if.c revision 1.427
1 /* $NetBSD: if.c,v 1.427 2018/06/01 07:16:23 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 1.427 2018/06/01 07:16:23 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 *, int); 206 static void if_attachdomain1(struct ifnet *); 207 static int ifconf(u_long, void *); 208 static int if_transmit(struct ifnet *, struct mbuf *); 209 static int if_clone_create(const char *); 210 static int if_clone_destroy(const char *); 211 static void if_link_state_change_si(void *); 212 static void if_up_locked(struct ifnet *); 213 static void _if_down(struct ifnet *); 214 static void if_down_deactivated(struct ifnet *); 215 216 struct if_percpuq { 217 struct ifnet *ipq_ifp; 218 void *ipq_si; 219 struct percpu *ipq_ifqs; /* struct ifqueue */ 220 }; 221 222 static struct mbuf *if_percpuq_dequeue(struct if_percpuq *); 223 224 static void if_percpuq_drops(void *, void *, struct cpu_info *); 225 static int sysctl_percpuq_drops_handler(SYSCTLFN_PROTO); 226 static void sysctl_percpuq_setup(struct sysctllog **, const char *, 227 struct if_percpuq *); 228 229 struct if_deferred_start { 230 struct ifnet *ids_ifp; 231 void (*ids_if_start)(struct ifnet *); 232 void *ids_si; 233 }; 234 235 static void if_deferred_start_softint(void *); 236 static void if_deferred_start_common(struct ifnet *); 237 static void if_deferred_start_destroy(struct ifnet *); 238 239 #if defined(INET) || defined(INET6) 240 static void sysctl_net_pktq_setup(struct sysctllog **, int); 241 #endif 242 243 static void if_sysctl_setup(struct sysctllog **); 244 245 /* Compatibility vector functions */ 246 u_long (*vec_compat_cvtcmd)(u_long) = NULL; 247 int (*vec_compat_ifioctl)(struct socket *, u_long, u_long, void *, 248 struct lwp *) = NULL; 249 int (*vec_compat_ifconf)(struct lwp *, u_long, void *) = (void *)enosys; 250 int (*vec_compat_ifdatareq)(struct lwp *, u_long, void *) = (void *)enosys; 251 252 static int 253 if_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie, 254 void *arg0, void *arg1, void *arg2, void *arg3) 255 { 256 int result; 257 enum kauth_network_req req; 258 259 result = KAUTH_RESULT_DEFER; 260 req = (enum kauth_network_req)arg1; 261 262 if (action != KAUTH_NETWORK_INTERFACE) 263 return result; 264 265 if ((req == KAUTH_REQ_NETWORK_INTERFACE_GET) || 266 (req == KAUTH_REQ_NETWORK_INTERFACE_SET)) 267 result = KAUTH_RESULT_ALLOW; 268 269 return result; 270 } 271 272 /* 273 * Network interface utility routines. 274 * 275 * Routines with ifa_ifwith* names take sockaddr *'s as 276 * parameters. 277 */ 278 void 279 ifinit(void) 280 { 281 282 if_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 { 302 303 #ifdef NET_MPSAFE 304 printf("NET_MPSAFE enabled\n"); 305 #endif 306 307 mutex_init(&if_clone_mtx, MUTEX_DEFAULT, IPL_NONE); 308 309 TAILQ_INIT(&ifnet_list); 310 mutex_init(&ifnet_mtx, MUTEX_DEFAULT, IPL_NONE); 311 ifnet_psz = pserialize_create(); 312 ifnet_psref_class = psref_class_create("ifnet", IPL_SOFTNET); 313 ifa_psref_class = psref_class_create("ifa", IPL_SOFTNET); 314 PSLIST_INIT(&ifnet_pslist); 315 316 if_indexlim = 8; 317 318 if_pfil = pfil_head_create(PFIL_TYPE_IFNET, NULL); 319 KASSERT(if_pfil != NULL); 320 321 #if NETHER > 0 || NFDDI > 0 || defined(NETATALK) || NTOKEN > 0 || defined(WLAN) 322 etherinit(); 323 #endif 324 } 325 326 ifnet_t * 327 if_alloc(u_char type) 328 { 329 return kmem_zalloc(sizeof(ifnet_t), KM_SLEEP); 330 } 331 332 void 333 if_free(ifnet_t *ifp) 334 { 335 kmem_free(ifp, sizeof(ifnet_t)); 336 } 337 338 void 339 if_initname(struct ifnet *ifp, const char *name, int unit) 340 { 341 (void)snprintf(ifp->if_xname, sizeof(ifp->if_xname), 342 "%s%d", name, unit); 343 } 344 345 /* 346 * Null routines used while an interface is going away. These routines 347 * just return an error. 348 */ 349 350 int 351 if_nulloutput(struct ifnet *ifp, struct mbuf *m, 352 const struct sockaddr *so, const struct rtentry *rt) 353 { 354 355 return ENXIO; 356 } 357 358 void 359 if_nullinput(struct ifnet *ifp, struct mbuf *m) 360 { 361 362 /* Nothing. */ 363 } 364 365 void 366 if_nullstart(struct ifnet *ifp) 367 { 368 369 /* Nothing. */ 370 } 371 372 int 373 if_nulltransmit(struct ifnet *ifp, struct mbuf *m) 374 { 375 376 m_freem(m); 377 return ENXIO; 378 } 379 380 int 381 if_nullioctl(struct ifnet *ifp, u_long cmd, void *data) 382 { 383 384 return ENXIO; 385 } 386 387 int 388 if_nullinit(struct ifnet *ifp) 389 { 390 391 return ENXIO; 392 } 393 394 void 395 if_nullstop(struct ifnet *ifp, int disable) 396 { 397 398 /* Nothing. */ 399 } 400 401 void 402 if_nullslowtimo(struct ifnet *ifp) 403 { 404 405 /* Nothing. */ 406 } 407 408 void 409 if_nulldrain(struct ifnet *ifp) 410 { 411 412 /* Nothing. */ 413 } 414 415 void 416 if_set_sadl(struct ifnet *ifp, const void *lla, u_char addrlen, bool factory) 417 { 418 struct ifaddr *ifa; 419 struct sockaddr_dl *sdl; 420 421 ifp->if_addrlen = addrlen; 422 if_alloc_sadl(ifp); 423 ifa = ifp->if_dl; 424 sdl = satosdl(ifa->ifa_addr); 425 426 (void)sockaddr_dl_setaddr(sdl, sdl->sdl_len, lla, ifp->if_addrlen); 427 if (factory) { 428 KASSERT(ifp->if_hwdl == NULL); 429 ifp->if_hwdl = ifp->if_dl; 430 ifaref(ifp->if_hwdl); 431 } 432 /* TBD routing socket */ 433 } 434 435 struct ifaddr * 436 if_dl_create(const struct ifnet *ifp, const struct sockaddr_dl **sdlp) 437 { 438 unsigned socksize, ifasize; 439 int addrlen, namelen; 440 struct sockaddr_dl *mask, *sdl; 441 struct ifaddr *ifa; 442 443 namelen = strlen(ifp->if_xname); 444 addrlen = ifp->if_addrlen; 445 socksize = roundup(sockaddr_dl_measure(namelen, addrlen), sizeof(long)); 446 ifasize = sizeof(*ifa) + 2 * socksize; 447 ifa = malloc(ifasize, M_IFADDR, M_WAITOK|M_ZERO); 448 449 sdl = (struct sockaddr_dl *)(ifa + 1); 450 mask = (struct sockaddr_dl *)(socksize + (char *)sdl); 451 452 sockaddr_dl_init(sdl, socksize, ifp->if_index, ifp->if_type, 453 ifp->if_xname, namelen, NULL, addrlen); 454 mask->sdl_family = AF_LINK; 455 mask->sdl_len = sockaddr_dl_measure(namelen, 0); 456 memset(&mask->sdl_data[0], 0xff, namelen); 457 ifa->ifa_rtrequest = link_rtrequest; 458 ifa->ifa_addr = (struct sockaddr *)sdl; 459 ifa->ifa_netmask = (struct sockaddr *)mask; 460 ifa_psref_init(ifa); 461 462 *sdlp = sdl; 463 464 return ifa; 465 } 466 467 static void 468 if_sadl_setrefs(struct ifnet *ifp, struct ifaddr *ifa) 469 { 470 const struct sockaddr_dl *sdl; 471 472 ifp->if_dl = ifa; 473 ifaref(ifa); 474 sdl = satosdl(ifa->ifa_addr); 475 ifp->if_sadl = sdl; 476 } 477 478 /* 479 * Allocate the link level name for the specified interface. This 480 * is an attachment helper. It must be called after ifp->if_addrlen 481 * is initialized, which may not be the case when if_attach() is 482 * called. 483 */ 484 void 485 if_alloc_sadl(struct ifnet *ifp) 486 { 487 struct ifaddr *ifa; 488 const struct sockaddr_dl *sdl; 489 490 /* 491 * If the interface already has a link name, release it 492 * now. This is useful for interfaces that can change 493 * link types, and thus switch link names often. 494 */ 495 if (ifp->if_sadl != NULL) 496 if_free_sadl(ifp, 0); 497 498 ifa = if_dl_create(ifp, &sdl); 499 500 ifa_insert(ifp, ifa); 501 if_sadl_setrefs(ifp, ifa); 502 } 503 504 static void 505 if_deactivate_sadl(struct ifnet *ifp) 506 { 507 struct ifaddr *ifa; 508 509 KASSERT(ifp->if_dl != NULL); 510 511 ifa = ifp->if_dl; 512 513 ifp->if_sadl = NULL; 514 515 ifp->if_dl = NULL; 516 ifafree(ifa); 517 } 518 519 static void 520 if_replace_sadl(struct ifnet *ifp, struct ifaddr *ifa) 521 { 522 struct ifaddr *old; 523 524 KASSERT(ifp->if_dl != NULL); 525 526 old = ifp->if_dl; 527 528 ifaref(ifa); 529 /* XXX Update if_dl and if_sadl atomically */ 530 ifp->if_dl = ifa; 531 ifp->if_sadl = satosdl(ifa->ifa_addr); 532 533 ifafree(old); 534 } 535 536 void 537 if_activate_sadl(struct ifnet *ifp, struct ifaddr *ifa0, 538 const struct sockaddr_dl *sdl) 539 { 540 int s, ss; 541 struct ifaddr *ifa; 542 int bound = curlwp_bind(); 543 544 KASSERT(ifa_held(ifa0)); 545 546 s = splsoftnet(); 547 548 if_replace_sadl(ifp, ifa0); 549 550 ss = pserialize_read_enter(); 551 IFADDR_READER_FOREACH(ifa, ifp) { 552 struct psref psref; 553 ifa_acquire(ifa, &psref); 554 pserialize_read_exit(ss); 555 556 rtinit(ifa, RTM_LLINFO_UPD, 0); 557 558 ss = pserialize_read_enter(); 559 ifa_release(ifa, &psref); 560 } 561 pserialize_read_exit(ss); 562 563 splx(s); 564 curlwp_bindx(bound); 565 } 566 567 /* 568 * Free the link level name for the specified interface. This is 569 * a detach helper. This is called from if_detach(). 570 */ 571 static void 572 if_free_sadl(struct ifnet *ifp, int factory) 573 { 574 struct ifaddr *ifa; 575 int s; 576 577 if (factory && ifp->if_hwdl != NULL) { 578 ifa = ifp->if_hwdl; 579 ifp->if_hwdl = NULL; 580 ifafree(ifa); 581 } 582 583 ifa = ifp->if_dl; 584 if (ifa == NULL) { 585 KASSERT(ifp->if_sadl == NULL); 586 return; 587 } 588 589 KASSERT(ifp->if_sadl != NULL); 590 591 s = splsoftnet(); 592 rtinit(ifa, RTM_DELETE, 0); 593 ifa_remove(ifp, ifa); 594 if_deactivate_sadl(ifp); 595 splx(s); 596 } 597 598 static void 599 if_getindex(ifnet_t *ifp) 600 { 601 bool hitlimit = false; 602 603 ifp->if_index_gen = index_gen++; 604 605 ifp->if_index = if_index; 606 if (ifindex2ifnet == NULL) { 607 if_index++; 608 goto skip; 609 } 610 while (if_byindex(ifp->if_index)) { 611 /* 612 * If we hit USHRT_MAX, we skip back to 0 since 613 * there are a number of places where the value 614 * of if_index or if_index itself is compared 615 * to or stored in an unsigned short. By 616 * jumping back, we won't botch those assignments 617 * or comparisons. 618 */ 619 if (++if_index == 0) { 620 if_index = 1; 621 } else if (if_index == USHRT_MAX) { 622 /* 623 * However, if we have to jump back to 624 * zero *twice* without finding an empty 625 * slot in ifindex2ifnet[], then there 626 * there are too many (>65535) interfaces. 627 */ 628 if (hitlimit) { 629 panic("too many interfaces"); 630 } 631 hitlimit = true; 632 if_index = 1; 633 } 634 ifp->if_index = if_index; 635 } 636 skip: 637 /* 638 * ifindex2ifnet is indexed by if_index. Since if_index will 639 * grow dynamically, it should grow too. 640 */ 641 if (ifindex2ifnet == NULL || ifp->if_index >= if_indexlim) { 642 size_t m, n, oldlim; 643 void *q; 644 645 oldlim = if_indexlim; 646 while (ifp->if_index >= if_indexlim) 647 if_indexlim <<= 1; 648 649 /* grow ifindex2ifnet */ 650 m = oldlim * sizeof(struct ifnet *); 651 n = if_indexlim * sizeof(struct ifnet *); 652 q = malloc(n, M_IFADDR, M_WAITOK|M_ZERO); 653 if (ifindex2ifnet != NULL) { 654 memcpy(q, ifindex2ifnet, m); 655 free(ifindex2ifnet, M_IFADDR); 656 } 657 ifindex2ifnet = (struct ifnet **)q; 658 } 659 ifindex2ifnet[ifp->if_index] = ifp; 660 } 661 662 /* 663 * Initialize an interface and assign an index for it. 664 * 665 * It must be called prior to a device specific attach routine 666 * (e.g., ether_ifattach and ieee80211_ifattach) or if_alloc_sadl, 667 * and be followed by if_register: 668 * 669 * if_initialize(ifp); 670 * ether_ifattach(ifp, enaddr); 671 * if_register(ifp); 672 */ 673 int 674 if_initialize(ifnet_t *ifp) 675 { 676 int rv = 0; 677 678 KASSERT(if_indexlim > 0); 679 TAILQ_INIT(&ifp->if_addrlist); 680 681 /* 682 * Link level name is allocated later by a separate call to 683 * if_alloc_sadl(). 684 */ 685 686 if (ifp->if_snd.ifq_maxlen == 0) 687 ifp->if_snd.ifq_maxlen = ifqmaxlen; 688 689 ifp->if_broadcastaddr = 0; /* reliably crash if used uninitialized */ 690 691 ifp->if_link_state = LINK_STATE_UNKNOWN; 692 ifp->if_link_queue = -1; /* all bits set, see link_state_change() */ 693 694 ifp->if_capenable = 0; 695 ifp->if_csum_flags_tx = 0; 696 ifp->if_csum_flags_rx = 0; 697 698 #ifdef ALTQ 699 ifp->if_snd.altq_type = 0; 700 ifp->if_snd.altq_disc = NULL; 701 ifp->if_snd.altq_flags &= ALTQF_CANTCHANGE; 702 ifp->if_snd.altq_tbr = NULL; 703 ifp->if_snd.altq_ifp = ifp; 704 #endif 705 706 IFQ_LOCK_INIT(&ifp->if_snd); 707 708 ifp->if_pfil = pfil_head_create(PFIL_TYPE_IFNET, ifp); 709 pfil_run_ifhooks(if_pfil, PFIL_IFNET_ATTACH, ifp); 710 711 IF_AFDATA_LOCK_INIT(ifp); 712 713 if (if_is_link_state_changeable(ifp)) { 714 u_int flags = SOFTINT_NET; 715 flags |= if_is_mpsafe(ifp) ? SOFTINT_MPSAFE : 0; 716 ifp->if_link_si = softint_establish(flags, 717 if_link_state_change_si, ifp); 718 if (ifp->if_link_si == NULL) { 719 rv = ENOMEM; 720 goto fail; 721 } 722 } 723 724 PSLIST_ENTRY_INIT(ifp, if_pslist_entry); 725 PSLIST_INIT(&ifp->if_addr_pslist); 726 psref_target_init(&ifp->if_psref, ifnet_psref_class); 727 ifp->if_ioctl_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 728 LIST_INIT(&ifp->if_multiaddrs); 729 730 IFNET_GLOBAL_LOCK(); 731 if_getindex(ifp); 732 IFNET_GLOBAL_UNLOCK(); 733 734 return 0; 735 736 fail: 737 IF_AFDATA_LOCK_DESTROY(ifp); 738 739 pfil_run_ifhooks(if_pfil, PFIL_IFNET_DETACH, ifp); 740 (void)pfil_head_destroy(ifp->if_pfil); 741 742 IFQ_LOCK_DESTROY(&ifp->if_snd); 743 744 return rv; 745 } 746 747 /* 748 * Register an interface to the list of "active" interfaces. 749 */ 750 void 751 if_register(ifnet_t *ifp) 752 { 753 /* 754 * If the driver has not supplied its own if_ioctl, then 755 * supply the default. 756 */ 757 if (ifp->if_ioctl == NULL) 758 ifp->if_ioctl = ifioctl_common; 759 760 sysctl_sndq_setup(&ifp->if_sysctl_log, ifp->if_xname, &ifp->if_snd); 761 762 if (!STAILQ_EMPTY(&domains)) 763 if_attachdomain1(ifp); 764 765 /* Announce the interface. */ 766 rt_ifannouncemsg(ifp, IFAN_ARRIVAL); 767 768 if (ifp->if_slowtimo != NULL) { 769 ifp->if_slowtimo_ch = 770 kmem_zalloc(sizeof(*ifp->if_slowtimo_ch), KM_SLEEP); 771 callout_init(ifp->if_slowtimo_ch, 0); 772 callout_setfunc(ifp->if_slowtimo_ch, if_slowtimo, ifp); 773 if_slowtimo(ifp); 774 } 775 776 if (ifp->if_transmit == NULL || ifp->if_transmit == if_nulltransmit) 777 ifp->if_transmit = if_transmit; 778 779 IFNET_GLOBAL_LOCK(); 780 TAILQ_INSERT_TAIL(&ifnet_list, ifp, if_list); 781 IFNET_WRITER_INSERT_TAIL(ifp); 782 IFNET_GLOBAL_UNLOCK(); 783 } 784 785 /* 786 * The if_percpuq framework 787 * 788 * It allows network device drivers to execute the network stack 789 * in softint (so called softint-based if_input). It utilizes 790 * softint and percpu ifqueue. It doesn't distribute any packets 791 * between CPUs, unlike pktqueue(9). 792 * 793 * Currently we support two options for device drivers to apply the framework: 794 * - Use it implicitly with less changes 795 * - If you use if_attach in driver's _attach function and if_input in 796 * driver's Rx interrupt handler, a packet is queued and a softint handles 797 * the packet implicitly 798 * - Use it explicitly in each driver (recommended) 799 * - You can use if_percpuq_* directly in your driver 800 * - In this case, you need to allocate struct if_percpuq in driver's softc 801 * - See wm(4) as a reference implementation 802 */ 803 804 static void 805 if_percpuq_softint(void *arg) 806 { 807 struct if_percpuq *ipq = arg; 808 struct ifnet *ifp = ipq->ipq_ifp; 809 struct mbuf *m; 810 811 while ((m = if_percpuq_dequeue(ipq)) != NULL) { 812 ifp->if_ipackets++; 813 bpf_mtap(ifp, m); 814 815 ifp->_if_input(ifp, m); 816 } 817 } 818 819 static void 820 if_percpuq_init_ifq(void *p, void *arg __unused, struct cpu_info *ci __unused) 821 { 822 struct ifqueue *const ifq = p; 823 824 memset(ifq, 0, sizeof(*ifq)); 825 ifq->ifq_maxlen = IFQ_MAXLEN; 826 } 827 828 struct if_percpuq * 829 if_percpuq_create(struct ifnet *ifp) 830 { 831 struct if_percpuq *ipq; 832 u_int flags = SOFTINT_NET; 833 834 flags |= if_is_mpsafe(ifp) ? SOFTINT_MPSAFE : 0; 835 836 ipq = kmem_zalloc(sizeof(*ipq), KM_SLEEP); 837 ipq->ipq_ifp = ifp; 838 ipq->ipq_si = softint_establish(flags, if_percpuq_softint, ipq); 839 ipq->ipq_ifqs = percpu_alloc(sizeof(struct ifqueue)); 840 percpu_foreach(ipq->ipq_ifqs, &if_percpuq_init_ifq, NULL); 841 842 sysctl_percpuq_setup(&ifp->if_sysctl_log, ifp->if_xname, ipq); 843 844 return ipq; 845 } 846 847 static struct mbuf * 848 if_percpuq_dequeue(struct if_percpuq *ipq) 849 { 850 struct mbuf *m; 851 struct ifqueue *ifq; 852 int s; 853 854 s = splnet(); 855 ifq = percpu_getref(ipq->ipq_ifqs); 856 IF_DEQUEUE(ifq, m); 857 percpu_putref(ipq->ipq_ifqs); 858 splx(s); 859 860 return m; 861 } 862 863 static void 864 if_percpuq_purge_ifq(void *p, void *arg __unused, struct cpu_info *ci __unused) 865 { 866 struct ifqueue *const ifq = p; 867 868 IF_PURGE(ifq); 869 } 870 871 void 872 if_percpuq_destroy(struct if_percpuq *ipq) 873 { 874 875 /* if_detach may already destroy it */ 876 if (ipq == NULL) 877 return; 878 879 softint_disestablish(ipq->ipq_si); 880 percpu_foreach(ipq->ipq_ifqs, &if_percpuq_purge_ifq, NULL); 881 percpu_free(ipq->ipq_ifqs, sizeof(struct ifqueue)); 882 kmem_free(ipq, sizeof(*ipq)); 883 } 884 885 void 886 if_percpuq_enqueue(struct if_percpuq *ipq, struct mbuf *m) 887 { 888 struct ifqueue *ifq; 889 int s; 890 891 KASSERT(ipq != NULL); 892 893 s = splnet(); 894 ifq = percpu_getref(ipq->ipq_ifqs); 895 if (IF_QFULL(ifq)) { 896 IF_DROP(ifq); 897 percpu_putref(ipq->ipq_ifqs); 898 m_freem(m); 899 goto out; 900 } 901 IF_ENQUEUE(ifq, m); 902 percpu_putref(ipq->ipq_ifqs); 903 904 softint_schedule(ipq->ipq_si); 905 out: 906 splx(s); 907 } 908 909 static void 910 if_percpuq_drops(void *p, void *arg, struct cpu_info *ci __unused) 911 { 912 struct ifqueue *const ifq = p; 913 int *sum = arg; 914 915 *sum += ifq->ifq_drops; 916 } 917 918 static int 919 sysctl_percpuq_drops_handler(SYSCTLFN_ARGS) 920 { 921 struct sysctlnode node; 922 struct if_percpuq *ipq; 923 int sum = 0; 924 int error; 925 926 node = *rnode; 927 ipq = node.sysctl_data; 928 929 percpu_foreach(ipq->ipq_ifqs, if_percpuq_drops, &sum); 930 931 node.sysctl_data = ∑ 932 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 933 if (error != 0 || newp == NULL) 934 return error; 935 936 return 0; 937 } 938 939 static void 940 sysctl_percpuq_setup(struct sysctllog **clog, const char* ifname, 941 struct if_percpuq *ipq) 942 { 943 const struct sysctlnode *cnode, *rnode; 944 945 if (sysctl_createv(clog, 0, NULL, &rnode, 946 CTLFLAG_PERMANENT, 947 CTLTYPE_NODE, "interfaces", 948 SYSCTL_DESCR("Per-interface controls"), 949 NULL, 0, NULL, 0, 950 CTL_NET, CTL_CREATE, CTL_EOL) != 0) 951 goto bad; 952 953 if (sysctl_createv(clog, 0, &rnode, &rnode, 954 CTLFLAG_PERMANENT, 955 CTLTYPE_NODE, ifname, 956 SYSCTL_DESCR("Interface controls"), 957 NULL, 0, NULL, 0, 958 CTL_CREATE, CTL_EOL) != 0) 959 goto bad; 960 961 if (sysctl_createv(clog, 0, &rnode, &rnode, 962 CTLFLAG_PERMANENT, 963 CTLTYPE_NODE, "rcvq", 964 SYSCTL_DESCR("Interface input queue controls"), 965 NULL, 0, NULL, 0, 966 CTL_CREATE, CTL_EOL) != 0) 967 goto bad; 968 969 #ifdef NOTYET 970 /* XXX Should show each per-CPU queue length? */ 971 if (sysctl_createv(clog, 0, &rnode, &rnode, 972 CTLFLAG_PERMANENT, 973 CTLTYPE_INT, "len", 974 SYSCTL_DESCR("Current input queue length"), 975 sysctl_percpuq_len, 0, NULL, 0, 976 CTL_CREATE, CTL_EOL) != 0) 977 goto bad; 978 979 if (sysctl_createv(clog, 0, &rnode, &cnode, 980 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 981 CTLTYPE_INT, "maxlen", 982 SYSCTL_DESCR("Maximum allowed input queue length"), 983 sysctl_percpuq_maxlen_handler, 0, (void *)ipq, 0, 984 CTL_CREATE, CTL_EOL) != 0) 985 goto bad; 986 #endif 987 988 if (sysctl_createv(clog, 0, &rnode, &cnode, 989 CTLFLAG_PERMANENT, 990 CTLTYPE_INT, "drops", 991 SYSCTL_DESCR("Total packets dropped due to full input queue"), 992 sysctl_percpuq_drops_handler, 0, (void *)ipq, 0, 993 CTL_CREATE, CTL_EOL) != 0) 994 goto bad; 995 996 return; 997 bad: 998 printf("%s: could not attach sysctl nodes\n", ifname); 999 return; 1000 } 1001 1002 /* 1003 * The deferred if_start framework 1004 * 1005 * The common APIs to defer if_start to softint when if_start is requested 1006 * from a device driver running in hardware interrupt context. 1007 */ 1008 /* 1009 * Call ifp->if_start (or equivalent) in a dedicated softint for 1010 * deferred if_start. 1011 */ 1012 static void 1013 if_deferred_start_softint(void *arg) 1014 { 1015 struct if_deferred_start *ids = arg; 1016 struct ifnet *ifp = ids->ids_ifp; 1017 1018 ids->ids_if_start(ifp); 1019 } 1020 1021 /* 1022 * The default callback function for deferred if_start. 1023 */ 1024 static void 1025 if_deferred_start_common(struct ifnet *ifp) 1026 { 1027 int s; 1028 1029 s = splnet(); 1030 if_start_lock(ifp); 1031 splx(s); 1032 } 1033 1034 static inline bool 1035 if_snd_is_used(struct ifnet *ifp) 1036 { 1037 1038 return ifp->if_transmit == NULL || ifp->if_transmit == if_nulltransmit || 1039 ALTQ_IS_ENABLED(&ifp->if_snd); 1040 } 1041 1042 /* 1043 * Schedule deferred if_start. 1044 */ 1045 void 1046 if_schedule_deferred_start(struct ifnet *ifp) 1047 { 1048 1049 KASSERT(ifp->if_deferred_start != NULL); 1050 1051 if (if_snd_is_used(ifp) && IFQ_IS_EMPTY(&ifp->if_snd)) 1052 return; 1053 1054 softint_schedule(ifp->if_deferred_start->ids_si); 1055 } 1056 1057 /* 1058 * Create an instance of deferred if_start. A driver should call the function 1059 * only if the driver needs deferred if_start. Drivers can setup their own 1060 * deferred if_start function via 2nd argument. 1061 */ 1062 void 1063 if_deferred_start_init(struct ifnet *ifp, void (*func)(struct ifnet *)) 1064 { 1065 struct if_deferred_start *ids; 1066 u_int flags = SOFTINT_NET; 1067 1068 flags |= if_is_mpsafe(ifp) ? SOFTINT_MPSAFE : 0; 1069 1070 ids = kmem_zalloc(sizeof(*ids), KM_SLEEP); 1071 ids->ids_ifp = ifp; 1072 ids->ids_si = softint_establish(flags, if_deferred_start_softint, ids); 1073 if (func != NULL) 1074 ids->ids_if_start = func; 1075 else 1076 ids->ids_if_start = if_deferred_start_common; 1077 1078 ifp->if_deferred_start = ids; 1079 } 1080 1081 static void 1082 if_deferred_start_destroy(struct ifnet *ifp) 1083 { 1084 1085 if (ifp->if_deferred_start == NULL) 1086 return; 1087 1088 softint_disestablish(ifp->if_deferred_start->ids_si); 1089 kmem_free(ifp->if_deferred_start, sizeof(*ifp->if_deferred_start)); 1090 ifp->if_deferred_start = NULL; 1091 } 1092 1093 /* 1094 * The common interface input routine that is called by device drivers, 1095 * which should be used only when the driver's rx handler already runs 1096 * in softint. 1097 */ 1098 void 1099 if_input(struct ifnet *ifp, struct mbuf *m) 1100 { 1101 1102 KASSERT(ifp->if_percpuq == NULL); 1103 KASSERT(!cpu_intr_p()); 1104 1105 ifp->if_ipackets++; 1106 bpf_mtap(ifp, m); 1107 1108 ifp->_if_input(ifp, m); 1109 } 1110 1111 /* 1112 * DEPRECATED. Use if_initialize and if_register instead. 1113 * See the above comment of if_initialize. 1114 * 1115 * Note that it implicitly enables if_percpuq to make drivers easy to 1116 * migrate softint-based if_input without much changes. If you don't 1117 * want to enable it, use if_initialize instead. 1118 */ 1119 int 1120 if_attach(ifnet_t *ifp) 1121 { 1122 int rv; 1123 1124 rv = if_initialize(ifp); 1125 if (rv != 0) 1126 return rv; 1127 1128 ifp->if_percpuq = if_percpuq_create(ifp); 1129 if_register(ifp); 1130 1131 return 0; 1132 } 1133 1134 void 1135 if_attachdomain(void) 1136 { 1137 struct ifnet *ifp; 1138 int s; 1139 int bound = curlwp_bind(); 1140 1141 s = pserialize_read_enter(); 1142 IFNET_READER_FOREACH(ifp) { 1143 struct psref psref; 1144 psref_acquire(&psref, &ifp->if_psref, ifnet_psref_class); 1145 pserialize_read_exit(s); 1146 if_attachdomain1(ifp); 1147 s = pserialize_read_enter(); 1148 psref_release(&psref, &ifp->if_psref, ifnet_psref_class); 1149 } 1150 pserialize_read_exit(s); 1151 curlwp_bindx(bound); 1152 } 1153 1154 static void 1155 if_attachdomain1(struct ifnet *ifp) 1156 { 1157 struct domain *dp; 1158 int s; 1159 1160 s = splsoftnet(); 1161 1162 /* address family dependent data region */ 1163 memset(ifp->if_afdata, 0, sizeof(ifp->if_afdata)); 1164 DOMAIN_FOREACH(dp) { 1165 if (dp->dom_ifattach != NULL) 1166 ifp->if_afdata[dp->dom_family] = 1167 (*dp->dom_ifattach)(ifp); 1168 } 1169 1170 splx(s); 1171 } 1172 1173 /* 1174 * Deactivate an interface. This points all of the procedure 1175 * handles at error stubs. May be called from interrupt context. 1176 */ 1177 void 1178 if_deactivate(struct ifnet *ifp) 1179 { 1180 int s; 1181 1182 s = splsoftnet(); 1183 1184 ifp->if_output = if_nulloutput; 1185 ifp->_if_input = if_nullinput; 1186 ifp->if_start = if_nullstart; 1187 ifp->if_transmit = if_nulltransmit; 1188 ifp->if_ioctl = if_nullioctl; 1189 ifp->if_init = if_nullinit; 1190 ifp->if_stop = if_nullstop; 1191 ifp->if_slowtimo = if_nullslowtimo; 1192 ifp->if_drain = if_nulldrain; 1193 1194 /* No more packets may be enqueued. */ 1195 ifp->if_snd.ifq_maxlen = 0; 1196 1197 splx(s); 1198 } 1199 1200 bool 1201 if_is_deactivated(const struct ifnet *ifp) 1202 { 1203 1204 return ifp->if_output == if_nulloutput; 1205 } 1206 1207 void 1208 if_purgeaddrs(struct ifnet *ifp, int family, void (*purgeaddr)(struct ifaddr *)) 1209 { 1210 struct ifaddr *ifa, *nifa; 1211 int s; 1212 1213 s = pserialize_read_enter(); 1214 for (ifa = IFADDR_READER_FIRST(ifp); ifa; ifa = nifa) { 1215 nifa = IFADDR_READER_NEXT(ifa); 1216 if (ifa->ifa_addr->sa_family != family) 1217 continue; 1218 pserialize_read_exit(s); 1219 1220 (*purgeaddr)(ifa); 1221 1222 s = pserialize_read_enter(); 1223 } 1224 pserialize_read_exit(s); 1225 } 1226 1227 #ifdef IFAREF_DEBUG 1228 static struct ifaddr **ifa_list; 1229 static int ifa_list_size; 1230 1231 /* Depends on only one if_attach runs at once */ 1232 static void 1233 if_build_ifa_list(struct ifnet *ifp) 1234 { 1235 struct ifaddr *ifa; 1236 int i; 1237 1238 KASSERT(ifa_list == NULL); 1239 KASSERT(ifa_list_size == 0); 1240 1241 IFADDR_READER_FOREACH(ifa, ifp) 1242 ifa_list_size++; 1243 1244 ifa_list = kmem_alloc(sizeof(*ifa) * ifa_list_size, KM_SLEEP); 1245 i = 0; 1246 IFADDR_READER_FOREACH(ifa, ifp) { 1247 ifa_list[i++] = ifa; 1248 ifaref(ifa); 1249 } 1250 } 1251 1252 static void 1253 if_check_and_free_ifa_list(struct ifnet *ifp) 1254 { 1255 int i; 1256 struct ifaddr *ifa; 1257 1258 if (ifa_list == NULL) 1259 return; 1260 1261 for (i = 0; i < ifa_list_size; i++) { 1262 char buf[64]; 1263 1264 ifa = ifa_list[i]; 1265 sockaddr_format(ifa->ifa_addr, buf, sizeof(buf)); 1266 if (ifa->ifa_refcnt > 1) { 1267 log(LOG_WARNING, 1268 "ifa(%s) still referenced (refcnt=%d)\n", 1269 buf, ifa->ifa_refcnt - 1); 1270 } else 1271 log(LOG_DEBUG, 1272 "ifa(%s) not referenced (refcnt=%d)\n", 1273 buf, ifa->ifa_refcnt - 1); 1274 ifafree(ifa); 1275 } 1276 1277 kmem_free(ifa_list, sizeof(*ifa) * ifa_list_size); 1278 ifa_list = NULL; 1279 ifa_list_size = 0; 1280 } 1281 #endif 1282 1283 /* 1284 * Detach an interface from the list of "active" interfaces, 1285 * freeing any resources as we go along. 1286 * 1287 * NOTE: This routine must be called with a valid thread context, 1288 * as it may block. 1289 */ 1290 void 1291 if_detach(struct ifnet *ifp) 1292 { 1293 struct socket so; 1294 struct ifaddr *ifa; 1295 #ifdef IFAREF_DEBUG 1296 struct ifaddr *last_ifa = NULL; 1297 #endif 1298 struct domain *dp; 1299 const struct protosw *pr; 1300 int s, i, family, purged; 1301 uint64_t xc; 1302 1303 #ifdef IFAREF_DEBUG 1304 if_build_ifa_list(ifp); 1305 #endif 1306 /* 1307 * XXX It's kind of lame that we have to have the 1308 * XXX socket structure... 1309 */ 1310 memset(&so, 0, sizeof(so)); 1311 1312 s = splnet(); 1313 1314 sysctl_teardown(&ifp->if_sysctl_log); 1315 IFNET_LOCK(ifp); 1316 if_deactivate(ifp); 1317 IFNET_UNLOCK(ifp); 1318 1319 if (ifp->if_slowtimo != NULL && ifp->if_slowtimo_ch != NULL) { 1320 ifp->if_slowtimo = NULL; 1321 callout_halt(ifp->if_slowtimo_ch, NULL); 1322 callout_destroy(ifp->if_slowtimo_ch); 1323 kmem_free(ifp->if_slowtimo_ch, sizeof(*ifp->if_slowtimo_ch)); 1324 } 1325 if_deferred_start_destroy(ifp); 1326 1327 /* 1328 * Do an if_down() to give protocols a chance to do something. 1329 */ 1330 if_down_deactivated(ifp); 1331 1332 #ifdef ALTQ 1333 if (ALTQ_IS_ENABLED(&ifp->if_snd)) 1334 altq_disable(&ifp->if_snd); 1335 if (ALTQ_IS_ATTACHED(&ifp->if_snd)) 1336 altq_detach(&ifp->if_snd); 1337 #endif 1338 1339 #if NCARP > 0 1340 /* Remove the interface from any carp group it is a part of. */ 1341 if (ifp->if_carp != NULL && ifp->if_type != IFT_CARP) 1342 carp_ifdetach(ifp); 1343 #endif 1344 1345 /* 1346 * Rip all the addresses off the interface. This should make 1347 * all of the routes go away. 1348 * 1349 * pr_usrreq calls can remove an arbitrary number of ifaddrs 1350 * from the list, including our "cursor", ifa. For safety, 1351 * and to honor the TAILQ abstraction, I just restart the 1352 * loop after each removal. Note that the loop will exit 1353 * when all of the remaining ifaddrs belong to the AF_LINK 1354 * family. I am counting on the historical fact that at 1355 * least one pr_usrreq in each address domain removes at 1356 * least one ifaddr. 1357 */ 1358 again: 1359 /* 1360 * At this point, no other one tries to remove ifa in the list, 1361 * so we don't need to take a lock or psref. Avoid using 1362 * IFADDR_READER_FOREACH to pass over an inspection of contract 1363 * violations of pserialize. 1364 */ 1365 IFADDR_WRITER_FOREACH(ifa, ifp) { 1366 family = ifa->ifa_addr->sa_family; 1367 #ifdef IFAREF_DEBUG 1368 printf("if_detach: ifaddr %p, family %d, refcnt %d\n", 1369 ifa, family, ifa->ifa_refcnt); 1370 if (last_ifa != NULL && ifa == last_ifa) 1371 panic("if_detach: loop detected"); 1372 last_ifa = ifa; 1373 #endif 1374 if (family == AF_LINK) 1375 continue; 1376 dp = pffinddomain(family); 1377 KASSERTMSG(dp != NULL, "no domain for AF %d", family); 1378 /* 1379 * XXX These PURGEIF calls are redundant with the 1380 * purge-all-families calls below, but are left in for 1381 * now both to make a smaller change, and to avoid 1382 * unplanned interactions with clearing of 1383 * ifp->if_addrlist. 1384 */ 1385 purged = 0; 1386 for (pr = dp->dom_protosw; 1387 pr < dp->dom_protoswNPROTOSW; pr++) { 1388 so.so_proto = pr; 1389 if (pr->pr_usrreqs) { 1390 (void) (*pr->pr_usrreqs->pr_purgeif)(&so, ifp); 1391 purged = 1; 1392 } 1393 } 1394 if (purged == 0) { 1395 /* 1396 * XXX What's really the best thing to do 1397 * XXX here? --thorpej (at) NetBSD.org 1398 */ 1399 printf("if_detach: WARNING: AF %d not purged\n", 1400 family); 1401 ifa_remove(ifp, ifa); 1402 } 1403 goto again; 1404 } 1405 1406 if_free_sadl(ifp, 1); 1407 1408 restart: 1409 IFADDR_WRITER_FOREACH(ifa, ifp) { 1410 family = ifa->ifa_addr->sa_family; 1411 KASSERT(family == AF_LINK); 1412 ifa_remove(ifp, ifa); 1413 goto restart; 1414 } 1415 1416 /* Delete stray routes from the routing table. */ 1417 for (i = 0; i <= AF_MAX; i++) 1418 rt_delete_matched_entries(i, if_delroute_matcher, ifp); 1419 1420 DOMAIN_FOREACH(dp) { 1421 if (dp->dom_ifdetach != NULL && ifp->if_afdata[dp->dom_family]) 1422 { 1423 void *p = ifp->if_afdata[dp->dom_family]; 1424 if (p) { 1425 ifp->if_afdata[dp->dom_family] = NULL; 1426 (*dp->dom_ifdetach)(ifp, p); 1427 } 1428 } 1429 1430 /* 1431 * One would expect multicast memberships (INET and 1432 * INET6) on UDP sockets to be purged by the PURGEIF 1433 * calls above, but if all addresses were removed from 1434 * the interface prior to destruction, the calls will 1435 * not be made (e.g. ppp, for which pppd(8) generally 1436 * removes addresses before destroying the interface). 1437 * Because there is no invariant that multicast 1438 * memberships only exist for interfaces with IPv4 1439 * addresses, we must call PURGEIF regardless of 1440 * addresses. (Protocols which might store ifnet 1441 * pointers are marked with PR_PURGEIF.) 1442 */ 1443 for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) { 1444 so.so_proto = pr; 1445 if (pr->pr_usrreqs && pr->pr_flags & PR_PURGEIF) 1446 (void)(*pr->pr_usrreqs->pr_purgeif)(&so, ifp); 1447 } 1448 } 1449 1450 /* Wait for all readers to drain before freeing. */ 1451 IFNET_GLOBAL_LOCK(); 1452 ifindex2ifnet[ifp->if_index] = NULL; 1453 TAILQ_REMOVE(&ifnet_list, ifp, if_list); 1454 IFNET_WRITER_REMOVE(ifp); 1455 pserialize_perform(ifnet_psz); 1456 IFNET_GLOBAL_UNLOCK(); 1457 1458 psref_target_destroy(&ifp->if_psref, ifnet_psref_class); 1459 PSLIST_ENTRY_DESTROY(ifp, if_pslist_entry); 1460 1461 pfil_run_ifhooks(if_pfil, PFIL_IFNET_DETACH, ifp); 1462 (void)pfil_head_destroy(ifp->if_pfil); 1463 1464 /* Announce that the interface is gone. */ 1465 rt_ifannouncemsg(ifp, IFAN_DEPARTURE); 1466 1467 IF_AFDATA_LOCK_DESTROY(ifp); 1468 1469 if (if_is_link_state_changeable(ifp)) { 1470 softint_disestablish(ifp->if_link_si); 1471 ifp->if_link_si = NULL; 1472 } 1473 1474 /* 1475 * remove packets that came from ifp, from software interrupt queues. 1476 */ 1477 DOMAIN_FOREACH(dp) { 1478 for (i = 0; i < __arraycount(dp->dom_ifqueues); i++) { 1479 struct ifqueue *iq = dp->dom_ifqueues[i]; 1480 if (iq == NULL) 1481 break; 1482 dp->dom_ifqueues[i] = NULL; 1483 if_detach_queues(ifp, iq); 1484 } 1485 } 1486 1487 /* 1488 * IP queues have to be processed separately: net-queue barrier 1489 * ensures that the packets are dequeued while a cross-call will 1490 * ensure that the interrupts have completed. FIXME: not quite.. 1491 */ 1492 #ifdef INET 1493 pktq_barrier(ip_pktq); 1494 #endif 1495 #ifdef INET6 1496 if (in6_present) 1497 pktq_barrier(ip6_pktq); 1498 #endif 1499 xc = xc_broadcast(0, (xcfunc_t)nullop, NULL, NULL); 1500 xc_wait(xc); 1501 1502 if (ifp->if_percpuq != NULL) { 1503 if_percpuq_destroy(ifp->if_percpuq); 1504 ifp->if_percpuq = NULL; 1505 } 1506 1507 mutex_obj_free(ifp->if_ioctl_lock); 1508 ifp->if_ioctl_lock = NULL; 1509 mutex_obj_free(ifp->if_snd.ifq_lock); 1510 1511 splx(s); 1512 1513 #ifdef IFAREF_DEBUG 1514 if_check_and_free_ifa_list(ifp); 1515 #endif 1516 } 1517 1518 static void 1519 if_detach_queues(struct ifnet *ifp, struct ifqueue *q) 1520 { 1521 struct mbuf *m, *prev, *next; 1522 1523 prev = NULL; 1524 for (m = q->ifq_head; m != NULL; m = next) { 1525 KASSERT((m->m_flags & M_PKTHDR) != 0); 1526 1527 next = m->m_nextpkt; 1528 if (m->m_pkthdr.rcvif_index != ifp->if_index) { 1529 prev = m; 1530 continue; 1531 } 1532 1533 if (prev != NULL) 1534 prev->m_nextpkt = m->m_nextpkt; 1535 else 1536 q->ifq_head = m->m_nextpkt; 1537 if (q->ifq_tail == m) 1538 q->ifq_tail = prev; 1539 q->ifq_len--; 1540 1541 m->m_nextpkt = NULL; 1542 m_freem(m); 1543 IF_DROP(q); 1544 } 1545 } 1546 1547 /* 1548 * Callback for a radix tree walk to delete all references to an 1549 * ifnet. 1550 */ 1551 static int 1552 if_delroute_matcher(struct rtentry *rt, void *v) 1553 { 1554 struct ifnet *ifp = (struct ifnet *)v; 1555 1556 if (rt->rt_ifp == ifp) 1557 return 1; 1558 else 1559 return 0; 1560 } 1561 1562 /* 1563 * Create a clone network interface. 1564 */ 1565 static int 1566 if_clone_create(const char *name) 1567 { 1568 struct if_clone *ifc; 1569 int unit; 1570 struct ifnet *ifp; 1571 struct psref psref; 1572 1573 KASSERT(mutex_owned(&if_clone_mtx)); 1574 1575 ifc = if_clone_lookup(name, &unit); 1576 if (ifc == NULL) 1577 return EINVAL; 1578 1579 ifp = if_get(name, &psref); 1580 if (ifp != NULL) { 1581 if_put(ifp, &psref); 1582 return EEXIST; 1583 } 1584 1585 return (*ifc->ifc_create)(ifc, unit); 1586 } 1587 1588 /* 1589 * Destroy a clone network interface. 1590 */ 1591 static int 1592 if_clone_destroy(const char *name) 1593 { 1594 struct if_clone *ifc; 1595 struct ifnet *ifp; 1596 struct psref psref; 1597 1598 KASSERT(mutex_owned(&if_clone_mtx)); 1599 1600 ifc = if_clone_lookup(name, NULL); 1601 if (ifc == NULL) 1602 return EINVAL; 1603 1604 if (ifc->ifc_destroy == NULL) 1605 return EOPNOTSUPP; 1606 1607 ifp = if_get(name, &psref); 1608 if (ifp == NULL) 1609 return ENXIO; 1610 1611 /* We have to disable ioctls here */ 1612 IFNET_LOCK(ifp); 1613 ifp->if_ioctl = if_nullioctl; 1614 IFNET_UNLOCK(ifp); 1615 1616 /* 1617 * We cannot call ifc_destroy with holding ifp. 1618 * Releasing ifp here is safe thanks to if_clone_mtx. 1619 */ 1620 if_put(ifp, &psref); 1621 1622 return (*ifc->ifc_destroy)(ifp); 1623 } 1624 1625 static bool 1626 if_is_unit(const char *name) 1627 { 1628 1629 while(*name != '\0') { 1630 if (*name < '0' || *name > '9') 1631 return false; 1632 name++; 1633 } 1634 1635 return true; 1636 } 1637 1638 /* 1639 * Look up a network interface cloner. 1640 */ 1641 static struct if_clone * 1642 if_clone_lookup(const char *name, int *unitp) 1643 { 1644 struct if_clone *ifc; 1645 const char *cp; 1646 char *dp, ifname[IFNAMSIZ + 3]; 1647 int unit; 1648 1649 KASSERT(mutex_owned(&if_clone_mtx)); 1650 1651 strcpy(ifname, "if_"); 1652 /* separate interface name from unit */ 1653 /* TODO: search unit number from backward */ 1654 for (dp = ifname + 3, cp = name; cp - name < IFNAMSIZ && 1655 *cp && !if_is_unit(cp);) 1656 *dp++ = *cp++; 1657 1658 if (cp == name || cp - name == IFNAMSIZ || !*cp) 1659 return NULL; /* No name or unit number */ 1660 *dp++ = '\0'; 1661 1662 again: 1663 LIST_FOREACH(ifc, &if_cloners, ifc_list) { 1664 if (strcmp(ifname + 3, ifc->ifc_name) == 0) 1665 break; 1666 } 1667 1668 if (ifc == NULL) { 1669 int error; 1670 if (*ifname == '\0') 1671 return NULL; 1672 mutex_exit(&if_clone_mtx); 1673 error = module_autoload(ifname, MODULE_CLASS_DRIVER); 1674 mutex_enter(&if_clone_mtx); 1675 if (error) 1676 return NULL; 1677 *ifname = '\0'; 1678 goto again; 1679 } 1680 1681 unit = 0; 1682 while (cp - name < IFNAMSIZ && *cp) { 1683 if (*cp < '0' || *cp > '9' || unit >= INT_MAX / 10) { 1684 /* Bogus unit number. */ 1685 return NULL; 1686 } 1687 unit = (unit * 10) + (*cp++ - '0'); 1688 } 1689 1690 if (unitp != NULL) 1691 *unitp = unit; 1692 return ifc; 1693 } 1694 1695 /* 1696 * Register a network interface cloner. 1697 */ 1698 void 1699 if_clone_attach(struct if_clone *ifc) 1700 { 1701 1702 mutex_enter(&if_clone_mtx); 1703 LIST_INSERT_HEAD(&if_cloners, ifc, ifc_list); 1704 if_cloners_count++; 1705 mutex_exit(&if_clone_mtx); 1706 } 1707 1708 /* 1709 * Unregister a network interface cloner. 1710 */ 1711 void 1712 if_clone_detach(struct if_clone *ifc) 1713 { 1714 1715 mutex_enter(&if_clone_mtx); 1716 LIST_REMOVE(ifc, ifc_list); 1717 if_cloners_count--; 1718 mutex_exit(&if_clone_mtx); 1719 } 1720 1721 /* 1722 * Provide list of interface cloners to userspace. 1723 */ 1724 int 1725 if_clone_list(int buf_count, char *buffer, int *total) 1726 { 1727 char outbuf[IFNAMSIZ], *dst; 1728 struct if_clone *ifc; 1729 int count, error = 0; 1730 1731 mutex_enter(&if_clone_mtx); 1732 *total = if_cloners_count; 1733 if ((dst = buffer) == NULL) { 1734 /* Just asking how many there are. */ 1735 goto out; 1736 } 1737 1738 if (buf_count < 0) { 1739 error = EINVAL; 1740 goto out; 1741 } 1742 1743 count = (if_cloners_count < buf_count) ? 1744 if_cloners_count : buf_count; 1745 1746 for (ifc = LIST_FIRST(&if_cloners); ifc != NULL && count != 0; 1747 ifc = LIST_NEXT(ifc, ifc_list), count--, dst += IFNAMSIZ) { 1748 (void)strncpy(outbuf, ifc->ifc_name, sizeof(outbuf)); 1749 if (outbuf[sizeof(outbuf) - 1] != '\0') { 1750 error = ENAMETOOLONG; 1751 goto out; 1752 } 1753 error = copyout(outbuf, dst, sizeof(outbuf)); 1754 if (error != 0) 1755 break; 1756 } 1757 1758 out: 1759 mutex_exit(&if_clone_mtx); 1760 return error; 1761 } 1762 1763 void 1764 ifa_psref_init(struct ifaddr *ifa) 1765 { 1766 1767 psref_target_init(&ifa->ifa_psref, ifa_psref_class); 1768 } 1769 1770 void 1771 ifaref(struct ifaddr *ifa) 1772 { 1773 KASSERT(!ISSET(ifa->ifa_flags, IFA_DESTROYING)); 1774 ifa->ifa_refcnt++; 1775 } 1776 1777 void 1778 ifafree(struct ifaddr *ifa) 1779 { 1780 KASSERT(ifa != NULL); 1781 KASSERT(ifa->ifa_refcnt > 0); 1782 1783 if (--ifa->ifa_refcnt == 0) { 1784 free(ifa, M_IFADDR); 1785 } 1786 } 1787 1788 bool 1789 ifa_is_destroying(struct ifaddr *ifa) 1790 { 1791 1792 return ISSET(ifa->ifa_flags, IFA_DESTROYING); 1793 } 1794 1795 void 1796 ifa_insert(struct ifnet *ifp, struct ifaddr *ifa) 1797 { 1798 1799 ifa->ifa_ifp = ifp; 1800 1801 /* 1802 * Check MP-safety for IFEF_MPSAFE drivers. 1803 * Check !IFF_RUNNING for initialization routines that normally don't 1804 * take IFNET_LOCK but it's safe because there is no competitor. 1805 * XXX there are false positive cases because IFF_RUNNING can be off on 1806 * if_stop. 1807 */ 1808 KASSERT(!if_is_mpsafe(ifp) || !ISSET(ifp->if_flags, IFF_RUNNING) || 1809 IFNET_LOCKED(ifp)); 1810 1811 TAILQ_INSERT_TAIL(&ifp->if_addrlist, ifa, ifa_list); 1812 IFADDR_ENTRY_INIT(ifa); 1813 IFADDR_WRITER_INSERT_TAIL(ifp, ifa); 1814 1815 ifaref(ifa); 1816 } 1817 1818 void 1819 ifa_remove(struct ifnet *ifp, struct ifaddr *ifa) 1820 { 1821 1822 KASSERT(ifa->ifa_ifp == ifp); 1823 /* 1824 * Check MP-safety for IFEF_MPSAFE drivers. 1825 * if_is_deactivated indicates ifa_remove is called form if_detach 1826 * where is safe even if IFNET_LOCK isn't held. 1827 */ 1828 KASSERT(!if_is_mpsafe(ifp) || if_is_deactivated(ifp) || IFNET_LOCKED(ifp)); 1829 1830 TAILQ_REMOVE(&ifp->if_addrlist, ifa, ifa_list); 1831 IFADDR_WRITER_REMOVE(ifa); 1832 #ifdef NET_MPSAFE 1833 IFNET_GLOBAL_LOCK(); 1834 pserialize_perform(ifnet_psz); 1835 IFNET_GLOBAL_UNLOCK(); 1836 #endif 1837 1838 #ifdef NET_MPSAFE 1839 psref_target_destroy(&ifa->ifa_psref, ifa_psref_class); 1840 #endif 1841 IFADDR_ENTRY_DESTROY(ifa); 1842 ifafree(ifa); 1843 } 1844 1845 void 1846 ifa_acquire(struct ifaddr *ifa, struct psref *psref) 1847 { 1848 1849 psref_acquire(psref, &ifa->ifa_psref, ifa_psref_class); 1850 } 1851 1852 void 1853 ifa_release(struct ifaddr *ifa, struct psref *psref) 1854 { 1855 1856 if (ifa == NULL) 1857 return; 1858 1859 psref_release(psref, &ifa->ifa_psref, ifa_psref_class); 1860 } 1861 1862 bool 1863 ifa_held(struct ifaddr *ifa) 1864 { 1865 1866 return psref_held(&ifa->ifa_psref, ifa_psref_class); 1867 } 1868 1869 static inline int 1870 equal(const struct sockaddr *sa1, const struct sockaddr *sa2) 1871 { 1872 return sockaddr_cmp(sa1, sa2) == 0; 1873 } 1874 1875 /* 1876 * Locate an interface based on a complete address. 1877 */ 1878 /*ARGSUSED*/ 1879 struct ifaddr * 1880 ifa_ifwithaddr(const struct sockaddr *addr) 1881 { 1882 struct ifnet *ifp; 1883 struct ifaddr *ifa; 1884 1885 IFNET_READER_FOREACH(ifp) { 1886 if (if_is_deactivated(ifp)) 1887 continue; 1888 IFADDR_READER_FOREACH(ifa, ifp) { 1889 if (ifa->ifa_addr->sa_family != addr->sa_family) 1890 continue; 1891 if (equal(addr, ifa->ifa_addr)) 1892 return ifa; 1893 if ((ifp->if_flags & IFF_BROADCAST) && 1894 ifa->ifa_broadaddr && 1895 /* IP6 doesn't have broadcast */ 1896 ifa->ifa_broadaddr->sa_len != 0 && 1897 equal(ifa->ifa_broadaddr, addr)) 1898 return ifa; 1899 } 1900 } 1901 return NULL; 1902 } 1903 1904 struct ifaddr * 1905 ifa_ifwithaddr_psref(const struct sockaddr *addr, struct psref *psref) 1906 { 1907 struct ifaddr *ifa; 1908 int s = pserialize_read_enter(); 1909 1910 ifa = ifa_ifwithaddr(addr); 1911 if (ifa != NULL) 1912 ifa_acquire(ifa, psref); 1913 pserialize_read_exit(s); 1914 1915 return ifa; 1916 } 1917 1918 /* 1919 * Locate the point to point interface with a given destination address. 1920 */ 1921 /*ARGSUSED*/ 1922 struct ifaddr * 1923 ifa_ifwithdstaddr(const struct sockaddr *addr) 1924 { 1925 struct ifnet *ifp; 1926 struct ifaddr *ifa; 1927 1928 IFNET_READER_FOREACH(ifp) { 1929 if (if_is_deactivated(ifp)) 1930 continue; 1931 if ((ifp->if_flags & IFF_POINTOPOINT) == 0) 1932 continue; 1933 IFADDR_READER_FOREACH(ifa, ifp) { 1934 if (ifa->ifa_addr->sa_family != addr->sa_family || 1935 ifa->ifa_dstaddr == NULL) 1936 continue; 1937 if (equal(addr, ifa->ifa_dstaddr)) 1938 return ifa; 1939 } 1940 } 1941 1942 return NULL; 1943 } 1944 1945 struct ifaddr * 1946 ifa_ifwithdstaddr_psref(const struct sockaddr *addr, struct psref *psref) 1947 { 1948 struct ifaddr *ifa; 1949 int s; 1950 1951 s = pserialize_read_enter(); 1952 ifa = ifa_ifwithdstaddr(addr); 1953 if (ifa != NULL) 1954 ifa_acquire(ifa, psref); 1955 pserialize_read_exit(s); 1956 1957 return ifa; 1958 } 1959 1960 /* 1961 * Find an interface on a specific network. If many, choice 1962 * is most specific found. 1963 */ 1964 struct ifaddr * 1965 ifa_ifwithnet(const struct sockaddr *addr) 1966 { 1967 struct ifnet *ifp; 1968 struct ifaddr *ifa, *ifa_maybe = NULL; 1969 const struct sockaddr_dl *sdl; 1970 u_int af = addr->sa_family; 1971 const char *addr_data = addr->sa_data, *cplim; 1972 1973 if (af == AF_LINK) { 1974 sdl = satocsdl(addr); 1975 if (sdl->sdl_index && sdl->sdl_index < if_indexlim && 1976 ifindex2ifnet[sdl->sdl_index] && 1977 !if_is_deactivated(ifindex2ifnet[sdl->sdl_index])) { 1978 return ifindex2ifnet[sdl->sdl_index]->if_dl; 1979 } 1980 } 1981 #ifdef NETATALK 1982 if (af == AF_APPLETALK) { 1983 const struct sockaddr_at *sat, *sat2; 1984 sat = (const struct sockaddr_at *)addr; 1985 IFNET_READER_FOREACH(ifp) { 1986 if (if_is_deactivated(ifp)) 1987 continue; 1988 ifa = at_ifawithnet((const struct sockaddr_at *)addr, ifp); 1989 if (ifa == NULL) 1990 continue; 1991 sat2 = (struct sockaddr_at *)ifa->ifa_addr; 1992 if (sat2->sat_addr.s_net == sat->sat_addr.s_net) 1993 return ifa; /* exact match */ 1994 if (ifa_maybe == NULL) { 1995 /* else keep the if with the right range */ 1996 ifa_maybe = ifa; 1997 } 1998 } 1999 return ifa_maybe; 2000 } 2001 #endif 2002 IFNET_READER_FOREACH(ifp) { 2003 if (if_is_deactivated(ifp)) 2004 continue; 2005 IFADDR_READER_FOREACH(ifa, ifp) { 2006 const char *cp, *cp2, *cp3; 2007 2008 if (ifa->ifa_addr->sa_family != af || 2009 ifa->ifa_netmask == NULL) 2010 next: continue; 2011 cp = addr_data; 2012 cp2 = ifa->ifa_addr->sa_data; 2013 cp3 = ifa->ifa_netmask->sa_data; 2014 cplim = (const char *)ifa->ifa_netmask + 2015 ifa->ifa_netmask->sa_len; 2016 while (cp3 < cplim) { 2017 if ((*cp++ ^ *cp2++) & *cp3++) { 2018 /* want to continue for() loop */ 2019 goto next; 2020 } 2021 } 2022 if (ifa_maybe == NULL || 2023 rt_refines(ifa->ifa_netmask, 2024 ifa_maybe->ifa_netmask)) 2025 ifa_maybe = ifa; 2026 } 2027 } 2028 return ifa_maybe; 2029 } 2030 2031 struct ifaddr * 2032 ifa_ifwithnet_psref(const struct sockaddr *addr, struct psref *psref) 2033 { 2034 struct ifaddr *ifa; 2035 int s; 2036 2037 s = pserialize_read_enter(); 2038 ifa = ifa_ifwithnet(addr); 2039 if (ifa != NULL) 2040 ifa_acquire(ifa, psref); 2041 pserialize_read_exit(s); 2042 2043 return ifa; 2044 } 2045 2046 /* 2047 * Find the interface of the addresss. 2048 */ 2049 struct ifaddr * 2050 ifa_ifwithladdr(const struct sockaddr *addr) 2051 { 2052 struct ifaddr *ia; 2053 2054 if ((ia = ifa_ifwithaddr(addr)) || (ia = ifa_ifwithdstaddr(addr)) || 2055 (ia = ifa_ifwithnet(addr))) 2056 return ia; 2057 return NULL; 2058 } 2059 2060 struct ifaddr * 2061 ifa_ifwithladdr_psref(const struct sockaddr *addr, struct psref *psref) 2062 { 2063 struct ifaddr *ifa; 2064 int s; 2065 2066 s = pserialize_read_enter(); 2067 ifa = ifa_ifwithladdr(addr); 2068 if (ifa != NULL) 2069 ifa_acquire(ifa, psref); 2070 pserialize_read_exit(s); 2071 2072 return ifa; 2073 } 2074 2075 /* 2076 * Find an interface using a specific address family 2077 */ 2078 struct ifaddr * 2079 ifa_ifwithaf(int af) 2080 { 2081 struct ifnet *ifp; 2082 struct ifaddr *ifa = NULL; 2083 int s; 2084 2085 s = pserialize_read_enter(); 2086 IFNET_READER_FOREACH(ifp) { 2087 if (if_is_deactivated(ifp)) 2088 continue; 2089 IFADDR_READER_FOREACH(ifa, ifp) { 2090 if (ifa->ifa_addr->sa_family == af) 2091 goto out; 2092 } 2093 } 2094 out: 2095 pserialize_read_exit(s); 2096 return ifa; 2097 } 2098 2099 /* 2100 * Find an interface address specific to an interface best matching 2101 * a given address. 2102 */ 2103 struct ifaddr * 2104 ifaof_ifpforaddr(const struct sockaddr *addr, struct ifnet *ifp) 2105 { 2106 struct ifaddr *ifa; 2107 const char *cp, *cp2, *cp3; 2108 const char *cplim; 2109 struct ifaddr *ifa_maybe = 0; 2110 u_int af = addr->sa_family; 2111 2112 if (if_is_deactivated(ifp)) 2113 return NULL; 2114 2115 if (af >= AF_MAX) 2116 return NULL; 2117 2118 IFADDR_READER_FOREACH(ifa, ifp) { 2119 if (ifa->ifa_addr->sa_family != af) 2120 continue; 2121 ifa_maybe = ifa; 2122 if (ifa->ifa_netmask == NULL) { 2123 if (equal(addr, ifa->ifa_addr) || 2124 (ifa->ifa_dstaddr && 2125 equal(addr, ifa->ifa_dstaddr))) 2126 return ifa; 2127 continue; 2128 } 2129 cp = addr->sa_data; 2130 cp2 = ifa->ifa_addr->sa_data; 2131 cp3 = ifa->ifa_netmask->sa_data; 2132 cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask; 2133 for (; cp3 < cplim; cp3++) { 2134 if ((*cp++ ^ *cp2++) & *cp3) 2135 break; 2136 } 2137 if (cp3 == cplim) 2138 return ifa; 2139 } 2140 return ifa_maybe; 2141 } 2142 2143 struct ifaddr * 2144 ifaof_ifpforaddr_psref(const struct sockaddr *addr, struct ifnet *ifp, 2145 struct psref *psref) 2146 { 2147 struct ifaddr *ifa; 2148 int s; 2149 2150 s = pserialize_read_enter(); 2151 ifa = ifaof_ifpforaddr(addr, ifp); 2152 if (ifa != NULL) 2153 ifa_acquire(ifa, psref); 2154 pserialize_read_exit(s); 2155 2156 return ifa; 2157 } 2158 2159 /* 2160 * Default action when installing a route with a Link Level gateway. 2161 * Lookup an appropriate real ifa to point to. 2162 * This should be moved to /sys/net/link.c eventually. 2163 */ 2164 void 2165 link_rtrequest(int cmd, struct rtentry *rt, const struct rt_addrinfo *info) 2166 { 2167 struct ifaddr *ifa; 2168 const struct sockaddr *dst; 2169 struct ifnet *ifp; 2170 struct psref psref; 2171 2172 if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL || 2173 (ifp = ifa->ifa_ifp) == NULL || (dst = rt_getkey(rt)) == NULL) 2174 return; 2175 if ((ifa = ifaof_ifpforaddr_psref(dst, ifp, &psref)) != NULL) { 2176 rt_replace_ifa(rt, ifa); 2177 if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest) 2178 ifa->ifa_rtrequest(cmd, rt, info); 2179 ifa_release(ifa, &psref); 2180 } 2181 } 2182 2183 /* 2184 * bitmask macros to manage a densely packed link_state change queue. 2185 * Because we need to store LINK_STATE_UNKNOWN(0), LINK_STATE_DOWN(1) and 2186 * LINK_STATE_UP(2) we need 2 bits for each state change. 2187 * As a state change to store is 0, treat all bits set as an unset item. 2188 */ 2189 #define LQ_ITEM_BITS 2 2190 #define LQ_ITEM_MASK ((1 << LQ_ITEM_BITS) - 1) 2191 #define LQ_MASK(i) (LQ_ITEM_MASK << (i) * LQ_ITEM_BITS) 2192 #define LINK_STATE_UNSET LQ_ITEM_MASK 2193 #define LQ_ITEM(q, i) (((q) & LQ_MASK((i))) >> (i) * LQ_ITEM_BITS) 2194 #define LQ_STORE(q, i, v) \ 2195 do { \ 2196 (q) &= ~LQ_MASK((i)); \ 2197 (q) |= (v) << (i) * LQ_ITEM_BITS; \ 2198 } while (0 /* CONSTCOND */) 2199 #define LQ_MAX(q) ((sizeof((q)) * NBBY) / LQ_ITEM_BITS) 2200 #define LQ_POP(q, v) \ 2201 do { \ 2202 (v) = LQ_ITEM((q), 0); \ 2203 (q) >>= LQ_ITEM_BITS; \ 2204 (q) |= LINK_STATE_UNSET << (LQ_MAX((q)) - 1) * LQ_ITEM_BITS; \ 2205 } while (0 /* CONSTCOND */) 2206 #define LQ_PUSH(q, v) \ 2207 do { \ 2208 (q) >>= LQ_ITEM_BITS; \ 2209 (q) |= (v) << (LQ_MAX((q)) - 1) * LQ_ITEM_BITS; \ 2210 } while (0 /* CONSTCOND */) 2211 #define LQ_FIND_UNSET(q, i) \ 2212 for ((i) = 0; i < LQ_MAX((q)); (i)++) { \ 2213 if (LQ_ITEM((q), (i)) == LINK_STATE_UNSET) \ 2214 break; \ 2215 } 2216 2217 /* 2218 * XXX reusing (ifp)->if_snd->ifq_lock rather than having another spin mutex 2219 * for each ifnet. It doesn't matter because: 2220 * - if IFEF_MPSAFE is enabled, if_snd isn't used and lock contentions on 2221 * ifq_lock don't happen 2222 * - if IFEF_MPSAFE is disabled, there is no lock contention on ifq_lock 2223 * because if_snd, if_link_state_change and if_link_state_change_softint 2224 * are all called with KERNEL_LOCK 2225 */ 2226 #define IF_LINK_STATE_CHANGE_LOCK(ifp) \ 2227 mutex_enter((ifp)->if_snd.ifq_lock) 2228 #define IF_LINK_STATE_CHANGE_UNLOCK(ifp) \ 2229 mutex_exit((ifp)->if_snd.ifq_lock) 2230 2231 /* 2232 * Handle a change in the interface link state and 2233 * queue notifications. 2234 */ 2235 void 2236 if_link_state_change(struct ifnet *ifp, int link_state) 2237 { 2238 int idx; 2239 2240 KASSERTMSG(if_is_link_state_changeable(ifp), 2241 "%s: IFEF_NO_LINK_STATE_CHANGE must not be set, but if_extflags=0x%x", 2242 ifp->if_xname, ifp->if_extflags); 2243 2244 /* Ensure change is to a valid state */ 2245 switch (link_state) { 2246 case LINK_STATE_UNKNOWN: /* FALLTHROUGH */ 2247 case LINK_STATE_DOWN: /* FALLTHROUGH */ 2248 case LINK_STATE_UP: 2249 break; 2250 default: 2251 #ifdef DEBUG 2252 printf("%s: invalid link state %d\n", 2253 ifp->if_xname, link_state); 2254 #endif 2255 return; 2256 } 2257 2258 IF_LINK_STATE_CHANGE_LOCK(ifp); 2259 2260 /* Find the last unset event in the queue. */ 2261 LQ_FIND_UNSET(ifp->if_link_queue, idx); 2262 2263 /* 2264 * Ensure link_state doesn't match the last event in the queue. 2265 * ifp->if_link_state is not checked and set here because 2266 * that would present an inconsistent picture to the system. 2267 */ 2268 if (idx != 0 && 2269 LQ_ITEM(ifp->if_link_queue, idx - 1) == (uint8_t)link_state) 2270 goto out; 2271 2272 /* Handle queue overflow. */ 2273 if (idx == LQ_MAX(ifp->if_link_queue)) { 2274 uint8_t lost; 2275 2276 /* 2277 * The DOWN state must be protected from being pushed off 2278 * the queue to ensure that userland will always be 2279 * in a sane state. 2280 * Because DOWN is protected, there is no need to protect 2281 * UNKNOWN. 2282 * It should be invalid to change from any other state to 2283 * UNKNOWN anyway ... 2284 */ 2285 lost = LQ_ITEM(ifp->if_link_queue, 0); 2286 LQ_PUSH(ifp->if_link_queue, (uint8_t)link_state); 2287 if (lost == LINK_STATE_DOWN) { 2288 lost = LQ_ITEM(ifp->if_link_queue, 0); 2289 LQ_STORE(ifp->if_link_queue, 0, LINK_STATE_DOWN); 2290 } 2291 printf("%s: lost link state change %s\n", 2292 ifp->if_xname, 2293 lost == LINK_STATE_UP ? "UP" : 2294 lost == LINK_STATE_DOWN ? "DOWN" : 2295 "UNKNOWN"); 2296 } else 2297 LQ_STORE(ifp->if_link_queue, idx, (uint8_t)link_state); 2298 2299 softint_schedule(ifp->if_link_si); 2300 2301 out: 2302 IF_LINK_STATE_CHANGE_UNLOCK(ifp); 2303 } 2304 2305 /* 2306 * Handle interface link state change notifications. 2307 */ 2308 void 2309 if_link_state_change_softint(struct ifnet *ifp, int link_state) 2310 { 2311 struct domain *dp; 2312 int s = splnet(); 2313 bool notify; 2314 2315 KASSERT(!cpu_intr_p()); 2316 2317 IF_LINK_STATE_CHANGE_LOCK(ifp); 2318 2319 /* Ensure the change is still valid. */ 2320 if (ifp->if_link_state == link_state) { 2321 IF_LINK_STATE_CHANGE_UNLOCK(ifp); 2322 return; 2323 } 2324 2325 #ifdef DEBUG 2326 log(LOG_DEBUG, "%s: link state %s (was %s)\n", ifp->if_xname, 2327 link_state == LINK_STATE_UP ? "UP" : 2328 link_state == LINK_STATE_DOWN ? "DOWN" : 2329 "UNKNOWN", 2330 ifp->if_link_state == LINK_STATE_UP ? "UP" : 2331 ifp->if_link_state == LINK_STATE_DOWN ? "DOWN" : 2332 "UNKNOWN"); 2333 #endif 2334 2335 /* 2336 * When going from UNKNOWN to UP, we need to mark existing 2337 * addresses as tentative and restart DAD as we may have 2338 * erroneously not found a duplicate. 2339 * 2340 * This needs to happen before rt_ifmsg to avoid a race where 2341 * listeners would have an address and expect it to work right 2342 * away. 2343 */ 2344 notify = (link_state == LINK_STATE_UP && 2345 ifp->if_link_state == LINK_STATE_UNKNOWN); 2346 ifp->if_link_state = link_state; 2347 /* The following routines may sleep so release the spin mutex */ 2348 IF_LINK_STATE_CHANGE_UNLOCK(ifp); 2349 2350 KERNEL_LOCK_UNLESS_NET_MPSAFE(); 2351 if (notify) { 2352 DOMAIN_FOREACH(dp) { 2353 if (dp->dom_if_link_state_change != NULL) 2354 dp->dom_if_link_state_change(ifp, 2355 LINK_STATE_DOWN); 2356 } 2357 } 2358 2359 /* Notify that the link state has changed. */ 2360 rt_ifmsg(ifp); 2361 2362 #if NCARP > 0 2363 if (ifp->if_carp) 2364 carp_carpdev_state(ifp); 2365 #endif 2366 2367 DOMAIN_FOREACH(dp) { 2368 if (dp->dom_if_link_state_change != NULL) 2369 dp->dom_if_link_state_change(ifp, link_state); 2370 } 2371 KERNEL_UNLOCK_UNLESS_NET_MPSAFE(); 2372 splx(s); 2373 } 2374 2375 /* 2376 * Process the interface link state change queue. 2377 */ 2378 static void 2379 if_link_state_change_si(void *arg) 2380 { 2381 struct ifnet *ifp = arg; 2382 int s; 2383 uint8_t state; 2384 bool schedule; 2385 2386 SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE(); 2387 s = splnet(); 2388 2389 /* Pop a link state change from the queue and process it. */ 2390 IF_LINK_STATE_CHANGE_LOCK(ifp); 2391 LQ_POP(ifp->if_link_queue, state); 2392 IF_LINK_STATE_CHANGE_UNLOCK(ifp); 2393 2394 if_link_state_change_softint(ifp, state); 2395 2396 /* If there is a link state change to come, schedule it. */ 2397 IF_LINK_STATE_CHANGE_LOCK(ifp); 2398 schedule = (LQ_ITEM(ifp->if_link_queue, 0) != LINK_STATE_UNSET); 2399 IF_LINK_STATE_CHANGE_UNLOCK(ifp); 2400 if (schedule) 2401 softint_schedule(ifp->if_link_si); 2402 2403 splx(s); 2404 SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE(); 2405 } 2406 2407 /* 2408 * Default action when installing a local route on a point-to-point 2409 * interface. 2410 */ 2411 void 2412 p2p_rtrequest(int req, struct rtentry *rt, 2413 __unused const struct rt_addrinfo *info) 2414 { 2415 struct ifnet *ifp = rt->rt_ifp; 2416 struct ifaddr *ifa, *lo0ifa; 2417 int s = pserialize_read_enter(); 2418 2419 switch (req) { 2420 case RTM_ADD: 2421 if ((rt->rt_flags & RTF_LOCAL) == 0) 2422 break; 2423 2424 rt->rt_ifp = lo0ifp; 2425 2426 IFADDR_READER_FOREACH(ifa, ifp) { 2427 if (equal(rt_getkey(rt), ifa->ifa_addr)) 2428 break; 2429 } 2430 if (ifa == NULL) 2431 break; 2432 2433 /* 2434 * Ensure lo0 has an address of the same family. 2435 */ 2436 IFADDR_READER_FOREACH(lo0ifa, lo0ifp) { 2437 if (lo0ifa->ifa_addr->sa_family == 2438 ifa->ifa_addr->sa_family) 2439 break; 2440 } 2441 if (lo0ifa == NULL) 2442 break; 2443 2444 /* 2445 * Make sure to set rt->rt_ifa to the interface 2446 * address we are using, otherwise we will have trouble 2447 * with source address selection. 2448 */ 2449 if (ifa != rt->rt_ifa) 2450 rt_replace_ifa(rt, ifa); 2451 break; 2452 case RTM_DELETE: 2453 default: 2454 break; 2455 } 2456 pserialize_read_exit(s); 2457 } 2458 2459 static void 2460 _if_down(struct ifnet *ifp) 2461 { 2462 struct ifaddr *ifa; 2463 struct domain *dp; 2464 int s, bound; 2465 struct psref psref; 2466 2467 ifp->if_flags &= ~IFF_UP; 2468 nanotime(&ifp->if_lastchange); 2469 2470 bound = curlwp_bind(); 2471 s = pserialize_read_enter(); 2472 IFADDR_READER_FOREACH(ifa, ifp) { 2473 ifa_acquire(ifa, &psref); 2474 pserialize_read_exit(s); 2475 2476 pfctlinput(PRC_IFDOWN, ifa->ifa_addr); 2477 2478 s = pserialize_read_enter(); 2479 ifa_release(ifa, &psref); 2480 } 2481 pserialize_read_exit(s); 2482 curlwp_bindx(bound); 2483 2484 IFQ_PURGE(&ifp->if_snd); 2485 #if NCARP > 0 2486 if (ifp->if_carp) 2487 carp_carpdev_state(ifp); 2488 #endif 2489 rt_ifmsg(ifp); 2490 DOMAIN_FOREACH(dp) { 2491 if (dp->dom_if_down) 2492 dp->dom_if_down(ifp); 2493 } 2494 } 2495 2496 static void 2497 if_down_deactivated(struct ifnet *ifp) 2498 { 2499 2500 KASSERT(if_is_deactivated(ifp)); 2501 _if_down(ifp); 2502 } 2503 2504 void 2505 if_down_locked(struct ifnet *ifp) 2506 { 2507 2508 KASSERT(IFNET_LOCKED(ifp)); 2509 _if_down(ifp); 2510 } 2511 2512 /* 2513 * Mark an interface down and notify protocols of 2514 * the transition. 2515 * NOTE: must be called at splsoftnet or equivalent. 2516 */ 2517 void 2518 if_down(struct ifnet *ifp) 2519 { 2520 2521 IFNET_LOCK(ifp); 2522 if_down_locked(ifp); 2523 IFNET_UNLOCK(ifp); 2524 } 2525 2526 /* 2527 * Must be called with holding if_ioctl_lock. 2528 */ 2529 static void 2530 if_up_locked(struct ifnet *ifp) 2531 { 2532 #ifdef notyet 2533 struct ifaddr *ifa; 2534 #endif 2535 struct domain *dp; 2536 2537 KASSERT(IFNET_LOCKED(ifp)); 2538 2539 KASSERT(!if_is_deactivated(ifp)); 2540 ifp->if_flags |= IFF_UP; 2541 nanotime(&ifp->if_lastchange); 2542 #ifdef notyet 2543 /* this has no effect on IP, and will kill all ISO connections XXX */ 2544 IFADDR_READER_FOREACH(ifa, ifp) 2545 pfctlinput(PRC_IFUP, ifa->ifa_addr); 2546 #endif 2547 #if NCARP > 0 2548 if (ifp->if_carp) 2549 carp_carpdev_state(ifp); 2550 #endif 2551 rt_ifmsg(ifp); 2552 DOMAIN_FOREACH(dp) { 2553 if (dp->dom_if_up) 2554 dp->dom_if_up(ifp); 2555 } 2556 } 2557 2558 /* 2559 * Handle interface slowtimo timer routine. Called 2560 * from softclock, we decrement timer (if set) and 2561 * call the appropriate interface routine on expiration. 2562 */ 2563 static void 2564 if_slowtimo(void *arg) 2565 { 2566 void (*slowtimo)(struct ifnet *); 2567 struct ifnet *ifp = arg; 2568 int s; 2569 2570 slowtimo = ifp->if_slowtimo; 2571 if (__predict_false(slowtimo == NULL)) 2572 return; 2573 2574 s = splnet(); 2575 if (ifp->if_timer != 0 && --ifp->if_timer == 0) 2576 (*slowtimo)(ifp); 2577 2578 splx(s); 2579 2580 if (__predict_true(ifp->if_slowtimo != NULL)) 2581 callout_schedule(ifp->if_slowtimo_ch, hz / IFNET_SLOWHZ); 2582 } 2583 2584 /* 2585 * Mark an interface up and notify protocols of 2586 * the transition. 2587 * NOTE: must be called at splsoftnet or equivalent. 2588 */ 2589 void 2590 if_up(struct ifnet *ifp) 2591 { 2592 2593 IFNET_LOCK(ifp); 2594 if_up_locked(ifp); 2595 IFNET_UNLOCK(ifp); 2596 } 2597 2598 /* 2599 * Set/clear promiscuous mode on interface ifp based on the truth value 2600 * of pswitch. The calls are reference counted so that only the first 2601 * "on" request actually has an effect, as does the final "off" request. 2602 * Results are undefined if the "off" and "on" requests are not matched. 2603 */ 2604 int 2605 ifpromisc_locked(struct ifnet *ifp, int pswitch) 2606 { 2607 int pcount, ret = 0; 2608 short nflags; 2609 2610 KASSERT(IFNET_LOCKED(ifp)); 2611 2612 pcount = ifp->if_pcount; 2613 if (pswitch) { 2614 /* 2615 * Allow the device to be "placed" into promiscuous 2616 * mode even if it is not configured up. It will 2617 * consult IFF_PROMISC when it is brought up. 2618 */ 2619 if (ifp->if_pcount++ != 0) 2620 goto out; 2621 nflags = ifp->if_flags | IFF_PROMISC; 2622 } else { 2623 if (--ifp->if_pcount > 0) 2624 goto out; 2625 nflags = ifp->if_flags & ~IFF_PROMISC; 2626 } 2627 ret = if_flags_set(ifp, nflags); 2628 /* Restore interface state if not successful. */ 2629 if (ret != 0) { 2630 ifp->if_pcount = pcount; 2631 } 2632 out: 2633 return ret; 2634 } 2635 2636 int 2637 ifpromisc(struct ifnet *ifp, int pswitch) 2638 { 2639 int e; 2640 2641 IFNET_LOCK(ifp); 2642 e = ifpromisc_locked(ifp, pswitch); 2643 IFNET_UNLOCK(ifp); 2644 2645 return e; 2646 } 2647 2648 /* 2649 * Map interface name to 2650 * interface structure pointer. 2651 */ 2652 struct ifnet * 2653 ifunit(const char *name) 2654 { 2655 struct ifnet *ifp; 2656 const char *cp = name; 2657 u_int unit = 0; 2658 u_int i; 2659 int s; 2660 2661 /* 2662 * If the entire name is a number, treat it as an ifindex. 2663 */ 2664 for (i = 0; i < IFNAMSIZ && *cp >= '0' && *cp <= '9'; i++, cp++) { 2665 unit = unit * 10 + (*cp - '0'); 2666 } 2667 2668 /* 2669 * If the number took all of the name, then it's a valid ifindex. 2670 */ 2671 if (i == IFNAMSIZ || (cp != name && *cp == '\0')) 2672 return if_byindex(unit); 2673 2674 ifp = NULL; 2675 s = pserialize_read_enter(); 2676 IFNET_READER_FOREACH(ifp) { 2677 if (if_is_deactivated(ifp)) 2678 continue; 2679 if (strcmp(ifp->if_xname, name) == 0) 2680 goto out; 2681 } 2682 out: 2683 pserialize_read_exit(s); 2684 return ifp; 2685 } 2686 2687 /* 2688 * Get a reference of an ifnet object by an interface name. 2689 * The returned reference is protected by psref(9). The caller 2690 * must release a returned reference by if_put after use. 2691 */ 2692 struct ifnet * 2693 if_get(const char *name, struct psref *psref) 2694 { 2695 struct ifnet *ifp; 2696 const char *cp = name; 2697 u_int unit = 0; 2698 u_int i; 2699 int s; 2700 2701 /* 2702 * If the entire name is a number, treat it as an ifindex. 2703 */ 2704 for (i = 0; i < IFNAMSIZ && *cp >= '0' && *cp <= '9'; i++, cp++) { 2705 unit = unit * 10 + (*cp - '0'); 2706 } 2707 2708 /* 2709 * If the number took all of the name, then it's a valid ifindex. 2710 */ 2711 if (i == IFNAMSIZ || (cp != name && *cp == '\0')) 2712 return if_get_byindex(unit, psref); 2713 2714 ifp = NULL; 2715 s = pserialize_read_enter(); 2716 IFNET_READER_FOREACH(ifp) { 2717 if (if_is_deactivated(ifp)) 2718 continue; 2719 if (strcmp(ifp->if_xname, name) == 0) { 2720 psref_acquire(psref, &ifp->if_psref, 2721 ifnet_psref_class); 2722 goto out; 2723 } 2724 } 2725 out: 2726 pserialize_read_exit(s); 2727 return ifp; 2728 } 2729 2730 /* 2731 * Release a reference of an ifnet object given by if_get, if_get_byindex 2732 * or if_get_bylla. 2733 */ 2734 void 2735 if_put(const struct ifnet *ifp, struct psref *psref) 2736 { 2737 2738 if (ifp == NULL) 2739 return; 2740 2741 psref_release(psref, &ifp->if_psref, ifnet_psref_class); 2742 } 2743 2744 /* 2745 * Return ifp having idx. Return NULL if not found. Normally if_byindex 2746 * should be used. 2747 */ 2748 ifnet_t * 2749 _if_byindex(u_int idx) 2750 { 2751 2752 return (__predict_true(idx < if_indexlim)) ? ifindex2ifnet[idx] : NULL; 2753 } 2754 2755 /* 2756 * Return ifp having idx. Return NULL if not found or the found ifp is 2757 * already deactivated. 2758 */ 2759 ifnet_t * 2760 if_byindex(u_int idx) 2761 { 2762 ifnet_t *ifp; 2763 2764 ifp = _if_byindex(idx); 2765 if (ifp != NULL && if_is_deactivated(ifp)) 2766 ifp = NULL; 2767 return ifp; 2768 } 2769 2770 /* 2771 * Get a reference of an ifnet object by an interface index. 2772 * The returned reference is protected by psref(9). The caller 2773 * must release a returned reference by if_put after use. 2774 */ 2775 ifnet_t * 2776 if_get_byindex(u_int idx, struct psref *psref) 2777 { 2778 ifnet_t *ifp; 2779 int s; 2780 2781 s = pserialize_read_enter(); 2782 ifp = if_byindex(idx); 2783 if (__predict_true(ifp != NULL)) 2784 psref_acquire(psref, &ifp->if_psref, ifnet_psref_class); 2785 pserialize_read_exit(s); 2786 2787 return ifp; 2788 } 2789 2790 ifnet_t * 2791 if_get_bylla(const void *lla, unsigned char lla_len, struct psref *psref) 2792 { 2793 ifnet_t *ifp; 2794 int s; 2795 2796 s = pserialize_read_enter(); 2797 IFNET_READER_FOREACH(ifp) { 2798 if (if_is_deactivated(ifp)) 2799 continue; 2800 if (ifp->if_addrlen != lla_len) 2801 continue; 2802 if (memcmp(lla, CLLADDR(ifp->if_sadl), lla_len) == 0) { 2803 psref_acquire(psref, &ifp->if_psref, 2804 ifnet_psref_class); 2805 break; 2806 } 2807 } 2808 pserialize_read_exit(s); 2809 2810 return ifp; 2811 } 2812 2813 /* 2814 * Note that it's safe only if the passed ifp is guaranteed to not be freed, 2815 * for example using pserialize or the ifp is already held or some other 2816 * object is held which guarantes the ifp to not be freed indirectly. 2817 */ 2818 void 2819 if_acquire(struct ifnet *ifp, struct psref *psref) 2820 { 2821 2822 KASSERT(ifp->if_index != 0); 2823 psref_acquire(psref, &ifp->if_psref, ifnet_psref_class); 2824 } 2825 2826 bool 2827 if_held(struct ifnet *ifp) 2828 { 2829 2830 return psref_held(&ifp->if_psref, ifnet_psref_class); 2831 } 2832 2833 /* 2834 * Some tunnel interfaces can nest, e.g. IPv4 over IPv4 gif(4) tunnel over IPv4. 2835 * Check the tunnel nesting count. 2836 * Return > 0, if tunnel nesting count is more than limit. 2837 * Return 0, if tunnel nesting count is equal or less than limit. 2838 */ 2839 int 2840 if_tunnel_check_nesting(struct ifnet *ifp, struct mbuf *m, int limit) 2841 { 2842 struct m_tag *mtag; 2843 int *count; 2844 2845 mtag = m_tag_find(m, PACKET_TAG_TUNNEL_INFO, NULL); 2846 if (mtag != NULL) { 2847 count = (int *)(mtag + 1); 2848 if (++(*count) > limit) { 2849 log(LOG_NOTICE, 2850 "%s: recursively called too many times(%d)\n", 2851 ifp->if_xname, *count); 2852 return EIO; 2853 } 2854 } else { 2855 mtag = m_tag_get(PACKET_TAG_TUNNEL_INFO, sizeof(*count), 2856 M_NOWAIT); 2857 if (mtag != NULL) { 2858 m_tag_prepend(m, mtag); 2859 count = (int *)(mtag + 1); 2860 *count = 0; 2861 } else { 2862 log(LOG_DEBUG, 2863 "%s: m_tag_get() failed, recursion calls are not prevented.\n", 2864 ifp->if_xname); 2865 } 2866 } 2867 2868 return 0; 2869 } 2870 2871 /* common */ 2872 int 2873 ifioctl_common(struct ifnet *ifp, u_long cmd, void *data) 2874 { 2875 int s; 2876 struct ifreq *ifr; 2877 struct ifcapreq *ifcr; 2878 struct ifdatareq *ifdr; 2879 2880 switch (cmd) { 2881 case SIOCSIFCAP: 2882 ifcr = data; 2883 if ((ifcr->ifcr_capenable & ~ifp->if_capabilities) != 0) 2884 return EINVAL; 2885 2886 if (ifcr->ifcr_capenable == ifp->if_capenable) 2887 return 0; 2888 2889 ifp->if_capenable = ifcr->ifcr_capenable; 2890 2891 /* Pre-compute the checksum flags mask. */ 2892 ifp->if_csum_flags_tx = 0; 2893 ifp->if_csum_flags_rx = 0; 2894 if (ifp->if_capenable & IFCAP_CSUM_IPv4_Tx) { 2895 ifp->if_csum_flags_tx |= M_CSUM_IPv4; 2896 } 2897 if (ifp->if_capenable & IFCAP_CSUM_IPv4_Rx) { 2898 ifp->if_csum_flags_rx |= M_CSUM_IPv4; 2899 } 2900 2901 if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Tx) { 2902 ifp->if_csum_flags_tx |= M_CSUM_TCPv4; 2903 } 2904 if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Rx) { 2905 ifp->if_csum_flags_rx |= M_CSUM_TCPv4; 2906 } 2907 2908 if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Tx) { 2909 ifp->if_csum_flags_tx |= M_CSUM_UDPv4; 2910 } 2911 if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Rx) { 2912 ifp->if_csum_flags_rx |= M_CSUM_UDPv4; 2913 } 2914 2915 if (ifp->if_capenable & IFCAP_CSUM_TCPv6_Tx) { 2916 ifp->if_csum_flags_tx |= M_CSUM_TCPv6; 2917 } 2918 if (ifp->if_capenable & IFCAP_CSUM_TCPv6_Rx) { 2919 ifp->if_csum_flags_rx |= M_CSUM_TCPv6; 2920 } 2921 2922 if (ifp->if_capenable & IFCAP_CSUM_UDPv6_Tx) { 2923 ifp->if_csum_flags_tx |= M_CSUM_UDPv6; 2924 } 2925 if (ifp->if_capenable & IFCAP_CSUM_UDPv6_Rx) { 2926 ifp->if_csum_flags_rx |= M_CSUM_UDPv6; 2927 } 2928 if (ifp->if_flags & IFF_UP) 2929 return ENETRESET; 2930 return 0; 2931 case SIOCSIFFLAGS: 2932 ifr = data; 2933 /* 2934 * If if_is_mpsafe(ifp), KERNEL_LOCK isn't held here, but if_up 2935 * and if_down aren't MP-safe yet, so we must hold the lock. 2936 */ 2937 KERNEL_LOCK_IF_IFP_MPSAFE(ifp); 2938 if (ifp->if_flags & IFF_UP && (ifr->ifr_flags & IFF_UP) == 0) { 2939 s = splsoftnet(); 2940 if_down_locked(ifp); 2941 splx(s); 2942 } 2943 if (ifr->ifr_flags & IFF_UP && (ifp->if_flags & IFF_UP) == 0) { 2944 s = splsoftnet(); 2945 if_up_locked(ifp); 2946 splx(s); 2947 } 2948 KERNEL_UNLOCK_IF_IFP_MPSAFE(ifp); 2949 ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) | 2950 (ifr->ifr_flags &~ IFF_CANTCHANGE); 2951 break; 2952 case SIOCGIFFLAGS: 2953 ifr = data; 2954 ifr->ifr_flags = ifp->if_flags; 2955 break; 2956 2957 case SIOCGIFMETRIC: 2958 ifr = data; 2959 ifr->ifr_metric = ifp->if_metric; 2960 break; 2961 2962 case SIOCGIFMTU: 2963 ifr = data; 2964 ifr->ifr_mtu = ifp->if_mtu; 2965 break; 2966 2967 case SIOCGIFDLT: 2968 ifr = data; 2969 ifr->ifr_dlt = ifp->if_dlt; 2970 break; 2971 2972 case SIOCGIFCAP: 2973 ifcr = data; 2974 ifcr->ifcr_capabilities = ifp->if_capabilities; 2975 ifcr->ifcr_capenable = ifp->if_capenable; 2976 break; 2977 2978 case SIOCSIFMETRIC: 2979 ifr = data; 2980 ifp->if_metric = ifr->ifr_metric; 2981 break; 2982 2983 case SIOCGIFDATA: 2984 ifdr = data; 2985 ifdr->ifdr_data = ifp->if_data; 2986 break; 2987 2988 case SIOCGIFINDEX: 2989 ifr = data; 2990 ifr->ifr_index = ifp->if_index; 2991 break; 2992 2993 case SIOCZIFDATA: 2994 ifdr = data; 2995 ifdr->ifdr_data = ifp->if_data; 2996 /* 2997 * Assumes that the volatile counters that can be 2998 * zero'ed are at the end of if_data. 2999 */ 3000 memset(&ifp->if_data.ifi_ipackets, 0, sizeof(ifp->if_data) - 3001 offsetof(struct if_data, ifi_ipackets)); 3002 /* 3003 * The memset() clears to the bottm of if_data. In the area, 3004 * if_lastchange is included. Please be careful if new entry 3005 * will be added into if_data or rewite this. 3006 * 3007 * And also, update if_lastchnage. 3008 */ 3009 getnanotime(&ifp->if_lastchange); 3010 break; 3011 case SIOCSIFMTU: 3012 ifr = data; 3013 if (ifp->if_mtu == ifr->ifr_mtu) 3014 break; 3015 ifp->if_mtu = ifr->ifr_mtu; 3016 /* 3017 * If the link MTU changed, do network layer specific procedure. 3018 */ 3019 #ifdef INET6 3020 KERNEL_LOCK_UNLESS_NET_MPSAFE(); 3021 if (in6_present) 3022 nd6_setmtu(ifp); 3023 KERNEL_UNLOCK_UNLESS_NET_MPSAFE(); 3024 #endif 3025 return ENETRESET; 3026 default: 3027 return ENOTTY; 3028 } 3029 return 0; 3030 } 3031 3032 int 3033 ifaddrpref_ioctl(struct socket *so, u_long cmd, void *data, struct ifnet *ifp) 3034 { 3035 struct if_addrprefreq *ifap = (struct if_addrprefreq *)data; 3036 struct ifaddr *ifa; 3037 const struct sockaddr *any, *sa; 3038 union { 3039 struct sockaddr sa; 3040 struct sockaddr_storage ss; 3041 } u, v; 3042 int s, error = 0; 3043 3044 switch (cmd) { 3045 case SIOCSIFADDRPREF: 3046 if (kauth_authorize_network(curlwp->l_cred, KAUTH_NETWORK_INTERFACE, 3047 KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd, 3048 NULL) != 0) 3049 return EPERM; 3050 case SIOCGIFADDRPREF: 3051 break; 3052 default: 3053 return EOPNOTSUPP; 3054 } 3055 3056 /* sanity checks */ 3057 if (data == NULL || ifp == NULL) { 3058 panic("invalid argument to %s", __func__); 3059 /*NOTREACHED*/ 3060 } 3061 3062 /* address must be specified on ADD and DELETE */ 3063 sa = sstocsa(&ifap->ifap_addr); 3064 if (sa->sa_family != sofamily(so)) 3065 return EINVAL; 3066 if ((any = sockaddr_any(sa)) == NULL || sa->sa_len != any->sa_len) 3067 return EINVAL; 3068 3069 sockaddr_externalize(&v.sa, sizeof(v.ss), sa); 3070 3071 s = pserialize_read_enter(); 3072 IFADDR_READER_FOREACH(ifa, ifp) { 3073 if (ifa->ifa_addr->sa_family != sa->sa_family) 3074 continue; 3075 sockaddr_externalize(&u.sa, sizeof(u.ss), ifa->ifa_addr); 3076 if (sockaddr_cmp(&u.sa, &v.sa) == 0) 3077 break; 3078 } 3079 if (ifa == NULL) { 3080 error = EADDRNOTAVAIL; 3081 goto out; 3082 } 3083 3084 switch (cmd) { 3085 case SIOCSIFADDRPREF: 3086 ifa->ifa_preference = ifap->ifap_preference; 3087 goto out; 3088 case SIOCGIFADDRPREF: 3089 /* fill in the if_laddrreq structure */ 3090 (void)sockaddr_copy(sstosa(&ifap->ifap_addr), 3091 sizeof(ifap->ifap_addr), ifa->ifa_addr); 3092 ifap->ifap_preference = ifa->ifa_preference; 3093 goto out; 3094 default: 3095 error = EOPNOTSUPP; 3096 } 3097 out: 3098 pserialize_read_exit(s); 3099 return error; 3100 } 3101 3102 /* 3103 * Interface ioctls. 3104 */ 3105 static int 3106 doifioctl(struct socket *so, u_long cmd, void *data, struct lwp *l) 3107 { 3108 struct ifnet *ifp; 3109 struct ifreq *ifr; 3110 int error = 0; 3111 #if defined(COMPAT_OSOCK) || defined(COMPAT_OIFREQ) 3112 u_long ocmd = cmd; 3113 #endif 3114 short oif_flags; 3115 #ifdef COMPAT_OIFREQ 3116 struct ifreq ifrb; 3117 struct oifreq *oifr = NULL; 3118 #endif 3119 int r; 3120 struct psref psref; 3121 int bound; 3122 3123 switch (cmd) { 3124 case SIOCGIFCONF: 3125 return ifconf(cmd, data); 3126 case SIOCINITIFADDR: 3127 return EPERM; 3128 default: 3129 error = (*vec_compat_ifconf)(l, cmd, data); 3130 if (error != ENOSYS) 3131 return error; 3132 error = (*vec_compat_ifdatareq)(l, cmd, data); 3133 if (error != ENOSYS) 3134 return error; 3135 break; 3136 } 3137 3138 ifr = data; 3139 #ifdef COMPAT_OIFREQ 3140 if (vec_compat_cvtcmd) { 3141 cmd = (*vec_compat_cvtcmd)(cmd); 3142 if (cmd != ocmd) { 3143 oifr = data; 3144 data = ifr = &ifrb; 3145 ifreqo2n(oifr, ifr); 3146 } 3147 } 3148 #endif 3149 3150 switch (cmd) { 3151 case SIOCIFCREATE: 3152 case SIOCIFDESTROY: 3153 bound = curlwp_bind(); 3154 if (l != NULL) { 3155 ifp = if_get(ifr->ifr_name, &psref); 3156 error = kauth_authorize_network(l->l_cred, 3157 KAUTH_NETWORK_INTERFACE, 3158 KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, 3159 (void *)cmd, NULL); 3160 if (ifp != NULL) 3161 if_put(ifp, &psref); 3162 if (error != 0) { 3163 curlwp_bindx(bound); 3164 return error; 3165 } 3166 } 3167 KERNEL_LOCK_UNLESS_NET_MPSAFE(); 3168 mutex_enter(&if_clone_mtx); 3169 r = (cmd == SIOCIFCREATE) ? 3170 if_clone_create(ifr->ifr_name) : 3171 if_clone_destroy(ifr->ifr_name); 3172 mutex_exit(&if_clone_mtx); 3173 KERNEL_UNLOCK_UNLESS_NET_MPSAFE(); 3174 curlwp_bindx(bound); 3175 return r; 3176 3177 case SIOCIFGCLONERS: 3178 { 3179 struct if_clonereq *req = (struct if_clonereq *)data; 3180 return if_clone_list(req->ifcr_count, req->ifcr_buffer, 3181 &req->ifcr_total); 3182 } 3183 } 3184 3185 bound = curlwp_bind(); 3186 ifp = if_get(ifr->ifr_name, &psref); 3187 if (ifp == NULL) { 3188 curlwp_bindx(bound); 3189 return ENXIO; 3190 } 3191 3192 switch (cmd) { 3193 case SIOCALIFADDR: 3194 case SIOCDLIFADDR: 3195 case SIOCSIFADDRPREF: 3196 case SIOCSIFFLAGS: 3197 case SIOCSIFCAP: 3198 case SIOCSIFMETRIC: 3199 case SIOCZIFDATA: 3200 case SIOCSIFMTU: 3201 case SIOCSIFPHYADDR: 3202 case SIOCDIFPHYADDR: 3203 #ifdef INET6 3204 case SIOCSIFPHYADDR_IN6: 3205 #endif 3206 case SIOCSLIFPHYADDR: 3207 case SIOCADDMULTI: 3208 case SIOCDELMULTI: 3209 case SIOCSIFMEDIA: 3210 case SIOCSDRVSPEC: 3211 case SIOCG80211: 3212 case SIOCS80211: 3213 case SIOCS80211NWID: 3214 case SIOCS80211NWKEY: 3215 case SIOCS80211POWER: 3216 case SIOCS80211BSSID: 3217 case SIOCS80211CHANNEL: 3218 case SIOCSLINKSTR: 3219 if (l != NULL) { 3220 error = kauth_authorize_network(l->l_cred, 3221 KAUTH_NETWORK_INTERFACE, 3222 KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, 3223 (void *)cmd, NULL); 3224 if (error != 0) 3225 goto out; 3226 } 3227 } 3228 3229 oif_flags = ifp->if_flags; 3230 3231 KERNEL_LOCK_UNLESS_IFP_MPSAFE(ifp); 3232 IFNET_LOCK(ifp); 3233 3234 error = (*ifp->if_ioctl)(ifp, cmd, data); 3235 if (error != ENOTTY) 3236 ; 3237 else if (so->so_proto == NULL) 3238 error = EOPNOTSUPP; 3239 else { 3240 KERNEL_LOCK_IF_IFP_MPSAFE(ifp); 3241 #ifdef COMPAT_OSOCK 3242 if (vec_compat_ifioctl != NULL) 3243 error = (*vec_compat_ifioctl)(so, ocmd, cmd, data, l); 3244 else 3245 #endif 3246 error = (*so->so_proto->pr_usrreqs->pr_ioctl)(so, 3247 cmd, data, ifp); 3248 KERNEL_UNLOCK_IF_IFP_MPSAFE(ifp); 3249 } 3250 3251 if (((oif_flags ^ ifp->if_flags) & IFF_UP) != 0) { 3252 if ((ifp->if_flags & IFF_UP) != 0) { 3253 int s = splsoftnet(); 3254 if_up_locked(ifp); 3255 splx(s); 3256 } 3257 } 3258 #ifdef COMPAT_OIFREQ 3259 if (cmd != ocmd) 3260 ifreqn2o(oifr, ifr); 3261 #endif 3262 3263 IFNET_UNLOCK(ifp); 3264 KERNEL_UNLOCK_UNLESS_IFP_MPSAFE(ifp); 3265 out: 3266 if_put(ifp, &psref); 3267 curlwp_bindx(bound); 3268 return error; 3269 } 3270 3271 /* 3272 * Return interface configuration 3273 * of system. List may be used 3274 * in later ioctl's (above) to get 3275 * other information. 3276 * 3277 * Each record is a struct ifreq. Before the addition of 3278 * sockaddr_storage, the API rule was that sockaddr flavors that did 3279 * not fit would extend beyond the struct ifreq, with the next struct 3280 * ifreq starting sa_len beyond the struct sockaddr. Because the 3281 * union in struct ifreq includes struct sockaddr_storage, every kind 3282 * of sockaddr must fit. Thus, there are no longer any overlength 3283 * records. 3284 * 3285 * Records are added to the user buffer if they fit, and ifc_len is 3286 * adjusted to the length that was written. Thus, the user is only 3287 * assured of getting the complete list if ifc_len on return is at 3288 * least sizeof(struct ifreq) less than it was on entry. 3289 * 3290 * If the user buffer pointer is NULL, this routine copies no data and 3291 * returns the amount of space that would be needed. 3292 * 3293 * Invariants: 3294 * ifrp points to the next part of the user's buffer to be used. If 3295 * ifrp != NULL, space holds the number of bytes remaining that we may 3296 * write at ifrp. Otherwise, space holds the number of bytes that 3297 * would have been written had there been adequate space. 3298 */ 3299 /*ARGSUSED*/ 3300 static int 3301 ifconf(u_long cmd, void *data) 3302 { 3303 struct ifconf *ifc = (struct ifconf *)data; 3304 struct ifnet *ifp; 3305 struct ifaddr *ifa; 3306 struct ifreq ifr, *ifrp = NULL; 3307 int space = 0, error = 0; 3308 const int sz = (int)sizeof(struct ifreq); 3309 const bool docopy = ifc->ifc_req != NULL; 3310 int s; 3311 int bound; 3312 struct psref psref; 3313 3314 if (docopy) { 3315 space = ifc->ifc_len; 3316 ifrp = ifc->ifc_req; 3317 } 3318 3319 bound = curlwp_bind(); 3320 s = pserialize_read_enter(); 3321 IFNET_READER_FOREACH(ifp) { 3322 psref_acquire(&psref, &ifp->if_psref, ifnet_psref_class); 3323 pserialize_read_exit(s); 3324 3325 (void)strncpy(ifr.ifr_name, ifp->if_xname, 3326 sizeof(ifr.ifr_name)); 3327 if (ifr.ifr_name[sizeof(ifr.ifr_name) - 1] != '\0') { 3328 error = ENAMETOOLONG; 3329 goto release_exit; 3330 } 3331 if (IFADDR_READER_EMPTY(ifp)) { 3332 /* Interface with no addresses - send zero sockaddr. */ 3333 memset(&ifr.ifr_addr, 0, sizeof(ifr.ifr_addr)); 3334 if (!docopy) { 3335 space += sz; 3336 goto next; 3337 } 3338 if (space >= sz) { 3339 error = copyout(&ifr, ifrp, sz); 3340 if (error != 0) 3341 goto release_exit; 3342 ifrp++; 3343 space -= sz; 3344 } 3345 } 3346 3347 s = pserialize_read_enter(); 3348 IFADDR_READER_FOREACH(ifa, ifp) { 3349 struct sockaddr *sa = ifa->ifa_addr; 3350 /* all sockaddrs must fit in sockaddr_storage */ 3351 KASSERT(sa->sa_len <= sizeof(ifr.ifr_ifru)); 3352 3353 if (!docopy) { 3354 space += sz; 3355 continue; 3356 } 3357 memcpy(&ifr.ifr_space, sa, sa->sa_len); 3358 pserialize_read_exit(s); 3359 3360 if (space >= sz) { 3361 error = copyout(&ifr, ifrp, sz); 3362 if (error != 0) 3363 goto release_exit; 3364 ifrp++; space -= sz; 3365 } 3366 s = pserialize_read_enter(); 3367 } 3368 pserialize_read_exit(s); 3369 3370 next: 3371 s = pserialize_read_enter(); 3372 psref_release(&psref, &ifp->if_psref, ifnet_psref_class); 3373 } 3374 pserialize_read_exit(s); 3375 curlwp_bindx(bound); 3376 3377 if (docopy) { 3378 KASSERT(0 <= space && space <= ifc->ifc_len); 3379 ifc->ifc_len -= space; 3380 } else { 3381 KASSERT(space >= 0); 3382 ifc->ifc_len = space; 3383 } 3384 return (0); 3385 3386 release_exit: 3387 psref_release(&psref, &ifp->if_psref, ifnet_psref_class); 3388 curlwp_bindx(bound); 3389 return error; 3390 } 3391 3392 int 3393 ifreq_setaddr(u_long cmd, struct ifreq *ifr, const struct sockaddr *sa) 3394 { 3395 uint8_t len = sizeof(ifr->ifr_ifru.ifru_space); 3396 #ifdef COMPAT_OIFREQ 3397 struct ifreq ifrb; 3398 struct oifreq *oifr = NULL; 3399 u_long ocmd = cmd; 3400 3401 if (vec_compat_cvtcmd) { 3402 cmd = (*vec_compat_cvtcmd)(cmd); 3403 if (cmd != ocmd) { 3404 oifr = (struct oifreq *)(void *)ifr; 3405 ifr = &ifrb; 3406 ifreqo2n(oifr, ifr); 3407 len = sizeof(oifr->ifr_addr); 3408 } 3409 } 3410 #endif 3411 if (len < sa->sa_len) 3412 return EFBIG; 3413 3414 memset(&ifr->ifr_addr, 0, len); 3415 sockaddr_copy(&ifr->ifr_addr, len, sa); 3416 3417 #ifdef COMPAT_OIFREQ 3418 if (cmd != ocmd) 3419 ifreqn2o(oifr, ifr); 3420 #endif 3421 return 0; 3422 } 3423 3424 /* 3425 * wrapper function for the drivers which doesn't have if_transmit(). 3426 */ 3427 static int 3428 if_transmit(struct ifnet *ifp, struct mbuf *m) 3429 { 3430 int s, error; 3431 size_t pktlen = m->m_pkthdr.len; 3432 bool mcast = (m->m_flags & M_MCAST) != 0; 3433 3434 s = splnet(); 3435 3436 IFQ_ENQUEUE(&ifp->if_snd, m, error); 3437 if (error != 0) { 3438 /* mbuf is already freed */ 3439 goto out; 3440 } 3441 3442 ifp->if_obytes += pktlen; 3443 if (mcast) 3444 ifp->if_omcasts++; 3445 3446 if ((ifp->if_flags & IFF_OACTIVE) == 0) 3447 if_start_lock(ifp); 3448 out: 3449 splx(s); 3450 3451 return error; 3452 } 3453 3454 int 3455 if_transmit_lock(struct ifnet *ifp, struct mbuf *m) 3456 { 3457 int error; 3458 3459 #ifdef ALTQ 3460 KERNEL_LOCK(1, NULL); 3461 if (ALTQ_IS_ENABLED(&ifp->if_snd)) { 3462 error = if_transmit(ifp, m); 3463 KERNEL_UNLOCK_ONE(NULL); 3464 } else { 3465 KERNEL_UNLOCK_ONE(NULL); 3466 error = (*ifp->if_transmit)(ifp, m); 3467 /* mbuf is alredy freed */ 3468 } 3469 #else /* !ALTQ */ 3470 error = (*ifp->if_transmit)(ifp, m); 3471 /* mbuf is alredy freed */ 3472 #endif /* !ALTQ */ 3473 3474 return error; 3475 } 3476 3477 /* 3478 * Queue message on interface, and start output if interface 3479 * not yet active. 3480 */ 3481 int 3482 ifq_enqueue(struct ifnet *ifp, struct mbuf *m) 3483 { 3484 3485 return if_transmit_lock(ifp, m); 3486 } 3487 3488 /* 3489 * Queue message on interface, possibly using a second fast queue 3490 */ 3491 int 3492 ifq_enqueue2(struct ifnet *ifp, struct ifqueue *ifq, struct mbuf *m) 3493 { 3494 int error = 0; 3495 3496 if (ifq != NULL 3497 #ifdef ALTQ 3498 && ALTQ_IS_ENABLED(&ifp->if_snd) == 0 3499 #endif 3500 ) { 3501 if (IF_QFULL(ifq)) { 3502 IF_DROP(&ifp->if_snd); 3503 m_freem(m); 3504 if (error == 0) 3505 error = ENOBUFS; 3506 } else 3507 IF_ENQUEUE(ifq, m); 3508 } else 3509 IFQ_ENQUEUE(&ifp->if_snd, m, error); 3510 if (error != 0) { 3511 ++ifp->if_oerrors; 3512 return error; 3513 } 3514 return 0; 3515 } 3516 3517 int 3518 if_addr_init(ifnet_t *ifp, struct ifaddr *ifa, const bool src) 3519 { 3520 int rc; 3521 3522 KASSERT(IFNET_LOCKED(ifp)); 3523 if (ifp->if_initaddr != NULL) 3524 rc = (*ifp->if_initaddr)(ifp, ifa, src); 3525 else if (src || 3526 (rc = (*ifp->if_ioctl)(ifp, SIOCSIFDSTADDR, ifa)) == ENOTTY) 3527 rc = (*ifp->if_ioctl)(ifp, SIOCINITIFADDR, ifa); 3528 3529 return rc; 3530 } 3531 3532 int 3533 if_do_dad(struct ifnet *ifp) 3534 { 3535 if ((ifp->if_flags & IFF_LOOPBACK) != 0) 3536 return 0; 3537 3538 switch (ifp->if_type) { 3539 case IFT_FAITH: 3540 /* 3541 * These interfaces do not have the IFF_LOOPBACK flag, 3542 * but loop packets back. We do not have to do DAD on such 3543 * interfaces. We should even omit it, because loop-backed 3544 * responses would confuse the DAD procedure. 3545 */ 3546 return 0; 3547 default: 3548 /* 3549 * Our DAD routine requires the interface up and running. 3550 * However, some interfaces can be up before the RUNNING 3551 * status. Additionaly, users may try to assign addresses 3552 * before the interface becomes up (or running). 3553 * We simply skip DAD in such a case as a work around. 3554 * XXX: we should rather mark "tentative" on such addresses, 3555 * and do DAD after the interface becomes ready. 3556 */ 3557 if ((ifp->if_flags & (IFF_UP|IFF_RUNNING)) != 3558 (IFF_UP|IFF_RUNNING)) 3559 return 0; 3560 3561 return 1; 3562 } 3563 } 3564 3565 int 3566 if_flags_set(ifnet_t *ifp, const short flags) 3567 { 3568 int rc; 3569 3570 KASSERT(IFNET_LOCKED(ifp)); 3571 3572 if (ifp->if_setflags != NULL) 3573 rc = (*ifp->if_setflags)(ifp, flags); 3574 else { 3575 short cantflags, chgdflags; 3576 struct ifreq ifr; 3577 3578 chgdflags = ifp->if_flags ^ flags; 3579 cantflags = chgdflags & IFF_CANTCHANGE; 3580 3581 if (cantflags != 0) 3582 ifp->if_flags ^= cantflags; 3583 3584 /* Traditionally, we do not call if_ioctl after 3585 * setting/clearing only IFF_PROMISC if the interface 3586 * isn't IFF_UP. Uphold that tradition. 3587 */ 3588 if (chgdflags == IFF_PROMISC && (ifp->if_flags & IFF_UP) == 0) 3589 return 0; 3590 3591 memset(&ifr, 0, sizeof(ifr)); 3592 3593 ifr.ifr_flags = flags & ~IFF_CANTCHANGE; 3594 rc = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, &ifr); 3595 3596 if (rc != 0 && cantflags != 0) 3597 ifp->if_flags ^= cantflags; 3598 } 3599 3600 return rc; 3601 } 3602 3603 int 3604 if_mcast_op(ifnet_t *ifp, const unsigned long cmd, const struct sockaddr *sa) 3605 { 3606 int rc; 3607 struct ifreq ifr; 3608 3609 /* There remain some paths that don't hold IFNET_LOCK yet */ 3610 #ifdef NET_MPSAFE 3611 /* CARP and MROUTING still don't deal with the lock yet */ 3612 #if (!defined(NCARP) || (NCARP == 0)) && !defined(MROUTING) 3613 KASSERT(IFNET_LOCKED(ifp)); 3614 #endif 3615 #endif 3616 if (ifp->if_mcastop != NULL) 3617 rc = (*ifp->if_mcastop)(ifp, cmd, sa); 3618 else { 3619 ifreq_setaddr(cmd, &ifr, sa); 3620 rc = (*ifp->if_ioctl)(ifp, cmd, &ifr); 3621 } 3622 3623 return rc; 3624 } 3625 3626 static void 3627 sysctl_sndq_setup(struct sysctllog **clog, const char *ifname, 3628 struct ifaltq *ifq) 3629 { 3630 const struct sysctlnode *cnode, *rnode; 3631 3632 if (sysctl_createv(clog, 0, NULL, &rnode, 3633 CTLFLAG_PERMANENT, 3634 CTLTYPE_NODE, "interfaces", 3635 SYSCTL_DESCR("Per-interface controls"), 3636 NULL, 0, NULL, 0, 3637 CTL_NET, CTL_CREATE, CTL_EOL) != 0) 3638 goto bad; 3639 3640 if (sysctl_createv(clog, 0, &rnode, &rnode, 3641 CTLFLAG_PERMANENT, 3642 CTLTYPE_NODE, ifname, 3643 SYSCTL_DESCR("Interface controls"), 3644 NULL, 0, NULL, 0, 3645 CTL_CREATE, CTL_EOL) != 0) 3646 goto bad; 3647 3648 if (sysctl_createv(clog, 0, &rnode, &rnode, 3649 CTLFLAG_PERMANENT, 3650 CTLTYPE_NODE, "sndq", 3651 SYSCTL_DESCR("Interface output queue controls"), 3652 NULL, 0, NULL, 0, 3653 CTL_CREATE, CTL_EOL) != 0) 3654 goto bad; 3655 3656 if (sysctl_createv(clog, 0, &rnode, &cnode, 3657 CTLFLAG_PERMANENT, 3658 CTLTYPE_INT, "len", 3659 SYSCTL_DESCR("Current output queue length"), 3660 NULL, 0, &ifq->ifq_len, 0, 3661 CTL_CREATE, CTL_EOL) != 0) 3662 goto bad; 3663 3664 if (sysctl_createv(clog, 0, &rnode, &cnode, 3665 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 3666 CTLTYPE_INT, "maxlen", 3667 SYSCTL_DESCR("Maximum allowed output queue length"), 3668 NULL, 0, &ifq->ifq_maxlen, 0, 3669 CTL_CREATE, CTL_EOL) != 0) 3670 goto bad; 3671 3672 if (sysctl_createv(clog, 0, &rnode, &cnode, 3673 CTLFLAG_PERMANENT, 3674 CTLTYPE_INT, "drops", 3675 SYSCTL_DESCR("Packets dropped due to full output queue"), 3676 NULL, 0, &ifq->ifq_drops, 0, 3677 CTL_CREATE, CTL_EOL) != 0) 3678 goto bad; 3679 3680 return; 3681 bad: 3682 printf("%s: could not attach sysctl nodes\n", ifname); 3683 return; 3684 } 3685 3686 #if defined(INET) || defined(INET6) 3687 3688 #define SYSCTL_NET_PKTQ(q, cn, c) \ 3689 static int \ 3690 sysctl_net_##q##_##cn(SYSCTLFN_ARGS) \ 3691 { \ 3692 return sysctl_pktq_count(SYSCTLFN_CALL(rnode), q, c); \ 3693 } 3694 3695 #if defined(INET) 3696 static int 3697 sysctl_net_ip_pktq_maxlen(SYSCTLFN_ARGS) 3698 { 3699 return sysctl_pktq_maxlen(SYSCTLFN_CALL(rnode), ip_pktq); 3700 } 3701 SYSCTL_NET_PKTQ(ip_pktq, items, PKTQ_NITEMS) 3702 SYSCTL_NET_PKTQ(ip_pktq, drops, PKTQ_DROPS) 3703 #endif 3704 3705 #if defined(INET6) 3706 static int 3707 sysctl_net_ip6_pktq_maxlen(SYSCTLFN_ARGS) 3708 { 3709 return sysctl_pktq_maxlen(SYSCTLFN_CALL(rnode), ip6_pktq); 3710 } 3711 SYSCTL_NET_PKTQ(ip6_pktq, items, PKTQ_NITEMS) 3712 SYSCTL_NET_PKTQ(ip6_pktq, drops, PKTQ_DROPS) 3713 #endif 3714 3715 static void 3716 sysctl_net_pktq_setup(struct sysctllog **clog, int pf) 3717 { 3718 sysctlfn len_func = NULL, maxlen_func = NULL, drops_func = NULL; 3719 const char *pfname = NULL, *ipname = NULL; 3720 int ipn = 0, qid = 0; 3721 3722 switch (pf) { 3723 #if defined(INET) 3724 case PF_INET: 3725 len_func = sysctl_net_ip_pktq_items; 3726 maxlen_func = sysctl_net_ip_pktq_maxlen; 3727 drops_func = sysctl_net_ip_pktq_drops; 3728 pfname = "inet", ipn = IPPROTO_IP; 3729 ipname = "ip", qid = IPCTL_IFQ; 3730 break; 3731 #endif 3732 #if defined(INET6) 3733 case PF_INET6: 3734 len_func = sysctl_net_ip6_pktq_items; 3735 maxlen_func = sysctl_net_ip6_pktq_maxlen; 3736 drops_func = sysctl_net_ip6_pktq_drops; 3737 pfname = "inet6", ipn = IPPROTO_IPV6; 3738 ipname = "ip6", qid = IPV6CTL_IFQ; 3739 break; 3740 #endif 3741 default: 3742 KASSERT(false); 3743 } 3744 3745 sysctl_createv(clog, 0, NULL, NULL, 3746 CTLFLAG_PERMANENT, 3747 CTLTYPE_NODE, pfname, NULL, 3748 NULL, 0, NULL, 0, 3749 CTL_NET, pf, CTL_EOL); 3750 sysctl_createv(clog, 0, NULL, NULL, 3751 CTLFLAG_PERMANENT, 3752 CTLTYPE_NODE, ipname, NULL, 3753 NULL, 0, NULL, 0, 3754 CTL_NET, pf, ipn, CTL_EOL); 3755 sysctl_createv(clog, 0, NULL, NULL, 3756 CTLFLAG_PERMANENT, 3757 CTLTYPE_NODE, "ifq", 3758 SYSCTL_DESCR("Protocol input queue controls"), 3759 NULL, 0, NULL, 0, 3760 CTL_NET, pf, ipn, qid, CTL_EOL); 3761 3762 sysctl_createv(clog, 0, NULL, NULL, 3763 CTLFLAG_PERMANENT, 3764 CTLTYPE_INT, "len", 3765 SYSCTL_DESCR("Current input queue length"), 3766 len_func, 0, NULL, 0, 3767 CTL_NET, pf, ipn, qid, IFQCTL_LEN, CTL_EOL); 3768 sysctl_createv(clog, 0, NULL, NULL, 3769 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 3770 CTLTYPE_INT, "maxlen", 3771 SYSCTL_DESCR("Maximum allowed input queue length"), 3772 maxlen_func, 0, NULL, 0, 3773 CTL_NET, pf, ipn, qid, IFQCTL_MAXLEN, CTL_EOL); 3774 sysctl_createv(clog, 0, NULL, NULL, 3775 CTLFLAG_PERMANENT, 3776 CTLTYPE_INT, "drops", 3777 SYSCTL_DESCR("Packets dropped due to full input queue"), 3778 drops_func, 0, NULL, 0, 3779 CTL_NET, pf, ipn, qid, IFQCTL_DROPS, CTL_EOL); 3780 } 3781 #endif /* INET || INET6 */ 3782 3783 static int 3784 if_sdl_sysctl(SYSCTLFN_ARGS) 3785 { 3786 struct ifnet *ifp; 3787 const struct sockaddr_dl *sdl; 3788 struct psref psref; 3789 int error = 0; 3790 int bound; 3791 3792 if (namelen != 1) 3793 return EINVAL; 3794 3795 bound = curlwp_bind(); 3796 ifp = if_get_byindex(name[0], &psref); 3797 if (ifp == NULL) { 3798 error = ENODEV; 3799 goto out0; 3800 } 3801 3802 sdl = ifp->if_sadl; 3803 if (sdl == NULL) { 3804 *oldlenp = 0; 3805 goto out1; 3806 } 3807 3808 if (oldp == NULL) { 3809 *oldlenp = sdl->sdl_alen; 3810 goto out1; 3811 } 3812 3813 if (*oldlenp >= sdl->sdl_alen) 3814 *oldlenp = sdl->sdl_alen; 3815 error = sysctl_copyout(l, &sdl->sdl_data[sdl->sdl_nlen], oldp, *oldlenp); 3816 out1: 3817 if_put(ifp, &psref); 3818 out0: 3819 curlwp_bindx(bound); 3820 return error; 3821 } 3822 3823 static void 3824 if_sysctl_setup(struct sysctllog **clog) 3825 { 3826 const struct sysctlnode *rnode = NULL; 3827 3828 sysctl_createv(clog, 0, NULL, &rnode, 3829 CTLFLAG_PERMANENT, 3830 CTLTYPE_NODE, "sdl", 3831 SYSCTL_DESCR("Get active link-layer address"), 3832 if_sdl_sysctl, 0, NULL, 0, 3833 CTL_NET, CTL_CREATE, CTL_EOL); 3834 3835 #if defined(INET) 3836 sysctl_net_pktq_setup(NULL, PF_INET); 3837 #endif 3838 #ifdef INET6 3839 if (in6_present) 3840 sysctl_net_pktq_setup(NULL, PF_INET6); 3841 #endif 3842 } 3843
Indexes created Wed Sep 24 05:09:52 GMT 2025