if.c revision 1.352
1 /* $NetBSD: if.c,v 1.352 2016/07/04 04:35:09 knakahara Exp $ */ 2 3 /*- 4 * Copyright (c) 1999, 2000, 2001, 2008 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by William Studenmund and Jason R. Thorpe. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 */ 31 32 /* 33 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. 34 * All rights reserved. 35 * 36 * Redistribution and use in source and binary forms, with or without 37 * modification, are permitted provided that the following conditions 38 * are met: 39 * 1. Redistributions of source code must retain the above copyright 40 * notice, this list of conditions and the following disclaimer. 41 * 2. Redistributions in binary form must reproduce the above copyright 42 * notice, this list of conditions and the following disclaimer in the 43 * documentation and/or other materials provided with the distribution. 44 * 3. Neither the name of the project nor the names of its contributors 45 * may be used to endorse or promote products derived from this software 46 * without specific prior written permission. 47 * 48 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND 49 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 50 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 51 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE 52 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 53 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 54 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 55 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 56 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 57 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 58 * SUCH DAMAGE. 59 */ 60 61 /* 62 * Copyright (c) 1980, 1986, 1993 63 * The Regents of the University of California. All rights reserved. 64 * 65 * Redistribution and use in source and binary forms, with or without 66 * modification, are permitted provided that the following conditions 67 * are met: 68 * 1. Redistributions of source code must retain the above copyright 69 * notice, this list of conditions and the following disclaimer. 70 * 2. Redistributions in binary form must reproduce the above copyright 71 * notice, this list of conditions and the following disclaimer in the 72 * documentation and/or other materials provided with the distribution. 73 * 3. Neither the name of the University nor the names of its contributors 74 * may be used to endorse or promote products derived from this software 75 * without specific prior written permission. 76 * 77 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 78 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 79 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 80 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 81 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 82 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 83 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 84 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 85 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 86 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 87 * SUCH DAMAGE. 88 * 89 * @(#)if.c 8.5 (Berkeley) 1/9/95 90 */ 91 92 #include <sys/cdefs.h> 93 __KERNEL_RCSID(0, "$NetBSD: if.c,v 1.352 2016/07/04 04:35:09 knakahara Exp $"); 94 95 #if defined(_KERNEL_OPT) 96 #include "opt_inet.h" 97 #include "opt_ipsec.h" 98 99 #include "opt_atalk.h" 100 #include "opt_natm.h" 101 #include "opt_wlan.h" 102 #include "opt_net_mpsafe.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 #ifndef IPSEC 142 #include <netinet/ip_encap.h> 143 #endif 144 145 #ifdef INET6 146 #include <netinet6/in6_var.h> 147 #include <netinet6/nd6.h> 148 #endif 149 150 #include "ether.h" 151 #include "fddi.h" 152 #include "token.h" 153 154 #include "carp.h" 155 #if NCARP > 0 156 #include <netinet/ip_carp.h> 157 #endif 158 159 #include <compat/sys/sockio.h> 160 #include <compat/sys/socket.h> 161 162 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address"); 163 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address"); 164 165 /* 166 * Global list of interfaces. 167 */ 168 /* DEPRECATED. Remove it once kvm(3) users disappeared */ 169 struct ifnet_head ifnet_list; 170 171 struct pslist_head ifnet_pslist; 172 static ifnet_t ** ifindex2ifnet = NULL; 173 static u_int if_index = 1; 174 static size_t if_indexlim = 0; 175 static uint64_t index_gen; 176 /* Mutex to protect the above objects. */ 177 kmutex_t ifnet_mtx __cacheline_aligned; 178 struct psref_class *ifnet_psref_class __read_mostly; 179 static pserialize_t ifnet_psz; 180 181 static kmutex_t if_clone_mtx; 182 183 struct ifnet *lo0ifp; 184 int ifqmaxlen = IFQ_MAXLEN; 185 186 static int if_rt_walktree(struct rtentry *, void *); 187 188 static struct if_clone *if_clone_lookup(const char *, int *); 189 190 static LIST_HEAD(, if_clone) if_cloners = LIST_HEAD_INITIALIZER(if_cloners); 191 static int if_cloners_count; 192 193 /* Packet filtering hook for interfaces. */ 194 pfil_head_t * if_pfil; 195 196 static kauth_listener_t if_listener; 197 198 static int doifioctl(struct socket *, u_long, void *, struct lwp *); 199 static void if_detach_queues(struct ifnet *, struct ifqueue *); 200 static void sysctl_sndq_setup(struct sysctllog **, const char *, 201 struct ifaltq *); 202 static void if_slowtimo(void *); 203 static void if_free_sadl(struct ifnet *); 204 static void if_attachdomain1(struct ifnet *); 205 static int ifconf(u_long, void *); 206 static int if_transmit(struct ifnet *, struct mbuf *); 207 static int if_clone_create(const char *); 208 static int if_clone_destroy(const char *); 209 static void if_link_state_change_si(void *); 210 211 struct if_percpuq { 212 struct ifnet *ipq_ifp; 213 void *ipq_si; 214 struct percpu *ipq_ifqs; /* struct ifqueue */ 215 }; 216 217 static struct mbuf *if_percpuq_dequeue(struct if_percpuq *); 218 219 static void if_percpuq_drops(void *, void *, struct cpu_info *); 220 static int sysctl_percpuq_drops_handler(SYSCTLFN_PROTO); 221 static void sysctl_percpuq_setup(struct sysctllog **, const char *, 222 struct if_percpuq *); 223 224 #if defined(INET) || defined(INET6) 225 static void sysctl_net_pktq_setup(struct sysctllog **, int); 226 #endif 227 228 static int 229 if_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie, 230 void *arg0, void *arg1, void *arg2, void *arg3) 231 { 232 int result; 233 enum kauth_network_req req; 234 235 result = KAUTH_RESULT_DEFER; 236 req = (enum kauth_network_req)arg1; 237 238 if (action != KAUTH_NETWORK_INTERFACE) 239 return result; 240 241 if ((req == KAUTH_REQ_NETWORK_INTERFACE_GET) || 242 (req == KAUTH_REQ_NETWORK_INTERFACE_SET)) 243 result = KAUTH_RESULT_ALLOW; 244 245 return result; 246 } 247 248 /* 249 * Network interface utility routines. 250 * 251 * Routines with ifa_ifwith* names take sockaddr *'s as 252 * parameters. 253 */ 254 void 255 ifinit(void) 256 { 257 #if defined(INET) 258 sysctl_net_pktq_setup(NULL, PF_INET); 259 #endif 260 #ifdef INET6 261 if (in6_present) 262 sysctl_net_pktq_setup(NULL, PF_INET6); 263 #endif 264 265 #ifndef IPSEC 266 encapinit(); 267 #endif 268 269 if_listener = kauth_listen_scope(KAUTH_SCOPE_NETWORK, 270 if_listener_cb, NULL); 271 272 /* interfaces are available, inform socket code */ 273 ifioctl = doifioctl; 274 } 275 276 /* 277 * XXX Initialization before configure(). 278 * XXX hack to get pfil_add_hook working in autoconf. 279 */ 280 void 281 ifinit1(void) 282 { 283 mutex_init(&if_clone_mtx, MUTEX_DEFAULT, IPL_NONE); 284 TAILQ_INIT(&ifnet_list); 285 mutex_init(&ifnet_mtx, MUTEX_DEFAULT, IPL_NONE); 286 ifnet_psz = pserialize_create(); 287 ifnet_psref_class = psref_class_create("ifnet", IPL_SOFTNET); 288 PSLIST_INIT(&ifnet_pslist); 289 if_indexlim = 8; 290 291 if_pfil = pfil_head_create(PFIL_TYPE_IFNET, NULL); 292 KASSERT(if_pfil != NULL); 293 294 #if NETHER > 0 || NFDDI > 0 || defined(NETATALK) || NTOKEN > 0 || defined(WLAN) 295 etherinit(); 296 #endif 297 } 298 299 ifnet_t * 300 if_alloc(u_char type) 301 { 302 return kmem_zalloc(sizeof(ifnet_t), KM_SLEEP); 303 } 304 305 void 306 if_free(ifnet_t *ifp) 307 { 308 kmem_free(ifp, sizeof(ifnet_t)); 309 } 310 311 void 312 if_initname(struct ifnet *ifp, const char *name, int unit) 313 { 314 (void)snprintf(ifp->if_xname, sizeof(ifp->if_xname), 315 "%s%d", name, unit); 316 } 317 318 /* 319 * Null routines used while an interface is going away. These routines 320 * just return an error. 321 */ 322 323 int 324 if_nulloutput(struct ifnet *ifp, struct mbuf *m, 325 const struct sockaddr *so, const struct rtentry *rt) 326 { 327 328 return ENXIO; 329 } 330 331 void 332 if_nullinput(struct ifnet *ifp, struct mbuf *m) 333 { 334 335 /* Nothing. */ 336 } 337 338 void 339 if_nullstart(struct ifnet *ifp) 340 { 341 342 /* Nothing. */ 343 } 344 345 int 346 if_nulltransmit(struct ifnet *ifp, struct mbuf *m) 347 { 348 349 return ENXIO; 350 } 351 352 int 353 if_nullioctl(struct ifnet *ifp, u_long cmd, void *data) 354 { 355 356 return ENXIO; 357 } 358 359 int 360 if_nullinit(struct ifnet *ifp) 361 { 362 363 return ENXIO; 364 } 365 366 void 367 if_nullstop(struct ifnet *ifp, int disable) 368 { 369 370 /* Nothing. */ 371 } 372 373 void 374 if_nullslowtimo(struct ifnet *ifp) 375 { 376 377 /* Nothing. */ 378 } 379 380 void 381 if_nulldrain(struct ifnet *ifp) 382 { 383 384 /* Nothing. */ 385 } 386 387 void 388 if_set_sadl(struct ifnet *ifp, const void *lla, u_char addrlen, bool factory) 389 { 390 struct ifaddr *ifa; 391 struct sockaddr_dl *sdl; 392 393 ifp->if_addrlen = addrlen; 394 if_alloc_sadl(ifp); 395 ifa = ifp->if_dl; 396 sdl = satosdl(ifa->ifa_addr); 397 398 (void)sockaddr_dl_setaddr(sdl, sdl->sdl_len, lla, ifp->if_addrlen); 399 if (factory) { 400 ifp->if_hwdl = ifp->if_dl; 401 ifaref(ifp->if_hwdl); 402 } 403 /* TBD routing socket */ 404 } 405 406 struct ifaddr * 407 if_dl_create(const struct ifnet *ifp, const struct sockaddr_dl **sdlp) 408 { 409 unsigned socksize, ifasize; 410 int addrlen, namelen; 411 struct sockaddr_dl *mask, *sdl; 412 struct ifaddr *ifa; 413 414 namelen = strlen(ifp->if_xname); 415 addrlen = ifp->if_addrlen; 416 socksize = roundup(sockaddr_dl_measure(namelen, addrlen), sizeof(long)); 417 ifasize = sizeof(*ifa) + 2 * socksize; 418 ifa = (struct ifaddr *)malloc(ifasize, M_IFADDR, M_WAITOK|M_ZERO); 419 420 sdl = (struct sockaddr_dl *)(ifa + 1); 421 mask = (struct sockaddr_dl *)(socksize + (char *)sdl); 422 423 sockaddr_dl_init(sdl, socksize, ifp->if_index, ifp->if_type, 424 ifp->if_xname, namelen, NULL, addrlen); 425 mask->sdl_len = sockaddr_dl_measure(namelen, 0); 426 memset(&mask->sdl_data[0], 0xff, namelen); 427 ifa->ifa_rtrequest = link_rtrequest; 428 ifa->ifa_addr = (struct sockaddr *)sdl; 429 ifa->ifa_netmask = (struct sockaddr *)mask; 430 431 *sdlp = sdl; 432 433 return ifa; 434 } 435 436 static void 437 if_sadl_setrefs(struct ifnet *ifp, struct ifaddr *ifa) 438 { 439 const struct sockaddr_dl *sdl; 440 441 ifp->if_dl = ifa; 442 ifaref(ifa); 443 sdl = satosdl(ifa->ifa_addr); 444 ifp->if_sadl = sdl; 445 } 446 447 /* 448 * Allocate the link level name for the specified interface. This 449 * is an attachment helper. It must be called after ifp->if_addrlen 450 * is initialized, which may not be the case when if_attach() is 451 * called. 452 */ 453 void 454 if_alloc_sadl(struct ifnet *ifp) 455 { 456 struct ifaddr *ifa; 457 const struct sockaddr_dl *sdl; 458 459 /* 460 * If the interface already has a link name, release it 461 * now. This is useful for interfaces that can change 462 * link types, and thus switch link names often. 463 */ 464 if (ifp->if_sadl != NULL) 465 if_free_sadl(ifp); 466 467 ifa = if_dl_create(ifp, &sdl); 468 469 ifa_insert(ifp, ifa); 470 if_sadl_setrefs(ifp, ifa); 471 } 472 473 static void 474 if_deactivate_sadl(struct ifnet *ifp) 475 { 476 struct ifaddr *ifa; 477 478 KASSERT(ifp->if_dl != NULL); 479 480 ifa = ifp->if_dl; 481 482 ifp->if_sadl = NULL; 483 484 ifp->if_dl = NULL; 485 ifafree(ifa); 486 } 487 488 void 489 if_activate_sadl(struct ifnet *ifp, struct ifaddr *ifa, 490 const struct sockaddr_dl *sdl) 491 { 492 int s; 493 494 s = splnet(); 495 496 if_deactivate_sadl(ifp); 497 498 if_sadl_setrefs(ifp, ifa); 499 IFADDR_FOREACH(ifa, ifp) 500 rtinit(ifa, RTM_LLINFO_UPD, 0); 501 splx(s); 502 } 503 504 /* 505 * Free the link level name for the specified interface. This is 506 * a detach helper. This is called from if_detach(). 507 */ 508 static void 509 if_free_sadl(struct ifnet *ifp) 510 { 511 struct ifaddr *ifa; 512 int s; 513 514 ifa = ifp->if_dl; 515 if (ifa == NULL) { 516 KASSERT(ifp->if_sadl == NULL); 517 return; 518 } 519 520 KASSERT(ifp->if_sadl != NULL); 521 522 s = splnet(); 523 rtinit(ifa, RTM_DELETE, 0); 524 ifa_remove(ifp, ifa); 525 if_deactivate_sadl(ifp); 526 if (ifp->if_hwdl == ifa) { 527 ifafree(ifa); 528 ifp->if_hwdl = NULL; 529 } 530 splx(s); 531 } 532 533 static void 534 if_getindex(ifnet_t *ifp) 535 { 536 bool hitlimit = false; 537 538 ifp->if_index_gen = index_gen++; 539 540 ifp->if_index = if_index; 541 if (ifindex2ifnet == NULL) { 542 if_index++; 543 goto skip; 544 } 545 while (if_byindex(ifp->if_index)) { 546 /* 547 * If we hit USHRT_MAX, we skip back to 0 since 548 * there are a number of places where the value 549 * of if_index or if_index itself is compared 550 * to or stored in an unsigned short. By 551 * jumping back, we won't botch those assignments 552 * or comparisons. 553 */ 554 if (++if_index == 0) { 555 if_index = 1; 556 } else if (if_index == USHRT_MAX) { 557 /* 558 * However, if we have to jump back to 559 * zero *twice* without finding an empty 560 * slot in ifindex2ifnet[], then there 561 * there are too many (>65535) interfaces. 562 */ 563 if (hitlimit) { 564 panic("too many interfaces"); 565 } 566 hitlimit = true; 567 if_index = 1; 568 } 569 ifp->if_index = if_index; 570 } 571 skip: 572 /* 573 * ifindex2ifnet is indexed by if_index. Since if_index will 574 * grow dynamically, it should grow too. 575 */ 576 if (ifindex2ifnet == NULL || ifp->if_index >= if_indexlim) { 577 size_t m, n, oldlim; 578 void *q; 579 580 oldlim = if_indexlim; 581 while (ifp->if_index >= if_indexlim) 582 if_indexlim <<= 1; 583 584 /* grow ifindex2ifnet */ 585 m = oldlim * sizeof(struct ifnet *); 586 n = if_indexlim * sizeof(struct ifnet *); 587 q = malloc(n, M_IFADDR, M_WAITOK|M_ZERO); 588 if (ifindex2ifnet != NULL) { 589 memcpy(q, ifindex2ifnet, m); 590 free(ifindex2ifnet, M_IFADDR); 591 } 592 ifindex2ifnet = (struct ifnet **)q; 593 } 594 ifindex2ifnet[ifp->if_index] = ifp; 595 } 596 597 /* 598 * Initialize an interface and assign an index for it. 599 * 600 * It must be called prior to a device specific attach routine 601 * (e.g., ether_ifattach and ieee80211_ifattach) or if_alloc_sadl, 602 * and be followed by if_register: 603 * 604 * if_initialize(ifp); 605 * ether_ifattach(ifp, enaddr); 606 * if_register(ifp); 607 */ 608 void 609 if_initialize(ifnet_t *ifp) 610 { 611 KASSERT(if_indexlim > 0); 612 TAILQ_INIT(&ifp->if_addrlist); 613 614 /* 615 * Link level name is allocated later by a separate call to 616 * if_alloc_sadl(). 617 */ 618 619 if (ifp->if_snd.ifq_maxlen == 0) 620 ifp->if_snd.ifq_maxlen = ifqmaxlen; 621 622 ifp->if_broadcastaddr = 0; /* reliably crash if used uninitialized */ 623 624 ifp->if_link_state = LINK_STATE_UNKNOWN; 625 ifp->if_link_queue = -1; /* all bits set, see link_state_change() */ 626 627 ifp->if_capenable = 0; 628 ifp->if_csum_flags_tx = 0; 629 ifp->if_csum_flags_rx = 0; 630 631 #ifdef ALTQ 632 ifp->if_snd.altq_type = 0; 633 ifp->if_snd.altq_disc = NULL; 634 ifp->if_snd.altq_flags &= ALTQF_CANTCHANGE; 635 ifp->if_snd.altq_tbr = NULL; 636 ifp->if_snd.altq_ifp = ifp; 637 #endif 638 639 #ifdef NET_MPSAFE 640 ifp->if_snd.ifq_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NET); 641 #else 642 ifp->if_snd.ifq_lock = NULL; 643 #endif 644 645 ifp->if_pfil = pfil_head_create(PFIL_TYPE_IFNET, ifp); 646 (void)pfil_run_hooks(if_pfil, 647 (struct mbuf **)PFIL_IFNET_ATTACH, ifp, PFIL_IFNET); 648 649 IF_AFDATA_LOCK_INIT(ifp); 650 651 if (if_is_link_state_changeable(ifp)) { 652 ifp->if_link_si = softint_establish(SOFTINT_NET, 653 if_link_state_change_si, ifp); 654 if (ifp->if_link_si == NULL) 655 panic("%s: softint_establish() failed", __func__); 656 } 657 658 PSLIST_ENTRY_INIT(ifp, if_pslist_entry); 659 psref_target_init(&ifp->if_psref, ifnet_psref_class); 660 ifp->if_ioctl_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 661 662 IFNET_LOCK(); 663 if_getindex(ifp); 664 IFNET_UNLOCK(); 665 } 666 667 /* 668 * Register an interface to the list of "active" interfaces. 669 */ 670 void 671 if_register(ifnet_t *ifp) 672 { 673 /* 674 * If the driver has not supplied its own if_ioctl, then 675 * supply the default. 676 */ 677 if (ifp->if_ioctl == NULL) 678 ifp->if_ioctl = ifioctl_common; 679 680 sysctl_sndq_setup(&ifp->if_sysctl_log, ifp->if_xname, &ifp->if_snd); 681 682 if (!STAILQ_EMPTY(&domains)) 683 if_attachdomain1(ifp); 684 685 /* Announce the interface. */ 686 rt_ifannouncemsg(ifp, IFAN_ARRIVAL); 687 688 if (ifp->if_slowtimo != NULL) { 689 ifp->if_slowtimo_ch = 690 kmem_zalloc(sizeof(*ifp->if_slowtimo_ch), KM_SLEEP); 691 callout_init(ifp->if_slowtimo_ch, 0); 692 callout_setfunc(ifp->if_slowtimo_ch, if_slowtimo, ifp); 693 if_slowtimo(ifp); 694 } 695 696 if (ifp->if_transmit == NULL || ifp->if_transmit == if_nulltransmit) 697 ifp->if_transmit = if_transmit; 698 699 IFNET_LOCK(); 700 TAILQ_INSERT_TAIL(&ifnet_list, ifp, if_list); 701 IFNET_WRITER_INSERT_TAIL(ifp); 702 IFNET_UNLOCK(); 703 } 704 705 /* 706 * The if_percpuq framework 707 * 708 * It allows network device drivers to execute the network stack 709 * in softint (so called softint-based if_input). It utilizes 710 * softint and percpu ifqueue. It doesn't distribute any packets 711 * between CPUs, unlike pktqueue(9). 712 * 713 * Currently we support two options for device drivers to apply the framework: 714 * - Use it implicitly with less changes 715 * - If you use if_attach in driver's _attach function and if_input in 716 * driver's Rx interrupt handler, a packet is queued and a softint handles 717 * the packet implicitly 718 * - Use it explicitly in each driver (recommended) 719 * - You can use if_percpuq_* directly in your driver 720 * - In this case, you need to allocate struct if_percpuq in driver's softc 721 * - See wm(4) as a reference implementation 722 */ 723 724 static void 725 if_percpuq_softint(void *arg) 726 { 727 struct if_percpuq *ipq = arg; 728 struct ifnet *ifp = ipq->ipq_ifp; 729 struct mbuf *m; 730 731 while ((m = if_percpuq_dequeue(ipq)) != NULL) 732 ifp->_if_input(ifp, m); 733 } 734 735 static void 736 if_percpuq_init_ifq(void *p, void *arg __unused, struct cpu_info *ci __unused) 737 { 738 struct ifqueue *const ifq = p; 739 740 memset(ifq, 0, sizeof(*ifq)); 741 ifq->ifq_maxlen = IFQ_MAXLEN; 742 } 743 744 struct if_percpuq * 745 if_percpuq_create(struct ifnet *ifp) 746 { 747 struct if_percpuq *ipq; 748 749 ipq = kmem_zalloc(sizeof(*ipq), KM_SLEEP); 750 if (ipq == NULL) 751 panic("kmem_zalloc failed"); 752 753 ipq->ipq_ifp = ifp; 754 ipq->ipq_si = softint_establish(SOFTINT_NET|SOFTINT_MPSAFE, 755 if_percpuq_softint, ipq); 756 ipq->ipq_ifqs = percpu_alloc(sizeof(struct ifqueue)); 757 percpu_foreach(ipq->ipq_ifqs, &if_percpuq_init_ifq, NULL); 758 759 sysctl_percpuq_setup(&ifp->if_sysctl_log, ifp->if_xname, ipq); 760 761 return ipq; 762 } 763 764 static struct mbuf * 765 if_percpuq_dequeue(struct if_percpuq *ipq) 766 { 767 struct mbuf *m; 768 struct ifqueue *ifq; 769 int s; 770 771 s = splnet(); 772 ifq = percpu_getref(ipq->ipq_ifqs); 773 IF_DEQUEUE(ifq, m); 774 percpu_putref(ipq->ipq_ifqs); 775 splx(s); 776 777 return m; 778 } 779 780 static void 781 if_percpuq_purge_ifq(void *p, void *arg __unused, struct cpu_info *ci __unused) 782 { 783 struct ifqueue *const ifq = p; 784 785 IF_PURGE(ifq); 786 } 787 788 void 789 if_percpuq_destroy(struct if_percpuq *ipq) 790 { 791 792 /* if_detach may already destroy it */ 793 if (ipq == NULL) 794 return; 795 796 softint_disestablish(ipq->ipq_si); 797 percpu_foreach(ipq->ipq_ifqs, &if_percpuq_purge_ifq, NULL); 798 percpu_free(ipq->ipq_ifqs, sizeof(struct ifqueue)); 799 } 800 801 void 802 if_percpuq_enqueue(struct if_percpuq *ipq, struct mbuf *m) 803 { 804 struct ifqueue *ifq; 805 int s; 806 807 KASSERT(ipq != NULL); 808 809 s = splnet(); 810 ifq = percpu_getref(ipq->ipq_ifqs); 811 if (IF_QFULL(ifq)) { 812 IF_DROP(ifq); 813 percpu_putref(ipq->ipq_ifqs); 814 m_freem(m); 815 goto out; 816 } 817 IF_ENQUEUE(ifq, m); 818 percpu_putref(ipq->ipq_ifqs); 819 820 softint_schedule(ipq->ipq_si); 821 out: 822 splx(s); 823 } 824 825 static void 826 if_percpuq_drops(void *p, void *arg, struct cpu_info *ci __unused) 827 { 828 struct ifqueue *const ifq = p; 829 int *sum = arg; 830 831 *sum += ifq->ifq_drops; 832 } 833 834 static int 835 sysctl_percpuq_drops_handler(SYSCTLFN_ARGS) 836 { 837 struct sysctlnode node; 838 struct if_percpuq *ipq; 839 int sum = 0; 840 int error; 841 842 node = *rnode; 843 ipq = node.sysctl_data; 844 845 percpu_foreach(ipq->ipq_ifqs, if_percpuq_drops, &sum); 846 847 node.sysctl_data = ∑ 848 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 849 if (error != 0 || newp == NULL) 850 return error; 851 852 return 0; 853 } 854 855 static void 856 sysctl_percpuq_setup(struct sysctllog **clog, const char* ifname, 857 struct if_percpuq *ipq) 858 { 859 const struct sysctlnode *cnode, *rnode; 860 861 if (sysctl_createv(clog, 0, NULL, &rnode, 862 CTLFLAG_PERMANENT, 863 CTLTYPE_NODE, "interfaces", 864 SYSCTL_DESCR("Per-interface controls"), 865 NULL, 0, NULL, 0, 866 CTL_NET, CTL_CREATE, CTL_EOL) != 0) 867 goto bad; 868 869 if (sysctl_createv(clog, 0, &rnode, &rnode, 870 CTLFLAG_PERMANENT, 871 CTLTYPE_NODE, ifname, 872 SYSCTL_DESCR("Interface controls"), 873 NULL, 0, NULL, 0, 874 CTL_CREATE, CTL_EOL) != 0) 875 goto bad; 876 877 if (sysctl_createv(clog, 0, &rnode, &rnode, 878 CTLFLAG_PERMANENT, 879 CTLTYPE_NODE, "rcvq", 880 SYSCTL_DESCR("Interface input queue controls"), 881 NULL, 0, NULL, 0, 882 CTL_CREATE, CTL_EOL) != 0) 883 goto bad; 884 885 #ifdef NOTYET 886 /* XXX Should show each per-CPU queue length? */ 887 if (sysctl_createv(clog, 0, &rnode, &rnode, 888 CTLFLAG_PERMANENT, 889 CTLTYPE_INT, "len", 890 SYSCTL_DESCR("Current input queue length"), 891 sysctl_percpuq_len, 0, NULL, 0, 892 CTL_CREATE, CTL_EOL) != 0) 893 goto bad; 894 895 if (sysctl_createv(clog, 0, &rnode, &cnode, 896 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 897 CTLTYPE_INT, "maxlen", 898 SYSCTL_DESCR("Maximum allowed input queue length"), 899 sysctl_percpuq_maxlen_handler, 0, (void *)ipq, 0, 900 CTL_CREATE, CTL_EOL) != 0) 901 goto bad; 902 #endif 903 904 if (sysctl_createv(clog, 0, &rnode, &cnode, 905 CTLFLAG_PERMANENT, 906 CTLTYPE_INT, "drops", 907 SYSCTL_DESCR("Total packets dropped due to full input queue"), 908 sysctl_percpuq_drops_handler, 0, (void *)ipq, 0, 909 CTL_CREATE, CTL_EOL) != 0) 910 goto bad; 911 912 return; 913 bad: 914 printf("%s: could not attach sysctl nodes\n", ifname); 915 return; 916 } 917 918 919 /* 920 * The common interface input routine that is called by device drivers, 921 * which should be used only when the driver's rx handler already runs 922 * in softint. 923 */ 924 void 925 if_input(struct ifnet *ifp, struct mbuf *m) 926 { 927 928 KASSERT(ifp->if_percpuq == NULL); 929 KASSERT(!cpu_intr_p()); 930 931 ifp->_if_input(ifp, m); 932 } 933 934 /* 935 * DEPRECATED. Use if_initialize and if_register instead. 936 * See the above comment of if_initialize. 937 * 938 * Note that it implicitly enables if_percpuq to make drivers easy to 939 * migrate softint-based if_input without much changes. If you don't 940 * want to enable it, use if_initialize instead. 941 */ 942 void 943 if_attach(ifnet_t *ifp) 944 { 945 946 if_initialize(ifp); 947 ifp->if_percpuq = if_percpuq_create(ifp); 948 if_register(ifp); 949 } 950 951 void 952 if_attachdomain(void) 953 { 954 struct ifnet *ifp; 955 int s; 956 int bound = curlwp_bind(); 957 958 s = pserialize_read_enter(); 959 IFNET_READER_FOREACH(ifp) { 960 struct psref psref; 961 psref_acquire(&psref, &ifp->if_psref, ifnet_psref_class); 962 pserialize_read_exit(s); 963 if_attachdomain1(ifp); 964 s = pserialize_read_enter(); 965 psref_release(&psref, &ifp->if_psref, ifnet_psref_class); 966 } 967 pserialize_read_exit(s); 968 curlwp_bindx(bound); 969 } 970 971 static void 972 if_attachdomain1(struct ifnet *ifp) 973 { 974 struct domain *dp; 975 int s; 976 977 s = splnet(); 978 979 /* address family dependent data region */ 980 memset(ifp->if_afdata, 0, sizeof(ifp->if_afdata)); 981 DOMAIN_FOREACH(dp) { 982 if (dp->dom_ifattach != NULL) 983 ifp->if_afdata[dp->dom_family] = 984 (*dp->dom_ifattach)(ifp); 985 } 986 987 splx(s); 988 } 989 990 /* 991 * Deactivate an interface. This points all of the procedure 992 * handles at error stubs. May be called from interrupt context. 993 */ 994 void 995 if_deactivate(struct ifnet *ifp) 996 { 997 int s; 998 999 s = splnet(); 1000 1001 ifp->if_output = if_nulloutput; 1002 ifp->_if_input = if_nullinput; 1003 ifp->if_start = if_nullstart; 1004 ifp->if_transmit = if_nulltransmit; 1005 ifp->if_ioctl = if_nullioctl; 1006 ifp->if_init = if_nullinit; 1007 ifp->if_stop = if_nullstop; 1008 ifp->if_slowtimo = if_nullslowtimo; 1009 ifp->if_drain = if_nulldrain; 1010 1011 /* No more packets may be enqueued. */ 1012 ifp->if_snd.ifq_maxlen = 0; 1013 1014 splx(s); 1015 } 1016 1017 bool 1018 if_is_deactivated(struct ifnet *ifp) 1019 { 1020 1021 return ifp->if_output == if_nulloutput; 1022 } 1023 1024 void 1025 if_purgeaddrs(struct ifnet *ifp, int family, void (*purgeaddr)(struct ifaddr *)) 1026 { 1027 struct ifaddr *ifa, *nifa; 1028 1029 IFADDR_FOREACH_SAFE(ifa, ifp, nifa) { 1030 if (ifa->ifa_addr->sa_family != family) 1031 continue; 1032 (*purgeaddr)(ifa); 1033 } 1034 } 1035 1036 #ifdef IFAREF_DEBUG 1037 static struct ifaddr **ifa_list; 1038 static int ifa_list_size; 1039 1040 /* Depends on only one if_attach runs at once */ 1041 static void 1042 if_build_ifa_list(struct ifnet *ifp) 1043 { 1044 struct ifaddr *ifa; 1045 int i; 1046 1047 KASSERT(ifa_list == NULL); 1048 KASSERT(ifa_list_size == 0); 1049 1050 IFADDR_FOREACH(ifa, ifp) 1051 ifa_list_size++; 1052 1053 ifa_list = kmem_alloc(sizeof(*ifa) * ifa_list_size, KM_SLEEP); 1054 if (ifa_list == NULL) 1055 return; 1056 1057 i = 0; 1058 IFADDR_FOREACH(ifa, ifp) { 1059 ifa_list[i++] = ifa; 1060 ifaref(ifa); 1061 } 1062 } 1063 1064 static void 1065 if_check_and_free_ifa_list(struct ifnet *ifp) 1066 { 1067 int i; 1068 struct ifaddr *ifa; 1069 1070 if (ifa_list == NULL) 1071 return; 1072 1073 for (i = 0; i < ifa_list_size; i++) { 1074 char buf[64]; 1075 1076 ifa = ifa_list[i]; 1077 sockaddr_format(ifa->ifa_addr, buf, sizeof(buf)); 1078 if (ifa->ifa_refcnt > 1) { 1079 log(LOG_WARNING, 1080 "ifa(%s) still referenced (refcnt=%d)\n", 1081 buf, ifa->ifa_refcnt - 1); 1082 } else 1083 log(LOG_DEBUG, 1084 "ifa(%s) not referenced (refcnt=%d)\n", 1085 buf, ifa->ifa_refcnt - 1); 1086 ifafree(ifa); 1087 } 1088 1089 kmem_free(ifa_list, sizeof(*ifa) * ifa_list_size); 1090 ifa_list = NULL; 1091 ifa_list_size = 0; 1092 } 1093 #endif 1094 1095 /* 1096 * Detach an interface from the list of "active" interfaces, 1097 * freeing any resources as we go along. 1098 * 1099 * NOTE: This routine must be called with a valid thread context, 1100 * as it may block. 1101 */ 1102 void 1103 if_detach(struct ifnet *ifp) 1104 { 1105 struct socket so; 1106 struct ifaddr *ifa; 1107 #ifdef IFAREF_DEBUG 1108 struct ifaddr *last_ifa = NULL; 1109 #endif 1110 struct domain *dp; 1111 const struct protosw *pr; 1112 int s, i, family, purged; 1113 uint64_t xc; 1114 1115 #ifdef IFAREF_DEBUG 1116 if_build_ifa_list(ifp); 1117 #endif 1118 /* 1119 * XXX It's kind of lame that we have to have the 1120 * XXX socket structure... 1121 */ 1122 memset(&so, 0, sizeof(so)); 1123 1124 s = splnet(); 1125 1126 sysctl_teardown(&ifp->if_sysctl_log); 1127 mutex_enter(ifp->if_ioctl_lock); 1128 if_deactivate(ifp); 1129 mutex_exit(ifp->if_ioctl_lock); 1130 1131 IFNET_LOCK(); 1132 ifindex2ifnet[ifp->if_index] = NULL; 1133 TAILQ_REMOVE(&ifnet_list, ifp, if_list); 1134 IFNET_WRITER_REMOVE(ifp); 1135 pserialize_perform(ifnet_psz); 1136 IFNET_UNLOCK(); 1137 1138 mutex_obj_free(ifp->if_ioctl_lock); 1139 ifp->if_ioctl_lock = NULL; 1140 1141 if (ifp->if_slowtimo != NULL && ifp->if_slowtimo_ch != NULL) { 1142 ifp->if_slowtimo = NULL; 1143 callout_halt(ifp->if_slowtimo_ch, NULL); 1144 callout_destroy(ifp->if_slowtimo_ch); 1145 kmem_free(ifp->if_slowtimo_ch, sizeof(*ifp->if_slowtimo_ch)); 1146 } 1147 1148 /* 1149 * Do an if_down() to give protocols a chance to do something. 1150 */ 1151 if_down(ifp); 1152 1153 #ifdef ALTQ 1154 if (ALTQ_IS_ENABLED(&ifp->if_snd)) 1155 altq_disable(&ifp->if_snd); 1156 if (ALTQ_IS_ATTACHED(&ifp->if_snd)) 1157 altq_detach(&ifp->if_snd); 1158 #endif 1159 1160 if (ifp->if_snd.ifq_lock) 1161 mutex_obj_free(ifp->if_snd.ifq_lock); 1162 1163 #if NCARP > 0 1164 /* Remove the interface from any carp group it is a part of. */ 1165 if (ifp->if_carp != NULL && ifp->if_type != IFT_CARP) 1166 carp_ifdetach(ifp); 1167 #endif 1168 1169 /* 1170 * Rip all the addresses off the interface. This should make 1171 * all of the routes go away. 1172 * 1173 * pr_usrreq calls can remove an arbitrary number of ifaddrs 1174 * from the list, including our "cursor", ifa. For safety, 1175 * and to honor the TAILQ abstraction, I just restart the 1176 * loop after each removal. Note that the loop will exit 1177 * when all of the remaining ifaddrs belong to the AF_LINK 1178 * family. I am counting on the historical fact that at 1179 * least one pr_usrreq in each address domain removes at 1180 * least one ifaddr. 1181 */ 1182 again: 1183 IFADDR_FOREACH(ifa, ifp) { 1184 family = ifa->ifa_addr->sa_family; 1185 #ifdef IFAREF_DEBUG 1186 printf("if_detach: ifaddr %p, family %d, refcnt %d\n", 1187 ifa, family, ifa->ifa_refcnt); 1188 if (last_ifa != NULL && ifa == last_ifa) 1189 panic("if_detach: loop detected"); 1190 last_ifa = ifa; 1191 #endif 1192 if (family == AF_LINK) 1193 continue; 1194 dp = pffinddomain(family); 1195 #ifdef DIAGNOSTIC 1196 if (dp == NULL) 1197 panic("if_detach: no domain for AF %d", 1198 family); 1199 #endif 1200 /* 1201 * XXX These PURGEIF calls are redundant with the 1202 * purge-all-families calls below, but are left in for 1203 * now both to make a smaller change, and to avoid 1204 * unplanned interactions with clearing of 1205 * ifp->if_addrlist. 1206 */ 1207 purged = 0; 1208 for (pr = dp->dom_protosw; 1209 pr < dp->dom_protoswNPROTOSW; pr++) { 1210 so.so_proto = pr; 1211 if (pr->pr_usrreqs) { 1212 (void) (*pr->pr_usrreqs->pr_purgeif)(&so, ifp); 1213 purged = 1; 1214 } 1215 } 1216 if (purged == 0) { 1217 /* 1218 * XXX What's really the best thing to do 1219 * XXX here? --thorpej (at) NetBSD.org 1220 */ 1221 printf("if_detach: WARNING: AF %d not purged\n", 1222 family); 1223 ifa_remove(ifp, ifa); 1224 } 1225 goto again; 1226 } 1227 1228 if_free_sadl(ifp); 1229 1230 /* Walk the routing table looking for stragglers. */ 1231 for (i = 0; i <= AF_MAX; i++) { 1232 while (rt_walktree(i, if_rt_walktree, ifp) == ERESTART) 1233 continue; 1234 } 1235 1236 DOMAIN_FOREACH(dp) { 1237 if (dp->dom_ifdetach != NULL && ifp->if_afdata[dp->dom_family]) 1238 { 1239 void *p = ifp->if_afdata[dp->dom_family]; 1240 if (p) { 1241 ifp->if_afdata[dp->dom_family] = NULL; 1242 (*dp->dom_ifdetach)(ifp, p); 1243 } 1244 } 1245 1246 /* 1247 * One would expect multicast memberships (INET and 1248 * INET6) on UDP sockets to be purged by the PURGEIF 1249 * calls above, but if all addresses were removed from 1250 * the interface prior to destruction, the calls will 1251 * not be made (e.g. ppp, for which pppd(8) generally 1252 * removes addresses before destroying the interface). 1253 * Because there is no invariant that multicast 1254 * memberships only exist for interfaces with IPv4 1255 * addresses, we must call PURGEIF regardless of 1256 * addresses. (Protocols which might store ifnet 1257 * pointers are marked with PR_PURGEIF.) 1258 */ 1259 for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) { 1260 so.so_proto = pr; 1261 if (pr->pr_usrreqs && pr->pr_flags & PR_PURGEIF) 1262 (void)(*pr->pr_usrreqs->pr_purgeif)(&so, ifp); 1263 } 1264 } 1265 1266 (void)pfil_run_hooks(if_pfil, 1267 (struct mbuf **)PFIL_IFNET_DETACH, ifp, PFIL_IFNET); 1268 (void)pfil_head_destroy(ifp->if_pfil); 1269 1270 /* Announce that the interface is gone. */ 1271 rt_ifannouncemsg(ifp, IFAN_DEPARTURE); 1272 1273 IF_AFDATA_LOCK_DESTROY(ifp); 1274 1275 if (if_is_link_state_changeable(ifp)) { 1276 softint_disestablish(ifp->if_link_si); 1277 ifp->if_link_si = NULL; 1278 } 1279 1280 /* 1281 * remove packets that came from ifp, from software interrupt queues. 1282 */ 1283 DOMAIN_FOREACH(dp) { 1284 for (i = 0; i < __arraycount(dp->dom_ifqueues); i++) { 1285 struct ifqueue *iq = dp->dom_ifqueues[i]; 1286 if (iq == NULL) 1287 break; 1288 dp->dom_ifqueues[i] = NULL; 1289 if_detach_queues(ifp, iq); 1290 } 1291 } 1292 1293 /* 1294 * IP queues have to be processed separately: net-queue barrier 1295 * ensures that the packets are dequeued while a cross-call will 1296 * ensure that the interrupts have completed. FIXME: not quite.. 1297 */ 1298 #ifdef INET 1299 pktq_barrier(ip_pktq); 1300 #endif 1301 #ifdef INET6 1302 if (in6_present) 1303 pktq_barrier(ip6_pktq); 1304 #endif 1305 xc = xc_broadcast(0, (xcfunc_t)nullop, NULL, NULL); 1306 xc_wait(xc); 1307 1308 /* Wait for all readers to drain before freeing. */ 1309 psref_target_destroy(&ifp->if_psref, ifnet_psref_class); 1310 PSLIST_ENTRY_DESTROY(ifp, if_pslist_entry); 1311 1312 if (ifp->if_percpuq != NULL) { 1313 if_percpuq_destroy(ifp->if_percpuq); 1314 ifp->if_percpuq = NULL; 1315 } 1316 1317 splx(s); 1318 1319 #ifdef IFAREF_DEBUG 1320 if_check_and_free_ifa_list(ifp); 1321 #endif 1322 } 1323 1324 static void 1325 if_detach_queues(struct ifnet *ifp, struct ifqueue *q) 1326 { 1327 struct mbuf *m, *prev, *next; 1328 1329 prev = NULL; 1330 for (m = q->ifq_head; m != NULL; m = next) { 1331 KASSERT((m->m_flags & M_PKTHDR) != 0); 1332 1333 next = m->m_nextpkt; 1334 if (m->m_pkthdr.rcvif_index != ifp->if_index) { 1335 prev = m; 1336 continue; 1337 } 1338 1339 if (prev != NULL) 1340 prev->m_nextpkt = m->m_nextpkt; 1341 else 1342 q->ifq_head = m->m_nextpkt; 1343 if (q->ifq_tail == m) 1344 q->ifq_tail = prev; 1345 q->ifq_len--; 1346 1347 m->m_nextpkt = NULL; 1348 m_freem(m); 1349 IF_DROP(q); 1350 } 1351 } 1352 1353 /* 1354 * Callback for a radix tree walk to delete all references to an 1355 * ifnet. 1356 */ 1357 static int 1358 if_rt_walktree(struct rtentry *rt, void *v) 1359 { 1360 struct ifnet *ifp = (struct ifnet *)v; 1361 int error; 1362 struct rtentry *retrt; 1363 1364 if (rt->rt_ifp != ifp) 1365 return 0; 1366 1367 /* Delete the entry. */ 1368 error = rtrequest(RTM_DELETE, rt_getkey(rt), rt->rt_gateway, 1369 rt_mask(rt), rt->rt_flags, &retrt); 1370 if (error == 0) { 1371 KASSERT(retrt == rt); 1372 KASSERT((retrt->rt_flags & RTF_UP) == 0); 1373 retrt->rt_ifp = NULL; 1374 rtfree(retrt); 1375 } else { 1376 printf("%s: warning: unable to delete rtentry @ %p, " 1377 "error = %d\n", ifp->if_xname, rt, error); 1378 } 1379 return ERESTART; 1380 } 1381 1382 /* 1383 * Create a clone network interface. 1384 */ 1385 static int 1386 if_clone_create(const char *name) 1387 { 1388 struct if_clone *ifc; 1389 int unit; 1390 struct ifnet *ifp; 1391 struct psref psref; 1392 1393 ifc = if_clone_lookup(name, &unit); 1394 if (ifc == NULL) 1395 return EINVAL; 1396 1397 ifp = if_get(name, &psref); 1398 if (ifp != NULL) { 1399 if_put(ifp, &psref); 1400 return EEXIST; 1401 } 1402 1403 return (*ifc->ifc_create)(ifc, unit); 1404 } 1405 1406 /* 1407 * Destroy a clone network interface. 1408 */ 1409 static int 1410 if_clone_destroy(const char *name) 1411 { 1412 struct if_clone *ifc; 1413 struct ifnet *ifp; 1414 struct psref psref; 1415 1416 ifc = if_clone_lookup(name, NULL); 1417 if (ifc == NULL) 1418 return EINVAL; 1419 1420 if (ifc->ifc_destroy == NULL) 1421 return EOPNOTSUPP; 1422 1423 ifp = if_get(name, &psref); 1424 if (ifp == NULL) 1425 return ENXIO; 1426 1427 /* We have to disable ioctls here */ 1428 mutex_enter(ifp->if_ioctl_lock); 1429 ifp->if_ioctl = if_nullioctl; 1430 mutex_exit(ifp->if_ioctl_lock); 1431 1432 /* 1433 * We cannot call ifc_destroy with holding ifp. 1434 * Releasing ifp here is safe thanks to if_clone_mtx. 1435 */ 1436 if_put(ifp, &psref); 1437 1438 return (*ifc->ifc_destroy)(ifp); 1439 } 1440 1441 /* 1442 * Look up a network interface cloner. 1443 */ 1444 static struct if_clone * 1445 if_clone_lookup(const char *name, int *unitp) 1446 { 1447 struct if_clone *ifc; 1448 const char *cp; 1449 char *dp, ifname[IFNAMSIZ + 3]; 1450 int unit; 1451 1452 strcpy(ifname, "if_"); 1453 /* separate interface name from unit */ 1454 for (dp = ifname + 3, cp = name; cp - name < IFNAMSIZ && 1455 *cp && (*cp < '0' || *cp > '9');) 1456 *dp++ = *cp++; 1457 1458 if (cp == name || cp - name == IFNAMSIZ || !*cp) 1459 return NULL; /* No name or unit number */ 1460 *dp++ = '\0'; 1461 1462 again: 1463 LIST_FOREACH(ifc, &if_cloners, ifc_list) { 1464 if (strcmp(ifname + 3, ifc->ifc_name) == 0) 1465 break; 1466 } 1467 1468 if (ifc == NULL) { 1469 if (*ifname == '\0' || 1470 module_autoload(ifname, MODULE_CLASS_DRIVER)) 1471 return NULL; 1472 *ifname = '\0'; 1473 goto again; 1474 } 1475 1476 unit = 0; 1477 while (cp - name < IFNAMSIZ && *cp) { 1478 if (*cp < '0' || *cp > '9' || unit >= INT_MAX / 10) { 1479 /* Bogus unit number. */ 1480 return NULL; 1481 } 1482 unit = (unit * 10) + (*cp++ - '0'); 1483 } 1484 1485 if (unitp != NULL) 1486 *unitp = unit; 1487 return ifc; 1488 } 1489 1490 /* 1491 * Register a network interface cloner. 1492 */ 1493 void 1494 if_clone_attach(struct if_clone *ifc) 1495 { 1496 1497 LIST_INSERT_HEAD(&if_cloners, ifc, ifc_list); 1498 if_cloners_count++; 1499 } 1500 1501 /* 1502 * Unregister a network interface cloner. 1503 */ 1504 void 1505 if_clone_detach(struct if_clone *ifc) 1506 { 1507 1508 LIST_REMOVE(ifc, ifc_list); 1509 if_cloners_count--; 1510 } 1511 1512 /* 1513 * Provide list of interface cloners to userspace. 1514 */ 1515 int 1516 if_clone_list(int buf_count, char *buffer, int *total) 1517 { 1518 char outbuf[IFNAMSIZ], *dst; 1519 struct if_clone *ifc; 1520 int count, error = 0; 1521 1522 *total = if_cloners_count; 1523 if ((dst = buffer) == NULL) { 1524 /* Just asking how many there are. */ 1525 return 0; 1526 } 1527 1528 if (buf_count < 0) 1529 return EINVAL; 1530 1531 count = (if_cloners_count < buf_count) ? 1532 if_cloners_count : buf_count; 1533 1534 for (ifc = LIST_FIRST(&if_cloners); ifc != NULL && count != 0; 1535 ifc = LIST_NEXT(ifc, ifc_list), count--, dst += IFNAMSIZ) { 1536 (void)strncpy(outbuf, ifc->ifc_name, sizeof(outbuf)); 1537 if (outbuf[sizeof(outbuf) - 1] != '\0') 1538 return ENAMETOOLONG; 1539 error = copyout(outbuf, dst, sizeof(outbuf)); 1540 if (error != 0) 1541 break; 1542 } 1543 1544 return error; 1545 } 1546 1547 void 1548 ifaref(struct ifaddr *ifa) 1549 { 1550 ifa->ifa_refcnt++; 1551 } 1552 1553 void 1554 ifafree(struct ifaddr *ifa) 1555 { 1556 KASSERT(ifa != NULL); 1557 KASSERT(ifa->ifa_refcnt > 0); 1558 1559 if (--ifa->ifa_refcnt == 0) { 1560 free(ifa, M_IFADDR); 1561 } 1562 } 1563 1564 void 1565 ifa_insert(struct ifnet *ifp, struct ifaddr *ifa) 1566 { 1567 ifa->ifa_ifp = ifp; 1568 TAILQ_INSERT_TAIL(&ifp->if_addrlist, ifa, ifa_list); 1569 ifaref(ifa); 1570 } 1571 1572 void 1573 ifa_remove(struct ifnet *ifp, struct ifaddr *ifa) 1574 { 1575 KASSERT(ifa->ifa_ifp == ifp); 1576 TAILQ_REMOVE(&ifp->if_addrlist, ifa, ifa_list); 1577 ifafree(ifa); 1578 } 1579 1580 static inline int 1581 equal(const struct sockaddr *sa1, const struct sockaddr *sa2) 1582 { 1583 return sockaddr_cmp(sa1, sa2) == 0; 1584 } 1585 1586 /* 1587 * Locate an interface based on a complete address. 1588 */ 1589 /*ARGSUSED*/ 1590 struct ifaddr * 1591 ifa_ifwithaddr(const struct sockaddr *addr) 1592 { 1593 struct ifnet *ifp; 1594 struct ifaddr *ifa; 1595 int s; 1596 1597 s = pserialize_read_enter(); 1598 IFNET_READER_FOREACH(ifp) { 1599 if (if_is_deactivated(ifp)) 1600 continue; 1601 IFADDR_FOREACH(ifa, ifp) { 1602 if (ifa->ifa_addr->sa_family != addr->sa_family) 1603 continue; 1604 if (equal(addr, ifa->ifa_addr)) 1605 return ifa; 1606 if ((ifp->if_flags & IFF_BROADCAST) && 1607 ifa->ifa_broadaddr && 1608 /* IP6 doesn't have broadcast */ 1609 ifa->ifa_broadaddr->sa_len != 0 && 1610 equal(ifa->ifa_broadaddr, addr)) 1611 return ifa; 1612 } 1613 } 1614 pserialize_read_exit(s); 1615 return NULL; 1616 } 1617 1618 /* 1619 * Locate the point to point interface with a given destination address. 1620 */ 1621 /*ARGSUSED*/ 1622 struct ifaddr * 1623 ifa_ifwithdstaddr(const struct sockaddr *addr) 1624 { 1625 struct ifnet *ifp; 1626 struct ifaddr *ifa; 1627 int s; 1628 1629 s = pserialize_read_enter(); 1630 IFNET_READER_FOREACH(ifp) { 1631 if (if_is_deactivated(ifp)) 1632 continue; 1633 if ((ifp->if_flags & IFF_POINTOPOINT) == 0) 1634 continue; 1635 IFADDR_FOREACH(ifa, ifp) { 1636 if (ifa->ifa_addr->sa_family != addr->sa_family || 1637 ifa->ifa_dstaddr == NULL) 1638 continue; 1639 if (equal(addr, ifa->ifa_dstaddr)) 1640 return ifa; 1641 } 1642 } 1643 pserialize_read_exit(s); 1644 return NULL; 1645 } 1646 1647 /* 1648 * Find an interface on a specific network. If many, choice 1649 * is most specific found. 1650 */ 1651 struct ifaddr * 1652 ifa_ifwithnet(const struct sockaddr *addr) 1653 { 1654 struct ifnet *ifp; 1655 struct ifaddr *ifa; 1656 const struct sockaddr_dl *sdl; 1657 struct ifaddr *ifa_maybe = 0; 1658 u_int af = addr->sa_family; 1659 const char *addr_data = addr->sa_data, *cplim; 1660 int s; 1661 1662 if (af == AF_LINK) { 1663 sdl = satocsdl(addr); 1664 if (sdl->sdl_index && sdl->sdl_index < if_indexlim && 1665 ifindex2ifnet[sdl->sdl_index] && 1666 !if_is_deactivated(ifindex2ifnet[sdl->sdl_index])) { 1667 return ifindex2ifnet[sdl->sdl_index]->if_dl; 1668 } 1669 } 1670 #ifdef NETATALK 1671 if (af == AF_APPLETALK) { 1672 const struct sockaddr_at *sat, *sat2; 1673 sat = (const struct sockaddr_at *)addr; 1674 s = pserialize_read_enter(); 1675 IFNET_READER_FOREACH(ifp) { 1676 if (if_is_deactivated(ifp)) 1677 continue; 1678 ifa = at_ifawithnet((const struct sockaddr_at *)addr, ifp); 1679 if (ifa == NULL) 1680 continue; 1681 sat2 = (struct sockaddr_at *)ifa->ifa_addr; 1682 if (sat2->sat_addr.s_net == sat->sat_addr.s_net) 1683 return ifa; /* exact match */ 1684 if (ifa_maybe == NULL) { 1685 /* else keep the if with the right range */ 1686 ifa_maybe = ifa; 1687 } 1688 } 1689 pserialize_read_exit(s); 1690 return ifa_maybe; 1691 } 1692 #endif 1693 s = pserialize_read_enter(); 1694 IFNET_READER_FOREACH(ifp) { 1695 if (if_is_deactivated(ifp)) 1696 continue; 1697 IFADDR_FOREACH(ifa, ifp) { 1698 const char *cp, *cp2, *cp3; 1699 1700 if (ifa->ifa_addr->sa_family != af || 1701 ifa->ifa_netmask == NULL) 1702 next: continue; 1703 cp = addr_data; 1704 cp2 = ifa->ifa_addr->sa_data; 1705 cp3 = ifa->ifa_netmask->sa_data; 1706 cplim = (const char *)ifa->ifa_netmask + 1707 ifa->ifa_netmask->sa_len; 1708 while (cp3 < cplim) { 1709 if ((*cp++ ^ *cp2++) & *cp3++) { 1710 /* want to continue for() loop */ 1711 goto next; 1712 } 1713 } 1714 if (ifa_maybe == NULL || 1715 rt_refines(ifa->ifa_netmask, 1716 ifa_maybe->ifa_netmask)) 1717 ifa_maybe = ifa; 1718 } 1719 } 1720 pserialize_read_exit(s); 1721 return ifa_maybe; 1722 } 1723 1724 /* 1725 * Find the interface of the addresss. 1726 */ 1727 struct ifaddr * 1728 ifa_ifwithladdr(const struct sockaddr *addr) 1729 { 1730 struct ifaddr *ia; 1731 1732 if ((ia = ifa_ifwithaddr(addr)) || (ia = ifa_ifwithdstaddr(addr)) || 1733 (ia = ifa_ifwithnet(addr))) 1734 return ia; 1735 return NULL; 1736 } 1737 1738 /* 1739 * Find an interface using a specific address family 1740 */ 1741 struct ifaddr * 1742 ifa_ifwithaf(int af) 1743 { 1744 struct ifnet *ifp; 1745 struct ifaddr *ifa = NULL; 1746 int s; 1747 1748 s = pserialize_read_enter(); 1749 IFNET_READER_FOREACH(ifp) { 1750 if (if_is_deactivated(ifp)) 1751 continue; 1752 IFADDR_FOREACH(ifa, ifp) { 1753 if (ifa->ifa_addr->sa_family == af) 1754 goto out; 1755 } 1756 } 1757 out: 1758 pserialize_read_exit(s); 1759 return ifa; 1760 } 1761 1762 /* 1763 * Find an interface address specific to an interface best matching 1764 * a given address. 1765 */ 1766 struct ifaddr * 1767 ifaof_ifpforaddr(const struct sockaddr *addr, struct ifnet *ifp) 1768 { 1769 struct ifaddr *ifa; 1770 const char *cp, *cp2, *cp3; 1771 const char *cplim; 1772 struct ifaddr *ifa_maybe = 0; 1773 u_int af = addr->sa_family; 1774 1775 if (if_is_deactivated(ifp)) 1776 return NULL; 1777 1778 if (af >= AF_MAX) 1779 return NULL; 1780 1781 IFADDR_FOREACH(ifa, ifp) { 1782 if (ifa->ifa_addr->sa_family != af) 1783 continue; 1784 ifa_maybe = ifa; 1785 if (ifa->ifa_netmask == NULL) { 1786 if (equal(addr, ifa->ifa_addr) || 1787 (ifa->ifa_dstaddr && 1788 equal(addr, ifa->ifa_dstaddr))) 1789 return ifa; 1790 continue; 1791 } 1792 cp = addr->sa_data; 1793 cp2 = ifa->ifa_addr->sa_data; 1794 cp3 = ifa->ifa_netmask->sa_data; 1795 cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask; 1796 for (; cp3 < cplim; cp3++) { 1797 if ((*cp++ ^ *cp2++) & *cp3) 1798 break; 1799 } 1800 if (cp3 == cplim) 1801 return ifa; 1802 } 1803 return ifa_maybe; 1804 } 1805 1806 /* 1807 * Default action when installing a route with a Link Level gateway. 1808 * Lookup an appropriate real ifa to point to. 1809 * This should be moved to /sys/net/link.c eventually. 1810 */ 1811 void 1812 link_rtrequest(int cmd, struct rtentry *rt, const struct rt_addrinfo *info) 1813 { 1814 struct ifaddr *ifa; 1815 const struct sockaddr *dst; 1816 struct ifnet *ifp; 1817 1818 if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL || 1819 (ifp = ifa->ifa_ifp) == NULL || (dst = rt_getkey(rt)) == NULL) 1820 return; 1821 if ((ifa = ifaof_ifpforaddr(dst, ifp)) != NULL) { 1822 rt_replace_ifa(rt, ifa); 1823 if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest) 1824 ifa->ifa_rtrequest(cmd, rt, info); 1825 } 1826 } 1827 1828 /* 1829 * bitmask macros to manage a densely packed link_state change queue. 1830 * Because we need to store LINK_STATE_UNKNOWN(0), LINK_STATE_DOWN(1) and 1831 * LINK_STATE_UP(2) we need 2 bits for each state change. 1832 * As a state change to store is 0, treat all bits set as an unset item. 1833 */ 1834 #define LQ_ITEM_BITS 2 1835 #define LQ_ITEM_MASK ((1 << LQ_ITEM_BITS) - 1) 1836 #define LQ_MASK(i) (LQ_ITEM_MASK << (i) * LQ_ITEM_BITS) 1837 #define LINK_STATE_UNSET LQ_ITEM_MASK 1838 #define LQ_ITEM(q, i) (((q) & LQ_MASK((i))) >> (i) * LQ_ITEM_BITS) 1839 #define LQ_STORE(q, i, v) \ 1840 do { \ 1841 (q) &= ~LQ_MASK((i)); \ 1842 (q) |= (v) << (i) * LQ_ITEM_BITS; \ 1843 } while (0 /* CONSTCOND */) 1844 #define LQ_MAX(q) ((sizeof((q)) * NBBY) / LQ_ITEM_BITS) 1845 #define LQ_POP(q, v) \ 1846 do { \ 1847 (v) = LQ_ITEM((q), 0); \ 1848 (q) >>= LQ_ITEM_BITS; \ 1849 (q) |= LINK_STATE_UNSET << (LQ_MAX((q)) - 1) * LQ_ITEM_BITS; \ 1850 } while (0 /* CONSTCOND */) 1851 #define LQ_PUSH(q, v) \ 1852 do { \ 1853 (q) >>= LQ_ITEM_BITS; \ 1854 (q) |= (v) << (LQ_MAX((q)) - 1) * LQ_ITEM_BITS; \ 1855 } while (0 /* CONSTCOND */) 1856 #define LQ_FIND_UNSET(q, i) \ 1857 for ((i) = 0; i < LQ_MAX((q)); (i)++) { \ 1858 if (LQ_ITEM((q), (i)) == LINK_STATE_UNSET) \ 1859 break; \ 1860 } 1861 /* 1862 * Handle a change in the interface link state and 1863 * queue notifications. 1864 */ 1865 void 1866 if_link_state_change(struct ifnet *ifp, int link_state) 1867 { 1868 int s, idx; 1869 1870 KASSERTMSG(if_is_link_state_changeable(ifp), 1871 "%s: IFEF_NO_LINK_STATE_CHANGE must not be set, but if_extflags=0x%x", 1872 ifp->if_xname, ifp->if_extflags); 1873 1874 /* Ensure change is to a valid state */ 1875 switch (link_state) { 1876 case LINK_STATE_UNKNOWN: /* FALLTHROUGH */ 1877 case LINK_STATE_DOWN: /* FALLTHROUGH */ 1878 case LINK_STATE_UP: 1879 break; 1880 default: 1881 #ifdef DEBUG 1882 printf("%s: invalid link state %d\n", 1883 ifp->if_xname, link_state); 1884 #endif 1885 return; 1886 } 1887 1888 s = splnet(); 1889 1890 /* Find the last unset event in the queue. */ 1891 LQ_FIND_UNSET(ifp->if_link_queue, idx); 1892 1893 /* 1894 * Ensure link_state doesn't match the last event in the queue. 1895 * ifp->if_link_state is not checked and set here because 1896 * that would present an inconsistent picture to the system. 1897 */ 1898 if (idx != 0 && 1899 LQ_ITEM(ifp->if_link_queue, idx - 1) == (uint8_t)link_state) 1900 goto out; 1901 1902 /* Handle queue overflow. */ 1903 if (idx == LQ_MAX(ifp->if_link_queue)) { 1904 uint8_t lost; 1905 1906 /* 1907 * The DOWN state must be protected from being pushed off 1908 * the queue to ensure that userland will always be 1909 * in a sane state. 1910 * Because DOWN is protected, there is no need to protect 1911 * UNKNOWN. 1912 * It should be invalid to change from any other state to 1913 * UNKNOWN anyway ... 1914 */ 1915 lost = LQ_ITEM(ifp->if_link_queue, 0); 1916 LQ_PUSH(ifp->if_link_queue, (uint8_t)link_state); 1917 if (lost == LINK_STATE_DOWN) { 1918 lost = LQ_ITEM(ifp->if_link_queue, 0); 1919 LQ_STORE(ifp->if_link_queue, 0, LINK_STATE_DOWN); 1920 } 1921 printf("%s: lost link state change %s\n", 1922 ifp->if_xname, 1923 lost == LINK_STATE_UP ? "UP" : 1924 lost == LINK_STATE_DOWN ? "DOWN" : 1925 "UNKNOWN"); 1926 } else 1927 LQ_STORE(ifp->if_link_queue, idx, (uint8_t)link_state); 1928 1929 softint_schedule(ifp->if_link_si); 1930 1931 out: 1932 splx(s); 1933 } 1934 1935 /* 1936 * Handle interface link state change notifications. 1937 * Must be called at splnet(). 1938 */ 1939 static void 1940 if_link_state_change0(struct ifnet *ifp, int link_state) 1941 { 1942 struct domain *dp; 1943 1944 /* Ensure the change is still valid. */ 1945 if (ifp->if_link_state == link_state) 1946 return; 1947 1948 #ifdef DEBUG 1949 log(LOG_DEBUG, "%s: link state %s (was %s)\n", ifp->if_xname, 1950 link_state == LINK_STATE_UP ? "UP" : 1951 link_state == LINK_STATE_DOWN ? "DOWN" : 1952 "UNKNOWN", 1953 ifp->if_link_state == LINK_STATE_UP ? "UP" : 1954 ifp->if_link_state == LINK_STATE_DOWN ? "DOWN" : 1955 "UNKNOWN"); 1956 #endif 1957 1958 /* 1959 * When going from UNKNOWN to UP, we need to mark existing 1960 * addresses as tentative and restart DAD as we may have 1961 * erroneously not found a duplicate. 1962 * 1963 * This needs to happen before rt_ifmsg to avoid a race where 1964 * listeners would have an address and expect it to work right 1965 * away. 1966 */ 1967 if (link_state == LINK_STATE_UP && 1968 ifp->if_link_state == LINK_STATE_UNKNOWN) 1969 { 1970 DOMAIN_FOREACH(dp) { 1971 if (dp->dom_if_link_state_change != NULL) 1972 dp->dom_if_link_state_change(ifp, 1973 LINK_STATE_DOWN); 1974 } 1975 } 1976 1977 ifp->if_link_state = link_state; 1978 1979 /* Notify that the link state has changed. */ 1980 rt_ifmsg(ifp); 1981 1982 #if NCARP > 0 1983 if (ifp->if_carp) 1984 carp_carpdev_state(ifp); 1985 #endif 1986 1987 DOMAIN_FOREACH(dp) { 1988 if (dp->dom_if_link_state_change != NULL) 1989 dp->dom_if_link_state_change(ifp, link_state); 1990 } 1991 } 1992 1993 /* 1994 * Process the interface link state change queue. 1995 */ 1996 static void 1997 if_link_state_change_si(void *arg) 1998 { 1999 struct ifnet *ifp = arg; 2000 int s; 2001 uint8_t state; 2002 2003 s = splnet(); 2004 2005 /* Pop a link state change from the queue and process it. */ 2006 LQ_POP(ifp->if_link_queue, state); 2007 if_link_state_change0(ifp, state); 2008 2009 /* If there is a link state change to come, schedule it. */ 2010 if (LQ_ITEM(ifp->if_link_queue, 0) != LINK_STATE_UNSET) 2011 softint_schedule(ifp->if_link_si); 2012 2013 splx(s); 2014 } 2015 2016 /* 2017 * Default action when installing a local route on a point-to-point 2018 * interface. 2019 */ 2020 void 2021 p2p_rtrequest(int req, struct rtentry *rt, 2022 __unused const struct rt_addrinfo *info) 2023 { 2024 struct ifnet *ifp = rt->rt_ifp; 2025 struct ifaddr *ifa, *lo0ifa; 2026 2027 switch (req) { 2028 case RTM_ADD: 2029 if ((rt->rt_flags & RTF_LOCAL) == 0) 2030 break; 2031 2032 rt->rt_ifp = lo0ifp; 2033 2034 IFADDR_FOREACH(ifa, ifp) { 2035 if (equal(rt_getkey(rt), ifa->ifa_addr)) 2036 break; 2037 } 2038 if (ifa == NULL) 2039 break; 2040 2041 /* 2042 * Ensure lo0 has an address of the same family. 2043 */ 2044 IFADDR_FOREACH(lo0ifa, lo0ifp) { 2045 if (lo0ifa->ifa_addr->sa_family == 2046 ifa->ifa_addr->sa_family) 2047 break; 2048 } 2049 if (lo0ifa == NULL) 2050 break; 2051 2052 /* 2053 * Make sure to set rt->rt_ifa to the interface 2054 * address we are using, otherwise we will have trouble 2055 * with source address selection. 2056 */ 2057 if (ifa != rt->rt_ifa) 2058 rt_replace_ifa(rt, ifa); 2059 break; 2060 case RTM_DELETE: 2061 default: 2062 break; 2063 } 2064 } 2065 2066 /* 2067 * Mark an interface down and notify protocols of 2068 * the transition. 2069 * NOTE: must be called at splsoftnet or equivalent. 2070 */ 2071 void 2072 if_down(struct ifnet *ifp) 2073 { 2074 struct ifaddr *ifa; 2075 struct domain *dp; 2076 2077 ifp->if_flags &= ~IFF_UP; 2078 nanotime(&ifp->if_lastchange); 2079 IFADDR_FOREACH(ifa, ifp) 2080 pfctlinput(PRC_IFDOWN, ifa->ifa_addr); 2081 IFQ_PURGE(&ifp->if_snd); 2082 #if NCARP > 0 2083 if (ifp->if_carp) 2084 carp_carpdev_state(ifp); 2085 #endif 2086 rt_ifmsg(ifp); 2087 DOMAIN_FOREACH(dp) { 2088 if (dp->dom_if_down) 2089 dp->dom_if_down(ifp); 2090 } 2091 } 2092 2093 /* 2094 * Mark an interface up and notify protocols of 2095 * the transition. 2096 * NOTE: must be called at splsoftnet or equivalent. 2097 */ 2098 void 2099 if_up(struct ifnet *ifp) 2100 { 2101 #ifdef notyet 2102 struct ifaddr *ifa; 2103 #endif 2104 struct domain *dp; 2105 2106 ifp->if_flags |= IFF_UP; 2107 nanotime(&ifp->if_lastchange); 2108 #ifdef notyet 2109 /* this has no effect on IP, and will kill all ISO connections XXX */ 2110 IFADDR_FOREACH(ifa, ifp) 2111 pfctlinput(PRC_IFUP, ifa->ifa_addr); 2112 #endif 2113 #if NCARP > 0 2114 if (ifp->if_carp) 2115 carp_carpdev_state(ifp); 2116 #endif 2117 rt_ifmsg(ifp); 2118 DOMAIN_FOREACH(dp) { 2119 if (dp->dom_if_up) 2120 dp->dom_if_up(ifp); 2121 } 2122 } 2123 2124 /* 2125 * Handle interface slowtimo timer routine. Called 2126 * from softclock, we decrement timer (if set) and 2127 * call the appropriate interface routine on expiration. 2128 */ 2129 static void 2130 if_slowtimo(void *arg) 2131 { 2132 void (*slowtimo)(struct ifnet *); 2133 struct ifnet *ifp = arg; 2134 int s; 2135 2136 slowtimo = ifp->if_slowtimo; 2137 if (__predict_false(slowtimo == NULL)) 2138 return; 2139 2140 s = splnet(); 2141 if (ifp->if_timer != 0 && --ifp->if_timer == 0) 2142 (*slowtimo)(ifp); 2143 2144 splx(s); 2145 2146 if (__predict_true(ifp->if_slowtimo != NULL)) 2147 callout_schedule(ifp->if_slowtimo_ch, hz / IFNET_SLOWHZ); 2148 } 2149 2150 /* 2151 * Set/clear promiscuous mode on interface ifp based on the truth value 2152 * of pswitch. The calls are reference counted so that only the first 2153 * "on" request actually has an effect, as does the final "off" request. 2154 * Results are undefined if the "off" and "on" requests are not matched. 2155 */ 2156 int 2157 ifpromisc(struct ifnet *ifp, int pswitch) 2158 { 2159 int pcount, ret; 2160 short nflags; 2161 2162 pcount = ifp->if_pcount; 2163 if (pswitch) { 2164 /* 2165 * Allow the device to be "placed" into promiscuous 2166 * mode even if it is not configured up. It will 2167 * consult IFF_PROMISC when it is brought up. 2168 */ 2169 if (ifp->if_pcount++ != 0) 2170 return 0; 2171 nflags = ifp->if_flags | IFF_PROMISC; 2172 } else { 2173 if (--ifp->if_pcount > 0) 2174 return 0; 2175 nflags = ifp->if_flags & ~IFF_PROMISC; 2176 } 2177 ret = if_flags_set(ifp, nflags); 2178 /* Restore interface state if not successful. */ 2179 if (ret != 0) { 2180 ifp->if_pcount = pcount; 2181 } 2182 return ret; 2183 } 2184 2185 /* 2186 * Map interface name to 2187 * interface structure pointer. 2188 */ 2189 struct ifnet * 2190 ifunit(const char *name) 2191 { 2192 struct ifnet *ifp; 2193 const char *cp = name; 2194 u_int unit = 0; 2195 u_int i; 2196 int s; 2197 2198 /* 2199 * If the entire name is a number, treat it as an ifindex. 2200 */ 2201 for (i = 0; i < IFNAMSIZ && *cp >= '0' && *cp <= '9'; i++, cp++) { 2202 unit = unit * 10 + (*cp - '0'); 2203 } 2204 2205 /* 2206 * If the number took all of the name, then it's a valid ifindex. 2207 */ 2208 if (i == IFNAMSIZ || (cp != name && *cp == '\0')) { 2209 if (unit >= if_indexlim) 2210 return NULL; 2211 ifp = ifindex2ifnet[unit]; 2212 if (ifp == NULL || if_is_deactivated(ifp)) 2213 return NULL; 2214 return ifp; 2215 } 2216 2217 ifp = NULL; 2218 s = pserialize_read_enter(); 2219 IFNET_READER_FOREACH(ifp) { 2220 if (if_is_deactivated(ifp)) 2221 continue; 2222 if (strcmp(ifp->if_xname, name) == 0) 2223 goto out; 2224 } 2225 out: 2226 pserialize_read_exit(s); 2227 return ifp; 2228 } 2229 2230 /* 2231 * Get a reference of an ifnet object by an interface name. 2232 * The returned reference is protected by psref(9). The caller 2233 * must release a returned reference by if_put after use. 2234 */ 2235 struct ifnet * 2236 if_get(const char *name, struct psref *psref) 2237 { 2238 struct ifnet *ifp; 2239 const char *cp = name; 2240 u_int unit = 0; 2241 u_int i; 2242 int s; 2243 2244 /* 2245 * If the entire name is a number, treat it as an ifindex. 2246 */ 2247 for (i = 0; i < IFNAMSIZ && *cp >= '0' && *cp <= '9'; i++, cp++) { 2248 unit = unit * 10 + (*cp - '0'); 2249 } 2250 2251 /* 2252 * If the number took all of the name, then it's a valid ifindex. 2253 */ 2254 if (i == IFNAMSIZ || (cp != name && *cp == '\0')) { 2255 if (unit >= if_indexlim) 2256 return NULL; 2257 ifp = ifindex2ifnet[unit]; 2258 if (ifp == NULL || if_is_deactivated(ifp)) 2259 return NULL; 2260 return ifp; 2261 } 2262 2263 ifp = NULL; 2264 s = pserialize_read_enter(); 2265 IFNET_READER_FOREACH(ifp) { 2266 if (if_is_deactivated(ifp)) 2267 continue; 2268 if (strcmp(ifp->if_xname, name) == 0) { 2269 psref_acquire(psref, &ifp->if_psref, 2270 ifnet_psref_class); 2271 goto out; 2272 } 2273 } 2274 out: 2275 pserialize_read_exit(s); 2276 return ifp; 2277 } 2278 2279 /* 2280 * Release a reference of an ifnet object given by if_get or 2281 * if_get_byindex. 2282 */ 2283 void 2284 if_put(const struct ifnet *ifp, struct psref *psref) 2285 { 2286 2287 if (ifp == NULL) 2288 return; 2289 2290 psref_release(psref, &ifp->if_psref, ifnet_psref_class); 2291 } 2292 2293 ifnet_t * 2294 if_byindex(u_int idx) 2295 { 2296 return (idx < if_indexlim) ? ifindex2ifnet[idx] : NULL; 2297 } 2298 2299 /* 2300 * Get a reference of an ifnet object by an interface index. 2301 * The returned reference is protected by psref(9). The caller 2302 * must release a returned reference by if_put after use. 2303 */ 2304 ifnet_t * 2305 if_get_byindex(u_int idx, struct psref *psref) 2306 { 2307 ifnet_t *ifp; 2308 int s; 2309 2310 s = pserialize_read_enter(); 2311 ifp = (__predict_true(idx < if_indexlim)) ? ifindex2ifnet[idx] : NULL; 2312 if (__predict_true(ifp != NULL)) 2313 psref_acquire(psref, &ifp->if_psref, ifnet_psref_class); 2314 pserialize_read_exit(s); 2315 2316 return ifp; 2317 } 2318 2319 /* 2320 * XXX it's safe only if the passed ifp is guaranteed to not be freed, 2321 * for example the ifp is already held or some other object is held which 2322 * guarantes the ifp to not be freed indirectly. 2323 */ 2324 void 2325 if_acquire_NOMPSAFE(struct ifnet *ifp, struct psref *psref) 2326 { 2327 2328 KASSERT(ifp->if_index != 0); 2329 psref_acquire(psref, &ifp->if_psref, ifnet_psref_class); 2330 } 2331 2332 bool 2333 if_held(struct ifnet *ifp) 2334 { 2335 2336 return psref_held(&ifp->if_psref, ifnet_psref_class); 2337 } 2338 2339 2340 /* common */ 2341 int 2342 ifioctl_common(struct ifnet *ifp, u_long cmd, void *data) 2343 { 2344 int s; 2345 struct ifreq *ifr; 2346 struct ifcapreq *ifcr; 2347 struct ifdatareq *ifdr; 2348 2349 switch (cmd) { 2350 case SIOCSIFCAP: 2351 ifcr = data; 2352 if ((ifcr->ifcr_capenable & ~ifp->if_capabilities) != 0) 2353 return EINVAL; 2354 2355 if (ifcr->ifcr_capenable == ifp->if_capenable) 2356 return 0; 2357 2358 ifp->if_capenable = ifcr->ifcr_capenable; 2359 2360 /* Pre-compute the checksum flags mask. */ 2361 ifp->if_csum_flags_tx = 0; 2362 ifp->if_csum_flags_rx = 0; 2363 if (ifp->if_capenable & IFCAP_CSUM_IPv4_Tx) { 2364 ifp->if_csum_flags_tx |= M_CSUM_IPv4; 2365 } 2366 if (ifp->if_capenable & IFCAP_CSUM_IPv4_Rx) { 2367 ifp->if_csum_flags_rx |= M_CSUM_IPv4; 2368 } 2369 2370 if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Tx) { 2371 ifp->if_csum_flags_tx |= M_CSUM_TCPv4; 2372 } 2373 if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Rx) { 2374 ifp->if_csum_flags_rx |= M_CSUM_TCPv4; 2375 } 2376 2377 if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Tx) { 2378 ifp->if_csum_flags_tx |= M_CSUM_UDPv4; 2379 } 2380 if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Rx) { 2381 ifp->if_csum_flags_rx |= M_CSUM_UDPv4; 2382 } 2383 2384 if (ifp->if_capenable & IFCAP_CSUM_TCPv6_Tx) { 2385 ifp->if_csum_flags_tx |= M_CSUM_TCPv6; 2386 } 2387 if (ifp->if_capenable & IFCAP_CSUM_TCPv6_Rx) { 2388 ifp->if_csum_flags_rx |= M_CSUM_TCPv6; 2389 } 2390 2391 if (ifp->if_capenable & IFCAP_CSUM_UDPv6_Tx) { 2392 ifp->if_csum_flags_tx |= M_CSUM_UDPv6; 2393 } 2394 if (ifp->if_capenable & IFCAP_CSUM_UDPv6_Rx) { 2395 ifp->if_csum_flags_rx |= M_CSUM_UDPv6; 2396 } 2397 if (ifp->if_flags & IFF_UP) 2398 return ENETRESET; 2399 return 0; 2400 case SIOCSIFFLAGS: 2401 ifr = data; 2402 if (ifp->if_flags & IFF_UP && (ifr->ifr_flags & IFF_UP) == 0) { 2403 s = splnet(); 2404 if_down(ifp); 2405 splx(s); 2406 } 2407 if (ifr->ifr_flags & IFF_UP && (ifp->if_flags & IFF_UP) == 0) { 2408 s = splnet(); 2409 if_up(ifp); 2410 splx(s); 2411 } 2412 ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) | 2413 (ifr->ifr_flags &~ IFF_CANTCHANGE); 2414 break; 2415 case SIOCGIFFLAGS: 2416 ifr = data; 2417 ifr->ifr_flags = ifp->if_flags; 2418 break; 2419 2420 case SIOCGIFMETRIC: 2421 ifr = data; 2422 ifr->ifr_metric = ifp->if_metric; 2423 break; 2424 2425 case SIOCGIFMTU: 2426 ifr = data; 2427 ifr->ifr_mtu = ifp->if_mtu; 2428 break; 2429 2430 case SIOCGIFDLT: 2431 ifr = data; 2432 ifr->ifr_dlt = ifp->if_dlt; 2433 break; 2434 2435 case SIOCGIFCAP: 2436 ifcr = data; 2437 ifcr->ifcr_capabilities = ifp->if_capabilities; 2438 ifcr->ifcr_capenable = ifp->if_capenable; 2439 break; 2440 2441 case SIOCSIFMETRIC: 2442 ifr = data; 2443 ifp->if_metric = ifr->ifr_metric; 2444 break; 2445 2446 case SIOCGIFDATA: 2447 ifdr = data; 2448 ifdr->ifdr_data = ifp->if_data; 2449 break; 2450 2451 case SIOCGIFINDEX: 2452 ifr = data; 2453 ifr->ifr_index = ifp->if_index; 2454 break; 2455 2456 case SIOCZIFDATA: 2457 ifdr = data; 2458 ifdr->ifdr_data = ifp->if_data; 2459 /* 2460 * Assumes that the volatile counters that can be 2461 * zero'ed are at the end of if_data. 2462 */ 2463 memset(&ifp->if_data.ifi_ipackets, 0, sizeof(ifp->if_data) - 2464 offsetof(struct if_data, ifi_ipackets)); 2465 /* 2466 * The memset() clears to the bottm of if_data. In the area, 2467 * if_lastchange is included. Please be careful if new entry 2468 * will be added into if_data or rewite this. 2469 * 2470 * And also, update if_lastchnage. 2471 */ 2472 getnanotime(&ifp->if_lastchange); 2473 break; 2474 case SIOCSIFMTU: 2475 ifr = data; 2476 if (ifp->if_mtu == ifr->ifr_mtu) 2477 break; 2478 ifp->if_mtu = ifr->ifr_mtu; 2479 /* 2480 * If the link MTU changed, do network layer specific procedure. 2481 */ 2482 #ifdef INET6 2483 if (in6_present) 2484 nd6_setmtu(ifp); 2485 #endif 2486 return ENETRESET; 2487 default: 2488 return ENOTTY; 2489 } 2490 return 0; 2491 } 2492 2493 int 2494 ifaddrpref_ioctl(struct socket *so, u_long cmd, void *data, struct ifnet *ifp) 2495 { 2496 struct if_addrprefreq *ifap = (struct if_addrprefreq *)data; 2497 struct ifaddr *ifa; 2498 const struct sockaddr *any, *sa; 2499 union { 2500 struct sockaddr sa; 2501 struct sockaddr_storage ss; 2502 } u, v; 2503 2504 switch (cmd) { 2505 case SIOCSIFADDRPREF: 2506 if (kauth_authorize_network(curlwp->l_cred, KAUTH_NETWORK_INTERFACE, 2507 KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd, 2508 NULL) != 0) 2509 return EPERM; 2510 case SIOCGIFADDRPREF: 2511 break; 2512 default: 2513 return EOPNOTSUPP; 2514 } 2515 2516 /* sanity checks */ 2517 if (data == NULL || ifp == NULL) { 2518 panic("invalid argument to %s", __func__); 2519 /*NOTREACHED*/ 2520 } 2521 2522 /* address must be specified on ADD and DELETE */ 2523 sa = sstocsa(&ifap->ifap_addr); 2524 if (sa->sa_family != sofamily(so)) 2525 return EINVAL; 2526 if ((any = sockaddr_any(sa)) == NULL || sa->sa_len != any->sa_len) 2527 return EINVAL; 2528 2529 sockaddr_externalize(&v.sa, sizeof(v.ss), sa); 2530 2531 IFADDR_FOREACH(ifa, ifp) { 2532 if (ifa->ifa_addr->sa_family != sa->sa_family) 2533 continue; 2534 sockaddr_externalize(&u.sa, sizeof(u.ss), ifa->ifa_addr); 2535 if (sockaddr_cmp(&u.sa, &v.sa) == 0) 2536 break; 2537 } 2538 if (ifa == NULL) 2539 return EADDRNOTAVAIL; 2540 2541 switch (cmd) { 2542 case SIOCSIFADDRPREF: 2543 ifa->ifa_preference = ifap->ifap_preference; 2544 return 0; 2545 case SIOCGIFADDRPREF: 2546 /* fill in the if_laddrreq structure */ 2547 (void)sockaddr_copy(sstosa(&ifap->ifap_addr), 2548 sizeof(ifap->ifap_addr), ifa->ifa_addr); 2549 ifap->ifap_preference = ifa->ifa_preference; 2550 return 0; 2551 default: 2552 return EOPNOTSUPP; 2553 } 2554 } 2555 2556 /* 2557 * Interface ioctls. 2558 */ 2559 static int 2560 doifioctl(struct socket *so, u_long cmd, void *data, struct lwp *l) 2561 { 2562 struct ifnet *ifp; 2563 struct ifreq *ifr; 2564 int error = 0; 2565 #if defined(COMPAT_OSOCK) || defined(COMPAT_OIFREQ) 2566 u_long ocmd = cmd; 2567 #endif 2568 short oif_flags; 2569 #ifdef COMPAT_OIFREQ 2570 struct ifreq ifrb; 2571 struct oifreq *oifr = NULL; 2572 #endif 2573 int r; 2574 struct psref psref; 2575 int bound; 2576 2577 switch (cmd) { 2578 #ifdef COMPAT_OIFREQ 2579 case OSIOCGIFCONF: 2580 case OOSIOCGIFCONF: 2581 return compat_ifconf(cmd, data); 2582 #endif 2583 #ifdef COMPAT_OIFDATA 2584 case OSIOCGIFDATA: 2585 case OSIOCZIFDATA: 2586 return compat_ifdatareq(l, cmd, data); 2587 #endif 2588 case SIOCGIFCONF: 2589 return ifconf(cmd, data); 2590 case SIOCINITIFADDR: 2591 return EPERM; 2592 } 2593 2594 #ifdef COMPAT_OIFREQ 2595 cmd = compat_cvtcmd(cmd); 2596 if (cmd != ocmd) { 2597 oifr = data; 2598 data = ifr = &ifrb; 2599 ifreqo2n(oifr, ifr); 2600 } else 2601 #endif 2602 ifr = data; 2603 2604 switch (cmd) { 2605 case SIOCIFCREATE: 2606 case SIOCIFDESTROY: 2607 bound = curlwp_bind(); 2608 if (l != NULL) { 2609 ifp = if_get(ifr->ifr_name, &psref); 2610 error = kauth_authorize_network(l->l_cred, 2611 KAUTH_NETWORK_INTERFACE, 2612 KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, 2613 (void *)cmd, NULL); 2614 if (ifp != NULL) 2615 if_put(ifp, &psref); 2616 if (error != 0) { 2617 curlwp_bindx(bound); 2618 return error; 2619 } 2620 } 2621 mutex_enter(&if_clone_mtx); 2622 r = (cmd == SIOCIFCREATE) ? 2623 if_clone_create(ifr->ifr_name) : 2624 if_clone_destroy(ifr->ifr_name); 2625 mutex_exit(&if_clone_mtx); 2626 curlwp_bindx(bound); 2627 return r; 2628 2629 case SIOCIFGCLONERS: 2630 { 2631 struct if_clonereq *req = (struct if_clonereq *)data; 2632 return if_clone_list(req->ifcr_count, req->ifcr_buffer, 2633 &req->ifcr_total); 2634 } 2635 } 2636 2637 bound = curlwp_bind(); 2638 ifp = if_get(ifr->ifr_name, &psref); 2639 if (ifp == NULL) { 2640 curlwp_bindx(bound); 2641 return ENXIO; 2642 } 2643 2644 switch (cmd) { 2645 case SIOCALIFADDR: 2646 case SIOCDLIFADDR: 2647 case SIOCSIFADDRPREF: 2648 case SIOCSIFFLAGS: 2649 case SIOCSIFCAP: 2650 case SIOCSIFMETRIC: 2651 case SIOCZIFDATA: 2652 case SIOCSIFMTU: 2653 case SIOCSIFPHYADDR: 2654 case SIOCDIFPHYADDR: 2655 #ifdef INET6 2656 case SIOCSIFPHYADDR_IN6: 2657 #endif 2658 case SIOCSLIFPHYADDR: 2659 case SIOCADDMULTI: 2660 case SIOCDELMULTI: 2661 case SIOCSIFMEDIA: 2662 case SIOCSDRVSPEC: 2663 case SIOCG80211: 2664 case SIOCS80211: 2665 case SIOCS80211NWID: 2666 case SIOCS80211NWKEY: 2667 case SIOCS80211POWER: 2668 case SIOCS80211BSSID: 2669 case SIOCS80211CHANNEL: 2670 case SIOCSLINKSTR: 2671 if (l != NULL) { 2672 error = kauth_authorize_network(l->l_cred, 2673 KAUTH_NETWORK_INTERFACE, 2674 KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, 2675 (void *)cmd, NULL); 2676 if (error != 0) 2677 goto out; 2678 } 2679 } 2680 2681 oif_flags = ifp->if_flags; 2682 2683 mutex_enter(ifp->if_ioctl_lock); 2684 2685 error = (*ifp->if_ioctl)(ifp, cmd, data); 2686 if (error != ENOTTY) 2687 ; 2688 else if (so->so_proto == NULL) 2689 error = EOPNOTSUPP; 2690 else { 2691 #ifdef COMPAT_OSOCK 2692 error = compat_ifioctl(so, ocmd, cmd, data, l); 2693 #else 2694 error = (*so->so_proto->pr_usrreqs->pr_ioctl)(so, 2695 cmd, data, ifp); 2696 #endif 2697 } 2698 2699 if (((oif_flags ^ ifp->if_flags) & IFF_UP) != 0) { 2700 if ((ifp->if_flags & IFF_UP) != 0) { 2701 int s = splnet(); 2702 if_up(ifp); 2703 splx(s); 2704 } 2705 } 2706 #ifdef COMPAT_OIFREQ 2707 if (cmd != ocmd) 2708 ifreqn2o(oifr, ifr); 2709 #endif 2710 2711 mutex_exit(ifp->if_ioctl_lock); 2712 out: 2713 if_put(ifp, &psref); 2714 curlwp_bindx(bound); 2715 return error; 2716 } 2717 2718 /* 2719 * Return interface configuration 2720 * of system. List may be used 2721 * in later ioctl's (above) to get 2722 * other information. 2723 * 2724 * Each record is a struct ifreq. Before the addition of 2725 * sockaddr_storage, the API rule was that sockaddr flavors that did 2726 * not fit would extend beyond the struct ifreq, with the next struct 2727 * ifreq starting sa_len beyond the struct sockaddr. Because the 2728 * union in struct ifreq includes struct sockaddr_storage, every kind 2729 * of sockaddr must fit. Thus, there are no longer any overlength 2730 * records. 2731 * 2732 * Records are added to the user buffer if they fit, and ifc_len is 2733 * adjusted to the length that was written. Thus, the user is only 2734 * assured of getting the complete list if ifc_len on return is at 2735 * least sizeof(struct ifreq) less than it was on entry. 2736 * 2737 * If the user buffer pointer is NULL, this routine copies no data and 2738 * returns the amount of space that would be needed. 2739 * 2740 * Invariants: 2741 * ifrp points to the next part of the user's buffer to be used. If 2742 * ifrp != NULL, space holds the number of bytes remaining that we may 2743 * write at ifrp. Otherwise, space holds the number of bytes that 2744 * would have been written had there been adequate space. 2745 */ 2746 /*ARGSUSED*/ 2747 static int 2748 ifconf(u_long cmd, void *data) 2749 { 2750 struct ifconf *ifc = (struct ifconf *)data; 2751 struct ifnet *ifp; 2752 struct ifaddr *ifa; 2753 struct ifreq ifr, *ifrp = NULL; 2754 int space = 0, error = 0; 2755 const int sz = (int)sizeof(struct ifreq); 2756 const bool docopy = ifc->ifc_req != NULL; 2757 int s; 2758 int bound; 2759 struct psref psref; 2760 2761 if (docopy) { 2762 space = ifc->ifc_len; 2763 ifrp = ifc->ifc_req; 2764 } 2765 2766 bound = curlwp_bind(); 2767 s = pserialize_read_enter(); 2768 IFNET_READER_FOREACH(ifp) { 2769 psref_acquire(&psref, &ifp->if_psref, ifnet_psref_class); 2770 pserialize_read_exit(s); 2771 2772 (void)strncpy(ifr.ifr_name, ifp->if_xname, 2773 sizeof(ifr.ifr_name)); 2774 if (ifr.ifr_name[sizeof(ifr.ifr_name) - 1] != '\0') { 2775 error = ENAMETOOLONG; 2776 goto release_exit; 2777 } 2778 if (IFADDR_EMPTY(ifp)) { 2779 /* Interface with no addresses - send zero sockaddr. */ 2780 memset(&ifr.ifr_addr, 0, sizeof(ifr.ifr_addr)); 2781 if (!docopy) { 2782 space += sz; 2783 continue; 2784 } 2785 if (space >= sz) { 2786 error = copyout(&ifr, ifrp, sz); 2787 if (error != 0) 2788 goto release_exit; 2789 ifrp++; 2790 space -= sz; 2791 } 2792 } 2793 2794 IFADDR_FOREACH(ifa, ifp) { 2795 struct sockaddr *sa = ifa->ifa_addr; 2796 /* all sockaddrs must fit in sockaddr_storage */ 2797 KASSERT(sa->sa_len <= sizeof(ifr.ifr_ifru)); 2798 2799 if (!docopy) { 2800 space += sz; 2801 continue; 2802 } 2803 memcpy(&ifr.ifr_space, sa, sa->sa_len); 2804 if (space >= sz) { 2805 error = copyout(&ifr, ifrp, sz); 2806 if (error != 0) 2807 goto release_exit; 2808 ifrp++; space -= sz; 2809 } 2810 } 2811 2812 s = pserialize_read_enter(); 2813 psref_release(&psref, &ifp->if_psref, ifnet_psref_class); 2814 } 2815 pserialize_read_exit(s); 2816 curlwp_bindx(bound); 2817 2818 if (docopy) { 2819 KASSERT(0 <= space && space <= ifc->ifc_len); 2820 ifc->ifc_len -= space; 2821 } else { 2822 KASSERT(space >= 0); 2823 ifc->ifc_len = space; 2824 } 2825 return (0); 2826 2827 release_exit: 2828 psref_release(&psref, &ifp->if_psref, ifnet_psref_class); 2829 curlwp_bindx(bound); 2830 return error; 2831 } 2832 2833 int 2834 ifreq_setaddr(u_long cmd, struct ifreq *ifr, const struct sockaddr *sa) 2835 { 2836 uint8_t len; 2837 #ifdef COMPAT_OIFREQ 2838 struct ifreq ifrb; 2839 struct oifreq *oifr = NULL; 2840 u_long ocmd = cmd; 2841 cmd = compat_cvtcmd(cmd); 2842 if (cmd != ocmd) { 2843 oifr = (struct oifreq *)(void *)ifr; 2844 ifr = &ifrb; 2845 ifreqo2n(oifr, ifr); 2846 len = sizeof(oifr->ifr_addr); 2847 } else 2848 #endif 2849 len = sizeof(ifr->ifr_ifru.ifru_space); 2850 2851 if (len < sa->sa_len) 2852 return EFBIG; 2853 2854 memset(&ifr->ifr_addr, 0, len); 2855 sockaddr_copy(&ifr->ifr_addr, len, sa); 2856 2857 #ifdef COMPAT_OIFREQ 2858 if (cmd != ocmd) 2859 ifreqn2o(oifr, ifr); 2860 #endif 2861 return 0; 2862 } 2863 2864 /* 2865 * wrapper function for the drivers which doesn't have if_transmit(). 2866 */ 2867 static int 2868 if_transmit(struct ifnet *ifp, struct mbuf *m) 2869 { 2870 int s, error; 2871 2872 s = splnet(); 2873 2874 /* 2875 * If NET_MPSAFE is not defined , IFQ_LOCK() is nop. 2876 * use KERNEL_LOCK instead of ifq_lock. 2877 */ 2878 #ifndef NET_MPSAFE 2879 KERNEL_LOCK(1, NULL); 2880 #endif 2881 IFQ_ENQUEUE(&ifp->if_snd, m, error); 2882 #ifndef NET_MPSAFE 2883 KERNEL_UNLOCK_ONE(NULL); 2884 #endif 2885 if (error != 0) { 2886 /* mbuf is already freed */ 2887 goto out; 2888 } 2889 2890 ifp->if_obytes += m->m_pkthdr.len;; 2891 if (m->m_flags & M_MCAST) 2892 ifp->if_omcasts++; 2893 2894 if ((ifp->if_flags & IFF_OACTIVE) == 0) 2895 if_start_lock(ifp); 2896 out: 2897 splx(s); 2898 2899 return error; 2900 } 2901 2902 int 2903 if_transmit_lock(struct ifnet *ifp, struct mbuf *m) 2904 { 2905 int error; 2906 2907 #ifdef ALTQ 2908 KERNEL_LOCK(1, NULL); 2909 if (ALTQ_IS_ENABLED(&ifp->if_snd)) { 2910 error = if_transmit(ifp, m); 2911 KERNEL_UNLOCK_ONE(NULL); 2912 } else { 2913 KERNEL_UNLOCK_ONE(NULL); 2914 error = (*ifp->if_transmit)(ifp, m); 2915 } 2916 #else /* !ALTQ */ 2917 error = (*ifp->if_transmit)(ifp, m); 2918 #endif /* !ALTQ */ 2919 2920 return error; 2921 } 2922 2923 /* 2924 * Queue message on interface, and start output if interface 2925 * not yet active. 2926 */ 2927 int 2928 ifq_enqueue(struct ifnet *ifp, struct mbuf *m) 2929 { 2930 2931 return if_transmit_lock(ifp, m); 2932 } 2933 2934 /* 2935 * Queue message on interface, possibly using a second fast queue 2936 */ 2937 int 2938 ifq_enqueue2(struct ifnet *ifp, struct ifqueue *ifq, struct mbuf *m) 2939 { 2940 int error = 0; 2941 2942 if (ifq != NULL 2943 #ifdef ALTQ 2944 && ALTQ_IS_ENABLED(&ifp->if_snd) == 0 2945 #endif 2946 ) { 2947 if (IF_QFULL(ifq)) { 2948 IF_DROP(&ifp->if_snd); 2949 m_freem(m); 2950 if (error == 0) 2951 error = ENOBUFS; 2952 } else 2953 IF_ENQUEUE(ifq, m); 2954 } else 2955 IFQ_ENQUEUE(&ifp->if_snd, m, error); 2956 if (error != 0) { 2957 ++ifp->if_oerrors; 2958 return error; 2959 } 2960 return 0; 2961 } 2962 2963 int 2964 if_addr_init(ifnet_t *ifp, struct ifaddr *ifa, const bool src) 2965 { 2966 int rc; 2967 2968 if (ifp->if_initaddr != NULL) 2969 rc = (*ifp->if_initaddr)(ifp, ifa, src); 2970 else if (src || 2971 /* FIXME: may not hold if_ioctl_lock */ 2972 (rc = (*ifp->if_ioctl)(ifp, SIOCSIFDSTADDR, ifa)) == ENOTTY) 2973 rc = (*ifp->if_ioctl)(ifp, SIOCINITIFADDR, ifa); 2974 2975 return rc; 2976 } 2977 2978 int 2979 if_do_dad(struct ifnet *ifp) 2980 { 2981 if ((ifp->if_flags & IFF_LOOPBACK) != 0) 2982 return 0; 2983 2984 switch (ifp->if_type) { 2985 case IFT_FAITH: 2986 /* 2987 * These interfaces do not have the IFF_LOOPBACK flag, 2988 * but loop packets back. We do not have to do DAD on such 2989 * interfaces. We should even omit it, because loop-backed 2990 * responses would confuse the DAD procedure. 2991 */ 2992 return 0; 2993 default: 2994 /* 2995 * Our DAD routine requires the interface up and running. 2996 * However, some interfaces can be up before the RUNNING 2997 * status. Additionaly, users may try to assign addresses 2998 * before the interface becomes up (or running). 2999 * We simply skip DAD in such a case as a work around. 3000 * XXX: we should rather mark "tentative" on such addresses, 3001 * and do DAD after the interface becomes ready. 3002 */ 3003 if ((ifp->if_flags & (IFF_UP|IFF_RUNNING)) != 3004 (IFF_UP|IFF_RUNNING)) 3005 return 0; 3006 3007 return 1; 3008 } 3009 } 3010 3011 int 3012 if_flags_set(ifnet_t *ifp, const short flags) 3013 { 3014 int rc; 3015 3016 if (ifp->if_setflags != NULL) 3017 rc = (*ifp->if_setflags)(ifp, flags); 3018 else { 3019 short cantflags, chgdflags; 3020 struct ifreq ifr; 3021 3022 chgdflags = ifp->if_flags ^ flags; 3023 cantflags = chgdflags & IFF_CANTCHANGE; 3024 3025 if (cantflags != 0) 3026 ifp->if_flags ^= cantflags; 3027 3028 /* Traditionally, we do not call if_ioctl after 3029 * setting/clearing only IFF_PROMISC if the interface 3030 * isn't IFF_UP. Uphold that tradition. 3031 */ 3032 if (chgdflags == IFF_PROMISC && (ifp->if_flags & IFF_UP) == 0) 3033 return 0; 3034 3035 memset(&ifr, 0, sizeof(ifr)); 3036 3037 ifr.ifr_flags = flags & ~IFF_CANTCHANGE; 3038 /* FIXME: may not hold if_ioctl_lock */ 3039 rc = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, &ifr); 3040 3041 if (rc != 0 && cantflags != 0) 3042 ifp->if_flags ^= cantflags; 3043 } 3044 3045 return rc; 3046 } 3047 3048 int 3049 if_mcast_op(ifnet_t *ifp, const unsigned long cmd, const struct sockaddr *sa) 3050 { 3051 int rc; 3052 struct ifreq ifr; 3053 3054 if (ifp->if_mcastop != NULL) 3055 rc = (*ifp->if_mcastop)(ifp, cmd, sa); 3056 else { 3057 ifreq_setaddr(cmd, &ifr, sa); 3058 rc = (*ifp->if_ioctl)(ifp, cmd, &ifr); 3059 } 3060 3061 return rc; 3062 } 3063 3064 static void 3065 sysctl_sndq_setup(struct sysctllog **clog, const char *ifname, 3066 struct ifaltq *ifq) 3067 { 3068 const struct sysctlnode *cnode, *rnode; 3069 3070 if (sysctl_createv(clog, 0, NULL, &rnode, 3071 CTLFLAG_PERMANENT, 3072 CTLTYPE_NODE, "interfaces", 3073 SYSCTL_DESCR("Per-interface controls"), 3074 NULL, 0, NULL, 0, 3075 CTL_NET, CTL_CREATE, CTL_EOL) != 0) 3076 goto bad; 3077 3078 if (sysctl_createv(clog, 0, &rnode, &rnode, 3079 CTLFLAG_PERMANENT, 3080 CTLTYPE_NODE, ifname, 3081 SYSCTL_DESCR("Interface controls"), 3082 NULL, 0, NULL, 0, 3083 CTL_CREATE, CTL_EOL) != 0) 3084 goto bad; 3085 3086 if (sysctl_createv(clog, 0, &rnode, &rnode, 3087 CTLFLAG_PERMANENT, 3088 CTLTYPE_NODE, "sndq", 3089 SYSCTL_DESCR("Interface output queue controls"), 3090 NULL, 0, NULL, 0, 3091 CTL_CREATE, CTL_EOL) != 0) 3092 goto bad; 3093 3094 if (sysctl_createv(clog, 0, &rnode, &cnode, 3095 CTLFLAG_PERMANENT, 3096 CTLTYPE_INT, "len", 3097 SYSCTL_DESCR("Current output queue length"), 3098 NULL, 0, &ifq->ifq_len, 0, 3099 CTL_CREATE, CTL_EOL) != 0) 3100 goto bad; 3101 3102 if (sysctl_createv(clog, 0, &rnode, &cnode, 3103 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 3104 CTLTYPE_INT, "maxlen", 3105 SYSCTL_DESCR("Maximum allowed output queue length"), 3106 NULL, 0, &ifq->ifq_maxlen, 0, 3107 CTL_CREATE, CTL_EOL) != 0) 3108 goto bad; 3109 3110 if (sysctl_createv(clog, 0, &rnode, &cnode, 3111 CTLFLAG_PERMANENT, 3112 CTLTYPE_INT, "drops", 3113 SYSCTL_DESCR("Packets dropped due to full output queue"), 3114 NULL, 0, &ifq->ifq_drops, 0, 3115 CTL_CREATE, CTL_EOL) != 0) 3116 goto bad; 3117 3118 return; 3119 bad: 3120 printf("%s: could not attach sysctl nodes\n", ifname); 3121 return; 3122 } 3123 3124 #if defined(INET) || defined(INET6) 3125 3126 #define SYSCTL_NET_PKTQ(q, cn, c) \ 3127 static int \ 3128 sysctl_net_##q##_##cn(SYSCTLFN_ARGS) \ 3129 { \ 3130 return sysctl_pktq_count(SYSCTLFN_CALL(rnode), q, c); \ 3131 } 3132 3133 #if defined(INET) 3134 static int 3135 sysctl_net_ip_pktq_maxlen(SYSCTLFN_ARGS) 3136 { 3137 return sysctl_pktq_maxlen(SYSCTLFN_CALL(rnode), ip_pktq); 3138 } 3139 SYSCTL_NET_PKTQ(ip_pktq, items, PKTQ_NITEMS) 3140 SYSCTL_NET_PKTQ(ip_pktq, drops, PKTQ_DROPS) 3141 #endif 3142 3143 #if defined(INET6) 3144 static int 3145 sysctl_net_ip6_pktq_maxlen(SYSCTLFN_ARGS) 3146 { 3147 return sysctl_pktq_maxlen(SYSCTLFN_CALL(rnode), ip6_pktq); 3148 } 3149 SYSCTL_NET_PKTQ(ip6_pktq, items, PKTQ_NITEMS) 3150 SYSCTL_NET_PKTQ(ip6_pktq, drops, PKTQ_DROPS) 3151 #endif 3152 3153 static void 3154 sysctl_net_pktq_setup(struct sysctllog **clog, int pf) 3155 { 3156 sysctlfn len_func = NULL, maxlen_func = NULL, drops_func = NULL; 3157 const char *pfname = NULL, *ipname = NULL; 3158 int ipn = 0, qid = 0; 3159 3160 switch (pf) { 3161 #if defined(INET) 3162 case PF_INET: 3163 len_func = sysctl_net_ip_pktq_items; 3164 maxlen_func = sysctl_net_ip_pktq_maxlen; 3165 drops_func = sysctl_net_ip_pktq_drops; 3166 pfname = "inet", ipn = IPPROTO_IP; 3167 ipname = "ip", qid = IPCTL_IFQ; 3168 break; 3169 #endif 3170 #if defined(INET6) 3171 case PF_INET6: 3172 len_func = sysctl_net_ip6_pktq_items; 3173 maxlen_func = sysctl_net_ip6_pktq_maxlen; 3174 drops_func = sysctl_net_ip6_pktq_drops; 3175 pfname = "inet6", ipn = IPPROTO_IPV6; 3176 ipname = "ip6", qid = IPV6CTL_IFQ; 3177 break; 3178 #endif 3179 default: 3180 KASSERT(false); 3181 } 3182 3183 sysctl_createv(clog, 0, NULL, NULL, 3184 CTLFLAG_PERMANENT, 3185 CTLTYPE_NODE, pfname, NULL, 3186 NULL, 0, NULL, 0, 3187 CTL_NET, pf, CTL_EOL); 3188 sysctl_createv(clog, 0, NULL, NULL, 3189 CTLFLAG_PERMANENT, 3190 CTLTYPE_NODE, ipname, NULL, 3191 NULL, 0, NULL, 0, 3192 CTL_NET, pf, ipn, CTL_EOL); 3193 sysctl_createv(clog, 0, NULL, NULL, 3194 CTLFLAG_PERMANENT, 3195 CTLTYPE_NODE, "ifq", 3196 SYSCTL_DESCR("Protocol input queue controls"), 3197 NULL, 0, NULL, 0, 3198 CTL_NET, pf, ipn, qid, CTL_EOL); 3199 3200 sysctl_createv(clog, 0, NULL, NULL, 3201 CTLFLAG_PERMANENT, 3202 CTLTYPE_INT, "len", 3203 SYSCTL_DESCR("Current input queue length"), 3204 len_func, 0, NULL, 0, 3205 CTL_NET, pf, ipn, qid, IFQCTL_LEN, CTL_EOL); 3206 sysctl_createv(clog, 0, NULL, NULL, 3207 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 3208 CTLTYPE_INT, "maxlen", 3209 SYSCTL_DESCR("Maximum allowed input queue length"), 3210 maxlen_func, 0, NULL, 0, 3211 CTL_NET, pf, ipn, qid, IFQCTL_MAXLEN, CTL_EOL); 3212 sysctl_createv(clog, 0, NULL, NULL, 3213 CTLFLAG_PERMANENT, 3214 CTLTYPE_INT, "drops", 3215 SYSCTL_DESCR("Packets dropped due to full input queue"), 3216 drops_func, 0, NULL, 0, 3217 CTL_NET, pf, ipn, qid, IFQCTL_DROPS, CTL_EOL); 3218 } 3219 #endif /* INET || INET6 */ 3220 3221 static int 3222 if_sdl_sysctl(SYSCTLFN_ARGS) 3223 { 3224 struct ifnet *ifp; 3225 const struct sockaddr_dl *sdl; 3226 struct psref psref; 3227 int error = 0; 3228 int bound; 3229 3230 if (namelen != 1) 3231 return EINVAL; 3232 3233 bound = curlwp_bind(); 3234 ifp = if_get_byindex(name[0], &psref); 3235 if (ifp == NULL) { 3236 error = ENODEV; 3237 goto out0; 3238 } 3239 3240 sdl = ifp->if_sadl; 3241 if (sdl == NULL) { 3242 *oldlenp = 0; 3243 goto out1; 3244 } 3245 3246 if (oldp == NULL) { 3247 *oldlenp = sdl->sdl_alen; 3248 goto out1; 3249 } 3250 3251 if (*oldlenp >= sdl->sdl_alen) 3252 *oldlenp = sdl->sdl_alen; 3253 error = sysctl_copyout(l, &sdl->sdl_data[sdl->sdl_nlen], oldp, *oldlenp); 3254 out1: 3255 if_put(ifp, &psref); 3256 out0: 3257 curlwp_bindx(bound); 3258 return error; 3259 } 3260 3261 SYSCTL_SETUP(sysctl_net_sdl_setup, "sysctl net.sdl subtree setup") 3262 { 3263 const struct sysctlnode *rnode = NULL; 3264 3265 sysctl_createv(clog, 0, NULL, &rnode, 3266 CTLFLAG_PERMANENT, 3267 CTLTYPE_NODE, "sdl", 3268 SYSCTL_DESCR("Get active link-layer address"), 3269 if_sdl_sysctl, 0, NULL, 0, 3270 CTL_NET, CTL_CREATE, CTL_EOL); 3271 } 3272
Indexes created Tue Sep 23 14:10:03 GMT 2025