if.c revision 1.447
1 /* $NetBSD: if.c,v 1.447 2019/03/23 09:48:04 pgoyette Exp $ */ 2 3 /*- 4 * Copyright (c) 1999, 2000, 2001, 2008 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by William Studenmund and Jason R. Thorpe. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 */ 31 32 /* 33 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. 34 * All rights reserved. 35 * 36 * Redistribution and use in source and binary forms, with or without 37 * modification, are permitted provided that the following conditions 38 * are met: 39 * 1. Redistributions of source code must retain the above copyright 40 * notice, this list of conditions and the following disclaimer. 41 * 2. Redistributions in binary form must reproduce the above copyright 42 * notice, this list of conditions and the following disclaimer in the 43 * documentation and/or other materials provided with the distribution. 44 * 3. Neither the name of the project nor the names of its contributors 45 * may be used to endorse or promote products derived from this software 46 * without specific prior written permission. 47 * 48 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND 49 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 50 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 51 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE 52 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 53 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 54 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 55 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 56 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 57 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 58 * SUCH DAMAGE. 59 */ 60 61 /* 62 * Copyright (c) 1980, 1986, 1993 63 * The Regents of the University of California. All rights reserved. 64 * 65 * Redistribution and use in source and binary forms, with or without 66 * modification, are permitted provided that the following conditions 67 * are met: 68 * 1. Redistributions of source code must retain the above copyright 69 * notice, this list of conditions and the following disclaimer. 70 * 2. Redistributions in binary form must reproduce the above copyright 71 * notice, this list of conditions and the following disclaimer in the 72 * documentation and/or other materials provided with the distribution. 73 * 3. Neither the name of the University nor the names of its contributors 74 * may be used to endorse or promote products derived from this software 75 * without specific prior written permission. 76 * 77 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 78 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 79 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 80 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 81 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 82 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 83 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 84 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 85 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 86 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 87 * SUCH DAMAGE. 88 * 89 * @(#)if.c 8.5 (Berkeley) 1/9/95 90 */ 91 92 #include <sys/cdefs.h> 93 __KERNEL_RCSID(0, "$NetBSD: if.c,v 1.447 2019/03/23 09:48:04 pgoyette Exp $"); 94 95 #if defined(_KERNEL_OPT) 96 #include "opt_inet.h" 97 #include "opt_ipsec.h" 98 #include "opt_atalk.h" 99 #include "opt_wlan.h" 100 #include "opt_net_mpsafe.h" 101 #include "opt_mrouting.h" 102 #endif 103 104 #include <sys/param.h> 105 #include <sys/mbuf.h> 106 #include <sys/systm.h> 107 #include <sys/callout.h> 108 #include <sys/proc.h> 109 #include <sys/socket.h> 110 #include <sys/socketvar.h> 111 #include <sys/domain.h> 112 #include <sys/protosw.h> 113 #include <sys/kernel.h> 114 #include <sys/ioctl.h> 115 #include <sys/sysctl.h> 116 #include <sys/syslog.h> 117 #include <sys/kauth.h> 118 #include <sys/kmem.h> 119 #include <sys/xcall.h> 120 #include <sys/cpu.h> 121 #include <sys/intr.h> 122 #include <sys/module_hook.h> 123 #include <sys/compat_stub.h> 124 125 #include <net/if.h> 126 #include <net/if_dl.h> 127 #include <net/if_ether.h> 128 #include <net/if_media.h> 129 #include <net80211/ieee80211.h> 130 #include <net80211/ieee80211_ioctl.h> 131 #include <net/if_types.h> 132 #include <net/route.h> 133 #include <net/netisr.h> 134 #include <sys/module.h> 135 #ifdef NETATALK 136 #include <netatalk/at_extern.h> 137 #include <netatalk/at.h> 138 #endif 139 #include <net/pfil.h> 140 #include <netinet/in.h> 141 #include <netinet/in_var.h> 142 #include <netinet/ip_encap.h> 143 #include <net/bpf.h> 144 145 #ifdef INET6 146 #include <netinet6/in6_var.h> 147 #include <netinet6/nd6.h> 148 #endif 149 150 #include "ether.h" 151 #include "fddi.h" 152 #include "token.h" 153 154 #include "bridge.h" 155 #if NBRIDGE > 0 156 #include <net/if_bridgevar.h> 157 #endif 158 159 #include "carp.h" 160 #if NCARP > 0 161 #include <netinet/ip_carp.h> 162 #endif 163 164 #include <compat/sys/sockio.h> 165 #include <compat/sys/socket.h> 166 167 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address"); 168 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address"); 169 170 /* 171 * Global list of interfaces. 172 */ 173 /* DEPRECATED. Remove it once kvm(3) users disappeared */ 174 struct ifnet_head ifnet_list; 175 176 struct pslist_head ifnet_pslist; 177 static ifnet_t ** ifindex2ifnet = NULL; 178 static u_int if_index = 1; 179 static size_t if_indexlim = 0; 180 static uint64_t index_gen; 181 /* Mutex to protect the above objects. */ 182 kmutex_t ifnet_mtx __cacheline_aligned; 183 static struct psref_class *ifnet_psref_class __read_mostly; 184 static pserialize_t ifnet_psz; 185 186 static kmutex_t if_clone_mtx; 187 188 struct ifnet *lo0ifp; 189 int ifqmaxlen = IFQ_MAXLEN; 190 191 struct psref_class *ifa_psref_class __read_mostly; 192 193 static int if_delroute_matcher(struct rtentry *, void *); 194 195 static bool if_is_unit(const char *); 196 static struct if_clone *if_clone_lookup(const char *, int *); 197 198 static LIST_HEAD(, if_clone) if_cloners = LIST_HEAD_INITIALIZER(if_cloners); 199 static int if_cloners_count; 200 201 /* Packet filtering hook for interfaces. */ 202 pfil_head_t * if_pfil __read_mostly; 203 204 static kauth_listener_t if_listener; 205 206 static int doifioctl(struct socket *, u_long, void *, struct lwp *); 207 static void if_detach_queues(struct ifnet *, struct ifqueue *); 208 static void sysctl_sndq_setup(struct sysctllog **, const char *, 209 struct ifaltq *); 210 static void if_slowtimo(void *); 211 static void if_attachdomain1(struct ifnet *); 212 static int ifconf(u_long, void *); 213 static int if_transmit(struct ifnet *, struct mbuf *); 214 static int if_clone_create(const char *); 215 static int if_clone_destroy(const char *); 216 static void if_link_state_change_si(void *); 217 static void if_up_locked(struct ifnet *); 218 static void _if_down(struct ifnet *); 219 static void if_down_deactivated(struct ifnet *); 220 221 struct if_percpuq { 222 struct ifnet *ipq_ifp; 223 void *ipq_si; 224 struct percpu *ipq_ifqs; /* struct ifqueue */ 225 }; 226 227 static struct mbuf *if_percpuq_dequeue(struct if_percpuq *); 228 229 static void if_percpuq_drops(void *, void *, struct cpu_info *); 230 static int sysctl_percpuq_drops_handler(SYSCTLFN_PROTO); 231 static void sysctl_percpuq_setup(struct sysctllog **, const char *, 232 struct if_percpuq *); 233 234 struct if_deferred_start { 235 struct ifnet *ids_ifp; 236 void (*ids_if_start)(struct ifnet *); 237 void *ids_si; 238 }; 239 240 static void if_deferred_start_softint(void *); 241 static void if_deferred_start_common(struct ifnet *); 242 static void if_deferred_start_destroy(struct ifnet *); 243 244 #if defined(INET) || defined(INET6) 245 static void sysctl_net_pktq_setup(struct sysctllog **, int); 246 #endif 247 248 /* 249 * Hook for if_vlan - needed by if_agr 250 */ 251 struct if_vlan_vlan_input_hook_t if_vlan_vlan_input_hook; 252 253 static void if_sysctl_setup(struct sysctllog **); 254 255 static int 256 if_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie, 257 void *arg0, void *arg1, void *arg2, void *arg3) 258 { 259 int result; 260 enum kauth_network_req req; 261 262 result = KAUTH_RESULT_DEFER; 263 req = (enum kauth_network_req)arg1; 264 265 if (action != KAUTH_NETWORK_INTERFACE) 266 return result; 267 268 if ((req == KAUTH_REQ_NETWORK_INTERFACE_GET) || 269 (req == KAUTH_REQ_NETWORK_INTERFACE_SET)) 270 result = KAUTH_RESULT_ALLOW; 271 272 return result; 273 } 274 275 /* 276 * Network interface utility routines. 277 * 278 * Routines with ifa_ifwith* names take sockaddr *'s as 279 * parameters. 280 */ 281 void 282 ifinit(void) 283 { 284 285 #if (defined(INET) || defined(INET6)) 286 encapinit(); 287 #endif 288 289 if_listener = kauth_listen_scope(KAUTH_SCOPE_NETWORK, 290 if_listener_cb, NULL); 291 292 /* interfaces are available, inform socket code */ 293 ifioctl = doifioctl; 294 } 295 296 /* 297 * XXX Initialization before configure(). 298 * XXX hack to get pfil_add_hook working in autoconf. 299 */ 300 void 301 ifinit1(void) 302 { 303 304 #ifdef NET_MPSAFE 305 printf("NET_MPSAFE enabled\n"); 306 #endif 307 308 mutex_init(&if_clone_mtx, MUTEX_DEFAULT, IPL_NONE); 309 310 TAILQ_INIT(&ifnet_list); 311 mutex_init(&ifnet_mtx, MUTEX_DEFAULT, IPL_NONE); 312 ifnet_psz = pserialize_create(); 313 ifnet_psref_class = psref_class_create("ifnet", IPL_SOFTNET); 314 ifa_psref_class = psref_class_create("ifa", IPL_SOFTNET); 315 PSLIST_INIT(&ifnet_pslist); 316 317 if_indexlim = 8; 318 319 if_pfil = pfil_head_create(PFIL_TYPE_IFNET, NULL); 320 KASSERT(if_pfil != NULL); 321 322 #if NETHER > 0 || NFDDI > 0 || defined(NETATALK) || NTOKEN > 0 || defined(WLAN) 323 etherinit(); 324 #endif 325 } 326 327 /* XXX must be after domaininit() */ 328 void 329 ifinit_post(void) 330 { 331 332 if_sysctl_setup(NULL); 333 } 334 335 ifnet_t * 336 if_alloc(u_char type) 337 { 338 return kmem_zalloc(sizeof(ifnet_t), KM_SLEEP); 339 } 340 341 void 342 if_free(ifnet_t *ifp) 343 { 344 kmem_free(ifp, sizeof(ifnet_t)); 345 } 346 347 void 348 if_initname(struct ifnet *ifp, const char *name, int unit) 349 { 350 (void)snprintf(ifp->if_xname, sizeof(ifp->if_xname), 351 "%s%d", name, unit); 352 } 353 354 /* 355 * Null routines used while an interface is going away. These routines 356 * just return an error. 357 */ 358 359 int 360 if_nulloutput(struct ifnet *ifp, struct mbuf *m, 361 const struct sockaddr *so, const struct rtentry *rt) 362 { 363 364 return ENXIO; 365 } 366 367 void 368 if_nullinput(struct ifnet *ifp, struct mbuf *m) 369 { 370 371 /* Nothing. */ 372 } 373 374 void 375 if_nullstart(struct ifnet *ifp) 376 { 377 378 /* Nothing. */ 379 } 380 381 int 382 if_nulltransmit(struct ifnet *ifp, struct mbuf *m) 383 { 384 385 m_freem(m); 386 return ENXIO; 387 } 388 389 int 390 if_nullioctl(struct ifnet *ifp, u_long cmd, void *data) 391 { 392 393 return ENXIO; 394 } 395 396 int 397 if_nullinit(struct ifnet *ifp) 398 { 399 400 return ENXIO; 401 } 402 403 void 404 if_nullstop(struct ifnet *ifp, int disable) 405 { 406 407 /* Nothing. */ 408 } 409 410 void 411 if_nullslowtimo(struct ifnet *ifp) 412 { 413 414 /* Nothing. */ 415 } 416 417 void 418 if_nulldrain(struct ifnet *ifp) 419 { 420 421 /* Nothing. */ 422 } 423 424 void 425 if_set_sadl(struct ifnet *ifp, const void *lla, u_char addrlen, bool factory) 426 { 427 struct ifaddr *ifa; 428 struct sockaddr_dl *sdl; 429 430 ifp->if_addrlen = addrlen; 431 if_alloc_sadl(ifp); 432 ifa = ifp->if_dl; 433 sdl = satosdl(ifa->ifa_addr); 434 435 (void)sockaddr_dl_setaddr(sdl, sdl->sdl_len, lla, ifp->if_addrlen); 436 if (factory) { 437 KASSERT(ifp->if_hwdl == NULL); 438 ifp->if_hwdl = ifp->if_dl; 439 ifaref(ifp->if_hwdl); 440 } 441 /* TBD routing socket */ 442 } 443 444 struct ifaddr * 445 if_dl_create(const struct ifnet *ifp, const struct sockaddr_dl **sdlp) 446 { 447 unsigned socksize, ifasize; 448 int addrlen, namelen; 449 struct sockaddr_dl *mask, *sdl; 450 struct ifaddr *ifa; 451 452 namelen = strlen(ifp->if_xname); 453 addrlen = ifp->if_addrlen; 454 socksize = roundup(sockaddr_dl_measure(namelen, addrlen), sizeof(long)); 455 ifasize = sizeof(*ifa) + 2 * socksize; 456 ifa = malloc(ifasize, M_IFADDR, M_WAITOK|M_ZERO); 457 458 sdl = (struct sockaddr_dl *)(ifa + 1); 459 mask = (struct sockaddr_dl *)(socksize + (char *)sdl); 460 461 sockaddr_dl_init(sdl, socksize, ifp->if_index, ifp->if_type, 462 ifp->if_xname, namelen, NULL, addrlen); 463 mask->sdl_family = AF_LINK; 464 mask->sdl_len = sockaddr_dl_measure(namelen, 0); 465 memset(&mask->sdl_data[0], 0xff, namelen); 466 ifa->ifa_rtrequest = link_rtrequest; 467 ifa->ifa_addr = (struct sockaddr *)sdl; 468 ifa->ifa_netmask = (struct sockaddr *)mask; 469 ifa_psref_init(ifa); 470 471 *sdlp = sdl; 472 473 return ifa; 474 } 475 476 static void 477 if_sadl_setrefs(struct ifnet *ifp, struct ifaddr *ifa) 478 { 479 const struct sockaddr_dl *sdl; 480 481 ifp->if_dl = ifa; 482 ifaref(ifa); 483 sdl = satosdl(ifa->ifa_addr); 484 ifp->if_sadl = sdl; 485 } 486 487 /* 488 * Allocate the link level name for the specified interface. This 489 * is an attachment helper. It must be called after ifp->if_addrlen 490 * is initialized, which may not be the case when if_attach() is 491 * called. 492 */ 493 void 494 if_alloc_sadl(struct ifnet *ifp) 495 { 496 struct ifaddr *ifa; 497 const struct sockaddr_dl *sdl; 498 499 /* 500 * If the interface already has a link name, release it 501 * now. This is useful for interfaces that can change 502 * link types, and thus switch link names often. 503 */ 504 if (ifp->if_sadl != NULL) 505 if_free_sadl(ifp, 0); 506 507 ifa = if_dl_create(ifp, &sdl); 508 509 ifa_insert(ifp, ifa); 510 if_sadl_setrefs(ifp, ifa); 511 } 512 513 static void 514 if_deactivate_sadl(struct ifnet *ifp) 515 { 516 struct ifaddr *ifa; 517 518 KASSERT(ifp->if_dl != NULL); 519 520 ifa = ifp->if_dl; 521 522 ifp->if_sadl = NULL; 523 524 ifp->if_dl = NULL; 525 ifafree(ifa); 526 } 527 528 static void 529 if_replace_sadl(struct ifnet *ifp, struct ifaddr *ifa) 530 { 531 struct ifaddr *old; 532 533 KASSERT(ifp->if_dl != NULL); 534 535 old = ifp->if_dl; 536 537 ifaref(ifa); 538 /* XXX Update if_dl and if_sadl atomically */ 539 ifp->if_dl = ifa; 540 ifp->if_sadl = satosdl(ifa->ifa_addr); 541 542 ifafree(old); 543 } 544 545 void 546 if_activate_sadl(struct ifnet *ifp, struct ifaddr *ifa0, 547 const struct sockaddr_dl *sdl) 548 { 549 int s, ss; 550 struct ifaddr *ifa; 551 int bound = curlwp_bind(); 552 553 KASSERT(ifa_held(ifa0)); 554 555 s = splsoftnet(); 556 557 if_replace_sadl(ifp, ifa0); 558 559 ss = pserialize_read_enter(); 560 IFADDR_READER_FOREACH(ifa, ifp) { 561 struct psref psref; 562 ifa_acquire(ifa, &psref); 563 pserialize_read_exit(ss); 564 565 rtinit(ifa, RTM_LLINFO_UPD, 0); 566 567 ss = pserialize_read_enter(); 568 ifa_release(ifa, &psref); 569 } 570 pserialize_read_exit(ss); 571 572 splx(s); 573 curlwp_bindx(bound); 574 } 575 576 /* 577 * Free the link level name for the specified interface. This is 578 * a detach helper. This is called from if_detach(). 579 */ 580 void 581 if_free_sadl(struct ifnet *ifp, int factory) 582 { 583 struct ifaddr *ifa; 584 int s; 585 586 if (factory && ifp->if_hwdl != NULL) { 587 ifa = ifp->if_hwdl; 588 ifp->if_hwdl = NULL; 589 ifafree(ifa); 590 } 591 592 ifa = ifp->if_dl; 593 if (ifa == NULL) { 594 KASSERT(ifp->if_sadl == NULL); 595 return; 596 } 597 598 KASSERT(ifp->if_sadl != NULL); 599 600 s = splsoftnet(); 601 KASSERT(ifa->ifa_addr->sa_family == AF_LINK); 602 ifa_remove(ifp, ifa); 603 if_deactivate_sadl(ifp); 604 splx(s); 605 } 606 607 static void 608 if_getindex(ifnet_t *ifp) 609 { 610 bool hitlimit = false; 611 612 ifp->if_index_gen = index_gen++; 613 614 ifp->if_index = if_index; 615 if (ifindex2ifnet == NULL) { 616 if_index++; 617 goto skip; 618 } 619 while (if_byindex(ifp->if_index)) { 620 /* 621 * If we hit USHRT_MAX, we skip back to 0 since 622 * there are a number of places where the value 623 * of if_index or if_index itself is compared 624 * to or stored in an unsigned short. By 625 * jumping back, we won't botch those assignments 626 * or comparisons. 627 */ 628 if (++if_index == 0) { 629 if_index = 1; 630 } else if (if_index == USHRT_MAX) { 631 /* 632 * However, if we have to jump back to 633 * zero *twice* without finding an empty 634 * slot in ifindex2ifnet[], then there 635 * there are too many (>65535) interfaces. 636 */ 637 if (hitlimit) { 638 panic("too many interfaces"); 639 } 640 hitlimit = true; 641 if_index = 1; 642 } 643 ifp->if_index = if_index; 644 } 645 skip: 646 /* 647 * ifindex2ifnet is indexed by if_index. Since if_index will 648 * grow dynamically, it should grow too. 649 */ 650 if (ifindex2ifnet == NULL || ifp->if_index >= if_indexlim) { 651 size_t m, n, oldlim; 652 void *q; 653 654 oldlim = if_indexlim; 655 while (ifp->if_index >= if_indexlim) 656 if_indexlim <<= 1; 657 658 /* grow ifindex2ifnet */ 659 m = oldlim * sizeof(struct ifnet *); 660 n = if_indexlim * sizeof(struct ifnet *); 661 q = malloc(n, M_IFADDR, M_WAITOK|M_ZERO); 662 if (ifindex2ifnet != NULL) { 663 memcpy(q, ifindex2ifnet, m); 664 free(ifindex2ifnet, M_IFADDR); 665 } 666 ifindex2ifnet = (struct ifnet **)q; 667 } 668 ifindex2ifnet[ifp->if_index] = ifp; 669 } 670 671 /* 672 * Initialize an interface and assign an index for it. 673 * 674 * It must be called prior to a device specific attach routine 675 * (e.g., ether_ifattach and ieee80211_ifattach) or if_alloc_sadl, 676 * and be followed by if_register: 677 * 678 * if_initialize(ifp); 679 * ether_ifattach(ifp, enaddr); 680 * if_register(ifp); 681 */ 682 int 683 if_initialize(ifnet_t *ifp) 684 { 685 int rv = 0; 686 687 KASSERT(if_indexlim > 0); 688 TAILQ_INIT(&ifp->if_addrlist); 689 690 /* 691 * Link level name is allocated later by a separate call to 692 * if_alloc_sadl(). 693 */ 694 695 if (ifp->if_snd.ifq_maxlen == 0) 696 ifp->if_snd.ifq_maxlen = ifqmaxlen; 697 698 ifp->if_broadcastaddr = 0; /* reliably crash if used uninitialized */ 699 700 ifp->if_link_state = LINK_STATE_UNKNOWN; 701 ifp->if_link_queue = -1; /* all bits set, see link_state_change() */ 702 703 ifp->if_capenable = 0; 704 ifp->if_csum_flags_tx = 0; 705 ifp->if_csum_flags_rx = 0; 706 707 #ifdef ALTQ 708 ifp->if_snd.altq_type = 0; 709 ifp->if_snd.altq_disc = NULL; 710 ifp->if_snd.altq_flags &= ALTQF_CANTCHANGE; 711 ifp->if_snd.altq_tbr = NULL; 712 ifp->if_snd.altq_ifp = ifp; 713 #endif 714 715 IFQ_LOCK_INIT(&ifp->if_snd); 716 717 ifp->if_pfil = pfil_head_create(PFIL_TYPE_IFNET, ifp); 718 pfil_run_ifhooks(if_pfil, PFIL_IFNET_ATTACH, ifp); 719 720 IF_AFDATA_LOCK_INIT(ifp); 721 722 if (if_is_link_state_changeable(ifp)) { 723 u_int flags = SOFTINT_NET; 724 flags |= if_is_mpsafe(ifp) ? SOFTINT_MPSAFE : 0; 725 ifp->if_link_si = softint_establish(flags, 726 if_link_state_change_si, ifp); 727 if (ifp->if_link_si == NULL) { 728 rv = ENOMEM; 729 goto fail; 730 } 731 } 732 733 PSLIST_ENTRY_INIT(ifp, if_pslist_entry); 734 PSLIST_INIT(&ifp->if_addr_pslist); 735 psref_target_init(&ifp->if_psref, ifnet_psref_class); 736 ifp->if_ioctl_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 737 LIST_INIT(&ifp->if_multiaddrs); 738 739 IFNET_GLOBAL_LOCK(); 740 if_getindex(ifp); 741 IFNET_GLOBAL_UNLOCK(); 742 743 return 0; 744 745 fail: 746 IF_AFDATA_LOCK_DESTROY(ifp); 747 748 pfil_run_ifhooks(if_pfil, PFIL_IFNET_DETACH, ifp); 749 (void)pfil_head_destroy(ifp->if_pfil); 750 751 IFQ_LOCK_DESTROY(&ifp->if_snd); 752 753 return rv; 754 } 755 756 /* 757 * Register an interface to the list of "active" interfaces. 758 */ 759 void 760 if_register(ifnet_t *ifp) 761 { 762 /* 763 * If the driver has not supplied its own if_ioctl, then 764 * supply the default. 765 */ 766 if (ifp->if_ioctl == NULL) 767 ifp->if_ioctl = ifioctl_common; 768 769 sysctl_sndq_setup(&ifp->if_sysctl_log, ifp->if_xname, &ifp->if_snd); 770 771 if (!STAILQ_EMPTY(&domains)) 772 if_attachdomain1(ifp); 773 774 /* Announce the interface. */ 775 rt_ifannouncemsg(ifp, IFAN_ARRIVAL); 776 777 if (ifp->if_slowtimo != NULL) { 778 ifp->if_slowtimo_ch = 779 kmem_zalloc(sizeof(*ifp->if_slowtimo_ch), KM_SLEEP); 780 callout_init(ifp->if_slowtimo_ch, 0); 781 callout_setfunc(ifp->if_slowtimo_ch, if_slowtimo, ifp); 782 if_slowtimo(ifp); 783 } 784 785 if (ifp->if_transmit == NULL || ifp->if_transmit == if_nulltransmit) 786 ifp->if_transmit = if_transmit; 787 788 IFNET_GLOBAL_LOCK(); 789 TAILQ_INSERT_TAIL(&ifnet_list, ifp, if_list); 790 IFNET_WRITER_INSERT_TAIL(ifp); 791 IFNET_GLOBAL_UNLOCK(); 792 } 793 794 /* 795 * The if_percpuq framework 796 * 797 * It allows network device drivers to execute the network stack 798 * in softint (so called softint-based if_input). It utilizes 799 * softint and percpu ifqueue. It doesn't distribute any packets 800 * between CPUs, unlike pktqueue(9). 801 * 802 * Currently we support two options for device drivers to apply the framework: 803 * - Use it implicitly with less changes 804 * - If you use if_attach in driver's _attach function and if_input in 805 * driver's Rx interrupt handler, a packet is queued and a softint handles 806 * the packet implicitly 807 * - Use it explicitly in each driver (recommended) 808 * - You can use if_percpuq_* directly in your driver 809 * - In this case, you need to allocate struct if_percpuq in driver's softc 810 * - See wm(4) as a reference implementation 811 */ 812 813 static void 814 if_percpuq_softint(void *arg) 815 { 816 struct if_percpuq *ipq = arg; 817 struct ifnet *ifp = ipq->ipq_ifp; 818 struct mbuf *m; 819 820 while ((m = if_percpuq_dequeue(ipq)) != NULL) { 821 ifp->if_ipackets++; 822 bpf_mtap(ifp, m, BPF_D_IN); 823 824 ifp->_if_input(ifp, m); 825 } 826 } 827 828 static void 829 if_percpuq_init_ifq(void *p, void *arg __unused, struct cpu_info *ci __unused) 830 { 831 struct ifqueue *const ifq = p; 832 833 memset(ifq, 0, sizeof(*ifq)); 834 ifq->ifq_maxlen = IFQ_MAXLEN; 835 } 836 837 struct if_percpuq * 838 if_percpuq_create(struct ifnet *ifp) 839 { 840 struct if_percpuq *ipq; 841 u_int flags = SOFTINT_NET; 842 843 flags |= if_is_mpsafe(ifp) ? SOFTINT_MPSAFE : 0; 844 845 ipq = kmem_zalloc(sizeof(*ipq), KM_SLEEP); 846 ipq->ipq_ifp = ifp; 847 ipq->ipq_si = softint_establish(flags, if_percpuq_softint, ipq); 848 ipq->ipq_ifqs = percpu_alloc(sizeof(struct ifqueue)); 849 percpu_foreach(ipq->ipq_ifqs, &if_percpuq_init_ifq, NULL); 850 851 sysctl_percpuq_setup(&ifp->if_sysctl_log, ifp->if_xname, ipq); 852 853 return ipq; 854 } 855 856 static struct mbuf * 857 if_percpuq_dequeue(struct if_percpuq *ipq) 858 { 859 struct mbuf *m; 860 struct ifqueue *ifq; 861 int s; 862 863 s = splnet(); 864 ifq = percpu_getref(ipq->ipq_ifqs); 865 IF_DEQUEUE(ifq, m); 866 percpu_putref(ipq->ipq_ifqs); 867 splx(s); 868 869 return m; 870 } 871 872 static void 873 if_percpuq_purge_ifq(void *p, void *arg __unused, struct cpu_info *ci __unused) 874 { 875 struct ifqueue *const ifq = p; 876 877 IF_PURGE(ifq); 878 } 879 880 void 881 if_percpuq_destroy(struct if_percpuq *ipq) 882 { 883 884 /* if_detach may already destroy it */ 885 if (ipq == NULL) 886 return; 887 888 softint_disestablish(ipq->ipq_si); 889 percpu_foreach(ipq->ipq_ifqs, &if_percpuq_purge_ifq, NULL); 890 percpu_free(ipq->ipq_ifqs, sizeof(struct ifqueue)); 891 kmem_free(ipq, sizeof(*ipq)); 892 } 893 894 void 895 if_percpuq_enqueue(struct if_percpuq *ipq, struct mbuf *m) 896 { 897 struct ifqueue *ifq; 898 int s; 899 900 KASSERT(ipq != NULL); 901 902 s = splnet(); 903 ifq = percpu_getref(ipq->ipq_ifqs); 904 if (IF_QFULL(ifq)) { 905 IF_DROP(ifq); 906 percpu_putref(ipq->ipq_ifqs); 907 m_freem(m); 908 goto out; 909 } 910 IF_ENQUEUE(ifq, m); 911 percpu_putref(ipq->ipq_ifqs); 912 913 softint_schedule(ipq->ipq_si); 914 out: 915 splx(s); 916 } 917 918 static void 919 if_percpuq_drops(void *p, void *arg, struct cpu_info *ci __unused) 920 { 921 struct ifqueue *const ifq = p; 922 int *sum = arg; 923 924 *sum += ifq->ifq_drops; 925 } 926 927 static int 928 sysctl_percpuq_drops_handler(SYSCTLFN_ARGS) 929 { 930 struct sysctlnode node; 931 struct if_percpuq *ipq; 932 int sum = 0; 933 int error; 934 935 node = *rnode; 936 ipq = node.sysctl_data; 937 938 percpu_foreach(ipq->ipq_ifqs, if_percpuq_drops, &sum); 939 940 node.sysctl_data = ∑ 941 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 942 if (error != 0 || newp == NULL) 943 return error; 944 945 return 0; 946 } 947 948 static void 949 sysctl_percpuq_setup(struct sysctllog **clog, const char* ifname, 950 struct if_percpuq *ipq) 951 { 952 const struct sysctlnode *cnode, *rnode; 953 954 if (sysctl_createv(clog, 0, NULL, &rnode, 955 CTLFLAG_PERMANENT, 956 CTLTYPE_NODE, "interfaces", 957 SYSCTL_DESCR("Per-interface controls"), 958 NULL, 0, NULL, 0, 959 CTL_NET, CTL_CREATE, CTL_EOL) != 0) 960 goto bad; 961 962 if (sysctl_createv(clog, 0, &rnode, &rnode, 963 CTLFLAG_PERMANENT, 964 CTLTYPE_NODE, ifname, 965 SYSCTL_DESCR("Interface controls"), 966 NULL, 0, NULL, 0, 967 CTL_CREATE, CTL_EOL) != 0) 968 goto bad; 969 970 if (sysctl_createv(clog, 0, &rnode, &rnode, 971 CTLFLAG_PERMANENT, 972 CTLTYPE_NODE, "rcvq", 973 SYSCTL_DESCR("Interface input queue controls"), 974 NULL, 0, NULL, 0, 975 CTL_CREATE, CTL_EOL) != 0) 976 goto bad; 977 978 #ifdef NOTYET 979 /* XXX Should show each per-CPU queue length? */ 980 if (sysctl_createv(clog, 0, &rnode, &rnode, 981 CTLFLAG_PERMANENT, 982 CTLTYPE_INT, "len", 983 SYSCTL_DESCR("Current input queue length"), 984 sysctl_percpuq_len, 0, NULL, 0, 985 CTL_CREATE, CTL_EOL) != 0) 986 goto bad; 987 988 if (sysctl_createv(clog, 0, &rnode, &cnode, 989 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 990 CTLTYPE_INT, "maxlen", 991 SYSCTL_DESCR("Maximum allowed input queue length"), 992 sysctl_percpuq_maxlen_handler, 0, (void *)ipq, 0, 993 CTL_CREATE, CTL_EOL) != 0) 994 goto bad; 995 #endif 996 997 if (sysctl_createv(clog, 0, &rnode, &cnode, 998 CTLFLAG_PERMANENT, 999 CTLTYPE_INT, "drops", 1000 SYSCTL_DESCR("Total packets dropped due to full input queue"), 1001 sysctl_percpuq_drops_handler, 0, (void *)ipq, 0, 1002 CTL_CREATE, CTL_EOL) != 0) 1003 goto bad; 1004 1005 return; 1006 bad: 1007 printf("%s: could not attach sysctl nodes\n", ifname); 1008 return; 1009 } 1010 1011 /* 1012 * The deferred if_start framework 1013 * 1014 * The common APIs to defer if_start to softint when if_start is requested 1015 * from a device driver running in hardware interrupt context. 1016 */ 1017 /* 1018 * Call ifp->if_start (or equivalent) in a dedicated softint for 1019 * deferred if_start. 1020 */ 1021 static void 1022 if_deferred_start_softint(void *arg) 1023 { 1024 struct if_deferred_start *ids = arg; 1025 struct ifnet *ifp = ids->ids_ifp; 1026 1027 ids->ids_if_start(ifp); 1028 } 1029 1030 /* 1031 * The default callback function for deferred if_start. 1032 */ 1033 static void 1034 if_deferred_start_common(struct ifnet *ifp) 1035 { 1036 int s; 1037 1038 s = splnet(); 1039 if_start_lock(ifp); 1040 splx(s); 1041 } 1042 1043 static inline bool 1044 if_snd_is_used(struct ifnet *ifp) 1045 { 1046 1047 return ifp->if_transmit == NULL || ifp->if_transmit == if_nulltransmit || 1048 ifp->if_transmit == if_transmit || 1049 ALTQ_IS_ENABLED(&ifp->if_snd); 1050 } 1051 1052 /* 1053 * Schedule deferred if_start. 1054 */ 1055 void 1056 if_schedule_deferred_start(struct ifnet *ifp) 1057 { 1058 1059 KASSERT(ifp->if_deferred_start != NULL); 1060 1061 if (if_snd_is_used(ifp) && IFQ_IS_EMPTY(&ifp->if_snd)) 1062 return; 1063 1064 softint_schedule(ifp->if_deferred_start->ids_si); 1065 } 1066 1067 /* 1068 * Create an instance of deferred if_start. A driver should call the function 1069 * only if the driver needs deferred if_start. Drivers can setup their own 1070 * deferred if_start function via 2nd argument. 1071 */ 1072 void 1073 if_deferred_start_init(struct ifnet *ifp, void (*func)(struct ifnet *)) 1074 { 1075 struct if_deferred_start *ids; 1076 u_int flags = SOFTINT_NET; 1077 1078 flags |= if_is_mpsafe(ifp) ? SOFTINT_MPSAFE : 0; 1079 1080 ids = kmem_zalloc(sizeof(*ids), KM_SLEEP); 1081 ids->ids_ifp = ifp; 1082 ids->ids_si = softint_establish(flags, if_deferred_start_softint, ids); 1083 if (func != NULL) 1084 ids->ids_if_start = func; 1085 else 1086 ids->ids_if_start = if_deferred_start_common; 1087 1088 ifp->if_deferred_start = ids; 1089 } 1090 1091 static void 1092 if_deferred_start_destroy(struct ifnet *ifp) 1093 { 1094 1095 if (ifp->if_deferred_start == NULL) 1096 return; 1097 1098 softint_disestablish(ifp->if_deferred_start->ids_si); 1099 kmem_free(ifp->if_deferred_start, sizeof(*ifp->if_deferred_start)); 1100 ifp->if_deferred_start = NULL; 1101 } 1102 1103 /* 1104 * The common interface input routine that is called by device drivers, 1105 * which should be used only when the driver's rx handler already runs 1106 * in softint. 1107 */ 1108 void 1109 if_input(struct ifnet *ifp, struct mbuf *m) 1110 { 1111 1112 KASSERT(ifp->if_percpuq == NULL); 1113 KASSERT(!cpu_intr_p()); 1114 1115 ifp->if_ipackets++; 1116 bpf_mtap(ifp, m, BPF_D_IN); 1117 1118 ifp->_if_input(ifp, m); 1119 } 1120 1121 /* 1122 * DEPRECATED. Use if_initialize and if_register instead. 1123 * See the above comment of if_initialize. 1124 * 1125 * Note that it implicitly enables if_percpuq to make drivers easy to 1126 * migrate softint-based if_input without much changes. If you don't 1127 * want to enable it, use if_initialize instead. 1128 */ 1129 int 1130 if_attach(ifnet_t *ifp) 1131 { 1132 int rv; 1133 1134 rv = if_initialize(ifp); 1135 if (rv != 0) 1136 return rv; 1137 1138 ifp->if_percpuq = if_percpuq_create(ifp); 1139 if_register(ifp); 1140 1141 return 0; 1142 } 1143 1144 void 1145 if_attachdomain(void) 1146 { 1147 struct ifnet *ifp; 1148 int s; 1149 int bound = curlwp_bind(); 1150 1151 s = pserialize_read_enter(); 1152 IFNET_READER_FOREACH(ifp) { 1153 struct psref psref; 1154 psref_acquire(&psref, &ifp->if_psref, ifnet_psref_class); 1155 pserialize_read_exit(s); 1156 if_attachdomain1(ifp); 1157 s = pserialize_read_enter(); 1158 psref_release(&psref, &ifp->if_psref, ifnet_psref_class); 1159 } 1160 pserialize_read_exit(s); 1161 curlwp_bindx(bound); 1162 } 1163 1164 static void 1165 if_attachdomain1(struct ifnet *ifp) 1166 { 1167 struct domain *dp; 1168 int s; 1169 1170 s = splsoftnet(); 1171 1172 /* address family dependent data region */ 1173 memset(ifp->if_afdata, 0, sizeof(ifp->if_afdata)); 1174 DOMAIN_FOREACH(dp) { 1175 if (dp->dom_ifattach != NULL) 1176 ifp->if_afdata[dp->dom_family] = 1177 (*dp->dom_ifattach)(ifp); 1178 } 1179 1180 splx(s); 1181 } 1182 1183 /* 1184 * Deactivate an interface. This points all of the procedure 1185 * handles at error stubs. May be called from interrupt context. 1186 */ 1187 void 1188 if_deactivate(struct ifnet *ifp) 1189 { 1190 int s; 1191 1192 s = splsoftnet(); 1193 1194 ifp->if_output = if_nulloutput; 1195 ifp->_if_input = if_nullinput; 1196 ifp->if_start = if_nullstart; 1197 ifp->if_transmit = if_nulltransmit; 1198 ifp->if_ioctl = if_nullioctl; 1199 ifp->if_init = if_nullinit; 1200 ifp->if_stop = if_nullstop; 1201 ifp->if_slowtimo = if_nullslowtimo; 1202 ifp->if_drain = if_nulldrain; 1203 1204 /* No more packets may be enqueued. */ 1205 ifp->if_snd.ifq_maxlen = 0; 1206 1207 splx(s); 1208 } 1209 1210 bool 1211 if_is_deactivated(const struct ifnet *ifp) 1212 { 1213 1214 return ifp->if_output == if_nulloutput; 1215 } 1216 1217 void 1218 if_purgeaddrs(struct ifnet *ifp, int family, void (*purgeaddr)(struct ifaddr *)) 1219 { 1220 struct ifaddr *ifa, *nifa; 1221 int s; 1222 1223 s = pserialize_read_enter(); 1224 for (ifa = IFADDR_READER_FIRST(ifp); ifa; ifa = nifa) { 1225 nifa = IFADDR_READER_NEXT(ifa); 1226 if (ifa->ifa_addr->sa_family != family) 1227 continue; 1228 pserialize_read_exit(s); 1229 1230 (*purgeaddr)(ifa); 1231 1232 s = pserialize_read_enter(); 1233 } 1234 pserialize_read_exit(s); 1235 } 1236 1237 #ifdef IFAREF_DEBUG 1238 static struct ifaddr **ifa_list; 1239 static int ifa_list_size; 1240 1241 /* Depends on only one if_attach runs at once */ 1242 static void 1243 if_build_ifa_list(struct ifnet *ifp) 1244 { 1245 struct ifaddr *ifa; 1246 int i; 1247 1248 KASSERT(ifa_list == NULL); 1249 KASSERT(ifa_list_size == 0); 1250 1251 IFADDR_READER_FOREACH(ifa, ifp) 1252 ifa_list_size++; 1253 1254 ifa_list = kmem_alloc(sizeof(*ifa) * ifa_list_size, KM_SLEEP); 1255 i = 0; 1256 IFADDR_READER_FOREACH(ifa, ifp) { 1257 ifa_list[i++] = ifa; 1258 ifaref(ifa); 1259 } 1260 } 1261 1262 static void 1263 if_check_and_free_ifa_list(struct ifnet *ifp) 1264 { 1265 int i; 1266 struct ifaddr *ifa; 1267 1268 if (ifa_list == NULL) 1269 return; 1270 1271 for (i = 0; i < ifa_list_size; i++) { 1272 char buf[64]; 1273 1274 ifa = ifa_list[i]; 1275 sockaddr_format(ifa->ifa_addr, buf, sizeof(buf)); 1276 if (ifa->ifa_refcnt > 1) { 1277 log(LOG_WARNING, 1278 "ifa(%s) still referenced (refcnt=%d)\n", 1279 buf, ifa->ifa_refcnt - 1); 1280 } else 1281 log(LOG_DEBUG, 1282 "ifa(%s) not referenced (refcnt=%d)\n", 1283 buf, ifa->ifa_refcnt - 1); 1284 ifafree(ifa); 1285 } 1286 1287 kmem_free(ifa_list, sizeof(*ifa) * ifa_list_size); 1288 ifa_list = NULL; 1289 ifa_list_size = 0; 1290 } 1291 #endif 1292 1293 /* 1294 * Detach an interface from the list of "active" interfaces, 1295 * freeing any resources as we go along. 1296 * 1297 * NOTE: This routine must be called with a valid thread context, 1298 * as it may block. 1299 */ 1300 void 1301 if_detach(struct ifnet *ifp) 1302 { 1303 struct socket so; 1304 struct ifaddr *ifa; 1305 #ifdef IFAREF_DEBUG 1306 struct ifaddr *last_ifa = NULL; 1307 #endif 1308 struct domain *dp; 1309 const struct protosw *pr; 1310 int s, i, family, purged; 1311 uint64_t xc; 1312 1313 #ifdef IFAREF_DEBUG 1314 if_build_ifa_list(ifp); 1315 #endif 1316 /* 1317 * XXX It's kind of lame that we have to have the 1318 * XXX socket structure... 1319 */ 1320 memset(&so, 0, sizeof(so)); 1321 1322 s = splnet(); 1323 1324 sysctl_teardown(&ifp->if_sysctl_log); 1325 IFNET_LOCK(ifp); 1326 if_deactivate(ifp); 1327 IFNET_UNLOCK(ifp); 1328 1329 /* 1330 * Unlink from the list and wait for all readers to leave 1331 * from pserialize read sections. Note that we can't do 1332 * psref_target_destroy here. See below. 1333 */ 1334 IFNET_GLOBAL_LOCK(); 1335 ifindex2ifnet[ifp->if_index] = NULL; 1336 TAILQ_REMOVE(&ifnet_list, ifp, if_list); 1337 IFNET_WRITER_REMOVE(ifp); 1338 pserialize_perform(ifnet_psz); 1339 IFNET_GLOBAL_UNLOCK(); 1340 1341 if (ifp->if_slowtimo != NULL && ifp->if_slowtimo_ch != NULL) { 1342 ifp->if_slowtimo = NULL; 1343 callout_halt(ifp->if_slowtimo_ch, NULL); 1344 callout_destroy(ifp->if_slowtimo_ch); 1345 kmem_free(ifp->if_slowtimo_ch, sizeof(*ifp->if_slowtimo_ch)); 1346 } 1347 if_deferred_start_destroy(ifp); 1348 1349 /* 1350 * Do an if_down() to give protocols a chance to do something. 1351 */ 1352 if_down_deactivated(ifp); 1353 1354 #ifdef ALTQ 1355 if (ALTQ_IS_ENABLED(&ifp->if_snd)) 1356 altq_disable(&ifp->if_snd); 1357 if (ALTQ_IS_ATTACHED(&ifp->if_snd)) 1358 altq_detach(&ifp->if_snd); 1359 #endif 1360 1361 #if NCARP > 0 1362 /* Remove the interface from any carp group it is a part of. */ 1363 if (ifp->if_carp != NULL && ifp->if_type != IFT_CARP) 1364 carp_ifdetach(ifp); 1365 #endif 1366 1367 /* 1368 * Rip all the addresses off the interface. This should make 1369 * all of the routes go away. 1370 * 1371 * pr_usrreq calls can remove an arbitrary number of ifaddrs 1372 * from the list, including our "cursor", ifa. For safety, 1373 * and to honor the TAILQ abstraction, I just restart the 1374 * loop after each removal. Note that the loop will exit 1375 * when all of the remaining ifaddrs belong to the AF_LINK 1376 * family. I am counting on the historical fact that at 1377 * least one pr_usrreq in each address domain removes at 1378 * least one ifaddr. 1379 */ 1380 again: 1381 /* 1382 * At this point, no other one tries to remove ifa in the list, 1383 * so we don't need to take a lock or psref. Avoid using 1384 * IFADDR_READER_FOREACH to pass over an inspection of contract 1385 * violations of pserialize. 1386 */ 1387 IFADDR_WRITER_FOREACH(ifa, ifp) { 1388 family = ifa->ifa_addr->sa_family; 1389 #ifdef IFAREF_DEBUG 1390 printf("if_detach: ifaddr %p, family %d, refcnt %d\n", 1391 ifa, family, ifa->ifa_refcnt); 1392 if (last_ifa != NULL && ifa == last_ifa) 1393 panic("if_detach: loop detected"); 1394 last_ifa = ifa; 1395 #endif 1396 if (family == AF_LINK) 1397 continue; 1398 dp = pffinddomain(family); 1399 KASSERTMSG(dp != NULL, "no domain for AF %d", family); 1400 /* 1401 * XXX These PURGEIF calls are redundant with the 1402 * purge-all-families calls below, but are left in for 1403 * now both to make a smaller change, and to avoid 1404 * unplanned interactions with clearing of 1405 * ifp->if_addrlist. 1406 */ 1407 purged = 0; 1408 for (pr = dp->dom_protosw; 1409 pr < dp->dom_protoswNPROTOSW; pr++) { 1410 so.so_proto = pr; 1411 if (pr->pr_usrreqs) { 1412 (void) (*pr->pr_usrreqs->pr_purgeif)(&so, ifp); 1413 purged = 1; 1414 } 1415 } 1416 if (purged == 0) { 1417 /* 1418 * XXX What's really the best thing to do 1419 * XXX here? --thorpej (at) NetBSD.org 1420 */ 1421 printf("if_detach: WARNING: AF %d not purged\n", 1422 family); 1423 ifa_remove(ifp, ifa); 1424 } 1425 goto again; 1426 } 1427 1428 if_free_sadl(ifp, 1); 1429 1430 restart: 1431 IFADDR_WRITER_FOREACH(ifa, ifp) { 1432 family = ifa->ifa_addr->sa_family; 1433 KASSERT(family == AF_LINK); 1434 ifa_remove(ifp, ifa); 1435 goto restart; 1436 } 1437 1438 /* Delete stray routes from the routing table. */ 1439 for (i = 0; i <= AF_MAX; i++) 1440 rt_delete_matched_entries(i, if_delroute_matcher, ifp); 1441 1442 DOMAIN_FOREACH(dp) { 1443 if (dp->dom_ifdetach != NULL && ifp->if_afdata[dp->dom_family]) 1444 { 1445 void *p = ifp->if_afdata[dp->dom_family]; 1446 if (p) { 1447 ifp->if_afdata[dp->dom_family] = NULL; 1448 (*dp->dom_ifdetach)(ifp, p); 1449 } 1450 } 1451 1452 /* 1453 * One would expect multicast memberships (INET and 1454 * INET6) on UDP sockets to be purged by the PURGEIF 1455 * calls above, but if all addresses were removed from 1456 * the interface prior to destruction, the calls will 1457 * not be made (e.g. ppp, for which pppd(8) generally 1458 * removes addresses before destroying the interface). 1459 * Because there is no invariant that multicast 1460 * memberships only exist for interfaces with IPv4 1461 * addresses, we must call PURGEIF regardless of 1462 * addresses. (Protocols which might store ifnet 1463 * pointers are marked with PR_PURGEIF.) 1464 */ 1465 for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) { 1466 so.so_proto = pr; 1467 if (pr->pr_usrreqs && pr->pr_flags & PR_PURGEIF) 1468 (void)(*pr->pr_usrreqs->pr_purgeif)(&so, ifp); 1469 } 1470 } 1471 1472 /* 1473 * Must be done after the above pr_purgeif because if_psref may be 1474 * still used in pr_purgeif. 1475 */ 1476 psref_target_destroy(&ifp->if_psref, ifnet_psref_class); 1477 PSLIST_ENTRY_DESTROY(ifp, if_pslist_entry); 1478 1479 pfil_run_ifhooks(if_pfil, PFIL_IFNET_DETACH, ifp); 1480 (void)pfil_head_destroy(ifp->if_pfil); 1481 1482 /* Announce that the interface is gone. */ 1483 rt_ifannouncemsg(ifp, IFAN_DEPARTURE); 1484 1485 IF_AFDATA_LOCK_DESTROY(ifp); 1486 1487 if (if_is_link_state_changeable(ifp)) { 1488 softint_disestablish(ifp->if_link_si); 1489 ifp->if_link_si = NULL; 1490 } 1491 1492 /* 1493 * remove packets that came from ifp, from software interrupt queues. 1494 */ 1495 DOMAIN_FOREACH(dp) { 1496 for (i = 0; i < __arraycount(dp->dom_ifqueues); i++) { 1497 struct ifqueue *iq = dp->dom_ifqueues[i]; 1498 if (iq == NULL) 1499 break; 1500 dp->dom_ifqueues[i] = NULL; 1501 if_detach_queues(ifp, iq); 1502 } 1503 } 1504 1505 /* 1506 * IP queues have to be processed separately: net-queue barrier 1507 * ensures that the packets are dequeued while a cross-call will 1508 * ensure that the interrupts have completed. FIXME: not quite.. 1509 */ 1510 #ifdef INET 1511 pktq_barrier(ip_pktq); 1512 #endif 1513 #ifdef INET6 1514 if (in6_present) 1515 pktq_barrier(ip6_pktq); 1516 #endif 1517 xc = xc_broadcast(0, (xcfunc_t)nullop, NULL, NULL); 1518 xc_wait(xc); 1519 1520 if (ifp->if_percpuq != NULL) { 1521 if_percpuq_destroy(ifp->if_percpuq); 1522 ifp->if_percpuq = NULL; 1523 } 1524 1525 mutex_obj_free(ifp->if_ioctl_lock); 1526 ifp->if_ioctl_lock = NULL; 1527 mutex_obj_free(ifp->if_snd.ifq_lock); 1528 1529 splx(s); 1530 1531 #ifdef IFAREF_DEBUG 1532 if_check_and_free_ifa_list(ifp); 1533 #endif 1534 } 1535 1536 static void 1537 if_detach_queues(struct ifnet *ifp, struct ifqueue *q) 1538 { 1539 struct mbuf *m, *prev, *next; 1540 1541 prev = NULL; 1542 for (m = q->ifq_head; m != NULL; m = next) { 1543 KASSERT((m->m_flags & M_PKTHDR) != 0); 1544 1545 next = m->m_nextpkt; 1546 if (m->m_pkthdr.rcvif_index != ifp->if_index) { 1547 prev = m; 1548 continue; 1549 } 1550 1551 if (prev != NULL) 1552 prev->m_nextpkt = m->m_nextpkt; 1553 else 1554 q->ifq_head = m->m_nextpkt; 1555 if (q->ifq_tail == m) 1556 q->ifq_tail = prev; 1557 q->ifq_len--; 1558 1559 m->m_nextpkt = NULL; 1560 m_freem(m); 1561 IF_DROP(q); 1562 } 1563 } 1564 1565 /* 1566 * Callback for a radix tree walk to delete all references to an 1567 * ifnet. 1568 */ 1569 static int 1570 if_delroute_matcher(struct rtentry *rt, void *v) 1571 { 1572 struct ifnet *ifp = (struct ifnet *)v; 1573 1574 if (rt->rt_ifp == ifp) 1575 return 1; 1576 else 1577 return 0; 1578 } 1579 1580 /* 1581 * Create a clone network interface. 1582 */ 1583 static int 1584 if_clone_create(const char *name) 1585 { 1586 struct if_clone *ifc; 1587 int unit; 1588 struct ifnet *ifp; 1589 struct psref psref; 1590 1591 KASSERT(mutex_owned(&if_clone_mtx)); 1592 1593 ifc = if_clone_lookup(name, &unit); 1594 if (ifc == NULL) 1595 return EINVAL; 1596 1597 ifp = if_get(name, &psref); 1598 if (ifp != NULL) { 1599 if_put(ifp, &psref); 1600 return EEXIST; 1601 } 1602 1603 return (*ifc->ifc_create)(ifc, unit); 1604 } 1605 1606 /* 1607 * Destroy a clone network interface. 1608 */ 1609 static int 1610 if_clone_destroy(const char *name) 1611 { 1612 struct if_clone *ifc; 1613 struct ifnet *ifp; 1614 struct psref psref; 1615 1616 KASSERT(mutex_owned(&if_clone_mtx)); 1617 1618 ifc = if_clone_lookup(name, NULL); 1619 if (ifc == NULL) 1620 return EINVAL; 1621 1622 if (ifc->ifc_destroy == NULL) 1623 return EOPNOTSUPP; 1624 1625 ifp = if_get(name, &psref); 1626 if (ifp == NULL) 1627 return ENXIO; 1628 1629 /* We have to disable ioctls here */ 1630 IFNET_LOCK(ifp); 1631 ifp->if_ioctl = if_nullioctl; 1632 IFNET_UNLOCK(ifp); 1633 1634 /* 1635 * We cannot call ifc_destroy with holding ifp. 1636 * Releasing ifp here is safe thanks to if_clone_mtx. 1637 */ 1638 if_put(ifp, &psref); 1639 1640 return (*ifc->ifc_destroy)(ifp); 1641 } 1642 1643 static bool 1644 if_is_unit(const char *name) 1645 { 1646 1647 while(*name != '\0') { 1648 if (*name < '0' || *name > '9') 1649 return false; 1650 name++; 1651 } 1652 1653 return true; 1654 } 1655 1656 /* 1657 * Look up a network interface cloner. 1658 */ 1659 static struct if_clone * 1660 if_clone_lookup(const char *name, int *unitp) 1661 { 1662 struct if_clone *ifc; 1663 const char *cp; 1664 char *dp, ifname[IFNAMSIZ + 3]; 1665 int unit; 1666 1667 KASSERT(mutex_owned(&if_clone_mtx)); 1668 1669 strcpy(ifname, "if_"); 1670 /* separate interface name from unit */ 1671 /* TODO: search unit number from backward */ 1672 for (dp = ifname + 3, cp = name; cp - name < IFNAMSIZ && 1673 *cp && !if_is_unit(cp);) 1674 *dp++ = *cp++; 1675 1676 if (cp == name || cp - name == IFNAMSIZ || !*cp) 1677 return NULL; /* No name or unit number */ 1678 *dp++ = '\0'; 1679 1680 again: 1681 LIST_FOREACH(ifc, &if_cloners, ifc_list) { 1682 if (strcmp(ifname + 3, ifc->ifc_name) == 0) 1683 break; 1684 } 1685 1686 if (ifc == NULL) { 1687 int error; 1688 if (*ifname == '\0') 1689 return NULL; 1690 mutex_exit(&if_clone_mtx); 1691 error = module_autoload(ifname, MODULE_CLASS_DRIVER); 1692 mutex_enter(&if_clone_mtx); 1693 if (error) 1694 return NULL; 1695 *ifname = '\0'; 1696 goto again; 1697 } 1698 1699 unit = 0; 1700 while (cp - name < IFNAMSIZ && *cp) { 1701 if (*cp < '0' || *cp > '9' || unit >= INT_MAX / 10) { 1702 /* Bogus unit number. */ 1703 return NULL; 1704 } 1705 unit = (unit * 10) + (*cp++ - '0'); 1706 } 1707 1708 if (unitp != NULL) 1709 *unitp = unit; 1710 return ifc; 1711 } 1712 1713 /* 1714 * Register a network interface cloner. 1715 */ 1716 void 1717 if_clone_attach(struct if_clone *ifc) 1718 { 1719 1720 mutex_enter(&if_clone_mtx); 1721 LIST_INSERT_HEAD(&if_cloners, ifc, ifc_list); 1722 if_cloners_count++; 1723 mutex_exit(&if_clone_mtx); 1724 } 1725 1726 /* 1727 * Unregister a network interface cloner. 1728 */ 1729 void 1730 if_clone_detach(struct if_clone *ifc) 1731 { 1732 1733 mutex_enter(&if_clone_mtx); 1734 LIST_REMOVE(ifc, ifc_list); 1735 if_cloners_count--; 1736 mutex_exit(&if_clone_mtx); 1737 } 1738 1739 /* 1740 * Provide list of interface cloners to userspace. 1741 */ 1742 int 1743 if_clone_list(int buf_count, char *buffer, int *total) 1744 { 1745 char outbuf[IFNAMSIZ], *dst; 1746 struct if_clone *ifc; 1747 int count, error = 0; 1748 1749 mutex_enter(&if_clone_mtx); 1750 *total = if_cloners_count; 1751 if ((dst = buffer) == NULL) { 1752 /* Just asking how many there are. */ 1753 goto out; 1754 } 1755 1756 if (buf_count < 0) { 1757 error = EINVAL; 1758 goto out; 1759 } 1760 1761 count = (if_cloners_count < buf_count) ? 1762 if_cloners_count : buf_count; 1763 1764 for (ifc = LIST_FIRST(&if_cloners); ifc != NULL && count != 0; 1765 ifc = LIST_NEXT(ifc, ifc_list), count--, dst += IFNAMSIZ) { 1766 (void)strncpy(outbuf, ifc->ifc_name, sizeof(outbuf)); 1767 if (outbuf[sizeof(outbuf) - 1] != '\0') { 1768 error = ENAMETOOLONG; 1769 goto out; 1770 } 1771 error = copyout(outbuf, dst, sizeof(outbuf)); 1772 if (error != 0) 1773 break; 1774 } 1775 1776 out: 1777 mutex_exit(&if_clone_mtx); 1778 return error; 1779 } 1780 1781 void 1782 ifa_psref_init(struct ifaddr *ifa) 1783 { 1784 1785 psref_target_init(&ifa->ifa_psref, ifa_psref_class); 1786 } 1787 1788 void 1789 ifaref(struct ifaddr *ifa) 1790 { 1791 1792 atomic_inc_uint(&ifa->ifa_refcnt); 1793 } 1794 1795 void 1796 ifafree(struct ifaddr *ifa) 1797 { 1798 KASSERT(ifa != NULL); 1799 KASSERT(ifa->ifa_refcnt > 0); 1800 1801 if (atomic_dec_uint_nv(&ifa->ifa_refcnt) == 0) { 1802 free(ifa, M_IFADDR); 1803 } 1804 } 1805 1806 bool 1807 ifa_is_destroying(struct ifaddr *ifa) 1808 { 1809 1810 return ISSET(ifa->ifa_flags, IFA_DESTROYING); 1811 } 1812 1813 void 1814 ifa_insert(struct ifnet *ifp, struct ifaddr *ifa) 1815 { 1816 1817 ifa->ifa_ifp = ifp; 1818 1819 /* 1820 * Check MP-safety for IFEF_MPSAFE drivers. 1821 * Check !IFF_RUNNING for initialization routines that normally don't 1822 * take IFNET_LOCK but it's safe because there is no competitor. 1823 * XXX there are false positive cases because IFF_RUNNING can be off on 1824 * if_stop. 1825 */ 1826 KASSERT(!if_is_mpsafe(ifp) || !ISSET(ifp->if_flags, IFF_RUNNING) || 1827 IFNET_LOCKED(ifp)); 1828 1829 TAILQ_INSERT_TAIL(&ifp->if_addrlist, ifa, ifa_list); 1830 IFADDR_ENTRY_INIT(ifa); 1831 IFADDR_WRITER_INSERT_TAIL(ifp, ifa); 1832 1833 ifaref(ifa); 1834 } 1835 1836 void 1837 ifa_remove(struct ifnet *ifp, struct ifaddr *ifa) 1838 { 1839 1840 KASSERT(ifa->ifa_ifp == ifp); 1841 /* 1842 * Check MP-safety for IFEF_MPSAFE drivers. 1843 * if_is_deactivated indicates ifa_remove is called form if_detach 1844 * where is safe even if IFNET_LOCK isn't held. 1845 */ 1846 KASSERT(!if_is_mpsafe(ifp) || if_is_deactivated(ifp) || IFNET_LOCKED(ifp)); 1847 1848 TAILQ_REMOVE(&ifp->if_addrlist, ifa, ifa_list); 1849 IFADDR_WRITER_REMOVE(ifa); 1850 #ifdef NET_MPSAFE 1851 IFNET_GLOBAL_LOCK(); 1852 pserialize_perform(ifnet_psz); 1853 IFNET_GLOBAL_UNLOCK(); 1854 #endif 1855 1856 #ifdef NET_MPSAFE 1857 psref_target_destroy(&ifa->ifa_psref, ifa_psref_class); 1858 #endif 1859 IFADDR_ENTRY_DESTROY(ifa); 1860 ifafree(ifa); 1861 } 1862 1863 void 1864 ifa_acquire(struct ifaddr *ifa, struct psref *psref) 1865 { 1866 1867 psref_acquire(psref, &ifa->ifa_psref, ifa_psref_class); 1868 } 1869 1870 void 1871 ifa_release(struct ifaddr *ifa, struct psref *psref) 1872 { 1873 1874 if (ifa == NULL) 1875 return; 1876 1877 psref_release(psref, &ifa->ifa_psref, ifa_psref_class); 1878 } 1879 1880 bool 1881 ifa_held(struct ifaddr *ifa) 1882 { 1883 1884 return psref_held(&ifa->ifa_psref, ifa_psref_class); 1885 } 1886 1887 static inline int 1888 equal(const struct sockaddr *sa1, const struct sockaddr *sa2) 1889 { 1890 return sockaddr_cmp(sa1, sa2) == 0; 1891 } 1892 1893 /* 1894 * Locate an interface based on a complete address. 1895 */ 1896 /*ARGSUSED*/ 1897 struct ifaddr * 1898 ifa_ifwithaddr(const struct sockaddr *addr) 1899 { 1900 struct ifnet *ifp; 1901 struct ifaddr *ifa; 1902 1903 IFNET_READER_FOREACH(ifp) { 1904 if (if_is_deactivated(ifp)) 1905 continue; 1906 IFADDR_READER_FOREACH(ifa, ifp) { 1907 if (ifa->ifa_addr->sa_family != addr->sa_family) 1908 continue; 1909 if (equal(addr, ifa->ifa_addr)) 1910 return ifa; 1911 if ((ifp->if_flags & IFF_BROADCAST) && 1912 ifa->ifa_broadaddr && 1913 /* IP6 doesn't have broadcast */ 1914 ifa->ifa_broadaddr->sa_len != 0 && 1915 equal(ifa->ifa_broadaddr, addr)) 1916 return ifa; 1917 } 1918 } 1919 return NULL; 1920 } 1921 1922 struct ifaddr * 1923 ifa_ifwithaddr_psref(const struct sockaddr *addr, struct psref *psref) 1924 { 1925 struct ifaddr *ifa; 1926 int s = pserialize_read_enter(); 1927 1928 ifa = ifa_ifwithaddr(addr); 1929 if (ifa != NULL) 1930 ifa_acquire(ifa, psref); 1931 pserialize_read_exit(s); 1932 1933 return ifa; 1934 } 1935 1936 /* 1937 * Locate the point to point interface with a given destination address. 1938 */ 1939 /*ARGSUSED*/ 1940 struct ifaddr * 1941 ifa_ifwithdstaddr(const struct sockaddr *addr) 1942 { 1943 struct ifnet *ifp; 1944 struct ifaddr *ifa; 1945 1946 IFNET_READER_FOREACH(ifp) { 1947 if (if_is_deactivated(ifp)) 1948 continue; 1949 if ((ifp->if_flags & IFF_POINTOPOINT) == 0) 1950 continue; 1951 IFADDR_READER_FOREACH(ifa, ifp) { 1952 if (ifa->ifa_addr->sa_family != addr->sa_family || 1953 ifa->ifa_dstaddr == NULL) 1954 continue; 1955 if (equal(addr, ifa->ifa_dstaddr)) 1956 return ifa; 1957 } 1958 } 1959 1960 return NULL; 1961 } 1962 1963 struct ifaddr * 1964 ifa_ifwithdstaddr_psref(const struct sockaddr *addr, struct psref *psref) 1965 { 1966 struct ifaddr *ifa; 1967 int s; 1968 1969 s = pserialize_read_enter(); 1970 ifa = ifa_ifwithdstaddr(addr); 1971 if (ifa != NULL) 1972 ifa_acquire(ifa, psref); 1973 pserialize_read_exit(s); 1974 1975 return ifa; 1976 } 1977 1978 /* 1979 * Find an interface on a specific network. If many, choice 1980 * is most specific found. 1981 */ 1982 struct ifaddr * 1983 ifa_ifwithnet(const struct sockaddr *addr) 1984 { 1985 struct ifnet *ifp; 1986 struct ifaddr *ifa, *ifa_maybe = NULL; 1987 const struct sockaddr_dl *sdl; 1988 u_int af = addr->sa_family; 1989 const char *addr_data = addr->sa_data, *cplim; 1990 1991 if (af == AF_LINK) { 1992 sdl = satocsdl(addr); 1993 if (sdl->sdl_index && sdl->sdl_index < if_indexlim && 1994 ifindex2ifnet[sdl->sdl_index] && 1995 !if_is_deactivated(ifindex2ifnet[sdl->sdl_index])) { 1996 return ifindex2ifnet[sdl->sdl_index]->if_dl; 1997 } 1998 } 1999 #ifdef NETATALK 2000 if (af == AF_APPLETALK) { 2001 const struct sockaddr_at *sat, *sat2; 2002 sat = (const struct sockaddr_at *)addr; 2003 IFNET_READER_FOREACH(ifp) { 2004 if (if_is_deactivated(ifp)) 2005 continue; 2006 ifa = at_ifawithnet((const struct sockaddr_at *)addr, ifp); 2007 if (ifa == NULL) 2008 continue; 2009 sat2 = (struct sockaddr_at *)ifa->ifa_addr; 2010 if (sat2->sat_addr.s_net == sat->sat_addr.s_net) 2011 return ifa; /* exact match */ 2012 if (ifa_maybe == NULL) { 2013 /* else keep the if with the right range */ 2014 ifa_maybe = ifa; 2015 } 2016 } 2017 return ifa_maybe; 2018 } 2019 #endif 2020 IFNET_READER_FOREACH(ifp) { 2021 if (if_is_deactivated(ifp)) 2022 continue; 2023 IFADDR_READER_FOREACH(ifa, ifp) { 2024 const char *cp, *cp2, *cp3; 2025 2026 if (ifa->ifa_addr->sa_family != af || 2027 ifa->ifa_netmask == NULL) 2028 next: continue; 2029 cp = addr_data; 2030 cp2 = ifa->ifa_addr->sa_data; 2031 cp3 = ifa->ifa_netmask->sa_data; 2032 cplim = (const char *)ifa->ifa_netmask + 2033 ifa->ifa_netmask->sa_len; 2034 while (cp3 < cplim) { 2035 if ((*cp++ ^ *cp2++) & *cp3++) { 2036 /* want to continue for() loop */ 2037 goto next; 2038 } 2039 } 2040 if (ifa_maybe == NULL || 2041 rt_refines(ifa->ifa_netmask, 2042 ifa_maybe->ifa_netmask)) 2043 ifa_maybe = ifa; 2044 } 2045 } 2046 return ifa_maybe; 2047 } 2048 2049 struct ifaddr * 2050 ifa_ifwithnet_psref(const struct sockaddr *addr, struct psref *psref) 2051 { 2052 struct ifaddr *ifa; 2053 int s; 2054 2055 s = pserialize_read_enter(); 2056 ifa = ifa_ifwithnet(addr); 2057 if (ifa != NULL) 2058 ifa_acquire(ifa, psref); 2059 pserialize_read_exit(s); 2060 2061 return ifa; 2062 } 2063 2064 /* 2065 * Find the interface of the addresss. 2066 */ 2067 struct ifaddr * 2068 ifa_ifwithladdr(const struct sockaddr *addr) 2069 { 2070 struct ifaddr *ia; 2071 2072 if ((ia = ifa_ifwithaddr(addr)) || (ia = ifa_ifwithdstaddr(addr)) || 2073 (ia = ifa_ifwithnet(addr))) 2074 return ia; 2075 return NULL; 2076 } 2077 2078 struct ifaddr * 2079 ifa_ifwithladdr_psref(const struct sockaddr *addr, struct psref *psref) 2080 { 2081 struct ifaddr *ifa; 2082 int s; 2083 2084 s = pserialize_read_enter(); 2085 ifa = ifa_ifwithladdr(addr); 2086 if (ifa != NULL) 2087 ifa_acquire(ifa, psref); 2088 pserialize_read_exit(s); 2089 2090 return ifa; 2091 } 2092 2093 /* 2094 * Find an interface using a specific address family 2095 */ 2096 struct ifaddr * 2097 ifa_ifwithaf(int af) 2098 { 2099 struct ifnet *ifp; 2100 struct ifaddr *ifa = NULL; 2101 int s; 2102 2103 s = pserialize_read_enter(); 2104 IFNET_READER_FOREACH(ifp) { 2105 if (if_is_deactivated(ifp)) 2106 continue; 2107 IFADDR_READER_FOREACH(ifa, ifp) { 2108 if (ifa->ifa_addr->sa_family == af) 2109 goto out; 2110 } 2111 } 2112 out: 2113 pserialize_read_exit(s); 2114 return ifa; 2115 } 2116 2117 /* 2118 * Find an interface address specific to an interface best matching 2119 * a given address. 2120 */ 2121 struct ifaddr * 2122 ifaof_ifpforaddr(const struct sockaddr *addr, struct ifnet *ifp) 2123 { 2124 struct ifaddr *ifa; 2125 const char *cp, *cp2, *cp3; 2126 const char *cplim; 2127 struct ifaddr *ifa_maybe = 0; 2128 u_int af = addr->sa_family; 2129 2130 if (if_is_deactivated(ifp)) 2131 return NULL; 2132 2133 if (af >= AF_MAX) 2134 return NULL; 2135 2136 IFADDR_READER_FOREACH(ifa, ifp) { 2137 if (ifa->ifa_addr->sa_family != af) 2138 continue; 2139 ifa_maybe = ifa; 2140 if (ifa->ifa_netmask == NULL) { 2141 if (equal(addr, ifa->ifa_addr) || 2142 (ifa->ifa_dstaddr && 2143 equal(addr, ifa->ifa_dstaddr))) 2144 return ifa; 2145 continue; 2146 } 2147 cp = addr->sa_data; 2148 cp2 = ifa->ifa_addr->sa_data; 2149 cp3 = ifa->ifa_netmask->sa_data; 2150 cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask; 2151 for (; cp3 < cplim; cp3++) { 2152 if ((*cp++ ^ *cp2++) & *cp3) 2153 break; 2154 } 2155 if (cp3 == cplim) 2156 return ifa; 2157 } 2158 return ifa_maybe; 2159 } 2160 2161 struct ifaddr * 2162 ifaof_ifpforaddr_psref(const struct sockaddr *addr, struct ifnet *ifp, 2163 struct psref *psref) 2164 { 2165 struct ifaddr *ifa; 2166 int s; 2167 2168 s = pserialize_read_enter(); 2169 ifa = ifaof_ifpforaddr(addr, ifp); 2170 if (ifa != NULL) 2171 ifa_acquire(ifa, psref); 2172 pserialize_read_exit(s); 2173 2174 return ifa; 2175 } 2176 2177 /* 2178 * Default action when installing a route with a Link Level gateway. 2179 * Lookup an appropriate real ifa to point to. 2180 * This should be moved to /sys/net/link.c eventually. 2181 */ 2182 void 2183 link_rtrequest(int cmd, struct rtentry *rt, const struct rt_addrinfo *info) 2184 { 2185 struct ifaddr *ifa; 2186 const struct sockaddr *dst; 2187 struct ifnet *ifp; 2188 struct psref psref; 2189 2190 if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL || 2191 (ifp = ifa->ifa_ifp) == NULL || (dst = rt_getkey(rt)) == NULL || 2192 ISSET(info->rti_flags, RTF_DONTCHANGEIFA)) 2193 return; 2194 if ((ifa = ifaof_ifpforaddr_psref(dst, ifp, &psref)) != NULL) { 2195 rt_replace_ifa(rt, ifa); 2196 if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest) 2197 ifa->ifa_rtrequest(cmd, rt, info); 2198 ifa_release(ifa, &psref); 2199 } 2200 } 2201 2202 /* 2203 * bitmask macros to manage a densely packed link_state change queue. 2204 * Because we need to store LINK_STATE_UNKNOWN(0), LINK_STATE_DOWN(1) and 2205 * LINK_STATE_UP(2) we need 2 bits for each state change. 2206 * As a state change to store is 0, treat all bits set as an unset item. 2207 */ 2208 #define LQ_ITEM_BITS 2 2209 #define LQ_ITEM_MASK ((1 << LQ_ITEM_BITS) - 1) 2210 #define LQ_MASK(i) (LQ_ITEM_MASK << (i) * LQ_ITEM_BITS) 2211 #define LINK_STATE_UNSET LQ_ITEM_MASK 2212 #define LQ_ITEM(q, i) (((q) & LQ_MASK((i))) >> (i) * LQ_ITEM_BITS) 2213 #define LQ_STORE(q, i, v) \ 2214 do { \ 2215 (q) &= ~LQ_MASK((i)); \ 2216 (q) |= (v) << (i) * LQ_ITEM_BITS; \ 2217 } while (0 /* CONSTCOND */) 2218 #define LQ_MAX(q) ((sizeof((q)) * NBBY) / LQ_ITEM_BITS) 2219 #define LQ_POP(q, v) \ 2220 do { \ 2221 (v) = LQ_ITEM((q), 0); \ 2222 (q) >>= LQ_ITEM_BITS; \ 2223 (q) |= LINK_STATE_UNSET << (LQ_MAX((q)) - 1) * LQ_ITEM_BITS; \ 2224 } while (0 /* CONSTCOND */) 2225 #define LQ_PUSH(q, v) \ 2226 do { \ 2227 (q) >>= LQ_ITEM_BITS; \ 2228 (q) |= (v) << (LQ_MAX((q)) - 1) * LQ_ITEM_BITS; \ 2229 } while (0 /* CONSTCOND */) 2230 #define LQ_FIND_UNSET(q, i) \ 2231 for ((i) = 0; i < LQ_MAX((q)); (i)++) { \ 2232 if (LQ_ITEM((q), (i)) == LINK_STATE_UNSET) \ 2233 break; \ 2234 } 2235 2236 /* 2237 * XXX reusing (ifp)->if_snd->ifq_lock rather than having another spin mutex 2238 * for each ifnet. It doesn't matter because: 2239 * - if IFEF_MPSAFE is enabled, if_snd isn't used and lock contentions on 2240 * ifq_lock don't happen 2241 * - if IFEF_MPSAFE is disabled, there is no lock contention on ifq_lock 2242 * because if_snd, if_link_state_change and if_link_state_change_softint 2243 * are all called with KERNEL_LOCK 2244 */ 2245 #define IF_LINK_STATE_CHANGE_LOCK(ifp) \ 2246 mutex_enter((ifp)->if_snd.ifq_lock) 2247 #define IF_LINK_STATE_CHANGE_UNLOCK(ifp) \ 2248 mutex_exit((ifp)->if_snd.ifq_lock) 2249 2250 /* 2251 * Handle a change in the interface link state and 2252 * queue notifications. 2253 */ 2254 void 2255 if_link_state_change(struct ifnet *ifp, int link_state) 2256 { 2257 int idx; 2258 2259 KASSERTMSG(if_is_link_state_changeable(ifp), 2260 "%s: IFEF_NO_LINK_STATE_CHANGE must not be set, but if_extflags=0x%x", 2261 ifp->if_xname, ifp->if_extflags); 2262 2263 /* Ensure change is to a valid state */ 2264 switch (link_state) { 2265 case LINK_STATE_UNKNOWN: /* FALLTHROUGH */ 2266 case LINK_STATE_DOWN: /* FALLTHROUGH */ 2267 case LINK_STATE_UP: 2268 break; 2269 default: 2270 #ifdef DEBUG 2271 printf("%s: invalid link state %d\n", 2272 ifp->if_xname, link_state); 2273 #endif 2274 return; 2275 } 2276 2277 IF_LINK_STATE_CHANGE_LOCK(ifp); 2278 2279 /* Find the last unset event in the queue. */ 2280 LQ_FIND_UNSET(ifp->if_link_queue, idx); 2281 2282 /* 2283 * Ensure link_state doesn't match the last event in the queue. 2284 * ifp->if_link_state is not checked and set here because 2285 * that would present an inconsistent picture to the system. 2286 */ 2287 if (idx != 0 && 2288 LQ_ITEM(ifp->if_link_queue, idx - 1) == (uint8_t)link_state) 2289 goto out; 2290 2291 /* Handle queue overflow. */ 2292 if (idx == LQ_MAX(ifp->if_link_queue)) { 2293 uint8_t lost; 2294 2295 /* 2296 * The DOWN state must be protected from being pushed off 2297 * the queue to ensure that userland will always be 2298 * in a sane state. 2299 * Because DOWN is protected, there is no need to protect 2300 * UNKNOWN. 2301 * It should be invalid to change from any other state to 2302 * UNKNOWN anyway ... 2303 */ 2304 lost = LQ_ITEM(ifp->if_link_queue, 0); 2305 LQ_PUSH(ifp->if_link_queue, (uint8_t)link_state); 2306 if (lost == LINK_STATE_DOWN) { 2307 lost = LQ_ITEM(ifp->if_link_queue, 0); 2308 LQ_STORE(ifp->if_link_queue, 0, LINK_STATE_DOWN); 2309 } 2310 printf("%s: lost link state change %s\n", 2311 ifp->if_xname, 2312 lost == LINK_STATE_UP ? "UP" : 2313 lost == LINK_STATE_DOWN ? "DOWN" : 2314 "UNKNOWN"); 2315 } else 2316 LQ_STORE(ifp->if_link_queue, idx, (uint8_t)link_state); 2317 2318 softint_schedule(ifp->if_link_si); 2319 2320 out: 2321 IF_LINK_STATE_CHANGE_UNLOCK(ifp); 2322 } 2323 2324 /* 2325 * Handle interface link state change notifications. 2326 */ 2327 void 2328 if_link_state_change_softint(struct ifnet *ifp, int link_state) 2329 { 2330 struct domain *dp; 2331 int s = splnet(); 2332 bool notify; 2333 2334 KASSERT(!cpu_intr_p()); 2335 2336 IF_LINK_STATE_CHANGE_LOCK(ifp); 2337 2338 /* Ensure the change is still valid. */ 2339 if (ifp->if_link_state == link_state) { 2340 IF_LINK_STATE_CHANGE_UNLOCK(ifp); 2341 splx(s); 2342 return; 2343 } 2344 2345 #ifdef DEBUG 2346 log(LOG_DEBUG, "%s: link state %s (was %s)\n", ifp->if_xname, 2347 link_state == LINK_STATE_UP ? "UP" : 2348 link_state == LINK_STATE_DOWN ? "DOWN" : 2349 "UNKNOWN", 2350 ifp->if_link_state == LINK_STATE_UP ? "UP" : 2351 ifp->if_link_state == LINK_STATE_DOWN ? "DOWN" : 2352 "UNKNOWN"); 2353 #endif 2354 2355 /* 2356 * When going from UNKNOWN to UP, we need to mark existing 2357 * addresses as tentative and restart DAD as we may have 2358 * erroneously not found a duplicate. 2359 * 2360 * This needs to happen before rt_ifmsg to avoid a race where 2361 * listeners would have an address and expect it to work right 2362 * away. 2363 */ 2364 notify = (link_state == LINK_STATE_UP && 2365 ifp->if_link_state == LINK_STATE_UNKNOWN); 2366 ifp->if_link_state = link_state; 2367 /* The following routines may sleep so release the spin mutex */ 2368 IF_LINK_STATE_CHANGE_UNLOCK(ifp); 2369 2370 KERNEL_LOCK_UNLESS_NET_MPSAFE(); 2371 if (notify) { 2372 DOMAIN_FOREACH(dp) { 2373 if (dp->dom_if_link_state_change != NULL) 2374 dp->dom_if_link_state_change(ifp, 2375 LINK_STATE_DOWN); 2376 } 2377 } 2378 2379 /* Notify that the link state has changed. */ 2380 rt_ifmsg(ifp); 2381 2382 #if NCARP > 0 2383 if (ifp->if_carp) 2384 carp_carpdev_state(ifp); 2385 #endif 2386 2387 DOMAIN_FOREACH(dp) { 2388 if (dp->dom_if_link_state_change != NULL) 2389 dp->dom_if_link_state_change(ifp, link_state); 2390 } 2391 KERNEL_UNLOCK_UNLESS_NET_MPSAFE(); 2392 splx(s); 2393 } 2394 2395 /* 2396 * Process the interface link state change queue. 2397 */ 2398 static void 2399 if_link_state_change_si(void *arg) 2400 { 2401 struct ifnet *ifp = arg; 2402 int s; 2403 uint8_t state; 2404 bool schedule; 2405 2406 SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE(); 2407 s = splnet(); 2408 2409 /* Pop a link state change from the queue and process it. */ 2410 IF_LINK_STATE_CHANGE_LOCK(ifp); 2411 LQ_POP(ifp->if_link_queue, state); 2412 IF_LINK_STATE_CHANGE_UNLOCK(ifp); 2413 2414 if_link_state_change_softint(ifp, state); 2415 2416 /* If there is a link state change to come, schedule it. */ 2417 IF_LINK_STATE_CHANGE_LOCK(ifp); 2418 schedule = (LQ_ITEM(ifp->if_link_queue, 0) != LINK_STATE_UNSET); 2419 IF_LINK_STATE_CHANGE_UNLOCK(ifp); 2420 if (schedule) 2421 softint_schedule(ifp->if_link_si); 2422 2423 splx(s); 2424 SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE(); 2425 } 2426 2427 /* 2428 * Default action when installing a local route on a point-to-point 2429 * interface. 2430 */ 2431 void 2432 p2p_rtrequest(int req, struct rtentry *rt, 2433 __unused const struct rt_addrinfo *info) 2434 { 2435 struct ifnet *ifp = rt->rt_ifp; 2436 struct ifaddr *ifa, *lo0ifa; 2437 int s = pserialize_read_enter(); 2438 2439 switch (req) { 2440 case RTM_ADD: 2441 if ((rt->rt_flags & RTF_LOCAL) == 0) 2442 break; 2443 2444 rt->rt_ifp = lo0ifp; 2445 2446 if (ISSET(info->rti_flags, RTF_DONTCHANGEIFA)) 2447 break; 2448 2449 IFADDR_READER_FOREACH(ifa, ifp) { 2450 if (equal(rt_getkey(rt), ifa->ifa_addr)) 2451 break; 2452 } 2453 if (ifa == NULL) 2454 break; 2455 2456 /* 2457 * Ensure lo0 has an address of the same family. 2458 */ 2459 IFADDR_READER_FOREACH(lo0ifa, lo0ifp) { 2460 if (lo0ifa->ifa_addr->sa_family == 2461 ifa->ifa_addr->sa_family) 2462 break; 2463 } 2464 if (lo0ifa == NULL) 2465 break; 2466 2467 /* 2468 * Make sure to set rt->rt_ifa to the interface 2469 * address we are using, otherwise we will have trouble 2470 * with source address selection. 2471 */ 2472 if (ifa != rt->rt_ifa) 2473 rt_replace_ifa(rt, ifa); 2474 break; 2475 case RTM_DELETE: 2476 default: 2477 break; 2478 } 2479 pserialize_read_exit(s); 2480 } 2481 2482 static void 2483 _if_down(struct ifnet *ifp) 2484 { 2485 struct ifaddr *ifa; 2486 struct domain *dp; 2487 int s, bound; 2488 struct psref psref; 2489 2490 ifp->if_flags &= ~IFF_UP; 2491 nanotime(&ifp->if_lastchange); 2492 2493 bound = curlwp_bind(); 2494 s = pserialize_read_enter(); 2495 IFADDR_READER_FOREACH(ifa, ifp) { 2496 ifa_acquire(ifa, &psref); 2497 pserialize_read_exit(s); 2498 2499 pfctlinput(PRC_IFDOWN, ifa->ifa_addr); 2500 2501 s = pserialize_read_enter(); 2502 ifa_release(ifa, &psref); 2503 } 2504 pserialize_read_exit(s); 2505 curlwp_bindx(bound); 2506 2507 IFQ_PURGE(&ifp->if_snd); 2508 #if NCARP > 0 2509 if (ifp->if_carp) 2510 carp_carpdev_state(ifp); 2511 #endif 2512 rt_ifmsg(ifp); 2513 DOMAIN_FOREACH(dp) { 2514 if (dp->dom_if_down) 2515 dp->dom_if_down(ifp); 2516 } 2517 } 2518 2519 static void 2520 if_down_deactivated(struct ifnet *ifp) 2521 { 2522 2523 KASSERT(if_is_deactivated(ifp)); 2524 _if_down(ifp); 2525 } 2526 2527 void 2528 if_down_locked(struct ifnet *ifp) 2529 { 2530 2531 KASSERT(IFNET_LOCKED(ifp)); 2532 _if_down(ifp); 2533 } 2534 2535 /* 2536 * Mark an interface down and notify protocols of 2537 * the transition. 2538 * NOTE: must be called at splsoftnet or equivalent. 2539 */ 2540 void 2541 if_down(struct ifnet *ifp) 2542 { 2543 2544 IFNET_LOCK(ifp); 2545 if_down_locked(ifp); 2546 IFNET_UNLOCK(ifp); 2547 } 2548 2549 /* 2550 * Must be called with holding if_ioctl_lock. 2551 */ 2552 static void 2553 if_up_locked(struct ifnet *ifp) 2554 { 2555 #ifdef notyet 2556 struct ifaddr *ifa; 2557 #endif 2558 struct domain *dp; 2559 2560 KASSERT(IFNET_LOCKED(ifp)); 2561 2562 KASSERT(!if_is_deactivated(ifp)); 2563 ifp->if_flags |= IFF_UP; 2564 nanotime(&ifp->if_lastchange); 2565 #ifdef notyet 2566 /* this has no effect on IP, and will kill all ISO connections XXX */ 2567 IFADDR_READER_FOREACH(ifa, ifp) 2568 pfctlinput(PRC_IFUP, ifa->ifa_addr); 2569 #endif 2570 #if NCARP > 0 2571 if (ifp->if_carp) 2572 carp_carpdev_state(ifp); 2573 #endif 2574 rt_ifmsg(ifp); 2575 DOMAIN_FOREACH(dp) { 2576 if (dp->dom_if_up) 2577 dp->dom_if_up(ifp); 2578 } 2579 } 2580 2581 /* 2582 * Handle interface slowtimo timer routine. Called 2583 * from softclock, we decrement timer (if set) and 2584 * call the appropriate interface routine on expiration. 2585 */ 2586 static void 2587 if_slowtimo(void *arg) 2588 { 2589 void (*slowtimo)(struct ifnet *); 2590 struct ifnet *ifp = arg; 2591 int s; 2592 2593 slowtimo = ifp->if_slowtimo; 2594 if (__predict_false(slowtimo == NULL)) 2595 return; 2596 2597 s = splnet(); 2598 if (ifp->if_timer != 0 && --ifp->if_timer == 0) 2599 (*slowtimo)(ifp); 2600 2601 splx(s); 2602 2603 if (__predict_true(ifp->if_slowtimo != NULL)) 2604 callout_schedule(ifp->if_slowtimo_ch, hz / IFNET_SLOWHZ); 2605 } 2606 2607 /* 2608 * Mark an interface up and notify protocols of 2609 * the transition. 2610 * NOTE: must be called at splsoftnet or equivalent. 2611 */ 2612 void 2613 if_up(struct ifnet *ifp) 2614 { 2615 2616 IFNET_LOCK(ifp); 2617 if_up_locked(ifp); 2618 IFNET_UNLOCK(ifp); 2619 } 2620 2621 /* 2622 * Set/clear promiscuous mode on interface ifp based on the truth value 2623 * of pswitch. The calls are reference counted so that only the first 2624 * "on" request actually has an effect, as does the final "off" request. 2625 * Results are undefined if the "off" and "on" requests are not matched. 2626 */ 2627 int 2628 ifpromisc_locked(struct ifnet *ifp, int pswitch) 2629 { 2630 int pcount, ret = 0; 2631 short nflags; 2632 2633 KASSERT(IFNET_LOCKED(ifp)); 2634 2635 pcount = ifp->if_pcount; 2636 if (pswitch) { 2637 /* 2638 * Allow the device to be "placed" into promiscuous 2639 * mode even if it is not configured up. It will 2640 * consult IFF_PROMISC when it is brought up. 2641 */ 2642 if (ifp->if_pcount++ != 0) 2643 goto out; 2644 nflags = ifp->if_flags | IFF_PROMISC; 2645 } else { 2646 if (--ifp->if_pcount > 0) 2647 goto out; 2648 nflags = ifp->if_flags & ~IFF_PROMISC; 2649 } 2650 ret = if_flags_set(ifp, nflags); 2651 /* Restore interface state if not successful. */ 2652 if (ret != 0) { 2653 ifp->if_pcount = pcount; 2654 } 2655 out: 2656 return ret; 2657 } 2658 2659 int 2660 ifpromisc(struct ifnet *ifp, int pswitch) 2661 { 2662 int e; 2663 2664 IFNET_LOCK(ifp); 2665 e = ifpromisc_locked(ifp, pswitch); 2666 IFNET_UNLOCK(ifp); 2667 2668 return e; 2669 } 2670 2671 /* 2672 * Map interface name to 2673 * interface structure pointer. 2674 */ 2675 struct ifnet * 2676 ifunit(const char *name) 2677 { 2678 struct ifnet *ifp; 2679 const char *cp = name; 2680 u_int unit = 0; 2681 u_int i; 2682 int s; 2683 2684 /* 2685 * If the entire name is a number, treat it as an ifindex. 2686 */ 2687 for (i = 0; i < IFNAMSIZ && *cp >= '0' && *cp <= '9'; i++, cp++) { 2688 unit = unit * 10 + (*cp - '0'); 2689 } 2690 2691 /* 2692 * If the number took all of the name, then it's a valid ifindex. 2693 */ 2694 if (i == IFNAMSIZ || (cp != name && *cp == '\0')) 2695 return if_byindex(unit); 2696 2697 ifp = NULL; 2698 s = pserialize_read_enter(); 2699 IFNET_READER_FOREACH(ifp) { 2700 if (if_is_deactivated(ifp)) 2701 continue; 2702 if (strcmp(ifp->if_xname, name) == 0) 2703 goto out; 2704 } 2705 out: 2706 pserialize_read_exit(s); 2707 return ifp; 2708 } 2709 2710 /* 2711 * Get a reference of an ifnet object by an interface name. 2712 * The returned reference is protected by psref(9). The caller 2713 * must release a returned reference by if_put after use. 2714 */ 2715 struct ifnet * 2716 if_get(const char *name, struct psref *psref) 2717 { 2718 struct ifnet *ifp; 2719 const char *cp = name; 2720 u_int unit = 0; 2721 u_int i; 2722 int s; 2723 2724 /* 2725 * If the entire name is a number, treat it as an ifindex. 2726 */ 2727 for (i = 0; i < IFNAMSIZ && *cp >= '0' && *cp <= '9'; i++, cp++) { 2728 unit = unit * 10 + (*cp - '0'); 2729 } 2730 2731 /* 2732 * If the number took all of the name, then it's a valid ifindex. 2733 */ 2734 if (i == IFNAMSIZ || (cp != name && *cp == '\0')) 2735 return if_get_byindex(unit, psref); 2736 2737 ifp = NULL; 2738 s = pserialize_read_enter(); 2739 IFNET_READER_FOREACH(ifp) { 2740 if (if_is_deactivated(ifp)) 2741 continue; 2742 if (strcmp(ifp->if_xname, name) == 0) { 2743 psref_acquire(psref, &ifp->if_psref, 2744 ifnet_psref_class); 2745 goto out; 2746 } 2747 } 2748 out: 2749 pserialize_read_exit(s); 2750 return ifp; 2751 } 2752 2753 /* 2754 * Release a reference of an ifnet object given by if_get, if_get_byindex 2755 * or if_get_bylla. 2756 */ 2757 void 2758 if_put(const struct ifnet *ifp, struct psref *psref) 2759 { 2760 2761 if (ifp == NULL) 2762 return; 2763 2764 psref_release(psref, &ifp->if_psref, ifnet_psref_class); 2765 } 2766 2767 /* 2768 * Return ifp having idx. Return NULL if not found. Normally if_byindex 2769 * should be used. 2770 */ 2771 ifnet_t * 2772 _if_byindex(u_int idx) 2773 { 2774 2775 return (__predict_true(idx < if_indexlim)) ? ifindex2ifnet[idx] : NULL; 2776 } 2777 2778 /* 2779 * Return ifp having idx. Return NULL if not found or the found ifp is 2780 * already deactivated. 2781 */ 2782 ifnet_t * 2783 if_byindex(u_int idx) 2784 { 2785 ifnet_t *ifp; 2786 2787 ifp = _if_byindex(idx); 2788 if (ifp != NULL && if_is_deactivated(ifp)) 2789 ifp = NULL; 2790 return ifp; 2791 } 2792 2793 /* 2794 * Get a reference of an ifnet object by an interface index. 2795 * The returned reference is protected by psref(9). The caller 2796 * must release a returned reference by if_put after use. 2797 */ 2798 ifnet_t * 2799 if_get_byindex(u_int idx, struct psref *psref) 2800 { 2801 ifnet_t *ifp; 2802 int s; 2803 2804 s = pserialize_read_enter(); 2805 ifp = if_byindex(idx); 2806 if (__predict_true(ifp != NULL)) 2807 psref_acquire(psref, &ifp->if_psref, ifnet_psref_class); 2808 pserialize_read_exit(s); 2809 2810 return ifp; 2811 } 2812 2813 ifnet_t * 2814 if_get_bylla(const void *lla, unsigned char lla_len, struct psref *psref) 2815 { 2816 ifnet_t *ifp; 2817 int s; 2818 2819 s = pserialize_read_enter(); 2820 IFNET_READER_FOREACH(ifp) { 2821 if (if_is_deactivated(ifp)) 2822 continue; 2823 if (ifp->if_addrlen != lla_len) 2824 continue; 2825 if (memcmp(lla, CLLADDR(ifp->if_sadl), lla_len) == 0) { 2826 psref_acquire(psref, &ifp->if_psref, 2827 ifnet_psref_class); 2828 break; 2829 } 2830 } 2831 pserialize_read_exit(s); 2832 2833 return ifp; 2834 } 2835 2836 /* 2837 * Note that it's safe only if the passed ifp is guaranteed to not be freed, 2838 * for example using pserialize or the ifp is already held or some other 2839 * object is held which guarantes the ifp to not be freed indirectly. 2840 */ 2841 void 2842 if_acquire(struct ifnet *ifp, struct psref *psref) 2843 { 2844 2845 KASSERT(ifp->if_index != 0); 2846 psref_acquire(psref, &ifp->if_psref, ifnet_psref_class); 2847 } 2848 2849 bool 2850 if_held(struct ifnet *ifp) 2851 { 2852 2853 return psref_held(&ifp->if_psref, ifnet_psref_class); 2854 } 2855 2856 /* 2857 * Some tunnel interfaces can nest, e.g. IPv4 over IPv4 gif(4) tunnel over IPv4. 2858 * Check the tunnel nesting count. 2859 * Return > 0, if tunnel nesting count is more than limit. 2860 * Return 0, if tunnel nesting count is equal or less than limit. 2861 */ 2862 int 2863 if_tunnel_check_nesting(struct ifnet *ifp, struct mbuf *m, int limit) 2864 { 2865 struct m_tag *mtag; 2866 int *count; 2867 2868 mtag = m_tag_find(m, PACKET_TAG_TUNNEL_INFO); 2869 if (mtag != NULL) { 2870 count = (int *)(mtag + 1); 2871 if (++(*count) > limit) { 2872 log(LOG_NOTICE, 2873 "%s: recursively called too many times(%d)\n", 2874 ifp->if_xname, *count); 2875 return EIO; 2876 } 2877 } else { 2878 mtag = m_tag_get(PACKET_TAG_TUNNEL_INFO, sizeof(*count), 2879 M_NOWAIT); 2880 if (mtag != NULL) { 2881 m_tag_prepend(m, mtag); 2882 count = (int *)(mtag + 1); 2883 *count = 0; 2884 } else { 2885 log(LOG_DEBUG, 2886 "%s: m_tag_get() failed, recursion calls are not prevented.\n", 2887 ifp->if_xname); 2888 } 2889 } 2890 2891 return 0; 2892 } 2893 2894 /* common */ 2895 int 2896 ifioctl_common(struct ifnet *ifp, u_long cmd, void *data) 2897 { 2898 int s; 2899 struct ifreq *ifr; 2900 struct ifcapreq *ifcr; 2901 struct ifdatareq *ifdr; 2902 2903 switch (cmd) { 2904 case SIOCSIFCAP: 2905 ifcr = data; 2906 if ((ifcr->ifcr_capenable & ~ifp->if_capabilities) != 0) 2907 return EINVAL; 2908 2909 if (ifcr->ifcr_capenable == ifp->if_capenable) 2910 return 0; 2911 2912 ifp->if_capenable = ifcr->ifcr_capenable; 2913 2914 /* Pre-compute the checksum flags mask. */ 2915 ifp->if_csum_flags_tx = 0; 2916 ifp->if_csum_flags_rx = 0; 2917 if (ifp->if_capenable & IFCAP_CSUM_IPv4_Tx) 2918 ifp->if_csum_flags_tx |= M_CSUM_IPv4; 2919 if (ifp->if_capenable & IFCAP_CSUM_IPv4_Rx) 2920 ifp->if_csum_flags_rx |= M_CSUM_IPv4; 2921 2922 if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Tx) 2923 ifp->if_csum_flags_tx |= M_CSUM_TCPv4; 2924 if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Rx) 2925 ifp->if_csum_flags_rx |= M_CSUM_TCPv4; 2926 2927 if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Tx) 2928 ifp->if_csum_flags_tx |= M_CSUM_UDPv4; 2929 if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Rx) 2930 ifp->if_csum_flags_rx |= M_CSUM_UDPv4; 2931 2932 if (ifp->if_capenable & IFCAP_CSUM_TCPv6_Tx) 2933 ifp->if_csum_flags_tx |= M_CSUM_TCPv6; 2934 if (ifp->if_capenable & IFCAP_CSUM_TCPv6_Rx) 2935 ifp->if_csum_flags_rx |= M_CSUM_TCPv6; 2936 2937 if (ifp->if_capenable & IFCAP_CSUM_UDPv6_Tx) 2938 ifp->if_csum_flags_tx |= M_CSUM_UDPv6; 2939 if (ifp->if_capenable & IFCAP_CSUM_UDPv6_Rx) 2940 ifp->if_csum_flags_rx |= M_CSUM_UDPv6; 2941 2942 if (ifp->if_capenable & IFCAP_TSOv4) 2943 ifp->if_csum_flags_tx |= M_CSUM_TSOv4; 2944 if (ifp->if_capenable & IFCAP_TSOv6) 2945 ifp->if_csum_flags_tx |= M_CSUM_TSOv6; 2946 2947 #if NBRIDGE > 0 2948 if (ifp->if_bridge != NULL) 2949 bridge_calc_csum_flags(ifp->if_bridge); 2950 #endif 2951 2952 if (ifp->if_flags & IFF_UP) 2953 return ENETRESET; 2954 return 0; 2955 case SIOCSIFFLAGS: 2956 ifr = data; 2957 /* 2958 * If if_is_mpsafe(ifp), KERNEL_LOCK isn't held here, but if_up 2959 * and if_down aren't MP-safe yet, so we must hold the lock. 2960 */ 2961 KERNEL_LOCK_IF_IFP_MPSAFE(ifp); 2962 if (ifp->if_flags & IFF_UP && (ifr->ifr_flags & IFF_UP) == 0) { 2963 s = splsoftnet(); 2964 if_down_locked(ifp); 2965 splx(s); 2966 } 2967 if (ifr->ifr_flags & IFF_UP && (ifp->if_flags & IFF_UP) == 0) { 2968 s = splsoftnet(); 2969 if_up_locked(ifp); 2970 splx(s); 2971 } 2972 KERNEL_UNLOCK_IF_IFP_MPSAFE(ifp); 2973 ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) | 2974 (ifr->ifr_flags &~ IFF_CANTCHANGE); 2975 break; 2976 case SIOCGIFFLAGS: 2977 ifr = data; 2978 ifr->ifr_flags = ifp->if_flags; 2979 break; 2980 2981 case SIOCGIFMETRIC: 2982 ifr = data; 2983 ifr->ifr_metric = ifp->if_metric; 2984 break; 2985 2986 case SIOCGIFMTU: 2987 ifr = data; 2988 ifr->ifr_mtu = ifp->if_mtu; 2989 break; 2990 2991 case SIOCGIFDLT: 2992 ifr = data; 2993 ifr->ifr_dlt = ifp->if_dlt; 2994 break; 2995 2996 case SIOCGIFCAP: 2997 ifcr = data; 2998 ifcr->ifcr_capabilities = ifp->if_capabilities; 2999 ifcr->ifcr_capenable = ifp->if_capenable; 3000 break; 3001 3002 case SIOCSIFMETRIC: 3003 ifr = data; 3004 ifp->if_metric = ifr->ifr_metric; 3005 break; 3006 3007 case SIOCGIFDATA: 3008 ifdr = data; 3009 ifdr->ifdr_data = ifp->if_data; 3010 break; 3011 3012 case SIOCGIFINDEX: 3013 ifr = data; 3014 ifr->ifr_index = ifp->if_index; 3015 break; 3016 3017 case SIOCZIFDATA: 3018 ifdr = data; 3019 ifdr->ifdr_data = ifp->if_data; 3020 /* 3021 * Assumes that the volatile counters that can be 3022 * zero'ed are at the end of if_data. 3023 */ 3024 memset(&ifp->if_data.ifi_ipackets, 0, sizeof(ifp->if_data) - 3025 offsetof(struct if_data, ifi_ipackets)); 3026 /* 3027 * The memset() clears to the bottm of if_data. In the area, 3028 * if_lastchange is included. Please be careful if new entry 3029 * will be added into if_data or rewite this. 3030 * 3031 * And also, update if_lastchnage. 3032 */ 3033 getnanotime(&ifp->if_lastchange); 3034 break; 3035 case SIOCSIFMTU: 3036 ifr = data; 3037 if (ifp->if_mtu == ifr->ifr_mtu) 3038 break; 3039 ifp->if_mtu = ifr->ifr_mtu; 3040 /* 3041 * If the link MTU changed, do network layer specific procedure. 3042 */ 3043 #ifdef INET6 3044 KERNEL_LOCK_UNLESS_NET_MPSAFE(); 3045 if (in6_present) 3046 nd6_setmtu(ifp); 3047 KERNEL_UNLOCK_UNLESS_NET_MPSAFE(); 3048 #endif 3049 return ENETRESET; 3050 default: 3051 return ENOTTY; 3052 } 3053 return 0; 3054 } 3055 3056 int 3057 ifaddrpref_ioctl(struct socket *so, u_long cmd, void *data, struct ifnet *ifp) 3058 { 3059 struct if_addrprefreq *ifap = (struct if_addrprefreq *)data; 3060 struct ifaddr *ifa; 3061 const struct sockaddr *any, *sa; 3062 union { 3063 struct sockaddr sa; 3064 struct sockaddr_storage ss; 3065 } u, v; 3066 int s, error = 0; 3067 3068 switch (cmd) { 3069 case SIOCSIFADDRPREF: 3070 if (kauth_authorize_network(curlwp->l_cred, KAUTH_NETWORK_INTERFACE, 3071 KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd, 3072 NULL) != 0) 3073 return EPERM; 3074 case SIOCGIFADDRPREF: 3075 break; 3076 default: 3077 return EOPNOTSUPP; 3078 } 3079 3080 /* sanity checks */ 3081 if (data == NULL || ifp == NULL) { 3082 panic("invalid argument to %s", __func__); 3083 /*NOTREACHED*/ 3084 } 3085 3086 /* address must be specified on ADD and DELETE */ 3087 sa = sstocsa(&ifap->ifap_addr); 3088 if (sa->sa_family != sofamily(so)) 3089 return EINVAL; 3090 if ((any = sockaddr_any(sa)) == NULL || sa->sa_len != any->sa_len) 3091 return EINVAL; 3092 3093 sockaddr_externalize(&v.sa, sizeof(v.ss), sa); 3094 3095 s = pserialize_read_enter(); 3096 IFADDR_READER_FOREACH(ifa, ifp) { 3097 if (ifa->ifa_addr->sa_family != sa->sa_family) 3098 continue; 3099 sockaddr_externalize(&u.sa, sizeof(u.ss), ifa->ifa_addr); 3100 if (sockaddr_cmp(&u.sa, &v.sa) == 0) 3101 break; 3102 } 3103 if (ifa == NULL) { 3104 error = EADDRNOTAVAIL; 3105 goto out; 3106 } 3107 3108 switch (cmd) { 3109 case SIOCSIFADDRPREF: 3110 ifa->ifa_preference = ifap->ifap_preference; 3111 goto out; 3112 case SIOCGIFADDRPREF: 3113 /* fill in the if_laddrreq structure */ 3114 (void)sockaddr_copy(sstosa(&ifap->ifap_addr), 3115 sizeof(ifap->ifap_addr), ifa->ifa_addr); 3116 ifap->ifap_preference = ifa->ifa_preference; 3117 goto out; 3118 default: 3119 error = EOPNOTSUPP; 3120 } 3121 out: 3122 pserialize_read_exit(s); 3123 return error; 3124 } 3125 3126 /* 3127 * Interface ioctls. 3128 */ 3129 static int 3130 doifioctl(struct socket *so, u_long cmd, void *data, struct lwp *l) 3131 { 3132 struct ifnet *ifp; 3133 struct ifreq *ifr; 3134 int error = 0, hook; 3135 u_long ocmd = cmd; 3136 short oif_flags; 3137 struct ifreq ifrb; 3138 struct oifreq *oifr = NULL; 3139 int r; 3140 struct psref psref; 3141 int bound; 3142 3143 switch (cmd) { 3144 case SIOCGIFCONF: 3145 return ifconf(cmd, data); 3146 case SIOCINITIFADDR: 3147 return EPERM; 3148 default: 3149 MODULE_HOOK_CALL(uipc_syscalls_40_hook, (cmd, data), enosys(), 3150 error); 3151 if (error != ENOSYS) 3152 return error; 3153 MODULE_HOOK_CALL(uipc_syscalls_50_hook, (l, cmd, data), 3154 enosys(), error); 3155 if (error != ENOSYS) 3156 return error; 3157 error = 0; 3158 break; 3159 } 3160 3161 ifr = data; 3162 MODULE_HOOK_CALL(if_cvtcmd_43_hook, (&cmd, ocmd), enosys(), hook); 3163 if (hook != ENOSYS) { 3164 if (cmd != ocmd) { 3165 oifr = data; 3166 data = ifr = &ifrb; 3167 ifreqo2n(oifr, ifr); 3168 } 3169 } 3170 3171 switch (cmd) { 3172 case SIOCIFCREATE: 3173 case SIOCIFDESTROY: 3174 bound = curlwp_bind(); 3175 if (l != NULL) { 3176 ifp = if_get(ifr->ifr_name, &psref); 3177 error = kauth_authorize_network(l->l_cred, 3178 KAUTH_NETWORK_INTERFACE, 3179 KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, 3180 (void *)cmd, NULL); 3181 if (ifp != NULL) 3182 if_put(ifp, &psref); 3183 if (error != 0) { 3184 curlwp_bindx(bound); 3185 return error; 3186 } 3187 } 3188 KERNEL_LOCK_UNLESS_NET_MPSAFE(); 3189 mutex_enter(&if_clone_mtx); 3190 r = (cmd == SIOCIFCREATE) ? 3191 if_clone_create(ifr->ifr_name) : 3192 if_clone_destroy(ifr->ifr_name); 3193 mutex_exit(&if_clone_mtx); 3194 KERNEL_UNLOCK_UNLESS_NET_MPSAFE(); 3195 curlwp_bindx(bound); 3196 return r; 3197 3198 case SIOCIFGCLONERS: 3199 { 3200 struct if_clonereq *req = (struct if_clonereq *)data; 3201 return if_clone_list(req->ifcr_count, req->ifcr_buffer, 3202 &req->ifcr_total); 3203 } 3204 } 3205 3206 bound = curlwp_bind(); 3207 ifp = if_get(ifr->ifr_name, &psref); 3208 if (ifp == NULL) { 3209 curlwp_bindx(bound); 3210 return ENXIO; 3211 } 3212 3213 switch (cmd) { 3214 case SIOCALIFADDR: 3215 case SIOCDLIFADDR: 3216 case SIOCSIFADDRPREF: 3217 case SIOCSIFFLAGS: 3218 case SIOCSIFCAP: 3219 case SIOCSIFMETRIC: 3220 case SIOCZIFDATA: 3221 case SIOCSIFMTU: 3222 case SIOCSIFPHYADDR: 3223 case SIOCDIFPHYADDR: 3224 #ifdef INET6 3225 case SIOCSIFPHYADDR_IN6: 3226 #endif 3227 case SIOCSLIFPHYADDR: 3228 case SIOCADDMULTI: 3229 case SIOCDELMULTI: 3230 case SIOCSETHERCAP: 3231 case SIOCSIFMEDIA: 3232 case SIOCSDRVSPEC: 3233 case SIOCG80211: 3234 case SIOCS80211: 3235 case SIOCS80211NWID: 3236 case SIOCS80211NWKEY: 3237 case SIOCS80211POWER: 3238 case SIOCS80211BSSID: 3239 case SIOCS80211CHANNEL: 3240 case SIOCSLINKSTR: 3241 if (l != NULL) { 3242 error = kauth_authorize_network(l->l_cred, 3243 KAUTH_NETWORK_INTERFACE, 3244 KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, 3245 (void *)cmd, NULL); 3246 if (error != 0) 3247 goto out; 3248 } 3249 } 3250 3251 oif_flags = ifp->if_flags; 3252 3253 KERNEL_LOCK_UNLESS_IFP_MPSAFE(ifp); 3254 IFNET_LOCK(ifp); 3255 3256 error = (*ifp->if_ioctl)(ifp, cmd, data); 3257 if (error != ENOTTY) 3258 ; 3259 else if (so->so_proto == NULL) 3260 error = EOPNOTSUPP; 3261 else { 3262 KERNEL_LOCK_IF_IFP_MPSAFE(ifp); 3263 MODULE_HOOK_CALL(if_ifioctl_43_hook, 3264 (so, ocmd, cmd, data, l), enosys(), error); 3265 if (error == ENOSYS) 3266 error = (*so->so_proto->pr_usrreqs->pr_ioctl)(so, 3267 cmd, data, ifp); 3268 KERNEL_UNLOCK_IF_IFP_MPSAFE(ifp); 3269 } 3270 3271 if (((oif_flags ^ ifp->if_flags) & IFF_UP) != 0) { 3272 if ((ifp->if_flags & IFF_UP) != 0) { 3273 int s = splsoftnet(); 3274 if_up_locked(ifp); 3275 splx(s); 3276 } 3277 } 3278 if (cmd != ocmd) 3279 ifreqn2o(oifr, ifr); 3280 3281 IFNET_UNLOCK(ifp); 3282 KERNEL_UNLOCK_UNLESS_IFP_MPSAFE(ifp); 3283 out: 3284 if_put(ifp, &psref); 3285 curlwp_bindx(bound); 3286 return error; 3287 } 3288 3289 /* 3290 * Return interface configuration 3291 * of system. List may be used 3292 * in later ioctl's (above) to get 3293 * other information. 3294 * 3295 * Each record is a struct ifreq. Before the addition of 3296 * sockaddr_storage, the API rule was that sockaddr flavors that did 3297 * not fit would extend beyond the struct ifreq, with the next struct 3298 * ifreq starting sa_len beyond the struct sockaddr. Because the 3299 * union in struct ifreq includes struct sockaddr_storage, every kind 3300 * of sockaddr must fit. Thus, there are no longer any overlength 3301 * records. 3302 * 3303 * Records are added to the user buffer if they fit, and ifc_len is 3304 * adjusted to the length that was written. Thus, the user is only 3305 * assured of getting the complete list if ifc_len on return is at 3306 * least sizeof(struct ifreq) less than it was on entry. 3307 * 3308 * If the user buffer pointer is NULL, this routine copies no data and 3309 * returns the amount of space that would be needed. 3310 * 3311 * Invariants: 3312 * ifrp points to the next part of the user's buffer to be used. If 3313 * ifrp != NULL, space holds the number of bytes remaining that we may 3314 * write at ifrp. Otherwise, space holds the number of bytes that 3315 * would have been written had there been adequate space. 3316 */ 3317 /*ARGSUSED*/ 3318 static int 3319 ifconf(u_long cmd, void *data) 3320 { 3321 struct ifconf *ifc = (struct ifconf *)data; 3322 struct ifnet *ifp; 3323 struct ifaddr *ifa; 3324 struct ifreq ifr, *ifrp = NULL; 3325 int space = 0, error = 0; 3326 const int sz = (int)sizeof(struct ifreq); 3327 const bool docopy = ifc->ifc_req != NULL; 3328 int s; 3329 int bound; 3330 struct psref psref; 3331 3332 if (docopy) { 3333 space = ifc->ifc_len; 3334 ifrp = ifc->ifc_req; 3335 } 3336 3337 bound = curlwp_bind(); 3338 s = pserialize_read_enter(); 3339 IFNET_READER_FOREACH(ifp) { 3340 psref_acquire(&psref, &ifp->if_psref, ifnet_psref_class); 3341 pserialize_read_exit(s); 3342 3343 (void)strncpy(ifr.ifr_name, ifp->if_xname, 3344 sizeof(ifr.ifr_name)); 3345 if (ifr.ifr_name[sizeof(ifr.ifr_name) - 1] != '\0') { 3346 error = ENAMETOOLONG; 3347 goto release_exit; 3348 } 3349 if (IFADDR_READER_EMPTY(ifp)) { 3350 /* Interface with no addresses - send zero sockaddr. */ 3351 memset(&ifr.ifr_addr, 0, sizeof(ifr.ifr_addr)); 3352 if (!docopy) { 3353 space += sz; 3354 goto next; 3355 } 3356 if (space >= sz) { 3357 error = copyout(&ifr, ifrp, sz); 3358 if (error != 0) 3359 goto release_exit; 3360 ifrp++; 3361 space -= sz; 3362 } 3363 } 3364 3365 s = pserialize_read_enter(); 3366 IFADDR_READER_FOREACH(ifa, ifp) { 3367 struct sockaddr *sa = ifa->ifa_addr; 3368 /* all sockaddrs must fit in sockaddr_storage */ 3369 KASSERT(sa->sa_len <= sizeof(ifr.ifr_ifru)); 3370 3371 if (!docopy) { 3372 space += sz; 3373 continue; 3374 } 3375 memcpy(&ifr.ifr_space, sa, sa->sa_len); 3376 pserialize_read_exit(s); 3377 3378 if (space >= sz) { 3379 error = copyout(&ifr, ifrp, sz); 3380 if (error != 0) 3381 goto release_exit; 3382 ifrp++; space -= sz; 3383 } 3384 s = pserialize_read_enter(); 3385 } 3386 pserialize_read_exit(s); 3387 3388 next: 3389 s = pserialize_read_enter(); 3390 psref_release(&psref, &ifp->if_psref, ifnet_psref_class); 3391 } 3392 pserialize_read_exit(s); 3393 curlwp_bindx(bound); 3394 3395 if (docopy) { 3396 KASSERT(0 <= space && space <= ifc->ifc_len); 3397 ifc->ifc_len -= space; 3398 } else { 3399 KASSERT(space >= 0); 3400 ifc->ifc_len = space; 3401 } 3402 return (0); 3403 3404 release_exit: 3405 psref_release(&psref, &ifp->if_psref, ifnet_psref_class); 3406 curlwp_bindx(bound); 3407 return error; 3408 } 3409 3410 int 3411 ifreq_setaddr(u_long cmd, struct ifreq *ifr, const struct sockaddr *sa) 3412 { 3413 uint8_t len = sizeof(ifr->ifr_ifru.ifru_space); 3414 struct ifreq ifrb; 3415 struct oifreq *oifr = NULL; 3416 u_long ocmd = cmd; 3417 int hook; 3418 3419 MODULE_HOOK_CALL(if_cvtcmd_43_hook, (&cmd, ocmd), enosys(), hook); 3420 if (hook != ENOSYS) { 3421 if (cmd != ocmd) { 3422 oifr = (struct oifreq *)(void *)ifr; 3423 ifr = &ifrb; 3424 ifreqo2n(oifr, ifr); 3425 len = sizeof(oifr->ifr_addr); 3426 } 3427 } 3428 3429 if (len < sa->sa_len) 3430 return EFBIG; 3431 3432 memset(&ifr->ifr_addr, 0, len); 3433 sockaddr_copy(&ifr->ifr_addr, len, sa); 3434 3435 if (cmd != ocmd) 3436 ifreqn2o(oifr, ifr); 3437 return 0; 3438 } 3439 3440 /* 3441 * wrapper function for the drivers which doesn't have if_transmit(). 3442 */ 3443 static int 3444 if_transmit(struct ifnet *ifp, struct mbuf *m) 3445 { 3446 int s, error; 3447 size_t pktlen = m->m_pkthdr.len; 3448 bool mcast = (m->m_flags & M_MCAST) != 0; 3449 3450 s = splnet(); 3451 3452 IFQ_ENQUEUE(&ifp->if_snd, m, error); 3453 if (error != 0) { 3454 /* mbuf is already freed */ 3455 goto out; 3456 } 3457 3458 ifp->if_obytes += pktlen; 3459 if (mcast) 3460 ifp->if_omcasts++; 3461 3462 if ((ifp->if_flags & IFF_OACTIVE) == 0) 3463 if_start_lock(ifp); 3464 out: 3465 splx(s); 3466 3467 return error; 3468 } 3469 3470 int 3471 if_transmit_lock(struct ifnet *ifp, struct mbuf *m) 3472 { 3473 int error; 3474 3475 #ifdef ALTQ 3476 KERNEL_LOCK(1, NULL); 3477 if (ALTQ_IS_ENABLED(&ifp->if_snd)) { 3478 error = if_transmit(ifp, m); 3479 KERNEL_UNLOCK_ONE(NULL); 3480 } else { 3481 KERNEL_UNLOCK_ONE(NULL); 3482 error = (*ifp->if_transmit)(ifp, m); 3483 /* mbuf is alredy freed */ 3484 } 3485 #else /* !ALTQ */ 3486 error = (*ifp->if_transmit)(ifp, m); 3487 /* mbuf is alredy freed */ 3488 #endif /* !ALTQ */ 3489 3490 return error; 3491 } 3492 3493 /* 3494 * Queue message on interface, and start output if interface 3495 * not yet active. 3496 */ 3497 int 3498 ifq_enqueue(struct ifnet *ifp, struct mbuf *m) 3499 { 3500 3501 return if_transmit_lock(ifp, m); 3502 } 3503 3504 /* 3505 * Queue message on interface, possibly using a second fast queue 3506 */ 3507 int 3508 ifq_enqueue2(struct ifnet *ifp, struct ifqueue *ifq, struct mbuf *m) 3509 { 3510 int error = 0; 3511 3512 if (ifq != NULL 3513 #ifdef ALTQ 3514 && ALTQ_IS_ENABLED(&ifp->if_snd) == 0 3515 #endif 3516 ) { 3517 if (IF_QFULL(ifq)) { 3518 IF_DROP(&ifp->if_snd); 3519 m_freem(m); 3520 if (error == 0) 3521 error = ENOBUFS; 3522 } else 3523 IF_ENQUEUE(ifq, m); 3524 } else 3525 IFQ_ENQUEUE(&ifp->if_snd, m, error); 3526 if (error != 0) { 3527 ++ifp->if_oerrors; 3528 return error; 3529 } 3530 return 0; 3531 } 3532 3533 int 3534 if_addr_init(ifnet_t *ifp, struct ifaddr *ifa, const bool src) 3535 { 3536 int rc; 3537 3538 KASSERT(IFNET_LOCKED(ifp)); 3539 if (ifp->if_initaddr != NULL) 3540 rc = (*ifp->if_initaddr)(ifp, ifa, src); 3541 else if (src || 3542 (rc = (*ifp->if_ioctl)(ifp, SIOCSIFDSTADDR, ifa)) == ENOTTY) 3543 rc = (*ifp->if_ioctl)(ifp, SIOCINITIFADDR, ifa); 3544 3545 return rc; 3546 } 3547 3548 int 3549 if_do_dad(struct ifnet *ifp) 3550 { 3551 if ((ifp->if_flags & IFF_LOOPBACK) != 0) 3552 return 0; 3553 3554 switch (ifp->if_type) { 3555 case IFT_FAITH: 3556 /* 3557 * These interfaces do not have the IFF_LOOPBACK flag, 3558 * but loop packets back. We do not have to do DAD on such 3559 * interfaces. We should even omit it, because loop-backed 3560 * responses would confuse the DAD procedure. 3561 */ 3562 return 0; 3563 default: 3564 /* 3565 * Our DAD routine requires the interface up and running. 3566 * However, some interfaces can be up before the RUNNING 3567 * status. Additionaly, users may try to assign addresses 3568 * before the interface becomes up (or running). 3569 * We simply skip DAD in such a case as a work around. 3570 * XXX: we should rather mark "tentative" on such addresses, 3571 * and do DAD after the interface becomes ready. 3572 */ 3573 if ((ifp->if_flags & (IFF_UP|IFF_RUNNING)) != 3574 (IFF_UP|IFF_RUNNING)) 3575 return 0; 3576 3577 return 1; 3578 } 3579 } 3580 3581 int 3582 if_flags_set(ifnet_t *ifp, const short flags) 3583 { 3584 int rc; 3585 3586 KASSERT(IFNET_LOCKED(ifp)); 3587 3588 if (ifp->if_setflags != NULL) 3589 rc = (*ifp->if_setflags)(ifp, flags); 3590 else { 3591 short cantflags, chgdflags; 3592 struct ifreq ifr; 3593 3594 chgdflags = ifp->if_flags ^ flags; 3595 cantflags = chgdflags & IFF_CANTCHANGE; 3596 3597 if (cantflags != 0) 3598 ifp->if_flags ^= cantflags; 3599 3600 /* Traditionally, we do not call if_ioctl after 3601 * setting/clearing only IFF_PROMISC if the interface 3602 * isn't IFF_UP. Uphold that tradition. 3603 */ 3604 if (chgdflags == IFF_PROMISC && (ifp->if_flags & IFF_UP) == 0) 3605 return 0; 3606 3607 memset(&ifr, 0, sizeof(ifr)); 3608 3609 ifr.ifr_flags = flags & ~IFF_CANTCHANGE; 3610 rc = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, &ifr); 3611 3612 if (rc != 0 && cantflags != 0) 3613 ifp->if_flags ^= cantflags; 3614 } 3615 3616 return rc; 3617 } 3618 3619 int 3620 if_mcast_op(ifnet_t *ifp, const unsigned long cmd, const struct sockaddr *sa) 3621 { 3622 int rc; 3623 struct ifreq ifr; 3624 3625 /* There remain some paths that don't hold IFNET_LOCK yet */ 3626 #ifdef NET_MPSAFE 3627 /* CARP and MROUTING still don't deal with the lock yet */ 3628 #if (!defined(NCARP) || (NCARP == 0)) && !defined(MROUTING) 3629 KASSERT(IFNET_LOCKED(ifp)); 3630 #endif 3631 #endif 3632 if (ifp->if_mcastop != NULL) 3633 rc = (*ifp->if_mcastop)(ifp, cmd, sa); 3634 else { 3635 ifreq_setaddr(cmd, &ifr, sa); 3636 rc = (*ifp->if_ioctl)(ifp, cmd, &ifr); 3637 } 3638 3639 return rc; 3640 } 3641 3642 static void 3643 sysctl_sndq_setup(struct sysctllog **clog, const char *ifname, 3644 struct ifaltq *ifq) 3645 { 3646 const struct sysctlnode *cnode, *rnode; 3647 3648 if (sysctl_createv(clog, 0, NULL, &rnode, 3649 CTLFLAG_PERMANENT, 3650 CTLTYPE_NODE, "interfaces", 3651 SYSCTL_DESCR("Per-interface controls"), 3652 NULL, 0, NULL, 0, 3653 CTL_NET, CTL_CREATE, CTL_EOL) != 0) 3654 goto bad; 3655 3656 if (sysctl_createv(clog, 0, &rnode, &rnode, 3657 CTLFLAG_PERMANENT, 3658 CTLTYPE_NODE, ifname, 3659 SYSCTL_DESCR("Interface controls"), 3660 NULL, 0, NULL, 0, 3661 CTL_CREATE, CTL_EOL) != 0) 3662 goto bad; 3663 3664 if (sysctl_createv(clog, 0, &rnode, &rnode, 3665 CTLFLAG_PERMANENT, 3666 CTLTYPE_NODE, "sndq", 3667 SYSCTL_DESCR("Interface output queue controls"), 3668 NULL, 0, NULL, 0, 3669 CTL_CREATE, CTL_EOL) != 0) 3670 goto bad; 3671 3672 if (sysctl_createv(clog, 0, &rnode, &cnode, 3673 CTLFLAG_PERMANENT, 3674 CTLTYPE_INT, "len", 3675 SYSCTL_DESCR("Current output queue length"), 3676 NULL, 0, &ifq->ifq_len, 0, 3677 CTL_CREATE, CTL_EOL) != 0) 3678 goto bad; 3679 3680 if (sysctl_createv(clog, 0, &rnode, &cnode, 3681 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 3682 CTLTYPE_INT, "maxlen", 3683 SYSCTL_DESCR("Maximum allowed output queue length"), 3684 NULL, 0, &ifq->ifq_maxlen, 0, 3685 CTL_CREATE, CTL_EOL) != 0) 3686 goto bad; 3687 3688 if (sysctl_createv(clog, 0, &rnode, &cnode, 3689 CTLFLAG_PERMANENT, 3690 CTLTYPE_INT, "drops", 3691 SYSCTL_DESCR("Packets dropped due to full output queue"), 3692 NULL, 0, &ifq->ifq_drops, 0, 3693 CTL_CREATE, CTL_EOL) != 0) 3694 goto bad; 3695 3696 return; 3697 bad: 3698 printf("%s: could not attach sysctl nodes\n", ifname); 3699 return; 3700 } 3701 3702 #if defined(INET) || defined(INET6) 3703 3704 #define SYSCTL_NET_PKTQ(q, cn, c) \ 3705 static int \ 3706 sysctl_net_##q##_##cn(SYSCTLFN_ARGS) \ 3707 { \ 3708 return sysctl_pktq_count(SYSCTLFN_CALL(rnode), q, c); \ 3709 } 3710 3711 #if defined(INET) 3712 static int 3713 sysctl_net_ip_pktq_maxlen(SYSCTLFN_ARGS) 3714 { 3715 return sysctl_pktq_maxlen(SYSCTLFN_CALL(rnode), ip_pktq); 3716 } 3717 SYSCTL_NET_PKTQ(ip_pktq, items, PKTQ_NITEMS) 3718 SYSCTL_NET_PKTQ(ip_pktq, drops, PKTQ_DROPS) 3719 #endif 3720 3721 #if defined(INET6) 3722 static int 3723 sysctl_net_ip6_pktq_maxlen(SYSCTLFN_ARGS) 3724 { 3725 return sysctl_pktq_maxlen(SYSCTLFN_CALL(rnode), ip6_pktq); 3726 } 3727 SYSCTL_NET_PKTQ(ip6_pktq, items, PKTQ_NITEMS) 3728 SYSCTL_NET_PKTQ(ip6_pktq, drops, PKTQ_DROPS) 3729 #endif 3730 3731 static void 3732 sysctl_net_pktq_setup(struct sysctllog **clog, int pf) 3733 { 3734 sysctlfn len_func = NULL, maxlen_func = NULL, drops_func = NULL; 3735 const char *pfname = NULL, *ipname = NULL; 3736 int ipn = 0, qid = 0; 3737 3738 switch (pf) { 3739 #if defined(INET) 3740 case PF_INET: 3741 len_func = sysctl_net_ip_pktq_items; 3742 maxlen_func = sysctl_net_ip_pktq_maxlen; 3743 drops_func = sysctl_net_ip_pktq_drops; 3744 pfname = "inet", ipn = IPPROTO_IP; 3745 ipname = "ip", qid = IPCTL_IFQ; 3746 break; 3747 #endif 3748 #if defined(INET6) 3749 case PF_INET6: 3750 len_func = sysctl_net_ip6_pktq_items; 3751 maxlen_func = sysctl_net_ip6_pktq_maxlen; 3752 drops_func = sysctl_net_ip6_pktq_drops; 3753 pfname = "inet6", ipn = IPPROTO_IPV6; 3754 ipname = "ip6", qid = IPV6CTL_IFQ; 3755 break; 3756 #endif 3757 default: 3758 KASSERT(false); 3759 } 3760 3761 sysctl_createv(clog, 0, NULL, NULL, 3762 CTLFLAG_PERMANENT, 3763 CTLTYPE_NODE, pfname, NULL, 3764 NULL, 0, NULL, 0, 3765 CTL_NET, pf, CTL_EOL); 3766 sysctl_createv(clog, 0, NULL, NULL, 3767 CTLFLAG_PERMANENT, 3768 CTLTYPE_NODE, ipname, NULL, 3769 NULL, 0, NULL, 0, 3770 CTL_NET, pf, ipn, CTL_EOL); 3771 sysctl_createv(clog, 0, NULL, NULL, 3772 CTLFLAG_PERMANENT, 3773 CTLTYPE_NODE, "ifq", 3774 SYSCTL_DESCR("Protocol input queue controls"), 3775 NULL, 0, NULL, 0, 3776 CTL_NET, pf, ipn, qid, CTL_EOL); 3777 3778 sysctl_createv(clog, 0, NULL, NULL, 3779 CTLFLAG_PERMANENT, 3780 CTLTYPE_QUAD, "len", 3781 SYSCTL_DESCR("Current input queue length"), 3782 len_func, 0, NULL, 0, 3783 CTL_NET, pf, ipn, qid, IFQCTL_LEN, CTL_EOL); 3784 sysctl_createv(clog, 0, NULL, NULL, 3785 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 3786 CTLTYPE_INT, "maxlen", 3787 SYSCTL_DESCR("Maximum allowed input queue length"), 3788 maxlen_func, 0, NULL, 0, 3789 CTL_NET, pf, ipn, qid, IFQCTL_MAXLEN, CTL_EOL); 3790 sysctl_createv(clog, 0, NULL, NULL, 3791 CTLFLAG_PERMANENT, 3792 CTLTYPE_QUAD, "drops", 3793 SYSCTL_DESCR("Packets dropped due to full input queue"), 3794 drops_func, 0, NULL, 0, 3795 CTL_NET, pf, ipn, qid, IFQCTL_DROPS, CTL_EOL); 3796 } 3797 #endif /* INET || INET6 */ 3798 3799 static int 3800 if_sdl_sysctl(SYSCTLFN_ARGS) 3801 { 3802 struct ifnet *ifp; 3803 const struct sockaddr_dl *sdl; 3804 struct psref psref; 3805 int error = 0; 3806 int bound; 3807 3808 if (namelen != 1) 3809 return EINVAL; 3810 3811 bound = curlwp_bind(); 3812 ifp = if_get_byindex(name[0], &psref); 3813 if (ifp == NULL) { 3814 error = ENODEV; 3815 goto out0; 3816 } 3817 3818 sdl = ifp->if_sadl; 3819 if (sdl == NULL) { 3820 *oldlenp = 0; 3821 goto out1; 3822 } 3823 3824 if (oldp == NULL) { 3825 *oldlenp = sdl->sdl_alen; 3826 goto out1; 3827 } 3828 3829 if (*oldlenp >= sdl->sdl_alen) 3830 *oldlenp = sdl->sdl_alen; 3831 error = sysctl_copyout(l, &sdl->sdl_data[sdl->sdl_nlen], oldp, *oldlenp); 3832 out1: 3833 if_put(ifp, &psref); 3834 out0: 3835 curlwp_bindx(bound); 3836 return error; 3837 } 3838 3839 static void 3840 if_sysctl_setup(struct sysctllog **clog) 3841 { 3842 const struct sysctlnode *rnode = NULL; 3843 3844 sysctl_createv(clog, 0, NULL, &rnode, 3845 CTLFLAG_PERMANENT, 3846 CTLTYPE_NODE, "sdl", 3847 SYSCTL_DESCR("Get active link-layer address"), 3848 if_sdl_sysctl, 0, NULL, 0, 3849 CTL_NET, CTL_CREATE, CTL_EOL); 3850 3851 #if defined(INET) 3852 sysctl_net_pktq_setup(NULL, PF_INET); 3853 #endif 3854 #ifdef INET6 3855 if (in6_present) 3856 sysctl_net_pktq_setup(NULL, PF_INET6); 3857 #endif 3858 } 3859
Indexes created Mon Sep 22 13:09:51 GMT 2025