 184 IF_START_KTR_STRING, IF_START_KTR_ARGS); 185 KTR_INFO(KTR_IF_START, if_start, contend_sched, 3, 186 IF_START_KTR_STRING, IF_START_KTR_ARGS); 187 KTR_INFO(KTR_IF_START, if_start, chase_sched, 4, 188 IF_START_KTR_STRING, IF_START_KTR_ARGS); 189 #define logifstart(name, arg) KTR_LOG(if_start_ ## name, arg) 190 #endif 191 192 TAILQ_HEAD(, ifg_group) ifg_head = TAILQ_HEAD_INITIALIZER(ifg_head); 193 194 /* 195 * Network interface utility routines. 196 * 197 * Routines with ifa_ifwith* names take sockaddr *'s as 198 * parameters. 199 */ 200 /* ARGSUSED*/ 201 void 202 ifinit(void *dummy) 203 { 204 struct ifnet *ifp; 205 206 callout_init(&if_slowtimo_timer); 207 208 crit_enter(); 209 TAILQ_FOREACH(ifp, &ifnet, if_link) { 210 if (ifp->if_snd.altq_maxlen == 0) { 211 if_printf(ifp, "XXX: driver didn't set altq_maxlen\n"); 212 ifq_set_maxlen(&ifp->if_snd, ifqmaxlen); 213 } 214 } 215 crit_exit(); 216 217 if_slowtimo(0); 218 } 219 220 static void 221 ifsq_ifstart_ipifunc(void *arg) 222 { 223 struct ifaltq_subque *ifsq = arg; 224 struct lwkt_msg *lmsg = ifsq_get_ifstart_lmsg(ifsq, mycpuid); 225 226 crit_enter(); 227 if (lmsg->ms_flags & MSGF_DONE) 228 lwkt_sendmsg_oncpu(netisr_cpuport(mycpuid), lmsg); 229 crit_exit(); 230 } 231 232 static __inline void 233 ifsq_stage_remove(struct ifsubq_stage_head *head, struct ifsubq_stage *stage) 234 { 235 KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED); 236 TAILQ_REMOVE(&head->stg_head, stage, stg_link); 237 stage->stg_flags &= ~(IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED); 238 stage->stg_cnt = 0; 239 stage->stg_len = 0; 240 } 241 242 static __inline void 243 ifsq_stage_insert(struct ifsubq_stage_head *head, struct ifsubq_stage *stage) 244 { 245 KKASSERT((stage->stg_flags & 246 (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0); 247 stage->stg_flags |= IFSQ_STAGE_FLAG_QUED; 248 TAILQ_INSERT_TAIL(&head->stg_head, stage, stg_link); 249 } 250 251 /* 252 * Schedule ifnet.if_start on the subqueue owner CPU 253 */ 254 static void 255 ifsq_ifstart_schedule(struct ifaltq_subque *ifsq, int force) 256 { 257 int cpu; 258 259 if (!force && curthread->td_type == TD_TYPE_NETISR && 260 ifsq_stage_cntmax > 0) { 261 struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid); 262 263 stage->stg_cnt = 0; 264 stage->stg_len = 0; 265 if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0) 266 ifsq_stage_insert(&ifsubq_stage_heads[mycpuid], stage); 267 stage->stg_flags |= IFSQ_STAGE_FLAG_SCHED; 268 return; 269 } 270 271 cpu = ifsq_get_cpuid(ifsq); 272 if (cpu != mycpuid) 273 lwkt_send_ipiq(globaldata_find(cpu), ifsq_ifstart_ipifunc, ifsq); 274 else 275 ifsq_ifstart_ipifunc(ifsq); 276 } 277 278 /* 279 * NOTE: 280 * This function will release ifnet.if_start subqueue interlock, 281 * if ifnet.if_start for the subqueue does not need to be scheduled 282 */ 283 static __inline int 284 ifsq_ifstart_need_schedule(struct ifaltq_subque *ifsq, int running) 285 { 286 if (!running || ifsq_is_empty(ifsq) 287 #ifdef ALTQ 288 || ifsq->ifsq_altq->altq_tbr != NULL 289 #endif 290 ) { 291 ALTQ_SQ_LOCK(ifsq); 292 /* 293 * ifnet.if_start subqueue interlock is released, if: 294 * 1) Hardware can not take any packets, due to 295 * o interface is marked down 296 * o hardware queue is full (ifsq_is_oactive) 297 * Under the second situation, hardware interrupt 298 * or polling(4) will call/schedule ifnet.if_start 299 * on the subqueue when hardware queue is ready 300 * 2) There is no packet in the subqueue. 301 * Further ifq_dispatch or ifq_handoff will call/ 302 * schedule ifnet.if_start on the subqueue. 303 * 3) TBR is used and it does not allow further 304 * dequeueing. 305 * TBR callout will call ifnet.if_start on the 306 * subqueue. 307 */ 308 if (!running || !ifsq_data_ready(ifsq)) { 309 ifsq_clr_started(ifsq); 310 ALTQ_SQ_UNLOCK(ifsq); 311 return 0; 312 } 313 ALTQ_SQ_UNLOCK(ifsq); 314 } 315 return 1; 316 } 317 318 static void 319 ifsq_ifstart_dispatch(netmsg_t msg) 320 { 321 struct lwkt_msg *lmsg = &msg->base.lmsg; 322 struct ifaltq_subque *ifsq = lmsg->u.ms_resultp; 323 struct ifnet *ifp = ifsq_get_ifp(ifsq); 324 struct globaldata *gd = mycpu; 325 int running = 0, need_sched; 326 327 crit_enter_gd(gd); 328 329 lwkt_replymsg(lmsg, 0); /* reply ASAP */ 330 331 if (gd->gd_cpuid != ifsq_get_cpuid(ifsq)) { 332 /* 333 * We need to chase the subqueue owner CPU change. 334 */ 335 ifsq_ifstart_schedule(ifsq, 1); 336 crit_exit_gd(gd); 337 return; 338 } 339 340 ifsq_serialize_hw(ifsq); 341 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) { 342 ifp->if_start(ifp, ifsq); 343 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) 344 running = 1; 345 } 346 need_sched = ifsq_ifstart_need_schedule(ifsq, running); 347 ifsq_deserialize_hw(ifsq); 348 349 if (need_sched) { 350 /* 351 * More data need to be transmitted, ifnet.if_start is 352 * scheduled on the subqueue owner CPU, and we keep going. 353 * NOTE: ifnet.if_start subqueue interlock is not released. 354 */ 355 ifsq_ifstart_schedule(ifsq, 0); 356 } 357 358 crit_exit_gd(gd); 359 } 360 361 /* Device driver ifnet.if_start helper function */ 362 void 363 ifsq_devstart(struct ifaltq_subque *ifsq) 364 { 365 struct ifnet *ifp = ifsq_get_ifp(ifsq); 366 int running = 0; 367 368 ASSERT_ALTQ_SQ_SERIALIZED_HW(ifsq); 369 370 ALTQ_SQ_LOCK(ifsq); 371 if (ifsq_is_started(ifsq) || !ifsq_data_ready(ifsq)) { 372 ALTQ_SQ_UNLOCK(ifsq); 373 return; 374 } 375 ifsq_set_started(ifsq); 376 ALTQ_SQ_UNLOCK(ifsq); 377 378 ifp->if_start(ifp, ifsq); 379 380 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) 381 running = 1; 382 383 if (ifsq_ifstart_need_schedule(ifsq, running)) { 384 /* 385 * More data need to be transmitted, ifnet.if_start is 386 * scheduled on ifnet's CPU, and we keep going. 387 * NOTE: ifnet.if_start interlock is not released. 388 */ 389 ifsq_ifstart_schedule(ifsq, 0); 390 } 391 } 392 393 void 394 if_devstart(struct ifnet *ifp) 395 { 396 ifsq_devstart(ifq_get_subq_default(&ifp->if_snd)); 397 } 398 399 /* Device driver ifnet.if_start schedule helper function */ 400 void 401 ifsq_devstart_sched(struct ifaltq_subque *ifsq) 402 { 403 ifsq_ifstart_schedule(ifsq, 1); 404 } 405 406 void 407 if_devstart_sched(struct ifnet *ifp) 408 { 409 ifsq_devstart_sched(ifq_get_subq_default(&ifp->if_snd)); 410 } 411 412 static void 413 if_default_serialize(struct ifnet *ifp, enum ifnet_serialize slz __unused) 414 { 415 lwkt_serialize_enter(ifp->if_serializer); 416 } 417 418 static void 419 if_default_deserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused) 420 { 421 lwkt_serialize_exit(ifp->if_serializer); 422 } 423 424 static int 425 if_default_tryserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused) 426 { 427 return lwkt_serialize_try(ifp->if_serializer); 428 } 429 430 #ifdef INVARIANTS 431 static void 432 if_default_serialize_assert(struct ifnet *ifp, 433 enum ifnet_serialize slz __unused, 434 boolean_t serialized) 435 { 436 if (serialized) 437 ASSERT_SERIALIZED(ifp->if_serializer); 438 else 439 ASSERT_NOT_SERIALIZED(ifp->if_serializer); 440 } 441 #endif 442 443 /* 444 * Attach an interface to the list of "active" interfaces. 445 * 446 * The serializer is optional. 447 */ 448 void 449 if_attach(struct ifnet *ifp, lwkt_serialize_t serializer) 450 { 451 unsigned socksize, ifasize; 452 int namelen, masklen; 453 struct sockaddr_dl *sdl; 454 struct ifaddr *ifa; 455 struct ifaltq *ifq; 456 int i, q; 457 458 static int if_indexlim = 8; 459 460 if (ifp->if_serialize != NULL) { 461 KASSERT(ifp->if_deserialize != NULL && 462 ifp->if_tryserialize != NULL && 463 ifp->if_serialize_assert != NULL, 464 ("serialize functions are partially setup")); 465 466 /* 467 * If the device supplies serialize functions, 468 * then clear if_serializer to catch any invalid 469 * usage of this field. 470 */ 471 KASSERT(serializer == NULL, 472 ("both serialize functions and default serializer " 473 "are supplied")); 474 ifp->if_serializer = NULL; 475 } else { 476 KASSERT(ifp->if_deserialize == NULL && 477 ifp->if_tryserialize == NULL && 478 ifp->if_serialize_assert == NULL, 479 ("serialize functions are partially setup")); 480 ifp->if_serialize = if_default_serialize; 481 ifp->if_deserialize = if_default_deserialize; 482 ifp->if_tryserialize = if_default_tryserialize; 483 #ifdef INVARIANTS 484 ifp->if_serialize_assert = if_default_serialize_assert; 485 #endif 486 487 /* 488 * The serializer can be passed in from the device, 489 * allowing the same serializer to be used for both 490 * the interrupt interlock and the device queue. 491 * If not specified, the netif structure will use an 492 * embedded serializer. 493 */ 494 if (serializer == NULL) { 495 serializer = &ifp->if_default_serializer; 496 lwkt_serialize_init(serializer); 497 } 498 ifp->if_serializer = serializer; 499 } 500 501 mtx_init(&ifp->if_ioctl_mtx); 502 mtx_lock(&ifp->if_ioctl_mtx); 503 504 lwkt_gettoken(&ifnet_token); /* protect if_index and ifnet tailq */ 505 ifp->if_index = ++if_index; 506 507 /* 508 * XXX - 509 * The old code would work if the interface passed a pre-existing 510 * chain of ifaddrs to this code. We don't trust our callers to 511 * properly initialize the tailq, however, so we no longer allow 512 * this unlikely case. 513 */ 514 ifp->if_addrheads = kmalloc(ncpus * sizeof(struct ifaddrhead), 515 M_IFADDR, M_WAITOK | M_ZERO); 516 for (i = 0; i < ncpus; ++i) 517 TAILQ_INIT(&ifp->if_addrheads[i]); 518 519 TAILQ_INIT(&ifp->if_prefixhead); 520 TAILQ_INIT(&ifp->if_multiaddrs); 521 TAILQ_INIT(&ifp->if_groups); 522 getmicrotime(&ifp->if_lastchange); 523 if (ifindex2ifnet == NULL || if_index >= if_indexlim) { 524 unsigned int n; 525 struct ifnet **q; 526 527 if_indexlim <<= 1; 528 529 /* grow ifindex2ifnet */ 530 n = if_indexlim * sizeof(*q); 531 q = kmalloc(n, M_IFADDR, M_WAITOK | M_ZERO); 532 if (ifindex2ifnet) { 533 bcopy(ifindex2ifnet, q, n/2); 534 kfree(ifindex2ifnet, M_IFADDR); 535 } 536 ifindex2ifnet = q; 537 } 538 539 ifindex2ifnet[if_index] = ifp; 540 541 /* 542 * create a Link Level name for this device 543 */ 544 namelen = strlen(ifp->if_xname); 545 masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + namelen; 546 socksize = masklen + ifp->if_addrlen; 547 #define ROUNDUP(a) (1 + (((a) - 1) | (sizeof(long) - 1))) 548 if (socksize < sizeof(*sdl)) 549 socksize = sizeof(*sdl); 550 socksize = ROUNDUP(socksize); 551 #undef ROUNDUP 552 ifasize = sizeof(struct ifaddr) + 2 * socksize; 553 ifa = ifa_create(ifasize, M_WAITOK); 554 sdl = (struct sockaddr_dl *)(ifa + 1); 555 sdl->sdl_len = socksize; 556 sdl->sdl_family = AF_LINK; 557 bcopy(ifp->if_xname, sdl->sdl_data, namelen); 558 sdl->sdl_nlen = namelen; 559 sdl->sdl_index = ifp->if_index; 560 sdl->sdl_type = ifp->if_type; 561 ifp->if_lladdr = ifa; 562 ifa->ifa_ifp = ifp; 563 ifa->ifa_rtrequest = link_rtrequest; 564 ifa->ifa_addr = (struct sockaddr *)sdl; 565 sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl); 566 ifa->ifa_netmask = (struct sockaddr *)sdl; 567 sdl->sdl_len = masklen; 568 while (namelen != 0) 569 sdl->sdl_data[--namelen] = 0xff; 570 ifa_iflink(ifa, ifp, 0 /* Insert head */); 571 572 ifp->if_data_pcpu = kmalloc_cachealign( 573 ncpus * sizeof(struct ifdata_pcpu), M_DEVBUF, M_WAITOK | M_ZERO); 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 = kmalloc_cachealign( 588 ifq->altq_subq_cnt * sizeof(struct ifaltq_subque), 589 M_DEVBUF, M_WAITOK | M_ZERO); 590 591 if (ifq->altq_maxlen == 0) { 592 if_printf(ifp, "driver didn't set altq_maxlen\n"); 593 ifq_set_maxlen(ifq, ifqmaxlen); 594 } 595 596 for (q = 0; q < ifq->altq_subq_cnt; ++q) { 597 struct ifaltq_subque *ifsq = &ifq->altq_subq[q]; 598 599 ALTQ_SQ_LOCK_INIT(ifsq); 600 ifsq->ifsq_index = q; 601 602 ifsq->ifsq_altq = ifq; 603 ifsq->ifsq_ifp = ifp; 604 605 ifsq->ifsq_maxlen = ifq->altq_maxlen; 606 ifsq->ifsq_maxbcnt = ifsq->ifsq_maxlen * MCLBYTES; 607 ifsq->ifsq_prepended = NULL; 608 ifsq->ifsq_started = 0; 609 ifsq->ifsq_hw_oactive = 0; 610 ifsq_set_cpuid(ifsq, 0); 611 if (ifp->if_serializer != NULL) 612 ifsq_set_hw_serialize(ifsq, ifp->if_serializer); 613 614 ifsq->ifsq_stage = 615 kmalloc_cachealign(ncpus * sizeof(struct ifsubq_stage), 616 M_DEVBUF, M_WAITOK | M_ZERO); 617 for (i = 0; i < ncpus; ++i) 618 ifsq->ifsq_stage[i].stg_subq = ifsq; 619 620 ifsq->ifsq_ifstart_nmsg = 621 kmalloc(ncpus * sizeof(struct netmsg_base), 622 M_LWKTMSG, M_WAITOK); 623 for (i = 0; i < ncpus; ++i) { 624 netmsg_init(&ifsq->ifsq_ifstart_nmsg[i], NULL, 625 &netisr_adone_rport, 0, ifsq_ifstart_dispatch); 626 ifsq->ifsq_ifstart_nmsg[i].lmsg.u.ms_resultp = ifsq; 627 } 628 } 629 ifq_set_classic(ifq); 630 631 if (!SLIST_EMPTY(&domains)) 632 if_attachdomain1(ifp); 633 634 TAILQ_INSERT_TAIL(&ifnet, ifp, if_link); 635 lwkt_reltoken(&ifnet_token); 636 637 /* Announce the interface. */ 638 EVENTHANDLER_INVOKE(ifnet_attach_event, ifp); 639 devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL); 640 rt_ifannouncemsg(ifp, IFAN_ARRIVAL); 641 642 mtx_unlock(&ifp->if_ioctl_mtx); 643 } 644 645 static void 646 if_attachdomain(void *dummy) 647 { 648 struct ifnet *ifp; 649 650 crit_enter(); 651 TAILQ_FOREACH(ifp, &ifnet, if_list) 652 if_attachdomain1(ifp); 653 crit_exit(); 654 } 655 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_FIRST, 656 if_attachdomain, NULL); 657 658 static void 659 if_attachdomain1(struct ifnet *ifp) 660 { 661 struct domain *dp; 662 663 crit_enter(); 664 665 /* address family dependent data region */ 666 bzero(ifp->if_afdata, sizeof(ifp->if_afdata)); 667 SLIST_FOREACH(dp, &domains, dom_next) 668 if (dp->dom_ifattach) 669 ifp->if_afdata[dp->dom_family] = 670 (*dp->dom_ifattach)(ifp); 671 crit_exit(); 672 } 673 674 /* 675 * Purge all addresses whose type is _not_ AF_LINK 676 */ 677 void 678 if_purgeaddrs_nolink(struct ifnet *ifp) 679 { 680 struct ifaddr_container *ifac, *next; 681 682 TAILQ_FOREACH_MUTABLE(ifac, &ifp->if_addrheads[mycpuid], 683 ifa_link, next) { 684 struct ifaddr *ifa = ifac->ifa; 685 686 /* Leave link ifaddr as it is */ 687 if (ifa->ifa_addr->sa_family == AF_LINK) 688 continue; 689 #ifdef INET 690 /* XXX: Ugly!! ad hoc just for INET */ 691 if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET) { 692 struct ifaliasreq ifr; 693 #ifdef IFADDR_DEBUG_VERBOSE 694 int i; 695 696 kprintf("purge in4 addr %p: ", ifa); 697 for (i = 0; i < ncpus; ++i) 698 kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt); 699 kprintf("\n"); 700 #endif 701 702 bzero(&ifr, sizeof ifr); 703 ifr.ifra_addr = *ifa->ifa_addr; 704 if (ifa->ifa_dstaddr) 705 ifr.ifra_broadaddr = *ifa->ifa_dstaddr; 706 if (in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp, 707 NULL) == 0) 708 continue; 709 } 710 #endif /* INET */ 711 #ifdef INET6 712 if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET6) { 713 #ifdef IFADDR_DEBUG_VERBOSE 714 int i; 715 716 kprintf("purge in6 addr %p: ", ifa); 717 for (i = 0; i < ncpus; ++i) 718 kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt); 719 kprintf("\n"); 720 #endif 721 722 in6_purgeaddr(ifa); 723 /* ifp_addrhead is already updated */ 724 continue; 725 } 726 #endif /* INET6 */ 727 ifa_ifunlink(ifa, ifp); 728 ifa_destroy(ifa); 729 } 730 } 731 732 static void 733 ifq_stage_detach_handler(netmsg_t nmsg) 734 { 735 struct ifaltq *ifq = nmsg->lmsg.u.ms_resultp; 736 int q; 737 738 for (q = 0; q < ifq->altq_subq_cnt; ++q) { 739 struct ifaltq_subque *ifsq = &ifq->altq_subq[q]; 740 struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid); 741 742 if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED) 743 ifsq_stage_remove(&ifsubq_stage_heads[mycpuid], stage); 744 } 745 lwkt_replymsg(&nmsg->lmsg, 0); 746 } 747 748 static void 749 ifq_stage_detach(struct ifaltq *ifq) 750 { 751 struct netmsg_base base; 752 int cpu; 753 754 netmsg_init(&base, NULL, &curthread->td_msgport, 0, 755 ifq_stage_detach_handler); 756 base.lmsg.u.ms_resultp = ifq; 757 758 for (cpu = 0; cpu < ncpus; ++cpu) 759 lwkt_domsg(netisr_cpuport(cpu), &base.lmsg, 0); 760 } 761 762 struct netmsg_if_rtdel { 763 struct netmsg_base base; 764 struct ifnet *ifp; 765 }; 766 767 static void 768 if_rtdel_dispatch(netmsg_t msg) 769 { 770 struct netmsg_if_rtdel *rmsg = (void *)msg; 771 int i, nextcpu, cpu; 772 773 cpu = mycpuid; 774 for (i = 1; i <= AF_MAX; i++) { 775 struct radix_node_head *rnh; 776 777 if ((rnh = rt_tables[cpu][i]) == NULL) 778 continue; 779 rnh->rnh_walktree(rnh, if_rtdel, rmsg->ifp); 780 } 781 782 nextcpu = cpu + 1; 783 if (nextcpu < ncpus) 784 lwkt_forwardmsg(netisr_cpuport(nextcpu), &rmsg->base.lmsg); 785 else 786 lwkt_replymsg(&rmsg->base.lmsg, 0); 787 } 788 789 /* 790 * Detach an interface, removing it from the 791 * list of "active" interfaces. 792 */ 793 void 794 if_detach(struct ifnet *ifp) 795 { 796 struct netmsg_if_rtdel msg; 797 struct domain *dp; 798 int q; 799 800 EVENTHANDLER_INVOKE(ifnet_detach_event, ifp); 801 802 /* 803 * Remove routes and flush queues. 804 */ 805 crit_enter(); 806 #ifdef IFPOLL_ENABLE 807 if (ifp->if_flags & IFF_NPOLLING) 808 ifpoll_deregister(ifp); 809 #endif 810 if_down(ifp); 811 812 #ifdef ALTQ 813 if (ifq_is_enabled(&ifp->if_snd)) 814 altq_disable(&ifp->if_snd); 815 if (ifq_is_attached(&ifp->if_snd)) 816 altq_detach(&ifp->if_snd); 817 #endif 818 819 /* 820 * Clean up all addresses. 821 */ 822 ifp->if_lladdr = NULL; 823 824 if_purgeaddrs_nolink(ifp); 825 if (!TAILQ_EMPTY(&ifp->if_addrheads[mycpuid])) { 826 struct ifaddr *ifa; 827 828 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa; 829 KASSERT(ifa->ifa_addr->sa_family == AF_LINK, 830 ("non-link ifaddr is left on if_addrheads")); 831 832 ifa_ifunlink(ifa, ifp); 833 ifa_destroy(ifa); 834 KASSERT(TAILQ_EMPTY(&ifp->if_addrheads[mycpuid]), 835 ("there are still ifaddrs left on if_addrheads")); 836 } 837 838 #ifdef INET 839 /* 840 * Remove all IPv4 kernel structures related to ifp. 841 */ 842 in_ifdetach(ifp); 843 #endif 844 845 #ifdef INET6 846 /* 847 * Remove all IPv6 kernel structs related to ifp. This should be done 848 * before removing routing entries below, since IPv6 interface direct 849 * routes are expected to be removed by the IPv6-specific kernel API. 850 * Otherwise, the kernel will detect some inconsistency and bark it. 851 */ 852 in6_ifdetach(ifp); 853 #endif 854 855 /* 856 * Delete all remaining routes using this interface 857 */ 858 netmsg_init(&msg.base, NULL, &curthread->td_msgport, MSGF_PRIORITY, 859 if_rtdel_dispatch); 860 msg.ifp = ifp; 861 rt_domsg_global(&msg.base); 862 863 /* Announce that the interface is gone. */ 864 rt_ifannouncemsg(ifp, IFAN_DEPARTURE); 865 devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL); 866 867 SLIST_FOREACH(dp, &domains, dom_next) 868 if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family]) 869 (*dp->dom_ifdetach)(ifp, 870 ifp->if_afdata[dp->dom_family]); 871 872 /* 873 * Remove interface from ifindex2ifp[] and maybe decrement if_index. 874 */ 875 lwkt_gettoken(&ifnet_token); 876 ifindex2ifnet[ifp->if_index] = NULL; 877 while (if_index > 0 && ifindex2ifnet[if_index] == NULL) 878 if_index--; 879 TAILQ_REMOVE(&ifnet, ifp, if_link); 880 lwkt_reltoken(&ifnet_token); 881 882 kfree(ifp->if_addrheads, M_IFADDR); 883 884 lwkt_synchronize_ipiqs("if_detach"); 885 ifq_stage_detach(&ifp->if_snd); 886 887 for (q = 0; q < ifp->if_snd.altq_subq_cnt; ++q) { 888 struct ifaltq_subque *ifsq = &ifp->if_snd.altq_subq[q]; 889 890 kfree(ifsq->ifsq_ifstart_nmsg, M_LWKTMSG); 891 kfree(ifsq->ifsq_stage, M_DEVBUF); 892 } 893 kfree(ifp->if_snd.altq_subq, M_DEVBUF); 894 895 kfree(ifp->if_data_pcpu, M_DEVBUF); 896 897 crit_exit(); 898 } 899 900 /* 901 * Create interface group without members 902 */ 903 struct ifg_group * 904 if_creategroup(const char *groupname) 905 { 906 struct ifg_group *ifg = NULL; 907 908 if ((ifg = (struct ifg_group *)kmalloc(sizeof(struct ifg_group), 909 M_TEMP, M_NOWAIT)) == NULL) 910 return (NULL); 911 912 strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group)); 913 ifg->ifg_refcnt = 0; 914 ifg->ifg_carp_demoted = 0; 915 TAILQ_INIT(&ifg->ifg_members); 916 #if NPF > 0 917 pfi_attach_ifgroup(ifg); 918 #endif 919 TAILQ_INSERT_TAIL(&ifg_head, ifg, ifg_next); 920 921 return (ifg); 922 } 923 924 /* 925 * Add a group to an interface 926 */ 927 int 928 if_addgroup(struct ifnet *ifp, const char *groupname) 929 { 930 struct ifg_list *ifgl; 931 struct ifg_group *ifg = NULL; 932 struct ifg_member *ifgm; 933 934 if (groupname[0] && groupname[strlen(groupname) - 1] >= '0' && 935 groupname[strlen(groupname) - 1] <= '9') 936 return (EINVAL); 937 938 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) 939 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname)) 940 return (EEXIST); 941 942 if ((ifgl = kmalloc(sizeof(*ifgl), M_TEMP, M_NOWAIT)) == NULL) 943 return (ENOMEM); 944 945 if ((ifgm = kmalloc(sizeof(*ifgm), M_TEMP, M_NOWAIT)) == NULL) { 946 kfree(ifgl, M_TEMP); 947 return (ENOMEM); 948 } 949 950 TAILQ_FOREACH(ifg, &ifg_head, ifg_next) 951 if (!strcmp(ifg->ifg_group, groupname)) 952 break; 953 954 if (ifg == NULL && (ifg = if_creategroup(groupname)) == NULL) { 955 kfree(ifgl, M_TEMP); 956 kfree(ifgm, M_TEMP); 957 return (ENOMEM); 958 } 959 960 ifg->ifg_refcnt++; 961 ifgl->ifgl_group = ifg; 962 ifgm->ifgm_ifp = ifp; 963 964 TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next); 965 TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next); 966 967 #if NPF > 0 968 pfi_group_change(groupname); 969 #endif 970 971 return (0); 972 } 973 974 /* 975 * Remove a group from an interface 976 */ 977 int 978 if_delgroup(struct ifnet *ifp, const char *groupname) 979 { 980 struct ifg_list *ifgl; 981 struct ifg_member *ifgm; 982 983 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) 984 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname)) 985 break; 986 if (ifgl == NULL) 987 return (ENOENT); 988 989 TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next); 990 991 TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next) 992 if (ifgm->ifgm_ifp == ifp) 993 break; 994 995 if (ifgm != NULL) { 996 TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next); 997 kfree(ifgm, M_TEMP); 998 } 999 1000 if (--ifgl->ifgl_group->ifg_refcnt == 0) { 1001 TAILQ_REMOVE(&ifg_head, ifgl->ifgl_group, ifg_next); 1002 #if NPF > 0 1003 pfi_detach_ifgroup(ifgl->ifgl_group); 1004 #endif 1005 kfree(ifgl->ifgl_group, M_TEMP); 1006 } 1007 1008 kfree(ifgl, M_TEMP); 1009 1010 #if NPF > 0 1011 pfi_group_change(groupname); 1012 #endif 1013 1014 return (0); 1015 } 1016 1017 /* 1018 * Stores all groups from an interface in memory pointed 1019 * to by data 1020 */ 1021 int 1022 if_getgroup(caddr_t data, struct ifnet *ifp) 1023 { 1024 int len, error; 1025 struct ifg_list *ifgl; 1026 struct ifg_req ifgrq, *ifgp; 1027 struct ifgroupreq *ifgr = (struct ifgroupreq *)data; 1028 1029 if (ifgr->ifgr_len == 0) { 1030 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) 1031 ifgr->ifgr_len += sizeof(struct ifg_req); 1032 return (0); 1033 } 1034 1035 len = ifgr->ifgr_len; 1036 ifgp = ifgr->ifgr_groups; 1037 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) { 1038 if (len < sizeof(ifgrq)) 1039 return (EINVAL); 1040 bzero(&ifgrq, sizeof ifgrq); 1041 strlcpy(ifgrq.ifgrq_group, ifgl->ifgl_group->ifg_group, 1042 sizeof(ifgrq.ifgrq_group)); 1043 if ((error = copyout((caddr_t)&ifgrq, (caddr_t)ifgp, 1044 sizeof(struct ifg_req)))) 1045 return (error); 1046 len -= sizeof(ifgrq); 1047 ifgp++; 1048 } 1049 1050 return (0); 1051 } 1052 1053 /* 1054 * Stores all members of a group in memory pointed to by data 1055 */ 1056 int 1057 if_getgroupmembers(caddr_t data) 1058 { 1059 struct ifgroupreq *ifgr = (struct ifgroupreq *)data; 1060 struct ifg_group *ifg; 1061 struct ifg_member *ifgm; 1062 struct ifg_req ifgrq, *ifgp; 1063 int len, error; 1064 1065 TAILQ_FOREACH(ifg, &ifg_head, ifg_next) 1066 if (!strcmp(ifg->ifg_group, ifgr->ifgr_name)) 1067 break; 1068 if (ifg == NULL) 1069 return (ENOENT); 1070 1071 if (ifgr->ifgr_len == 0) { 1072 TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) 1073 ifgr->ifgr_len += sizeof(ifgrq); 1074 return (0); 1075 } 1076 1077 len = ifgr->ifgr_len; 1078 ifgp = ifgr->ifgr_groups; 1079 TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) { 1080 if (len < sizeof(ifgrq)) 1081 return (EINVAL); 1082 bzero(&ifgrq, sizeof ifgrq); 1083 strlcpy(ifgrq.ifgrq_member, ifgm->ifgm_ifp->if_xname, 1084 sizeof(ifgrq.ifgrq_member)); 1085 if ((error = copyout((caddr_t)&ifgrq, (caddr_t)ifgp, 1086 sizeof(struct ifg_req)))) 1087 return (error); 1088 len -= sizeof(ifgrq); 1089 ifgp++; 1090 } 1091 1092 return (0); 1093 } 1094 1095 /* 1096 * Delete Routes for a Network Interface 1097 * 1098 * Called for each routing entry via the rnh->rnh_walktree() call above 1099 * to delete all route entries referencing a detaching network interface. 1100 * 1101 * Arguments: 1102 * rn pointer to node in the routing table 1103 * arg argument passed to rnh->rnh_walktree() - detaching interface 1104 * 1105 * Returns: 1106 * 0 successful 1107 * errno failed - reason indicated 1108 * 1109 */ 1110 static int 1111 if_rtdel(struct radix_node *rn, void *arg) 1112 { 1113 struct rtentry *rt = (struct rtentry *)rn; 1114 struct ifnet *ifp = arg; 1115 int err; 1116 1117 if (rt->rt_ifp == ifp) { 1118 1119 /* 1120 * Protect (sorta) against walktree recursion problems 1121 * with cloned routes 1122 */ 1123 if (!(rt->rt_flags & RTF_UP)) 1124 return (0); 1125 1126 err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway, 1127 rt_mask(rt), rt->rt_flags, 1128 NULL); 1129 if (err) { 1130 log(LOG_WARNING, "if_rtdel: error %d\n", err); 1131 } 1132 } 1133 1134 return (0); 1135 } 1136 1137 /* 1138 * Locate an interface based on a complete address. 1139 */ 1140 struct ifaddr * 1141 ifa_ifwithaddr(struct sockaddr *addr) 1142 { 1143 struct ifnet *ifp; 1144 1145 TAILQ_FOREACH(ifp, &ifnet, if_link) { 1146 struct ifaddr_container *ifac; 1147 1148 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) { 1149 struct ifaddr *ifa = ifac->ifa; 1150 1151 if (ifa->ifa_addr->sa_family != addr->sa_family) 1152 continue; 1153 if (sa_equal(addr, ifa->ifa_addr)) 1154 return (ifa); 1155 if ((ifp->if_flags & IFF_BROADCAST) && 1156 ifa->ifa_broadaddr && 1157 /* IPv6 doesn't have broadcast */ 1158 ifa->ifa_broadaddr->sa_len != 0 && 1159 sa_equal(ifa->ifa_broadaddr, addr)) 1160 return (ifa); 1161 } 1162 } 1163 return (NULL); 1164 } 1165 /* 1166 * Locate the point to point interface with a given destination address. 1167 */ 1168 struct ifaddr * 1169 ifa_ifwithdstaddr(struct sockaddr *addr) 1170 { 1171 struct ifnet *ifp; 1172 1173 TAILQ_FOREACH(ifp, &ifnet, if_link) { 1174 struct ifaddr_container *ifac; 1175 1176 if (!(ifp->if_flags & IFF_POINTOPOINT)) 1177 continue; 1178 1179 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) { 1180 struct ifaddr *ifa = ifac->ifa; 1181 1182 if (ifa->ifa_addr->sa_family != addr->sa_family) 1183 continue; 1184 if (ifa->ifa_dstaddr && 1185 sa_equal(addr, ifa->ifa_dstaddr)) 1186 return (ifa); 1187 } 1188 } 1189 return (NULL); 1190 } 1191 1192 /* 1193 * Find an interface on a specific network. If many, choice 1194 * is most specific found. 1195 */ 1196 struct ifaddr * 1197 ifa_ifwithnet(struct sockaddr *addr) 1198 { 1199 struct ifnet *ifp; 1200 struct ifaddr *ifa_maybe = NULL; 1201 u_int af = addr->sa_family; 1202 char *addr_data = addr->sa_data, *cplim; 1203 1204 /* 1205 * AF_LINK addresses can be looked up directly by their index number, 1206 * so do that if we can. 1207 */ 1208 if (af == AF_LINK) { 1209 struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr; 1210 1211 if (sdl->sdl_index && sdl->sdl_index <= if_index) 1212 return (ifindex2ifnet[sdl->sdl_index]->if_lladdr); 1213 } 1214 1215 /* 1216 * Scan though each interface, looking for ones that have 1217 * addresses in this address family. 1218 */ 1219 TAILQ_FOREACH(ifp, &ifnet, if_link) { 1220 struct ifaddr_container *ifac; 1221 1222 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) { 1223 struct ifaddr *ifa = ifac->ifa; 1224 char *cp, *cp2, *cp3; 1225 1226 if (ifa->ifa_addr->sa_family != af) 1227 next: continue; 1228 if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT) { 1229 /* 1230 * This is a bit broken as it doesn't 1231 * take into account that the remote end may 1232 * be a single node in the network we are 1233 * looking for. 1234 * The trouble is that we don't know the 1235 * netmask for the remote end. 1236 */ 1237 if (ifa->ifa_dstaddr != NULL && 1238 sa_equal(addr, ifa->ifa_dstaddr)) 1239 return (ifa); 1240 } else { 1241 /* 1242 * if we have a special address handler, 1243 * then use it instead of the generic one. 1244 */ 1245 if (ifa->ifa_claim_addr) { 1246 if ((*ifa->ifa_claim_addr)(ifa, addr)) { 1247 return (ifa); 1248 } else { 1249 continue; 1250 } 1251 } 1252 1253 /* 1254 * Scan all the bits in the ifa's address. 1255 * If a bit dissagrees with what we are 1256 * looking for, mask it with the netmask 1257 * to see if it really matters. 1258 * (A byte at a time) 1259 */ 1260 if (ifa->ifa_netmask == 0) 1261 continue; 1262 cp = addr_data; 1263 cp2 = ifa->ifa_addr->sa_data; 1264 cp3 = ifa->ifa_netmask->sa_data; 1265 cplim = ifa->ifa_netmask->sa_len + 1266 (char *)ifa->ifa_netmask; 1267 while (cp3 < cplim) 1268 if ((*cp++ ^ *cp2++) & *cp3++) 1269 goto next; /* next address! */ 1270 /* 1271 * If the netmask of what we just found 1272 * is more specific than what we had before 1273 * (if we had one) then remember the new one 1274 * before continuing to search 1275 * for an even better one. 1276 */ 1277 if (ifa_maybe == NULL || 1278 rn_refines((char *)ifa->ifa_netmask, 1279 (char *)ifa_maybe->ifa_netmask)) 1280 ifa_maybe = ifa; 1281 } 1282 } 1283 } 1284 return (ifa_maybe); 1285 } 1286 1287 /* 1288 * Find an interface address specific to an interface best matching 1289 * a given address. 1290 */ 1291 struct ifaddr * 1292 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp) 1293 { 1294 struct ifaddr_container *ifac; 1295 char *cp, *cp2, *cp3; 1296 char *cplim; 1297 struct ifaddr *ifa_maybe = NULL; 1298 u_int af = addr->sa_family; 1299 1300 if (af >= AF_MAX) 1301 return (0); 1302 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) { 1303 struct ifaddr *ifa = ifac->ifa; 1304 1305 if (ifa->ifa_addr->sa_family != af) 1306 continue; 1307 if (ifa_maybe == NULL) 1308 ifa_maybe = ifa; 1309 if (ifa->ifa_netmask == NULL) { 1310 if (sa_equal(addr, ifa->ifa_addr) || 1311 (ifa->ifa_dstaddr != NULL && 1312 sa_equal(addr, ifa->ifa_dstaddr))) 1313 return (ifa); 1314 continue; 1315 } 1316 if (ifp->if_flags & IFF_POINTOPOINT) { 1317 if (sa_equal(addr, ifa->ifa_dstaddr)) 1318 return (ifa); 1319 } else { 1320 cp = addr->sa_data; 1321 cp2 = ifa->ifa_addr->sa_data; 1322 cp3 = ifa->ifa_netmask->sa_data; 1323 cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask; 1324 for (; cp3 < cplim; cp3++) 1325 if ((*cp++ ^ *cp2++) & *cp3) 1326 break; 1327 if (cp3 == cplim) 1328 return (ifa); 1329 } 1330 } 1331 return (ifa_maybe); 1332 } 1333 1334 /* 1335 * Default action when installing a route with a Link Level gateway. 1336 * Lookup an appropriate real ifa to point to. 1337 * This should be moved to /sys/net/link.c eventually. 1338 */ 1339 static void 1340 link_rtrequest(int cmd, struct rtentry *rt) 1341 { 1342 struct ifaddr *ifa; 1343 struct sockaddr *dst; 1344 struct ifnet *ifp; 1345 1346 if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL || 1347 (ifp = ifa->ifa_ifp) == NULL || (dst = rt_key(rt)) == NULL) 1348 return; 1349 ifa = ifaof_ifpforaddr(dst, ifp); 1350 if (ifa != NULL) { 1351 IFAFREE(rt->rt_ifa); 1352 IFAREF(ifa); 1353 rt->rt_ifa = ifa; 1354 if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest) 1355 ifa->ifa_rtrequest(cmd, rt); 1356 } 1357 } 1358 1359 /* 1360 * Mark an interface down and notify protocols of 1361 * the transition. 1362 * NOTE: must be called at splnet or eqivalent. 1363 */ 1364 void 1365 if_unroute(struct ifnet *ifp, int flag, int fam) 1366 { 1367 struct ifaddr_container *ifac; 1368 1369 ifp->if_flags &= ~flag; 1370 getmicrotime(&ifp->if_lastchange); 1371 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) { 1372 struct ifaddr *ifa = ifac->ifa; 1373 1374 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family)) 1375 kpfctlinput(PRC_IFDOWN, ifa->ifa_addr); 1376 } 1377 ifq_purge_all(&ifp->if_snd); 1378 rt_ifmsg(ifp); 1379 } 1380 1381 /* 1382 * Mark an interface up and notify protocols of 1383 * the transition. 1384 * NOTE: must be called at splnet or eqivalent. 1385 */ 1386 void 1387 if_route(struct ifnet *ifp, int flag, int fam) 1388 { 1389 struct ifaddr_container *ifac; 1390 1391 ifq_purge_all(&ifp->if_snd); 1392 ifp->if_flags |= flag; 1393 getmicrotime(&ifp->if_lastchange); 1394 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) { 1395 struct ifaddr *ifa = ifac->ifa; 1396 1397 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family)) 1398 kpfctlinput(PRC_IFUP, ifa->ifa_addr); 1399 } 1400 rt_ifmsg(ifp); 1401 #ifdef INET6 1402 in6_if_up(ifp); 1403 #endif 1404 } 1405 1406 /* 1407 * Mark an interface down and notify protocols of the transition. An 1408 * interface going down is also considered to be a synchronizing event. 1409 * We must ensure that all packet processing related to the interface 1410 * has completed before we return so e.g. the caller can free the ifnet 1411 * structure that the mbufs may be referencing. 1412 * 1413 * NOTE: must be called at splnet or eqivalent. 1414 */ 1415 void 1416 if_down(struct ifnet *ifp) 1417 { 1418 if_unroute(ifp, IFF_UP, AF_UNSPEC); 1419 netmsg_service_sync(); 1420 } 1421 1422 /* 1423 * Mark an interface up and notify protocols of 1424 * the transition. 1425 * NOTE: must be called at splnet or eqivalent. 1426 */ 1427 void 1428 if_up(struct ifnet *ifp) 1429 { 1430 if_route(ifp, IFF_UP, AF_UNSPEC); 1431 } 1432 1433 /* 1434 * Process a link state change. 1435 * NOTE: must be called at splsoftnet or equivalent. 1436 */ 1437 void 1438 if_link_state_change(struct ifnet *ifp) 1439 { 1440 int link_state = ifp->if_link_state; 1441 1442 rt_ifmsg(ifp); 1443 devctl_notify("IFNET", ifp->if_xname, 1444 (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL); 1445 } 1446 1447 /* 1448 * Handle interface watchdog timer routines. Called 1449 * from softclock, we decrement timers (if set) and 1450 * call the appropriate interface routine on expiration. 1451 */ 1452 static void 1453 if_slowtimo(void *arg) 1454 { 1455 struct ifnet *ifp; 1456 1457 crit_enter(); 1458 1459 TAILQ_FOREACH(ifp, &ifnet, if_link) { 1460 if (if_stats_compat) { 1461 IFNET_STAT_GET(ifp, ipackets, ifp->if_ipackets); 1462 IFNET_STAT_GET(ifp, ierrors, ifp->if_ierrors); 1463 IFNET_STAT_GET(ifp, opackets, ifp->if_opackets); 1464 IFNET_STAT_GET(ifp, oerrors, ifp->if_oerrors); 1465 IFNET_STAT_GET(ifp, collisions, ifp->if_collisions); 1466 IFNET_STAT_GET(ifp, ibytes, ifp->if_ibytes); 1467 IFNET_STAT_GET(ifp, obytes, ifp->if_obytes); 1468 IFNET_STAT_GET(ifp, imcasts, ifp->if_imcasts); 1469 IFNET_STAT_GET(ifp, omcasts, ifp->if_omcasts); 1470 IFNET_STAT_GET(ifp, iqdrops, ifp->if_iqdrops); 1471 IFNET_STAT_GET(ifp, noproto, ifp->if_noproto); 1472 } 1473 1474 if (ifp->if_timer == 0 || --ifp->if_timer) 1475 continue; 1476 if (ifp->if_watchdog) { 1477 if (ifnet_tryserialize_all(ifp)) { 1478 (*ifp->if_watchdog)(ifp); 1479 ifnet_deserialize_all(ifp); 1480 } else { 1481 /* try again next timeout */ 1482 ++ifp->if_timer; 1483 } 1484 } 1485 } 1486 1487 crit_exit(); 1488 1489 callout_reset(&if_slowtimo_timer, hz / IFNET_SLOWHZ, if_slowtimo, NULL); 1490 } 1491 1492 /* 1493 * Map interface name to 1494 * interface structure pointer. 1495 */ 1496 struct ifnet * 1497 ifunit(const char *name) 1498 { 1499 struct ifnet *ifp; 1500 1501 /* 1502 * Search all the interfaces for this name/number 1503 */ 1504 1505 TAILQ_FOREACH(ifp, &ifnet, if_link) { 1506 if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0) 1507 break; 1508 } 1509 return (ifp); 1510 } 1511 1512 1513 /* 1514 * Map interface name in a sockaddr_dl to 1515 * interface structure pointer. 1516 */ 1517 struct ifnet * 1518 if_withname(struct sockaddr *sa) 1519 { 1520 char ifname[IFNAMSIZ+1]; 1521 struct sockaddr_dl *sdl = (struct sockaddr_dl *)sa; 1522 1523 if ( (sa->sa_family != AF_LINK) || (sdl->sdl_nlen == 0) || 1524 (sdl->sdl_nlen > IFNAMSIZ) ) 1525 return NULL; 1526 1527 /* 1528 * ifunit wants a null-terminated name. It may not be null-terminated 1529 * in the sockaddr. We don't want to change the caller's sockaddr, 1530 * and there might not be room to put the trailing null anyway, so we 1531 * make a local copy that we know we can null terminate safely. 1532 */ 1533 1534 bcopy(sdl->sdl_data, ifname, sdl->sdl_nlen); 1535 ifname[sdl->sdl_nlen] = '\0'; 1536 return ifunit(ifname); 1537 } 1538 1539 1540 /* 1541 * Interface ioctls. 1542 */ 1543 int 1544 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct ucred *cred) 1545 { 1546 struct ifnet *ifp; 1547 struct ifreq *ifr; 1548 struct ifstat *ifs; 1549 int error; 1550 short oif_flags; 1551 int new_flags; 1552 #ifdef COMPAT_43 1553 int ocmd; 1554 #endif 1555 size_t namelen, onamelen; 1556 char new_name[IFNAMSIZ]; 1557 struct ifaddr *ifa; 1558 struct sockaddr_dl *sdl; 1559 1560 switch (cmd) { 1561 case SIOCGIFCONF: 1562 case OSIOCGIFCONF: 1563 return (ifconf(cmd, data, cred)); 1564 default: 1565 break; 1566 } 1567 1568 ifr = (struct ifreq *)data; 1569 1570 switch (cmd) { 1571 case SIOCIFCREATE: 1572 case SIOCIFCREATE2: 1573 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0) 1574 return (error); 1575 return (if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name), 1576 cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL)); 1577 case SIOCIFDESTROY: 1578 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0) 1579 return (error); 1580 return (if_clone_destroy(ifr->ifr_name)); 1581 case SIOCIFGCLONERS: 1582 return (if_clone_list((struct if_clonereq *)data)); 1583 default: 1584 break; 1585 } 1586 1587 /* 1588 * Nominal ioctl through interface, lookup the ifp and obtain a 1589 * lock to serialize the ifconfig ioctl operation. 1590 */ 1591 ifp = ifunit(ifr->ifr_name); 1592 if (ifp == NULL) 1593 return (ENXIO); 1594 error = 0; 1595 mtx_lock(&ifp->if_ioctl_mtx); 1596 1597 switch (cmd) { 1598 case SIOCGIFINDEX: 1599 ifr->ifr_index = ifp->if_index; 1600 break; 1601 1602 case SIOCGIFFLAGS: 1603 ifr->ifr_flags = ifp->if_flags; 1604 ifr->ifr_flagshigh = ifp->if_flags >> 16; 1605 break; 1606 1607 case SIOCGIFCAP: 1608 ifr->ifr_reqcap = ifp->if_capabilities; 1609 ifr->ifr_curcap = ifp->if_capenable; 1610 break; 1611 1612 case SIOCGIFMETRIC: 1613 ifr->ifr_metric = ifp->if_metric; 1614 break; 1615 1616 case SIOCGIFMTU: 1617 ifr->ifr_mtu = ifp->if_mtu; 1618 break; 1619 1620 case SIOCGIFTSOLEN: 1621 ifr->ifr_tsolen = ifp->if_tsolen; 1622 break; 1623 1624 case SIOCGIFDATA: 1625 error = copyout((caddr_t)&ifp->if_data, ifr->ifr_data, 1626 sizeof(ifp->if_data)); 1627 break; 1628 1629 case SIOCGIFPHYS: 1630 ifr->ifr_phys = ifp->if_physical; 1631 break; 1632 1633 case SIOCGIFPOLLCPU: 1634 ifr->ifr_pollcpu = -1; 1635 break; 1636 1637 case SIOCSIFPOLLCPU: 1638 break; 1639 1640 case SIOCSIFFLAGS: 1641 error = priv_check_cred(cred, PRIV_ROOT, 0); 1642 if (error) 1643 break; 1644 new_flags = (ifr->ifr_flags & 0xffff) | 1645 (ifr->ifr_flagshigh << 16); 1646 if (ifp->if_flags & IFF_SMART) { 1647 /* Smart drivers twiddle their own routes */ 1648 } else if (ifp->if_flags & IFF_UP && 1649 (new_flags & IFF_UP) == 0) { 1650 crit_enter(); 1651 if_down(ifp); 1652 crit_exit(); 1653 } else if (new_flags & IFF_UP && 1654 (ifp->if_flags & IFF_UP) == 0) { 1655 crit_enter(); 1656 if_up(ifp); 1657 crit_exit(); 1658 } 1659 1660 #ifdef IFPOLL_ENABLE 1661 if ((new_flags ^ ifp->if_flags) & IFF_NPOLLING) { 1662 if (new_flags & IFF_NPOLLING) 1663 ifpoll_register(ifp); 1664 else 1665 ifpoll_deregister(ifp); 1666 } 1667 #endif 1668 1669 ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) | 1670 (new_flags &~ IFF_CANTCHANGE); 1671 if (new_flags & IFF_PPROMISC) { 1672 /* Permanently promiscuous mode requested */ 1673 ifp->if_flags |= IFF_PROMISC; 1674 } else if (ifp->if_pcount == 0) { 1675 ifp->if_flags &= ~IFF_PROMISC; 1676 } 1677 if (ifp->if_ioctl) { 1678 ifnet_serialize_all(ifp); 1679 ifp->if_ioctl(ifp, cmd, data, cred); 1680 ifnet_deserialize_all(ifp); 1681 } 1682 getmicrotime(&ifp->if_lastchange); 1683 break; 1684 1685 case SIOCSIFCAP: 1686 error = priv_check_cred(cred, PRIV_ROOT, 0); 1687 if (error) 1688 break; 1689 if (ifr->ifr_reqcap & ~ifp->if_capabilities) { 1690 error = EINVAL; 1691 break; 1692 } 1693 ifnet_serialize_all(ifp); 1694 ifp->if_ioctl(ifp, cmd, data, cred); 1695 ifnet_deserialize_all(ifp); 1696 break; 1697 1698 case SIOCSIFNAME: 1699 error = priv_check_cred(cred, PRIV_ROOT, 0); 1700 if (error) 1701 break; 1702 error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL); 1703 if (error) 1704 break; 1705 if (new_name[0] == '\0') { 1706 error = EINVAL; 1707 break; 1708 } 1709 if (ifunit(new_name) != NULL) { 1710 error = EEXIST; 1711 break; 1712 } 1713 1714 EVENTHANDLER_INVOKE(ifnet_detach_event, ifp); 1715 1716 /* Announce the departure of the interface. */ 1717 rt_ifannouncemsg(ifp, IFAN_DEPARTURE); 1718 1719 strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname)); 1720 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa; 1721 /* XXX IFA_LOCK(ifa); */ 1722 sdl = (struct sockaddr_dl *)ifa->ifa_addr; 1723 namelen = strlen(new_name); 1724 onamelen = sdl->sdl_nlen; 1725 /* 1726 * Move the address if needed. This is safe because we 1727 * allocate space for a name of length IFNAMSIZ when we 1728 * create this in if_attach(). 1729 */ 1730 if (namelen != onamelen) { 1731 bcopy(sdl->sdl_data + onamelen, 1732 sdl->sdl_data + namelen, sdl->sdl_alen); 1733 } 1734 bcopy(new_name, sdl->sdl_data, namelen); 1735 sdl->sdl_nlen = namelen; 1736 sdl = (struct sockaddr_dl *)ifa->ifa_netmask; 1737 bzero(sdl->sdl_data, onamelen); 1738 while (namelen != 0) 1739 sdl->sdl_data[--namelen] = 0xff; 1740 /* XXX IFA_UNLOCK(ifa) */ 1741 1742 EVENTHANDLER_INVOKE(ifnet_attach_event, ifp); 1743 1744 /* Announce the return of the interface. */ 1745 rt_ifannouncemsg(ifp, IFAN_ARRIVAL); 1746 break; 1747 1748 case SIOCSIFMETRIC: 1749 error = priv_check_cred(cred, PRIV_ROOT, 0); 1750 if (error) 1751 break; 1752 ifp->if_metric = ifr->ifr_metric; 1753 getmicrotime(&ifp->if_lastchange); 1754 break; 1755 1756 case SIOCSIFPHYS: 1757 error = priv_check_cred(cred, PRIV_ROOT, 0); 1758 if (error) 1759 break; 1760 if (ifp->if_ioctl == NULL) { 1761 error = EOPNOTSUPP; 1762 break; 1763 } 1764 ifnet_serialize_all(ifp); 1765 error = ifp->if_ioctl(ifp, cmd, data, cred); 1766 ifnet_deserialize_all(ifp); 1767 if (error == 0) 1768 getmicrotime(&ifp->if_lastchange); 1769 break; 1770 1771 case SIOCSIFMTU: 1772 { 1773 u_long oldmtu = ifp->if_mtu; 1774 1775 error = priv_check_cred(cred, PRIV_ROOT, 0); 1776 if (error) 1777 break; 1778 if (ifp->if_ioctl == NULL) { 1779 error = EOPNOTSUPP; 1780 break; 1781 } 1782 if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU) { 1783 error = EINVAL; 1784 break; 1785 } 1786 ifnet_serialize_all(ifp); 1787 error = ifp->if_ioctl(ifp, cmd, data, cred); 1788 ifnet_deserialize_all(ifp); 1789 if (error == 0) { 1790 getmicrotime(&ifp->if_lastchange); 1791 rt_ifmsg(ifp); 1792 } 1793 /* 1794 * If the link MTU changed, do network layer specific procedure. 1795 */ 1796 if (ifp->if_mtu != oldmtu) { 1797 #ifdef INET6 1798 nd6_setmtu(ifp); 1799 #endif 1800 } 1801 break; 1802 } 1803 1804 case SIOCSIFTSOLEN: 1805 error = priv_check_cred(cred, PRIV_ROOT, 0); 1806 if (error) 1807 break; 1808 1809 /* XXX need driver supplied upper limit */ 1810 if (ifr->ifr_tsolen <= 0) { 1811 error = EINVAL; 1812 break; 1813 } 1814 ifp->if_tsolen = ifr->ifr_tsolen; 1815 break; 1816 1817 case SIOCADDMULTI: 1818 case SIOCDELMULTI: 1819 error = priv_check_cred(cred, PRIV_ROOT, 0); 1820 if (error) 1821 break; 1822 1823 /* Don't allow group membership on non-multicast interfaces. */ 1824 if ((ifp->if_flags & IFF_MULTICAST) == 0) { 1825 error = EOPNOTSUPP; 1826 break; 1827 } 1828 1829 /* Don't let users screw up protocols' entries. */ 1830 if (ifr->ifr_addr.sa_family != AF_LINK) { 1831 error = EINVAL; 1832 break; 1833 } 1834 1835 if (cmd == SIOCADDMULTI) { 1836 struct ifmultiaddr *ifma; 1837 error = if_addmulti(ifp, &ifr->ifr_addr, &ifma); 1838 } else { 1839 error = if_delmulti(ifp, &ifr->ifr_addr); 1840 } 1841 if (error == 0) 1842 getmicrotime(&ifp->if_lastchange); 1843 break; 1844 1845 case SIOCSIFPHYADDR: 1846 case SIOCDIFPHYADDR: 1847 #ifdef INET6 1848 case SIOCSIFPHYADDR_IN6: 1849 #endif 1850 case SIOCSLIFPHYADDR: 1851 case SIOCSIFMEDIA: 1852 case SIOCSIFGENERIC: 1853 error = priv_check_cred(cred, PRIV_ROOT, 0); 1854 if (error) 1855 break; 1856 if (ifp->if_ioctl == 0) { 1857 error = EOPNOTSUPP; 1858 break; 1859 } 1860 ifnet_serialize_all(ifp); 1861 error = ifp->if_ioctl(ifp, cmd, data, cred); 1862 ifnet_deserialize_all(ifp); 1863 if (error == 0) 1864 getmicrotime(&ifp->if_lastchange); 1865 break; 1866 1867 case SIOCGIFSTATUS: 1868 ifs = (struct ifstat *)data; 1869 ifs->ascii[0] = '\0'; 1870 /* fall through */ 1871 case SIOCGIFPSRCADDR: 1872 case SIOCGIFPDSTADDR: 1873 case SIOCGLIFPHYADDR: 1874 case SIOCGIFMEDIA: 1875 case SIOCGIFGENERIC: 1876 if (ifp->if_ioctl == NULL) { 1877 error = EOPNOTSUPP; 1878 break; 1879 } 1880 ifnet_serialize_all(ifp); 1881 error = ifp->if_ioctl(ifp, cmd, data, cred); 1882 ifnet_deserialize_all(ifp); 1883 break; 1884 1885 case SIOCSIFLLADDR: 1886 error = priv_check_cred(cred, PRIV_ROOT, 0); 1887 if (error) 1888 break; 1889 error = if_setlladdr(ifp, ifr->ifr_addr.sa_data, 1890 ifr->ifr_addr.sa_len); 1891 EVENTHANDLER_INVOKE(iflladdr_event, ifp); 1892 break; 1893 1894 default: 1895 oif_flags = ifp->if_flags; 1896 if (so->so_proto == 0) { 1897 error = EOPNOTSUPP; 1898 break; 1899 } 1900 #ifndef COMPAT_43 1901 error = so_pru_control_direct(so, cmd, data, ifp); 1902 #else 1903 ocmd = cmd; 1904 1905 switch (cmd) { 1906 case SIOCSIFDSTADDR: 1907 case SIOCSIFADDR: 1908 case SIOCSIFBRDADDR: 1909 case SIOCSIFNETMASK: 1910 #if BYTE_ORDER != BIG_ENDIAN 1911 if (ifr->ifr_addr.sa_family == 0 && 1912 ifr->ifr_addr.sa_len < 16) { 1913 ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len; 1914 ifr->ifr_addr.sa_len = 16; 1915 } 1916 #else 1917 if (ifr->ifr_addr.sa_len == 0) 1918 ifr->ifr_addr.sa_len = 16; 1919 #endif 1920 break; 1921 case OSIOCGIFADDR: 1922 cmd = SIOCGIFADDR; 1923 break; 1924 case OSIOCGIFDSTADDR: 1925 cmd = SIOCGIFDSTADDR; 1926 break; 1927 case OSIOCGIFBRDADDR: 1928 cmd = SIOCGIFBRDADDR; 1929 break; 1930 case OSIOCGIFNETMASK: 1931 cmd = SIOCGIFNETMASK; 1932 break; 1933 default: 1934 break; 1935 } 1936 1937 error = so_pru_control_direct(so, cmd, data, ifp); 1938 1939 switch (ocmd) { 1940 case OSIOCGIFADDR: 1941 case OSIOCGIFDSTADDR: 1942 case OSIOCGIFBRDADDR: 1943 case OSIOCGIFNETMASK: 1944 *(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family; 1945 break; 1946 } 1947 #endif /* COMPAT_43 */ 1948 1949 if ((oif_flags ^ ifp->if_flags) & IFF_UP) { 1950 #ifdef INET6 1951 DELAY(100);/* XXX: temporary workaround for fxp issue*/ 1952 if (ifp->if_flags & IFF_UP) { 1953 crit_enter(); 1954 in6_if_up(ifp); 1955 crit_exit(); 1956 } 1957 #endif 1958 } 1959 break; 1960 } 1961 1962 mtx_unlock(&ifp->if_ioctl_mtx); 1963 return (error); 1964 } 1965 1966 /* 1967 * Set/clear promiscuous mode on interface ifp based on the truth value 1968 * of pswitch. The calls are reference counted so that only the first 1969 * "on" request actually has an effect, as does the final "off" request. 1970 * Results are undefined if the "off" and "on" requests are not matched. 1971 */ 1972 int 1973 ifpromisc(struct ifnet *ifp, int pswitch) 1974 { 1975 struct ifreq ifr; 1976 int error; 1977 int oldflags; 1978 1979 oldflags = ifp->if_flags; 1980 if (ifp->if_flags & IFF_PPROMISC) { 1981 /* Do nothing if device is in permanently promiscuous mode */ 1982 ifp->if_pcount += pswitch ? 1 : -1; 1983 return (0); 1984 } 1985 if (pswitch) { 1986 /* 1987 * If the device is not configured up, we cannot put it in 1988 * promiscuous mode. 1989 */ 1990 if ((ifp->if_flags & IFF_UP) == 0) 1991 return (ENETDOWN); 1992 if (ifp->if_pcount++ != 0) 1993 return (0); 1994 ifp->if_flags |= IFF_PROMISC; 1995 log(LOG_INFO, "%s: promiscuous mode enabled\n", 1996 ifp->if_xname); 1997 } else { 1998 if (--ifp->if_pcount > 0) 1999 return (0); 2000 ifp->if_flags &= ~IFF_PROMISC; 2001 log(LOG_INFO, "%s: promiscuous mode disabled\n", 2002 ifp->if_xname); 2003 } 2004 ifr.ifr_flags = ifp->if_flags; 2005 ifr.ifr_flagshigh = ifp->if_flags >> 16; 2006 ifnet_serialize_all(ifp); 2007 error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr, NULL); 2008 ifnet_deserialize_all(ifp); 2009 if (error == 0) 2010 rt_ifmsg(ifp); 2011 else 2012 ifp->if_flags = oldflags; 2013 return error; 2014 } 2015 2016 /* 2017 * Return interface configuration 2018 * of system. List may be used 2019 * in later ioctl's (above) to get 2020 * other information. 2021 */ 2022 static int 2023 ifconf(u_long cmd, caddr_t data, struct ucred *cred) 2024 { 2025 struct ifconf *ifc = (struct ifconf *)data; 2026 struct ifnet *ifp; 2027 struct sockaddr *sa; 2028 struct ifreq ifr, *ifrp; 2029 int space = ifc->ifc_len, error = 0; 2030 2031 ifrp = ifc->ifc_req; 2032 TAILQ_FOREACH(ifp, &ifnet, if_link) { 2033 struct ifaddr_container *ifac; 2034 int addrs; 2035 2036 if (space <= sizeof ifr) 2037 break; 2038 2039 /* 2040 * Zero the stack declared structure first to prevent 2041 * memory disclosure. 2042 */ 2043 bzero(&ifr, sizeof(ifr)); 2044 if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name)) 2045 >= sizeof(ifr.ifr_name)) { 2046 error = ENAMETOOLONG; 2047 break; 2048 } 2049 2050 addrs = 0; 2051 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) { 2052 struct ifaddr *ifa = ifac->ifa; 2053 2054 if (space <= sizeof ifr) 2055 break; 2056 sa = ifa->ifa_addr; 2057 if (cred->cr_prison && 2058 prison_if(cred, sa)) 2059 continue; 2060 addrs++; 2061 #ifdef COMPAT_43 2062 if (cmd == OSIOCGIFCONF) { 2063 struct osockaddr *osa = 2064 (struct osockaddr *)&ifr.ifr_addr; 2065 ifr.ifr_addr = *sa; 2066 osa->sa_family = sa->sa_family; 2067 error = copyout(&ifr, ifrp, sizeof ifr); 2068 ifrp++; 2069 } else 2070 #endif 2071 if (sa->sa_len <= sizeof(*sa)) { 2072 ifr.ifr_addr = *sa; 2073 error = copyout(&ifr, ifrp, sizeof ifr); 2074 ifrp++; 2075 } else { 2076 if (space < (sizeof ifr) + sa->sa_len - 2077 sizeof(*sa)) 2078 break; 2079 space -= sa->sa_len - sizeof(*sa); 2080 error = copyout(&ifr, ifrp, 2081 sizeof ifr.ifr_name); 2082 if (error == 0) 2083 error = copyout(sa, &ifrp->ifr_addr, 2084 sa->sa_len); 2085 ifrp = (struct ifreq *) 2086 (sa->sa_len + (caddr_t)&ifrp->ifr_addr); 2087 } 2088 if (error) 2089 break; 2090 space -= sizeof ifr; 2091 } 2092 if (error) 2093 break; 2094 if (!addrs) { 2095 bzero(&ifr.ifr_addr, sizeof ifr.ifr_addr); 2096 error = copyout(&ifr, ifrp, sizeof ifr); 2097 if (error) 2098 break; 2099 space -= sizeof ifr; 2100 ifrp++; 2101 } 2102 } 2103 ifc->ifc_len -= space; 2104 return (error); 2105 } 2106 2107 /* 2108 * Just like if_promisc(), but for all-multicast-reception mode. 2109 */ 2110 int 2111 if_allmulti(struct ifnet *ifp, int onswitch) 2112 { 2113 int error = 0; 2114 struct ifreq ifr; 2115 2116 crit_enter(); 2117 2118 if (onswitch) { 2119 if (ifp->if_amcount++ == 0) { 2120 ifp->if_flags |= IFF_ALLMULTI; 2121 ifr.ifr_flags = ifp->if_flags; 2122 ifr.ifr_flagshigh = ifp->if_flags >> 16; 2123 ifnet_serialize_all(ifp); 2124 error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr, 2125 NULL); 2126 ifnet_deserialize_all(ifp); 2127 } 2128 } else { 2129 if (ifp->if_amcount > 1) { 2130 ifp->if_amcount--; 2131 } else { 2132 ifp->if_amcount = 0; 2133 ifp->if_flags &= ~IFF_ALLMULTI; 2134 ifr.ifr_flags = ifp->if_flags; 2135 ifr.ifr_flagshigh = ifp->if_flags >> 16; 2136 ifnet_serialize_all(ifp); 2137 error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr, 2138 NULL); 2139 ifnet_deserialize_all(ifp); 2140 } 2141 } 2142 2143 crit_exit(); 2144 2145 if (error == 0) 2146 rt_ifmsg(ifp); 2147 return error; 2148 } 2149 2150 /* 2151 * Add a multicast listenership to the interface in question. 2152 * The link layer provides a routine which converts 2153 */ 2154 int 2155 if_addmulti( 2156 struct ifnet *ifp, /* interface to manipulate */ 2157 struct sockaddr *sa, /* address to add */ 2158 struct ifmultiaddr **retifma) 2159 { 2160 struct sockaddr *llsa, *dupsa; 2161 int error; 2162 struct ifmultiaddr *ifma; 2163 2164 /* 2165 * If the matching multicast address already exists 2166 * then don't add a new one, just add a reference 2167 */ 2168 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 2169 if (sa_equal(sa, ifma->ifma_addr)) { 2170 ifma->ifma_refcount++; 2171 if (retifma) 2172 *retifma = ifma; 2173 return 0; 2174 } 2175 } 2176 2177 /* 2178 * Give the link layer a chance to accept/reject it, and also 2179 * find out which AF_LINK address this maps to, if it isn't one 2180 * already. 2181 */ 2182 if (ifp->if_resolvemulti) { 2183 ifnet_serialize_all(ifp); 2184 error = ifp->if_resolvemulti(ifp, &llsa, sa); 2185 ifnet_deserialize_all(ifp); 2186 if (error) 2187 return error; 2188 } else { 2189 llsa = NULL; 2190 } 2191 2192 ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_WAITOK); 2193 dupsa = kmalloc(sa->sa_len, M_IFMADDR, M_WAITOK); 2194 bcopy(sa, dupsa, sa->sa_len); 2195 2196 ifma->ifma_addr = dupsa; 2197 ifma->ifma_lladdr = llsa; 2198 ifma->ifma_ifp = ifp; 2199 ifma->ifma_refcount = 1; 2200 ifma->ifma_protospec = 0; 2201 rt_newmaddrmsg(RTM_NEWMADDR, ifma); 2202 2203 /* 2204 * Some network interfaces can scan the address list at 2205 * interrupt time; lock them out. 2206 */ 2207 crit_enter(); 2208 TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link); 2209 crit_exit(); 2210 if (retifma) 2211 *retifma = ifma; 2212 2213 if (llsa != NULL) { 2214 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 2215 if (sa_equal(ifma->ifma_addr, llsa)) 2216 break; 2217 } 2218 if (ifma) { 2219 ifma->ifma_refcount++; 2220 } else { 2221 ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_WAITOK); 2222 dupsa = kmalloc(llsa->sa_len, M_IFMADDR, M_WAITOK); 2223 bcopy(llsa, dupsa, llsa->sa_len); 2224 ifma->ifma_addr = dupsa; 2225 ifma->ifma_ifp = ifp; 2226 ifma->ifma_refcount = 1; 2227 crit_enter(); 2228 TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link); 2229 crit_exit(); 2230 } 2231 } 2232 /* 2233 * We are certain we have added something, so call down to the 2234 * interface to let them know about it. 2235 */ 2236 crit_enter(); 2237 ifnet_serialize_all(ifp); 2238 if (ifp->if_ioctl) 2239 ifp->if_ioctl(ifp, SIOCADDMULTI, 0, NULL); 2240 ifnet_deserialize_all(ifp); 2241 crit_exit(); 2242 2243 return 0; 2244 } 2245 2246 /* 2247 * Remove a reference to a multicast address on this interface. Yell 2248 * if the request does not match an existing membership. 2249 */ 2250 int 2251 if_delmulti(struct ifnet *ifp, struct sockaddr *sa) 2252 { 2253 struct ifmultiaddr *ifma; 2254 2255 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) 2256 if (sa_equal(sa, ifma->ifma_addr)) 2257 break; 2258 if (ifma == NULL) 2259 return ENOENT; 2260 2261 if (ifma->ifma_refcount > 1) { 2262 ifma->ifma_refcount--; 2263 return 0; 2264 } 2265 2266 rt_newmaddrmsg(RTM_DELMADDR, ifma); 2267 sa = ifma->ifma_lladdr; 2268 crit_enter(); 2269 TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link); 2270 /* 2271 * Make sure the interface driver is notified 2272 * in the case of a link layer mcast group being left. 2273 */ 2274 if (ifma->ifma_addr->sa_family == AF_LINK && sa == NULL) { 2275 ifnet_serialize_all(ifp); 2276 ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL); 2277 ifnet_deserialize_all(ifp); 2278 } 2279 crit_exit(); 2280 kfree(ifma->ifma_addr, M_IFMADDR); 2281 kfree(ifma, M_IFMADDR); 2282 if (sa == NULL) 2283 return 0; 2284 2285 /* 2286 * Now look for the link-layer address which corresponds to 2287 * this network address. It had been squirreled away in 2288 * ifma->ifma_lladdr for this purpose (so we don't have 2289 * to call ifp->if_resolvemulti() again), and we saved that 2290 * value in sa above. If some nasty deleted the 2291 * link-layer address out from underneath us, we can deal because 2292 * the address we stored was is not the same as the one which was 2293 * in the record for the link-layer address. (So we don't complain 2294 * in that case.) 2295 */ 2296 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) 2297 if (sa_equal(sa, ifma->ifma_addr)) 2298 break; 2299 if (ifma == NULL) 2300 return 0; 2301 2302 if (ifma->ifma_refcount > 1) { 2303 ifma->ifma_refcount--; 2304 return 0; 2305 } 2306 2307 crit_enter(); 2308 ifnet_serialize_all(ifp); 2309 TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link); 2310 ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL); 2311 ifnet_deserialize_all(ifp); 2312 crit_exit(); 2313 kfree(ifma->ifma_addr, M_IFMADDR); 2314 kfree(sa, M_IFMADDR); 2315 kfree(ifma, M_IFMADDR); 2316 2317 return 0; 2318 } 2319 2320 /* 2321 * Delete all multicast group membership for an interface. 2322 * Should be used to quickly flush all multicast filters. 2323 */ 2324 void 2325 if_delallmulti(struct ifnet *ifp) 2326 { 2327 struct ifmultiaddr *ifma; 2328 struct ifmultiaddr *next; 2329 2330 TAILQ_FOREACH_MUTABLE(ifma, &ifp->if_multiaddrs, ifma_link, next) 2331 if_delmulti(ifp, ifma->ifma_addr); 2332 } 2333 2334 2335 /* 2336 * Set the link layer address on an interface. 2337 * 2338 * At this time we only support certain types of interfaces, 2339 * and we don't allow the length of the address to change. 2340 */ 2341 int 2342 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len) 2343 { 2344 struct sockaddr_dl *sdl; 2345 struct ifreq ifr; 2346 2347 sdl = IF_LLSOCKADDR(ifp); 2348 if (sdl == NULL) 2349 return (EINVAL); 2350 if (len != sdl->sdl_alen) /* don't allow length to change */ 2351 return (EINVAL); 2352 switch (ifp->if_type) { 2353 case IFT_ETHER: /* these types use struct arpcom */ 2354 case IFT_XETHER: 2355 case IFT_L2VLAN: 2356 bcopy(lladdr, ((struct arpcom *)ifp->if_softc)->ac_enaddr, len); 2357 bcopy(lladdr, LLADDR(sdl), len); 2358 break; 2359 default: 2360 return (ENODEV); 2361 } 2362 /* 2363 * If the interface is already up, we need 2364 * to re-init it in order to reprogram its 2365 * address filter. 2366 */ 2367 ifnet_serialize_all(ifp); 2368 if ((ifp->if_flags & IFF_UP) != 0) { 2369 #ifdef INET 2370 struct ifaddr_container *ifac; 2371 #endif 2372 2373 ifp->if_flags &= ~IFF_UP; 2374 ifr.ifr_flags = ifp->if_flags; 2375 ifr.ifr_flagshigh = ifp->if_flags >> 16; 2376 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr, 2377 NULL); 2378 ifp->if_flags |= IFF_UP; 2379 ifr.ifr_flags = ifp->if_flags; 2380 ifr.ifr_flagshigh = ifp->if_flags >> 16; 2381 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr, 2382 NULL); 2383 #ifdef INET 2384 /* 2385 * Also send gratuitous ARPs to notify other nodes about 2386 * the address change. 2387 */ 2388 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) { 2389 struct ifaddr *ifa = ifac->ifa; 2390 2391 if (ifa->ifa_addr != NULL && 2392 ifa->ifa_addr->sa_family == AF_INET) 2393 arp_gratuitous(ifp, ifa); 2394 } 2395 #endif 2396 } 2397 ifnet_deserialize_all(ifp); 2398 return (0); 2399 } 2400 2401 struct ifmultiaddr * 2402 ifmaof_ifpforaddr(struct sockaddr *sa, struct ifnet *ifp) 2403 { 2404 struct ifmultiaddr *ifma; 2405 2406 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) 2407 if (sa_equal(ifma->ifma_addr, sa)) 2408 break; 2409 2410 return ifma; 2411 } 2412 2413 /* 2414 * This function locates the first real ethernet MAC from a network 2415 * card and loads it into node, returning 0 on success or ENOENT if 2416 * no suitable interfaces were found. It is used by the uuid code to 2417 * generate a unique 6-byte number. 2418 */ 2419 int 2420 if_getanyethermac(uint16_t *node, int minlen) 2421 { 2422 struct ifnet *ifp; 2423 struct sockaddr_dl *sdl; 2424 2425 TAILQ_FOREACH(ifp, &ifnet, if_link) { 2426 if (ifp->if_type != IFT_ETHER) 2427 continue; 2428 sdl = IF_LLSOCKADDR(ifp); 2429 if (sdl->sdl_alen < minlen) 2430 continue; 2431 bcopy(((struct arpcom *)ifp->if_softc)->ac_enaddr, node, 2432 minlen); 2433 return(0); 2434 } 2435 return (ENOENT); 2436 } 2437 2438 /* 2439 * The name argument must be a pointer to storage which will last as 2440 * long as the interface does. For physical devices, the result of 2441 * device_get_name(dev) is a good choice and for pseudo-devices a 2442 * static string works well. 2443 */ 2444 void 2445 if_initname(struct ifnet *ifp, const char *name, int unit) 2446 { 2447 ifp->if_dname = name; 2448 ifp->if_dunit = unit; 2449 if (unit != IF_DUNIT_NONE) 2450 ksnprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit); 2451 else 2452 strlcpy(ifp->if_xname, name, IFNAMSIZ); 2453 } 2454 2455 int 2456 if_printf(struct ifnet *ifp, const char *fmt, ...) 2457 { 2458 __va_list ap; 2459 int retval; 2460 2461 retval = kprintf("%s: ", ifp->if_xname); 2462 __va_start(ap, fmt); 2463 retval += kvprintf(fmt, ap); 2464 __va_end(ap); 2465 return (retval); 2466 } 2467 2468 struct ifnet * 2469 if_alloc(uint8_t type) 2470 { 2471 struct ifnet *ifp; 2472 size_t size; 2473 2474 /* 2475 * XXX temporary hack until arpcom is setup in if_l2com 2476 */ 2477 if (type == IFT_ETHER) 2478 size = sizeof(struct arpcom); 2479 else 2480 size = sizeof(struct ifnet); 2481 2482 ifp = kmalloc(size, M_IFNET, M_WAITOK|M_ZERO); 2483 2484 ifp->if_type = type; 2485 2486 if (if_com_alloc[type] != NULL) { 2487 ifp->if_l2com = if_com_alloc[type](type, ifp); 2488 if (ifp->if_l2com == NULL) { 2489 kfree(ifp, M_IFNET); 2490 return (NULL); 2491 } 2492 } 2493 return (ifp); 2494 } 2495 2496 void 2497 if_free(struct ifnet *ifp) 2498 { 2499 kfree(ifp, M_IFNET); 2500 } 2501 2502 void 2503 ifq_set_classic(struct ifaltq *ifq) 2504 { 2505 ifq_set_methods(ifq, ifq->altq_ifp->if_mapsubq, 2506 ifsq_classic_enqueue, ifsq_classic_dequeue, ifsq_classic_request); 2507 } 2508 2509 void 2510 ifq_set_methods(struct ifaltq *ifq, altq_mapsubq_t mapsubq, 2511 ifsq_enqueue_t enqueue, ifsq_dequeue_t dequeue, ifsq_request_t request) 2512 { 2513 int q; 2514 2515 KASSERT(mapsubq != NULL, ("mapsubq is not specified")); 2516 KASSERT(enqueue != NULL, ("enqueue is not specified")); 2517 KASSERT(dequeue != NULL, ("dequeue is not specified")); 2518 KASSERT(request != NULL, ("request is not specified")); 2519 2520 ifq->altq_mapsubq = mapsubq; 2521 for (q = 0; q < ifq->altq_subq_cnt; ++q) { 2522 struct ifaltq_subque *ifsq = &ifq->altq_subq[q]; 2523 2524 ifsq->ifsq_enqueue = enqueue; 2525 ifsq->ifsq_dequeue = dequeue; 2526 ifsq->ifsq_request = request; 2527 } 2528 } 2529 2530 static void 2531 ifsq_norm_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m) 2532 { 2533 m->m_nextpkt = NULL; 2534 if (ifsq->ifsq_norm_tail == NULL) 2535 ifsq->ifsq_norm_head = m; 2536 else 2537 ifsq->ifsq_norm_tail->m_nextpkt = m; 2538 ifsq->ifsq_norm_tail = m; 2539 ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len); 2540 } 2541 2542 static void 2543 ifsq_prio_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m) 2544 { 2545 m->m_nextpkt = NULL; 2546 if (ifsq->ifsq_prio_tail == NULL) 2547 ifsq->ifsq_prio_head = m; 2548 else 2549 ifsq->ifsq_prio_tail->m_nextpkt = m; 2550 ifsq->ifsq_prio_tail = m; 2551 ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len); 2552 ALTQ_SQ_PRIO_CNTR_INC(ifsq, m->m_pkthdr.len); 2553 } 2554 2555 static struct mbuf * 2556 ifsq_norm_dequeue(struct ifaltq_subque *ifsq) 2557 { 2558 struct mbuf *m; 2559 2560 m = ifsq->ifsq_norm_head; 2561 if (m != NULL) { 2562 if ((ifsq->ifsq_norm_head = m->m_nextpkt) == NULL) 2563 ifsq->ifsq_norm_tail = NULL; 2564 m->m_nextpkt = NULL; 2565 ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len); 2566 } 2567 return m; 2568 } 2569 2570 static struct mbuf * 2571 ifsq_prio_dequeue(struct ifaltq_subque *ifsq) 2572 { 2573 struct mbuf *m; 2574 2575 m = ifsq->ifsq_prio_head; 2576 if (m != NULL) { 2577 if ((ifsq->ifsq_prio_head = m->m_nextpkt) == NULL) 2578 ifsq->ifsq_prio_tail = NULL; 2579 m->m_nextpkt = NULL; 2580 ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len); 2581 ALTQ_SQ_PRIO_CNTR_DEC(ifsq, m->m_pkthdr.len); 2582 } 2583 return m; 2584 } 2585 2586 int 2587 ifsq_classic_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m, 2588 struct altq_pktattr *pa __unused) 2589 { 2590 M_ASSERTPKTHDR(m); 2591 if (ifsq->ifsq_len >= ifsq->ifsq_maxlen || 2592 ifsq->ifsq_bcnt >= ifsq->ifsq_maxbcnt) { 2593 if ((m->m_flags & M_PRIO) && 2594 ifsq->ifsq_prio_len < (ifsq->ifsq_maxlen / 2) && 2595 ifsq->ifsq_prio_bcnt < (ifsq->ifsq_maxbcnt / 2)) { 2596 struct mbuf *m_drop; 2597 2598 /* 2599 * Perform drop-head on normal queue 2600 */ 2601 m_drop = ifsq_norm_dequeue(ifsq); 2602 if (m_drop != NULL) { 2603 m_freem(m_drop); 2604 ifsq_prio_enqueue(ifsq, m); 2605 return 0; 2606 } 2607 /* XXX nothing could be dropped? */ 2608 } 2609 m_freem(m); 2610 return ENOBUFS; 2611 } else { 2612 if (m->m_flags & M_PRIO) 2613 ifsq_prio_enqueue(ifsq, m); 2614 else 2615 ifsq_norm_enqueue(ifsq, m); 2616 return 0; 2617 } 2618 } 2619 2620 struct mbuf * 2621 ifsq_classic_dequeue(struct ifaltq_subque *ifsq, int op) 2622 { 2623 struct mbuf *m; 2624 2625 switch (op) { 2626 case ALTDQ_POLL: 2627 m = ifsq->ifsq_prio_head; 2628 if (m == NULL) 2629 m = ifsq->ifsq_norm_head; 2630 break; 2631 2632 case ALTDQ_REMOVE: 2633 m = ifsq_prio_dequeue(ifsq); 2634 if (m == NULL) 2635 m = ifsq_norm_dequeue(ifsq); 2636 break; 2637 2638 default: 2639 panic("unsupported ALTQ dequeue op: %d", op); 2640 } 2641 return m; 2642 } 2643 2644 int 2645 ifsq_classic_request(struct ifaltq_subque *ifsq, int req, void *arg) 2646 { 2647 switch (req) { 2648 case ALTRQ_PURGE: 2649 for (;;) { 2650 struct mbuf *m; 2651 2652 m = ifsq_classic_dequeue(ifsq, ALTDQ_REMOVE); 2653 if (m == NULL) 2654 break; 2655 m_freem(m); 2656 } 2657 break; 2658 2659 default: 2660 panic("unsupported ALTQ request: %d", req); 2661 } 2662 return 0; 2663 } 2664 2665 static void 2666 ifsq_ifstart_try(struct ifaltq_subque *ifsq, int force_sched) 2667 { 2668 struct ifnet *ifp = ifsq_get_ifp(ifsq); 2669 int running = 0, need_sched; 2670 2671 /* 2672 * Try to do direct ifnet.if_start on the subqueue first, if there is 2673 * contention on the subqueue hardware serializer, ifnet.if_start on 2674 * the subqueue will be scheduled on the subqueue owner CPU. 2675 */ 2676 if (!ifsq_tryserialize_hw(ifsq)) { 2677 /* 2678 * Subqueue hardware serializer contention happened, 2679 * ifnet.if_start on the subqueue is scheduled on 2680 * the subqueue owner CPU, and we keep going. 2681 */ 2682 ifsq_ifstart_schedule(ifsq, 1); 2683 return; 2684 } 2685 2686 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) { 2687 ifp->if_start(ifp, ifsq); 2688 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) 2689 running = 1; 2690 } 2691 need_sched = ifsq_ifstart_need_schedule(ifsq, running); 2692 2693 ifsq_deserialize_hw(ifsq); 2694 2695 if (need_sched) { 2696 /* 2697 * More data need to be transmitted, ifnet.if_start on the 2698 * subqueue is scheduled on the subqueue owner CPU, and we 2699 * keep going. 2700 * NOTE: ifnet.if_start subqueue interlock is not released. 2701 */ 2702 ifsq_ifstart_schedule(ifsq, force_sched); 2703 } 2704 } 2705 2706 /* 2707 * Subqeue packets staging mechanism: 2708 * 2709 * The packets enqueued into the subqueue are staged to a certain amount 2710 * before the ifnet.if_start on the subqueue is called. In this way, the 2711 * driver could avoid writing to hardware registers upon every packet, 2712 * instead, hardware registers could be written when certain amount of 2713 * packets are put onto hardware TX ring. The measurement on several modern 2714 * NICs (emx(4), igb(4), bnx(4), bge(4), jme(4)) shows that the hardware 2715 * registers writing aggregation could save ~20% CPU time when 18bytes UDP 2716 * datagrams are transmitted at 1.48Mpps. The performance improvement by 2717 * hardware registers writing aggeregation is also mentioned by Luigi Rizzo's 2718 * netmap paper (http://info.iet.unipi.it/~luigi/netmap/). 2719 * 2720 * Subqueue packets staging is performed for two entry points into drivers' 2721 * transmission function: 2722 * - Direct ifnet.if_start calling on the subqueue, i.e. ifsq_ifstart_try() 2723 * - ifnet.if_start scheduling on the subqueue, i.e. ifsq_ifstart_schedule() 2724 * 2725 * Subqueue packets staging will be stopped upon any of the following 2726 * conditions: 2727 * - If the count of packets enqueued on the current CPU is great than or 2728 * equal to ifsq_stage_cntmax. (XXX this should be per-interface) 2729 * - If the total length of packets enqueued on the current CPU is great 2730 * than or equal to the hardware's MTU - max_protohdr. max_protohdr is 2731 * cut from the hardware's MTU mainly bacause a full TCP segment's size 2732 * is usually less than hardware's MTU. 2733 * - ifsq_ifstart_schedule() is not pending on the current CPU and 2734 * ifnet.if_start subqueue interlock (ifaltq_subq.ifsq_started) is not 2735 * released. 2736 * - The if_start_rollup(), which is registered as low priority netisr 2737 * rollup function, is called; probably because no more work is pending 2738 * for netisr. 2739 * 2740 * NOTE: 2741 * Currently subqueue packet staging is only performed in netisr threads. 2742 */ 2743 int 2744 ifq_dispatch(struct ifnet *ifp, struct mbuf *m, struct altq_pktattr *pa) 2745 { 2746 struct ifaltq *ifq = &ifp->if_snd; 2747 struct ifaltq_subque *ifsq; 2748 int error, start = 0, len, mcast = 0, avoid_start = 0; 2749 struct ifsubq_stage_head *head = NULL; 2750 struct ifsubq_stage *stage = NULL; 2751 struct globaldata *gd = mycpu; 2752 struct thread *td = gd->gd_curthread; 2753 2754 crit_enter_quick(td); 2755 2756 ifsq = ifq_map_subq(ifq, gd->gd_cpuid); 2757 ASSERT_ALTQ_SQ_NOT_SERIALIZED_HW(ifsq); 2758 2759 len = m->m_pkthdr.len; 2760 if (m->m_flags & M_MCAST) 2761 mcast = 1; 2762 2763 if (td->td_type == TD_TYPE_NETISR) { 2764 head = &ifsubq_stage_heads[mycpuid]; 2765 stage = ifsq_get_stage(ifsq, mycpuid); 2766 2767 stage->stg_cnt++; 2768 stage->stg_len += len; 2769 if (stage->stg_cnt < ifsq_stage_cntmax && 2770 stage->stg_len < (ifp->if_mtu - max_protohdr)) 2771 avoid_start = 1; 2772 } 2773 2774 ALTQ_SQ_LOCK(ifsq); 2775 error = ifsq_enqueue_locked(ifsq, m, pa); 2776 if (error) { 2777 if (!ifsq_data_ready(ifsq)) { 2778 ALTQ_SQ_UNLOCK(ifsq); 2779 crit_exit_quick(td); 2780 return error; 2781 } 2782 avoid_start = 0; 2783 } 2784 if (!ifsq_is_started(ifsq)) { 2785 if (avoid_start) { 2786 ALTQ_SQ_UNLOCK(ifsq); 2787 2788 KKASSERT(!error); 2789 if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0) 2790 ifsq_stage_insert(head, stage); 2791 2792 IFNET_STAT_INC(ifp, obytes, len); 2793 if (mcast) 2794 IFNET_STAT_INC(ifp, omcasts, 1); 2795 crit_exit_quick(td); 2796 return error; 2797 } 2798 2799 /* 2800 * Hold the subqueue interlock of ifnet.if_start 2801 */ 2802 ifsq_set_started(ifsq); 2803 start = 1; 2804 } 2805 ALTQ_SQ_UNLOCK(ifsq); 2806 2807 if (!error) { 2808 IFNET_STAT_INC(ifp, obytes, len); 2809 if (mcast) 2810 IFNET_STAT_INC(ifp, omcasts, 1); 2811 } 2812 2813 if (stage != NULL) { 2814 if (!start && (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)) { 2815 KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED); 2816 if (!avoid_start) { 2817 ifsq_stage_remove(head, stage); 2818 ifsq_ifstart_schedule(ifsq, 1); 2819 } 2820 crit_exit_quick(td); 2821 return error; 2822 } 2823 2824 if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED) { 2825 ifsq_stage_remove(head, stage); 2826 } else { 2827 stage->stg_cnt = 0; 2828 stage->stg_len = 0; 2829 } 2830 } 2831 2832 if (!start) { 2833 crit_exit_quick(td); 2834 return error; 2835 } 2836 2837 ifsq_ifstart_try(ifsq, 0); 2838 2839 crit_exit_quick(td); 2840 return error; 2841 } 2842 2843 void * 2844 ifa_create(int size, int flags) 2845 { 2846 struct ifaddr *ifa; 2847 int i; 2848 2849 KASSERT(size >= sizeof(*ifa), ("ifaddr size too small")); 2850 2851 ifa = kmalloc(size, M_IFADDR, flags | M_ZERO); 2852 if (ifa == NULL) 2853 return NULL; 2854 2855 ifa->ifa_containers = 2856 kmalloc_cachealign(ncpus * sizeof(struct ifaddr_container), 2857 M_IFADDR, M_WAITOK | M_ZERO); 2858 ifa->ifa_ncnt = ncpus; 2859 for (i = 0; i < ncpus; ++i) { 2860 struct ifaddr_container *ifac = &ifa->ifa_containers[i]; 2861 2862 ifac->ifa_magic = IFA_CONTAINER_MAGIC; 2863 ifac->ifa = ifa; 2864 ifac->ifa_refcnt = 1; 2865 } 2866 #ifdef IFADDR_DEBUG 2867 kprintf("alloc ifa %p %d\n", ifa, size); 2868 #endif 2869 return ifa; 2870 } 2871 2872 void 2873 ifac_free(struct ifaddr_container *ifac, int cpu_id) 2874 { 2875 struct ifaddr *ifa = ifac->ifa; 2876 2877 KKASSERT(ifac->ifa_magic == IFA_CONTAINER_MAGIC); 2878 KKASSERT(ifac->ifa_refcnt == 0); 2879 KASSERT(ifac->ifa_listmask == 0, 2880 ("ifa is still on %#x lists", ifac->ifa_listmask)); 2881 2882 ifac->ifa_magic = IFA_CONTAINER_DEAD; 2883 2884 #ifdef IFADDR_DEBUG_VERBOSE 2885 kprintf("try free ifa %p cpu_id %d\n", ifac->ifa, cpu_id); 2886 #endif 2887 2888 KASSERT(ifa->ifa_ncnt > 0 && ifa->ifa_ncnt <= ncpus, 2889 ("invalid # of ifac, %d", ifa->ifa_ncnt)); 2890 if (atomic_fetchadd_int(&ifa->ifa_ncnt, -1) == 1) { 2891 #ifdef IFADDR_DEBUG 2892 kprintf("free ifa %p\n", ifa); 2893 #endif 2894 kfree(ifa->ifa_containers, M_IFADDR); 2895 kfree(ifa, M_IFADDR); 2896 } 2897 } 2898 2899 static void 2900 ifa_iflink_dispatch(netmsg_t nmsg) 2901 { 2902 struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg; 2903 struct ifaddr *ifa = msg->ifa; 2904 struct ifnet *ifp = msg->ifp; 2905 int cpu = mycpuid; 2906 struct ifaddr_container *ifac; 2907 2908 crit_enter(); 2909 2910 ifac = &ifa->ifa_containers[cpu]; 2911 ASSERT_IFAC_VALID(ifac); 2912 KASSERT((ifac->ifa_listmask & IFA_LIST_IFADDRHEAD) == 0, 2913 ("ifaddr is on if_addrheads")); 2914 2915 ifac->ifa_listmask |= IFA_LIST_IFADDRHEAD; 2916 if (msg->tail) 2917 TAILQ_INSERT_TAIL(&ifp->if_addrheads[cpu], ifac, ifa_link); 2918 else 2919 TAILQ_INSERT_HEAD(&ifp->if_addrheads[cpu], ifac, ifa_link); 2920 2921 crit_exit(); 2922 2923 ifa_forwardmsg(&nmsg->lmsg, cpu + 1); 2924 } 2925 2926 void 2927 ifa_iflink(struct ifaddr *ifa, struct ifnet *ifp, int tail) 2928 { 2929 struct netmsg_ifaddr msg; 2930 2931 netmsg_init(&msg.base, NULL, &curthread->td_msgport, 2932 0, ifa_iflink_dispatch); 2933 msg.ifa = ifa; 2934 msg.ifp = ifp; 2935 msg.tail = tail; 2936 2937 ifa_domsg(&msg.base.lmsg, 0); 2938 } 2939 2940 static void 2941 ifa_ifunlink_dispatch(netmsg_t nmsg) 2942 { 2943 struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg; 2944 struct ifaddr *ifa = msg->ifa; 2945 struct ifnet *ifp = msg->ifp; 2946 int cpu = mycpuid; 2947 struct ifaddr_container *ifac; 2948 2949 crit_enter(); 2950 2951 ifac = &ifa->ifa_containers[cpu]; 2952 ASSERT_IFAC_VALID(ifac); 2953 KASSERT(ifac->ifa_listmask & IFA_LIST_IFADDRHEAD, 2954 ("ifaddr is not on if_addrhead")); 2955 2956 TAILQ_REMOVE(&ifp->if_addrheads[cpu], ifac, ifa_link); 2957 ifac->ifa_listmask &= ~IFA_LIST_IFADDRHEAD; 2958 2959 crit_exit(); 2960 2961 ifa_forwardmsg(&nmsg->lmsg, cpu + 1); 2962 } 2963 2964 void 2965 ifa_ifunlink(struct ifaddr *ifa, struct ifnet *ifp) 2966 { 2967 struct netmsg_ifaddr msg; 2968 2969 netmsg_init(&msg.base, NULL, &curthread->td_msgport, 2970 0, ifa_ifunlink_dispatch); 2971 msg.ifa = ifa; 2972 msg.ifp = ifp; 2973 2974 ifa_domsg(&msg.base.lmsg, 0); 2975 } 2976 2977 static void 2978 ifa_destroy_dispatch(netmsg_t nmsg) 2979 { 2980 struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg; 2981 2982 IFAFREE(msg->ifa); 2983 ifa_forwardmsg(&nmsg->lmsg, mycpuid + 1); 2984 } 2985 2986 void 2987 ifa_destroy(struct ifaddr *ifa) 2988 { 2989 struct netmsg_ifaddr msg; 2990 2991 netmsg_init(&msg.base, NULL, &curthread->td_msgport, 2992 0, ifa_destroy_dispatch); 2993 msg.ifa = ifa; 2994 2995 ifa_domsg(&msg.base.lmsg, 0); 2996 } 2997 2998 struct lwkt_port * 2999 ifnet_portfn(int cpu) 3000 { 3001 return &ifnet_threads[cpu].td_msgport; 3002 } 3003 3004 void 3005 ifnet_forwardmsg(struct lwkt_msg *lmsg, int next_cpu) 3006 { 3007 KKASSERT(next_cpu > mycpuid && next_cpu <= ncpus); 3008 3009 if (next_cpu < ncpus) 3010 lwkt_forwardmsg(ifnet_portfn(next_cpu), lmsg); 3011 else 3012 lwkt_replymsg(lmsg, 0); 3013 } 3014 3015 int 3016 ifnet_domsg(struct lwkt_msg *lmsg, int cpu) 3017 { 3018 KKASSERT(cpu < ncpus); 3019 return lwkt_domsg(ifnet_portfn(cpu), lmsg, 0); 3020 } 3021 3022 void 3023 ifnet_sendmsg(struct lwkt_msg *lmsg, int cpu) 3024 { 3025 KKASSERT(cpu < ncpus); 3026 lwkt_sendmsg(ifnet_portfn(cpu), lmsg); 3027 } 3028 3029 /* 3030 * Generic netmsg service loop. Some protocols may roll their own but all 3031 * must do the basic command dispatch function call done here. 3032 */ 3033 static void 3034 ifnet_service_loop(void *arg __unused) 3035 { 3036 netmsg_t msg; 3037 3038 while ((msg = lwkt_waitport(&curthread->td_msgport, 0))) { 3039 KASSERT(msg->base.nm_dispatch, ("ifnet_service: badmsg")); 3040 msg->base.nm_dispatch(msg); 3041 } 3042 } 3043 3044 static void 3045 if_start_rollup(void) 3046 { 3047 struct ifsubq_stage_head *head = &ifsubq_stage_heads[mycpuid]; 3048 struct ifsubq_stage *stage; 3049 3050 crit_enter(); 3051 3052 while ((stage = TAILQ_FIRST(&head->stg_head)) != NULL) { 3053 struct ifaltq_subque *ifsq = stage->stg_subq; 3054 int is_sched = 0; 3055 3056 if (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED) 3057 is_sched = 1; 3058 ifsq_stage_remove(head, stage); 3059 3060 if (is_sched) { 3061 ifsq_ifstart_schedule(ifsq, 1); 3062 } else { 3063 int start = 0; 3064 3065 ALTQ_SQ_LOCK(ifsq); 3066 if (!ifsq_is_started(ifsq)) { 3067 /* 3068 * Hold the subqueue interlock of 3069 * ifnet.if_start 3070 */ 3071 ifsq_set_started(ifsq); 3072 start = 1; 3073 } 3074 ALTQ_SQ_UNLOCK(ifsq); 3075 3076 if (start) 3077 ifsq_ifstart_try(ifsq, 1); 3078 } 3079 KKASSERT((stage->stg_flags & 3080 (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0); 3081 } 3082 3083 crit_exit(); 3084 } 3085 3086 static void 3087 ifnetinit(void *dummy __unused) 3088 { 3089 int i; 3090 3091 for (i = 0; i < ncpus; ++i) { 3092 struct thread *thr = &ifnet_threads[i]; 3093 3094 lwkt_create(ifnet_service_loop, NULL, NULL, 3095 thr, TDF_NOSTART|TDF_FORCE_SPINPORT|TDF_FIXEDCPU, 3096 i, "ifnet %d", i); 3097 netmsg_service_port_init(&thr->td_msgport); 3098 lwkt_schedule(thr); 3099 } 3100 3101 for (i = 0; i < ncpus; ++i) 3102 TAILQ_INIT(&ifsubq_stage_heads[i].stg_head); 3103 netisr_register_rollup(if_start_rollup, NETISR_ROLLUP_PRIO_IFSTART); 3104 } 3105 3106 struct ifnet * 3107 ifnet_byindex(unsigned short idx) 3108 { 3109 if (idx > if_index) 3110 return NULL; 3111 return ifindex2ifnet[idx]; 3112 } 3113 3114 struct ifaddr * 3115 ifaddr_byindex(unsigned short idx) 3116 { 3117 struct ifnet *ifp; 3118 3119 ifp = ifnet_byindex(idx); 3120 if (!ifp) 3121 return NULL; 3122 return TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa; 3123 } 3124 3125 void 3126 if_register_com_alloc(u_char type, 3127 if_com_alloc_t *a, if_com_free_t *f) 3128 { 3129 3130 KASSERT(if_com_alloc[type] == NULL, 3131 ("if_register_com_alloc: %d already registered", type)); 3132 KASSERT(if_com_free[type] == NULL, 3133 ("if_register_com_alloc: %d free already registered", type)); 3134 3135 if_com_alloc[type] = a; 3136 if_com_free[type] = f; 3137 } 3138 3139 void 3140 if_deregister_com_alloc(u_char type) 3141 { 3142 3143 KASSERT(if_com_alloc[type] != NULL, 3144 ("if_deregister_com_alloc: %d not registered", type)); 3145 KASSERT(if_com_free[type] != NULL, 3146 ("if_deregister_com_alloc: %d free not registered", type)); 3147 if_com_alloc[type] = NULL; 3148 if_com_free[type] = NULL; 3149 } 3150 3151 int 3152 if_ring_count2(int cnt, int cnt_max) 3153 { 3154 int shift = 0; 3155 3156 KASSERT(cnt_max >= 1 && powerof2(cnt_max), 3157 ("invalid ring count max %d", cnt_max)); 3158 3159 if (cnt <= 0) 3160 cnt = cnt_max; 3161 if (cnt > ncpus2) 3162 cnt = ncpus2; 3163 if (cnt > cnt_max) 3164 cnt = cnt_max; 3165 3166 while ((1 << (shift + 1)) <= cnt) 3167 ++shift; 3168 cnt = 1 << shift; 3169 3170 KASSERT(cnt >= 1 && cnt <= ncpus2 && cnt <= cnt_max, 3171 ("calculate cnt %d, ncpus2 %d, cnt max %d", 3172 cnt, ncpus2, cnt_max)); 3173 return cnt; 3174 } 3175 3176 void 3177 ifq_set_maxlen(struct ifaltq *ifq, int len) 3178 { 3179 ifq->altq_maxlen = len + (ncpus * ifsq_stage_cntmax); 3180 } 3181 3182 int 3183 ifq_mapsubq_default(struct ifaltq *ifq __unused, int cpuid __unused) 3184 { 3185 return ALTQ_SUBQ_INDEX_DEFAULT; 3186 } 3187 3188 int 3189 ifq_mapsubq_mask(struct ifaltq *ifq, int cpuid) 3190 { 3191 return (cpuid & ifq->altq_subq_mask); 3192 } 3193 3194 static void 3195 ifsq_watchdog(void *arg) 3196 { 3197 struct ifsubq_watchdog *wd = arg; 3198 struct ifnet *ifp; 3199 3200 if (__predict_true(wd->wd_timer == 0 || --wd->wd_timer)) 3201 goto done; 3202 3203 ifp = ifsq_get_ifp(wd->wd_subq); 3204 if (ifnet_tryserialize_all(ifp)) { 3205 wd->wd_watchdog(wd->wd_subq); 3206 ifnet_deserialize_all(ifp); 3207 } else { 3208 /* try again next timeout */ 3209 wd->wd_timer = 1; 3210 } 3211 done: 3212 ifsq_watchdog_reset(wd); 3213 } 3214 3215 static void 3216 ifsq_watchdog_reset(struct ifsubq_watchdog *wd) 3217 { 3218 callout_reset_bycpu(&wd->wd_callout, hz, ifsq_watchdog, wd, 3219 ifsq_get_cpuid(wd->wd_subq)); 3220 } 3221 3222 void 3223 ifsq_watchdog_init(struct ifsubq_watchdog *wd, struct ifaltq_subque *ifsq, 3224 ifsq_watchdog_t watchdog) 3225 { 3226 callout_init_mp(&wd->wd_callout); 3227 wd->wd_timer = 0; 3228 wd->wd_subq = ifsq; 3229 wd->wd_watchdog = watchdog; 3230 } 3231 3232 void 3233 ifsq_watchdog_start(struct ifsubq_watchdog *wd) 3234 { 3235 wd->wd_timer = 0; 3236 ifsq_watchdog_reset(wd); 3237 } 3238 3239 void 3240 ifsq_watchdog_stop(struct ifsubq_watchdog *wd) 3241 { 3242 wd->wd_timer = 0; 3243 callout_stop(&wd->wd_callout); 3244 } 3245