if.c revision 1.433
1 /* $NetBSD: if.c,v 1.433 2018/08/10 10:31:01 knakahara 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 1.433 2018/08/10 10:31:01 knakahara 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_natm.h" 100 #include "opt_wlan.h" 101 #include "opt_net_mpsafe.h" 102 #include "opt_mrouting.h" 103 #endif 104 105 #include <sys/param.h> 106 #include <sys/mbuf.h> 107 #include <sys/systm.h> 108 #include <sys/callout.h> 109 #include <sys/proc.h> 110 #include <sys/socket.h> 111 #include <sys/socketvar.h> 112 #include <sys/domain.h> 113 #include <sys/protosw.h> 114 #include <sys/kernel.h> 115 #include <sys/ioctl.h> 116 #include <sys/sysctl.h> 117 #include <sys/syslog.h> 118 #include <sys/kauth.h> 119 #include <sys/kmem.h> 120 #include <sys/xcall.h> 121 #include <sys/cpu.h> 122 #include <sys/intr.h> 123 124 #include <net/if.h> 125 #include <net/if_dl.h> 126 #include <net/if_ether.h> 127 #include <net/if_media.h> 128 #include <net80211/ieee80211.h> 129 #include <net80211/ieee80211_ioctl.h> 130 #include <net/if_types.h> 131 #include <net/route.h> 132 #include <net/netisr.h> 133 #include <sys/module.h> 134 #ifdef NETATALK 135 #include <netatalk/at_extern.h> 136 #include <netatalk/at.h> 137 #endif 138 #include <net/pfil.h> 139 #include <netinet/in.h> 140 #include <netinet/in_var.h> 141 #include <netinet/ip_encap.h> 142 #include <net/bpf.h> 143 144 #ifdef INET6 145 #include <netinet6/in6_var.h> 146 #include <netinet6/nd6.h> 147 #endif 148 149 #include "ether.h" 150 #include "fddi.h" 151 #include "token.h" 152 153 #include "carp.h" 154 #if NCARP > 0 155 #include <netinet/ip_carp.h> 156 #endif 157 158 #include <compat/sys/sockio.h> 159 #include <compat/sys/socket.h> 160 161 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address"); 162 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address"); 163 164 /* 165 * Global list of interfaces. 166 */ 167 /* DEPRECATED. Remove it once kvm(3) users disappeared */ 168 struct ifnet_head ifnet_list; 169 170 struct pslist_head ifnet_pslist; 171 static ifnet_t ** ifindex2ifnet = NULL; 172 static u_int if_index = 1; 173 static size_t if_indexlim = 0; 174 static uint64_t index_gen; 175 /* Mutex to protect the above objects. */ 176 kmutex_t ifnet_mtx __cacheline_aligned; 177 static struct psref_class *ifnet_psref_class __read_mostly; 178 static pserialize_t ifnet_psz; 179 180 static kmutex_t if_clone_mtx; 181 182 struct ifnet *lo0ifp; 183 int ifqmaxlen = IFQ_MAXLEN; 184 185 struct psref_class *ifa_psref_class __read_mostly; 186 187 static int if_delroute_matcher(struct rtentry *, void *); 188 189 static bool if_is_unit(const char *); 190 static struct if_clone *if_clone_lookup(const char *, int *); 191 192 static LIST_HEAD(, if_clone) if_cloners = LIST_HEAD_INITIALIZER(if_cloners); 193 static int if_cloners_count; 194 195 /* Packet filtering hook for interfaces. */ 196 pfil_head_t * if_pfil __read_mostly; 197 198 static kauth_listener_t if_listener; 199 200 static int doifioctl(struct socket *, u_long, void *, struct lwp *); 201 static void if_detach_queues(struct ifnet *, struct ifqueue *); 202 static void sysctl_sndq_setup(struct sysctllog **, const char *, 203 struct ifaltq *); 204 static void if_slowtimo(void *); 205 static void if_free_sadl(struct ifnet *, int); 206 static void if_attachdomain1(struct ifnet *); 207 static int ifconf(u_long, void *); 208 static int if_transmit(struct ifnet *, struct mbuf *); 209 static int if_clone_create(const char *); 210 static int if_clone_destroy(const char *); 211 static void if_link_state_change_si(void *); 212 static void if_up_locked(struct ifnet *); 213 static void _if_down(struct ifnet *); 214 static void if_down_deactivated(struct ifnet *); 215 216 struct if_percpuq { 217 struct ifnet *ipq_ifp; 218 void *ipq_si; 219 struct percpu *ipq_ifqs; /* struct ifqueue */ 220 }; 221 222 static struct mbuf *if_percpuq_dequeue(struct if_percpuq *); 223 224 static void if_percpuq_drops(void *, void *, struct cpu_info *); 225 static int sysctl_percpuq_drops_handler(SYSCTLFN_PROTO); 226 static void sysctl_percpuq_setup(struct sysctllog **, const char *, 227 struct if_percpuq *); 228 229 struct if_deferred_start { 230 struct ifnet *ids_ifp; 231 void (*ids_if_start)(struct ifnet *); 232 void *ids_si; 233 }; 234 235 static void if_deferred_start_softint(void *); 236 static void if_deferred_start_common(struct ifnet *); 237 static void if_deferred_start_destroy(struct ifnet *); 238 239 #if defined(INET) || defined(INET6) 240 static void sysctl_net_pktq_setup(struct sysctllog **, int); 241 #endif 242 243 static void if_sysctl_setup(struct sysctllog **); 244 245 /* Compatibility vector functions */ 246 u_long (*vec_compat_cvtcmd)(u_long) = NULL; 247 int (*vec_compat_ifioctl)(struct socket *, u_long, u_long, void *, 248 struct lwp *) = NULL; 249 int (*vec_compat_ifconf)(struct lwp *, u_long, void *) = (void *)enosys; 250 int (*vec_compat_ifdatareq)(struct lwp *, u_long, void *) = (void *)enosys; 251 252 static int 253 if_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie, 254 void *arg0, void *arg1, void *arg2, void *arg3) 255 { 256 int result; 257 enum kauth_network_req req; 258 259 result = KAUTH_RESULT_DEFER; 260 req = (enum kauth_network_req)arg1; 261 262 if (action != KAUTH_NETWORK_INTERFACE) 263 return result; 264 265 if ((req == KAUTH_REQ_NETWORK_INTERFACE_GET) || 266 (req == KAUTH_REQ_NETWORK_INTERFACE_SET)) 267 result = KAUTH_RESULT_ALLOW; 268 269 return result; 270 } 271 272 /* 273 * Network interface utility routines. 274 * 275 * Routines with ifa_ifwith* names take sockaddr *'s as 276 * parameters. 277 */ 278 void 279 ifinit(void) 280 { 281 282 #if (defined(INET) || defined(INET6)) 283 encapinit(); 284 #endif 285 286 if_listener = kauth_listen_scope(KAUTH_SCOPE_NETWORK, 287 if_listener_cb, NULL); 288 289 /* interfaces are available, inform socket code */ 290 ifioctl = doifioctl; 291 } 292 293 /* 294 * XXX Initialization before configure(). 295 * XXX hack to get pfil_add_hook working in autoconf. 296 */ 297 void 298 ifinit1(void) 299 { 300 301 #ifdef NET_MPSAFE 302 printf("NET_MPSAFE enabled\n"); 303 #endif 304 305 mutex_init(&if_clone_mtx, MUTEX_DEFAULT, IPL_NONE); 306 307 TAILQ_INIT(&ifnet_list); 308 mutex_init(&ifnet_mtx, MUTEX_DEFAULT, IPL_NONE); 309 ifnet_psz = pserialize_create(); 310 ifnet_psref_class = psref_class_create("ifnet", IPL_SOFTNET); 311 ifa_psref_class = psref_class_create("ifa", IPL_SOFTNET); 312 PSLIST_INIT(&ifnet_pslist); 313 314 if_indexlim = 8; 315 316 if_pfil = pfil_head_create(PFIL_TYPE_IFNET, NULL); 317 KASSERT(if_pfil != NULL); 318 319 #if NETHER > 0 || NFDDI > 0 || defined(NETATALK) || NTOKEN > 0 || defined(WLAN) 320 etherinit(); 321 #endif 322 } 323 324 /* XXX must be after domaininit() */ 325 void 326 ifinit_post(void) 327 { 328 329 if_sysctl_setup(NULL); 330 } 331 332 ifnet_t * 333 if_alloc(u_char type) 334 { 335 return kmem_zalloc(sizeof(ifnet_t), KM_SLEEP); 336 } 337 338 void 339 if_free(ifnet_t *ifp) 340 { 341 kmem_free(ifp, sizeof(ifnet_t)); 342 } 343 344 void 345 if_initname(struct ifnet *ifp, const char *name, int unit) 346 { 347 (void)snprintf(ifp->if_xname, sizeof(ifp->if_xname), 348 "%s%d", name, unit); 349 } 350 351 /* 352 * Null routines used while an interface is going away. These routines 353 * just return an error. 354 */ 355 356 int 357 if_nulloutput(struct ifnet *ifp, struct mbuf *m, 358 const struct sockaddr *so, const struct rtentry *rt) 359 { 360 361 return ENXIO; 362 } 363 364 void 365 if_nullinput(struct ifnet *ifp, struct mbuf *m) 366 { 367 368 /* Nothing. */ 369 } 370 371 void 372 if_nullstart(struct ifnet *ifp) 373 { 374 375 /* Nothing. */ 376 } 377 378 int 379 if_nulltransmit(struct ifnet *ifp, struct mbuf *m) 380 { 381 382 m_freem(m); 383 return ENXIO; 384 } 385 386 int 387 if_nullioctl(struct ifnet *ifp, u_long cmd, void *data) 388 { 389 390 return ENXIO; 391 } 392 393 int 394 if_nullinit(struct ifnet *ifp) 395 { 396 397 return ENXIO; 398 } 399 400 void 401 if_nullstop(struct ifnet *ifp, int disable) 402 { 403 404 /* Nothing. */ 405 } 406 407 void 408 if_nullslowtimo(struct ifnet *ifp) 409 { 410 411 /* Nothing. */ 412 } 413 414 void 415 if_nulldrain(struct ifnet *ifp) 416 { 417 418 /* Nothing. */ 419 } 420 421 void 422 if_set_sadl(struct ifnet *ifp, const void *lla, u_char addrlen, bool factory) 423 { 424 struct ifaddr *ifa; 425 struct sockaddr_dl *sdl; 426 427 ifp->if_addrlen = addrlen; 428 if_alloc_sadl(ifp); 429 ifa = ifp->if_dl; 430 sdl = satosdl(ifa->ifa_addr); 431 432 (void)sockaddr_dl_setaddr(sdl, sdl->sdl_len, lla, ifp->if_addrlen); 433 if (factory) { 434 KASSERT(ifp->if_hwdl == NULL); 435 ifp->if_hwdl = ifp->if_dl; 436 ifaref(ifp->if_hwdl); 437 } 438 /* TBD routing socket */ 439 } 440 441 struct ifaddr * 442 if_dl_create(const struct ifnet *ifp, const struct sockaddr_dl **sdlp) 443 { 444 unsigned socksize, ifasize; 445 int addrlen, namelen; 446 struct sockaddr_dl *mask, *sdl; 447 struct ifaddr *ifa; 448 449 namelen = strlen(ifp->if_xname); 450 addrlen = ifp->if_addrlen; 451 socksize = roundup(sockaddr_dl_measure(namelen, addrlen), sizeof(long)); 452 ifasize = sizeof(*ifa) + 2 * socksize; 453 ifa = malloc(ifasize, M_IFADDR, M_WAITOK|M_ZERO); 454 455 sdl = (struct sockaddr_dl *)(ifa + 1); 456 mask = (struct sockaddr_dl *)(socksize + (char *)sdl); 457 458 sockaddr_dl_init(sdl, socksize, ifp->if_index, ifp->if_type, 459 ifp->if_xname, namelen, NULL, addrlen); 460 mask->sdl_family = AF_LINK; 461 mask->sdl_len = sockaddr_dl_measure(namelen, 0); 462 memset(&mask->sdl_data[0], 0xff, namelen); 463 ifa->ifa_rtrequest = link_rtrequest; 464 ifa->ifa_addr = (struct sockaddr *)sdl; 465 ifa->ifa_netmask = (struct sockaddr *)mask; 466 ifa_psref_init(ifa); 467 468 *sdlp = sdl; 469 470 return ifa; 471 } 472 473 static void 474 if_sadl_setrefs(struct ifnet *ifp, struct ifaddr *ifa) 475 { 476 const struct sockaddr_dl *sdl; 477 478 ifp->if_dl = ifa; 479 ifaref(ifa); 480 sdl = satosdl(ifa->ifa_addr); 481 ifp->if_sadl = sdl; 482 } 483 484 /* 485 * Allocate the link level name for the specified interface. This 486 * is an attachment helper. It must be called after ifp->if_addrlen 487 * is initialized, which may not be the case when if_attach() is 488 * called. 489 */ 490 void 491 if_alloc_sadl(struct ifnet *ifp) 492 { 493 struct ifaddr *ifa; 494 const struct sockaddr_dl *sdl; 495 496 /* 497 * If the interface already has a link name, release it 498 * now. This is useful for interfaces that can change 499 * link types, and thus switch link names often. 500 */ 501 if (ifp->if_sadl != NULL) 502 if_free_sadl(ifp, 0); 503 504 ifa = if_dl_create(ifp, &sdl); 505 506 ifa_insert(ifp, ifa); 507 if_sadl_setrefs(ifp, ifa); 508 } 509 510 static void 511 if_deactivate_sadl(struct ifnet *ifp) 512 { 513 struct ifaddr *ifa; 514 515 KASSERT(ifp->if_dl != NULL); 516 517 ifa = ifp->if_dl; 518 519 ifp->if_sadl = NULL; 520 521 ifp->if_dl = NULL; 522 ifafree(ifa); 523 } 524 525 static void 526 if_replace_sadl(struct ifnet *ifp, struct ifaddr *ifa) 527 { 528 struct ifaddr *old; 529 530 KASSERT(ifp->if_dl != NULL); 531 532 old = ifp->if_dl; 533 534 ifaref(ifa); 535 /* XXX Update if_dl and if_sadl atomically */ 536 ifp->if_dl = ifa; 537 ifp->if_sadl = satosdl(ifa->ifa_addr); 538 539 ifafree(old); 540 } 541 542 void 543 if_activate_sadl(struct ifnet *ifp, struct ifaddr *ifa0, 544 const struct sockaddr_dl *sdl) 545 { 546 int s, ss; 547 struct ifaddr *ifa; 548 int bound = curlwp_bind(); 549 550 KASSERT(ifa_held(ifa0)); 551 552 s = splsoftnet(); 553 554 if_replace_sadl(ifp, ifa0); 555 556 ss = pserialize_read_enter(); 557 IFADDR_READER_FOREACH(ifa, ifp) { 558 struct psref psref; 559 ifa_acquire(ifa, &psref); 560 pserialize_read_exit(ss); 561 562 rtinit(ifa, RTM_LLINFO_UPD, 0); 563 564 ss = pserialize_read_enter(); 565 ifa_release(ifa, &psref); 566 } 567 pserialize_read_exit(ss); 568 569 splx(s); 570 curlwp_bindx(bound); 571 } 572 573 /* 574 * Free the link level name for the specified interface. This is 575 * a detach helper. This is called from if_detach(). 576 */ 577 static void 578 if_free_sadl(struct ifnet *ifp, int factory) 579 { 580 struct ifaddr *ifa; 581 int s; 582 583 if (factory && ifp->if_hwdl != NULL) { 584 ifa = ifp->if_hwdl; 585 ifp->if_hwdl = NULL; 586 ifafree(ifa); 587 } 588 589 ifa = ifp->if_dl; 590 if (ifa == NULL) { 591 KASSERT(ifp->if_sadl == NULL); 592 return; 593 } 594 595 KASSERT(ifp->if_sadl != NULL); 596 597 s = splsoftnet(); 598 KASSERT(ifa->ifa_addr->sa_family == AF_LINK); 599 ifa_remove(ifp, ifa); 600 if_deactivate_sadl(ifp); 601 splx(s); 602 } 603 604 static void 605 if_getindex(ifnet_t *ifp) 606 { 607 bool hitlimit = false; 608 609 ifp->if_index_gen = index_gen++; 610 611 ifp->if_index = if_index; 612 if (ifindex2ifnet == NULL) { 613 if_index++; 614 goto skip; 615 } 616 while (if_byindex(ifp->if_index)) { 617 /* 618 * If we hit USHRT_MAX, we skip back to 0 since 619 * there are a number of places where the value 620 * of if_index or if_index itself is compared 621 * to or stored in an unsigned short. By 622 * jumping back, we won't botch those assignments 623 * or comparisons. 624 */ 625 if (++if_index == 0) { 626 if_index = 1; 627 } else if (if_index == USHRT_MAX) { 628 /* 629 * However, if we have to jump back to 630 * zero *twice* without finding an empty 631 * slot in ifindex2ifnet[], then there 632 * there are too many (>65535) interfaces. 633 */ 634 if (hitlimit) { 635 panic("too many interfaces"); 636 } 637 hitlimit = true; 638 if_index = 1; 639 } 640 ifp->if_index = if_index; 641 } 642 skip: 643 /* 644 * ifindex2ifnet is indexed by if_index. Since if_index will 645 * grow dynamically, it should grow too. 646 */ 647 if (ifindex2ifnet == NULL || ifp->if_index >= if_indexlim) { 648 size_t m, n, oldlim; 649 void *q; 650 651 oldlim = if_indexlim; 652 while (ifp->if_index >= if_indexlim) 653 if_indexlim <<= 1; 654 655 /* grow ifindex2ifnet */ 656 m = oldlim * sizeof(struct ifnet *); 657 n = if_indexlim * sizeof(struct ifnet *); 658 q = malloc(n, M_IFADDR, M_WAITOK|M_ZERO); 659 if (ifindex2ifnet != NULL) { 660 memcpy(q, ifindex2ifnet, m); 661 free(ifindex2ifnet, M_IFADDR); 662 } 663 ifindex2ifnet = (struct ifnet **)q; 664 } 665 ifindex2ifnet[ifp->if_index] = ifp; 666 } 667 668 /* 669 * Initialize an interface and assign an index for it. 670 * 671 * It must be called prior to a device specific attach routine 672 * (e.g., ether_ifattach and ieee80211_ifattach) or if_alloc_sadl, 673 * and be followed by if_register: 674 * 675 * if_initialize(ifp); 676 * ether_ifattach(ifp, enaddr); 677 * if_register(ifp); 678 */ 679 int 680 if_initialize(ifnet_t *ifp) 681 { 682 int rv = 0; 683 684 KASSERT(if_indexlim > 0); 685 TAILQ_INIT(&ifp->if_addrlist); 686 687 /* 688 * Link level name is allocated later by a separate call to 689 * if_alloc_sadl(). 690 */ 691 692 if (ifp->if_snd.ifq_maxlen == 0) 693 ifp->if_snd.ifq_maxlen = ifqmaxlen; 694 695 ifp->if_broadcastaddr = 0; /* reliably crash if used uninitialized */ 696 697 ifp->if_link_state = LINK_STATE_UNKNOWN; 698 ifp->if_link_queue = -1; /* all bits set, see link_state_change() */ 699 700 ifp->if_capenable = 0; 701 ifp->if_csum_flags_tx = 0; 702 ifp->if_csum_flags_rx = 0; 703 704 #ifdef ALTQ 705 ifp->if_snd.altq_type = 0; 706 ifp->if_snd.altq_disc = NULL; 707 ifp->if_snd.altq_flags &= ALTQF_CANTCHANGE; 708 ifp->if_snd.altq_tbr = NULL; 709 ifp->if_snd.altq_ifp = ifp; 710 #endif 711 712 IFQ_LOCK_INIT(&ifp->if_snd); 713 714 ifp->if_pfil = pfil_head_create(PFIL_TYPE_IFNET, ifp); 715 pfil_run_ifhooks(if_pfil, PFIL_IFNET_ATTACH, ifp); 716 717 IF_AFDATA_LOCK_INIT(ifp); 718 719 if (if_is_link_state_changeable(ifp)) { 720 u_int flags = SOFTINT_NET; 721 flags |= if_is_mpsafe(ifp) ? SOFTINT_MPSAFE : 0; 722 ifp->if_link_si = softint_establish(flags, 723 if_link_state_change_si, ifp); 724 if (ifp->if_link_si == NULL) { 725 rv = ENOMEM; 726 goto fail; 727 } 728 } 729 730 PSLIST_ENTRY_INIT(ifp, if_pslist_entry); 731 PSLIST_INIT(&ifp->if_addr_pslist); 732 psref_target_init(&ifp->if_psref, ifnet_psref_class); 733 ifp->if_ioctl_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 734 LIST_INIT(&ifp->if_multiaddrs); 735 736 IFNET_GLOBAL_LOCK(); 737 if_getindex(ifp); 738 IFNET_GLOBAL_UNLOCK(); 739 740 return 0; 741 742 fail: 743 IF_AFDATA_LOCK_DESTROY(ifp); 744 745 pfil_run_ifhooks(if_pfil, PFIL_IFNET_DETACH, ifp); 746 (void)pfil_head_destroy(ifp->if_pfil); 747 748 IFQ_LOCK_DESTROY(&ifp->if_snd); 749 750 return rv; 751 } 752 753 /* 754 * Register an interface to the list of "active" interfaces. 755 */ 756 void 757 if_register(ifnet_t *ifp) 758 { 759 /* 760 * If the driver has not supplied its own if_ioctl, then 761 * supply the default. 762 */ 763 if (ifp->if_ioctl == NULL) 764 ifp->if_ioctl = ifioctl_common; 765 766 sysctl_sndq_setup(&ifp->if_sysctl_log, ifp->if_xname, &ifp->if_snd); 767 768 if (!STAILQ_EMPTY(&domains)) 769 if_attachdomain1(ifp); 770 771 /* Announce the interface. */ 772 rt_ifannouncemsg(ifp, IFAN_ARRIVAL); 773 774 if (ifp->if_slowtimo != NULL) { 775 ifp->if_slowtimo_ch = 776 kmem_zalloc(sizeof(*ifp->if_slowtimo_ch), KM_SLEEP); 777 callout_init(ifp->if_slowtimo_ch, 0); 778 callout_setfunc(ifp->if_slowtimo_ch, if_slowtimo, ifp); 779 if_slowtimo(ifp); 780 } 781 782 if (ifp->if_transmit == NULL || ifp->if_transmit == if_nulltransmit) 783 ifp->if_transmit = if_transmit; 784 785 IFNET_GLOBAL_LOCK(); 786 TAILQ_INSERT_TAIL(&ifnet_list, ifp, if_list); 787 IFNET_WRITER_INSERT_TAIL(ifp); 788 IFNET_GLOBAL_UNLOCK(); 789 } 790 791 /* 792 * The if_percpuq framework 793 * 794 * It allows network device drivers to execute the network stack 795 * in softint (so called softint-based if_input). It utilizes 796 * softint and percpu ifqueue. It doesn't distribute any packets 797 * between CPUs, unlike pktqueue(9). 798 * 799 * Currently we support two options for device drivers to apply the framework: 800 * - Use it implicitly with less changes 801 * - If you use if_attach in driver's _attach function and if_input in 802 * driver's Rx interrupt handler, a packet is queued and a softint handles 803 * the packet implicitly 804 * - Use it explicitly in each driver (recommended) 805 * - You can use if_percpuq_* directly in your driver 806 * - In this case, you need to allocate struct if_percpuq in driver's softc 807 * - See wm(4) as a reference implementation 808 */ 809 810 static void 811 if_percpuq_softint(void *arg) 812 { 813 struct if_percpuq *ipq = arg; 814 struct ifnet *ifp = ipq->ipq_ifp; 815 struct mbuf *m; 816 817 while ((m = if_percpuq_dequeue(ipq)) != NULL) { 818 ifp->if_ipackets++; 819 bpf_mtap(ifp, m, BPF_D_IN); 820 821 ifp->_if_input(ifp, m); 822 } 823 } 824 825 static void 826 if_percpuq_init_ifq(void *p, void *arg __unused, struct cpu_info *ci __unused) 827 { 828 struct ifqueue *const ifq = p; 829 830 memset(ifq, 0, sizeof(*ifq)); 831 ifq->ifq_maxlen = IFQ_MAXLEN; 832 } 833 834 struct if_percpuq * 835 if_percpuq_create(struct ifnet *ifp) 836 { 837 struct if_percpuq *ipq; 838 u_int flags = SOFTINT_NET; 839 840 flags |= if_is_mpsafe(ifp) ? SOFTINT_MPSAFE : 0; 841 842 ipq = kmem_zalloc(sizeof(*ipq), KM_SLEEP); 843 ipq->ipq_ifp = ifp; 844 ipq->ipq_si = softint_establish(flags, if_percpuq_softint, ipq); 845 ipq->ipq_ifqs = percpu_alloc(sizeof(struct ifqueue)); 846 percpu_foreach(ipq->ipq_ifqs, &if_percpuq_init_ifq, NULL); 847 848 sysctl_percpuq_setup(&ifp->if_sysctl_log, ifp->if_xname, ipq); 849 850 return ipq; 851 } 852 853 static struct mbuf * 854 if_percpuq_dequeue(struct if_percpuq *ipq) 855 { 856 struct mbuf *m; 857 struct ifqueue *ifq; 858 int s; 859 860 s = splnet(); 861 ifq = percpu_getref(ipq->ipq_ifqs); 862 IF_DEQUEUE(ifq, m); 863 percpu_putref(ipq->ipq_ifqs); 864 splx(s); 865 866 return m; 867 } 868 869 static void 870 if_percpuq_purge_ifq(void *p, void *arg __unused, struct cpu_info *ci __unused) 871 { 872 struct ifqueue *const ifq = p; 873 874 IF_PURGE(ifq); 875 } 876 877 void 878 if_percpuq_destroy(struct if_percpuq *ipq) 879 { 880 881 /* if_detach may already destroy it */ 882 if (ipq == NULL) 883 return; 884 885 softint_disestablish(ipq->ipq_si); 886 percpu_foreach(ipq->ipq_ifqs, &if_percpuq_purge_ifq, NULL); 887 percpu_free(ipq->ipq_ifqs, sizeof(struct ifqueue)); 888 kmem_free(ipq, sizeof(*ipq)); 889 } 890 891 void 892 if_percpuq_enqueue(struct if_percpuq *ipq, struct mbuf *m) 893 { 894 struct ifqueue *ifq; 895 int s; 896 897 KASSERT(ipq != NULL); 898 899 s = splnet(); 900 ifq = percpu_getref(ipq->ipq_ifqs); 901 if (IF_QFULL(ifq)) { 902 IF_DROP(ifq); 903 percpu_putref(ipq->ipq_ifqs); 904 m_freem(m); 905 goto out; 906 } 907 IF_ENQUEUE(ifq, m); 908 percpu_putref(ipq->ipq_ifqs); 909 910 softint_schedule(ipq->ipq_si); 911 out: 912 splx(s); 913 } 914 915 static void 916 if_percpuq_drops(void *p, void *arg, struct cpu_info *ci __unused) 917 { 918 struct ifqueue *const ifq = p; 919 int *sum = arg; 920 921 *sum += ifq->ifq_drops; 922 } 923 924 static int 925 sysctl_percpuq_drops_handler(SYSCTLFN_ARGS) 926 { 927 struct sysctlnode node; 928 struct if_percpuq *ipq; 929 int sum = 0; 930 int error; 931 932 node = *rnode; 933 ipq = node.sysctl_data; 934 935 percpu_foreach(ipq->ipq_ifqs, if_percpuq_drops, &sum); 936 937 node.sysctl_data = ∑ 938 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 939 if (error != 0 || newp == NULL) 940 return error; 941 942 return 0; 943 } 944 945 static void 946 sysctl_percpuq_setup(struct sysctllog **clog, const char* ifname, 947 struct if_percpuq *ipq) 948 { 949 const struct sysctlnode *cnode, *rnode; 950 951 if (sysctl_createv(clog, 0, NULL, &rnode, 952 CTLFLAG_PERMANENT, 953 CTLTYPE_NODE, "interfaces", 954 SYSCTL_DESCR("Per-interface controls"), 955 NULL, 0, NULL, 0, 956 CTL_NET, CTL_CREATE, CTL_EOL) != 0) 957 goto bad; 958 959 if (sysctl_createv(clog, 0, &rnode, &rnode, 960 CTLFLAG_PERMANENT, 961 CTLTYPE_NODE, ifname, 962 SYSCTL_DESCR("Interface controls"), 963 NULL, 0, NULL, 0, 964 CTL_CREATE, CTL_EOL) != 0) 965 goto bad; 966 967 if (sysctl_createv(clog, 0, &rnode, &rnode, 968 CTLFLAG_PERMANENT, 969 CTLTYPE_NODE, "rcvq", 970 SYSCTL_DESCR("Interface input queue controls"), 971 NULL, 0, NULL, 0, 972 CTL_CREATE, CTL_EOL) != 0) 973 goto bad; 974 975 #ifdef NOTYET 976 /* XXX Should show each per-CPU queue length? */ 977 if (sysctl_createv(clog, 0, &rnode, &rnode, 978 CTLFLAG_PERMANENT, 979 CTLTYPE_INT, "len", 980 SYSCTL_DESCR("Current input queue length"), 981 sysctl_percpuq_len, 0, NULL, 0, 982 CTL_CREATE, CTL_EOL) != 0) 983 goto bad; 984 985 if (sysctl_createv(clog, 0, &rnode, &cnode, 986 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 987 CTLTYPE_INT, "maxlen", 988 SYSCTL_DESCR("Maximum allowed input queue length"), 989 sysctl_percpuq_maxlen_handler, 0, (void *)ipq, 0, 990 CTL_CREATE, CTL_EOL) != 0) 991 goto bad; 992 #endif 993 994 if (sysctl_createv(clog, 0, &rnode, &cnode, 995 CTLFLAG_PERMANENT, 996 CTLTYPE_INT, "drops", 997 SYSCTL_DESCR("Total packets dropped due to full input queue"), 998 sysctl_percpuq_drops_handler, 0, (void *)ipq, 0, 999 CTL_CREATE, CTL_EOL) != 0) 1000 goto bad; 1001 1002 return; 1003 bad: 1004 printf("%s: could not attach sysctl nodes\n", ifname); 1005 return; 1006 } 1007 1008 /* 1009 * The deferred if_start framework 1010 * 1011 * The common APIs to defer if_start to softint when if_start is requested 1012 * from a device driver running in hardware interrupt context. 1013 */ 1014 /* 1015 * Call ifp->if_start (or equivalent) in a dedicated softint for 1016 * deferred if_start. 1017 */ 1018 static void 1019 if_deferred_start_softint(void *arg) 1020 { 1021 struct if_deferred_start *ids = arg; 1022 struct ifnet *ifp = ids->ids_ifp; 1023 1024 ids->ids_if_start(ifp); 1025 } 1026 1027 /* 1028 * The default callback function for deferred if_start. 1029 */ 1030 static void 1031 if_deferred_start_common(struct ifnet *ifp) 1032 { 1033 int s; 1034 1035 s = splnet(); 1036 if_start_lock(ifp); 1037 splx(s); 1038 } 1039 1040 static inline bool 1041 if_snd_is_used(struct ifnet *ifp) 1042 { 1043 1044 return ifp->if_transmit == NULL || ifp->if_transmit == if_nulltransmit || 1045 ifp->if_transmit == if_transmit || 1046 ALTQ_IS_ENABLED(&ifp->if_snd); 1047 } 1048 1049 /* 1050 * Schedule deferred if_start. 1051 */ 1052 void 1053 if_schedule_deferred_start(struct ifnet *ifp) 1054 { 1055 1056 KASSERT(ifp->if_deferred_start != NULL); 1057 1058 if (if_snd_is_used(ifp) && IFQ_IS_EMPTY(&ifp->if_snd)) 1059 return; 1060 1061 softint_schedule(ifp->if_deferred_start->ids_si); 1062 } 1063 1064 /* 1065 * Create an instance of deferred if_start. A driver should call the function 1066 * only if the driver needs deferred if_start. Drivers can setup their own 1067 * deferred if_start function via 2nd argument. 1068 */ 1069 void 1070 if_deferred_start_init(struct ifnet *ifp, void (*func)(struct ifnet *)) 1071 { 1072 struct if_deferred_start *ids; 1073 u_int flags = SOFTINT_NET; 1074 1075 flags |= if_is_mpsafe(ifp) ? SOFTINT_MPSAFE : 0; 1076 1077 ids = kmem_zalloc(sizeof(*ids), KM_SLEEP); 1078 ids->ids_ifp = ifp; 1079 ids->ids_si = softint_establish(flags, if_deferred_start_softint, ids); 1080 if (func != NULL) 1081 ids->ids_if_start = func; 1082 else 1083 ids->ids_if_start = if_deferred_start_common; 1084 1085 ifp->if_deferred_start = ids; 1086 } 1087 1088 static void 1089 if_deferred_start_destroy(struct ifnet *ifp) 1090 { 1091 1092 if (ifp->if_deferred_start == NULL) 1093 return; 1094 1095 softint_disestablish(ifp->if_deferred_start->ids_si); 1096 kmem_free(ifp->if_deferred_start, sizeof(*ifp->if_deferred_start)); 1097 ifp->if_deferred_start = NULL; 1098 } 1099 1100 /* 1101 * The common interface input routine that is called by device drivers, 1102 * which should be used only when the driver's rx handler already runs 1103 * in softint. 1104 */ 1105 void 1106 if_input(struct ifnet *ifp, struct mbuf *m) 1107 { 1108 1109 KASSERT(ifp->if_percpuq == NULL); 1110 KASSERT(!cpu_intr_p()); 1111 1112 ifp->if_ipackets++; 1113 bpf_mtap(ifp, m, BPF_D_IN); 1114 1115 ifp->_if_input(ifp, m); 1116 } 1117 1118 /* 1119 * DEPRECATED. Use if_initialize and if_register instead. 1120 * See the above comment of if_initialize. 1121 * 1122 * Note that it implicitly enables if_percpuq to make drivers easy to 1123 * migrate softint-based if_input without much changes. If you don't 1124 * want to enable it, use if_initialize instead. 1125 */ 1126 int 1127 if_attach(ifnet_t *ifp) 1128 { 1129 int rv; 1130 1131 rv = if_initialize(ifp); 1132 if (rv != 0) 1133 return rv; 1134 1135 ifp->if_percpuq = if_percpuq_create(ifp); 1136 if_register(ifp); 1137 1138 return 0; 1139 } 1140 1141 void 1142 if_attachdomain(void) 1143 { 1144 struct ifnet *ifp; 1145 int s; 1146 int bound = curlwp_bind(); 1147 1148 s = pserialize_read_enter(); 1149 IFNET_READER_FOREACH(ifp) { 1150 struct psref psref; 1151 psref_acquire(&psref, &ifp->if_psref, ifnet_psref_class); 1152 pserialize_read_exit(s); 1153 if_attachdomain1(ifp); 1154 s = pserialize_read_enter(); 1155 psref_release(&psref, &ifp->if_psref, ifnet_psref_class); 1156 } 1157 pserialize_read_exit(s); 1158 curlwp_bindx(bound); 1159 } 1160 1161 static void 1162 if_attachdomain1(struct ifnet *ifp) 1163 { 1164 struct domain *dp; 1165 int s; 1166 1167 s = splsoftnet(); 1168 1169 /* address family dependent data region */ 1170 memset(ifp->if_afdata, 0, sizeof(ifp->if_afdata)); 1171 DOMAIN_FOREACH(dp) { 1172 if (dp->dom_ifattach != NULL) 1173 ifp->if_afdata[dp->dom_family] = 1174 (*dp->dom_ifattach)(ifp); 1175 } 1176 1177 splx(s); 1178 } 1179 1180 /* 1181 * Deactivate an interface. This points all of the procedure 1182 * handles at error stubs. May be called from interrupt context. 1183 */ 1184 void 1185 if_deactivate(struct ifnet *ifp) 1186 { 1187 int s; 1188 1189 s = splsoftnet(); 1190 1191 ifp->if_output = if_nulloutput; 1192 ifp->_if_input = if_nullinput; 1193 ifp->if_start = if_nullstart; 1194 ifp->if_transmit = if_nulltransmit; 1195 ifp->if_ioctl = if_nullioctl; 1196 ifp->if_init = if_nullinit; 1197 ifp->if_stop = if_nullstop; 1198 ifp->if_slowtimo = if_nullslowtimo; 1199 ifp->if_drain = if_nulldrain; 1200 1201 /* No more packets may be enqueued. */ 1202 ifp->if_snd.ifq_maxlen = 0; 1203 1204 splx(s); 1205 } 1206 1207 bool 1208 if_is_deactivated(const struct ifnet *ifp) 1209 { 1210 1211 return ifp->if_output == if_nulloutput; 1212 } 1213 1214 void 1215 if_purgeaddrs(struct ifnet *ifp, int family, void (*purgeaddr)(struct ifaddr *)) 1216 { 1217 struct ifaddr *ifa, *nifa; 1218 int s; 1219 1220 s = pserialize_read_enter(); 1221 for (ifa = IFADDR_READER_FIRST(ifp); ifa; ifa = nifa) { 1222 nifa = IFADDR_READER_NEXT(ifa); 1223 if (ifa->ifa_addr->sa_family != family) 1224 continue; 1225 pserialize_read_exit(s); 1226 1227 (*purgeaddr)(ifa); 1228 1229 s = pserialize_read_enter(); 1230 } 1231 pserialize_read_exit(s); 1232 } 1233 1234 #ifdef IFAREF_DEBUG 1235 static struct ifaddr **ifa_list; 1236 static int ifa_list_size; 1237 1238 /* Depends on only one if_attach runs at once */ 1239 static void 1240 if_build_ifa_list(struct ifnet *ifp) 1241 { 1242 struct ifaddr *ifa; 1243 int i; 1244 1245 KASSERT(ifa_list == NULL); 1246 KASSERT(ifa_list_size == 0); 1247 1248 IFADDR_READER_FOREACH(ifa, ifp) 1249 ifa_list_size++; 1250 1251 ifa_list = kmem_alloc(sizeof(*ifa) * ifa_list_size, KM_SLEEP); 1252 i = 0; 1253 IFADDR_READER_FOREACH(ifa, ifp) { 1254 ifa_list[i++] = ifa; 1255 ifaref(ifa); 1256 } 1257 } 1258 1259 static void 1260 if_check_and_free_ifa_list(struct ifnet *ifp) 1261 { 1262 int i; 1263 struct ifaddr *ifa; 1264 1265 if (ifa_list == NULL) 1266 return; 1267 1268 for (i = 0; i < ifa_list_size; i++) { 1269 char buf[64]; 1270 1271 ifa = ifa_list[i]; 1272 sockaddr_format(ifa->ifa_addr, buf, sizeof(buf)); 1273 if (ifa->ifa_refcnt > 1) { 1274 log(LOG_WARNING, 1275 "ifa(%s) still referenced (refcnt=%d)\n", 1276 buf, ifa->ifa_refcnt - 1); 1277 } else 1278 log(LOG_DEBUG, 1279 "ifa(%s) not referenced (refcnt=%d)\n", 1280 buf, ifa->ifa_refcnt - 1); 1281 ifafree(ifa); 1282 } 1283 1284 kmem_free(ifa_list, sizeof(*ifa) * ifa_list_size); 1285 ifa_list = NULL; 1286 ifa_list_size = 0; 1287 } 1288 #endif 1289 1290 /* 1291 * Detach an interface from the list of "active" interfaces, 1292 * freeing any resources as we go along. 1293 * 1294 * NOTE: This routine must be called with a valid thread context, 1295 * as it may block. 1296 */ 1297 void 1298 if_detach(struct ifnet *ifp) 1299 { 1300 struct socket so; 1301 struct ifaddr *ifa; 1302 #ifdef IFAREF_DEBUG 1303 struct ifaddr *last_ifa = NULL; 1304 #endif 1305 struct domain *dp; 1306 const struct protosw *pr; 1307 int s, i, family, purged; 1308 uint64_t xc; 1309 1310 #ifdef IFAREF_DEBUG 1311 if_build_ifa_list(ifp); 1312 #endif 1313 /* 1314 * XXX It's kind of lame that we have to have the 1315 * XXX socket structure... 1316 */ 1317 memset(&so, 0, sizeof(so)); 1318 1319 s = splnet(); 1320 1321 sysctl_teardown(&ifp->if_sysctl_log); 1322 IFNET_LOCK(ifp); 1323 if_deactivate(ifp); 1324 IFNET_UNLOCK(ifp); 1325 1326 if (ifp->if_slowtimo != NULL && ifp->if_slowtimo_ch != NULL) { 1327 ifp->if_slowtimo = NULL; 1328 callout_halt(ifp->if_slowtimo_ch, NULL); 1329 callout_destroy(ifp->if_slowtimo_ch); 1330 kmem_free(ifp->if_slowtimo_ch, sizeof(*ifp->if_slowtimo_ch)); 1331 } 1332 if_deferred_start_destroy(ifp); 1333 1334 /* 1335 * Do an if_down() to give protocols a chance to do something. 1336 */ 1337 if_down_deactivated(ifp); 1338 1339 #ifdef ALTQ 1340 if (ALTQ_IS_ENABLED(&ifp->if_snd)) 1341 altq_disable(&ifp->if_snd); 1342 if (ALTQ_IS_ATTACHED(&ifp->if_snd)) 1343 altq_detach(&ifp->if_snd); 1344 #endif 1345 1346 #if NCARP > 0 1347 /* Remove the interface from any carp group it is a part of. */ 1348 if (ifp->if_carp != NULL && ifp->if_type != IFT_CARP) 1349 carp_ifdetach(ifp); 1350 #endif 1351 1352 /* 1353 * Rip all the addresses off the interface. This should make 1354 * all of the routes go away. 1355 * 1356 * pr_usrreq calls can remove an arbitrary number of ifaddrs 1357 * from the list, including our "cursor", ifa. For safety, 1358 * and to honor the TAILQ abstraction, I just restart the 1359 * loop after each removal. Note that the loop will exit 1360 * when all of the remaining ifaddrs belong to the AF_LINK 1361 * family. I am counting on the historical fact that at 1362 * least one pr_usrreq in each address domain removes at 1363 * least one ifaddr. 1364 */ 1365 again: 1366 /* 1367 * At this point, no other one tries to remove ifa in the list, 1368 * so we don't need to take a lock or psref. Avoid using 1369 * IFADDR_READER_FOREACH to pass over an inspection of contract 1370 * violations of pserialize. 1371 */ 1372 IFADDR_WRITER_FOREACH(ifa, ifp) { 1373 family = ifa->ifa_addr->sa_family; 1374 #ifdef IFAREF_DEBUG 1375 printf("if_detach: ifaddr %p, family %d, refcnt %d\n", 1376 ifa, family, ifa->ifa_refcnt); 1377 if (last_ifa != NULL && ifa == last_ifa) 1378 panic("if_detach: loop detected"); 1379 last_ifa = ifa; 1380 #endif 1381 if (family == AF_LINK) 1382 continue; 1383 dp = pffinddomain(family); 1384 KASSERTMSG(dp != NULL, "no domain for AF %d", family); 1385 /* 1386 * XXX These PURGEIF calls are redundant with the 1387 * purge-all-families calls below, but are left in for 1388 * now both to make a smaller change, and to avoid 1389 * unplanned interactions with clearing of 1390 * ifp->if_addrlist. 1391 */ 1392 purged = 0; 1393 for (pr = dp->dom_protosw; 1394 pr < dp->dom_protoswNPROTOSW; pr++) { 1395 so.so_proto = pr; 1396 if (pr->pr_usrreqs) { 1397 (void) (*pr->pr_usrreqs->pr_purgeif)(&so, ifp); 1398 purged = 1; 1399 } 1400 } 1401 if (purged == 0) { 1402 /* 1403 * XXX What's really the best thing to do 1404 * XXX here? --thorpej (at) NetBSD.org 1405 */ 1406 printf("if_detach: WARNING: AF %d not purged\n", 1407 family); 1408 ifa_remove(ifp, ifa); 1409 } 1410 goto again; 1411 } 1412 1413 if_free_sadl(ifp, 1); 1414 1415 restart: 1416 IFADDR_WRITER_FOREACH(ifa, ifp) { 1417 family = ifa->ifa_addr->sa_family; 1418 KASSERT(family == AF_LINK); 1419 ifa_remove(ifp, ifa); 1420 goto restart; 1421 } 1422 1423 /* Delete stray routes from the routing table. */ 1424 for (i = 0; i <= AF_MAX; i++) 1425 rt_delete_matched_entries(i, if_delroute_matcher, ifp); 1426 1427 DOMAIN_FOREACH(dp) { 1428 if (dp->dom_ifdetach != NULL && ifp->if_afdata[dp->dom_family]) 1429 { 1430 void *p = ifp->if_afdata[dp->dom_family]; 1431 if (p) { 1432 ifp->if_afdata[dp->dom_family] = NULL; 1433 (*dp->dom_ifdetach)(ifp, p); 1434 } 1435 } 1436 1437 /* 1438 * One would expect multicast memberships (INET and 1439 * INET6) on UDP sockets to be purged by the PURGEIF 1440 * calls above, but if all addresses were removed from 1441 * the interface prior to destruction, the calls will 1442 * not be made (e.g. ppp, for which pppd(8) generally 1443 * removes addresses before destroying the interface). 1444 * Because there is no invariant that multicast 1445 * memberships only exist for interfaces with IPv4 1446 * addresses, we must call PURGEIF regardless of 1447 * addresses. (Protocols which might store ifnet 1448 * pointers are marked with PR_PURGEIF.) 1449 */ 1450 for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) { 1451 so.so_proto = pr; 1452 if (pr->pr_usrreqs && pr->pr_flags & PR_PURGEIF) 1453 (void)(*pr->pr_usrreqs->pr_purgeif)(&so, ifp); 1454 } 1455 } 1456 1457 /* Wait for all readers to drain before freeing. */ 1458 IFNET_GLOBAL_LOCK(); 1459 ifindex2ifnet[ifp->if_index] = NULL; 1460 TAILQ_REMOVE(&ifnet_list, ifp, if_list); 1461 IFNET_WRITER_REMOVE(ifp); 1462 pserialize_perform(ifnet_psz); 1463 IFNET_GLOBAL_UNLOCK(); 1464 1465 psref_target_destroy(&ifp->if_psref, ifnet_psref_class); 1466 PSLIST_ENTRY_DESTROY(ifp, if_pslist_entry); 1467 1468 pfil_run_ifhooks(if_pfil, PFIL_IFNET_DETACH, ifp); 1469 (void)pfil_head_destroy(ifp->if_pfil); 1470 1471 /* Announce that the interface is gone. */ 1472 rt_ifannouncemsg(ifp, IFAN_DEPARTURE); 1473 1474 IF_AFDATA_LOCK_DESTROY(ifp); 1475 1476 if (if_is_link_state_changeable(ifp)) { 1477 softint_disestablish(ifp->if_link_si); 1478 ifp->if_link_si = NULL; 1479 } 1480 1481 /* 1482 * remove packets that came from ifp, from software interrupt queues. 1483 */ 1484 DOMAIN_FOREACH(dp) { 1485 for (i = 0; i < __arraycount(dp->dom_ifqueues); i++) { 1486 struct ifqueue *iq = dp->dom_ifqueues[i]; 1487 if (iq == NULL) 1488 break; 1489 dp->dom_ifqueues[i] = NULL; 1490 if_detach_queues(ifp, iq); 1491 } 1492 } 1493 1494 /* 1495 * IP queues have to be processed separately: net-queue barrier 1496 * ensures that the packets are dequeued while a cross-call will 1497 * ensure that the interrupts have completed. FIXME: not quite.. 1498 */ 1499 #ifdef INET 1500 pktq_barrier(ip_pktq); 1501 #endif 1502 #ifdef INET6 1503 if (in6_present) 1504 pktq_barrier(ip6_pktq); 1505 #endif 1506 xc = xc_broadcast(0, (xcfunc_t)nullop, NULL, NULL); 1507 xc_wait(xc); 1508 1509 if (ifp->if_percpuq != NULL) { 1510 if_percpuq_destroy(ifp->if_percpuq); 1511 ifp->if_percpuq = NULL; 1512 } 1513 1514 mutex_obj_free(ifp->if_ioctl_lock); 1515 ifp->if_ioctl_lock = NULL; 1516 mutex_obj_free(ifp->if_snd.ifq_lock); 1517 1518 splx(s); 1519 1520 #ifdef IFAREF_DEBUG 1521 if_check_and_free_ifa_list(ifp); 1522 #endif 1523 } 1524 1525 static void 1526 if_detach_queues(struct ifnet *ifp, struct ifqueue *q) 1527 { 1528 struct mbuf *m, *prev, *next; 1529 1530 prev = NULL; 1531 for (m = q->ifq_head; m != NULL; m = next) { 1532 KASSERT((m->m_flags & M_PKTHDR) != 0); 1533 1534 next = m->m_nextpkt; 1535 if (m->m_pkthdr.rcvif_index != ifp->if_index) { 1536 prev = m; 1537 continue; 1538 } 1539 1540 if (prev != NULL) 1541 prev->m_nextpkt = m->m_nextpkt; 1542 else 1543 q->ifq_head = m->m_nextpkt; 1544 if (q->ifq_tail == m) 1545 q->ifq_tail = prev; 1546 q->ifq_len--; 1547 1548 m->m_nextpkt = NULL; 1549 m_freem(m); 1550 IF_DROP(q); 1551 } 1552 } 1553 1554 /* 1555 * Callback for a radix tree walk to delete all references to an 1556 * ifnet. 1557 */ 1558 static int 1559 if_delroute_matcher(struct rtentry *rt, void *v) 1560 { 1561 struct ifnet *ifp = (struct ifnet *)v; 1562 1563 if (rt->rt_ifp == ifp) 1564 return 1; 1565 else 1566 return 0; 1567 } 1568 1569 /* 1570 * Create a clone network interface. 1571 */ 1572 static int 1573 if_clone_create(const char *name) 1574 { 1575 struct if_clone *ifc; 1576 int unit; 1577 struct ifnet *ifp; 1578 struct psref psref; 1579 1580 KASSERT(mutex_owned(&if_clone_mtx)); 1581 1582 ifc = if_clone_lookup(name, &unit); 1583 if (ifc == NULL) 1584 return EINVAL; 1585 1586 ifp = if_get(name, &psref); 1587 if (ifp != NULL) { 1588 if_put(ifp, &psref); 1589 return EEXIST; 1590 } 1591 1592 return (*ifc->ifc_create)(ifc, unit); 1593 } 1594 1595 /* 1596 * Destroy a clone network interface. 1597 */ 1598 static int 1599 if_clone_destroy(const char *name) 1600 { 1601 struct if_clone *ifc; 1602 struct ifnet *ifp; 1603 struct psref psref; 1604 1605 KASSERT(mutex_owned(&if_clone_mtx)); 1606 1607 ifc = if_clone_lookup(name, NULL); 1608 if (ifc == NULL) 1609 return EINVAL; 1610 1611 if (ifc->ifc_destroy == NULL) 1612 return EOPNOTSUPP; 1613 1614 ifp = if_get(name, &psref); 1615 if (ifp == NULL) 1616 return ENXIO; 1617 1618 /* We have to disable ioctls here */ 1619 IFNET_LOCK(ifp); 1620 ifp->if_ioctl = if_nullioctl; 1621 IFNET_UNLOCK(ifp); 1622 1623 /* 1624 * We cannot call ifc_destroy with holding ifp. 1625 * Releasing ifp here is safe thanks to if_clone_mtx. 1626 */ 1627 if_put(ifp, &psref); 1628 1629 return (*ifc->ifc_destroy)(ifp); 1630 } 1631 1632 static bool 1633 if_is_unit(const char *name) 1634 { 1635 1636 while(*name != '\0') { 1637 if (*name < '0' || *name > '9') 1638 return false; 1639 name++; 1640 } 1641 1642 return true; 1643 } 1644 1645 /* 1646 * Look up a network interface cloner. 1647 */ 1648 static struct if_clone * 1649 if_clone_lookup(const char *name, int *unitp) 1650 { 1651 struct if_clone *ifc; 1652 const char *cp; 1653 char *dp, ifname[IFNAMSIZ + 3]; 1654 int unit; 1655 1656 KASSERT(mutex_owned(&if_clone_mtx)); 1657 1658 strcpy(ifname, "if_"); 1659 /* separate interface name from unit */ 1660 /* TODO: search unit number from backward */ 1661 for (dp = ifname + 3, cp = name; cp - name < IFNAMSIZ && 1662 *cp && !if_is_unit(cp);) 1663 *dp++ = *cp++; 1664 1665 if (cp == name || cp - name == IFNAMSIZ || !*cp) 1666 return NULL; /* No name or unit number */ 1667 *dp++ = '\0'; 1668 1669 again: 1670 LIST_FOREACH(ifc, &if_cloners, ifc_list) { 1671 if (strcmp(ifname + 3, ifc->ifc_name) == 0) 1672 break; 1673 } 1674 1675 if (ifc == NULL) { 1676 int error; 1677 if (*ifname == '\0') 1678 return NULL; 1679 mutex_exit(&if_clone_mtx); 1680 error = module_autoload(ifname, MODULE_CLASS_DRIVER); 1681 mutex_enter(&if_clone_mtx); 1682 if (error) 1683 return NULL; 1684 *ifname = '\0'; 1685 goto again; 1686 } 1687 1688 unit = 0; 1689 while (cp - name < IFNAMSIZ && *cp) { 1690 if (*cp < '0' || *cp > '9' || unit >= INT_MAX / 10) { 1691 /* Bogus unit number. */ 1692 return NULL; 1693 } 1694 unit = (unit * 10) + (*cp++ - '0'); 1695 } 1696 1697 if (unitp != NULL) 1698 *unitp = unit; 1699 return ifc; 1700 } 1701 1702 /* 1703 * Register a network interface cloner. 1704 */ 1705 void 1706 if_clone_attach(struct if_clone *ifc) 1707 { 1708 1709 mutex_enter(&if_clone_mtx); 1710 LIST_INSERT_HEAD(&if_cloners, ifc, ifc_list); 1711 if_cloners_count++; 1712 mutex_exit(&if_clone_mtx); 1713 } 1714 1715 /* 1716 * Unregister a network interface cloner. 1717 */ 1718 void 1719 if_clone_detach(struct if_clone *ifc) 1720 { 1721 1722 mutex_enter(&if_clone_mtx); 1723 LIST_REMOVE(ifc, ifc_list); 1724 if_cloners_count--; 1725 mutex_exit(&if_clone_mtx); 1726 } 1727 1728 /* 1729 * Provide list of interface cloners to userspace. 1730 */ 1731 int 1732 if_clone_list(int buf_count, char *buffer, int *total) 1733 { 1734 char outbuf[IFNAMSIZ], *dst; 1735 struct if_clone *ifc; 1736 int count, error = 0; 1737 1738 mutex_enter(&if_clone_mtx); 1739 *total = if_cloners_count; 1740 if ((dst = buffer) == NULL) { 1741 /* Just asking how many there are. */ 1742 goto out; 1743 } 1744 1745 if (buf_count < 0) { 1746 error = EINVAL; 1747 goto out; 1748 } 1749 1750 count = (if_cloners_count < buf_count) ? 1751 if_cloners_count : buf_count; 1752 1753 for (ifc = LIST_FIRST(&if_cloners); ifc != NULL && count != 0; 1754 ifc = LIST_NEXT(ifc, ifc_list), count--, dst += IFNAMSIZ) { 1755 (void)strncpy(outbuf, ifc->ifc_name, sizeof(outbuf)); 1756 if (outbuf[sizeof(outbuf) - 1] != '\0') { 1757 error = ENAMETOOLONG; 1758 goto out; 1759 } 1760 error = copyout(outbuf, dst, sizeof(outbuf)); 1761 if (error != 0) 1762 break; 1763 } 1764 1765 out: 1766 mutex_exit(&if_clone_mtx); 1767 return error; 1768 } 1769 1770 void 1771 ifa_psref_init(struct ifaddr *ifa) 1772 { 1773 1774 psref_target_init(&ifa->ifa_psref, ifa_psref_class); 1775 } 1776 1777 void 1778 ifaref(struct ifaddr *ifa) 1779 { 1780 KASSERT(!ISSET(ifa->ifa_flags, IFA_DESTROYING)); 1781 ifa->ifa_refcnt++; 1782 } 1783 1784 void 1785 ifafree(struct ifaddr *ifa) 1786 { 1787 KASSERT(ifa != NULL); 1788 KASSERT(ifa->ifa_refcnt > 0); 1789 1790 if (--ifa->ifa_refcnt == 0) { 1791 free(ifa, M_IFADDR); 1792 } 1793 } 1794 1795 bool 1796 ifa_is_destroying(struct ifaddr *ifa) 1797 { 1798 1799 return ISSET(ifa->ifa_flags, IFA_DESTROYING); 1800 } 1801 1802 void 1803 ifa_insert(struct ifnet *ifp, struct ifaddr *ifa) 1804 { 1805 1806 ifa->ifa_ifp = ifp; 1807 1808 /* 1809 * Check MP-safety for IFEF_MPSAFE drivers. 1810 * Check !IFF_RUNNING for initialization routines that normally don't 1811 * take IFNET_LOCK but it's safe because there is no competitor. 1812 * XXX there are false positive cases because IFF_RUNNING can be off on 1813 * if_stop. 1814 */ 1815 KASSERT(!if_is_mpsafe(ifp) || !ISSET(ifp->if_flags, IFF_RUNNING) || 1816 IFNET_LOCKED(ifp)); 1817 1818 TAILQ_INSERT_TAIL(&ifp->if_addrlist, ifa, ifa_list); 1819 IFADDR_ENTRY_INIT(ifa); 1820 IFADDR_WRITER_INSERT_TAIL(ifp, ifa); 1821 1822 ifaref(ifa); 1823 } 1824 1825 void 1826 ifa_remove(struct ifnet *ifp, struct ifaddr *ifa) 1827 { 1828 1829 KASSERT(ifa->ifa_ifp == ifp); 1830 /* 1831 * Check MP-safety for IFEF_MPSAFE drivers. 1832 * if_is_deactivated indicates ifa_remove is called form if_detach 1833 * where is safe even if IFNET_LOCK isn't held. 1834 */ 1835 KASSERT(!if_is_mpsafe(ifp) || if_is_deactivated(ifp) || IFNET_LOCKED(ifp)); 1836 1837 TAILQ_REMOVE(&ifp->if_addrlist, ifa, ifa_list); 1838 IFADDR_WRITER_REMOVE(ifa); 1839 #ifdef NET_MPSAFE 1840 IFNET_GLOBAL_LOCK(); 1841 pserialize_perform(ifnet_psz); 1842 IFNET_GLOBAL_UNLOCK(); 1843 #endif 1844 1845 #ifdef NET_MPSAFE 1846 psref_target_destroy(&ifa->ifa_psref, ifa_psref_class); 1847 #endif 1848 IFADDR_ENTRY_DESTROY(ifa); 1849 ifafree(ifa); 1850 } 1851 1852 void 1853 ifa_acquire(struct ifaddr *ifa, struct psref *psref) 1854 { 1855 1856 psref_acquire(psref, &ifa->ifa_psref, ifa_psref_class); 1857 } 1858 1859 void 1860 ifa_release(struct ifaddr *ifa, struct psref *psref) 1861 { 1862 1863 if (ifa == NULL) 1864 return; 1865 1866 psref_release(psref, &ifa->ifa_psref, ifa_psref_class); 1867 } 1868 1869 bool 1870 ifa_held(struct ifaddr *ifa) 1871 { 1872 1873 return psref_held(&ifa->ifa_psref, ifa_psref_class); 1874 } 1875 1876 static inline int 1877 equal(const struct sockaddr *sa1, const struct sockaddr *sa2) 1878 { 1879 return sockaddr_cmp(sa1, sa2) == 0; 1880 } 1881 1882 /* 1883 * Locate an interface based on a complete address. 1884 */ 1885 /*ARGSUSED*/ 1886 struct ifaddr * 1887 ifa_ifwithaddr(const struct sockaddr *addr) 1888 { 1889 struct ifnet *ifp; 1890 struct ifaddr *ifa; 1891 1892 IFNET_READER_FOREACH(ifp) { 1893 if (if_is_deactivated(ifp)) 1894 continue; 1895 IFADDR_READER_FOREACH(ifa, ifp) { 1896 if (ifa->ifa_addr->sa_family != addr->sa_family) 1897 continue; 1898 if (equal(addr, ifa->ifa_addr)) 1899 return ifa; 1900 if ((ifp->if_flags & IFF_BROADCAST) && 1901 ifa->ifa_broadaddr && 1902 /* IP6 doesn't have broadcast */ 1903 ifa->ifa_broadaddr->sa_len != 0 && 1904 equal(ifa->ifa_broadaddr, addr)) 1905 return ifa; 1906 } 1907 } 1908 return NULL; 1909 } 1910 1911 struct ifaddr * 1912 ifa_ifwithaddr_psref(const struct sockaddr *addr, struct psref *psref) 1913 { 1914 struct ifaddr *ifa; 1915 int s = pserialize_read_enter(); 1916 1917 ifa = ifa_ifwithaddr(addr); 1918 if (ifa != NULL) 1919 ifa_acquire(ifa, psref); 1920 pserialize_read_exit(s); 1921 1922 return ifa; 1923 } 1924 1925 /* 1926 * Locate the point to point interface with a given destination address. 1927 */ 1928 /*ARGSUSED*/ 1929 struct ifaddr * 1930 ifa_ifwithdstaddr(const struct sockaddr *addr) 1931 { 1932 struct ifnet *ifp; 1933 struct ifaddr *ifa; 1934 1935 IFNET_READER_FOREACH(ifp) { 1936 if (if_is_deactivated(ifp)) 1937 continue; 1938 if ((ifp->if_flags & IFF_POINTOPOINT) == 0) 1939 continue; 1940 IFADDR_READER_FOREACH(ifa, ifp) { 1941 if (ifa->ifa_addr->sa_family != addr->sa_family || 1942 ifa->ifa_dstaddr == NULL) 1943 continue; 1944 if (equal(addr, ifa->ifa_dstaddr)) 1945 return ifa; 1946 } 1947 } 1948 1949 return NULL; 1950 } 1951 1952 struct ifaddr * 1953 ifa_ifwithdstaddr_psref(const struct sockaddr *addr, struct psref *psref) 1954 { 1955 struct ifaddr *ifa; 1956 int s; 1957 1958 s = pserialize_read_enter(); 1959 ifa = ifa_ifwithdstaddr(addr); 1960 if (ifa != NULL) 1961 ifa_acquire(ifa, psref); 1962 pserialize_read_exit(s); 1963 1964 return ifa; 1965 } 1966 1967 /* 1968 * Find an interface on a specific network. If many, choice 1969 * is most specific found. 1970 */ 1971 struct ifaddr * 1972 ifa_ifwithnet(const struct sockaddr *addr) 1973 { 1974 struct ifnet *ifp; 1975 struct ifaddr *ifa, *ifa_maybe = NULL; 1976 const struct sockaddr_dl *sdl; 1977 u_int af = addr->sa_family; 1978 const char *addr_data = addr->sa_data, *cplim; 1979 1980 if (af == AF_LINK) { 1981 sdl = satocsdl(addr); 1982 if (sdl->sdl_index && sdl->sdl_index < if_indexlim && 1983 ifindex2ifnet[sdl->sdl_index] && 1984 !if_is_deactivated(ifindex2ifnet[sdl->sdl_index])) { 1985 return ifindex2ifnet[sdl->sdl_index]->if_dl; 1986 } 1987 } 1988 #ifdef NETATALK 1989 if (af == AF_APPLETALK) { 1990 const struct sockaddr_at *sat, *sat2; 1991 sat = (const struct sockaddr_at *)addr; 1992 IFNET_READER_FOREACH(ifp) { 1993 if (if_is_deactivated(ifp)) 1994 continue; 1995 ifa = at_ifawithnet((const struct sockaddr_at *)addr, ifp); 1996 if (ifa == NULL) 1997 continue; 1998 sat2 = (struct sockaddr_at *)ifa->ifa_addr; 1999 if (sat2->sat_addr.s_net == sat->sat_addr.s_net) 2000 return ifa; /* exact match */ 2001 if (ifa_maybe == NULL) { 2002 /* else keep the if with the right range */ 2003 ifa_maybe = ifa; 2004 } 2005 } 2006 return ifa_maybe; 2007 } 2008 #endif 2009 IFNET_READER_FOREACH(ifp) { 2010 if (if_is_deactivated(ifp)) 2011 continue; 2012 IFADDR_READER_FOREACH(ifa, ifp) { 2013 const char *cp, *cp2, *cp3; 2014 2015 if (ifa->ifa_addr->sa_family != af || 2016 ifa->ifa_netmask == NULL) 2017 next: continue; 2018 cp = addr_data; 2019 cp2 = ifa->ifa_addr->sa_data; 2020 cp3 = ifa->ifa_netmask->sa_data; 2021 cplim = (const char *)ifa->ifa_netmask + 2022 ifa->ifa_netmask->sa_len; 2023 while (cp3 < cplim) { 2024 if ((*cp++ ^ *cp2++) & *cp3++) { 2025 /* want to continue for() loop */ 2026 goto next; 2027 } 2028 } 2029 if (ifa_maybe == NULL || 2030 rt_refines(ifa->ifa_netmask, 2031 ifa_maybe->ifa_netmask)) 2032 ifa_maybe = ifa; 2033 } 2034 } 2035 return ifa_maybe; 2036 } 2037 2038 struct ifaddr * 2039 ifa_ifwithnet_psref(const struct sockaddr *addr, struct psref *psref) 2040 { 2041 struct ifaddr *ifa; 2042 int s; 2043 2044 s = pserialize_read_enter(); 2045 ifa = ifa_ifwithnet(addr); 2046 if (ifa != NULL) 2047 ifa_acquire(ifa, psref); 2048 pserialize_read_exit(s); 2049 2050 return ifa; 2051 } 2052 2053 /* 2054 * Find the interface of the addresss. 2055 */ 2056 struct ifaddr * 2057 ifa_ifwithladdr(const struct sockaddr *addr) 2058 { 2059 struct ifaddr *ia; 2060 2061 if ((ia = ifa_ifwithaddr(addr)) || (ia = ifa_ifwithdstaddr(addr)) || 2062 (ia = ifa_ifwithnet(addr))) 2063 return ia; 2064 return NULL; 2065 } 2066 2067 struct ifaddr * 2068 ifa_ifwithladdr_psref(const struct sockaddr *addr, struct psref *psref) 2069 { 2070 struct ifaddr *ifa; 2071 int s; 2072 2073 s = pserialize_read_enter(); 2074 ifa = ifa_ifwithladdr(addr); 2075 if (ifa != NULL) 2076 ifa_acquire(ifa, psref); 2077 pserialize_read_exit(s); 2078 2079 return ifa; 2080 } 2081 2082 /* 2083 * Find an interface using a specific address family 2084 */ 2085 struct ifaddr * 2086 ifa_ifwithaf(int af) 2087 { 2088 struct ifnet *ifp; 2089 struct ifaddr *ifa = NULL; 2090 int s; 2091 2092 s = pserialize_read_enter(); 2093 IFNET_READER_FOREACH(ifp) { 2094 if (if_is_deactivated(ifp)) 2095 continue; 2096 IFADDR_READER_FOREACH(ifa, ifp) { 2097 if (ifa->ifa_addr->sa_family == af) 2098 goto out; 2099 } 2100 } 2101 out: 2102 pserialize_read_exit(s); 2103 return ifa; 2104 } 2105 2106 /* 2107 * Find an interface address specific to an interface best matching 2108 * a given address. 2109 */ 2110 struct ifaddr * 2111 ifaof_ifpforaddr(const struct sockaddr *addr, struct ifnet *ifp) 2112 { 2113 struct ifaddr *ifa; 2114 const char *cp, *cp2, *cp3; 2115 const char *cplim; 2116 struct ifaddr *ifa_maybe = 0; 2117 u_int af = addr->sa_family; 2118 2119 if (if_is_deactivated(ifp)) 2120 return NULL; 2121 2122 if (af >= AF_MAX) 2123 return NULL; 2124 2125 IFADDR_READER_FOREACH(ifa, ifp) { 2126 if (ifa->ifa_addr->sa_family != af) 2127 continue; 2128 ifa_maybe = ifa; 2129 if (ifa->ifa_netmask == NULL) { 2130 if (equal(addr, ifa->ifa_addr) || 2131 (ifa->ifa_dstaddr && 2132 equal(addr, ifa->ifa_dstaddr))) 2133 return ifa; 2134 continue; 2135 } 2136 cp = addr->sa_data; 2137 cp2 = ifa->ifa_addr->sa_data; 2138 cp3 = ifa->ifa_netmask->sa_data; 2139 cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask; 2140 for (; cp3 < cplim; cp3++) { 2141 if ((*cp++ ^ *cp2++) & *cp3) 2142 break; 2143 } 2144 if (cp3 == cplim) 2145 return ifa; 2146 } 2147 return ifa_maybe; 2148 } 2149 2150 struct ifaddr * 2151 ifaof_ifpforaddr_psref(const struct sockaddr *addr, struct ifnet *ifp, 2152 struct psref *psref) 2153 { 2154 struct ifaddr *ifa; 2155 int s; 2156 2157 s = pserialize_read_enter(); 2158 ifa = ifaof_ifpforaddr(addr, ifp); 2159 if (ifa != NULL) 2160 ifa_acquire(ifa, psref); 2161 pserialize_read_exit(s); 2162 2163 return ifa; 2164 } 2165 2166 /* 2167 * Default action when installing a route with a Link Level gateway. 2168 * Lookup an appropriate real ifa to point to. 2169 * This should be moved to /sys/net/link.c eventually. 2170 */ 2171 void 2172 link_rtrequest(int cmd, struct rtentry *rt, const struct rt_addrinfo *info) 2173 { 2174 struct ifaddr *ifa; 2175 const struct sockaddr *dst; 2176 struct ifnet *ifp; 2177 struct psref psref; 2178 2179 if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL || 2180 (ifp = ifa->ifa_ifp) == NULL || (dst = rt_getkey(rt)) == NULL) 2181 return; 2182 if ((ifa = ifaof_ifpforaddr_psref(dst, ifp, &psref)) != NULL) { 2183 rt_replace_ifa(rt, ifa); 2184 if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest) 2185 ifa->ifa_rtrequest(cmd, rt, info); 2186 ifa_release(ifa, &psref); 2187 } 2188 } 2189 2190 /* 2191 * bitmask macros to manage a densely packed link_state change queue. 2192 * Because we need to store LINK_STATE_UNKNOWN(0), LINK_STATE_DOWN(1) and 2193 * LINK_STATE_UP(2) we need 2 bits for each state change. 2194 * As a state change to store is 0, treat all bits set as an unset item. 2195 */ 2196 #define LQ_ITEM_BITS 2 2197 #define LQ_ITEM_MASK ((1 << LQ_ITEM_BITS) - 1) 2198 #define LQ_MASK(i) (LQ_ITEM_MASK << (i) * LQ_ITEM_BITS) 2199 #define LINK_STATE_UNSET LQ_ITEM_MASK 2200 #define LQ_ITEM(q, i) (((q) & LQ_MASK((i))) >> (i) * LQ_ITEM_BITS) 2201 #define LQ_STORE(q, i, v) \ 2202 do { \ 2203 (q) &= ~LQ_MASK((i)); \ 2204 (q) |= (v) << (i) * LQ_ITEM_BITS; \ 2205 } while (0 /* CONSTCOND */) 2206 #define LQ_MAX(q) ((sizeof((q)) * NBBY) / LQ_ITEM_BITS) 2207 #define LQ_POP(q, v) \ 2208 do { \ 2209 (v) = LQ_ITEM((q), 0); \ 2210 (q) >>= LQ_ITEM_BITS; \ 2211 (q) |= LINK_STATE_UNSET << (LQ_MAX((q)) - 1) * LQ_ITEM_BITS; \ 2212 } while (0 /* CONSTCOND */) 2213 #define LQ_PUSH(q, v) \ 2214 do { \ 2215 (q) >>= LQ_ITEM_BITS; \ 2216 (q) |= (v) << (LQ_MAX((q)) - 1) * LQ_ITEM_BITS; \ 2217 } while (0 /* CONSTCOND */) 2218 #define LQ_FIND_UNSET(q, i) \ 2219 for ((i) = 0; i < LQ_MAX((q)); (i)++) { \ 2220 if (LQ_ITEM((q), (i)) == LINK_STATE_UNSET) \ 2221 break; \ 2222 } 2223 2224 /* 2225 * XXX reusing (ifp)->if_snd->ifq_lock rather than having another spin mutex 2226 * for each ifnet. It doesn't matter because: 2227 * - if IFEF_MPSAFE is enabled, if_snd isn't used and lock contentions on 2228 * ifq_lock don't happen 2229 * - if IFEF_MPSAFE is disabled, there is no lock contention on ifq_lock 2230 * because if_snd, if_link_state_change and if_link_state_change_softint 2231 * are all called with KERNEL_LOCK 2232 */ 2233 #define IF_LINK_STATE_CHANGE_LOCK(ifp) \ 2234 mutex_enter((ifp)->if_snd.ifq_lock) 2235 #define IF_LINK_STATE_CHANGE_UNLOCK(ifp) \ 2236 mutex_exit((ifp)->if_snd.ifq_lock) 2237 2238 /* 2239 * Handle a change in the interface link state and 2240 * queue notifications. 2241 */ 2242 void 2243 if_link_state_change(struct ifnet *ifp, int link_state) 2244 { 2245 int idx; 2246 2247 KASSERTMSG(if_is_link_state_changeable(ifp), 2248 "%s: IFEF_NO_LINK_STATE_CHANGE must not be set, but if_extflags=0x%x", 2249 ifp->if_xname, ifp->if_extflags); 2250 2251 /* Ensure change is to a valid state */ 2252 switch (link_state) { 2253 case LINK_STATE_UNKNOWN: /* FALLTHROUGH */ 2254 case LINK_STATE_DOWN: /* FALLTHROUGH */ 2255 case LINK_STATE_UP: 2256 break; 2257 default: 2258 #ifdef DEBUG 2259 printf("%s: invalid link state %d\n", 2260 ifp->if_xname, link_state); 2261 #endif 2262 return; 2263 } 2264 2265 IF_LINK_STATE_CHANGE_LOCK(ifp); 2266 2267 /* Find the last unset event in the queue. */ 2268 LQ_FIND_UNSET(ifp->if_link_queue, idx); 2269 2270 /* 2271 * Ensure link_state doesn't match the last event in the queue. 2272 * ifp->if_link_state is not checked and set here because 2273 * that would present an inconsistent picture to the system. 2274 */ 2275 if (idx != 0 && 2276 LQ_ITEM(ifp->if_link_queue, idx - 1) == (uint8_t)link_state) 2277 goto out; 2278 2279 /* Handle queue overflow. */ 2280 if (idx == LQ_MAX(ifp->if_link_queue)) { 2281 uint8_t lost; 2282 2283 /* 2284 * The DOWN state must be protected from being pushed off 2285 * the queue to ensure that userland will always be 2286 * in a sane state. 2287 * Because DOWN is protected, there is no need to protect 2288 * UNKNOWN. 2289 * It should be invalid to change from any other state to 2290 * UNKNOWN anyway ... 2291 */ 2292 lost = LQ_ITEM(ifp->if_link_queue, 0); 2293 LQ_PUSH(ifp->if_link_queue, (uint8_t)link_state); 2294 if (lost == LINK_STATE_DOWN) { 2295 lost = LQ_ITEM(ifp->if_link_queue, 0); 2296 LQ_STORE(ifp->if_link_queue, 0, LINK_STATE_DOWN); 2297 } 2298 printf("%s: lost link state change %s\n", 2299 ifp->if_xname, 2300 lost == LINK_STATE_UP ? "UP" : 2301 lost == LINK_STATE_DOWN ? "DOWN" : 2302 "UNKNOWN"); 2303 } else 2304 LQ_STORE(ifp->if_link_queue, idx, (uint8_t)link_state); 2305 2306 softint_schedule(ifp->if_link_si); 2307 2308 out: 2309 IF_LINK_STATE_CHANGE_UNLOCK(ifp); 2310 } 2311 2312 /* 2313 * Handle interface link state change notifications. 2314 */ 2315 void 2316 if_link_state_change_softint(struct ifnet *ifp, int link_state) 2317 { 2318 struct domain *dp; 2319 int s = splnet(); 2320 bool notify; 2321 2322 KASSERT(!cpu_intr_p()); 2323 2324 IF_LINK_STATE_CHANGE_LOCK(ifp); 2325 2326 /* Ensure the change is still valid. */ 2327 if (ifp->if_link_state == link_state) { 2328 IF_LINK_STATE_CHANGE_UNLOCK(ifp); 2329 return; 2330 } 2331 2332 #ifdef DEBUG 2333 log(LOG_DEBUG, "%s: link state %s (was %s)\n", ifp->if_xname, 2334 link_state == LINK_STATE_UP ? "UP" : 2335 link_state == LINK_STATE_DOWN ? "DOWN" : 2336 "UNKNOWN", 2337 ifp->if_link_state == LINK_STATE_UP ? "UP" : 2338 ifp->if_link_state == LINK_STATE_DOWN ? "DOWN" : 2339 "UNKNOWN"); 2340 #endif 2341 2342 /* 2343 * When going from UNKNOWN to UP, we need to mark existing 2344 * addresses as tentative and restart DAD as we may have 2345 * erroneously not found a duplicate. 2346 * 2347 * This needs to happen before rt_ifmsg to avoid a race where 2348 * listeners would have an address and expect it to work right 2349 * away. 2350 */ 2351 notify = (link_state == LINK_STATE_UP && 2352 ifp->if_link_state == LINK_STATE_UNKNOWN); 2353 ifp->if_link_state = link_state; 2354 /* The following routines may sleep so release the spin mutex */ 2355 IF_LINK_STATE_CHANGE_UNLOCK(ifp); 2356 2357 KERNEL_LOCK_UNLESS_NET_MPSAFE(); 2358 if (notify) { 2359 DOMAIN_FOREACH(dp) { 2360 if (dp->dom_if_link_state_change != NULL) 2361 dp->dom_if_link_state_change(ifp, 2362 LINK_STATE_DOWN); 2363 } 2364 } 2365 2366 /* Notify that the link state has changed. */ 2367 rt_ifmsg(ifp); 2368 2369 #if NCARP > 0 2370 if (ifp->if_carp) 2371 carp_carpdev_state(ifp); 2372 #endif 2373 2374 DOMAIN_FOREACH(dp) { 2375 if (dp->dom_if_link_state_change != NULL) 2376 dp->dom_if_link_state_change(ifp, link_state); 2377 } 2378 KERNEL_UNLOCK_UNLESS_NET_MPSAFE(); 2379 splx(s); 2380 } 2381 2382 /* 2383 * Process the interface link state change queue. 2384 */ 2385 static void 2386 if_link_state_change_si(void *arg) 2387 { 2388 struct ifnet *ifp = arg; 2389 int s; 2390 uint8_t state; 2391 bool schedule; 2392 2393 SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE(); 2394 s = splnet(); 2395 2396 /* Pop a link state change from the queue and process it. */ 2397 IF_LINK_STATE_CHANGE_LOCK(ifp); 2398 LQ_POP(ifp->if_link_queue, state); 2399 IF_LINK_STATE_CHANGE_UNLOCK(ifp); 2400 2401 if_link_state_change_softint(ifp, state); 2402 2403 /* If there is a link state change to come, schedule it. */ 2404 IF_LINK_STATE_CHANGE_LOCK(ifp); 2405 schedule = (LQ_ITEM(ifp->if_link_queue, 0) != LINK_STATE_UNSET); 2406 IF_LINK_STATE_CHANGE_UNLOCK(ifp); 2407 if (schedule) 2408 softint_schedule(ifp->if_link_si); 2409 2410 splx(s); 2411 SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE(); 2412 } 2413 2414 /* 2415 * Default action when installing a local route on a point-to-point 2416 * interface. 2417 */ 2418 void 2419 p2p_rtrequest(int req, struct rtentry *rt, 2420 __unused const struct rt_addrinfo *info) 2421 { 2422 struct ifnet *ifp = rt->rt_ifp; 2423 struct ifaddr *ifa, *lo0ifa; 2424 int s = pserialize_read_enter(); 2425 2426 switch (req) { 2427 case RTM_ADD: 2428 if ((rt->rt_flags & RTF_LOCAL) == 0) 2429 break; 2430 2431 rt->rt_ifp = lo0ifp; 2432 2433 IFADDR_READER_FOREACH(ifa, ifp) { 2434 if (equal(rt_getkey(rt), ifa->ifa_addr)) 2435 break; 2436 } 2437 if (ifa == NULL) 2438 break; 2439 2440 /* 2441 * Ensure lo0 has an address of the same family. 2442 */ 2443 IFADDR_READER_FOREACH(lo0ifa, lo0ifp) { 2444 if (lo0ifa->ifa_addr->sa_family == 2445 ifa->ifa_addr->sa_family) 2446 break; 2447 } 2448 if (lo0ifa == NULL) 2449 break; 2450 2451 /* 2452 * Make sure to set rt->rt_ifa to the interface 2453 * address we are using, otherwise we will have trouble 2454 * with source address selection. 2455 */ 2456 if (ifa != rt->rt_ifa) 2457 rt_replace_ifa(rt, ifa); 2458 break; 2459 case RTM_DELETE: 2460 default: 2461 break; 2462 } 2463 pserialize_read_exit(s); 2464 } 2465 2466 static void 2467 _if_down(struct ifnet *ifp) 2468 { 2469 struct ifaddr *ifa; 2470 struct domain *dp; 2471 int s, bound; 2472 struct psref psref; 2473 2474 ifp->if_flags &= ~IFF_UP; 2475 nanotime(&ifp->if_lastchange); 2476 2477 bound = curlwp_bind(); 2478 s = pserialize_read_enter(); 2479 IFADDR_READER_FOREACH(ifa, ifp) { 2480 ifa_acquire(ifa, &psref); 2481 pserialize_read_exit(s); 2482 2483 pfctlinput(PRC_IFDOWN, ifa->ifa_addr); 2484 2485 s = pserialize_read_enter(); 2486 ifa_release(ifa, &psref); 2487 } 2488 pserialize_read_exit(s); 2489 curlwp_bindx(bound); 2490 2491 IFQ_PURGE(&ifp->if_snd); 2492 #if NCARP > 0 2493 if (ifp->if_carp) 2494 carp_carpdev_state(ifp); 2495 #endif 2496 rt_ifmsg(ifp); 2497 DOMAIN_FOREACH(dp) { 2498 if (dp->dom_if_down) 2499 dp->dom_if_down(ifp); 2500 } 2501 } 2502 2503 static void 2504 if_down_deactivated(struct ifnet *ifp) 2505 { 2506 2507 KASSERT(if_is_deactivated(ifp)); 2508 _if_down(ifp); 2509 } 2510 2511 void 2512 if_down_locked(struct ifnet *ifp) 2513 { 2514 2515 KASSERT(IFNET_LOCKED(ifp)); 2516 _if_down(ifp); 2517 } 2518 2519 /* 2520 * Mark an interface down and notify protocols of 2521 * the transition. 2522 * NOTE: must be called at splsoftnet or equivalent. 2523 */ 2524 void 2525 if_down(struct ifnet *ifp) 2526 { 2527 2528 IFNET_LOCK(ifp); 2529 if_down_locked(ifp); 2530 IFNET_UNLOCK(ifp); 2531 } 2532 2533 /* 2534 * Must be called with holding if_ioctl_lock. 2535 */ 2536 static void 2537 if_up_locked(struct ifnet *ifp) 2538 { 2539 #ifdef notyet 2540 struct ifaddr *ifa; 2541 #endif 2542 struct domain *dp; 2543 2544 KASSERT(IFNET_LOCKED(ifp)); 2545 2546 KASSERT(!if_is_deactivated(ifp)); 2547 ifp->if_flags |= IFF_UP; 2548 nanotime(&ifp->if_lastchange); 2549 #ifdef notyet 2550 /* this has no effect on IP, and will kill all ISO connections XXX */ 2551 IFADDR_READER_FOREACH(ifa, ifp) 2552 pfctlinput(PRC_IFUP, ifa->ifa_addr); 2553 #endif 2554 #if NCARP > 0 2555 if (ifp->if_carp) 2556 carp_carpdev_state(ifp); 2557 #endif 2558 rt_ifmsg(ifp); 2559 DOMAIN_FOREACH(dp) { 2560 if (dp->dom_if_up) 2561 dp->dom_if_up(ifp); 2562 } 2563 } 2564 2565 /* 2566 * Handle interface slowtimo timer routine. Called 2567 * from softclock, we decrement timer (if set) and 2568 * call the appropriate interface routine on expiration. 2569 */ 2570 static void 2571 if_slowtimo(void *arg) 2572 { 2573 void (*slowtimo)(struct ifnet *); 2574 struct ifnet *ifp = arg; 2575 int s; 2576 2577 slowtimo = ifp->if_slowtimo; 2578 if (__predict_false(slowtimo == NULL)) 2579 return; 2580 2581 s = splnet(); 2582 if (ifp->if_timer != 0 && --ifp->if_timer == 0) 2583 (*slowtimo)(ifp); 2584 2585 splx(s); 2586 2587 if (__predict_true(ifp->if_slowtimo != NULL)) 2588 callout_schedule(ifp->if_slowtimo_ch, hz / IFNET_SLOWHZ); 2589 } 2590 2591 /* 2592 * Mark an interface up and notify protocols of 2593 * the transition. 2594 * NOTE: must be called at splsoftnet or equivalent. 2595 */ 2596 void 2597 if_up(struct ifnet *ifp) 2598 { 2599 2600 IFNET_LOCK(ifp); 2601 if_up_locked(ifp); 2602 IFNET_UNLOCK(ifp); 2603 } 2604 2605 /* 2606 * Set/clear promiscuous mode on interface ifp based on the truth value 2607 * of pswitch. The calls are reference counted so that only the first 2608 * "on" request actually has an effect, as does the final "off" request. 2609 * Results are undefined if the "off" and "on" requests are not matched. 2610 */ 2611 int 2612 ifpromisc_locked(struct ifnet *ifp, int pswitch) 2613 { 2614 int pcount, ret = 0; 2615 short nflags; 2616 2617 KASSERT(IFNET_LOCKED(ifp)); 2618 2619 pcount = ifp->if_pcount; 2620 if (pswitch) { 2621 /* 2622 * Allow the device to be "placed" into promiscuous 2623 * mode even if it is not configured up. It will 2624 * consult IFF_PROMISC when it is brought up. 2625 */ 2626 if (ifp->if_pcount++ != 0) 2627 goto out; 2628 nflags = ifp->if_flags | IFF_PROMISC; 2629 } else { 2630 if (--ifp->if_pcount > 0) 2631 goto out; 2632 nflags = ifp->if_flags & ~IFF_PROMISC; 2633 } 2634 ret = if_flags_set(ifp, nflags); 2635 /* Restore interface state if not successful. */ 2636 if (ret != 0) { 2637 ifp->if_pcount = pcount; 2638 } 2639 out: 2640 return ret; 2641 } 2642 2643 int 2644 ifpromisc(struct ifnet *ifp, int pswitch) 2645 { 2646 int e; 2647 2648 IFNET_LOCK(ifp); 2649 e = ifpromisc_locked(ifp, pswitch); 2650 IFNET_UNLOCK(ifp); 2651 2652 return e; 2653 } 2654 2655 /* 2656 * Map interface name to 2657 * interface structure pointer. 2658 */ 2659 struct ifnet * 2660 ifunit(const char *name) 2661 { 2662 struct ifnet *ifp; 2663 const char *cp = name; 2664 u_int unit = 0; 2665 u_int i; 2666 int s; 2667 2668 /* 2669 * If the entire name is a number, treat it as an ifindex. 2670 */ 2671 for (i = 0; i < IFNAMSIZ && *cp >= '0' && *cp <= '9'; i++, cp++) { 2672 unit = unit * 10 + (*cp - '0'); 2673 } 2674 2675 /* 2676 * If the number took all of the name, then it's a valid ifindex. 2677 */ 2678 if (i == IFNAMSIZ || (cp != name && *cp == '\0')) 2679 return if_byindex(unit); 2680 2681 ifp = NULL; 2682 s = pserialize_read_enter(); 2683 IFNET_READER_FOREACH(ifp) { 2684 if (if_is_deactivated(ifp)) 2685 continue; 2686 if (strcmp(ifp->if_xname, name) == 0) 2687 goto out; 2688 } 2689 out: 2690 pserialize_read_exit(s); 2691 return ifp; 2692 } 2693 2694 /* 2695 * Get a reference of an ifnet object by an interface name. 2696 * The returned reference is protected by psref(9). The caller 2697 * must release a returned reference by if_put after use. 2698 */ 2699 struct ifnet * 2700 if_get(const char *name, struct psref *psref) 2701 { 2702 struct ifnet *ifp; 2703 const char *cp = name; 2704 u_int unit = 0; 2705 u_int i; 2706 int s; 2707 2708 /* 2709 * If the entire name is a number, treat it as an ifindex. 2710 */ 2711 for (i = 0; i < IFNAMSIZ && *cp >= '0' && *cp <= '9'; i++, cp++) { 2712 unit = unit * 10 + (*cp - '0'); 2713 } 2714 2715 /* 2716 * If the number took all of the name, then it's a valid ifindex. 2717 */ 2718 if (i == IFNAMSIZ || (cp != name && *cp == '\0')) 2719 return if_get_byindex(unit, psref); 2720 2721 ifp = NULL; 2722 s = pserialize_read_enter(); 2723 IFNET_READER_FOREACH(ifp) { 2724 if (if_is_deactivated(ifp)) 2725 continue; 2726 if (strcmp(ifp->if_xname, name) == 0) { 2727 psref_acquire(psref, &ifp->if_psref, 2728 ifnet_psref_class); 2729 goto out; 2730 } 2731 } 2732 out: 2733 pserialize_read_exit(s); 2734 return ifp; 2735 } 2736 2737 /* 2738 * Release a reference of an ifnet object given by if_get, if_get_byindex 2739 * or if_get_bylla. 2740 */ 2741 void 2742 if_put(const struct ifnet *ifp, struct psref *psref) 2743 { 2744 2745 if (ifp == NULL) 2746 return; 2747 2748 psref_release(psref, &ifp->if_psref, ifnet_psref_class); 2749 } 2750 2751 /* 2752 * Return ifp having idx. Return NULL if not found. Normally if_byindex 2753 * should be used. 2754 */ 2755 ifnet_t * 2756 _if_byindex(u_int idx) 2757 { 2758 2759 return (__predict_true(idx < if_indexlim)) ? ifindex2ifnet[idx] : NULL; 2760 } 2761 2762 /* 2763 * Return ifp having idx. Return NULL if not found or the found ifp is 2764 * already deactivated. 2765 */ 2766 ifnet_t * 2767 if_byindex(u_int idx) 2768 { 2769 ifnet_t *ifp; 2770 2771 ifp = _if_byindex(idx); 2772 if (ifp != NULL && if_is_deactivated(ifp)) 2773 ifp = NULL; 2774 return ifp; 2775 } 2776 2777 /* 2778 * Get a reference of an ifnet object by an interface index. 2779 * The returned reference is protected by psref(9). The caller 2780 * must release a returned reference by if_put after use. 2781 */ 2782 ifnet_t * 2783 if_get_byindex(u_int idx, struct psref *psref) 2784 { 2785 ifnet_t *ifp; 2786 int s; 2787 2788 s = pserialize_read_enter(); 2789 ifp = if_byindex(idx); 2790 if (__predict_true(ifp != NULL)) 2791 psref_acquire(psref, &ifp->if_psref, ifnet_psref_class); 2792 pserialize_read_exit(s); 2793 2794 return ifp; 2795 } 2796 2797 ifnet_t * 2798 if_get_bylla(const void *lla, unsigned char lla_len, struct psref *psref) 2799 { 2800 ifnet_t *ifp; 2801 int s; 2802 2803 s = pserialize_read_enter(); 2804 IFNET_READER_FOREACH(ifp) { 2805 if (if_is_deactivated(ifp)) 2806 continue; 2807 if (ifp->if_addrlen != lla_len) 2808 continue; 2809 if (memcmp(lla, CLLADDR(ifp->if_sadl), lla_len) == 0) { 2810 psref_acquire(psref, &ifp->if_psref, 2811 ifnet_psref_class); 2812 break; 2813 } 2814 } 2815 pserialize_read_exit(s); 2816 2817 return ifp; 2818 } 2819 2820 /* 2821 * Note that it's safe only if the passed ifp is guaranteed to not be freed, 2822 * for example using pserialize or the ifp is already held or some other 2823 * object is held which guarantes the ifp to not be freed indirectly. 2824 */ 2825 void 2826 if_acquire(struct ifnet *ifp, struct psref *psref) 2827 { 2828 2829 KASSERT(ifp->if_index != 0); 2830 psref_acquire(psref, &ifp->if_psref, ifnet_psref_class); 2831 } 2832 2833 bool 2834 if_held(struct ifnet *ifp) 2835 { 2836 2837 return psref_held(&ifp->if_psref, ifnet_psref_class); 2838 } 2839 2840 /* 2841 * Some tunnel interfaces can nest, e.g. IPv4 over IPv4 gif(4) tunnel over IPv4. 2842 * Check the tunnel nesting count. 2843 * Return > 0, if tunnel nesting count is more than limit. 2844 * Return 0, if tunnel nesting count is equal or less than limit. 2845 */ 2846 int 2847 if_tunnel_check_nesting(struct ifnet *ifp, struct mbuf *m, int limit) 2848 { 2849 struct m_tag *mtag; 2850 int *count; 2851 2852 mtag = m_tag_find(m, PACKET_TAG_TUNNEL_INFO, NULL); 2853 if (mtag != NULL) { 2854 count = (int *)(mtag + 1); 2855 if (++(*count) > limit) { 2856 log(LOG_NOTICE, 2857 "%s: recursively called too many times(%d)\n", 2858 ifp->if_xname, *count); 2859 return EIO; 2860 } 2861 } else { 2862 mtag = m_tag_get(PACKET_TAG_TUNNEL_INFO, sizeof(*count), 2863 M_NOWAIT); 2864 if (mtag != NULL) { 2865 m_tag_prepend(m, mtag); 2866 count = (int *)(mtag + 1); 2867 *count = 0; 2868 } else { 2869 log(LOG_DEBUG, 2870 "%s: m_tag_get() failed, recursion calls are not prevented.\n", 2871 ifp->if_xname); 2872 } 2873 } 2874 2875 return 0; 2876 } 2877 2878 /* common */ 2879 int 2880 ifioctl_common(struct ifnet *ifp, u_long cmd, void *data) 2881 { 2882 int s; 2883 struct ifreq *ifr; 2884 struct ifcapreq *ifcr; 2885 struct ifdatareq *ifdr; 2886 2887 switch (cmd) { 2888 case SIOCSIFCAP: 2889 ifcr = data; 2890 if ((ifcr->ifcr_capenable & ~ifp->if_capabilities) != 0) 2891 return EINVAL; 2892 2893 if (ifcr->ifcr_capenable == ifp->if_capenable) 2894 return 0; 2895 2896 ifp->if_capenable = ifcr->ifcr_capenable; 2897 2898 /* Pre-compute the checksum flags mask. */ 2899 ifp->if_csum_flags_tx = 0; 2900 ifp->if_csum_flags_rx = 0; 2901 if (ifp->if_capenable & IFCAP_CSUM_IPv4_Tx) { 2902 ifp->if_csum_flags_tx |= M_CSUM_IPv4; 2903 } 2904 if (ifp->if_capenable & IFCAP_CSUM_IPv4_Rx) { 2905 ifp->if_csum_flags_rx |= M_CSUM_IPv4; 2906 } 2907 2908 if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Tx) { 2909 ifp->if_csum_flags_tx |= M_CSUM_TCPv4; 2910 } 2911 if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Rx) { 2912 ifp->if_csum_flags_rx |= M_CSUM_TCPv4; 2913 } 2914 2915 if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Tx) { 2916 ifp->if_csum_flags_tx |= M_CSUM_UDPv4; 2917 } 2918 if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Rx) { 2919 ifp->if_csum_flags_rx |= M_CSUM_UDPv4; 2920 } 2921 2922 if (ifp->if_capenable & IFCAP_CSUM_TCPv6_Tx) { 2923 ifp->if_csum_flags_tx |= M_CSUM_TCPv6; 2924 } 2925 if (ifp->if_capenable & IFCAP_CSUM_TCPv6_Rx) { 2926 ifp->if_csum_flags_rx |= M_CSUM_TCPv6; 2927 } 2928 2929 if (ifp->if_capenable & IFCAP_CSUM_UDPv6_Tx) { 2930 ifp->if_csum_flags_tx |= M_CSUM_UDPv6; 2931 } 2932 if (ifp->if_capenable & IFCAP_CSUM_UDPv6_Rx) { 2933 ifp->if_csum_flags_rx |= M_CSUM_UDPv6; 2934 } 2935 if (ifp->if_flags & IFF_UP) 2936 return ENETRESET; 2937 return 0; 2938 case SIOCSIFFLAGS: 2939 ifr = data; 2940 /* 2941 * If if_is_mpsafe(ifp), KERNEL_LOCK isn't held here, but if_up 2942 * and if_down aren't MP-safe yet, so we must hold the lock. 2943 */ 2944 KERNEL_LOCK_IF_IFP_MPSAFE(ifp); 2945 if (ifp->if_flags & IFF_UP && (ifr->ifr_flags & IFF_UP) == 0) { 2946 s = splsoftnet(); 2947 if_down_locked(ifp); 2948 splx(s); 2949 } 2950 if (ifr->ifr_flags & IFF_UP && (ifp->if_flags & IFF_UP) == 0) { 2951 s = splsoftnet(); 2952 if_up_locked(ifp); 2953 splx(s); 2954 } 2955 KERNEL_UNLOCK_IF_IFP_MPSAFE(ifp); 2956 ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) | 2957 (ifr->ifr_flags &~ IFF_CANTCHANGE); 2958 break; 2959 case SIOCGIFFLAGS: 2960 ifr = data; 2961 ifr->ifr_flags = ifp->if_flags; 2962 break; 2963 2964 case SIOCGIFMETRIC: 2965 ifr = data; 2966 ifr->ifr_metric = ifp->if_metric; 2967 break; 2968 2969 case SIOCGIFMTU: 2970 ifr = data; 2971 ifr->ifr_mtu = ifp->if_mtu; 2972 break; 2973 2974 case SIOCGIFDLT: 2975 ifr = data; 2976 ifr->ifr_dlt = ifp->if_dlt; 2977 break; 2978 2979 case SIOCGIFCAP: 2980 ifcr = data; 2981 ifcr->ifcr_capabilities = ifp->if_capabilities; 2982 ifcr->ifcr_capenable = ifp->if_capenable; 2983 break; 2984 2985 case SIOCSIFMETRIC: 2986 ifr = data; 2987 ifp->if_metric = ifr->ifr_metric; 2988 break; 2989 2990 case SIOCGIFDATA: 2991 ifdr = data; 2992 ifdr->ifdr_data = ifp->if_data; 2993 break; 2994 2995 case SIOCGIFINDEX: 2996 ifr = data; 2997 ifr->ifr_index = ifp->if_index; 2998 break; 2999 3000 case SIOCZIFDATA: 3001 ifdr = data; 3002 ifdr->ifdr_data = ifp->if_data; 3003 /* 3004 * Assumes that the volatile counters that can be 3005 * zero'ed are at the end of if_data. 3006 */ 3007 memset(&ifp->if_data.ifi_ipackets, 0, sizeof(ifp->if_data) - 3008 offsetof(struct if_data, ifi_ipackets)); 3009 /* 3010 * The memset() clears to the bottm of if_data. In the area, 3011 * if_lastchange is included. Please be careful if new entry 3012 * will be added into if_data or rewite this. 3013 * 3014 * And also, update if_lastchnage. 3015 */ 3016 getnanotime(&ifp->if_lastchange); 3017 break; 3018 case SIOCSIFMTU: 3019 ifr = data; 3020 if (ifp->if_mtu == ifr->ifr_mtu) 3021 break; 3022 ifp->if_mtu = ifr->ifr_mtu; 3023 /* 3024 * If the link MTU changed, do network layer specific procedure. 3025 */ 3026 #ifdef INET6 3027 KERNEL_LOCK_UNLESS_NET_MPSAFE(); 3028 if (in6_present) 3029 nd6_setmtu(ifp); 3030 KERNEL_UNLOCK_UNLESS_NET_MPSAFE(); 3031 #endif 3032 return ENETRESET; 3033 default: 3034 return ENOTTY; 3035 } 3036 return 0; 3037 } 3038 3039 int 3040 ifaddrpref_ioctl(struct socket *so, u_long cmd, void *data, struct ifnet *ifp) 3041 { 3042 struct if_addrprefreq *ifap = (struct if_addrprefreq *)data; 3043 struct ifaddr *ifa; 3044 const struct sockaddr *any, *sa; 3045 union { 3046 struct sockaddr sa; 3047 struct sockaddr_storage ss; 3048 } u, v; 3049 int s, error = 0; 3050 3051 switch (cmd) { 3052 case SIOCSIFADDRPREF: 3053 if (kauth_authorize_network(curlwp->l_cred, KAUTH_NETWORK_INTERFACE, 3054 KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd, 3055 NULL) != 0) 3056 return EPERM; 3057 case SIOCGIFADDRPREF: 3058 break; 3059 default: 3060 return EOPNOTSUPP; 3061 } 3062 3063 /* sanity checks */ 3064 if (data == NULL || ifp == NULL) { 3065 panic("invalid argument to %s", __func__); 3066 /*NOTREACHED*/ 3067 } 3068 3069 /* address must be specified on ADD and DELETE */ 3070 sa = sstocsa(&ifap->ifap_addr); 3071 if (sa->sa_family != sofamily(so)) 3072 return EINVAL; 3073 if ((any = sockaddr_any(sa)) == NULL || sa->sa_len != any->sa_len) 3074 return EINVAL; 3075 3076 sockaddr_externalize(&v.sa, sizeof(v.ss), sa); 3077 3078 s = pserialize_read_enter(); 3079 IFADDR_READER_FOREACH(ifa, ifp) { 3080 if (ifa->ifa_addr->sa_family != sa->sa_family) 3081 continue; 3082 sockaddr_externalize(&u.sa, sizeof(u.ss), ifa->ifa_addr); 3083 if (sockaddr_cmp(&u.sa, &v.sa) == 0) 3084 break; 3085 } 3086 if (ifa == NULL) { 3087 error = EADDRNOTAVAIL; 3088 goto out; 3089 } 3090 3091 switch (cmd) { 3092 case SIOCSIFADDRPREF: 3093 ifa->ifa_preference = ifap->ifap_preference; 3094 goto out; 3095 case SIOCGIFADDRPREF: 3096 /* fill in the if_laddrreq structure */ 3097 (void)sockaddr_copy(sstosa(&ifap->ifap_addr), 3098 sizeof(ifap->ifap_addr), ifa->ifa_addr); 3099 ifap->ifap_preference = ifa->ifa_preference; 3100 goto out; 3101 default: 3102 error = EOPNOTSUPP; 3103 } 3104 out: 3105 pserialize_read_exit(s); 3106 return error; 3107 } 3108 3109 /* 3110 * Interface ioctls. 3111 */ 3112 static int 3113 doifioctl(struct socket *so, u_long cmd, void *data, struct lwp *l) 3114 { 3115 struct ifnet *ifp; 3116 struct ifreq *ifr; 3117 int error = 0; 3118 #if defined(COMPAT_OSOCK) || defined(COMPAT_OIFREQ) 3119 u_long ocmd = cmd; 3120 #endif 3121 short oif_flags; 3122 #ifdef COMPAT_OIFREQ 3123 struct ifreq ifrb; 3124 struct oifreq *oifr = NULL; 3125 #endif 3126 int r; 3127 struct psref psref; 3128 int bound; 3129 3130 switch (cmd) { 3131 case SIOCGIFCONF: 3132 return ifconf(cmd, data); 3133 case SIOCINITIFADDR: 3134 return EPERM; 3135 default: 3136 error = (*vec_compat_ifconf)(l, cmd, data); 3137 if (error != ENOSYS) 3138 return error; 3139 error = (*vec_compat_ifdatareq)(l, cmd, data); 3140 if (error != ENOSYS) 3141 return error; 3142 break; 3143 } 3144 3145 ifr = data; 3146 #ifdef COMPAT_OIFREQ 3147 if (vec_compat_cvtcmd) { 3148 cmd = (*vec_compat_cvtcmd)(cmd); 3149 if (cmd != ocmd) { 3150 oifr = data; 3151 data = ifr = &ifrb; 3152 ifreqo2n(oifr, ifr); 3153 } 3154 } 3155 #endif 3156 3157 switch (cmd) { 3158 case SIOCIFCREATE: 3159 case SIOCIFDESTROY: 3160 bound = curlwp_bind(); 3161 if (l != NULL) { 3162 ifp = if_get(ifr->ifr_name, &psref); 3163 error = kauth_authorize_network(l->l_cred, 3164 KAUTH_NETWORK_INTERFACE, 3165 KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, 3166 (void *)cmd, NULL); 3167 if (ifp != NULL) 3168 if_put(ifp, &psref); 3169 if (error != 0) { 3170 curlwp_bindx(bound); 3171 return error; 3172 } 3173 } 3174 KERNEL_LOCK_UNLESS_NET_MPSAFE(); 3175 mutex_enter(&if_clone_mtx); 3176 r = (cmd == SIOCIFCREATE) ? 3177 if_clone_create(ifr->ifr_name) : 3178 if_clone_destroy(ifr->ifr_name); 3179 mutex_exit(&if_clone_mtx); 3180 KERNEL_UNLOCK_UNLESS_NET_MPSAFE(); 3181 curlwp_bindx(bound); 3182 return r; 3183 3184 case SIOCIFGCLONERS: 3185 { 3186 struct if_clonereq *req = (struct if_clonereq *)data; 3187 return if_clone_list(req->ifcr_count, req->ifcr_buffer, 3188 &req->ifcr_total); 3189 } 3190 } 3191 3192 bound = curlwp_bind(); 3193 ifp = if_get(ifr->ifr_name, &psref); 3194 if (ifp == NULL) { 3195 curlwp_bindx(bound); 3196 return ENXIO; 3197 } 3198 3199 switch (cmd) { 3200 case SIOCALIFADDR: 3201 case SIOCDLIFADDR: 3202 case SIOCSIFADDRPREF: 3203 case SIOCSIFFLAGS: 3204 case SIOCSIFCAP: 3205 case SIOCSIFMETRIC: 3206 case SIOCZIFDATA: 3207 case SIOCSIFMTU: 3208 case SIOCSIFPHYADDR: 3209 case SIOCDIFPHYADDR: 3210 #ifdef INET6 3211 case SIOCSIFPHYADDR_IN6: 3212 #endif 3213 case SIOCSLIFPHYADDR: 3214 case SIOCADDMULTI: 3215 case SIOCDELMULTI: 3216 case SIOCSIFMEDIA: 3217 case SIOCSDRVSPEC: 3218 case SIOCG80211: 3219 case SIOCS80211: 3220 case SIOCS80211NWID: 3221 case SIOCS80211NWKEY: 3222 case SIOCS80211POWER: 3223 case SIOCS80211BSSID: 3224 case SIOCS80211CHANNEL: 3225 case SIOCSLINKSTR: 3226 if (l != NULL) { 3227 error = kauth_authorize_network(l->l_cred, 3228 KAUTH_NETWORK_INTERFACE, 3229 KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, 3230 (void *)cmd, NULL); 3231 if (error != 0) 3232 goto out; 3233 } 3234 } 3235 3236 oif_flags = ifp->if_flags; 3237 3238 KERNEL_LOCK_UNLESS_IFP_MPSAFE(ifp); 3239 IFNET_LOCK(ifp); 3240 3241 error = (*ifp->if_ioctl)(ifp, cmd, data); 3242 if (error != ENOTTY) 3243 ; 3244 else if (so->so_proto == NULL) 3245 error = EOPNOTSUPP; 3246 else { 3247 KERNEL_LOCK_IF_IFP_MPSAFE(ifp); 3248 #ifdef COMPAT_OSOCK 3249 if (vec_compat_ifioctl != NULL) 3250 error = (*vec_compat_ifioctl)(so, ocmd, cmd, data, l); 3251 else 3252 #endif 3253 error = (*so->so_proto->pr_usrreqs->pr_ioctl)(so, 3254 cmd, data, ifp); 3255 KERNEL_UNLOCK_IF_IFP_MPSAFE(ifp); 3256 } 3257 3258 if (((oif_flags ^ ifp->if_flags) & IFF_UP) != 0) { 3259 if ((ifp->if_flags & IFF_UP) != 0) { 3260 int s = splsoftnet(); 3261 if_up_locked(ifp); 3262 splx(s); 3263 } 3264 } 3265 #ifdef COMPAT_OIFREQ 3266 if (cmd != ocmd) 3267 ifreqn2o(oifr, ifr); 3268 #endif 3269 3270 IFNET_UNLOCK(ifp); 3271 KERNEL_UNLOCK_UNLESS_IFP_MPSAFE(ifp); 3272 out: 3273 if_put(ifp, &psref); 3274 curlwp_bindx(bound); 3275 return error; 3276 } 3277 3278 /* 3279 * Return interface configuration 3280 * of system. List may be used 3281 * in later ioctl's (above) to get 3282 * other information. 3283 * 3284 * Each record is a struct ifreq. Before the addition of 3285 * sockaddr_storage, the API rule was that sockaddr flavors that did 3286 * not fit would extend beyond the struct ifreq, with the next struct 3287 * ifreq starting sa_len beyond the struct sockaddr. Because the 3288 * union in struct ifreq includes struct sockaddr_storage, every kind 3289 * of sockaddr must fit. Thus, there are no longer any overlength 3290 * records. 3291 * 3292 * Records are added to the user buffer if they fit, and ifc_len is 3293 * adjusted to the length that was written. Thus, the user is only 3294 * assured of getting the complete list if ifc_len on return is at 3295 * least sizeof(struct ifreq) less than it was on entry. 3296 * 3297 * If the user buffer pointer is NULL, this routine copies no data and 3298 * returns the amount of space that would be needed. 3299 * 3300 * Invariants: 3301 * ifrp points to the next part of the user's buffer to be used. If 3302 * ifrp != NULL, space holds the number of bytes remaining that we may 3303 * write at ifrp. Otherwise, space holds the number of bytes that 3304 * would have been written had there been adequate space. 3305 */ 3306 /*ARGSUSED*/ 3307 static int 3308 ifconf(u_long cmd, void *data) 3309 { 3310 struct ifconf *ifc = (struct ifconf *)data; 3311 struct ifnet *ifp; 3312 struct ifaddr *ifa; 3313 struct ifreq ifr, *ifrp = NULL; 3314 int space = 0, error = 0; 3315 const int sz = (int)sizeof(struct ifreq); 3316 const bool docopy = ifc->ifc_req != NULL; 3317 int s; 3318 int bound; 3319 struct psref psref; 3320 3321 if (docopy) { 3322 space = ifc->ifc_len; 3323 ifrp = ifc->ifc_req; 3324 } 3325 3326 bound = curlwp_bind(); 3327 s = pserialize_read_enter(); 3328 IFNET_READER_FOREACH(ifp) { 3329 psref_acquire(&psref, &ifp->if_psref, ifnet_psref_class); 3330 pserialize_read_exit(s); 3331 3332 (void)strncpy(ifr.ifr_name, ifp->if_xname, 3333 sizeof(ifr.ifr_name)); 3334 if (ifr.ifr_name[sizeof(ifr.ifr_name) - 1] != '\0') { 3335 error = ENAMETOOLONG; 3336 goto release_exit; 3337 } 3338 if (IFADDR_READER_EMPTY(ifp)) { 3339 /* Interface with no addresses - send zero sockaddr. */ 3340 memset(&ifr.ifr_addr, 0, sizeof(ifr.ifr_addr)); 3341 if (!docopy) { 3342 space += sz; 3343 goto next; 3344 } 3345 if (space >= sz) { 3346 error = copyout(&ifr, ifrp, sz); 3347 if (error != 0) 3348 goto release_exit; 3349 ifrp++; 3350 space -= sz; 3351 } 3352 } 3353 3354 s = pserialize_read_enter(); 3355 IFADDR_READER_FOREACH(ifa, ifp) { 3356 struct sockaddr *sa = ifa->ifa_addr; 3357 /* all sockaddrs must fit in sockaddr_storage */ 3358 KASSERT(sa->sa_len <= sizeof(ifr.ifr_ifru)); 3359 3360 if (!docopy) { 3361 space += sz; 3362 continue; 3363 } 3364 memcpy(&ifr.ifr_space, sa, sa->sa_len); 3365 pserialize_read_exit(s); 3366 3367 if (space >= sz) { 3368 error = copyout(&ifr, ifrp, sz); 3369 if (error != 0) 3370 goto release_exit; 3371 ifrp++; space -= sz; 3372 } 3373 s = pserialize_read_enter(); 3374 } 3375 pserialize_read_exit(s); 3376 3377 next: 3378 s = pserialize_read_enter(); 3379 psref_release(&psref, &ifp->if_psref, ifnet_psref_class); 3380 } 3381 pserialize_read_exit(s); 3382 curlwp_bindx(bound); 3383 3384 if (docopy) { 3385 KASSERT(0 <= space && space <= ifc->ifc_len); 3386 ifc->ifc_len -= space; 3387 } else { 3388 KASSERT(space >= 0); 3389 ifc->ifc_len = space; 3390 } 3391 return (0); 3392 3393 release_exit: 3394 psref_release(&psref, &ifp->if_psref, ifnet_psref_class); 3395 curlwp_bindx(bound); 3396 return error; 3397 } 3398 3399 int 3400 ifreq_setaddr(u_long cmd, struct ifreq *ifr, const struct sockaddr *sa) 3401 { 3402 uint8_t len = sizeof(ifr->ifr_ifru.ifru_space); 3403 #ifdef COMPAT_OIFREQ 3404 struct ifreq ifrb; 3405 struct oifreq *oifr = NULL; 3406 u_long ocmd = cmd; 3407 3408 if (vec_compat_cvtcmd) { 3409 cmd = (*vec_compat_cvtcmd)(cmd); 3410 if (cmd != ocmd) { 3411 oifr = (struct oifreq *)(void *)ifr; 3412 ifr = &ifrb; 3413 ifreqo2n(oifr, ifr); 3414 len = sizeof(oifr->ifr_addr); 3415 } 3416 } 3417 #endif 3418 if (len < sa->sa_len) 3419 return EFBIG; 3420 3421 memset(&ifr->ifr_addr, 0, len); 3422 sockaddr_copy(&ifr->ifr_addr, len, sa); 3423 3424 #ifdef COMPAT_OIFREQ 3425 if (cmd != ocmd) 3426 ifreqn2o(oifr, ifr); 3427 #endif 3428 return 0; 3429 } 3430 3431 /* 3432 * wrapper function for the drivers which doesn't have if_transmit(). 3433 */ 3434 static int 3435 if_transmit(struct ifnet *ifp, struct mbuf *m) 3436 { 3437 int s, error; 3438 size_t pktlen = m->m_pkthdr.len; 3439 bool mcast = (m->m_flags & M_MCAST) != 0; 3440 3441 s = splnet(); 3442 3443 IFQ_ENQUEUE(&ifp->if_snd, m, error); 3444 if (error != 0) { 3445 /* mbuf is already freed */ 3446 goto out; 3447 } 3448 3449 ifp->if_obytes += pktlen; 3450 if (mcast) 3451 ifp->if_omcasts++; 3452 3453 if ((ifp->if_flags & IFF_OACTIVE) == 0) 3454 if_start_lock(ifp); 3455 out: 3456 splx(s); 3457 3458 return error; 3459 } 3460 3461 int 3462 if_transmit_lock(struct ifnet *ifp, struct mbuf *m) 3463 { 3464 int error; 3465 3466 #ifdef ALTQ 3467 KERNEL_LOCK(1, NULL); 3468 if (ALTQ_IS_ENABLED(&ifp->if_snd)) { 3469 error = if_transmit(ifp, m); 3470 KERNEL_UNLOCK_ONE(NULL); 3471 } else { 3472 KERNEL_UNLOCK_ONE(NULL); 3473 error = (*ifp->if_transmit)(ifp, m); 3474 /* mbuf is alredy freed */ 3475 } 3476 #else /* !ALTQ */ 3477 error = (*ifp->if_transmit)(ifp, m); 3478 /* mbuf is alredy freed */ 3479 #endif /* !ALTQ */ 3480 3481 return error; 3482 } 3483 3484 /* 3485 * Queue message on interface, and start output if interface 3486 * not yet active. 3487 */ 3488 int 3489 ifq_enqueue(struct ifnet *ifp, struct mbuf *m) 3490 { 3491 3492 return if_transmit_lock(ifp, m); 3493 } 3494 3495 /* 3496 * Queue message on interface, possibly using a second fast queue 3497 */ 3498 int 3499 ifq_enqueue2(struct ifnet *ifp, struct ifqueue *ifq, struct mbuf *m) 3500 { 3501 int error = 0; 3502 3503 if (ifq != NULL 3504 #ifdef ALTQ 3505 && ALTQ_IS_ENABLED(&ifp->if_snd) == 0 3506 #endif 3507 ) { 3508 if (IF_QFULL(ifq)) { 3509 IF_DROP(&ifp->if_snd); 3510 m_freem(m); 3511 if (error == 0) 3512 error = ENOBUFS; 3513 } else 3514 IF_ENQUEUE(ifq, m); 3515 } else 3516 IFQ_ENQUEUE(&ifp->if_snd, m, error); 3517 if (error != 0) { 3518 ++ifp->if_oerrors; 3519 return error; 3520 } 3521 return 0; 3522 } 3523 3524 int 3525 if_addr_init(ifnet_t *ifp, struct ifaddr *ifa, const bool src) 3526 { 3527 int rc; 3528 3529 KASSERT(IFNET_LOCKED(ifp)); 3530 if (ifp->if_initaddr != NULL) 3531 rc = (*ifp->if_initaddr)(ifp, ifa, src); 3532 else if (src || 3533 (rc = (*ifp->if_ioctl)(ifp, SIOCSIFDSTADDR, ifa)) == ENOTTY) 3534 rc = (*ifp->if_ioctl)(ifp, SIOCINITIFADDR, ifa); 3535 3536 return rc; 3537 } 3538 3539 int 3540 if_do_dad(struct ifnet *ifp) 3541 { 3542 if ((ifp->if_flags & IFF_LOOPBACK) != 0) 3543 return 0; 3544 3545 switch (ifp->if_type) { 3546 case IFT_FAITH: 3547 /* 3548 * These interfaces do not have the IFF_LOOPBACK flag, 3549 * but loop packets back. We do not have to do DAD on such 3550 * interfaces. We should even omit it, because loop-backed 3551 * responses would confuse the DAD procedure. 3552 */ 3553 return 0; 3554 default: 3555 /* 3556 * Our DAD routine requires the interface up and running. 3557 * However, some interfaces can be up before the RUNNING 3558 * status. Additionaly, users may try to assign addresses 3559 * before the interface becomes up (or running). 3560 * We simply skip DAD in such a case as a work around. 3561 * XXX: we should rather mark "tentative" on such addresses, 3562 * and do DAD after the interface becomes ready. 3563 */ 3564 if ((ifp->if_flags & (IFF_UP|IFF_RUNNING)) != 3565 (IFF_UP|IFF_RUNNING)) 3566 return 0; 3567 3568 return 1; 3569 } 3570 } 3571 3572 int 3573 if_flags_set(ifnet_t *ifp, const short flags) 3574 { 3575 int rc; 3576 3577 KASSERT(IFNET_LOCKED(ifp)); 3578 3579 if (ifp->if_setflags != NULL) 3580 rc = (*ifp->if_setflags)(ifp, flags); 3581 else { 3582 short cantflags, chgdflags; 3583 struct ifreq ifr; 3584 3585 chgdflags = ifp->if_flags ^ flags; 3586 cantflags = chgdflags & IFF_CANTCHANGE; 3587 3588 if (cantflags != 0) 3589 ifp->if_flags ^= cantflags; 3590 3591 /* Traditionally, we do not call if_ioctl after 3592 * setting/clearing only IFF_PROMISC if the interface 3593 * isn't IFF_UP. Uphold that tradition. 3594 */ 3595 if (chgdflags == IFF_PROMISC && (ifp->if_flags & IFF_UP) == 0) 3596 return 0; 3597 3598 memset(&ifr, 0, sizeof(ifr)); 3599 3600 ifr.ifr_flags = flags & ~IFF_CANTCHANGE; 3601 rc = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, &ifr); 3602 3603 if (rc != 0 && cantflags != 0) 3604 ifp->if_flags ^= cantflags; 3605 } 3606 3607 return rc; 3608 } 3609 3610 int 3611 if_mcast_op(ifnet_t *ifp, const unsigned long cmd, const struct sockaddr *sa) 3612 { 3613 int rc; 3614 struct ifreq ifr; 3615 3616 /* There remain some paths that don't hold IFNET_LOCK yet */ 3617 #ifdef NET_MPSAFE 3618 /* CARP and MROUTING still don't deal with the lock yet */ 3619 #if (!defined(NCARP) || (NCARP == 0)) && !defined(MROUTING) 3620 KASSERT(IFNET_LOCKED(ifp)); 3621 #endif 3622 #endif 3623 if (ifp->if_mcastop != NULL) 3624 rc = (*ifp->if_mcastop)(ifp, cmd, sa); 3625 else { 3626 ifreq_setaddr(cmd, &ifr, sa); 3627 rc = (*ifp->if_ioctl)(ifp, cmd, &ifr); 3628 } 3629 3630 return rc; 3631 } 3632 3633 static void 3634 sysctl_sndq_setup(struct sysctllog **clog, const char *ifname, 3635 struct ifaltq *ifq) 3636 { 3637 const struct sysctlnode *cnode, *rnode; 3638 3639 if (sysctl_createv(clog, 0, NULL, &rnode, 3640 CTLFLAG_PERMANENT, 3641 CTLTYPE_NODE, "interfaces", 3642 SYSCTL_DESCR("Per-interface controls"), 3643 NULL, 0, NULL, 0, 3644 CTL_NET, CTL_CREATE, CTL_EOL) != 0) 3645 goto bad; 3646 3647 if (sysctl_createv(clog, 0, &rnode, &rnode, 3648 CTLFLAG_PERMANENT, 3649 CTLTYPE_NODE, ifname, 3650 SYSCTL_DESCR("Interface controls"), 3651 NULL, 0, NULL, 0, 3652 CTL_CREATE, CTL_EOL) != 0) 3653 goto bad; 3654 3655 if (sysctl_createv(clog, 0, &rnode, &rnode, 3656 CTLFLAG_PERMANENT, 3657 CTLTYPE_NODE, "sndq", 3658 SYSCTL_DESCR("Interface output queue controls"), 3659 NULL, 0, NULL, 0, 3660 CTL_CREATE, CTL_EOL) != 0) 3661 goto bad; 3662 3663 if (sysctl_createv(clog, 0, &rnode, &cnode, 3664 CTLFLAG_PERMANENT, 3665 CTLTYPE_INT, "len", 3666 SYSCTL_DESCR("Current output queue length"), 3667 NULL, 0, &ifq->ifq_len, 0, 3668 CTL_CREATE, CTL_EOL) != 0) 3669 goto bad; 3670 3671 if (sysctl_createv(clog, 0, &rnode, &cnode, 3672 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 3673 CTLTYPE_INT, "maxlen", 3674 SYSCTL_DESCR("Maximum allowed output queue length"), 3675 NULL, 0, &ifq->ifq_maxlen, 0, 3676 CTL_CREATE, CTL_EOL) != 0) 3677 goto bad; 3678 3679 if (sysctl_createv(clog, 0, &rnode, &cnode, 3680 CTLFLAG_PERMANENT, 3681 CTLTYPE_INT, "drops", 3682 SYSCTL_DESCR("Packets dropped due to full output queue"), 3683 NULL, 0, &ifq->ifq_drops, 0, 3684 CTL_CREATE, CTL_EOL) != 0) 3685 goto bad; 3686 3687 return; 3688 bad: 3689 printf("%s: could not attach sysctl nodes\n", ifname); 3690 return; 3691 } 3692 3693 #if defined(INET) || defined(INET6) 3694 3695 #define SYSCTL_NET_PKTQ(q, cn, c) \ 3696 static int \ 3697 sysctl_net_##q##_##cn(SYSCTLFN_ARGS) \ 3698 { \ 3699 return sysctl_pktq_count(SYSCTLFN_CALL(rnode), q, c); \ 3700 } 3701 3702 #if defined(INET) 3703 static int 3704 sysctl_net_ip_pktq_maxlen(SYSCTLFN_ARGS) 3705 { 3706 return sysctl_pktq_maxlen(SYSCTLFN_CALL(rnode), ip_pktq); 3707 } 3708 SYSCTL_NET_PKTQ(ip_pktq, items, PKTQ_NITEMS) 3709 SYSCTL_NET_PKTQ(ip_pktq, drops, PKTQ_DROPS) 3710 #endif 3711 3712 #if defined(INET6) 3713 static int 3714 sysctl_net_ip6_pktq_maxlen(SYSCTLFN_ARGS) 3715 { 3716 return sysctl_pktq_maxlen(SYSCTLFN_CALL(rnode), ip6_pktq); 3717 } 3718 SYSCTL_NET_PKTQ(ip6_pktq, items, PKTQ_NITEMS) 3719 SYSCTL_NET_PKTQ(ip6_pktq, drops, PKTQ_DROPS) 3720 #endif 3721 3722 static void 3723 sysctl_net_pktq_setup(struct sysctllog **clog, int pf) 3724 { 3725 sysctlfn len_func = NULL, maxlen_func = NULL, drops_func = NULL; 3726 const char *pfname = NULL, *ipname = NULL; 3727 int ipn = 0, qid = 0; 3728 3729 switch (pf) { 3730 #if defined(INET) 3731 case PF_INET: 3732 len_func = sysctl_net_ip_pktq_items; 3733 maxlen_func = sysctl_net_ip_pktq_maxlen; 3734 drops_func = sysctl_net_ip_pktq_drops; 3735 pfname = "inet", ipn = IPPROTO_IP; 3736 ipname = "ip", qid = IPCTL_IFQ; 3737 break; 3738 #endif 3739 #if defined(INET6) 3740 case PF_INET6: 3741 len_func = sysctl_net_ip6_pktq_items; 3742 maxlen_func = sysctl_net_ip6_pktq_maxlen; 3743 drops_func = sysctl_net_ip6_pktq_drops; 3744 pfname = "inet6", ipn = IPPROTO_IPV6; 3745 ipname = "ip6", qid = IPV6CTL_IFQ; 3746 break; 3747 #endif 3748 default: 3749 KASSERT(false); 3750 } 3751 3752 sysctl_createv(clog, 0, NULL, NULL, 3753 CTLFLAG_PERMANENT, 3754 CTLTYPE_NODE, pfname, NULL, 3755 NULL, 0, NULL, 0, 3756 CTL_NET, pf, CTL_EOL); 3757 sysctl_createv(clog, 0, NULL, NULL, 3758 CTLFLAG_PERMANENT, 3759 CTLTYPE_NODE, ipname, NULL, 3760 NULL, 0, NULL, 0, 3761 CTL_NET, pf, ipn, CTL_EOL); 3762 sysctl_createv(clog, 0, NULL, NULL, 3763 CTLFLAG_PERMANENT, 3764 CTLTYPE_NODE, "ifq", 3765 SYSCTL_DESCR("Protocol input queue controls"), 3766 NULL, 0, NULL, 0, 3767 CTL_NET, pf, ipn, qid, CTL_EOL); 3768 3769 sysctl_createv(clog, 0, NULL, NULL, 3770 CTLFLAG_PERMANENT, 3771 CTLTYPE_QUAD, "len", 3772 SYSCTL_DESCR("Current input queue length"), 3773 len_func, 0, NULL, 0, 3774 CTL_NET, pf, ipn, qid, IFQCTL_LEN, CTL_EOL); 3775 sysctl_createv(clog, 0, NULL, NULL, 3776 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 3777 CTLTYPE_INT, "maxlen", 3778 SYSCTL_DESCR("Maximum allowed input queue length"), 3779 maxlen_func, 0, NULL, 0, 3780 CTL_NET, pf, ipn, qid, IFQCTL_MAXLEN, CTL_EOL); 3781 sysctl_createv(clog, 0, NULL, NULL, 3782 CTLFLAG_PERMANENT, 3783 CTLTYPE_QUAD, "drops", 3784 SYSCTL_DESCR("Packets dropped due to full input queue"), 3785 drops_func, 0, NULL, 0, 3786 CTL_NET, pf, ipn, qid, IFQCTL_DROPS, CTL_EOL); 3787 } 3788 #endif /* INET || INET6 */ 3789 3790 static int 3791 if_sdl_sysctl(SYSCTLFN_ARGS) 3792 { 3793 struct ifnet *ifp; 3794 const struct sockaddr_dl *sdl; 3795 struct psref psref; 3796 int error = 0; 3797 int bound; 3798 3799 if (namelen != 1) 3800 return EINVAL; 3801 3802 bound = curlwp_bind(); 3803 ifp = if_get_byindex(name[0], &psref); 3804 if (ifp == NULL) { 3805 error = ENODEV; 3806 goto out0; 3807 } 3808 3809 sdl = ifp->if_sadl; 3810 if (sdl == NULL) { 3811 *oldlenp = 0; 3812 goto out1; 3813 } 3814 3815 if (oldp == NULL) { 3816 *oldlenp = sdl->sdl_alen; 3817 goto out1; 3818 } 3819 3820 if (*oldlenp >= sdl->sdl_alen) 3821 *oldlenp = sdl->sdl_alen; 3822 error = sysctl_copyout(l, &sdl->sdl_data[sdl->sdl_nlen], oldp, *oldlenp); 3823 out1: 3824 if_put(ifp, &psref); 3825 out0: 3826 curlwp_bindx(bound); 3827 return error; 3828 } 3829 3830 static void 3831 if_sysctl_setup(struct sysctllog **clog) 3832 { 3833 const struct sysctlnode *rnode = NULL; 3834 3835 sysctl_createv(clog, 0, NULL, &rnode, 3836 CTLFLAG_PERMANENT, 3837 CTLTYPE_NODE, "sdl", 3838 SYSCTL_DESCR("Get active link-layer address"), 3839 if_sdl_sysctl, 0, NULL, 0, 3840 CTL_NET, CTL_CREATE, CTL_EOL); 3841 3842 #if defined(INET) 3843 sysctl_net_pktq_setup(NULL, PF_INET); 3844 #endif 3845 #ifdef INET6 3846 if (in6_present) 3847 sysctl_net_pktq_setup(NULL, PF_INET6); 3848 #endif 3849 } 3850
Indexes created Tue Sep 23 14:10:03 GMT 2025