if.c revision 1.422
1 /* $NetBSD: if.c,v 1.422 2018/05/14 02:55:03 ozaki-r 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 1.422 2018/05/14 02:55:03 ozaki-r 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 *); 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_sysctl_setup(NULL); 283 284 #if (defined(INET) || defined(INET6)) 285 encapinit(); 286 #endif 287 288 if_listener = kauth_listen_scope(KAUTH_SCOPE_NETWORK, 289 if_listener_cb, NULL); 290 291 /* interfaces are available, inform socket code */ 292 ifioctl = doifioctl; 293 } 294 295 /* 296 * XXX Initialization before configure(). 297 * XXX hack to get pfil_add_hook working in autoconf. 298 */ 299 void 300 ifinit1(void) 301 { 302 mutex_init(&if_clone_mtx, MUTEX_DEFAULT, IPL_NONE); 303 304 TAILQ_INIT(&ifnet_list); 305 mutex_init(&ifnet_mtx, MUTEX_DEFAULT, IPL_NONE); 306 ifnet_psz = pserialize_create(); 307 ifnet_psref_class = psref_class_create("ifnet", IPL_SOFTNET); 308 ifa_psref_class = psref_class_create("ifa", IPL_SOFTNET); 309 PSLIST_INIT(&ifnet_pslist); 310 311 if_indexlim = 8; 312 313 if_pfil = pfil_head_create(PFIL_TYPE_IFNET, NULL); 314 KASSERT(if_pfil != NULL); 315 316 #if NETHER > 0 || NFDDI > 0 || defined(NETATALK) || NTOKEN > 0 || defined(WLAN) 317 etherinit(); 318 #endif 319 } 320 321 ifnet_t * 322 if_alloc(u_char type) 323 { 324 return kmem_zalloc(sizeof(ifnet_t), KM_SLEEP); 325 } 326 327 void 328 if_free(ifnet_t *ifp) 329 { 330 kmem_free(ifp, sizeof(ifnet_t)); 331 } 332 333 void 334 if_initname(struct ifnet *ifp, const char *name, int unit) 335 { 336 (void)snprintf(ifp->if_xname, sizeof(ifp->if_xname), 337 "%s%d", name, unit); 338 } 339 340 /* 341 * Null routines used while an interface is going away. These routines 342 * just return an error. 343 */ 344 345 int 346 if_nulloutput(struct ifnet *ifp, struct mbuf *m, 347 const struct sockaddr *so, const struct rtentry *rt) 348 { 349 350 return ENXIO; 351 } 352 353 void 354 if_nullinput(struct ifnet *ifp, struct mbuf *m) 355 { 356 357 /* Nothing. */ 358 } 359 360 void 361 if_nullstart(struct ifnet *ifp) 362 { 363 364 /* Nothing. */ 365 } 366 367 int 368 if_nulltransmit(struct ifnet *ifp, struct mbuf *m) 369 { 370 371 m_freem(m); 372 return ENXIO; 373 } 374 375 int 376 if_nullioctl(struct ifnet *ifp, u_long cmd, void *data) 377 { 378 379 return ENXIO; 380 } 381 382 int 383 if_nullinit(struct ifnet *ifp) 384 { 385 386 return ENXIO; 387 } 388 389 void 390 if_nullstop(struct ifnet *ifp, int disable) 391 { 392 393 /* Nothing. */ 394 } 395 396 void 397 if_nullslowtimo(struct ifnet *ifp) 398 { 399 400 /* Nothing. */ 401 } 402 403 void 404 if_nulldrain(struct ifnet *ifp) 405 { 406 407 /* Nothing. */ 408 } 409 410 void 411 if_set_sadl(struct ifnet *ifp, const void *lla, u_char addrlen, bool factory) 412 { 413 struct ifaddr *ifa; 414 struct sockaddr_dl *sdl; 415 416 ifp->if_addrlen = addrlen; 417 if_alloc_sadl(ifp); 418 ifa = ifp->if_dl; 419 sdl = satosdl(ifa->ifa_addr); 420 421 (void)sockaddr_dl_setaddr(sdl, sdl->sdl_len, lla, ifp->if_addrlen); 422 if (factory) { 423 ifp->if_hwdl = ifp->if_dl; 424 ifaref(ifp->if_hwdl); 425 } 426 /* TBD routing socket */ 427 } 428 429 struct ifaddr * 430 if_dl_create(const struct ifnet *ifp, const struct sockaddr_dl **sdlp) 431 { 432 unsigned socksize, ifasize; 433 int addrlen, namelen; 434 struct sockaddr_dl *mask, *sdl; 435 struct ifaddr *ifa; 436 437 namelen = strlen(ifp->if_xname); 438 addrlen = ifp->if_addrlen; 439 socksize = roundup(sockaddr_dl_measure(namelen, addrlen), sizeof(long)); 440 ifasize = sizeof(*ifa) + 2 * socksize; 441 ifa = malloc(ifasize, M_IFADDR, M_WAITOK|M_ZERO); 442 443 sdl = (struct sockaddr_dl *)(ifa + 1); 444 mask = (struct sockaddr_dl *)(socksize + (char *)sdl); 445 446 sockaddr_dl_init(sdl, socksize, ifp->if_index, ifp->if_type, 447 ifp->if_xname, namelen, NULL, addrlen); 448 mask->sdl_family = AF_LINK; 449 mask->sdl_len = sockaddr_dl_measure(namelen, 0); 450 memset(&mask->sdl_data[0], 0xff, namelen); 451 ifa->ifa_rtrequest = link_rtrequest; 452 ifa->ifa_addr = (struct sockaddr *)sdl; 453 ifa->ifa_netmask = (struct sockaddr *)mask; 454 ifa_psref_init(ifa); 455 456 *sdlp = sdl; 457 458 return ifa; 459 } 460 461 static void 462 if_sadl_setrefs(struct ifnet *ifp, struct ifaddr *ifa) 463 { 464 const struct sockaddr_dl *sdl; 465 466 ifp->if_dl = ifa; 467 ifaref(ifa); 468 sdl = satosdl(ifa->ifa_addr); 469 ifp->if_sadl = sdl; 470 } 471 472 /* 473 * Allocate the link level name for the specified interface. This 474 * is an attachment helper. It must be called after ifp->if_addrlen 475 * is initialized, which may not be the case when if_attach() is 476 * called. 477 */ 478 void 479 if_alloc_sadl(struct ifnet *ifp) 480 { 481 struct ifaddr *ifa; 482 const struct sockaddr_dl *sdl; 483 484 /* 485 * If the interface already has a link name, release it 486 * now. This is useful for interfaces that can change 487 * link types, and thus switch link names often. 488 */ 489 if (ifp->if_sadl != NULL) 490 if_free_sadl(ifp); 491 492 ifa = if_dl_create(ifp, &sdl); 493 494 ifa_insert(ifp, ifa); 495 if_sadl_setrefs(ifp, ifa); 496 } 497 498 static void 499 if_deactivate_sadl(struct ifnet *ifp) 500 { 501 struct ifaddr *ifa; 502 503 KASSERT(ifp->if_dl != NULL); 504 505 ifa = ifp->if_dl; 506 507 ifp->if_sadl = NULL; 508 509 ifp->if_dl = NULL; 510 ifafree(ifa); 511 } 512 513 static void 514 if_replace_sadl(struct ifnet *ifp, struct ifaddr *ifa) 515 { 516 struct ifaddr *old; 517 518 KASSERT(ifp->if_dl != NULL); 519 520 old = ifp->if_dl; 521 522 ifaref(ifa); 523 /* XXX Update if_dl and if_sadl atomically */ 524 ifp->if_dl = ifa; 525 ifp->if_sadl = satosdl(ifa->ifa_addr); 526 527 ifafree(old); 528 } 529 530 void 531 if_activate_sadl(struct ifnet *ifp, struct ifaddr *ifa0, 532 const struct sockaddr_dl *sdl) 533 { 534 int s, ss; 535 struct ifaddr *ifa; 536 int bound = curlwp_bind(); 537 538 KASSERT(ifa_held(ifa0)); 539 540 s = splsoftnet(); 541 542 if_replace_sadl(ifp, ifa0); 543 544 ss = pserialize_read_enter(); 545 IFADDR_READER_FOREACH(ifa, ifp) { 546 struct psref psref; 547 ifa_acquire(ifa, &psref); 548 pserialize_read_exit(ss); 549 550 rtinit(ifa, RTM_LLINFO_UPD, 0); 551 552 ss = pserialize_read_enter(); 553 ifa_release(ifa, &psref); 554 } 555 pserialize_read_exit(ss); 556 557 splx(s); 558 curlwp_bindx(bound); 559 } 560 561 /* 562 * Free the link level name for the specified interface. This is 563 * a detach helper. This is called from if_detach(). 564 */ 565 static void 566 if_free_sadl(struct ifnet *ifp) 567 { 568 struct ifaddr *ifa; 569 int s; 570 571 ifa = ifp->if_dl; 572 if (ifa == NULL) { 573 KASSERT(ifp->if_sadl == NULL); 574 return; 575 } 576 577 KASSERT(ifp->if_sadl != NULL); 578 579 s = splsoftnet(); 580 rtinit(ifa, RTM_DELETE, 0); 581 ifa_remove(ifp, ifa); 582 if_deactivate_sadl(ifp); 583 if (ifp->if_hwdl == ifa) { 584 ifafree(ifa); 585 ifp->if_hwdl = NULL; 586 } 587 splx(s); 588 } 589 590 static void 591 if_getindex(ifnet_t *ifp) 592 { 593 bool hitlimit = false; 594 595 ifp->if_index_gen = index_gen++; 596 597 ifp->if_index = if_index; 598 if (ifindex2ifnet == NULL) { 599 if_index++; 600 goto skip; 601 } 602 while (if_byindex(ifp->if_index)) { 603 /* 604 * If we hit USHRT_MAX, we skip back to 0 since 605 * there are a number of places where the value 606 * of if_index or if_index itself is compared 607 * to or stored in an unsigned short. By 608 * jumping back, we won't botch those assignments 609 * or comparisons. 610 */ 611 if (++if_index == 0) { 612 if_index = 1; 613 } else if (if_index == USHRT_MAX) { 614 /* 615 * However, if we have to jump back to 616 * zero *twice* without finding an empty 617 * slot in ifindex2ifnet[], then there 618 * there are too many (>65535) interfaces. 619 */ 620 if (hitlimit) { 621 panic("too many interfaces"); 622 } 623 hitlimit = true; 624 if_index = 1; 625 } 626 ifp->if_index = if_index; 627 } 628 skip: 629 /* 630 * ifindex2ifnet is indexed by if_index. Since if_index will 631 * grow dynamically, it should grow too. 632 */ 633 if (ifindex2ifnet == NULL || ifp->if_index >= if_indexlim) { 634 size_t m, n, oldlim; 635 void *q; 636 637 oldlim = if_indexlim; 638 while (ifp->if_index >= if_indexlim) 639 if_indexlim <<= 1; 640 641 /* grow ifindex2ifnet */ 642 m = oldlim * sizeof(struct ifnet *); 643 n = if_indexlim * sizeof(struct ifnet *); 644 q = malloc(n, M_IFADDR, M_WAITOK|M_ZERO); 645 if (ifindex2ifnet != NULL) { 646 memcpy(q, ifindex2ifnet, m); 647 free(ifindex2ifnet, M_IFADDR); 648 } 649 ifindex2ifnet = (struct ifnet **)q; 650 } 651 ifindex2ifnet[ifp->if_index] = ifp; 652 } 653 654 /* 655 * Initialize an interface and assign an index for it. 656 * 657 * It must be called prior to a device specific attach routine 658 * (e.g., ether_ifattach and ieee80211_ifattach) or if_alloc_sadl, 659 * and be followed by if_register: 660 * 661 * if_initialize(ifp); 662 * ether_ifattach(ifp, enaddr); 663 * if_register(ifp); 664 */ 665 int 666 if_initialize(ifnet_t *ifp) 667 { 668 int rv = 0; 669 670 KASSERT(if_indexlim > 0); 671 TAILQ_INIT(&ifp->if_addrlist); 672 673 /* 674 * Link level name is allocated later by a separate call to 675 * if_alloc_sadl(). 676 */ 677 678 if (ifp->if_snd.ifq_maxlen == 0) 679 ifp->if_snd.ifq_maxlen = ifqmaxlen; 680 681 ifp->if_broadcastaddr = 0; /* reliably crash if used uninitialized */ 682 683 ifp->if_link_state = LINK_STATE_UNKNOWN; 684 ifp->if_link_queue = -1; /* all bits set, see link_state_change() */ 685 686 ifp->if_capenable = 0; 687 ifp->if_csum_flags_tx = 0; 688 ifp->if_csum_flags_rx = 0; 689 690 #ifdef ALTQ 691 ifp->if_snd.altq_type = 0; 692 ifp->if_snd.altq_disc = NULL; 693 ifp->if_snd.altq_flags &= ALTQF_CANTCHANGE; 694 ifp->if_snd.altq_tbr = NULL; 695 ifp->if_snd.altq_ifp = ifp; 696 #endif 697 698 IFQ_LOCK_INIT(&ifp->if_snd); 699 700 ifp->if_pfil = pfil_head_create(PFIL_TYPE_IFNET, ifp); 701 pfil_run_ifhooks(if_pfil, PFIL_IFNET_ATTACH, ifp); 702 703 IF_AFDATA_LOCK_INIT(ifp); 704 705 if (if_is_link_state_changeable(ifp)) { 706 u_int flags = SOFTINT_NET; 707 flags |= if_is_mpsafe(ifp) ? SOFTINT_MPSAFE : 0; 708 ifp->if_link_si = softint_establish(flags, 709 if_link_state_change_si, ifp); 710 if (ifp->if_link_si == NULL) { 711 rv = ENOMEM; 712 goto fail; 713 } 714 } 715 716 PSLIST_ENTRY_INIT(ifp, if_pslist_entry); 717 PSLIST_INIT(&ifp->if_addr_pslist); 718 psref_target_init(&ifp->if_psref, ifnet_psref_class); 719 ifp->if_ioctl_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 720 LIST_INIT(&ifp->if_multiaddrs); 721 722 IFNET_GLOBAL_LOCK(); 723 if_getindex(ifp); 724 IFNET_GLOBAL_UNLOCK(); 725 726 return 0; 727 728 fail: 729 IF_AFDATA_LOCK_DESTROY(ifp); 730 731 pfil_run_ifhooks(if_pfil, PFIL_IFNET_DETACH, ifp); 732 (void)pfil_head_destroy(ifp->if_pfil); 733 734 IFQ_LOCK_DESTROY(&ifp->if_snd); 735 736 return rv; 737 } 738 739 /* 740 * Register an interface to the list of "active" interfaces. 741 */ 742 void 743 if_register(ifnet_t *ifp) 744 { 745 /* 746 * If the driver has not supplied its own if_ioctl, then 747 * supply the default. 748 */ 749 if (ifp->if_ioctl == NULL) 750 ifp->if_ioctl = ifioctl_common; 751 752 sysctl_sndq_setup(&ifp->if_sysctl_log, ifp->if_xname, &ifp->if_snd); 753 754 if (!STAILQ_EMPTY(&domains)) 755 if_attachdomain1(ifp); 756 757 /* Announce the interface. */ 758 rt_ifannouncemsg(ifp, IFAN_ARRIVAL); 759 760 if (ifp->if_slowtimo != NULL) { 761 ifp->if_slowtimo_ch = 762 kmem_zalloc(sizeof(*ifp->if_slowtimo_ch), KM_SLEEP); 763 callout_init(ifp->if_slowtimo_ch, 0); 764 callout_setfunc(ifp->if_slowtimo_ch, if_slowtimo, ifp); 765 if_slowtimo(ifp); 766 } 767 768 if (ifp->if_transmit == NULL || ifp->if_transmit == if_nulltransmit) 769 ifp->if_transmit = if_transmit; 770 771 IFNET_GLOBAL_LOCK(); 772 TAILQ_INSERT_TAIL(&ifnet_list, ifp, if_list); 773 IFNET_WRITER_INSERT_TAIL(ifp); 774 IFNET_GLOBAL_UNLOCK(); 775 } 776 777 /* 778 * The if_percpuq framework 779 * 780 * It allows network device drivers to execute the network stack 781 * in softint (so called softint-based if_input). It utilizes 782 * softint and percpu ifqueue. It doesn't distribute any packets 783 * between CPUs, unlike pktqueue(9). 784 * 785 * Currently we support two options for device drivers to apply the framework: 786 * - Use it implicitly with less changes 787 * - If you use if_attach in driver's _attach function and if_input in 788 * driver's Rx interrupt handler, a packet is queued and a softint handles 789 * the packet implicitly 790 * - Use it explicitly in each driver (recommended) 791 * - You can use if_percpuq_* directly in your driver 792 * - In this case, you need to allocate struct if_percpuq in driver's softc 793 * - See wm(4) as a reference implementation 794 */ 795 796 static void 797 if_percpuq_softint(void *arg) 798 { 799 struct if_percpuq *ipq = arg; 800 struct ifnet *ifp = ipq->ipq_ifp; 801 struct mbuf *m; 802 803 while ((m = if_percpuq_dequeue(ipq)) != NULL) { 804 ifp->if_ipackets++; 805 bpf_mtap(ifp, m); 806 807 ifp->_if_input(ifp, m); 808 } 809 } 810 811 static void 812 if_percpuq_init_ifq(void *p, void *arg __unused, struct cpu_info *ci __unused) 813 { 814 struct ifqueue *const ifq = p; 815 816 memset(ifq, 0, sizeof(*ifq)); 817 ifq->ifq_maxlen = IFQ_MAXLEN; 818 } 819 820 struct if_percpuq * 821 if_percpuq_create(struct ifnet *ifp) 822 { 823 struct if_percpuq *ipq; 824 u_int flags = SOFTINT_NET; 825 826 flags |= if_is_mpsafe(ifp) ? SOFTINT_MPSAFE : 0; 827 828 ipq = kmem_zalloc(sizeof(*ipq), KM_SLEEP); 829 ipq->ipq_ifp = ifp; 830 ipq->ipq_si = softint_establish(flags, if_percpuq_softint, ipq); 831 ipq->ipq_ifqs = percpu_alloc(sizeof(struct ifqueue)); 832 percpu_foreach(ipq->ipq_ifqs, &if_percpuq_init_ifq, NULL); 833 834 sysctl_percpuq_setup(&ifp->if_sysctl_log, ifp->if_xname, ipq); 835 836 return ipq; 837 } 838 839 static struct mbuf * 840 if_percpuq_dequeue(struct if_percpuq *ipq) 841 { 842 struct mbuf *m; 843 struct ifqueue *ifq; 844 int s; 845 846 s = splnet(); 847 ifq = percpu_getref(ipq->ipq_ifqs); 848 IF_DEQUEUE(ifq, m); 849 percpu_putref(ipq->ipq_ifqs); 850 splx(s); 851 852 return m; 853 } 854 855 static void 856 if_percpuq_purge_ifq(void *p, void *arg __unused, struct cpu_info *ci __unused) 857 { 858 struct ifqueue *const ifq = p; 859 860 IF_PURGE(ifq); 861 } 862 863 void 864 if_percpuq_destroy(struct if_percpuq *ipq) 865 { 866 867 /* if_detach may already destroy it */ 868 if (ipq == NULL) 869 return; 870 871 softint_disestablish(ipq->ipq_si); 872 percpu_foreach(ipq->ipq_ifqs, &if_percpuq_purge_ifq, NULL); 873 percpu_free(ipq->ipq_ifqs, sizeof(struct ifqueue)); 874 kmem_free(ipq, sizeof(*ipq)); 875 } 876 877 void 878 if_percpuq_enqueue(struct if_percpuq *ipq, struct mbuf *m) 879 { 880 struct ifqueue *ifq; 881 int s; 882 883 KASSERT(ipq != NULL); 884 885 s = splnet(); 886 ifq = percpu_getref(ipq->ipq_ifqs); 887 if (IF_QFULL(ifq)) { 888 IF_DROP(ifq); 889 percpu_putref(ipq->ipq_ifqs); 890 m_freem(m); 891 goto out; 892 } 893 IF_ENQUEUE(ifq, m); 894 percpu_putref(ipq->ipq_ifqs); 895 896 softint_schedule(ipq->ipq_si); 897 out: 898 splx(s); 899 } 900 901 static void 902 if_percpuq_drops(void *p, void *arg, struct cpu_info *ci __unused) 903 { 904 struct ifqueue *const ifq = p; 905 int *sum = arg; 906 907 *sum += ifq->ifq_drops; 908 } 909 910 static int 911 sysctl_percpuq_drops_handler(SYSCTLFN_ARGS) 912 { 913 struct sysctlnode node; 914 struct if_percpuq *ipq; 915 int sum = 0; 916 int error; 917 918 node = *rnode; 919 ipq = node.sysctl_data; 920 921 percpu_foreach(ipq->ipq_ifqs, if_percpuq_drops, &sum); 922 923 node.sysctl_data = ∑ 924 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 925 if (error != 0 || newp == NULL) 926 return error; 927 928 return 0; 929 } 930 931 static void 932 sysctl_percpuq_setup(struct sysctllog **clog, const char* ifname, 933 struct if_percpuq *ipq) 934 { 935 const struct sysctlnode *cnode, *rnode; 936 937 if (sysctl_createv(clog, 0, NULL, &rnode, 938 CTLFLAG_PERMANENT, 939 CTLTYPE_NODE, "interfaces", 940 SYSCTL_DESCR("Per-interface controls"), 941 NULL, 0, NULL, 0, 942 CTL_NET, CTL_CREATE, CTL_EOL) != 0) 943 goto bad; 944 945 if (sysctl_createv(clog, 0, &rnode, &rnode, 946 CTLFLAG_PERMANENT, 947 CTLTYPE_NODE, ifname, 948 SYSCTL_DESCR("Interface controls"), 949 NULL, 0, NULL, 0, 950 CTL_CREATE, CTL_EOL) != 0) 951 goto bad; 952 953 if (sysctl_createv(clog, 0, &rnode, &rnode, 954 CTLFLAG_PERMANENT, 955 CTLTYPE_NODE, "rcvq", 956 SYSCTL_DESCR("Interface input queue controls"), 957 NULL, 0, NULL, 0, 958 CTL_CREATE, CTL_EOL) != 0) 959 goto bad; 960 961 #ifdef NOTYET 962 /* XXX Should show each per-CPU queue length? */ 963 if (sysctl_createv(clog, 0, &rnode, &rnode, 964 CTLFLAG_PERMANENT, 965 CTLTYPE_INT, "len", 966 SYSCTL_DESCR("Current input queue length"), 967 sysctl_percpuq_len, 0, NULL, 0, 968 CTL_CREATE, CTL_EOL) != 0) 969 goto bad; 970 971 if (sysctl_createv(clog, 0, &rnode, &cnode, 972 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 973 CTLTYPE_INT, "maxlen", 974 SYSCTL_DESCR("Maximum allowed input queue length"), 975 sysctl_percpuq_maxlen_handler, 0, (void *)ipq, 0, 976 CTL_CREATE, CTL_EOL) != 0) 977 goto bad; 978 #endif 979 980 if (sysctl_createv(clog, 0, &rnode, &cnode, 981 CTLFLAG_PERMANENT, 982 CTLTYPE_INT, "drops", 983 SYSCTL_DESCR("Total packets dropped due to full input queue"), 984 sysctl_percpuq_drops_handler, 0, (void *)ipq, 0, 985 CTL_CREATE, CTL_EOL) != 0) 986 goto bad; 987 988 return; 989 bad: 990 printf("%s: could not attach sysctl nodes\n", ifname); 991 return; 992 } 993 994 /* 995 * The deferred if_start framework 996 * 997 * The common APIs to defer if_start to softint when if_start is requested 998 * from a device driver running in hardware interrupt context. 999 */ 1000 /* 1001 * Call ifp->if_start (or equivalent) in a dedicated softint for 1002 * deferred if_start. 1003 */ 1004 static void 1005 if_deferred_start_softint(void *arg) 1006 { 1007 struct if_deferred_start *ids = arg; 1008 struct ifnet *ifp = ids->ids_ifp; 1009 1010 ids->ids_if_start(ifp); 1011 } 1012 1013 /* 1014 * The default callback function for deferred if_start. 1015 */ 1016 static void 1017 if_deferred_start_common(struct ifnet *ifp) 1018 { 1019 int s; 1020 1021 s = splnet(); 1022 if_start_lock(ifp); 1023 splx(s); 1024 } 1025 1026 static inline bool 1027 if_snd_is_used(struct ifnet *ifp) 1028 { 1029 1030 return ifp->if_transmit == NULL || ifp->if_transmit == if_nulltransmit || 1031 ALTQ_IS_ENABLED(&ifp->if_snd); 1032 } 1033 1034 /* 1035 * Schedule deferred if_start. 1036 */ 1037 void 1038 if_schedule_deferred_start(struct ifnet *ifp) 1039 { 1040 1041 KASSERT(ifp->if_deferred_start != NULL); 1042 1043 if (if_snd_is_used(ifp) && IFQ_IS_EMPTY(&ifp->if_snd)) 1044 return; 1045 1046 softint_schedule(ifp->if_deferred_start->ids_si); 1047 } 1048 1049 /* 1050 * Create an instance of deferred if_start. A driver should call the function 1051 * only if the driver needs deferred if_start. Drivers can setup their own 1052 * deferred if_start function via 2nd argument. 1053 */ 1054 void 1055 if_deferred_start_init(struct ifnet *ifp, void (*func)(struct ifnet *)) 1056 { 1057 struct if_deferred_start *ids; 1058 1059 ids = kmem_zalloc(sizeof(*ids), KM_SLEEP); 1060 ids->ids_ifp = ifp; 1061 ids->ids_si = softint_establish(SOFTINT_NET|SOFTINT_MPSAFE, 1062 if_deferred_start_softint, ids); 1063 if (func != NULL) 1064 ids->ids_if_start = func; 1065 else 1066 ids->ids_if_start = if_deferred_start_common; 1067 1068 ifp->if_deferred_start = ids; 1069 } 1070 1071 static void 1072 if_deferred_start_destroy(struct ifnet *ifp) 1073 { 1074 1075 if (ifp->if_deferred_start == NULL) 1076 return; 1077 1078 softint_disestablish(ifp->if_deferred_start->ids_si); 1079 kmem_free(ifp->if_deferred_start, sizeof(*ifp->if_deferred_start)); 1080 ifp->if_deferred_start = NULL; 1081 } 1082 1083 /* 1084 * The common interface input routine that is called by device drivers, 1085 * which should be used only when the driver's rx handler already runs 1086 * in softint. 1087 */ 1088 void 1089 if_input(struct ifnet *ifp, struct mbuf *m) 1090 { 1091 1092 KASSERT(ifp->if_percpuq == NULL); 1093 KASSERT(!cpu_intr_p()); 1094 1095 ifp->if_ipackets++; 1096 bpf_mtap(ifp, m); 1097 1098 ifp->_if_input(ifp, m); 1099 } 1100 1101 /* 1102 * DEPRECATED. Use if_initialize and if_register instead. 1103 * See the above comment of if_initialize. 1104 * 1105 * Note that it implicitly enables if_percpuq to make drivers easy to 1106 * migrate softint-based if_input without much changes. If you don't 1107 * want to enable it, use if_initialize instead. 1108 */ 1109 int 1110 if_attach(ifnet_t *ifp) 1111 { 1112 int rv; 1113 1114 rv = if_initialize(ifp); 1115 if (rv != 0) 1116 return rv; 1117 1118 ifp->if_percpuq = if_percpuq_create(ifp); 1119 if_register(ifp); 1120 1121 return 0; 1122 } 1123 1124 void 1125 if_attachdomain(void) 1126 { 1127 struct ifnet *ifp; 1128 int s; 1129 int bound = curlwp_bind(); 1130 1131 s = pserialize_read_enter(); 1132 IFNET_READER_FOREACH(ifp) { 1133 struct psref psref; 1134 psref_acquire(&psref, &ifp->if_psref, ifnet_psref_class); 1135 pserialize_read_exit(s); 1136 if_attachdomain1(ifp); 1137 s = pserialize_read_enter(); 1138 psref_release(&psref, &ifp->if_psref, ifnet_psref_class); 1139 } 1140 pserialize_read_exit(s); 1141 curlwp_bindx(bound); 1142 } 1143 1144 static void 1145 if_attachdomain1(struct ifnet *ifp) 1146 { 1147 struct domain *dp; 1148 int s; 1149 1150 s = splsoftnet(); 1151 1152 /* address family dependent data region */ 1153 memset(ifp->if_afdata, 0, sizeof(ifp->if_afdata)); 1154 DOMAIN_FOREACH(dp) { 1155 if (dp->dom_ifattach != NULL) 1156 ifp->if_afdata[dp->dom_family] = 1157 (*dp->dom_ifattach)(ifp); 1158 } 1159 1160 splx(s); 1161 } 1162 1163 /* 1164 * Deactivate an interface. This points all of the procedure 1165 * handles at error stubs. May be called from interrupt context. 1166 */ 1167 void 1168 if_deactivate(struct ifnet *ifp) 1169 { 1170 int s; 1171 1172 s = splsoftnet(); 1173 1174 ifp->if_output = if_nulloutput; 1175 ifp->_if_input = if_nullinput; 1176 ifp->if_start = if_nullstart; 1177 ifp->if_transmit = if_nulltransmit; 1178 ifp->if_ioctl = if_nullioctl; 1179 ifp->if_init = if_nullinit; 1180 ifp->if_stop = if_nullstop; 1181 ifp->if_slowtimo = if_nullslowtimo; 1182 ifp->if_drain = if_nulldrain; 1183 1184 /* No more packets may be enqueued. */ 1185 ifp->if_snd.ifq_maxlen = 0; 1186 1187 splx(s); 1188 } 1189 1190 bool 1191 if_is_deactivated(const struct ifnet *ifp) 1192 { 1193 1194 return ifp->if_output == if_nulloutput; 1195 } 1196 1197 void 1198 if_purgeaddrs(struct ifnet *ifp, int family, void (*purgeaddr)(struct ifaddr *)) 1199 { 1200 struct ifaddr *ifa, *nifa; 1201 int s; 1202 1203 s = pserialize_read_enter(); 1204 for (ifa = IFADDR_READER_FIRST(ifp); ifa; ifa = nifa) { 1205 nifa = IFADDR_READER_NEXT(ifa); 1206 if (ifa->ifa_addr->sa_family != family) 1207 continue; 1208 pserialize_read_exit(s); 1209 1210 (*purgeaddr)(ifa); 1211 1212 s = pserialize_read_enter(); 1213 } 1214 pserialize_read_exit(s); 1215 } 1216 1217 #ifdef IFAREF_DEBUG 1218 static struct ifaddr **ifa_list; 1219 static int ifa_list_size; 1220 1221 /* Depends on only one if_attach runs at once */ 1222 static void 1223 if_build_ifa_list(struct ifnet *ifp) 1224 { 1225 struct ifaddr *ifa; 1226 int i; 1227 1228 KASSERT(ifa_list == NULL); 1229 KASSERT(ifa_list_size == 0); 1230 1231 IFADDR_READER_FOREACH(ifa, ifp) 1232 ifa_list_size++; 1233 1234 ifa_list = kmem_alloc(sizeof(*ifa) * ifa_list_size, KM_SLEEP); 1235 i = 0; 1236 IFADDR_READER_FOREACH(ifa, ifp) { 1237 ifa_list[i++] = ifa; 1238 ifaref(ifa); 1239 } 1240 } 1241 1242 static void 1243 if_check_and_free_ifa_list(struct ifnet *ifp) 1244 { 1245 int i; 1246 struct ifaddr *ifa; 1247 1248 if (ifa_list == NULL) 1249 return; 1250 1251 for (i = 0; i < ifa_list_size; i++) { 1252 char buf[64]; 1253 1254 ifa = ifa_list[i]; 1255 sockaddr_format(ifa->ifa_addr, buf, sizeof(buf)); 1256 if (ifa->ifa_refcnt > 1) { 1257 log(LOG_WARNING, 1258 "ifa(%s) still referenced (refcnt=%d)\n", 1259 buf, ifa->ifa_refcnt - 1); 1260 } else 1261 log(LOG_DEBUG, 1262 "ifa(%s) not referenced (refcnt=%d)\n", 1263 buf, ifa->ifa_refcnt - 1); 1264 ifafree(ifa); 1265 } 1266 1267 kmem_free(ifa_list, sizeof(*ifa) * ifa_list_size); 1268 ifa_list = NULL; 1269 ifa_list_size = 0; 1270 } 1271 #endif 1272 1273 /* 1274 * Detach an interface from the list of "active" interfaces, 1275 * freeing any resources as we go along. 1276 * 1277 * NOTE: This routine must be called with a valid thread context, 1278 * as it may block. 1279 */ 1280 void 1281 if_detach(struct ifnet *ifp) 1282 { 1283 struct socket so; 1284 struct ifaddr *ifa; 1285 #ifdef IFAREF_DEBUG 1286 struct ifaddr *last_ifa = NULL; 1287 #endif 1288 struct domain *dp; 1289 const struct protosw *pr; 1290 int s, i, family, purged; 1291 uint64_t xc; 1292 1293 #ifdef IFAREF_DEBUG 1294 if_build_ifa_list(ifp); 1295 #endif 1296 /* 1297 * XXX It's kind of lame that we have to have the 1298 * XXX socket structure... 1299 */ 1300 memset(&so, 0, sizeof(so)); 1301 1302 s = splnet(); 1303 1304 sysctl_teardown(&ifp->if_sysctl_log); 1305 IFNET_LOCK(ifp); 1306 if_deactivate(ifp); 1307 IFNET_UNLOCK(ifp); 1308 1309 if (ifp->if_slowtimo != NULL && ifp->if_slowtimo_ch != NULL) { 1310 ifp->if_slowtimo = NULL; 1311 callout_halt(ifp->if_slowtimo_ch, NULL); 1312 callout_destroy(ifp->if_slowtimo_ch); 1313 kmem_free(ifp->if_slowtimo_ch, sizeof(*ifp->if_slowtimo_ch)); 1314 } 1315 if_deferred_start_destroy(ifp); 1316 1317 /* 1318 * Do an if_down() to give protocols a chance to do something. 1319 */ 1320 if_down_deactivated(ifp); 1321 1322 #ifdef ALTQ 1323 if (ALTQ_IS_ENABLED(&ifp->if_snd)) 1324 altq_disable(&ifp->if_snd); 1325 if (ALTQ_IS_ATTACHED(&ifp->if_snd)) 1326 altq_detach(&ifp->if_snd); 1327 #endif 1328 1329 #if NCARP > 0 1330 /* Remove the interface from any carp group it is a part of. */ 1331 if (ifp->if_carp != NULL && ifp->if_type != IFT_CARP) 1332 carp_ifdetach(ifp); 1333 #endif 1334 1335 /* 1336 * Rip all the addresses off the interface. This should make 1337 * all of the routes go away. 1338 * 1339 * pr_usrreq calls can remove an arbitrary number of ifaddrs 1340 * from the list, including our "cursor", ifa. For safety, 1341 * and to honor the TAILQ abstraction, I just restart the 1342 * loop after each removal. Note that the loop will exit 1343 * when all of the remaining ifaddrs belong to the AF_LINK 1344 * family. I am counting on the historical fact that at 1345 * least one pr_usrreq in each address domain removes at 1346 * least one ifaddr. 1347 */ 1348 again: 1349 /* 1350 * At this point, no other one tries to remove ifa in the list, 1351 * so we don't need to take a lock or psref. Avoid using 1352 * IFADDR_READER_FOREACH to pass over an inspection of contract 1353 * violations of pserialize. 1354 */ 1355 IFADDR_WRITER_FOREACH(ifa, ifp) { 1356 family = ifa->ifa_addr->sa_family; 1357 #ifdef IFAREF_DEBUG 1358 printf("if_detach: ifaddr %p, family %d, refcnt %d\n", 1359 ifa, family, ifa->ifa_refcnt); 1360 if (last_ifa != NULL && ifa == last_ifa) 1361 panic("if_detach: loop detected"); 1362 last_ifa = ifa; 1363 #endif 1364 if (family == AF_LINK) 1365 continue; 1366 dp = pffinddomain(family); 1367 KASSERTMSG(dp != NULL, "no domain for AF %d", family); 1368 /* 1369 * XXX These PURGEIF calls are redundant with the 1370 * purge-all-families calls below, but are left in for 1371 * now both to make a smaller change, and to avoid 1372 * unplanned interactions with clearing of 1373 * ifp->if_addrlist. 1374 */ 1375 purged = 0; 1376 for (pr = dp->dom_protosw; 1377 pr < dp->dom_protoswNPROTOSW; pr++) { 1378 so.so_proto = pr; 1379 if (pr->pr_usrreqs) { 1380 (void) (*pr->pr_usrreqs->pr_purgeif)(&so, ifp); 1381 purged = 1; 1382 } 1383 } 1384 if (purged == 0) { 1385 /* 1386 * XXX What's really the best thing to do 1387 * XXX here? --thorpej (at) NetBSD.org 1388 */ 1389 printf("if_detach: WARNING: AF %d not purged\n", 1390 family); 1391 ifa_remove(ifp, ifa); 1392 } 1393 goto again; 1394 } 1395 1396 if_free_sadl(ifp); 1397 1398 /* Delete stray routes from the routing table. */ 1399 for (i = 0; i <= AF_MAX; i++) 1400 rt_delete_matched_entries(i, if_delroute_matcher, ifp); 1401 1402 DOMAIN_FOREACH(dp) { 1403 if (dp->dom_ifdetach != NULL && ifp->if_afdata[dp->dom_family]) 1404 { 1405 void *p = ifp->if_afdata[dp->dom_family]; 1406 if (p) { 1407 ifp->if_afdata[dp->dom_family] = NULL; 1408 (*dp->dom_ifdetach)(ifp, p); 1409 } 1410 } 1411 1412 /* 1413 * One would expect multicast memberships (INET and 1414 * INET6) on UDP sockets to be purged by the PURGEIF 1415 * calls above, but if all addresses were removed from 1416 * the interface prior to destruction, the calls will 1417 * not be made (e.g. ppp, for which pppd(8) generally 1418 * removes addresses before destroying the interface). 1419 * Because there is no invariant that multicast 1420 * memberships only exist for interfaces with IPv4 1421 * addresses, we must call PURGEIF regardless of 1422 * addresses. (Protocols which might store ifnet 1423 * pointers are marked with PR_PURGEIF.) 1424 */ 1425 for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) { 1426 so.so_proto = pr; 1427 if (pr->pr_usrreqs && pr->pr_flags & PR_PURGEIF) 1428 (void)(*pr->pr_usrreqs->pr_purgeif)(&so, ifp); 1429 } 1430 } 1431 1432 /* Wait for all readers to drain before freeing. */ 1433 IFNET_GLOBAL_LOCK(); 1434 ifindex2ifnet[ifp->if_index] = NULL; 1435 TAILQ_REMOVE(&ifnet_list, ifp, if_list); 1436 IFNET_WRITER_REMOVE(ifp); 1437 pserialize_perform(ifnet_psz); 1438 IFNET_GLOBAL_UNLOCK(); 1439 1440 psref_target_destroy(&ifp->if_psref, ifnet_psref_class); 1441 PSLIST_ENTRY_DESTROY(ifp, if_pslist_entry); 1442 1443 pfil_run_ifhooks(if_pfil, PFIL_IFNET_DETACH, ifp); 1444 (void)pfil_head_destroy(ifp->if_pfil); 1445 1446 /* Announce that the interface is gone. */ 1447 rt_ifannouncemsg(ifp, IFAN_DEPARTURE); 1448 1449 IF_AFDATA_LOCK_DESTROY(ifp); 1450 1451 if (if_is_link_state_changeable(ifp)) { 1452 softint_disestablish(ifp->if_link_si); 1453 ifp->if_link_si = NULL; 1454 } 1455 1456 /* 1457 * remove packets that came from ifp, from software interrupt queues. 1458 */ 1459 DOMAIN_FOREACH(dp) { 1460 for (i = 0; i < __arraycount(dp->dom_ifqueues); i++) { 1461 struct ifqueue *iq = dp->dom_ifqueues[i]; 1462 if (iq == NULL) 1463 break; 1464 dp->dom_ifqueues[i] = NULL; 1465 if_detach_queues(ifp, iq); 1466 } 1467 } 1468 1469 /* 1470 * IP queues have to be processed separately: net-queue barrier 1471 * ensures that the packets are dequeued while a cross-call will 1472 * ensure that the interrupts have completed. FIXME: not quite.. 1473 */ 1474 #ifdef INET 1475 pktq_barrier(ip_pktq); 1476 #endif 1477 #ifdef INET6 1478 if (in6_present) 1479 pktq_barrier(ip6_pktq); 1480 #endif 1481 xc = xc_broadcast(0, (xcfunc_t)nullop, NULL, NULL); 1482 xc_wait(xc); 1483 1484 if (ifp->if_percpuq != NULL) { 1485 if_percpuq_destroy(ifp->if_percpuq); 1486 ifp->if_percpuq = NULL; 1487 } 1488 1489 mutex_obj_free(ifp->if_ioctl_lock); 1490 ifp->if_ioctl_lock = NULL; 1491 mutex_obj_free(ifp->if_snd.ifq_lock); 1492 1493 splx(s); 1494 1495 #ifdef IFAREF_DEBUG 1496 if_check_and_free_ifa_list(ifp); 1497 #endif 1498 } 1499 1500 static void 1501 if_detach_queues(struct ifnet *ifp, struct ifqueue *q) 1502 { 1503 struct mbuf *m, *prev, *next; 1504 1505 prev = NULL; 1506 for (m = q->ifq_head; m != NULL; m = next) { 1507 KASSERT((m->m_flags & M_PKTHDR) != 0); 1508 1509 next = m->m_nextpkt; 1510 if (m->m_pkthdr.rcvif_index != ifp->if_index) { 1511 prev = m; 1512 continue; 1513 } 1514 1515 if (prev != NULL) 1516 prev->m_nextpkt = m->m_nextpkt; 1517 else 1518 q->ifq_head = m->m_nextpkt; 1519 if (q->ifq_tail == m) 1520 q->ifq_tail = prev; 1521 q->ifq_len--; 1522 1523 m->m_nextpkt = NULL; 1524 m_freem(m); 1525 IF_DROP(q); 1526 } 1527 } 1528 1529 /* 1530 * Callback for a radix tree walk to delete all references to an 1531 * ifnet. 1532 */ 1533 static int 1534 if_delroute_matcher(struct rtentry *rt, void *v) 1535 { 1536 struct ifnet *ifp = (struct ifnet *)v; 1537 1538 if (rt->rt_ifp == ifp) 1539 return 1; 1540 else 1541 return 0; 1542 } 1543 1544 /* 1545 * Create a clone network interface. 1546 */ 1547 static int 1548 if_clone_create(const char *name) 1549 { 1550 struct if_clone *ifc; 1551 int unit; 1552 struct ifnet *ifp; 1553 struct psref psref; 1554 1555 KASSERT(mutex_owned(&if_clone_mtx)); 1556 1557 ifc = if_clone_lookup(name, &unit); 1558 if (ifc == NULL) 1559 return EINVAL; 1560 1561 ifp = if_get(name, &psref); 1562 if (ifp != NULL) { 1563 if_put(ifp, &psref); 1564 return EEXIST; 1565 } 1566 1567 return (*ifc->ifc_create)(ifc, unit); 1568 } 1569 1570 /* 1571 * Destroy a clone network interface. 1572 */ 1573 static int 1574 if_clone_destroy(const char *name) 1575 { 1576 struct if_clone *ifc; 1577 struct ifnet *ifp; 1578 struct psref psref; 1579 1580 KASSERT(mutex_owned(&if_clone_mtx)); 1581 1582 ifc = if_clone_lookup(name, NULL); 1583 if (ifc == NULL) 1584 return EINVAL; 1585 1586 if (ifc->ifc_destroy == NULL) 1587 return EOPNOTSUPP; 1588 1589 ifp = if_get(name, &psref); 1590 if (ifp == NULL) 1591 return ENXIO; 1592 1593 /* We have to disable ioctls here */ 1594 IFNET_LOCK(ifp); 1595 ifp->if_ioctl = if_nullioctl; 1596 IFNET_UNLOCK(ifp); 1597 1598 /* 1599 * We cannot call ifc_destroy with holding ifp. 1600 * Releasing ifp here is safe thanks to if_clone_mtx. 1601 */ 1602 if_put(ifp, &psref); 1603 1604 return (*ifc->ifc_destroy)(ifp); 1605 } 1606 1607 static bool 1608 if_is_unit(const char *name) 1609 { 1610 1611 while(*name != '\0') { 1612 if (*name < '0' || *name > '9') 1613 return false; 1614 name++; 1615 } 1616 1617 return true; 1618 } 1619 1620 /* 1621 * Look up a network interface cloner. 1622 */ 1623 static struct if_clone * 1624 if_clone_lookup(const char *name, int *unitp) 1625 { 1626 struct if_clone *ifc; 1627 const char *cp; 1628 char *dp, ifname[IFNAMSIZ + 3]; 1629 int unit; 1630 1631 KASSERT(mutex_owned(&if_clone_mtx)); 1632 1633 strcpy(ifname, "if_"); 1634 /* separate interface name from unit */ 1635 /* TODO: search unit number from backward */ 1636 for (dp = ifname + 3, cp = name; cp - name < IFNAMSIZ && 1637 *cp && !if_is_unit(cp);) 1638 *dp++ = *cp++; 1639 1640 if (cp == name || cp - name == IFNAMSIZ || !*cp) 1641 return NULL; /* No name or unit number */ 1642 *dp++ = '\0'; 1643 1644 again: 1645 LIST_FOREACH(ifc, &if_cloners, ifc_list) { 1646 if (strcmp(ifname + 3, ifc->ifc_name) == 0) 1647 break; 1648 } 1649 1650 if (ifc == NULL) { 1651 int error; 1652 if (*ifname == '\0') 1653 return NULL; 1654 mutex_exit(&if_clone_mtx); 1655 error = module_autoload(ifname, MODULE_CLASS_DRIVER); 1656 mutex_enter(&if_clone_mtx); 1657 if (error) 1658 return NULL; 1659 *ifname = '\0'; 1660 goto again; 1661 } 1662 1663 unit = 0; 1664 while (cp - name < IFNAMSIZ && *cp) { 1665 if (*cp < '0' || *cp > '9' || unit >= INT_MAX / 10) { 1666 /* Bogus unit number. */ 1667 return NULL; 1668 } 1669 unit = (unit * 10) + (*cp++ - '0'); 1670 } 1671 1672 if (unitp != NULL) 1673 *unitp = unit; 1674 return ifc; 1675 } 1676 1677 /* 1678 * Register a network interface cloner. 1679 */ 1680 void 1681 if_clone_attach(struct if_clone *ifc) 1682 { 1683 1684 mutex_enter(&if_clone_mtx); 1685 LIST_INSERT_HEAD(&if_cloners, ifc, ifc_list); 1686 if_cloners_count++; 1687 mutex_exit(&if_clone_mtx); 1688 } 1689 1690 /* 1691 * Unregister a network interface cloner. 1692 */ 1693 void 1694 if_clone_detach(struct if_clone *ifc) 1695 { 1696 1697 mutex_enter(&if_clone_mtx); 1698 LIST_REMOVE(ifc, ifc_list); 1699 if_cloners_count--; 1700 mutex_exit(&if_clone_mtx); 1701 } 1702 1703 /* 1704 * Provide list of interface cloners to userspace. 1705 */ 1706 int 1707 if_clone_list(int buf_count, char *buffer, int *total) 1708 { 1709 char outbuf[IFNAMSIZ], *dst; 1710 struct if_clone *ifc; 1711 int count, error = 0; 1712 1713 mutex_enter(&if_clone_mtx); 1714 *total = if_cloners_count; 1715 if ((dst = buffer) == NULL) { 1716 /* Just asking how many there are. */ 1717 goto out; 1718 } 1719 1720 if (buf_count < 0) { 1721 error = EINVAL; 1722 goto out; 1723 } 1724 1725 count = (if_cloners_count < buf_count) ? 1726 if_cloners_count : buf_count; 1727 1728 for (ifc = LIST_FIRST(&if_cloners); ifc != NULL && count != 0; 1729 ifc = LIST_NEXT(ifc, ifc_list), count--, dst += IFNAMSIZ) { 1730 (void)strncpy(outbuf, ifc->ifc_name, sizeof(outbuf)); 1731 if (outbuf[sizeof(outbuf) - 1] != '\0') { 1732 error = ENAMETOOLONG; 1733 goto out; 1734 } 1735 error = copyout(outbuf, dst, sizeof(outbuf)); 1736 if (error != 0) 1737 break; 1738 } 1739 1740 out: 1741 mutex_exit(&if_clone_mtx); 1742 return error; 1743 } 1744 1745 void 1746 ifa_psref_init(struct ifaddr *ifa) 1747 { 1748 1749 psref_target_init(&ifa->ifa_psref, ifa_psref_class); 1750 } 1751 1752 void 1753 ifaref(struct ifaddr *ifa) 1754 { 1755 KASSERT(!ISSET(ifa->ifa_flags, IFA_DESTROYING)); 1756 ifa->ifa_refcnt++; 1757 } 1758 1759 void 1760 ifafree(struct ifaddr *ifa) 1761 { 1762 KASSERT(ifa != NULL); 1763 KASSERT(ifa->ifa_refcnt > 0); 1764 1765 if (--ifa->ifa_refcnt == 0) { 1766 free(ifa, M_IFADDR); 1767 } 1768 } 1769 1770 bool 1771 ifa_is_destroying(struct ifaddr *ifa) 1772 { 1773 1774 return ISSET(ifa->ifa_flags, IFA_DESTROYING); 1775 } 1776 1777 void 1778 ifa_insert(struct ifnet *ifp, struct ifaddr *ifa) 1779 { 1780 1781 ifa->ifa_ifp = ifp; 1782 1783 /* 1784 * Check MP-safety for IFEF_MPSAFE drivers. 1785 * Check !IFF_RUNNING for initialization routines that normally don't 1786 * take IFNET_LOCK but it's safe because there is no competitor. 1787 * XXX there are false positive cases because IFF_RUNNING can be off on 1788 * if_stop. 1789 */ 1790 KASSERT(!if_is_mpsafe(ifp) || !ISSET(ifp->if_flags, IFF_RUNNING) || 1791 IFNET_LOCKED(ifp)); 1792 1793 TAILQ_INSERT_TAIL(&ifp->if_addrlist, ifa, ifa_list); 1794 IFADDR_ENTRY_INIT(ifa); 1795 IFADDR_WRITER_INSERT_TAIL(ifp, ifa); 1796 1797 ifaref(ifa); 1798 } 1799 1800 void 1801 ifa_remove(struct ifnet *ifp, struct ifaddr *ifa) 1802 { 1803 1804 KASSERT(ifa->ifa_ifp == ifp); 1805 /* 1806 * Check MP-safety for IFEF_MPSAFE drivers. 1807 * if_is_deactivated indicates ifa_remove is called form if_detach 1808 * where is safe even if IFNET_LOCK isn't held. 1809 */ 1810 KASSERT(!if_is_mpsafe(ifp) || if_is_deactivated(ifp) || IFNET_LOCKED(ifp)); 1811 1812 TAILQ_REMOVE(&ifp->if_addrlist, ifa, ifa_list); 1813 IFADDR_WRITER_REMOVE(ifa); 1814 #ifdef NET_MPSAFE 1815 IFNET_GLOBAL_LOCK(); 1816 pserialize_perform(ifnet_psz); 1817 IFNET_GLOBAL_UNLOCK(); 1818 #endif 1819 1820 #ifdef NET_MPSAFE 1821 psref_target_destroy(&ifa->ifa_psref, ifa_psref_class); 1822 #endif 1823 IFADDR_ENTRY_DESTROY(ifa); 1824 ifafree(ifa); 1825 } 1826 1827 void 1828 ifa_acquire(struct ifaddr *ifa, struct psref *psref) 1829 { 1830 1831 psref_acquire(psref, &ifa->ifa_psref, ifa_psref_class); 1832 } 1833 1834 void 1835 ifa_release(struct ifaddr *ifa, struct psref *psref) 1836 { 1837 1838 if (ifa == NULL) 1839 return; 1840 1841 psref_release(psref, &ifa->ifa_psref, ifa_psref_class); 1842 } 1843 1844 bool 1845 ifa_held(struct ifaddr *ifa) 1846 { 1847 1848 return psref_held(&ifa->ifa_psref, ifa_psref_class); 1849 } 1850 1851 static inline int 1852 equal(const struct sockaddr *sa1, const struct sockaddr *sa2) 1853 { 1854 return sockaddr_cmp(sa1, sa2) == 0; 1855 } 1856 1857 /* 1858 * Locate an interface based on a complete address. 1859 */ 1860 /*ARGSUSED*/ 1861 struct ifaddr * 1862 ifa_ifwithaddr(const struct sockaddr *addr) 1863 { 1864 struct ifnet *ifp; 1865 struct ifaddr *ifa; 1866 1867 IFNET_READER_FOREACH(ifp) { 1868 if (if_is_deactivated(ifp)) 1869 continue; 1870 IFADDR_READER_FOREACH(ifa, ifp) { 1871 if (ifa->ifa_addr->sa_family != addr->sa_family) 1872 continue; 1873 if (equal(addr, ifa->ifa_addr)) 1874 return ifa; 1875 if ((ifp->if_flags & IFF_BROADCAST) && 1876 ifa->ifa_broadaddr && 1877 /* IP6 doesn't have broadcast */ 1878 ifa->ifa_broadaddr->sa_len != 0 && 1879 equal(ifa->ifa_broadaddr, addr)) 1880 return ifa; 1881 } 1882 } 1883 return NULL; 1884 } 1885 1886 struct ifaddr * 1887 ifa_ifwithaddr_psref(const struct sockaddr *addr, struct psref *psref) 1888 { 1889 struct ifaddr *ifa; 1890 int s = pserialize_read_enter(); 1891 1892 ifa = ifa_ifwithaddr(addr); 1893 if (ifa != NULL) 1894 ifa_acquire(ifa, psref); 1895 pserialize_read_exit(s); 1896 1897 return ifa; 1898 } 1899 1900 /* 1901 * Locate the point to point interface with a given destination address. 1902 */ 1903 /*ARGSUSED*/ 1904 struct ifaddr * 1905 ifa_ifwithdstaddr(const struct sockaddr *addr) 1906 { 1907 struct ifnet *ifp; 1908 struct ifaddr *ifa; 1909 1910 IFNET_READER_FOREACH(ifp) { 1911 if (if_is_deactivated(ifp)) 1912 continue; 1913 if ((ifp->if_flags & IFF_POINTOPOINT) == 0) 1914 continue; 1915 IFADDR_READER_FOREACH(ifa, ifp) { 1916 if (ifa->ifa_addr->sa_family != addr->sa_family || 1917 ifa->ifa_dstaddr == NULL) 1918 continue; 1919 if (equal(addr, ifa->ifa_dstaddr)) 1920 return ifa; 1921 } 1922 } 1923 1924 return NULL; 1925 } 1926 1927 struct ifaddr * 1928 ifa_ifwithdstaddr_psref(const struct sockaddr *addr, struct psref *psref) 1929 { 1930 struct ifaddr *ifa; 1931 int s; 1932 1933 s = pserialize_read_enter(); 1934 ifa = ifa_ifwithdstaddr(addr); 1935 if (ifa != NULL) 1936 ifa_acquire(ifa, psref); 1937 pserialize_read_exit(s); 1938 1939 return ifa; 1940 } 1941 1942 /* 1943 * Find an interface on a specific network. If many, choice 1944 * is most specific found. 1945 */ 1946 struct ifaddr * 1947 ifa_ifwithnet(const struct sockaddr *addr) 1948 { 1949 struct ifnet *ifp; 1950 struct ifaddr *ifa, *ifa_maybe = NULL; 1951 const struct sockaddr_dl *sdl; 1952 u_int af = addr->sa_family; 1953 const char *addr_data = addr->sa_data, *cplim; 1954 1955 if (af == AF_LINK) { 1956 sdl = satocsdl(addr); 1957 if (sdl->sdl_index && sdl->sdl_index < if_indexlim && 1958 ifindex2ifnet[sdl->sdl_index] && 1959 !if_is_deactivated(ifindex2ifnet[sdl->sdl_index])) { 1960 return ifindex2ifnet[sdl->sdl_index]->if_dl; 1961 } 1962 } 1963 #ifdef NETATALK 1964 if (af == AF_APPLETALK) { 1965 const struct sockaddr_at *sat, *sat2; 1966 sat = (const struct sockaddr_at *)addr; 1967 IFNET_READER_FOREACH(ifp) { 1968 if (if_is_deactivated(ifp)) 1969 continue; 1970 ifa = at_ifawithnet((const struct sockaddr_at *)addr, ifp); 1971 if (ifa == NULL) 1972 continue; 1973 sat2 = (struct sockaddr_at *)ifa->ifa_addr; 1974 if (sat2->sat_addr.s_net == sat->sat_addr.s_net) 1975 return ifa; /* exact match */ 1976 if (ifa_maybe == NULL) { 1977 /* else keep the if with the right range */ 1978 ifa_maybe = ifa; 1979 } 1980 } 1981 return ifa_maybe; 1982 } 1983 #endif 1984 IFNET_READER_FOREACH(ifp) { 1985 if (if_is_deactivated(ifp)) 1986 continue; 1987 IFADDR_READER_FOREACH(ifa, ifp) { 1988 const char *cp, *cp2, *cp3; 1989 1990 if (ifa->ifa_addr->sa_family != af || 1991 ifa->ifa_netmask == NULL) 1992 next: continue; 1993 cp = addr_data; 1994 cp2 = ifa->ifa_addr->sa_data; 1995 cp3 = ifa->ifa_netmask->sa_data; 1996 cplim = (const char *)ifa->ifa_netmask + 1997 ifa->ifa_netmask->sa_len; 1998 while (cp3 < cplim) { 1999 if ((*cp++ ^ *cp2++) & *cp3++) { 2000 /* want to continue for() loop */ 2001 goto next; 2002 } 2003 } 2004 if (ifa_maybe == NULL || 2005 rt_refines(ifa->ifa_netmask, 2006 ifa_maybe->ifa_netmask)) 2007 ifa_maybe = ifa; 2008 } 2009 } 2010 return ifa_maybe; 2011 } 2012 2013 struct ifaddr * 2014 ifa_ifwithnet_psref(const struct sockaddr *addr, struct psref *psref) 2015 { 2016 struct ifaddr *ifa; 2017 int s; 2018 2019 s = pserialize_read_enter(); 2020 ifa = ifa_ifwithnet(addr); 2021 if (ifa != NULL) 2022 ifa_acquire(ifa, psref); 2023 pserialize_read_exit(s); 2024 2025 return ifa; 2026 } 2027 2028 /* 2029 * Find the interface of the addresss. 2030 */ 2031 struct ifaddr * 2032 ifa_ifwithladdr(const struct sockaddr *addr) 2033 { 2034 struct ifaddr *ia; 2035 2036 if ((ia = ifa_ifwithaddr(addr)) || (ia = ifa_ifwithdstaddr(addr)) || 2037 (ia = ifa_ifwithnet(addr))) 2038 return ia; 2039 return NULL; 2040 } 2041 2042 struct ifaddr * 2043 ifa_ifwithladdr_psref(const struct sockaddr *addr, struct psref *psref) 2044 { 2045 struct ifaddr *ifa; 2046 int s; 2047 2048 s = pserialize_read_enter(); 2049 ifa = ifa_ifwithladdr(addr); 2050 if (ifa != NULL) 2051 ifa_acquire(ifa, psref); 2052 pserialize_read_exit(s); 2053 2054 return ifa; 2055 } 2056 2057 /* 2058 * Find an interface using a specific address family 2059 */ 2060 struct ifaddr * 2061 ifa_ifwithaf(int af) 2062 { 2063 struct ifnet *ifp; 2064 struct ifaddr *ifa = NULL; 2065 int s; 2066 2067 s = pserialize_read_enter(); 2068 IFNET_READER_FOREACH(ifp) { 2069 if (if_is_deactivated(ifp)) 2070 continue; 2071 IFADDR_READER_FOREACH(ifa, ifp) { 2072 if (ifa->ifa_addr->sa_family == af) 2073 goto out; 2074 } 2075 } 2076 out: 2077 pserialize_read_exit(s); 2078 return ifa; 2079 } 2080 2081 /* 2082 * Find an interface address specific to an interface best matching 2083 * a given address. 2084 */ 2085 struct ifaddr * 2086 ifaof_ifpforaddr(const struct sockaddr *addr, struct ifnet *ifp) 2087 { 2088 struct ifaddr *ifa; 2089 const char *cp, *cp2, *cp3; 2090 const char *cplim; 2091 struct ifaddr *ifa_maybe = 0; 2092 u_int af = addr->sa_family; 2093 2094 if (if_is_deactivated(ifp)) 2095 return NULL; 2096 2097 if (af >= AF_MAX) 2098 return NULL; 2099 2100 IFADDR_READER_FOREACH(ifa, ifp) { 2101 if (ifa->ifa_addr->sa_family != af) 2102 continue; 2103 ifa_maybe = ifa; 2104 if (ifa->ifa_netmask == NULL) { 2105 if (equal(addr, ifa->ifa_addr) || 2106 (ifa->ifa_dstaddr && 2107 equal(addr, ifa->ifa_dstaddr))) 2108 return ifa; 2109 continue; 2110 } 2111 cp = addr->sa_data; 2112 cp2 = ifa->ifa_addr->sa_data; 2113 cp3 = ifa->ifa_netmask->sa_data; 2114 cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask; 2115 for (; cp3 < cplim; cp3++) { 2116 if ((*cp++ ^ *cp2++) & *cp3) 2117 break; 2118 } 2119 if (cp3 == cplim) 2120 return ifa; 2121 } 2122 return ifa_maybe; 2123 } 2124 2125 struct ifaddr * 2126 ifaof_ifpforaddr_psref(const struct sockaddr *addr, struct ifnet *ifp, 2127 struct psref *psref) 2128 { 2129 struct ifaddr *ifa; 2130 int s; 2131 2132 s = pserialize_read_enter(); 2133 ifa = ifaof_ifpforaddr(addr, ifp); 2134 if (ifa != NULL) 2135 ifa_acquire(ifa, psref); 2136 pserialize_read_exit(s); 2137 2138 return ifa; 2139 } 2140 2141 /* 2142 * Default action when installing a route with a Link Level gateway. 2143 * Lookup an appropriate real ifa to point to. 2144 * This should be moved to /sys/net/link.c eventually. 2145 */ 2146 void 2147 link_rtrequest(int cmd, struct rtentry *rt, const struct rt_addrinfo *info) 2148 { 2149 struct ifaddr *ifa; 2150 const struct sockaddr *dst; 2151 struct ifnet *ifp; 2152 struct psref psref; 2153 2154 if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL || 2155 (ifp = ifa->ifa_ifp) == NULL || (dst = rt_getkey(rt)) == NULL) 2156 return; 2157 if ((ifa = ifaof_ifpforaddr_psref(dst, ifp, &psref)) != NULL) { 2158 rt_replace_ifa(rt, ifa); 2159 if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest) 2160 ifa->ifa_rtrequest(cmd, rt, info); 2161 ifa_release(ifa, &psref); 2162 } 2163 } 2164 2165 /* 2166 * bitmask macros to manage a densely packed link_state change queue. 2167 * Because we need to store LINK_STATE_UNKNOWN(0), LINK_STATE_DOWN(1) and 2168 * LINK_STATE_UP(2) we need 2 bits for each state change. 2169 * As a state change to store is 0, treat all bits set as an unset item. 2170 */ 2171 #define LQ_ITEM_BITS 2 2172 #define LQ_ITEM_MASK ((1 << LQ_ITEM_BITS) - 1) 2173 #define LQ_MASK(i) (LQ_ITEM_MASK << (i) * LQ_ITEM_BITS) 2174 #define LINK_STATE_UNSET LQ_ITEM_MASK 2175 #define LQ_ITEM(q, i) (((q) & LQ_MASK((i))) >> (i) * LQ_ITEM_BITS) 2176 #define LQ_STORE(q, i, v) \ 2177 do { \ 2178 (q) &= ~LQ_MASK((i)); \ 2179 (q) |= (v) << (i) * LQ_ITEM_BITS; \ 2180 } while (0 /* CONSTCOND */) 2181 #define LQ_MAX(q) ((sizeof((q)) * NBBY) / LQ_ITEM_BITS) 2182 #define LQ_POP(q, v) \ 2183 do { \ 2184 (v) = LQ_ITEM((q), 0); \ 2185 (q) >>= LQ_ITEM_BITS; \ 2186 (q) |= LINK_STATE_UNSET << (LQ_MAX((q)) - 1) * LQ_ITEM_BITS; \ 2187 } while (0 /* CONSTCOND */) 2188 #define LQ_PUSH(q, v) \ 2189 do { \ 2190 (q) >>= LQ_ITEM_BITS; \ 2191 (q) |= (v) << (LQ_MAX((q)) - 1) * LQ_ITEM_BITS; \ 2192 } while (0 /* CONSTCOND */) 2193 #define LQ_FIND_UNSET(q, i) \ 2194 for ((i) = 0; i < LQ_MAX((q)); (i)++) { \ 2195 if (LQ_ITEM((q), (i)) == LINK_STATE_UNSET) \ 2196 break; \ 2197 } 2198 2199 /* 2200 * XXX reusing (ifp)->if_snd->ifq_lock rather than having another spin mutex 2201 * for each ifnet. It doesn't matter because: 2202 * - if IFEF_MPSAFE is enabled, if_snd isn't used and lock contentions on 2203 * ifq_lock don't happen 2204 * - if IFEF_MPSAFE is disabled, there is no lock contention on ifq_lock 2205 * because if_snd, if_link_state_change and if_link_state_change_softint 2206 * are all called with KERNEL_LOCK 2207 */ 2208 #define IF_LINK_STATE_CHANGE_LOCK(ifp) \ 2209 mutex_enter((ifp)->if_snd.ifq_lock) 2210 #define IF_LINK_STATE_CHANGE_UNLOCK(ifp) \ 2211 mutex_exit((ifp)->if_snd.ifq_lock) 2212 2213 /* 2214 * Handle a change in the interface link state and 2215 * queue notifications. 2216 */ 2217 void 2218 if_link_state_change(struct ifnet *ifp, int link_state) 2219 { 2220 int idx; 2221 2222 KASSERTMSG(if_is_link_state_changeable(ifp), 2223 "%s: IFEF_NO_LINK_STATE_CHANGE must not be set, but if_extflags=0x%x", 2224 ifp->if_xname, ifp->if_extflags); 2225 2226 /* Ensure change is to a valid state */ 2227 switch (link_state) { 2228 case LINK_STATE_UNKNOWN: /* FALLTHROUGH */ 2229 case LINK_STATE_DOWN: /* FALLTHROUGH */ 2230 case LINK_STATE_UP: 2231 break; 2232 default: 2233 #ifdef DEBUG 2234 printf("%s: invalid link state %d\n", 2235 ifp->if_xname, link_state); 2236 #endif 2237 return; 2238 } 2239 2240 IF_LINK_STATE_CHANGE_LOCK(ifp); 2241 2242 /* Find the last unset event in the queue. */ 2243 LQ_FIND_UNSET(ifp->if_link_queue, idx); 2244 2245 /* 2246 * Ensure link_state doesn't match the last event in the queue. 2247 * ifp->if_link_state is not checked and set here because 2248 * that would present an inconsistent picture to the system. 2249 */ 2250 if (idx != 0 && 2251 LQ_ITEM(ifp->if_link_queue, idx - 1) == (uint8_t)link_state) 2252 goto out; 2253 2254 /* Handle queue overflow. */ 2255 if (idx == LQ_MAX(ifp->if_link_queue)) { 2256 uint8_t lost; 2257 2258 /* 2259 * The DOWN state must be protected from being pushed off 2260 * the queue to ensure that userland will always be 2261 * in a sane state. 2262 * Because DOWN is protected, there is no need to protect 2263 * UNKNOWN. 2264 * It should be invalid to change from any other state to 2265 * UNKNOWN anyway ... 2266 */ 2267 lost = LQ_ITEM(ifp->if_link_queue, 0); 2268 LQ_PUSH(ifp->if_link_queue, (uint8_t)link_state); 2269 if (lost == LINK_STATE_DOWN) { 2270 lost = LQ_ITEM(ifp->if_link_queue, 0); 2271 LQ_STORE(ifp->if_link_queue, 0, LINK_STATE_DOWN); 2272 } 2273 printf("%s: lost link state change %s\n", 2274 ifp->if_xname, 2275 lost == LINK_STATE_UP ? "UP" : 2276 lost == LINK_STATE_DOWN ? "DOWN" : 2277 "UNKNOWN"); 2278 } else 2279 LQ_STORE(ifp->if_link_queue, idx, (uint8_t)link_state); 2280 2281 softint_schedule(ifp->if_link_si); 2282 2283 out: 2284 IF_LINK_STATE_CHANGE_UNLOCK(ifp); 2285 } 2286 2287 /* 2288 * Handle interface link state change notifications. 2289 */ 2290 void 2291 if_link_state_change_softint(struct ifnet *ifp, int link_state) 2292 { 2293 struct domain *dp; 2294 int s = splnet(); 2295 bool notify; 2296 2297 KASSERT(!cpu_intr_p()); 2298 2299 IF_LINK_STATE_CHANGE_LOCK(ifp); 2300 2301 /* Ensure the change is still valid. */ 2302 if (ifp->if_link_state == link_state) { 2303 IF_LINK_STATE_CHANGE_UNLOCK(ifp); 2304 return; 2305 } 2306 2307 #ifdef DEBUG 2308 log(LOG_DEBUG, "%s: link state %s (was %s)\n", ifp->if_xname, 2309 link_state == LINK_STATE_UP ? "UP" : 2310 link_state == LINK_STATE_DOWN ? "DOWN" : 2311 "UNKNOWN", 2312 ifp->if_link_state == LINK_STATE_UP ? "UP" : 2313 ifp->if_link_state == LINK_STATE_DOWN ? "DOWN" : 2314 "UNKNOWN"); 2315 #endif 2316 2317 /* 2318 * When going from UNKNOWN to UP, we need to mark existing 2319 * addresses as tentative and restart DAD as we may have 2320 * erroneously not found a duplicate. 2321 * 2322 * This needs to happen before rt_ifmsg to avoid a race where 2323 * listeners would have an address and expect it to work right 2324 * away. 2325 */ 2326 notify = (link_state == LINK_STATE_UP && 2327 ifp->if_link_state == LINK_STATE_UNKNOWN); 2328 ifp->if_link_state = link_state; 2329 /* The following routines may sleep so release the spin mutex */ 2330 IF_LINK_STATE_CHANGE_UNLOCK(ifp); 2331 2332 KERNEL_LOCK_UNLESS_NET_MPSAFE(); 2333 if (notify) { 2334 DOMAIN_FOREACH(dp) { 2335 if (dp->dom_if_link_state_change != NULL) 2336 dp->dom_if_link_state_change(ifp, 2337 LINK_STATE_DOWN); 2338 } 2339 } 2340 2341 /* Notify that the link state has changed. */ 2342 rt_ifmsg(ifp); 2343 2344 #if NCARP > 0 2345 if (ifp->if_carp) 2346 carp_carpdev_state(ifp); 2347 #endif 2348 2349 DOMAIN_FOREACH(dp) { 2350 if (dp->dom_if_link_state_change != NULL) 2351 dp->dom_if_link_state_change(ifp, link_state); 2352 } 2353 KERNEL_UNLOCK_UNLESS_NET_MPSAFE(); 2354 splx(s); 2355 } 2356 2357 /* 2358 * Process the interface link state change queue. 2359 */ 2360 static void 2361 if_link_state_change_si(void *arg) 2362 { 2363 struct ifnet *ifp = arg; 2364 int s; 2365 uint8_t state; 2366 bool schedule; 2367 2368 SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE(); 2369 s = splnet(); 2370 2371 /* Pop a link state change from the queue and process it. */ 2372 IF_LINK_STATE_CHANGE_LOCK(ifp); 2373 LQ_POP(ifp->if_link_queue, state); 2374 IF_LINK_STATE_CHANGE_UNLOCK(ifp); 2375 2376 if_link_state_change_softint(ifp, state); 2377 2378 /* If there is a link state change to come, schedule it. */ 2379 IF_LINK_STATE_CHANGE_LOCK(ifp); 2380 schedule = (LQ_ITEM(ifp->if_link_queue, 0) != LINK_STATE_UNSET); 2381 IF_LINK_STATE_CHANGE_UNLOCK(ifp); 2382 if (schedule) 2383 softint_schedule(ifp->if_link_si); 2384 2385 splx(s); 2386 SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE(); 2387 } 2388 2389 /* 2390 * Default action when installing a local route on a point-to-point 2391 * interface. 2392 */ 2393 void 2394 p2p_rtrequest(int req, struct rtentry *rt, 2395 __unused const struct rt_addrinfo *info) 2396 { 2397 struct ifnet *ifp = rt->rt_ifp; 2398 struct ifaddr *ifa, *lo0ifa; 2399 int s = pserialize_read_enter(); 2400 2401 switch (req) { 2402 case RTM_ADD: 2403 if ((rt->rt_flags & RTF_LOCAL) == 0) 2404 break; 2405 2406 rt->rt_ifp = lo0ifp; 2407 2408 IFADDR_READER_FOREACH(ifa, ifp) { 2409 if (equal(rt_getkey(rt), ifa->ifa_addr)) 2410 break; 2411 } 2412 if (ifa == NULL) 2413 break; 2414 2415 /* 2416 * Ensure lo0 has an address of the same family. 2417 */ 2418 IFADDR_READER_FOREACH(lo0ifa, lo0ifp) { 2419 if (lo0ifa->ifa_addr->sa_family == 2420 ifa->ifa_addr->sa_family) 2421 break; 2422 } 2423 if (lo0ifa == NULL) 2424 break; 2425 2426 /* 2427 * Make sure to set rt->rt_ifa to the interface 2428 * address we are using, otherwise we will have trouble 2429 * with source address selection. 2430 */ 2431 if (ifa != rt->rt_ifa) 2432 rt_replace_ifa(rt, ifa); 2433 break; 2434 case RTM_DELETE: 2435 default: 2436 break; 2437 } 2438 pserialize_read_exit(s); 2439 } 2440 2441 static void 2442 _if_down(struct ifnet *ifp) 2443 { 2444 struct ifaddr *ifa; 2445 struct domain *dp; 2446 int s, bound; 2447 struct psref psref; 2448 2449 ifp->if_flags &= ~IFF_UP; 2450 nanotime(&ifp->if_lastchange); 2451 2452 bound = curlwp_bind(); 2453 s = pserialize_read_enter(); 2454 IFADDR_READER_FOREACH(ifa, ifp) { 2455 ifa_acquire(ifa, &psref); 2456 pserialize_read_exit(s); 2457 2458 pfctlinput(PRC_IFDOWN, ifa->ifa_addr); 2459 2460 s = pserialize_read_enter(); 2461 ifa_release(ifa, &psref); 2462 } 2463 pserialize_read_exit(s); 2464 curlwp_bindx(bound); 2465 2466 IFQ_PURGE(&ifp->if_snd); 2467 #if NCARP > 0 2468 if (ifp->if_carp) 2469 carp_carpdev_state(ifp); 2470 #endif 2471 rt_ifmsg(ifp); 2472 DOMAIN_FOREACH(dp) { 2473 if (dp->dom_if_down) 2474 dp->dom_if_down(ifp); 2475 } 2476 } 2477 2478 static void 2479 if_down_deactivated(struct ifnet *ifp) 2480 { 2481 2482 KASSERT(if_is_deactivated(ifp)); 2483 _if_down(ifp); 2484 } 2485 2486 void 2487 if_down_locked(struct ifnet *ifp) 2488 { 2489 2490 KASSERT(IFNET_LOCKED(ifp)); 2491 _if_down(ifp); 2492 } 2493 2494 /* 2495 * Mark an interface down and notify protocols of 2496 * the transition. 2497 * NOTE: must be called at splsoftnet or equivalent. 2498 */ 2499 void 2500 if_down(struct ifnet *ifp) 2501 { 2502 2503 IFNET_LOCK(ifp); 2504 if_down_locked(ifp); 2505 IFNET_UNLOCK(ifp); 2506 } 2507 2508 /* 2509 * Must be called with holding if_ioctl_lock. 2510 */ 2511 static void 2512 if_up_locked(struct ifnet *ifp) 2513 { 2514 #ifdef notyet 2515 struct ifaddr *ifa; 2516 #endif 2517 struct domain *dp; 2518 2519 KASSERT(IFNET_LOCKED(ifp)); 2520 2521 KASSERT(!if_is_deactivated(ifp)); 2522 ifp->if_flags |= IFF_UP; 2523 nanotime(&ifp->if_lastchange); 2524 #ifdef notyet 2525 /* this has no effect on IP, and will kill all ISO connections XXX */ 2526 IFADDR_READER_FOREACH(ifa, ifp) 2527 pfctlinput(PRC_IFUP, ifa->ifa_addr); 2528 #endif 2529 #if NCARP > 0 2530 if (ifp->if_carp) 2531 carp_carpdev_state(ifp); 2532 #endif 2533 rt_ifmsg(ifp); 2534 DOMAIN_FOREACH(dp) { 2535 if (dp->dom_if_up) 2536 dp->dom_if_up(ifp); 2537 } 2538 } 2539 2540 /* 2541 * Handle interface slowtimo timer routine. Called 2542 * from softclock, we decrement timer (if set) and 2543 * call the appropriate interface routine on expiration. 2544 */ 2545 static void 2546 if_slowtimo(void *arg) 2547 { 2548 void (*slowtimo)(struct ifnet *); 2549 struct ifnet *ifp = arg; 2550 int s; 2551 2552 slowtimo = ifp->if_slowtimo; 2553 if (__predict_false(slowtimo == NULL)) 2554 return; 2555 2556 s = splnet(); 2557 if (ifp->if_timer != 0 && --ifp->if_timer == 0) 2558 (*slowtimo)(ifp); 2559 2560 splx(s); 2561 2562 if (__predict_true(ifp->if_slowtimo != NULL)) 2563 callout_schedule(ifp->if_slowtimo_ch, hz / IFNET_SLOWHZ); 2564 } 2565 2566 /* 2567 * Mark an interface up and notify protocols of 2568 * the transition. 2569 * NOTE: must be called at splsoftnet or equivalent. 2570 */ 2571 void 2572 if_up(struct ifnet *ifp) 2573 { 2574 2575 IFNET_LOCK(ifp); 2576 if_up_locked(ifp); 2577 IFNET_UNLOCK(ifp); 2578 } 2579 2580 /* 2581 * Set/clear promiscuous mode on interface ifp based on the truth value 2582 * of pswitch. The calls are reference counted so that only the first 2583 * "on" request actually has an effect, as does the final "off" request. 2584 * Results are undefined if the "off" and "on" requests are not matched. 2585 */ 2586 int 2587 ifpromisc_locked(struct ifnet *ifp, int pswitch) 2588 { 2589 int pcount, ret = 0; 2590 short nflags; 2591 2592 KASSERT(IFNET_LOCKED(ifp)); 2593 2594 pcount = ifp->if_pcount; 2595 if (pswitch) { 2596 /* 2597 * Allow the device to be "placed" into promiscuous 2598 * mode even if it is not configured up. It will 2599 * consult IFF_PROMISC when it is brought up. 2600 */ 2601 if (ifp->if_pcount++ != 0) 2602 goto out; 2603 nflags = ifp->if_flags | IFF_PROMISC; 2604 } else { 2605 if (--ifp->if_pcount > 0) 2606 goto out; 2607 nflags = ifp->if_flags & ~IFF_PROMISC; 2608 } 2609 ret = if_flags_set(ifp, nflags); 2610 /* Restore interface state if not successful. */ 2611 if (ret != 0) { 2612 ifp->if_pcount = pcount; 2613 } 2614 out: 2615 return ret; 2616 } 2617 2618 int 2619 ifpromisc(struct ifnet *ifp, int pswitch) 2620 { 2621 int e; 2622 2623 IFNET_LOCK(ifp); 2624 e = ifpromisc_locked(ifp, pswitch); 2625 IFNET_UNLOCK(ifp); 2626 2627 return e; 2628 } 2629 2630 /* 2631 * Map interface name to 2632 * interface structure pointer. 2633 */ 2634 struct ifnet * 2635 ifunit(const char *name) 2636 { 2637 struct ifnet *ifp; 2638 const char *cp = name; 2639 u_int unit = 0; 2640 u_int i; 2641 int s; 2642 2643 /* 2644 * If the entire name is a number, treat it as an ifindex. 2645 */ 2646 for (i = 0; i < IFNAMSIZ && *cp >= '0' && *cp <= '9'; i++, cp++) { 2647 unit = unit * 10 + (*cp - '0'); 2648 } 2649 2650 /* 2651 * If the number took all of the name, then it's a valid ifindex. 2652 */ 2653 if (i == IFNAMSIZ || (cp != name && *cp == '\0')) 2654 return if_byindex(unit); 2655 2656 ifp = NULL; 2657 s = pserialize_read_enter(); 2658 IFNET_READER_FOREACH(ifp) { 2659 if (if_is_deactivated(ifp)) 2660 continue; 2661 if (strcmp(ifp->if_xname, name) == 0) 2662 goto out; 2663 } 2664 out: 2665 pserialize_read_exit(s); 2666 return ifp; 2667 } 2668 2669 /* 2670 * Get a reference of an ifnet object by an interface name. 2671 * The returned reference is protected by psref(9). The caller 2672 * must release a returned reference by if_put after use. 2673 */ 2674 struct ifnet * 2675 if_get(const char *name, struct psref *psref) 2676 { 2677 struct ifnet *ifp; 2678 const char *cp = name; 2679 u_int unit = 0; 2680 u_int i; 2681 int s; 2682 2683 /* 2684 * If the entire name is a number, treat it as an ifindex. 2685 */ 2686 for (i = 0; i < IFNAMSIZ && *cp >= '0' && *cp <= '9'; i++, cp++) { 2687 unit = unit * 10 + (*cp - '0'); 2688 } 2689 2690 /* 2691 * If the number took all of the name, then it's a valid ifindex. 2692 */ 2693 if (i == IFNAMSIZ || (cp != name && *cp == '\0')) 2694 return if_get_byindex(unit, psref); 2695 2696 ifp = NULL; 2697 s = pserialize_read_enter(); 2698 IFNET_READER_FOREACH(ifp) { 2699 if (if_is_deactivated(ifp)) 2700 continue; 2701 if (strcmp(ifp->if_xname, name) == 0) { 2702 psref_acquire(psref, &ifp->if_psref, 2703 ifnet_psref_class); 2704 goto out; 2705 } 2706 } 2707 out: 2708 pserialize_read_exit(s); 2709 return ifp; 2710 } 2711 2712 /* 2713 * Release a reference of an ifnet object given by if_get, if_get_byindex 2714 * or if_get_bylla. 2715 */ 2716 void 2717 if_put(const struct ifnet *ifp, struct psref *psref) 2718 { 2719 2720 if (ifp == NULL) 2721 return; 2722 2723 psref_release(psref, &ifp->if_psref, ifnet_psref_class); 2724 } 2725 2726 /* 2727 * Return ifp having idx. Return NULL if not found. Normally if_byindex 2728 * should be used. 2729 */ 2730 ifnet_t * 2731 _if_byindex(u_int idx) 2732 { 2733 2734 return (__predict_true(idx < if_indexlim)) ? ifindex2ifnet[idx] : NULL; 2735 } 2736 2737 /* 2738 * Return ifp having idx. Return NULL if not found or the found ifp is 2739 * already deactivated. 2740 */ 2741 ifnet_t * 2742 if_byindex(u_int idx) 2743 { 2744 ifnet_t *ifp; 2745 2746 ifp = _if_byindex(idx); 2747 if (ifp != NULL && if_is_deactivated(ifp)) 2748 ifp = NULL; 2749 return ifp; 2750 } 2751 2752 /* 2753 * Get a reference of an ifnet object by an interface index. 2754 * The returned reference is protected by psref(9). The caller 2755 * must release a returned reference by if_put after use. 2756 */ 2757 ifnet_t * 2758 if_get_byindex(u_int idx, struct psref *psref) 2759 { 2760 ifnet_t *ifp; 2761 int s; 2762 2763 s = pserialize_read_enter(); 2764 ifp = if_byindex(idx); 2765 if (__predict_true(ifp != NULL)) 2766 psref_acquire(psref, &ifp->if_psref, ifnet_psref_class); 2767 pserialize_read_exit(s); 2768 2769 return ifp; 2770 } 2771 2772 ifnet_t * 2773 if_get_bylla(const void *lla, unsigned char lla_len, struct psref *psref) 2774 { 2775 ifnet_t *ifp; 2776 int s; 2777 2778 s = pserialize_read_enter(); 2779 IFNET_READER_FOREACH(ifp) { 2780 if (if_is_deactivated(ifp)) 2781 continue; 2782 if (ifp->if_addrlen != lla_len) 2783 continue; 2784 if (memcmp(lla, CLLADDR(ifp->if_sadl), lla_len) == 0) { 2785 psref_acquire(psref, &ifp->if_psref, 2786 ifnet_psref_class); 2787 break; 2788 } 2789 } 2790 pserialize_read_exit(s); 2791 2792 return ifp; 2793 } 2794 2795 /* 2796 * Note that it's safe only if the passed ifp is guaranteed to not be freed, 2797 * for example using pserialize or the ifp is already held or some other 2798 * object is held which guarantes the ifp to not be freed indirectly. 2799 */ 2800 void 2801 if_acquire(struct ifnet *ifp, struct psref *psref) 2802 { 2803 2804 KASSERT(ifp->if_index != 0); 2805 psref_acquire(psref, &ifp->if_psref, ifnet_psref_class); 2806 } 2807 2808 bool 2809 if_held(struct ifnet *ifp) 2810 { 2811 2812 return psref_held(&ifp->if_psref, ifnet_psref_class); 2813 } 2814 2815 /* 2816 * Some tunnel interfaces can nest, e.g. IPv4 over IPv4 gif(4) tunnel over IPv4. 2817 * Check the tunnel nesting count. 2818 * Return > 0, if tunnel nesting count is more than limit. 2819 * Return 0, if tunnel nesting count is equal or less than limit. 2820 */ 2821 int 2822 if_tunnel_check_nesting(struct ifnet *ifp, struct mbuf *m, int limit) 2823 { 2824 struct m_tag *mtag; 2825 int *count; 2826 2827 mtag = m_tag_find(m, PACKET_TAG_TUNNEL_INFO, NULL); 2828 if (mtag != NULL) { 2829 count = (int *)(mtag + 1); 2830 if (++(*count) > limit) { 2831 log(LOG_NOTICE, 2832 "%s: recursively called too many times(%d)\n", 2833 ifp->if_xname, *count); 2834 return EIO; 2835 } 2836 } else { 2837 mtag = m_tag_get(PACKET_TAG_TUNNEL_INFO, sizeof(*count), 2838 M_NOWAIT); 2839 if (mtag != NULL) { 2840 m_tag_prepend(m, mtag); 2841 count = (int *)(mtag + 1); 2842 *count = 0; 2843 } else { 2844 log(LOG_DEBUG, 2845 "%s: m_tag_get() failed, recursion calls are not prevented.\n", 2846 ifp->if_xname); 2847 } 2848 } 2849 2850 return 0; 2851 } 2852 2853 /* common */ 2854 int 2855 ifioctl_common(struct ifnet *ifp, u_long cmd, void *data) 2856 { 2857 int s; 2858 struct ifreq *ifr; 2859 struct ifcapreq *ifcr; 2860 struct ifdatareq *ifdr; 2861 2862 switch (cmd) { 2863 case SIOCSIFCAP: 2864 ifcr = data; 2865 if ((ifcr->ifcr_capenable & ~ifp->if_capabilities) != 0) 2866 return EINVAL; 2867 2868 if (ifcr->ifcr_capenable == ifp->if_capenable) 2869 return 0; 2870 2871 ifp->if_capenable = ifcr->ifcr_capenable; 2872 2873 /* Pre-compute the checksum flags mask. */ 2874 ifp->if_csum_flags_tx = 0; 2875 ifp->if_csum_flags_rx = 0; 2876 if (ifp->if_capenable & IFCAP_CSUM_IPv4_Tx) { 2877 ifp->if_csum_flags_tx |= M_CSUM_IPv4; 2878 } 2879 if (ifp->if_capenable & IFCAP_CSUM_IPv4_Rx) { 2880 ifp->if_csum_flags_rx |= M_CSUM_IPv4; 2881 } 2882 2883 if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Tx) { 2884 ifp->if_csum_flags_tx |= M_CSUM_TCPv4; 2885 } 2886 if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Rx) { 2887 ifp->if_csum_flags_rx |= M_CSUM_TCPv4; 2888 } 2889 2890 if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Tx) { 2891 ifp->if_csum_flags_tx |= M_CSUM_UDPv4; 2892 } 2893 if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Rx) { 2894 ifp->if_csum_flags_rx |= M_CSUM_UDPv4; 2895 } 2896 2897 if (ifp->if_capenable & IFCAP_CSUM_TCPv6_Tx) { 2898 ifp->if_csum_flags_tx |= M_CSUM_TCPv6; 2899 } 2900 if (ifp->if_capenable & IFCAP_CSUM_TCPv6_Rx) { 2901 ifp->if_csum_flags_rx |= M_CSUM_TCPv6; 2902 } 2903 2904 if (ifp->if_capenable & IFCAP_CSUM_UDPv6_Tx) { 2905 ifp->if_csum_flags_tx |= M_CSUM_UDPv6; 2906 } 2907 if (ifp->if_capenable & IFCAP_CSUM_UDPv6_Rx) { 2908 ifp->if_csum_flags_rx |= M_CSUM_UDPv6; 2909 } 2910 if (ifp->if_flags & IFF_UP) 2911 return ENETRESET; 2912 return 0; 2913 case SIOCSIFFLAGS: 2914 ifr = data; 2915 /* 2916 * If if_is_mpsafe(ifp), KERNEL_LOCK isn't held here, but if_up 2917 * and if_down aren't MP-safe yet, so we must hold the lock. 2918 */ 2919 KERNEL_LOCK_IF_IFP_MPSAFE(ifp); 2920 if (ifp->if_flags & IFF_UP && (ifr->ifr_flags & IFF_UP) == 0) { 2921 s = splsoftnet(); 2922 if_down_locked(ifp); 2923 splx(s); 2924 } 2925 if (ifr->ifr_flags & IFF_UP && (ifp->if_flags & IFF_UP) == 0) { 2926 s = splsoftnet(); 2927 if_up_locked(ifp); 2928 splx(s); 2929 } 2930 KERNEL_UNLOCK_IF_IFP_MPSAFE(ifp); 2931 ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) | 2932 (ifr->ifr_flags &~ IFF_CANTCHANGE); 2933 break; 2934 case SIOCGIFFLAGS: 2935 ifr = data; 2936 ifr->ifr_flags = ifp->if_flags; 2937 break; 2938 2939 case SIOCGIFMETRIC: 2940 ifr = data; 2941 ifr->ifr_metric = ifp->if_metric; 2942 break; 2943 2944 case SIOCGIFMTU: 2945 ifr = data; 2946 ifr->ifr_mtu = ifp->if_mtu; 2947 break; 2948 2949 case SIOCGIFDLT: 2950 ifr = data; 2951 ifr->ifr_dlt = ifp->if_dlt; 2952 break; 2953 2954 case SIOCGIFCAP: 2955 ifcr = data; 2956 ifcr->ifcr_capabilities = ifp->if_capabilities; 2957 ifcr->ifcr_capenable = ifp->if_capenable; 2958 break; 2959 2960 case SIOCSIFMETRIC: 2961 ifr = data; 2962 ifp->if_metric = ifr->ifr_metric; 2963 break; 2964 2965 case SIOCGIFDATA: 2966 ifdr = data; 2967 ifdr->ifdr_data = ifp->if_data; 2968 break; 2969 2970 case SIOCGIFINDEX: 2971 ifr = data; 2972 ifr->ifr_index = ifp->if_index; 2973 break; 2974 2975 case SIOCZIFDATA: 2976 ifdr = data; 2977 ifdr->ifdr_data = ifp->if_data; 2978 /* 2979 * Assumes that the volatile counters that can be 2980 * zero'ed are at the end of if_data. 2981 */ 2982 memset(&ifp->if_data.ifi_ipackets, 0, sizeof(ifp->if_data) - 2983 offsetof(struct if_data, ifi_ipackets)); 2984 /* 2985 * The memset() clears to the bottm of if_data. In the area, 2986 * if_lastchange is included. Please be careful if new entry 2987 * will be added into if_data or rewite this. 2988 * 2989 * And also, update if_lastchnage. 2990 */ 2991 getnanotime(&ifp->if_lastchange); 2992 break; 2993 case SIOCSIFMTU: 2994 ifr = data; 2995 if (ifp->if_mtu == ifr->ifr_mtu) 2996 break; 2997 ifp->if_mtu = ifr->ifr_mtu; 2998 /* 2999 * If the link MTU changed, do network layer specific procedure. 3000 */ 3001 #ifdef INET6 3002 KERNEL_LOCK_UNLESS_NET_MPSAFE(); 3003 if (in6_present) 3004 nd6_setmtu(ifp); 3005 KERNEL_UNLOCK_UNLESS_NET_MPSAFE(); 3006 #endif 3007 return ENETRESET; 3008 default: 3009 return ENOTTY; 3010 } 3011 return 0; 3012 } 3013 3014 int 3015 ifaddrpref_ioctl(struct socket *so, u_long cmd, void *data, struct ifnet *ifp) 3016 { 3017 struct if_addrprefreq *ifap = (struct if_addrprefreq *)data; 3018 struct ifaddr *ifa; 3019 const struct sockaddr *any, *sa; 3020 union { 3021 struct sockaddr sa; 3022 struct sockaddr_storage ss; 3023 } u, v; 3024 int s, error = 0; 3025 3026 switch (cmd) { 3027 case SIOCSIFADDRPREF: 3028 if (kauth_authorize_network(curlwp->l_cred, KAUTH_NETWORK_INTERFACE, 3029 KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd, 3030 NULL) != 0) 3031 return EPERM; 3032 case SIOCGIFADDRPREF: 3033 break; 3034 default: 3035 return EOPNOTSUPP; 3036 } 3037 3038 /* sanity checks */ 3039 if (data == NULL || ifp == NULL) { 3040 panic("invalid argument to %s", __func__); 3041 /*NOTREACHED*/ 3042 } 3043 3044 /* address must be specified on ADD and DELETE */ 3045 sa = sstocsa(&ifap->ifap_addr); 3046 if (sa->sa_family != sofamily(so)) 3047 return EINVAL; 3048 if ((any = sockaddr_any(sa)) == NULL || sa->sa_len != any->sa_len) 3049 return EINVAL; 3050 3051 sockaddr_externalize(&v.sa, sizeof(v.ss), sa); 3052 3053 s = pserialize_read_enter(); 3054 IFADDR_READER_FOREACH(ifa, ifp) { 3055 if (ifa->ifa_addr->sa_family != sa->sa_family) 3056 continue; 3057 sockaddr_externalize(&u.sa, sizeof(u.ss), ifa->ifa_addr); 3058 if (sockaddr_cmp(&u.sa, &v.sa) == 0) 3059 break; 3060 } 3061 if (ifa == NULL) { 3062 error = EADDRNOTAVAIL; 3063 goto out; 3064 } 3065 3066 switch (cmd) { 3067 case SIOCSIFADDRPREF: 3068 ifa->ifa_preference = ifap->ifap_preference; 3069 goto out; 3070 case SIOCGIFADDRPREF: 3071 /* fill in the if_laddrreq structure */ 3072 (void)sockaddr_copy(sstosa(&ifap->ifap_addr), 3073 sizeof(ifap->ifap_addr), ifa->ifa_addr); 3074 ifap->ifap_preference = ifa->ifa_preference; 3075 goto out; 3076 default: 3077 error = EOPNOTSUPP; 3078 } 3079 out: 3080 pserialize_read_exit(s); 3081 return error; 3082 } 3083 3084 /* 3085 * Interface ioctls. 3086 */ 3087 static int 3088 doifioctl(struct socket *so, u_long cmd, void *data, struct lwp *l) 3089 { 3090 struct ifnet *ifp; 3091 struct ifreq *ifr; 3092 int error = 0; 3093 #if defined(COMPAT_OSOCK) || defined(COMPAT_OIFREQ) 3094 u_long ocmd = cmd; 3095 #endif 3096 short oif_flags; 3097 #ifdef COMPAT_OIFREQ 3098 struct ifreq ifrb; 3099 struct oifreq *oifr = NULL; 3100 #endif 3101 int r; 3102 struct psref psref; 3103 int bound; 3104 3105 switch (cmd) { 3106 case SIOCGIFCONF: 3107 return ifconf(cmd, data); 3108 case SIOCINITIFADDR: 3109 return EPERM; 3110 default: 3111 error = (*vec_compat_ifconf)(l, cmd, data); 3112 if (error != ENOSYS) 3113 return error; 3114 error = (*vec_compat_ifdatareq)(l, cmd, data); 3115 if (error != ENOSYS) 3116 return error; 3117 break; 3118 } 3119 3120 ifr = data; 3121 #ifdef COMPAT_OIFREQ 3122 if (vec_compat_cvtcmd) { 3123 cmd = (*vec_compat_cvtcmd)(cmd); 3124 if (cmd != ocmd) { 3125 oifr = data; 3126 data = ifr = &ifrb; 3127 ifreqo2n(oifr, ifr); 3128 } 3129 } 3130 #endif 3131 3132 switch (cmd) { 3133 case SIOCIFCREATE: 3134 case SIOCIFDESTROY: 3135 bound = curlwp_bind(); 3136 if (l != NULL) { 3137 ifp = if_get(ifr->ifr_name, &psref); 3138 error = kauth_authorize_network(l->l_cred, 3139 KAUTH_NETWORK_INTERFACE, 3140 KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, 3141 (void *)cmd, NULL); 3142 if (ifp != NULL) 3143 if_put(ifp, &psref); 3144 if (error != 0) { 3145 curlwp_bindx(bound); 3146 return error; 3147 } 3148 } 3149 KERNEL_LOCK_UNLESS_NET_MPSAFE(); 3150 mutex_enter(&if_clone_mtx); 3151 r = (cmd == SIOCIFCREATE) ? 3152 if_clone_create(ifr->ifr_name) : 3153 if_clone_destroy(ifr->ifr_name); 3154 mutex_exit(&if_clone_mtx); 3155 KERNEL_UNLOCK_UNLESS_NET_MPSAFE(); 3156 curlwp_bindx(bound); 3157 return r; 3158 3159 case SIOCIFGCLONERS: 3160 { 3161 struct if_clonereq *req = (struct if_clonereq *)data; 3162 return if_clone_list(req->ifcr_count, req->ifcr_buffer, 3163 &req->ifcr_total); 3164 } 3165 } 3166 3167 bound = curlwp_bind(); 3168 ifp = if_get(ifr->ifr_name, &psref); 3169 if (ifp == NULL) { 3170 curlwp_bindx(bound); 3171 return ENXIO; 3172 } 3173 3174 switch (cmd) { 3175 case SIOCALIFADDR: 3176 case SIOCDLIFADDR: 3177 case SIOCSIFADDRPREF: 3178 case SIOCSIFFLAGS: 3179 case SIOCSIFCAP: 3180 case SIOCSIFMETRIC: 3181 case SIOCZIFDATA: 3182 case SIOCSIFMTU: 3183 case SIOCSIFPHYADDR: 3184 case SIOCDIFPHYADDR: 3185 #ifdef INET6 3186 case SIOCSIFPHYADDR_IN6: 3187 #endif 3188 case SIOCSLIFPHYADDR: 3189 case SIOCADDMULTI: 3190 case SIOCDELMULTI: 3191 case SIOCSIFMEDIA: 3192 case SIOCSDRVSPEC: 3193 case SIOCG80211: 3194 case SIOCS80211: 3195 case SIOCS80211NWID: 3196 case SIOCS80211NWKEY: 3197 case SIOCS80211POWER: 3198 case SIOCS80211BSSID: 3199 case SIOCS80211CHANNEL: 3200 case SIOCSLINKSTR: 3201 if (l != NULL) { 3202 error = kauth_authorize_network(l->l_cred, 3203 KAUTH_NETWORK_INTERFACE, 3204 KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, 3205 (void *)cmd, NULL); 3206 if (error != 0) 3207 goto out; 3208 } 3209 } 3210 3211 oif_flags = ifp->if_flags; 3212 3213 KERNEL_LOCK_UNLESS_IFP_MPSAFE(ifp); 3214 IFNET_LOCK(ifp); 3215 3216 error = (*ifp->if_ioctl)(ifp, cmd, data); 3217 if (error != ENOTTY) 3218 ; 3219 else if (so->so_proto == NULL) 3220 error = EOPNOTSUPP; 3221 else { 3222 KERNEL_LOCK_IF_IFP_MPSAFE(ifp); 3223 #ifdef COMPAT_OSOCK 3224 if (vec_compat_ifioctl != NULL) 3225 error = (*vec_compat_ifioctl)(so, ocmd, cmd, data, l); 3226 else 3227 #endif 3228 error = (*so->so_proto->pr_usrreqs->pr_ioctl)(so, 3229 cmd, data, ifp); 3230 KERNEL_UNLOCK_IF_IFP_MPSAFE(ifp); 3231 } 3232 3233 if (((oif_flags ^ ifp->if_flags) & IFF_UP) != 0) { 3234 if ((ifp->if_flags & IFF_UP) != 0) { 3235 int s = splsoftnet(); 3236 if_up_locked(ifp); 3237 splx(s); 3238 } 3239 } 3240 #ifdef COMPAT_OIFREQ 3241 if (cmd != ocmd) 3242 ifreqn2o(oifr, ifr); 3243 #endif 3244 3245 IFNET_UNLOCK(ifp); 3246 KERNEL_UNLOCK_UNLESS_IFP_MPSAFE(ifp); 3247 out: 3248 if_put(ifp, &psref); 3249 curlwp_bindx(bound); 3250 return error; 3251 } 3252 3253 /* 3254 * Return interface configuration 3255 * of system. List may be used 3256 * in later ioctl's (above) to get 3257 * other information. 3258 * 3259 * Each record is a struct ifreq. Before the addition of 3260 * sockaddr_storage, the API rule was that sockaddr flavors that did 3261 * not fit would extend beyond the struct ifreq, with the next struct 3262 * ifreq starting sa_len beyond the struct sockaddr. Because the 3263 * union in struct ifreq includes struct sockaddr_storage, every kind 3264 * of sockaddr must fit. Thus, there are no longer any overlength 3265 * records. 3266 * 3267 * Records are added to the user buffer if they fit, and ifc_len is 3268 * adjusted to the length that was written. Thus, the user is only 3269 * assured of getting the complete list if ifc_len on return is at 3270 * least sizeof(struct ifreq) less than it was on entry. 3271 * 3272 * If the user buffer pointer is NULL, this routine copies no data and 3273 * returns the amount of space that would be needed. 3274 * 3275 * Invariants: 3276 * ifrp points to the next part of the user's buffer to be used. If 3277 * ifrp != NULL, space holds the number of bytes remaining that we may 3278 * write at ifrp. Otherwise, space holds the number of bytes that 3279 * would have been written had there been adequate space. 3280 */ 3281 /*ARGSUSED*/ 3282 static int 3283 ifconf(u_long cmd, void *data) 3284 { 3285 struct ifconf *ifc = (struct ifconf *)data; 3286 struct ifnet *ifp; 3287 struct ifaddr *ifa; 3288 struct ifreq ifr, *ifrp = NULL; 3289 int space = 0, error = 0; 3290 const int sz = (int)sizeof(struct ifreq); 3291 const bool docopy = ifc->ifc_req != NULL; 3292 int s; 3293 int bound; 3294 struct psref psref; 3295 3296 if (docopy) { 3297 space = ifc->ifc_len; 3298 ifrp = ifc->ifc_req; 3299 } 3300 3301 bound = curlwp_bind(); 3302 s = pserialize_read_enter(); 3303 IFNET_READER_FOREACH(ifp) { 3304 psref_acquire(&psref, &ifp->if_psref, ifnet_psref_class); 3305 pserialize_read_exit(s); 3306 3307 (void)strncpy(ifr.ifr_name, ifp->if_xname, 3308 sizeof(ifr.ifr_name)); 3309 if (ifr.ifr_name[sizeof(ifr.ifr_name) - 1] != '\0') { 3310 error = ENAMETOOLONG; 3311 goto release_exit; 3312 } 3313 if (IFADDR_READER_EMPTY(ifp)) { 3314 /* Interface with no addresses - send zero sockaddr. */ 3315 memset(&ifr.ifr_addr, 0, sizeof(ifr.ifr_addr)); 3316 if (!docopy) { 3317 space += sz; 3318 goto next; 3319 } 3320 if (space >= sz) { 3321 error = copyout(&ifr, ifrp, sz); 3322 if (error != 0) 3323 goto release_exit; 3324 ifrp++; 3325 space -= sz; 3326 } 3327 } 3328 3329 s = pserialize_read_enter(); 3330 IFADDR_READER_FOREACH(ifa, ifp) { 3331 struct sockaddr *sa = ifa->ifa_addr; 3332 /* all sockaddrs must fit in sockaddr_storage */ 3333 KASSERT(sa->sa_len <= sizeof(ifr.ifr_ifru)); 3334 3335 if (!docopy) { 3336 space += sz; 3337 continue; 3338 } 3339 memcpy(&ifr.ifr_space, sa, sa->sa_len); 3340 pserialize_read_exit(s); 3341 3342 if (space >= sz) { 3343 error = copyout(&ifr, ifrp, sz); 3344 if (error != 0) 3345 goto release_exit; 3346 ifrp++; space -= sz; 3347 } 3348 s = pserialize_read_enter(); 3349 } 3350 pserialize_read_exit(s); 3351 3352 next: 3353 s = pserialize_read_enter(); 3354 psref_release(&psref, &ifp->if_psref, ifnet_psref_class); 3355 } 3356 pserialize_read_exit(s); 3357 curlwp_bindx(bound); 3358 3359 if (docopy) { 3360 KASSERT(0 <= space && space <= ifc->ifc_len); 3361 ifc->ifc_len -= space; 3362 } else { 3363 KASSERT(space >= 0); 3364 ifc->ifc_len = space; 3365 } 3366 return (0); 3367 3368 release_exit: 3369 psref_release(&psref, &ifp->if_psref, ifnet_psref_class); 3370 curlwp_bindx(bound); 3371 return error; 3372 } 3373 3374 int 3375 ifreq_setaddr(u_long cmd, struct ifreq *ifr, const struct sockaddr *sa) 3376 { 3377 uint8_t len = sizeof(ifr->ifr_ifru.ifru_space); 3378 #ifdef COMPAT_OIFREQ 3379 struct ifreq ifrb; 3380 struct oifreq *oifr = NULL; 3381 u_long ocmd = cmd; 3382 3383 if (vec_compat_cvtcmd) { 3384 cmd = (*vec_compat_cvtcmd)(cmd); 3385 if (cmd != ocmd) { 3386 oifr = (struct oifreq *)(void *)ifr; 3387 ifr = &ifrb; 3388 ifreqo2n(oifr, ifr); 3389 len = sizeof(oifr->ifr_addr); 3390 } 3391 } 3392 #endif 3393 if (len < sa->sa_len) 3394 return EFBIG; 3395 3396 memset(&ifr->ifr_addr, 0, len); 3397 sockaddr_copy(&ifr->ifr_addr, len, sa); 3398 3399 #ifdef COMPAT_OIFREQ 3400 if (cmd != ocmd) 3401 ifreqn2o(oifr, ifr); 3402 #endif 3403 return 0; 3404 } 3405 3406 /* 3407 * wrapper function for the drivers which doesn't have if_transmit(). 3408 */ 3409 static int 3410 if_transmit(struct ifnet *ifp, struct mbuf *m) 3411 { 3412 int s, error; 3413 size_t pktlen = m->m_pkthdr.len; 3414 bool mcast = (m->m_flags & M_MCAST) != 0; 3415 3416 s = splnet(); 3417 3418 IFQ_ENQUEUE(&ifp->if_snd, m, error); 3419 if (error != 0) { 3420 /* mbuf is already freed */ 3421 goto out; 3422 } 3423 3424 ifp->if_obytes += pktlen; 3425 if (mcast) 3426 ifp->if_omcasts++; 3427 3428 if ((ifp->if_flags & IFF_OACTIVE) == 0) 3429 if_start_lock(ifp); 3430 out: 3431 splx(s); 3432 3433 return error; 3434 } 3435 3436 int 3437 if_transmit_lock(struct ifnet *ifp, struct mbuf *m) 3438 { 3439 int error; 3440 3441 #ifdef ALTQ 3442 KERNEL_LOCK(1, NULL); 3443 if (ALTQ_IS_ENABLED(&ifp->if_snd)) { 3444 error = if_transmit(ifp, m); 3445 KERNEL_UNLOCK_ONE(NULL); 3446 } else { 3447 KERNEL_UNLOCK_ONE(NULL); 3448 error = (*ifp->if_transmit)(ifp, m); 3449 /* mbuf is alredy freed */ 3450 } 3451 #else /* !ALTQ */ 3452 error = (*ifp->if_transmit)(ifp, m); 3453 /* mbuf is alredy freed */ 3454 #endif /* !ALTQ */ 3455 3456 return error; 3457 } 3458 3459 /* 3460 * Queue message on interface, and start output if interface 3461 * not yet active. 3462 */ 3463 int 3464 ifq_enqueue(struct ifnet *ifp, struct mbuf *m) 3465 { 3466 3467 return if_transmit_lock(ifp, m); 3468 } 3469 3470 /* 3471 * Queue message on interface, possibly using a second fast queue 3472 */ 3473 int 3474 ifq_enqueue2(struct ifnet *ifp, struct ifqueue *ifq, struct mbuf *m) 3475 { 3476 int error = 0; 3477 3478 if (ifq != NULL 3479 #ifdef ALTQ 3480 && ALTQ_IS_ENABLED(&ifp->if_snd) == 0 3481 #endif 3482 ) { 3483 if (IF_QFULL(ifq)) { 3484 IF_DROP(&ifp->if_snd); 3485 m_freem(m); 3486 if (error == 0) 3487 error = ENOBUFS; 3488 } else 3489 IF_ENQUEUE(ifq, m); 3490 } else 3491 IFQ_ENQUEUE(&ifp->if_snd, m, error); 3492 if (error != 0) { 3493 ++ifp->if_oerrors; 3494 return error; 3495 } 3496 return 0; 3497 } 3498 3499 int 3500 if_addr_init(ifnet_t *ifp, struct ifaddr *ifa, const bool src) 3501 { 3502 int rc; 3503 3504 KASSERT(IFNET_LOCKED(ifp)); 3505 if (ifp->if_initaddr != NULL) 3506 rc = (*ifp->if_initaddr)(ifp, ifa, src); 3507 else if (src || 3508 (rc = (*ifp->if_ioctl)(ifp, SIOCSIFDSTADDR, ifa)) == ENOTTY) 3509 rc = (*ifp->if_ioctl)(ifp, SIOCINITIFADDR, ifa); 3510 3511 return rc; 3512 } 3513 3514 int 3515 if_do_dad(struct ifnet *ifp) 3516 { 3517 if ((ifp->if_flags & IFF_LOOPBACK) != 0) 3518 return 0; 3519 3520 switch (ifp->if_type) { 3521 case IFT_FAITH: 3522 /* 3523 * These interfaces do not have the IFF_LOOPBACK flag, 3524 * but loop packets back. We do not have to do DAD on such 3525 * interfaces. We should even omit it, because loop-backed 3526 * responses would confuse the DAD procedure. 3527 */ 3528 return 0; 3529 default: 3530 /* 3531 * Our DAD routine requires the interface up and running. 3532 * However, some interfaces can be up before the RUNNING 3533 * status. Additionaly, users may try to assign addresses 3534 * before the interface becomes up (or running). 3535 * We simply skip DAD in such a case as a work around. 3536 * XXX: we should rather mark "tentative" on such addresses, 3537 * and do DAD after the interface becomes ready. 3538 */ 3539 if ((ifp->if_flags & (IFF_UP|IFF_RUNNING)) != 3540 (IFF_UP|IFF_RUNNING)) 3541 return 0; 3542 3543 return 1; 3544 } 3545 } 3546 3547 int 3548 if_flags_set(ifnet_t *ifp, const short flags) 3549 { 3550 int rc; 3551 3552 KASSERT(IFNET_LOCKED(ifp)); 3553 3554 if (ifp->if_setflags != NULL) 3555 rc = (*ifp->if_setflags)(ifp, flags); 3556 else { 3557 short cantflags, chgdflags; 3558 struct ifreq ifr; 3559 3560 chgdflags = ifp->if_flags ^ flags; 3561 cantflags = chgdflags & IFF_CANTCHANGE; 3562 3563 if (cantflags != 0) 3564 ifp->if_flags ^= cantflags; 3565 3566 /* Traditionally, we do not call if_ioctl after 3567 * setting/clearing only IFF_PROMISC if the interface 3568 * isn't IFF_UP. Uphold that tradition. 3569 */ 3570 if (chgdflags == IFF_PROMISC && (ifp->if_flags & IFF_UP) == 0) 3571 return 0; 3572 3573 memset(&ifr, 0, sizeof(ifr)); 3574 3575 ifr.ifr_flags = flags & ~IFF_CANTCHANGE; 3576 rc = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, &ifr); 3577 3578 if (rc != 0 && cantflags != 0) 3579 ifp->if_flags ^= cantflags; 3580 } 3581 3582 return rc; 3583 } 3584 3585 int 3586 if_mcast_op(ifnet_t *ifp, const unsigned long cmd, const struct sockaddr *sa) 3587 { 3588 int rc; 3589 struct ifreq ifr; 3590 3591 /* CARP and MROUTING still don't deal with the lock yet */ 3592 #if (!defined(NCARP) || (NCARP == 0)) && !defined(MROUTING) 3593 KASSERT(IFNET_LOCKED(ifp)); 3594 #endif 3595 if (ifp->if_mcastop != NULL) 3596 rc = (*ifp->if_mcastop)(ifp, cmd, sa); 3597 else { 3598 ifreq_setaddr(cmd, &ifr, sa); 3599 rc = (*ifp->if_ioctl)(ifp, cmd, &ifr); 3600 } 3601 3602 return rc; 3603 } 3604 3605 static void 3606 sysctl_sndq_setup(struct sysctllog **clog, const char *ifname, 3607 struct ifaltq *ifq) 3608 { 3609 const struct sysctlnode *cnode, *rnode; 3610 3611 if (sysctl_createv(clog, 0, NULL, &rnode, 3612 CTLFLAG_PERMANENT, 3613 CTLTYPE_NODE, "interfaces", 3614 SYSCTL_DESCR("Per-interface controls"), 3615 NULL, 0, NULL, 0, 3616 CTL_NET, CTL_CREATE, CTL_EOL) != 0) 3617 goto bad; 3618 3619 if (sysctl_createv(clog, 0, &rnode, &rnode, 3620 CTLFLAG_PERMANENT, 3621 CTLTYPE_NODE, ifname, 3622 SYSCTL_DESCR("Interface controls"), 3623 NULL, 0, NULL, 0, 3624 CTL_CREATE, CTL_EOL) != 0) 3625 goto bad; 3626 3627 if (sysctl_createv(clog, 0, &rnode, &rnode, 3628 CTLFLAG_PERMANENT, 3629 CTLTYPE_NODE, "sndq", 3630 SYSCTL_DESCR("Interface output queue controls"), 3631 NULL, 0, NULL, 0, 3632 CTL_CREATE, CTL_EOL) != 0) 3633 goto bad; 3634 3635 if (sysctl_createv(clog, 0, &rnode, &cnode, 3636 CTLFLAG_PERMANENT, 3637 CTLTYPE_INT, "len", 3638 SYSCTL_DESCR("Current output queue length"), 3639 NULL, 0, &ifq->ifq_len, 0, 3640 CTL_CREATE, CTL_EOL) != 0) 3641 goto bad; 3642 3643 if (sysctl_createv(clog, 0, &rnode, &cnode, 3644 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 3645 CTLTYPE_INT, "maxlen", 3646 SYSCTL_DESCR("Maximum allowed output queue length"), 3647 NULL, 0, &ifq->ifq_maxlen, 0, 3648 CTL_CREATE, CTL_EOL) != 0) 3649 goto bad; 3650 3651 if (sysctl_createv(clog, 0, &rnode, &cnode, 3652 CTLFLAG_PERMANENT, 3653 CTLTYPE_INT, "drops", 3654 SYSCTL_DESCR("Packets dropped due to full output queue"), 3655 NULL, 0, &ifq->ifq_drops, 0, 3656 CTL_CREATE, CTL_EOL) != 0) 3657 goto bad; 3658 3659 return; 3660 bad: 3661 printf("%s: could not attach sysctl nodes\n", ifname); 3662 return; 3663 } 3664 3665 #if defined(INET) || defined(INET6) 3666 3667 #define SYSCTL_NET_PKTQ(q, cn, c) \ 3668 static int \ 3669 sysctl_net_##q##_##cn(SYSCTLFN_ARGS) \ 3670 { \ 3671 return sysctl_pktq_count(SYSCTLFN_CALL(rnode), q, c); \ 3672 } 3673 3674 #if defined(INET) 3675 static int 3676 sysctl_net_ip_pktq_maxlen(SYSCTLFN_ARGS) 3677 { 3678 return sysctl_pktq_maxlen(SYSCTLFN_CALL(rnode), ip_pktq); 3679 } 3680 SYSCTL_NET_PKTQ(ip_pktq, items, PKTQ_NITEMS) 3681 SYSCTL_NET_PKTQ(ip_pktq, drops, PKTQ_DROPS) 3682 #endif 3683 3684 #if defined(INET6) 3685 static int 3686 sysctl_net_ip6_pktq_maxlen(SYSCTLFN_ARGS) 3687 { 3688 return sysctl_pktq_maxlen(SYSCTLFN_CALL(rnode), ip6_pktq); 3689 } 3690 SYSCTL_NET_PKTQ(ip6_pktq, items, PKTQ_NITEMS) 3691 SYSCTL_NET_PKTQ(ip6_pktq, drops, PKTQ_DROPS) 3692 #endif 3693 3694 static void 3695 sysctl_net_pktq_setup(struct sysctllog **clog, int pf) 3696 { 3697 sysctlfn len_func = NULL, maxlen_func = NULL, drops_func = NULL; 3698 const char *pfname = NULL, *ipname = NULL; 3699 int ipn = 0, qid = 0; 3700 3701 switch (pf) { 3702 #if defined(INET) 3703 case PF_INET: 3704 len_func = sysctl_net_ip_pktq_items; 3705 maxlen_func = sysctl_net_ip_pktq_maxlen; 3706 drops_func = sysctl_net_ip_pktq_drops; 3707 pfname = "inet", ipn = IPPROTO_IP; 3708 ipname = "ip", qid = IPCTL_IFQ; 3709 break; 3710 #endif 3711 #if defined(INET6) 3712 case PF_INET6: 3713 len_func = sysctl_net_ip6_pktq_items; 3714 maxlen_func = sysctl_net_ip6_pktq_maxlen; 3715 drops_func = sysctl_net_ip6_pktq_drops; 3716 pfname = "inet6", ipn = IPPROTO_IPV6; 3717 ipname = "ip6", qid = IPV6CTL_IFQ; 3718 break; 3719 #endif 3720 default: 3721 KASSERT(false); 3722 } 3723 3724 sysctl_createv(clog, 0, NULL, NULL, 3725 CTLFLAG_PERMANENT, 3726 CTLTYPE_NODE, pfname, NULL, 3727 NULL, 0, NULL, 0, 3728 CTL_NET, pf, CTL_EOL); 3729 sysctl_createv(clog, 0, NULL, NULL, 3730 CTLFLAG_PERMANENT, 3731 CTLTYPE_NODE, ipname, NULL, 3732 NULL, 0, NULL, 0, 3733 CTL_NET, pf, ipn, CTL_EOL); 3734 sysctl_createv(clog, 0, NULL, NULL, 3735 CTLFLAG_PERMANENT, 3736 CTLTYPE_NODE, "ifq", 3737 SYSCTL_DESCR("Protocol input queue controls"), 3738 NULL, 0, NULL, 0, 3739 CTL_NET, pf, ipn, qid, CTL_EOL); 3740 3741 sysctl_createv(clog, 0, NULL, NULL, 3742 CTLFLAG_PERMANENT, 3743 CTLTYPE_INT, "len", 3744 SYSCTL_DESCR("Current input queue length"), 3745 len_func, 0, NULL, 0, 3746 CTL_NET, pf, ipn, qid, IFQCTL_LEN, CTL_EOL); 3747 sysctl_createv(clog, 0, NULL, NULL, 3748 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 3749 CTLTYPE_INT, "maxlen", 3750 SYSCTL_DESCR("Maximum allowed input queue length"), 3751 maxlen_func, 0, NULL, 0, 3752 CTL_NET, pf, ipn, qid, IFQCTL_MAXLEN, CTL_EOL); 3753 sysctl_createv(clog, 0, NULL, NULL, 3754 CTLFLAG_PERMANENT, 3755 CTLTYPE_INT, "drops", 3756 SYSCTL_DESCR("Packets dropped due to full input queue"), 3757 drops_func, 0, NULL, 0, 3758 CTL_NET, pf, ipn, qid, IFQCTL_DROPS, CTL_EOL); 3759 } 3760 #endif /* INET || INET6 */ 3761 3762 static int 3763 if_sdl_sysctl(SYSCTLFN_ARGS) 3764 { 3765 struct ifnet *ifp; 3766 const struct sockaddr_dl *sdl; 3767 struct psref psref; 3768 int error = 0; 3769 int bound; 3770 3771 if (namelen != 1) 3772 return EINVAL; 3773 3774 bound = curlwp_bind(); 3775 ifp = if_get_byindex(name[0], &psref); 3776 if (ifp == NULL) { 3777 error = ENODEV; 3778 goto out0; 3779 } 3780 3781 sdl = ifp->if_sadl; 3782 if (sdl == NULL) { 3783 *oldlenp = 0; 3784 goto out1; 3785 } 3786 3787 if (oldp == NULL) { 3788 *oldlenp = sdl->sdl_alen; 3789 goto out1; 3790 } 3791 3792 if (*oldlenp >= sdl->sdl_alen) 3793 *oldlenp = sdl->sdl_alen; 3794 error = sysctl_copyout(l, &sdl->sdl_data[sdl->sdl_nlen], oldp, *oldlenp); 3795 out1: 3796 if_put(ifp, &psref); 3797 out0: 3798 curlwp_bindx(bound); 3799 return error; 3800 } 3801 3802 static void 3803 if_sysctl_setup(struct sysctllog **clog) 3804 { 3805 const struct sysctlnode *rnode = NULL; 3806 3807 sysctl_createv(clog, 0, NULL, &rnode, 3808 CTLFLAG_PERMANENT, 3809 CTLTYPE_NODE, "sdl", 3810 SYSCTL_DESCR("Get active link-layer address"), 3811 if_sdl_sysctl, 0, NULL, 0, 3812 CTL_NET, CTL_CREATE, CTL_EOL); 3813 3814 #if defined(INET) 3815 sysctl_net_pktq_setup(NULL, PF_INET); 3816 #endif 3817 #ifdef INET6 3818 if (in6_present) 3819 sysctl_net_pktq_setup(NULL, PF_INET6); 3820 #endif 3821 } 3822
Indexes created Tue Sep 23 14:10:03 GMT 2025