if.c revision 1.335
1 /* $NetBSD: if.c,v 1.335 2016/05/16 01:06:31 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 1.335 2016/05/16 01:06:31 ozaki-r Exp $"); 94 95 #if defined(_KERNEL_OPT) 96 #include "opt_inet.h" 97 98 #include "opt_atalk.h" 99 #include "opt_natm.h" 100 #include "opt_wlan.h" 101 #include "opt_net_mpsafe.h" 102 #endif 103 104 #include <sys/param.h> 105 #include <sys/mbuf.h> 106 #include <sys/systm.h> 107 #include <sys/callout.h> 108 #include <sys/proc.h> 109 #include <sys/socket.h> 110 #include <sys/socketvar.h> 111 #include <sys/domain.h> 112 #include <sys/protosw.h> 113 #include <sys/kernel.h> 114 #include <sys/ioctl.h> 115 #include <sys/sysctl.h> 116 #include <sys/syslog.h> 117 #include <sys/kauth.h> 118 #include <sys/kmem.h> 119 #include <sys/xcall.h> 120 #include <sys/cpu.h> 121 #include <sys/intr.h> 122 123 #include <net/if.h> 124 #include <net/if_dl.h> 125 #include <net/if_ether.h> 126 #include <net/if_media.h> 127 #include <net80211/ieee80211.h> 128 #include <net80211/ieee80211_ioctl.h> 129 #include <net/if_types.h> 130 #include <net/route.h> 131 #include <net/netisr.h> 132 #include <sys/module.h> 133 #ifdef NETATALK 134 #include <netatalk/at_extern.h> 135 #include <netatalk/at.h> 136 #endif 137 #include <net/pfil.h> 138 #include <netinet/in.h> 139 #include <netinet/in_var.h> 140 141 #ifdef INET6 142 #include <netinet6/in6_var.h> 143 #include <netinet6/nd6.h> 144 #endif 145 146 #include "ether.h" 147 #include "fddi.h" 148 #include "token.h" 149 150 #include "carp.h" 151 #if NCARP > 0 152 #include <netinet/ip_carp.h> 153 #endif 154 155 #include <compat/sys/sockio.h> 156 #include <compat/sys/socket.h> 157 158 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address"); 159 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address"); 160 161 /* 162 * Global list of interfaces. 163 */ 164 /* DEPRECATED. Remove it once kvm(3) users disappeared */ 165 struct ifnet_head ifnet_list; 166 167 struct pslist_head ifnet_pslist; 168 static ifnet_t ** ifindex2ifnet = NULL; 169 static u_int if_index = 1; 170 static size_t if_indexlim = 0; 171 static uint64_t index_gen; 172 /* Mutex to protect the above objects. */ 173 kmutex_t ifnet_mtx __cacheline_aligned; 174 struct psref_class *ifnet_psref_class __read_mostly; 175 static pserialize_t ifnet_psz; 176 177 static kmutex_t if_clone_mtx; 178 179 struct ifnet *lo0ifp; 180 int ifqmaxlen = IFQ_MAXLEN; 181 182 static int if_rt_walktree(struct rtentry *, void *); 183 184 static struct if_clone *if_clone_lookup(const char *, int *); 185 186 static LIST_HEAD(, if_clone) if_cloners = LIST_HEAD_INITIALIZER(if_cloners); 187 static int if_cloners_count; 188 189 /* Packet filtering hook for interfaces. */ 190 pfil_head_t * if_pfil; 191 192 static kauth_listener_t if_listener; 193 194 static int doifioctl(struct socket *, u_long, void *, struct lwp *); 195 static int ifioctl_attach(struct ifnet *); 196 static void ifioctl_detach(struct ifnet *); 197 static void ifnet_lock_enter(struct ifnet_lock *); 198 static void ifnet_lock_exit(struct ifnet_lock *); 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_clone_create(const char *); 207 static int if_clone_destroy(const char *); 208 static void if_link_state_change_si(void *); 209 210 struct if_percpuq { 211 struct ifnet *ipq_ifp; 212 void *ipq_si; 213 struct percpu *ipq_ifqs; /* struct ifqueue */ 214 }; 215 216 static struct mbuf *if_percpuq_dequeue(struct if_percpuq *); 217 218 static void if_percpuq_drops(void *, void *, struct cpu_info *); 219 static int sysctl_percpuq_drops_handler(SYSCTLFN_PROTO); 220 static void sysctl_percpuq_setup(struct sysctllog **, const char *, 221 struct if_percpuq *); 222 223 #if defined(INET) || defined(INET6) 224 static void sysctl_net_pktq_setup(struct sysctllog **, int); 225 #endif 226 227 static int 228 if_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie, 229 void *arg0, void *arg1, void *arg2, void *arg3) 230 { 231 int result; 232 enum kauth_network_req req; 233 234 result = KAUTH_RESULT_DEFER; 235 req = (enum kauth_network_req)arg1; 236 237 if (action != KAUTH_NETWORK_INTERFACE) 238 return result; 239 240 if ((req == KAUTH_REQ_NETWORK_INTERFACE_GET) || 241 (req == KAUTH_REQ_NETWORK_INTERFACE_SET)) 242 result = KAUTH_RESULT_ALLOW; 243 244 return result; 245 } 246 247 /* 248 * Network interface utility routines. 249 * 250 * Routines with ifa_ifwith* names take sockaddr *'s as 251 * parameters. 252 */ 253 void 254 ifinit(void) 255 { 256 #if defined(INET) 257 sysctl_net_pktq_setup(NULL, PF_INET); 258 #endif 259 #ifdef INET6 260 if (in6_present) 261 sysctl_net_pktq_setup(NULL, PF_INET6); 262 #endif 263 264 if_listener = kauth_listen_scope(KAUTH_SCOPE_NETWORK, 265 if_listener_cb, NULL); 266 267 /* interfaces are available, inform socket code */ 268 ifioctl = doifioctl; 269 } 270 271 /* 272 * XXX Initialization before configure(). 273 * XXX hack to get pfil_add_hook working in autoconf. 274 */ 275 void 276 ifinit1(void) 277 { 278 mutex_init(&if_clone_mtx, MUTEX_DEFAULT, IPL_NONE); 279 TAILQ_INIT(&ifnet_list); 280 mutex_init(&ifnet_mtx, MUTEX_DEFAULT, IPL_NONE); 281 ifnet_psz = pserialize_create(); 282 ifnet_psref_class = psref_class_create("ifnet", IPL_SOFTNET); 283 PSLIST_INIT(&ifnet_pslist); 284 if_indexlim = 8; 285 286 if_pfil = pfil_head_create(PFIL_TYPE_IFNET, NULL); 287 KASSERT(if_pfil != NULL); 288 289 #if NETHER > 0 || NFDDI > 0 || defined(NETATALK) || NTOKEN > 0 || defined(WLAN) 290 etherinit(); 291 #endif 292 } 293 294 ifnet_t * 295 if_alloc(u_char type) 296 { 297 return kmem_zalloc(sizeof(ifnet_t), KM_SLEEP); 298 } 299 300 void 301 if_free(ifnet_t *ifp) 302 { 303 kmem_free(ifp, sizeof(ifnet_t)); 304 } 305 306 void 307 if_initname(struct ifnet *ifp, const char *name, int unit) 308 { 309 (void)snprintf(ifp->if_xname, sizeof(ifp->if_xname), 310 "%s%d", name, unit); 311 } 312 313 /* 314 * Null routines used while an interface is going away. These routines 315 * just return an error. 316 */ 317 318 int 319 if_nulloutput(struct ifnet *ifp, struct mbuf *m, 320 const struct sockaddr *so, const struct rtentry *rt) 321 { 322 323 return ENXIO; 324 } 325 326 void 327 if_nullinput(struct ifnet *ifp, struct mbuf *m) 328 { 329 330 /* Nothing. */ 331 } 332 333 void 334 if_nullstart(struct ifnet *ifp) 335 { 336 337 /* Nothing. */ 338 } 339 340 int 341 if_nulltransmit(struct ifnet *ifp, struct mbuf *m) 342 { 343 344 return ENXIO; 345 } 346 347 int 348 if_nullioctl(struct ifnet *ifp, u_long cmd, void *data) 349 { 350 351 /* Wake ifioctl_detach(), who may wait for all threads to 352 * quit the critical section. 353 */ 354 cv_signal(&ifp->if_ioctl_lock->il_emptied); 355 return ENXIO; 356 } 357 358 int 359 if_nullinit(struct ifnet *ifp) 360 { 361 362 return ENXIO; 363 } 364 365 void 366 if_nullstop(struct ifnet *ifp, int disable) 367 { 368 369 /* Nothing. */ 370 } 371 372 void 373 if_nullslowtimo(struct ifnet *ifp) 374 { 375 376 /* Nothing. */ 377 } 378 379 void 380 if_nulldrain(struct ifnet *ifp) 381 { 382 383 /* Nothing. */ 384 } 385 386 void 387 if_set_sadl(struct ifnet *ifp, const void *lla, u_char addrlen, bool factory) 388 { 389 struct ifaddr *ifa; 390 struct sockaddr_dl *sdl; 391 392 ifp->if_addrlen = addrlen; 393 if_alloc_sadl(ifp); 394 ifa = ifp->if_dl; 395 sdl = satosdl(ifa->ifa_addr); 396 397 (void)sockaddr_dl_setaddr(sdl, sdl->sdl_len, lla, ifp->if_addrlen); 398 if (factory) { 399 ifp->if_hwdl = ifp->if_dl; 400 ifaref(ifp->if_hwdl); 401 } 402 /* TBD routing socket */ 403 } 404 405 struct ifaddr * 406 if_dl_create(const struct ifnet *ifp, const struct sockaddr_dl **sdlp) 407 { 408 unsigned socksize, ifasize; 409 int addrlen, namelen; 410 struct sockaddr_dl *mask, *sdl; 411 struct ifaddr *ifa; 412 413 namelen = strlen(ifp->if_xname); 414 addrlen = ifp->if_addrlen; 415 socksize = roundup(sockaddr_dl_measure(namelen, addrlen), sizeof(long)); 416 ifasize = sizeof(*ifa) + 2 * socksize; 417 ifa = (struct ifaddr *)malloc(ifasize, M_IFADDR, M_WAITOK|M_ZERO); 418 419 sdl = (struct sockaddr_dl *)(ifa + 1); 420 mask = (struct sockaddr_dl *)(socksize + (char *)sdl); 421 422 sockaddr_dl_init(sdl, socksize, ifp->if_index, ifp->if_type, 423 ifp->if_xname, namelen, NULL, addrlen); 424 mask->sdl_len = sockaddr_dl_measure(namelen, 0); 425 memset(&mask->sdl_data[0], 0xff, namelen); 426 ifa->ifa_rtrequest = link_rtrequest; 427 ifa->ifa_addr = (struct sockaddr *)sdl; 428 ifa->ifa_netmask = (struct sockaddr *)mask; 429 430 *sdlp = sdl; 431 432 return ifa; 433 } 434 435 static void 436 if_sadl_setrefs(struct ifnet *ifp, struct ifaddr *ifa) 437 { 438 const struct sockaddr_dl *sdl; 439 440 ifp->if_dl = ifa; 441 ifaref(ifa); 442 sdl = satosdl(ifa->ifa_addr); 443 ifp->if_sadl = sdl; 444 } 445 446 /* 447 * Allocate the link level name for the specified interface. This 448 * is an attachment helper. It must be called after ifp->if_addrlen 449 * is initialized, which may not be the case when if_attach() is 450 * called. 451 */ 452 void 453 if_alloc_sadl(struct ifnet *ifp) 454 { 455 struct ifaddr *ifa; 456 const struct sockaddr_dl *sdl; 457 458 /* 459 * If the interface already has a link name, release it 460 * now. This is useful for interfaces that can change 461 * link types, and thus switch link names often. 462 */ 463 if (ifp->if_sadl != NULL) 464 if_free_sadl(ifp); 465 466 ifa = if_dl_create(ifp, &sdl); 467 468 ifa_insert(ifp, ifa); 469 if_sadl_setrefs(ifp, ifa); 470 } 471 472 static void 473 if_deactivate_sadl(struct ifnet *ifp) 474 { 475 struct ifaddr *ifa; 476 477 KASSERT(ifp->if_dl != NULL); 478 479 ifa = ifp->if_dl; 480 481 ifp->if_sadl = NULL; 482 483 ifp->if_dl = NULL; 484 ifafree(ifa); 485 } 486 487 void 488 if_activate_sadl(struct ifnet *ifp, struct ifaddr *ifa, 489 const struct sockaddr_dl *sdl) 490 { 491 int s; 492 493 s = splnet(); 494 495 if_deactivate_sadl(ifp); 496 497 if_sadl_setrefs(ifp, ifa); 498 IFADDR_FOREACH(ifa, ifp) 499 rtinit(ifa, RTM_LLINFO_UPD, 0); 500 splx(s); 501 } 502 503 /* 504 * Free the link level name for the specified interface. This is 505 * a detach helper. This is called from if_detach(). 506 */ 507 static void 508 if_free_sadl(struct ifnet *ifp) 509 { 510 struct ifaddr *ifa; 511 int s; 512 513 ifa = ifp->if_dl; 514 if (ifa == NULL) { 515 KASSERT(ifp->if_sadl == NULL); 516 return; 517 } 518 519 KASSERT(ifp->if_sadl != NULL); 520 521 s = splnet(); 522 rtinit(ifa, RTM_DELETE, 0); 523 ifa_remove(ifp, ifa); 524 if_deactivate_sadl(ifp); 525 if (ifp->if_hwdl == ifa) { 526 ifafree(ifa); 527 ifp->if_hwdl = NULL; 528 } 529 splx(s); 530 } 531 532 static void 533 if_getindex(ifnet_t *ifp) 534 { 535 bool hitlimit = false; 536 537 ifp->if_index_gen = index_gen++; 538 539 ifp->if_index = if_index; 540 if (ifindex2ifnet == NULL) { 541 if_index++; 542 goto skip; 543 } 544 while (if_byindex(ifp->if_index)) { 545 /* 546 * If we hit USHRT_MAX, we skip back to 0 since 547 * there are a number of places where the value 548 * of if_index or if_index itself is compared 549 * to or stored in an unsigned short. By 550 * jumping back, we won't botch those assignments 551 * or comparisons. 552 */ 553 if (++if_index == 0) { 554 if_index = 1; 555 } else if (if_index == USHRT_MAX) { 556 /* 557 * However, if we have to jump back to 558 * zero *twice* without finding an empty 559 * slot in ifindex2ifnet[], then there 560 * there are too many (>65535) interfaces. 561 */ 562 if (hitlimit) { 563 panic("too many interfaces"); 564 } 565 hitlimit = true; 566 if_index = 1; 567 } 568 ifp->if_index = if_index; 569 } 570 skip: 571 /* 572 * ifindex2ifnet is indexed by if_index. Since if_index will 573 * grow dynamically, it should grow too. 574 */ 575 if (ifindex2ifnet == NULL || ifp->if_index >= if_indexlim) { 576 size_t m, n, oldlim; 577 void *q; 578 579 oldlim = if_indexlim; 580 while (ifp->if_index >= if_indexlim) 581 if_indexlim <<= 1; 582 583 /* grow ifindex2ifnet */ 584 m = oldlim * sizeof(struct ifnet *); 585 n = if_indexlim * sizeof(struct ifnet *); 586 q = malloc(n, M_IFADDR, M_WAITOK|M_ZERO); 587 if (ifindex2ifnet != NULL) { 588 memcpy(q, ifindex2ifnet, m); 589 free(ifindex2ifnet, M_IFADDR); 590 } 591 ifindex2ifnet = (struct ifnet **)q; 592 } 593 ifindex2ifnet[ifp->if_index] = ifp; 594 } 595 596 /* 597 * Initialize an interface and assign an index for it. 598 * 599 * It must be called prior to a device specific attach routine 600 * (e.g., ether_ifattach and ieee80211_ifattach) or if_alloc_sadl, 601 * and be followed by if_register: 602 * 603 * if_initialize(ifp); 604 * ether_ifattach(ifp, enaddr); 605 * if_register(ifp); 606 */ 607 void 608 if_initialize(ifnet_t *ifp) 609 { 610 KASSERT(if_indexlim > 0); 611 TAILQ_INIT(&ifp->if_addrlist); 612 613 /* 614 * Link level name is allocated later by a separate call to 615 * if_alloc_sadl(). 616 */ 617 618 if (ifp->if_snd.ifq_maxlen == 0) 619 ifp->if_snd.ifq_maxlen = ifqmaxlen; 620 621 ifp->if_broadcastaddr = 0; /* reliably crash if used uninitialized */ 622 623 ifp->if_link_state = LINK_STATE_UNKNOWN; 624 ifp->if_link_queue = -1; /* all bits set, see link_state_change() */ 625 626 ifp->if_capenable = 0; 627 ifp->if_csum_flags_tx = 0; 628 ifp->if_csum_flags_rx = 0; 629 630 #ifdef ALTQ 631 ifp->if_snd.altq_type = 0; 632 ifp->if_snd.altq_disc = NULL; 633 ifp->if_snd.altq_flags &= ALTQF_CANTCHANGE; 634 ifp->if_snd.altq_tbr = NULL; 635 ifp->if_snd.altq_ifp = ifp; 636 #endif 637 638 #ifdef NET_MPSAFE 639 ifp->if_snd.ifq_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NET); 640 #else 641 ifp->if_snd.ifq_lock = NULL; 642 #endif 643 644 ifp->if_pfil = pfil_head_create(PFIL_TYPE_IFNET, ifp); 645 (void)pfil_run_hooks(if_pfil, 646 (struct mbuf **)PFIL_IFNET_ATTACH, ifp, PFIL_IFNET); 647 648 IF_AFDATA_LOCK_INIT(ifp); 649 650 ifp->if_link_si = softint_establish(SOFTINT_NET, if_link_state_change_si, ifp); 651 if (ifp->if_link_si == NULL) 652 panic("%s: softint_establish() failed", __func__); 653 654 PSLIST_ENTRY_INIT(ifp, if_pslist_entry); 655 psref_target_init(&ifp->if_psref, ifnet_psref_class); 656 657 IFNET_LOCK(); 658 if_getindex(ifp); 659 IFNET_UNLOCK(); 660 } 661 662 /* 663 * Register an interface to the list of "active" interfaces. 664 */ 665 void 666 if_register(ifnet_t *ifp) 667 { 668 if (ifioctl_attach(ifp) != 0) 669 panic("%s: ifioctl_attach() failed", __func__); 670 671 sysctl_sndq_setup(&ifp->if_sysctl_log, ifp->if_xname, &ifp->if_snd); 672 673 if (!STAILQ_EMPTY(&domains)) 674 if_attachdomain1(ifp); 675 676 /* Announce the interface. */ 677 rt_ifannouncemsg(ifp, IFAN_ARRIVAL); 678 679 if (ifp->if_slowtimo != NULL) { 680 ifp->if_slowtimo_ch = 681 kmem_zalloc(sizeof(*ifp->if_slowtimo_ch), KM_SLEEP); 682 callout_init(ifp->if_slowtimo_ch, 0); 683 callout_setfunc(ifp->if_slowtimo_ch, if_slowtimo, ifp); 684 if_slowtimo(ifp); 685 } 686 687 if (ifp->if_transmit == NULL || ifp->if_transmit == if_nulltransmit) 688 ifp->if_transmit = if_transmit; 689 690 IFNET_LOCK(); 691 TAILQ_INSERT_TAIL(&ifnet_list, ifp, if_list); 692 IFNET_WRITER_INSERT_TAIL(ifp); 693 IFNET_UNLOCK(); 694 } 695 696 /* 697 * The if_percpuq framework 698 * 699 * It allows network device drivers to execute the network stack 700 * in softint (so called softint-based if_input). It utilizes 701 * softint and percpu ifqueue. It doesn't distribute any packets 702 * between CPUs, unlike pktqueue(9). 703 * 704 * Currently we support two options for device drivers to apply the framework: 705 * - Use it implicitly with less changes 706 * - If you use if_attach in driver's _attach function and if_input in 707 * driver's Rx interrupt handler, a packet is queued and a softint handles 708 * the packet implicitly 709 * - Use it explicitly in each driver (recommended) 710 * - You can use if_percpuq_* directly in your driver 711 * - In this case, you need to allocate struct if_percpuq in driver's softc 712 * - See wm(4) as a reference implementation 713 */ 714 715 static void 716 if_percpuq_softint(void *arg) 717 { 718 struct if_percpuq *ipq = arg; 719 struct ifnet *ifp = ipq->ipq_ifp; 720 struct mbuf *m; 721 722 while ((m = if_percpuq_dequeue(ipq)) != NULL) 723 ifp->_if_input(ifp, m); 724 } 725 726 static void 727 if_percpuq_init_ifq(void *p, void *arg __unused, struct cpu_info *ci __unused) 728 { 729 struct ifqueue *const ifq = p; 730 731 memset(ifq, 0, sizeof(*ifq)); 732 ifq->ifq_maxlen = IFQ_MAXLEN; 733 } 734 735 struct if_percpuq * 736 if_percpuq_create(struct ifnet *ifp) 737 { 738 struct if_percpuq *ipq; 739 740 ipq = kmem_zalloc(sizeof(*ipq), KM_SLEEP); 741 if (ipq == NULL) 742 panic("kmem_zalloc failed"); 743 744 ipq->ipq_ifp = ifp; 745 ipq->ipq_si = softint_establish(SOFTINT_NET|SOFTINT_MPSAFE, 746 if_percpuq_softint, ipq); 747 ipq->ipq_ifqs = percpu_alloc(sizeof(struct ifqueue)); 748 percpu_foreach(ipq->ipq_ifqs, &if_percpuq_init_ifq, NULL); 749 750 sysctl_percpuq_setup(&ifp->if_sysctl_log, ifp->if_xname, ipq); 751 752 return ipq; 753 } 754 755 static struct mbuf * 756 if_percpuq_dequeue(struct if_percpuq *ipq) 757 { 758 struct mbuf *m; 759 struct ifqueue *ifq; 760 int s; 761 762 s = splnet(); 763 ifq = percpu_getref(ipq->ipq_ifqs); 764 IF_DEQUEUE(ifq, m); 765 percpu_putref(ipq->ipq_ifqs); 766 splx(s); 767 768 return m; 769 } 770 771 static void 772 if_percpuq_purge_ifq(void *p, void *arg __unused, struct cpu_info *ci __unused) 773 { 774 struct ifqueue *const ifq = p; 775 776 IF_PURGE(ifq); 777 } 778 779 void 780 if_percpuq_destroy(struct if_percpuq *ipq) 781 { 782 783 /* if_detach may already destroy it */ 784 if (ipq == NULL) 785 return; 786 787 softint_disestablish(ipq->ipq_si); 788 percpu_foreach(ipq->ipq_ifqs, &if_percpuq_purge_ifq, NULL); 789 percpu_free(ipq->ipq_ifqs, sizeof(struct ifqueue)); 790 } 791 792 void 793 if_percpuq_enqueue(struct if_percpuq *ipq, struct mbuf *m) 794 { 795 struct ifqueue *ifq; 796 int s; 797 798 KASSERT(ipq != NULL); 799 800 s = splnet(); 801 ifq = percpu_getref(ipq->ipq_ifqs); 802 if (IF_QFULL(ifq)) { 803 IF_DROP(ifq); 804 percpu_putref(ipq->ipq_ifqs); 805 m_freem(m); 806 goto out; 807 } 808 IF_ENQUEUE(ifq, m); 809 percpu_putref(ipq->ipq_ifqs); 810 811 softint_schedule(ipq->ipq_si); 812 out: 813 splx(s); 814 } 815 816 static void 817 if_percpuq_drops(void *p, void *arg, struct cpu_info *ci __unused) 818 { 819 struct ifqueue *const ifq = p; 820 int *sum = arg; 821 822 *sum += ifq->ifq_drops; 823 } 824 825 static int 826 sysctl_percpuq_drops_handler(SYSCTLFN_ARGS) 827 { 828 struct sysctlnode node; 829 struct if_percpuq *ipq; 830 int sum = 0; 831 int error; 832 833 node = *rnode; 834 ipq = node.sysctl_data; 835 836 percpu_foreach(ipq->ipq_ifqs, if_percpuq_drops, &sum); 837 838 node.sysctl_data = ∑ 839 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 840 if (error != 0 || newp == NULL) 841 return error; 842 843 return 0; 844 } 845 846 static void 847 sysctl_percpuq_setup(struct sysctllog **clog, const char* ifname, 848 struct if_percpuq *ipq) 849 { 850 const struct sysctlnode *cnode, *rnode; 851 852 if (sysctl_createv(clog, 0, NULL, &rnode, 853 CTLFLAG_PERMANENT, 854 CTLTYPE_NODE, "interfaces", 855 SYSCTL_DESCR("Per-interface controls"), 856 NULL, 0, NULL, 0, 857 CTL_NET, CTL_CREATE, CTL_EOL) != 0) 858 goto bad; 859 860 if (sysctl_createv(clog, 0, &rnode, &rnode, 861 CTLFLAG_PERMANENT, 862 CTLTYPE_NODE, ifname, 863 SYSCTL_DESCR("Interface controls"), 864 NULL, 0, NULL, 0, 865 CTL_CREATE, CTL_EOL) != 0) 866 goto bad; 867 868 if (sysctl_createv(clog, 0, &rnode, &rnode, 869 CTLFLAG_PERMANENT, 870 CTLTYPE_NODE, "rcvq", 871 SYSCTL_DESCR("Interface input queue controls"), 872 NULL, 0, NULL, 0, 873 CTL_CREATE, CTL_EOL) != 0) 874 goto bad; 875 876 #ifdef NOTYET 877 /* XXX Should show each per-CPU queue length? */ 878 if (sysctl_createv(clog, 0, &rnode, &rnode, 879 CTLFLAG_PERMANENT, 880 CTLTYPE_INT, "len", 881 SYSCTL_DESCR("Current input queue length"), 882 sysctl_percpuq_len, 0, NULL, 0, 883 CTL_CREATE, CTL_EOL) != 0) 884 goto bad; 885 886 if (sysctl_createv(clog, 0, &rnode, &cnode, 887 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 888 CTLTYPE_INT, "maxlen", 889 SYSCTL_DESCR("Maximum allowed input queue length"), 890 sysctl_percpuq_maxlen_handler, 0, (void *)ipq, 0, 891 CTL_CREATE, CTL_EOL) != 0) 892 goto bad; 893 #endif 894 895 if (sysctl_createv(clog, 0, &rnode, &cnode, 896 CTLFLAG_PERMANENT, 897 CTLTYPE_INT, "drops", 898 SYSCTL_DESCR("Total packets dropped due to full input queue"), 899 sysctl_percpuq_drops_handler, 0, (void *)ipq, 0, 900 CTL_CREATE, CTL_EOL) != 0) 901 goto bad; 902 903 return; 904 bad: 905 printf("%s: could not attach sysctl nodes\n", ifname); 906 return; 907 } 908 909 910 /* 911 * The common interface input routine that is called by device drivers, 912 * which should be used only when the driver's rx handler already runs 913 * in softint. 914 */ 915 void 916 if_input(struct ifnet *ifp, struct mbuf *m) 917 { 918 919 KASSERT(ifp->if_percpuq == NULL); 920 KASSERT(!cpu_intr_p()); 921 922 ifp->_if_input(ifp, m); 923 } 924 925 /* 926 * DEPRECATED. Use if_initialize and if_register instead. 927 * See the above comment of if_initialize. 928 * 929 * Note that it implicitly enables if_percpuq to make drivers easy to 930 * migrate softint-based if_input without much changes. If you don't 931 * want to enable it, use if_initialize instead. 932 */ 933 void 934 if_attach(ifnet_t *ifp) 935 { 936 937 if_initialize(ifp); 938 ifp->if_percpuq = if_percpuq_create(ifp); 939 if_register(ifp); 940 } 941 942 void 943 if_attachdomain(void) 944 { 945 struct ifnet *ifp; 946 int s; 947 int bound = curlwp->l_pflag & LP_BOUND; 948 949 curlwp->l_pflag |= LP_BOUND; 950 s = pserialize_read_enter(); 951 IFNET_READER_FOREACH(ifp) { 952 struct psref psref; 953 psref_acquire(&psref, &ifp->if_psref, ifnet_psref_class); 954 pserialize_read_exit(s); 955 if_attachdomain1(ifp); 956 s = pserialize_read_enter(); 957 psref_release(&psref, &ifp->if_psref, ifnet_psref_class); 958 } 959 pserialize_read_exit(s); 960 curlwp->l_pflag ^= bound ^ LP_BOUND; 961 } 962 963 static void 964 if_attachdomain1(struct ifnet *ifp) 965 { 966 struct domain *dp; 967 int s; 968 969 s = splnet(); 970 971 /* address family dependent data region */ 972 memset(ifp->if_afdata, 0, sizeof(ifp->if_afdata)); 973 DOMAIN_FOREACH(dp) { 974 if (dp->dom_ifattach != NULL) 975 ifp->if_afdata[dp->dom_family] = 976 (*dp->dom_ifattach)(ifp); 977 } 978 979 splx(s); 980 } 981 982 /* 983 * Deactivate an interface. This points all of the procedure 984 * handles at error stubs. May be called from interrupt context. 985 */ 986 void 987 if_deactivate(struct ifnet *ifp) 988 { 989 int s; 990 991 s = splnet(); 992 993 ifp->if_output = if_nulloutput; 994 ifp->_if_input = if_nullinput; 995 ifp->if_start = if_nullstart; 996 ifp->if_transmit = if_nulltransmit; 997 ifp->if_ioctl = if_nullioctl; 998 ifp->if_init = if_nullinit; 999 ifp->if_stop = if_nullstop; 1000 ifp->if_slowtimo = if_nullslowtimo; 1001 ifp->if_drain = if_nulldrain; 1002 1003 /* No more packets may be enqueued. */ 1004 ifp->if_snd.ifq_maxlen = 0; 1005 1006 splx(s); 1007 } 1008 1009 void 1010 if_purgeaddrs(struct ifnet *ifp, int family, void (*purgeaddr)(struct ifaddr *)) 1011 { 1012 struct ifaddr *ifa, *nifa; 1013 1014 IFADDR_FOREACH_SAFE(ifa, ifp, nifa) { 1015 if (ifa->ifa_addr->sa_family != family) 1016 continue; 1017 (*purgeaddr)(ifa); 1018 } 1019 } 1020 1021 /* 1022 * Detach an interface from the list of "active" interfaces, 1023 * freeing any resources as we go along. 1024 * 1025 * NOTE: This routine must be called with a valid thread context, 1026 * as it may block. 1027 */ 1028 void 1029 if_detach(struct ifnet *ifp) 1030 { 1031 struct socket so; 1032 struct ifaddr *ifa; 1033 #ifdef IFAREF_DEBUG 1034 struct ifaddr *last_ifa = NULL; 1035 #endif 1036 struct domain *dp; 1037 const struct protosw *pr; 1038 int s, i, family, purged; 1039 uint64_t xc; 1040 1041 /* 1042 * XXX It's kind of lame that we have to have the 1043 * XXX socket structure... 1044 */ 1045 memset(&so, 0, sizeof(so)); 1046 1047 s = splnet(); 1048 1049 sysctl_teardown(&ifp->if_sysctl_log); 1050 1051 IFNET_LOCK(); 1052 ifindex2ifnet[ifp->if_index] = NULL; 1053 TAILQ_REMOVE(&ifnet_list, ifp, if_list); 1054 IFNET_WRITER_REMOVE(ifp); 1055 pserialize_perform(ifnet_psz); 1056 IFNET_UNLOCK(); 1057 1058 /* Wait for all readers to drain before freeing. */ 1059 psref_target_destroy(&ifp->if_psref, ifnet_psref_class); 1060 PSLIST_ENTRY_DESTROY(ifp, if_pslist_entry); 1061 1062 if (ifp->if_slowtimo != NULL) { 1063 ifp->if_slowtimo = NULL; 1064 callout_halt(ifp->if_slowtimo_ch, NULL); 1065 callout_destroy(ifp->if_slowtimo_ch); 1066 kmem_free(ifp->if_slowtimo_ch, sizeof(*ifp->if_slowtimo_ch)); 1067 } 1068 1069 /* 1070 * Do an if_down() to give protocols a chance to do something. 1071 */ 1072 if_down(ifp); 1073 1074 #ifdef ALTQ 1075 if (ALTQ_IS_ENABLED(&ifp->if_snd)) 1076 altq_disable(&ifp->if_snd); 1077 if (ALTQ_IS_ATTACHED(&ifp->if_snd)) 1078 altq_detach(&ifp->if_snd); 1079 #endif 1080 1081 if (ifp->if_snd.ifq_lock) 1082 mutex_obj_free(ifp->if_snd.ifq_lock); 1083 1084 #if NCARP > 0 1085 /* Remove the interface from any carp group it is a part of. */ 1086 if (ifp->if_carp != NULL && ifp->if_type != IFT_CARP) 1087 carp_ifdetach(ifp); 1088 #endif 1089 1090 /* 1091 * Rip all the addresses off the interface. This should make 1092 * all of the routes go away. 1093 * 1094 * pr_usrreq calls can remove an arbitrary number of ifaddrs 1095 * from the list, including our "cursor", ifa. For safety, 1096 * and to honor the TAILQ abstraction, I just restart the 1097 * loop after each removal. Note that the loop will exit 1098 * when all of the remaining ifaddrs belong to the AF_LINK 1099 * family. I am counting on the historical fact that at 1100 * least one pr_usrreq in each address domain removes at 1101 * least one ifaddr. 1102 */ 1103 again: 1104 IFADDR_FOREACH(ifa, ifp) { 1105 family = ifa->ifa_addr->sa_family; 1106 #ifdef IFAREF_DEBUG 1107 printf("if_detach: ifaddr %p, family %d, refcnt %d\n", 1108 ifa, family, ifa->ifa_refcnt); 1109 if (last_ifa != NULL && ifa == last_ifa) 1110 panic("if_detach: loop detected"); 1111 last_ifa = ifa; 1112 #endif 1113 if (family == AF_LINK) 1114 continue; 1115 dp = pffinddomain(family); 1116 #ifdef DIAGNOSTIC 1117 if (dp == NULL) 1118 panic("if_detach: no domain for AF %d", 1119 family); 1120 #endif 1121 /* 1122 * XXX These PURGEIF calls are redundant with the 1123 * purge-all-families calls below, but are left in for 1124 * now both to make a smaller change, and to avoid 1125 * unplanned interactions with clearing of 1126 * ifp->if_addrlist. 1127 */ 1128 purged = 0; 1129 for (pr = dp->dom_protosw; 1130 pr < dp->dom_protoswNPROTOSW; pr++) { 1131 so.so_proto = pr; 1132 if (pr->pr_usrreqs) { 1133 (void) (*pr->pr_usrreqs->pr_purgeif)(&so, ifp); 1134 purged = 1; 1135 } 1136 } 1137 if (purged == 0) { 1138 /* 1139 * XXX What's really the best thing to do 1140 * XXX here? --thorpej (at) NetBSD.org 1141 */ 1142 printf("if_detach: WARNING: AF %d not purged\n", 1143 family); 1144 ifa_remove(ifp, ifa); 1145 } 1146 goto again; 1147 } 1148 1149 if_free_sadl(ifp); 1150 1151 /* Walk the routing table looking for stragglers. */ 1152 for (i = 0; i <= AF_MAX; i++) { 1153 while (rt_walktree(i, if_rt_walktree, ifp) == ERESTART) 1154 continue; 1155 } 1156 1157 DOMAIN_FOREACH(dp) { 1158 if (dp->dom_ifdetach != NULL && ifp->if_afdata[dp->dom_family]) 1159 { 1160 void *p = ifp->if_afdata[dp->dom_family]; 1161 if (p) { 1162 ifp->if_afdata[dp->dom_family] = NULL; 1163 (*dp->dom_ifdetach)(ifp, p); 1164 } 1165 } 1166 1167 /* 1168 * One would expect multicast memberships (INET and 1169 * INET6) on UDP sockets to be purged by the PURGEIF 1170 * calls above, but if all addresses were removed from 1171 * the interface prior to destruction, the calls will 1172 * not be made (e.g. ppp, for which pppd(8) generally 1173 * removes addresses before destroying the interface). 1174 * Because there is no invariant that multicast 1175 * memberships only exist for interfaces with IPv4 1176 * addresses, we must call PURGEIF regardless of 1177 * addresses. (Protocols which might store ifnet 1178 * pointers are marked with PR_PURGEIF.) 1179 */ 1180 for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) { 1181 so.so_proto = pr; 1182 if (pr->pr_usrreqs && pr->pr_flags & PR_PURGEIF) 1183 (void)(*pr->pr_usrreqs->pr_purgeif)(&so, ifp); 1184 } 1185 } 1186 1187 (void)pfil_run_hooks(if_pfil, 1188 (struct mbuf **)PFIL_IFNET_DETACH, ifp, PFIL_IFNET); 1189 (void)pfil_head_destroy(ifp->if_pfil); 1190 1191 /* Announce that the interface is gone. */ 1192 rt_ifannouncemsg(ifp, IFAN_DEPARTURE); 1193 1194 ifioctl_detach(ifp); 1195 1196 IF_AFDATA_LOCK_DESTROY(ifp); 1197 1198 softint_disestablish(ifp->if_link_si); 1199 ifp->if_link_si = NULL; 1200 1201 /* 1202 * remove packets that came from ifp, from software interrupt queues. 1203 */ 1204 DOMAIN_FOREACH(dp) { 1205 for (i = 0; i < __arraycount(dp->dom_ifqueues); i++) { 1206 struct ifqueue *iq = dp->dom_ifqueues[i]; 1207 if (iq == NULL) 1208 break; 1209 dp->dom_ifqueues[i] = NULL; 1210 if_detach_queues(ifp, iq); 1211 } 1212 } 1213 1214 /* 1215 * IP queues have to be processed separately: net-queue barrier 1216 * ensures that the packets are dequeued while a cross-call will 1217 * ensure that the interrupts have completed. FIXME: not quite.. 1218 */ 1219 #ifdef INET 1220 pktq_barrier(ip_pktq); 1221 #endif 1222 #ifdef INET6 1223 if (in6_present) 1224 pktq_barrier(ip6_pktq); 1225 #endif 1226 xc = xc_broadcast(0, (xcfunc_t)nullop, NULL, NULL); 1227 xc_wait(xc); 1228 1229 if (ifp->if_percpuq != NULL) { 1230 if_percpuq_destroy(ifp->if_percpuq); 1231 ifp->if_percpuq = NULL; 1232 } 1233 1234 splx(s); 1235 } 1236 1237 static void 1238 if_detach_queues(struct ifnet *ifp, struct ifqueue *q) 1239 { 1240 struct mbuf *m, *prev, *next; 1241 1242 prev = NULL; 1243 for (m = q->ifq_head; m != NULL; m = next) { 1244 KASSERT((m->m_flags & M_PKTHDR) != 0); 1245 1246 next = m->m_nextpkt; 1247 if (m->m_pkthdr.rcvif != ifp) { 1248 prev = m; 1249 continue; 1250 } 1251 1252 if (prev != NULL) 1253 prev->m_nextpkt = m->m_nextpkt; 1254 else 1255 q->ifq_head = m->m_nextpkt; 1256 if (q->ifq_tail == m) 1257 q->ifq_tail = prev; 1258 q->ifq_len--; 1259 1260 m->m_nextpkt = NULL; 1261 m_freem(m); 1262 IF_DROP(q); 1263 } 1264 } 1265 1266 /* 1267 * Callback for a radix tree walk to delete all references to an 1268 * ifnet. 1269 */ 1270 static int 1271 if_rt_walktree(struct rtentry *rt, void *v) 1272 { 1273 struct ifnet *ifp = (struct ifnet *)v; 1274 int error; 1275 struct rtentry *retrt; 1276 1277 if (rt->rt_ifp != ifp) 1278 return 0; 1279 1280 /* Delete the entry. */ 1281 error = rtrequest(RTM_DELETE, rt_getkey(rt), rt->rt_gateway, 1282 rt_mask(rt), rt->rt_flags, &retrt); 1283 if (error == 0) { 1284 KASSERT(retrt == rt); 1285 KASSERT((retrt->rt_flags & RTF_UP) == 0); 1286 retrt->rt_ifp = NULL; 1287 rtfree(retrt); 1288 } else { 1289 printf("%s: warning: unable to delete rtentry @ %p, " 1290 "error = %d\n", ifp->if_xname, rt, error); 1291 } 1292 return ERESTART; 1293 } 1294 1295 /* 1296 * Create a clone network interface. 1297 */ 1298 static int 1299 if_clone_create(const char *name) 1300 { 1301 struct if_clone *ifc; 1302 int unit; 1303 1304 ifc = if_clone_lookup(name, &unit); 1305 if (ifc == NULL) 1306 return EINVAL; 1307 1308 if (ifunit(name) != NULL) 1309 return EEXIST; 1310 1311 return (*ifc->ifc_create)(ifc, unit); 1312 } 1313 1314 /* 1315 * Destroy a clone network interface. 1316 */ 1317 static int 1318 if_clone_destroy(const char *name) 1319 { 1320 struct if_clone *ifc; 1321 struct ifnet *ifp; 1322 1323 ifc = if_clone_lookup(name, NULL); 1324 if (ifc == NULL) 1325 return EINVAL; 1326 1327 ifp = ifunit(name); 1328 if (ifp == NULL) 1329 return ENXIO; 1330 1331 if (ifc->ifc_destroy == NULL) 1332 return EOPNOTSUPP; 1333 1334 return (*ifc->ifc_destroy)(ifp); 1335 } 1336 1337 /* 1338 * Look up a network interface cloner. 1339 */ 1340 static struct if_clone * 1341 if_clone_lookup(const char *name, int *unitp) 1342 { 1343 struct if_clone *ifc; 1344 const char *cp; 1345 char *dp, ifname[IFNAMSIZ + 3]; 1346 int unit; 1347 1348 strcpy(ifname, "if_"); 1349 /* separate interface name from unit */ 1350 for (dp = ifname + 3, cp = name; cp - name < IFNAMSIZ && 1351 *cp && (*cp < '0' || *cp > '9');) 1352 *dp++ = *cp++; 1353 1354 if (cp == name || cp - name == IFNAMSIZ || !*cp) 1355 return NULL; /* No name or unit number */ 1356 *dp++ = '\0'; 1357 1358 again: 1359 LIST_FOREACH(ifc, &if_cloners, ifc_list) { 1360 if (strcmp(ifname + 3, ifc->ifc_name) == 0) 1361 break; 1362 } 1363 1364 if (ifc == NULL) { 1365 if (*ifname == '\0' || 1366 module_autoload(ifname, MODULE_CLASS_DRIVER)) 1367 return NULL; 1368 *ifname = '\0'; 1369 goto again; 1370 } 1371 1372 unit = 0; 1373 while (cp - name < IFNAMSIZ && *cp) { 1374 if (*cp < '0' || *cp > '9' || unit >= INT_MAX / 10) { 1375 /* Bogus unit number. */ 1376 return NULL; 1377 } 1378 unit = (unit * 10) + (*cp++ - '0'); 1379 } 1380 1381 if (unitp != NULL) 1382 *unitp = unit; 1383 return ifc; 1384 } 1385 1386 /* 1387 * Register a network interface cloner. 1388 */ 1389 void 1390 if_clone_attach(struct if_clone *ifc) 1391 { 1392 1393 LIST_INSERT_HEAD(&if_cloners, ifc, ifc_list); 1394 if_cloners_count++; 1395 } 1396 1397 /* 1398 * Unregister a network interface cloner. 1399 */ 1400 void 1401 if_clone_detach(struct if_clone *ifc) 1402 { 1403 1404 LIST_REMOVE(ifc, ifc_list); 1405 if_cloners_count--; 1406 } 1407 1408 /* 1409 * Provide list of interface cloners to userspace. 1410 */ 1411 int 1412 if_clone_list(int buf_count, char *buffer, int *total) 1413 { 1414 char outbuf[IFNAMSIZ], *dst; 1415 struct if_clone *ifc; 1416 int count, error = 0; 1417 1418 *total = if_cloners_count; 1419 if ((dst = buffer) == NULL) { 1420 /* Just asking how many there are. */ 1421 return 0; 1422 } 1423 1424 if (buf_count < 0) 1425 return EINVAL; 1426 1427 count = (if_cloners_count < buf_count) ? 1428 if_cloners_count : buf_count; 1429 1430 for (ifc = LIST_FIRST(&if_cloners); ifc != NULL && count != 0; 1431 ifc = LIST_NEXT(ifc, ifc_list), count--, dst += IFNAMSIZ) { 1432 (void)strncpy(outbuf, ifc->ifc_name, sizeof(outbuf)); 1433 if (outbuf[sizeof(outbuf) - 1] != '\0') 1434 return ENAMETOOLONG; 1435 error = copyout(outbuf, dst, sizeof(outbuf)); 1436 if (error != 0) 1437 break; 1438 } 1439 1440 return error; 1441 } 1442 1443 void 1444 ifaref(struct ifaddr *ifa) 1445 { 1446 ifa->ifa_refcnt++; 1447 } 1448 1449 void 1450 ifafree(struct ifaddr *ifa) 1451 { 1452 KASSERT(ifa != NULL); 1453 KASSERT(ifa->ifa_refcnt > 0); 1454 1455 if (--ifa->ifa_refcnt == 0) { 1456 free(ifa, M_IFADDR); 1457 } 1458 } 1459 1460 void 1461 ifa_insert(struct ifnet *ifp, struct ifaddr *ifa) 1462 { 1463 ifa->ifa_ifp = ifp; 1464 TAILQ_INSERT_TAIL(&ifp->if_addrlist, ifa, ifa_list); 1465 ifaref(ifa); 1466 } 1467 1468 void 1469 ifa_remove(struct ifnet *ifp, struct ifaddr *ifa) 1470 { 1471 KASSERT(ifa->ifa_ifp == ifp); 1472 TAILQ_REMOVE(&ifp->if_addrlist, ifa, ifa_list); 1473 ifafree(ifa); 1474 } 1475 1476 static inline int 1477 equal(const struct sockaddr *sa1, const struct sockaddr *sa2) 1478 { 1479 return sockaddr_cmp(sa1, sa2) == 0; 1480 } 1481 1482 /* 1483 * Locate an interface based on a complete address. 1484 */ 1485 /*ARGSUSED*/ 1486 struct ifaddr * 1487 ifa_ifwithaddr(const struct sockaddr *addr) 1488 { 1489 struct ifnet *ifp; 1490 struct ifaddr *ifa; 1491 int s; 1492 1493 s = pserialize_read_enter(); 1494 IFNET_READER_FOREACH(ifp) { 1495 if (ifp->if_output == if_nulloutput) 1496 continue; 1497 IFADDR_FOREACH(ifa, ifp) { 1498 if (ifa->ifa_addr->sa_family != addr->sa_family) 1499 continue; 1500 if (equal(addr, ifa->ifa_addr)) 1501 return ifa; 1502 if ((ifp->if_flags & IFF_BROADCAST) && 1503 ifa->ifa_broadaddr && 1504 /* IP6 doesn't have broadcast */ 1505 ifa->ifa_broadaddr->sa_len != 0 && 1506 equal(ifa->ifa_broadaddr, addr)) 1507 return ifa; 1508 } 1509 } 1510 pserialize_read_exit(s); 1511 return NULL; 1512 } 1513 1514 /* 1515 * Locate the point to point interface with a given destination address. 1516 */ 1517 /*ARGSUSED*/ 1518 struct ifaddr * 1519 ifa_ifwithdstaddr(const struct sockaddr *addr) 1520 { 1521 struct ifnet *ifp; 1522 struct ifaddr *ifa; 1523 int s; 1524 1525 s = pserialize_read_enter(); 1526 IFNET_READER_FOREACH(ifp) { 1527 if (ifp->if_output == if_nulloutput) 1528 continue; 1529 if ((ifp->if_flags & IFF_POINTOPOINT) == 0) 1530 continue; 1531 IFADDR_FOREACH(ifa, ifp) { 1532 if (ifa->ifa_addr->sa_family != addr->sa_family || 1533 ifa->ifa_dstaddr == NULL) 1534 continue; 1535 if (equal(addr, ifa->ifa_dstaddr)) 1536 return ifa; 1537 } 1538 } 1539 pserialize_read_exit(s); 1540 return NULL; 1541 } 1542 1543 /* 1544 * Find an interface on a specific network. If many, choice 1545 * is most specific found. 1546 */ 1547 struct ifaddr * 1548 ifa_ifwithnet(const struct sockaddr *addr) 1549 { 1550 struct ifnet *ifp; 1551 struct ifaddr *ifa; 1552 const struct sockaddr_dl *sdl; 1553 struct ifaddr *ifa_maybe = 0; 1554 u_int af = addr->sa_family; 1555 const char *addr_data = addr->sa_data, *cplim; 1556 int s; 1557 1558 if (af == AF_LINK) { 1559 sdl = satocsdl(addr); 1560 if (sdl->sdl_index && sdl->sdl_index < if_indexlim && 1561 ifindex2ifnet[sdl->sdl_index] && 1562 ifindex2ifnet[sdl->sdl_index]->if_output != if_nulloutput) { 1563 return ifindex2ifnet[sdl->sdl_index]->if_dl; 1564 } 1565 } 1566 #ifdef NETATALK 1567 if (af == AF_APPLETALK) { 1568 const struct sockaddr_at *sat, *sat2; 1569 sat = (const struct sockaddr_at *)addr; 1570 s = pserialize_read_enter(); 1571 IFNET_READER_FOREACH(ifp) { 1572 if (ifp->if_output == if_nulloutput) 1573 continue; 1574 ifa = at_ifawithnet((const struct sockaddr_at *)addr, ifp); 1575 if (ifa == NULL) 1576 continue; 1577 sat2 = (struct sockaddr_at *)ifa->ifa_addr; 1578 if (sat2->sat_addr.s_net == sat->sat_addr.s_net) 1579 return ifa; /* exact match */ 1580 if (ifa_maybe == NULL) { 1581 /* else keep the if with the right range */ 1582 ifa_maybe = ifa; 1583 } 1584 } 1585 pserialize_read_exit(s); 1586 return ifa_maybe; 1587 } 1588 #endif 1589 s = pserialize_read_enter(); 1590 IFNET_READER_FOREACH(ifp) { 1591 if (ifp->if_output == if_nulloutput) 1592 continue; 1593 IFADDR_FOREACH(ifa, ifp) { 1594 const char *cp, *cp2, *cp3; 1595 1596 if (ifa->ifa_addr->sa_family != af || 1597 ifa->ifa_netmask == NULL) 1598 next: continue; 1599 cp = addr_data; 1600 cp2 = ifa->ifa_addr->sa_data; 1601 cp3 = ifa->ifa_netmask->sa_data; 1602 cplim = (const char *)ifa->ifa_netmask + 1603 ifa->ifa_netmask->sa_len; 1604 while (cp3 < cplim) { 1605 if ((*cp++ ^ *cp2++) & *cp3++) { 1606 /* want to continue for() loop */ 1607 goto next; 1608 } 1609 } 1610 if (ifa_maybe == NULL || 1611 rt_refines(ifa->ifa_netmask, 1612 ifa_maybe->ifa_netmask)) 1613 ifa_maybe = ifa; 1614 } 1615 } 1616 pserialize_read_exit(s); 1617 return ifa_maybe; 1618 } 1619 1620 /* 1621 * Find the interface of the addresss. 1622 */ 1623 struct ifaddr * 1624 ifa_ifwithladdr(const struct sockaddr *addr) 1625 { 1626 struct ifaddr *ia; 1627 1628 if ((ia = ifa_ifwithaddr(addr)) || (ia = ifa_ifwithdstaddr(addr)) || 1629 (ia = ifa_ifwithnet(addr))) 1630 return ia; 1631 return NULL; 1632 } 1633 1634 /* 1635 * Find an interface using a specific address family 1636 */ 1637 struct ifaddr * 1638 ifa_ifwithaf(int af) 1639 { 1640 struct ifnet *ifp; 1641 struct ifaddr *ifa = NULL; 1642 int s; 1643 1644 s = pserialize_read_enter(); 1645 IFNET_READER_FOREACH(ifp) { 1646 if (ifp->if_output == if_nulloutput) 1647 continue; 1648 IFADDR_FOREACH(ifa, ifp) { 1649 if (ifa->ifa_addr->sa_family == af) 1650 goto out; 1651 } 1652 } 1653 out: 1654 pserialize_read_exit(s); 1655 return ifa; 1656 } 1657 1658 /* 1659 * Find an interface address specific to an interface best matching 1660 * a given address. 1661 */ 1662 struct ifaddr * 1663 ifaof_ifpforaddr(const struct sockaddr *addr, struct ifnet *ifp) 1664 { 1665 struct ifaddr *ifa; 1666 const char *cp, *cp2, *cp3; 1667 const char *cplim; 1668 struct ifaddr *ifa_maybe = 0; 1669 u_int af = addr->sa_family; 1670 1671 if (ifp->if_output == if_nulloutput) 1672 return NULL; 1673 1674 if (af >= AF_MAX) 1675 return NULL; 1676 1677 IFADDR_FOREACH(ifa, ifp) { 1678 if (ifa->ifa_addr->sa_family != af) 1679 continue; 1680 ifa_maybe = ifa; 1681 if (ifa->ifa_netmask == NULL) { 1682 if (equal(addr, ifa->ifa_addr) || 1683 (ifa->ifa_dstaddr && 1684 equal(addr, ifa->ifa_dstaddr))) 1685 return ifa; 1686 continue; 1687 } 1688 cp = addr->sa_data; 1689 cp2 = ifa->ifa_addr->sa_data; 1690 cp3 = ifa->ifa_netmask->sa_data; 1691 cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask; 1692 for (; cp3 < cplim; cp3++) { 1693 if ((*cp++ ^ *cp2++) & *cp3) 1694 break; 1695 } 1696 if (cp3 == cplim) 1697 return ifa; 1698 } 1699 return ifa_maybe; 1700 } 1701 1702 /* 1703 * Default action when installing a route with a Link Level gateway. 1704 * Lookup an appropriate real ifa to point to. 1705 * This should be moved to /sys/net/link.c eventually. 1706 */ 1707 void 1708 link_rtrequest(int cmd, struct rtentry *rt, const struct rt_addrinfo *info) 1709 { 1710 struct ifaddr *ifa; 1711 const struct sockaddr *dst; 1712 struct ifnet *ifp; 1713 1714 if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL || 1715 (ifp = ifa->ifa_ifp) == NULL || (dst = rt_getkey(rt)) == NULL) 1716 return; 1717 if ((ifa = ifaof_ifpforaddr(dst, ifp)) != NULL) { 1718 rt_replace_ifa(rt, ifa); 1719 if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest) 1720 ifa->ifa_rtrequest(cmd, rt, info); 1721 } 1722 } 1723 1724 /* 1725 * bitmask macros to manage a densely packed link_state change queue. 1726 * Because we need to store LINK_STATE_UNKNOWN(0), LINK_STATE_DOWN(1) and 1727 * LINK_STATE_UP(2) we need 2 bits for each state change. 1728 * As a state change to store is 0, treat all bits set as an unset item. 1729 */ 1730 #define LQ_ITEM_BITS 2 1731 #define LQ_ITEM_MASK ((1 << LQ_ITEM_BITS) - 1) 1732 #define LQ_MASK(i) (LQ_ITEM_MASK << (i) * LQ_ITEM_BITS) 1733 #define LINK_STATE_UNSET LQ_ITEM_MASK 1734 #define LQ_ITEM(q, i) (((q) & LQ_MASK((i))) >> (i) * LQ_ITEM_BITS) 1735 #define LQ_STORE(q, i, v) \ 1736 do { \ 1737 (q) &= ~LQ_MASK((i)); \ 1738 (q) |= (v) << (i) * LQ_ITEM_BITS; \ 1739 } while (0 /* CONSTCOND */) 1740 #define LQ_MAX(q) ((sizeof((q)) * NBBY) / LQ_ITEM_BITS) 1741 #define LQ_POP(q, v) \ 1742 do { \ 1743 (v) = LQ_ITEM((q), 0); \ 1744 (q) >>= LQ_ITEM_BITS; \ 1745 (q) |= LINK_STATE_UNSET << (LQ_MAX((q)) - 1) * LQ_ITEM_BITS; \ 1746 } while (0 /* CONSTCOND */) 1747 #define LQ_PUSH(q, v) \ 1748 do { \ 1749 (q) >>= LQ_ITEM_BITS; \ 1750 (q) |= (v) << (LQ_MAX((q)) - 1) * LQ_ITEM_BITS; \ 1751 } while (0 /* CONSTCOND */) 1752 #define LQ_FIND_UNSET(q, i) \ 1753 for ((i) = 0; i < LQ_MAX((q)); (i)++) { \ 1754 if (LQ_ITEM((q), (i)) == LINK_STATE_UNSET) \ 1755 break; \ 1756 } 1757 /* 1758 * Handle a change in the interface link state and 1759 * queue notifications. 1760 */ 1761 void 1762 if_link_state_change(struct ifnet *ifp, int link_state) 1763 { 1764 int s, idx; 1765 1766 /* Ensure change is to a valid state */ 1767 switch (link_state) { 1768 case LINK_STATE_UNKNOWN: /* FALLTHROUGH */ 1769 case LINK_STATE_DOWN: /* FALLTHROUGH */ 1770 case LINK_STATE_UP: 1771 break; 1772 default: 1773 #ifdef DEBUG 1774 printf("%s: invalid link state %d\n", 1775 ifp->if_xname, link_state); 1776 #endif 1777 return; 1778 } 1779 1780 s = splnet(); 1781 1782 /* Find the last unset event in the queue. */ 1783 LQ_FIND_UNSET(ifp->if_link_queue, idx); 1784 1785 /* 1786 * Ensure link_state doesn't match the last event in the queue. 1787 * ifp->if_link_state is not checked and set here because 1788 * that would present an inconsistent picture to the system. 1789 */ 1790 if (idx != 0 && 1791 LQ_ITEM(ifp->if_link_queue, idx - 1) == (uint8_t)link_state) 1792 goto out; 1793 1794 /* Handle queue overflow. */ 1795 if (idx == LQ_MAX(ifp->if_link_queue)) { 1796 uint8_t lost; 1797 1798 /* 1799 * The DOWN state must be protected from being pushed off 1800 * the queue to ensure that userland will always be 1801 * in a sane state. 1802 * Because DOWN is protected, there is no need to protect 1803 * UNKNOWN. 1804 * It should be invalid to change from any other state to 1805 * UNKNOWN anyway ... 1806 */ 1807 lost = LQ_ITEM(ifp->if_link_queue, 0); 1808 LQ_PUSH(ifp->if_link_queue, (uint8_t)link_state); 1809 if (lost == LINK_STATE_DOWN) { 1810 lost = LQ_ITEM(ifp->if_link_queue, 0); 1811 LQ_STORE(ifp->if_link_queue, 0, LINK_STATE_DOWN); 1812 } 1813 printf("%s: lost link state change %s\n", 1814 ifp->if_xname, 1815 lost == LINK_STATE_UP ? "UP" : 1816 lost == LINK_STATE_DOWN ? "DOWN" : 1817 "UNKNOWN"); 1818 } else 1819 LQ_STORE(ifp->if_link_queue, idx, (uint8_t)link_state); 1820 1821 softint_schedule(ifp->if_link_si); 1822 1823 out: 1824 splx(s); 1825 } 1826 1827 /* 1828 * Handle interface link state change notifications. 1829 * Must be called at splnet(). 1830 */ 1831 static void 1832 if_link_state_change0(struct ifnet *ifp, int link_state) 1833 { 1834 struct domain *dp; 1835 1836 /* Ensure the change is still valid. */ 1837 if (ifp->if_link_state == link_state) 1838 return; 1839 1840 #ifdef DEBUG 1841 log(LOG_DEBUG, "%s: link state %s (was %s)\n", ifp->if_xname, 1842 link_state == LINK_STATE_UP ? "UP" : 1843 link_state == LINK_STATE_DOWN ? "DOWN" : 1844 "UNKNOWN", 1845 ifp->if_link_state == LINK_STATE_UP ? "UP" : 1846 ifp->if_link_state == LINK_STATE_DOWN ? "DOWN" : 1847 "UNKNOWN"); 1848 #endif 1849 1850 /* 1851 * When going from UNKNOWN to UP, we need to mark existing 1852 * addresses as tentative and restart DAD as we may have 1853 * erroneously not found a duplicate. 1854 * 1855 * This needs to happen before rt_ifmsg to avoid a race where 1856 * listeners would have an address and expect it to work right 1857 * away. 1858 */ 1859 if (link_state == LINK_STATE_UP && 1860 ifp->if_link_state == LINK_STATE_UNKNOWN) 1861 { 1862 DOMAIN_FOREACH(dp) { 1863 if (dp->dom_if_link_state_change != NULL) 1864 dp->dom_if_link_state_change(ifp, 1865 LINK_STATE_DOWN); 1866 } 1867 } 1868 1869 ifp->if_link_state = link_state; 1870 1871 /* Notify that the link state has changed. */ 1872 rt_ifmsg(ifp); 1873 1874 #if NCARP > 0 1875 if (ifp->if_carp) 1876 carp_carpdev_state(ifp); 1877 #endif 1878 1879 DOMAIN_FOREACH(dp) { 1880 if (dp->dom_if_link_state_change != NULL) 1881 dp->dom_if_link_state_change(ifp, link_state); 1882 } 1883 } 1884 1885 /* 1886 * Process the interface link state change queue. 1887 */ 1888 static void 1889 if_link_state_change_si(void *arg) 1890 { 1891 struct ifnet *ifp = arg; 1892 int s; 1893 uint8_t state; 1894 1895 s = splnet(); 1896 1897 /* Pop a link state change from the queue and process it. */ 1898 LQ_POP(ifp->if_link_queue, state); 1899 if_link_state_change0(ifp, state); 1900 1901 /* If there is a link state change to come, schedule it. */ 1902 if (LQ_ITEM(ifp->if_link_queue, 0) != LINK_STATE_UNSET) 1903 softint_schedule(ifp->if_link_si); 1904 1905 splx(s); 1906 } 1907 1908 /* 1909 * Default action when installing a local route on a point-to-point 1910 * interface. 1911 */ 1912 void 1913 p2p_rtrequest(int req, struct rtentry *rt, 1914 __unused const struct rt_addrinfo *info) 1915 { 1916 struct ifnet *ifp = rt->rt_ifp; 1917 struct ifaddr *ifa, *lo0ifa; 1918 1919 switch (req) { 1920 case RTM_ADD: 1921 if ((rt->rt_flags & RTF_LOCAL) == 0) 1922 break; 1923 1924 IFADDR_FOREACH(ifa, ifp) { 1925 if (equal(rt_getkey(rt), ifa->ifa_addr)) 1926 break; 1927 } 1928 if (ifa == NULL) 1929 break; 1930 1931 /* 1932 * Ensure lo0 has an address of the same family. 1933 */ 1934 IFADDR_FOREACH(lo0ifa, lo0ifp) { 1935 if (lo0ifa->ifa_addr->sa_family == 1936 ifa->ifa_addr->sa_family) 1937 break; 1938 } 1939 if (lo0ifa == NULL) 1940 break; 1941 1942 rt->rt_ifp = lo0ifp; 1943 1944 /* 1945 * Make sure to set rt->rt_ifa to the interface 1946 * address we are using, otherwise we will have trouble 1947 * with source address selection. 1948 */ 1949 if (ifa != rt->rt_ifa) 1950 rt_replace_ifa(rt, ifa); 1951 break; 1952 case RTM_DELETE: 1953 default: 1954 break; 1955 } 1956 } 1957 1958 /* 1959 * Mark an interface down and notify protocols of 1960 * the transition. 1961 * NOTE: must be called at splsoftnet or equivalent. 1962 */ 1963 void 1964 if_down(struct ifnet *ifp) 1965 { 1966 struct ifaddr *ifa; 1967 struct domain *dp; 1968 1969 ifp->if_flags &= ~IFF_UP; 1970 nanotime(&ifp->if_lastchange); 1971 IFADDR_FOREACH(ifa, ifp) 1972 pfctlinput(PRC_IFDOWN, ifa->ifa_addr); 1973 IFQ_PURGE(&ifp->if_snd); 1974 #if NCARP > 0 1975 if (ifp->if_carp) 1976 carp_carpdev_state(ifp); 1977 #endif 1978 rt_ifmsg(ifp); 1979 DOMAIN_FOREACH(dp) { 1980 if (dp->dom_if_down) 1981 dp->dom_if_down(ifp); 1982 } 1983 } 1984 1985 /* 1986 * Mark an interface up and notify protocols of 1987 * the transition. 1988 * NOTE: must be called at splsoftnet or equivalent. 1989 */ 1990 void 1991 if_up(struct ifnet *ifp) 1992 { 1993 #ifdef notyet 1994 struct ifaddr *ifa; 1995 #endif 1996 struct domain *dp; 1997 1998 ifp->if_flags |= IFF_UP; 1999 nanotime(&ifp->if_lastchange); 2000 #ifdef notyet 2001 /* this has no effect on IP, and will kill all ISO connections XXX */ 2002 IFADDR_FOREACH(ifa, ifp) 2003 pfctlinput(PRC_IFUP, ifa->ifa_addr); 2004 #endif 2005 #if NCARP > 0 2006 if (ifp->if_carp) 2007 carp_carpdev_state(ifp); 2008 #endif 2009 rt_ifmsg(ifp); 2010 DOMAIN_FOREACH(dp) { 2011 if (dp->dom_if_up) 2012 dp->dom_if_up(ifp); 2013 } 2014 } 2015 2016 /* 2017 * Handle interface slowtimo timer routine. Called 2018 * from softclock, we decrement timer (if set) and 2019 * call the appropriate interface routine on expiration. 2020 */ 2021 static void 2022 if_slowtimo(void *arg) 2023 { 2024 void (*slowtimo)(struct ifnet *); 2025 struct ifnet *ifp = arg; 2026 int s; 2027 2028 slowtimo = ifp->if_slowtimo; 2029 if (__predict_false(slowtimo == NULL)) 2030 return; 2031 2032 s = splnet(); 2033 if (ifp->if_timer != 0 && --ifp->if_timer == 0) 2034 (*slowtimo)(ifp); 2035 2036 splx(s); 2037 2038 if (__predict_true(ifp->if_slowtimo != NULL)) 2039 callout_schedule(ifp->if_slowtimo_ch, hz / IFNET_SLOWHZ); 2040 } 2041 2042 /* 2043 * Set/clear promiscuous mode on interface ifp based on the truth value 2044 * of pswitch. The calls are reference counted so that only the first 2045 * "on" request actually has an effect, as does the final "off" request. 2046 * Results are undefined if the "off" and "on" requests are not matched. 2047 */ 2048 int 2049 ifpromisc(struct ifnet *ifp, int pswitch) 2050 { 2051 int pcount, ret; 2052 short nflags; 2053 2054 pcount = ifp->if_pcount; 2055 if (pswitch) { 2056 /* 2057 * Allow the device to be "placed" into promiscuous 2058 * mode even if it is not configured up. It will 2059 * consult IFF_PROMISC when it is brought up. 2060 */ 2061 if (ifp->if_pcount++ != 0) 2062 return 0; 2063 nflags = ifp->if_flags | IFF_PROMISC; 2064 } else { 2065 if (--ifp->if_pcount > 0) 2066 return 0; 2067 nflags = ifp->if_flags & ~IFF_PROMISC; 2068 } 2069 ret = if_flags_set(ifp, nflags); 2070 /* Restore interface state if not successful. */ 2071 if (ret != 0) { 2072 ifp->if_pcount = pcount; 2073 } 2074 return ret; 2075 } 2076 2077 /* 2078 * Map interface name to 2079 * interface structure pointer. 2080 */ 2081 struct ifnet * 2082 ifunit(const char *name) 2083 { 2084 struct ifnet *ifp; 2085 const char *cp = name; 2086 u_int unit = 0; 2087 u_int i; 2088 int s; 2089 2090 /* 2091 * If the entire name is a number, treat it as an ifindex. 2092 */ 2093 for (i = 0; i < IFNAMSIZ && *cp >= '0' && *cp <= '9'; i++, cp++) { 2094 unit = unit * 10 + (*cp - '0'); 2095 } 2096 2097 /* 2098 * If the number took all of the name, then it's a valid ifindex. 2099 */ 2100 if (i == IFNAMSIZ || (cp != name && *cp == '\0')) { 2101 if (unit >= if_indexlim) 2102 return NULL; 2103 ifp = ifindex2ifnet[unit]; 2104 if (ifp == NULL || ifp->if_output == if_nulloutput) 2105 return NULL; 2106 return ifp; 2107 } 2108 2109 ifp = NULL; 2110 s = pserialize_read_enter(); 2111 IFNET_READER_FOREACH(ifp) { 2112 if (ifp->if_output == if_nulloutput) 2113 continue; 2114 if (strcmp(ifp->if_xname, name) == 0) 2115 goto out; 2116 } 2117 out: 2118 pserialize_read_exit(s); 2119 return ifp; 2120 } 2121 2122 /* 2123 * Get a reference of an ifnet object by an interface name. 2124 * The returned reference is protected by psref(9). The caller 2125 * must release a returned reference by if_put after use. 2126 */ 2127 struct ifnet * 2128 if_get(const char *name, struct psref *psref) 2129 { 2130 struct ifnet *ifp; 2131 const char *cp = name; 2132 u_int unit = 0; 2133 u_int i; 2134 int s; 2135 2136 /* 2137 * If the entire name is a number, treat it as an ifindex. 2138 */ 2139 for (i = 0; i < IFNAMSIZ && *cp >= '0' && *cp <= '9'; i++, cp++) { 2140 unit = unit * 10 + (*cp - '0'); 2141 } 2142 2143 /* 2144 * If the number took all of the name, then it's a valid ifindex. 2145 */ 2146 if (i == IFNAMSIZ || (cp != name && *cp == '\0')) { 2147 if (unit >= if_indexlim) 2148 return NULL; 2149 ifp = ifindex2ifnet[unit]; 2150 if (ifp == NULL || ifp->if_output == if_nulloutput) 2151 return NULL; 2152 return ifp; 2153 } 2154 2155 ifp = NULL; 2156 s = pserialize_read_enter(); 2157 IFNET_READER_FOREACH(ifp) { 2158 if (ifp->if_output == if_nulloutput) 2159 continue; 2160 if (strcmp(ifp->if_xname, name) == 0) { 2161 psref_acquire(psref, &ifp->if_psref, 2162 ifnet_psref_class); 2163 goto out; 2164 } 2165 } 2166 out: 2167 pserialize_read_exit(s); 2168 return ifp; 2169 } 2170 2171 /* 2172 * Release a reference of an ifnet object given by if_get or 2173 * if_get_byindex. 2174 */ 2175 void 2176 if_put(const struct ifnet *ifp, struct psref *psref) 2177 { 2178 2179 psref_release(psref, &ifp->if_psref, ifnet_psref_class); 2180 } 2181 2182 ifnet_t * 2183 if_byindex(u_int idx) 2184 { 2185 return (idx < if_indexlim) ? ifindex2ifnet[idx] : NULL; 2186 } 2187 2188 /* 2189 * Get a reference of an ifnet object by an interface index. 2190 * The returned reference is protected by psref(9). The caller 2191 * must release a returned reference by if_put after use. 2192 */ 2193 ifnet_t * 2194 if_get_byindex(u_int idx, struct psref *psref) 2195 { 2196 ifnet_t *ifp; 2197 int s; 2198 2199 s = pserialize_read_enter(); 2200 ifp = (idx < if_indexlim) ? ifindex2ifnet[idx] : NULL; 2201 if (ifp != NULL) 2202 psref_acquire(psref, &ifp->if_psref, ifnet_psref_class); 2203 pserialize_read_exit(s); 2204 2205 return ifp; 2206 } 2207 2208 /* common */ 2209 int 2210 ifioctl_common(struct ifnet *ifp, u_long cmd, void *data) 2211 { 2212 int s; 2213 struct ifreq *ifr; 2214 struct ifcapreq *ifcr; 2215 struct ifdatareq *ifdr; 2216 2217 switch (cmd) { 2218 case SIOCSIFCAP: 2219 ifcr = data; 2220 if ((ifcr->ifcr_capenable & ~ifp->if_capabilities) != 0) 2221 return EINVAL; 2222 2223 if (ifcr->ifcr_capenable == ifp->if_capenable) 2224 return 0; 2225 2226 ifp->if_capenable = ifcr->ifcr_capenable; 2227 2228 /* Pre-compute the checksum flags mask. */ 2229 ifp->if_csum_flags_tx = 0; 2230 ifp->if_csum_flags_rx = 0; 2231 if (ifp->if_capenable & IFCAP_CSUM_IPv4_Tx) { 2232 ifp->if_csum_flags_tx |= M_CSUM_IPv4; 2233 } 2234 if (ifp->if_capenable & IFCAP_CSUM_IPv4_Rx) { 2235 ifp->if_csum_flags_rx |= M_CSUM_IPv4; 2236 } 2237 2238 if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Tx) { 2239 ifp->if_csum_flags_tx |= M_CSUM_TCPv4; 2240 } 2241 if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Rx) { 2242 ifp->if_csum_flags_rx |= M_CSUM_TCPv4; 2243 } 2244 2245 if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Tx) { 2246 ifp->if_csum_flags_tx |= M_CSUM_UDPv4; 2247 } 2248 if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Rx) { 2249 ifp->if_csum_flags_rx |= M_CSUM_UDPv4; 2250 } 2251 2252 if (ifp->if_capenable & IFCAP_CSUM_TCPv6_Tx) { 2253 ifp->if_csum_flags_tx |= M_CSUM_TCPv6; 2254 } 2255 if (ifp->if_capenable & IFCAP_CSUM_TCPv6_Rx) { 2256 ifp->if_csum_flags_rx |= M_CSUM_TCPv6; 2257 } 2258 2259 if (ifp->if_capenable & IFCAP_CSUM_UDPv6_Tx) { 2260 ifp->if_csum_flags_tx |= M_CSUM_UDPv6; 2261 } 2262 if (ifp->if_capenable & IFCAP_CSUM_UDPv6_Rx) { 2263 ifp->if_csum_flags_rx |= M_CSUM_UDPv6; 2264 } 2265 if (ifp->if_flags & IFF_UP) 2266 return ENETRESET; 2267 return 0; 2268 case SIOCSIFFLAGS: 2269 ifr = data; 2270 if (ifp->if_flags & IFF_UP && (ifr->ifr_flags & IFF_UP) == 0) { 2271 s = splnet(); 2272 if_down(ifp); 2273 splx(s); 2274 } 2275 if (ifr->ifr_flags & IFF_UP && (ifp->if_flags & IFF_UP) == 0) { 2276 s = splnet(); 2277 if_up(ifp); 2278 splx(s); 2279 } 2280 ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) | 2281 (ifr->ifr_flags &~ IFF_CANTCHANGE); 2282 break; 2283 case SIOCGIFFLAGS: 2284 ifr = data; 2285 ifr->ifr_flags = ifp->if_flags; 2286 break; 2287 2288 case SIOCGIFMETRIC: 2289 ifr = data; 2290 ifr->ifr_metric = ifp->if_metric; 2291 break; 2292 2293 case SIOCGIFMTU: 2294 ifr = data; 2295 ifr->ifr_mtu = ifp->if_mtu; 2296 break; 2297 2298 case SIOCGIFDLT: 2299 ifr = data; 2300 ifr->ifr_dlt = ifp->if_dlt; 2301 break; 2302 2303 case SIOCGIFCAP: 2304 ifcr = data; 2305 ifcr->ifcr_capabilities = ifp->if_capabilities; 2306 ifcr->ifcr_capenable = ifp->if_capenable; 2307 break; 2308 2309 case SIOCSIFMETRIC: 2310 ifr = data; 2311 ifp->if_metric = ifr->ifr_metric; 2312 break; 2313 2314 case SIOCGIFDATA: 2315 ifdr = data; 2316 ifdr->ifdr_data = ifp->if_data; 2317 break; 2318 2319 case SIOCGIFINDEX: 2320 ifr = data; 2321 ifr->ifr_index = ifp->if_index; 2322 break; 2323 2324 case SIOCZIFDATA: 2325 ifdr = data; 2326 ifdr->ifdr_data = ifp->if_data; 2327 /* 2328 * Assumes that the volatile counters that can be 2329 * zero'ed are at the end of if_data. 2330 */ 2331 memset(&ifp->if_data.ifi_ipackets, 0, sizeof(ifp->if_data) - 2332 offsetof(struct if_data, ifi_ipackets)); 2333 /* 2334 * The memset() clears to the bottm of if_data. In the area, 2335 * if_lastchange is included. Please be careful if new entry 2336 * will be added into if_data or rewite this. 2337 * 2338 * And also, update if_lastchnage. 2339 */ 2340 getnanotime(&ifp->if_lastchange); 2341 break; 2342 case SIOCSIFMTU: 2343 ifr = data; 2344 if (ifp->if_mtu == ifr->ifr_mtu) 2345 break; 2346 ifp->if_mtu = ifr->ifr_mtu; 2347 /* 2348 * If the link MTU changed, do network layer specific procedure. 2349 */ 2350 #ifdef INET6 2351 if (in6_present) 2352 nd6_setmtu(ifp); 2353 #endif 2354 return ENETRESET; 2355 default: 2356 return ENOTTY; 2357 } 2358 return 0; 2359 } 2360 2361 int 2362 ifaddrpref_ioctl(struct socket *so, u_long cmd, void *data, struct ifnet *ifp) 2363 { 2364 struct if_addrprefreq *ifap = (struct if_addrprefreq *)data; 2365 struct ifaddr *ifa; 2366 const struct sockaddr *any, *sa; 2367 union { 2368 struct sockaddr sa; 2369 struct sockaddr_storage ss; 2370 } u, v; 2371 2372 switch (cmd) { 2373 case SIOCSIFADDRPREF: 2374 if (kauth_authorize_network(curlwp->l_cred, KAUTH_NETWORK_INTERFACE, 2375 KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd, 2376 NULL) != 0) 2377 return EPERM; 2378 case SIOCGIFADDRPREF: 2379 break; 2380 default: 2381 return EOPNOTSUPP; 2382 } 2383 2384 /* sanity checks */ 2385 if (data == NULL || ifp == NULL) { 2386 panic("invalid argument to %s", __func__); 2387 /*NOTREACHED*/ 2388 } 2389 2390 /* address must be specified on ADD and DELETE */ 2391 sa = sstocsa(&ifap->ifap_addr); 2392 if (sa->sa_family != sofamily(so)) 2393 return EINVAL; 2394 if ((any = sockaddr_any(sa)) == NULL || sa->sa_len != any->sa_len) 2395 return EINVAL; 2396 2397 sockaddr_externalize(&v.sa, sizeof(v.ss), sa); 2398 2399 IFADDR_FOREACH(ifa, ifp) { 2400 if (ifa->ifa_addr->sa_family != sa->sa_family) 2401 continue; 2402 sockaddr_externalize(&u.sa, sizeof(u.ss), ifa->ifa_addr); 2403 if (sockaddr_cmp(&u.sa, &v.sa) == 0) 2404 break; 2405 } 2406 if (ifa == NULL) 2407 return EADDRNOTAVAIL; 2408 2409 switch (cmd) { 2410 case SIOCSIFADDRPREF: 2411 ifa->ifa_preference = ifap->ifap_preference; 2412 return 0; 2413 case SIOCGIFADDRPREF: 2414 /* fill in the if_laddrreq structure */ 2415 (void)sockaddr_copy(sstosa(&ifap->ifap_addr), 2416 sizeof(ifap->ifap_addr), ifa->ifa_addr); 2417 ifap->ifap_preference = ifa->ifa_preference; 2418 return 0; 2419 default: 2420 return EOPNOTSUPP; 2421 } 2422 } 2423 2424 static void 2425 ifnet_lock_enter(struct ifnet_lock *il) 2426 { 2427 uint64_t *nenter; 2428 2429 /* Before trying to acquire the mutex, increase the count of threads 2430 * who have entered or who wait to enter the critical section. 2431 * Avoid one costly locked memory transaction by keeping a count for 2432 * each CPU. 2433 */ 2434 nenter = percpu_getref(il->il_nenter); 2435 (*nenter)++; 2436 percpu_putref(il->il_nenter); 2437 mutex_enter(&il->il_lock); 2438 } 2439 2440 static void 2441 ifnet_lock_exit(struct ifnet_lock *il) 2442 { 2443 /* Increase the count of threads who have exited the critical 2444 * section. Increase while we still hold the lock. 2445 */ 2446 il->il_nexit++; 2447 mutex_exit(&il->il_lock); 2448 } 2449 2450 /* 2451 * Interface ioctls. 2452 */ 2453 static int 2454 doifioctl(struct socket *so, u_long cmd, void *data, struct lwp *l) 2455 { 2456 struct ifnet *ifp; 2457 struct ifreq *ifr; 2458 int error = 0; 2459 #if defined(COMPAT_OSOCK) || defined(COMPAT_OIFREQ) 2460 u_long ocmd = cmd; 2461 #endif 2462 short oif_flags; 2463 #ifdef COMPAT_OIFREQ 2464 struct ifreq ifrb; 2465 struct oifreq *oifr = NULL; 2466 #endif 2467 int r; 2468 2469 switch (cmd) { 2470 #ifdef COMPAT_OIFREQ 2471 case OSIOCGIFCONF: 2472 case OOSIOCGIFCONF: 2473 return compat_ifconf(cmd, data); 2474 #endif 2475 #ifdef COMPAT_OIFDATA 2476 case OSIOCGIFDATA: 2477 case OSIOCZIFDATA: 2478 return compat_ifdatareq(l, cmd, data); 2479 #endif 2480 case SIOCGIFCONF: 2481 return ifconf(cmd, data); 2482 case SIOCINITIFADDR: 2483 return EPERM; 2484 } 2485 2486 #ifdef COMPAT_OIFREQ 2487 cmd = compat_cvtcmd(cmd); 2488 if (cmd != ocmd) { 2489 oifr = data; 2490 data = ifr = &ifrb; 2491 ifreqo2n(oifr, ifr); 2492 } else 2493 #endif 2494 ifr = data; 2495 2496 ifp = ifunit(ifr->ifr_name); 2497 2498 switch (cmd) { 2499 case SIOCIFCREATE: 2500 case SIOCIFDESTROY: 2501 if (l != NULL) { 2502 error = kauth_authorize_network(l->l_cred, 2503 KAUTH_NETWORK_INTERFACE, 2504 KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, 2505 (void *)cmd, NULL); 2506 if (error != 0) 2507 return error; 2508 } 2509 mutex_enter(&if_clone_mtx); 2510 r = (cmd == SIOCIFCREATE) ? 2511 if_clone_create(ifr->ifr_name) : 2512 if_clone_destroy(ifr->ifr_name); 2513 mutex_exit(&if_clone_mtx); 2514 return r; 2515 2516 case SIOCIFGCLONERS: 2517 { 2518 struct if_clonereq *req = (struct if_clonereq *)data; 2519 return if_clone_list(req->ifcr_count, req->ifcr_buffer, 2520 &req->ifcr_total); 2521 } 2522 } 2523 2524 if (ifp == NULL) 2525 return ENXIO; 2526 2527 switch (cmd) { 2528 case SIOCALIFADDR: 2529 case SIOCDLIFADDR: 2530 case SIOCSIFADDRPREF: 2531 case SIOCSIFFLAGS: 2532 case SIOCSIFCAP: 2533 case SIOCSIFMETRIC: 2534 case SIOCZIFDATA: 2535 case SIOCSIFMTU: 2536 case SIOCSIFPHYADDR: 2537 case SIOCDIFPHYADDR: 2538 #ifdef INET6 2539 case SIOCSIFPHYADDR_IN6: 2540 #endif 2541 case SIOCSLIFPHYADDR: 2542 case SIOCADDMULTI: 2543 case SIOCDELMULTI: 2544 case SIOCSIFMEDIA: 2545 case SIOCSDRVSPEC: 2546 case SIOCG80211: 2547 case SIOCS80211: 2548 case SIOCS80211NWID: 2549 case SIOCS80211NWKEY: 2550 case SIOCS80211POWER: 2551 case SIOCS80211BSSID: 2552 case SIOCS80211CHANNEL: 2553 case SIOCSLINKSTR: 2554 if (l != NULL) { 2555 error = kauth_authorize_network(l->l_cred, 2556 KAUTH_NETWORK_INTERFACE, 2557 KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, 2558 (void *)cmd, NULL); 2559 if (error != 0) 2560 return error; 2561 } 2562 } 2563 2564 oif_flags = ifp->if_flags; 2565 2566 ifnet_lock_enter(ifp->if_ioctl_lock); 2567 error = (*ifp->if_ioctl)(ifp, cmd, data); 2568 if (error != ENOTTY) 2569 ; 2570 else if (so->so_proto == NULL) 2571 error = EOPNOTSUPP; 2572 else { 2573 #ifdef COMPAT_OSOCK 2574 error = compat_ifioctl(so, ocmd, cmd, data, l); 2575 #else 2576 error = (*so->so_proto->pr_usrreqs->pr_ioctl)(so, 2577 cmd, data, ifp); 2578 #endif 2579 } 2580 2581 if (((oif_flags ^ ifp->if_flags) & IFF_UP) != 0) { 2582 if ((ifp->if_flags & IFF_UP) != 0) { 2583 int s = splnet(); 2584 if_up(ifp); 2585 splx(s); 2586 } 2587 } 2588 #ifdef COMPAT_OIFREQ 2589 if (cmd != ocmd) 2590 ifreqn2o(oifr, ifr); 2591 #endif 2592 2593 ifnet_lock_exit(ifp->if_ioctl_lock); 2594 return error; 2595 } 2596 2597 /* This callback adds to the sum in `arg' the number of 2598 * threads on `ci' who have entered or who wait to enter the 2599 * critical section. 2600 */ 2601 static void 2602 ifnet_lock_sum(void *p, void *arg, struct cpu_info *ci) 2603 { 2604 uint64_t *sum = arg, *nenter = p; 2605 2606 *sum += *nenter; 2607 } 2608 2609 /* Return the number of threads who have entered or who wait 2610 * to enter the critical section on all CPUs. 2611 */ 2612 static uint64_t 2613 ifnet_lock_entrances(struct ifnet_lock *il) 2614 { 2615 uint64_t sum = 0; 2616 2617 percpu_foreach(il->il_nenter, ifnet_lock_sum, &sum); 2618 2619 return sum; 2620 } 2621 2622 static int 2623 ifioctl_attach(struct ifnet *ifp) 2624 { 2625 struct ifnet_lock *il; 2626 2627 /* If the driver has not supplied its own if_ioctl, then 2628 * supply the default. 2629 */ 2630 if (ifp->if_ioctl == NULL) 2631 ifp->if_ioctl = ifioctl_common; 2632 2633 /* Create an ifnet_lock for synchronizing ifioctls. */ 2634 if ((il = kmem_zalloc(sizeof(*il), KM_SLEEP)) == NULL) 2635 return ENOMEM; 2636 2637 il->il_nenter = percpu_alloc(sizeof(uint64_t)); 2638 if (il->il_nenter == NULL) { 2639 kmem_free(il, sizeof(*il)); 2640 return ENOMEM; 2641 } 2642 2643 mutex_init(&il->il_lock, MUTEX_DEFAULT, IPL_NONE); 2644 cv_init(&il->il_emptied, ifp->if_xname); 2645 2646 ifp->if_ioctl_lock = il; 2647 2648 return 0; 2649 } 2650 2651 /* 2652 * This must not be called until after `ifp' has been withdrawn from the 2653 * ifnet tables so that ifioctl() cannot get a handle on it by calling 2654 * ifunit(). 2655 */ 2656 static void 2657 ifioctl_detach(struct ifnet *ifp) 2658 { 2659 struct ifnet_lock *il; 2660 2661 il = ifp->if_ioctl_lock; 2662 mutex_enter(&il->il_lock); 2663 /* Install if_nullioctl to make sure that any thread that 2664 * subsequently enters the critical section will quit it 2665 * immediately and signal the condition variable that we 2666 * wait on, below. 2667 */ 2668 ifp->if_ioctl = if_nullioctl; 2669 /* Sleep while threads are still in the critical section or 2670 * wait to enter it. 2671 */ 2672 while (ifnet_lock_entrances(il) != il->il_nexit) 2673 cv_wait(&il->il_emptied, &il->il_lock); 2674 /* At this point, we are the only thread still in the critical 2675 * section, and no new thread can get a handle on the ifioctl 2676 * lock, so it is safe to free its memory. 2677 */ 2678 mutex_exit(&il->il_lock); 2679 ifp->if_ioctl_lock = NULL; 2680 percpu_free(il->il_nenter, sizeof(uint64_t)); 2681 il->il_nenter = NULL; 2682 cv_destroy(&il->il_emptied); 2683 mutex_destroy(&il->il_lock); 2684 kmem_free(il, sizeof(*il)); 2685 } 2686 2687 /* 2688 * Return interface configuration 2689 * of system. List may be used 2690 * in later ioctl's (above) to get 2691 * other information. 2692 * 2693 * Each record is a struct ifreq. Before the addition of 2694 * sockaddr_storage, the API rule was that sockaddr flavors that did 2695 * not fit would extend beyond the struct ifreq, with the next struct 2696 * ifreq starting sa_len beyond the struct sockaddr. Because the 2697 * union in struct ifreq includes struct sockaddr_storage, every kind 2698 * of sockaddr must fit. Thus, there are no longer any overlength 2699 * records. 2700 * 2701 * Records are added to the user buffer if they fit, and ifc_len is 2702 * adjusted to the length that was written. Thus, the user is only 2703 * assured of getting the complete list if ifc_len on return is at 2704 * least sizeof(struct ifreq) less than it was on entry. 2705 * 2706 * If the user buffer pointer is NULL, this routine copies no data and 2707 * returns the amount of space that would be needed. 2708 * 2709 * Invariants: 2710 * ifrp points to the next part of the user's buffer to be used. If 2711 * ifrp != NULL, space holds the number of bytes remaining that we may 2712 * write at ifrp. Otherwise, space holds the number of bytes that 2713 * would have been written had there been adequate space. 2714 */ 2715 /*ARGSUSED*/ 2716 static int 2717 ifconf(u_long cmd, void *data) 2718 { 2719 struct ifconf *ifc = (struct ifconf *)data; 2720 struct ifnet *ifp; 2721 struct ifaddr *ifa; 2722 struct ifreq ifr, *ifrp = NULL; 2723 int space = 0, error = 0; 2724 const int sz = (int)sizeof(struct ifreq); 2725 const bool docopy = ifc->ifc_req != NULL; 2726 int s; 2727 int bound = curlwp->l_pflag & LP_BOUND; 2728 struct psref psref; 2729 2730 if (docopy) { 2731 space = ifc->ifc_len; 2732 ifrp = ifc->ifc_req; 2733 } 2734 2735 curlwp->l_pflag |= LP_BOUND; 2736 s = pserialize_read_enter(); 2737 IFNET_READER_FOREACH(ifp) { 2738 psref_acquire(&psref, &ifp->if_psref, ifnet_psref_class); 2739 pserialize_read_exit(s); 2740 2741 (void)strncpy(ifr.ifr_name, ifp->if_xname, 2742 sizeof(ifr.ifr_name)); 2743 if (ifr.ifr_name[sizeof(ifr.ifr_name) - 1] != '\0') { 2744 error = ENAMETOOLONG; 2745 goto release_exit; 2746 } 2747 if (IFADDR_EMPTY(ifp)) { 2748 /* Interface with no addresses - send zero sockaddr. */ 2749 memset(&ifr.ifr_addr, 0, sizeof(ifr.ifr_addr)); 2750 if (!docopy) { 2751 space += sz; 2752 continue; 2753 } 2754 if (space >= sz) { 2755 error = copyout(&ifr, ifrp, sz); 2756 if (error != 0) 2757 goto release_exit; 2758 ifrp++; 2759 space -= sz; 2760 } 2761 } 2762 2763 IFADDR_FOREACH(ifa, ifp) { 2764 struct sockaddr *sa = ifa->ifa_addr; 2765 /* all sockaddrs must fit in sockaddr_storage */ 2766 KASSERT(sa->sa_len <= sizeof(ifr.ifr_ifru)); 2767 2768 if (!docopy) { 2769 space += sz; 2770 continue; 2771 } 2772 memcpy(&ifr.ifr_space, sa, sa->sa_len); 2773 if (space >= sz) { 2774 error = copyout(&ifr, ifrp, sz); 2775 if (error != 0) 2776 goto release_exit; 2777 ifrp++; space -= sz; 2778 } 2779 } 2780 2781 s = pserialize_read_enter(); 2782 psref_release(&psref, &ifp->if_psref, ifnet_psref_class); 2783 } 2784 pserialize_read_exit(s); 2785 curlwp->l_pflag ^= bound ^ LP_BOUND; 2786 2787 if (docopy) { 2788 KASSERT(0 <= space && space <= ifc->ifc_len); 2789 ifc->ifc_len -= space; 2790 } else { 2791 KASSERT(space >= 0); 2792 ifc->ifc_len = space; 2793 } 2794 return (0); 2795 2796 release_exit: 2797 psref_release(&psref, &ifp->if_psref, ifnet_psref_class); 2798 curlwp->l_pflag ^= bound ^ LP_BOUND; 2799 return error; 2800 } 2801 2802 int 2803 ifreq_setaddr(u_long cmd, struct ifreq *ifr, const struct sockaddr *sa) 2804 { 2805 uint8_t len; 2806 #ifdef COMPAT_OIFREQ 2807 struct ifreq ifrb; 2808 struct oifreq *oifr = NULL; 2809 u_long ocmd = cmd; 2810 cmd = compat_cvtcmd(cmd); 2811 if (cmd != ocmd) { 2812 oifr = (struct oifreq *)(void *)ifr; 2813 ifr = &ifrb; 2814 ifreqo2n(oifr, ifr); 2815 len = sizeof(oifr->ifr_addr); 2816 } else 2817 #endif 2818 len = sizeof(ifr->ifr_ifru.ifru_space); 2819 2820 if (len < sa->sa_len) 2821 return EFBIG; 2822 2823 memset(&ifr->ifr_addr, 0, len); 2824 sockaddr_copy(&ifr->ifr_addr, len, sa); 2825 2826 #ifdef COMPAT_OIFREQ 2827 if (cmd != ocmd) 2828 ifreqn2o(oifr, ifr); 2829 #endif 2830 return 0; 2831 } 2832 2833 /* 2834 * wrapper function for the drivers which doesn't have if_transmit(). 2835 */ 2836 int 2837 if_transmit(struct ifnet *ifp, struct mbuf *m) 2838 { 2839 int s, error; 2840 2841 s = splnet(); 2842 2843 IFQ_ENQUEUE(&ifp->if_snd, m, error); 2844 if (error != 0) { 2845 /* mbuf is already freed */ 2846 goto out; 2847 } 2848 2849 ifp->if_obytes += m->m_pkthdr.len;; 2850 if (m->m_flags & M_MCAST) 2851 ifp->if_omcasts++; 2852 2853 if ((ifp->if_flags & IFF_OACTIVE) == 0) 2854 (*ifp->if_start)(ifp); 2855 out: 2856 splx(s); 2857 2858 return error; 2859 } 2860 2861 /* 2862 * Queue message on interface, and start output if interface 2863 * not yet active. 2864 */ 2865 int 2866 ifq_enqueue(struct ifnet *ifp, struct mbuf *m) 2867 { 2868 2869 return (*ifp->if_transmit)(ifp, m); 2870 } 2871 2872 /* 2873 * Queue message on interface, possibly using a second fast queue 2874 */ 2875 int 2876 ifq_enqueue2(struct ifnet *ifp, struct ifqueue *ifq, struct mbuf *m) 2877 { 2878 int error = 0; 2879 2880 if (ifq != NULL 2881 #ifdef ALTQ 2882 && ALTQ_IS_ENABLED(&ifp->if_snd) == 0 2883 #endif 2884 ) { 2885 if (IF_QFULL(ifq)) { 2886 IF_DROP(&ifp->if_snd); 2887 m_freem(m); 2888 if (error == 0) 2889 error = ENOBUFS; 2890 } else 2891 IF_ENQUEUE(ifq, m); 2892 } else 2893 IFQ_ENQUEUE(&ifp->if_snd, m, error); 2894 if (error != 0) { 2895 ++ifp->if_oerrors; 2896 return error; 2897 } 2898 return 0; 2899 } 2900 2901 int 2902 if_addr_init(ifnet_t *ifp, struct ifaddr *ifa, const bool src) 2903 { 2904 int rc; 2905 2906 if (ifp->if_initaddr != NULL) 2907 rc = (*ifp->if_initaddr)(ifp, ifa, src); 2908 else if (src || 2909 (rc = (*ifp->if_ioctl)(ifp, SIOCSIFDSTADDR, ifa)) == ENOTTY) 2910 rc = (*ifp->if_ioctl)(ifp, SIOCINITIFADDR, ifa); 2911 2912 return rc; 2913 } 2914 2915 int 2916 if_do_dad(struct ifnet *ifp) 2917 { 2918 if ((ifp->if_flags & IFF_LOOPBACK) != 0) 2919 return 0; 2920 2921 switch (ifp->if_type) { 2922 case IFT_FAITH: 2923 /* 2924 * These interfaces do not have the IFF_LOOPBACK flag, 2925 * but loop packets back. We do not have to do DAD on such 2926 * interfaces. We should even omit it, because loop-backed 2927 * responses would confuse the DAD procedure. 2928 */ 2929 return 0; 2930 default: 2931 /* 2932 * Our DAD routine requires the interface up and running. 2933 * However, some interfaces can be up before the RUNNING 2934 * status. Additionaly, users may try to assign addresses 2935 * before the interface becomes up (or running). 2936 * We simply skip DAD in such a case as a work around. 2937 * XXX: we should rather mark "tentative" on such addresses, 2938 * and do DAD after the interface becomes ready. 2939 */ 2940 if ((ifp->if_flags & (IFF_UP|IFF_RUNNING)) != 2941 (IFF_UP|IFF_RUNNING)) 2942 return 0; 2943 2944 return 1; 2945 } 2946 } 2947 2948 int 2949 if_flags_set(ifnet_t *ifp, const short flags) 2950 { 2951 int rc; 2952 2953 if (ifp->if_setflags != NULL) 2954 rc = (*ifp->if_setflags)(ifp, flags); 2955 else { 2956 short cantflags, chgdflags; 2957 struct ifreq ifr; 2958 2959 chgdflags = ifp->if_flags ^ flags; 2960 cantflags = chgdflags & IFF_CANTCHANGE; 2961 2962 if (cantflags != 0) 2963 ifp->if_flags ^= cantflags; 2964 2965 /* Traditionally, we do not call if_ioctl after 2966 * setting/clearing only IFF_PROMISC if the interface 2967 * isn't IFF_UP. Uphold that tradition. 2968 */ 2969 if (chgdflags == IFF_PROMISC && (ifp->if_flags & IFF_UP) == 0) 2970 return 0; 2971 2972 memset(&ifr, 0, sizeof(ifr)); 2973 2974 ifr.ifr_flags = flags & ~IFF_CANTCHANGE; 2975 rc = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, &ifr); 2976 2977 if (rc != 0 && cantflags != 0) 2978 ifp->if_flags ^= cantflags; 2979 } 2980 2981 return rc; 2982 } 2983 2984 int 2985 if_mcast_op(ifnet_t *ifp, const unsigned long cmd, const struct sockaddr *sa) 2986 { 2987 int rc; 2988 struct ifreq ifr; 2989 2990 if (ifp->if_mcastop != NULL) 2991 rc = (*ifp->if_mcastop)(ifp, cmd, sa); 2992 else { 2993 ifreq_setaddr(cmd, &ifr, sa); 2994 rc = (*ifp->if_ioctl)(ifp, cmd, &ifr); 2995 } 2996 2997 return rc; 2998 } 2999 3000 static void 3001 sysctl_sndq_setup(struct sysctllog **clog, const char *ifname, 3002 struct ifaltq *ifq) 3003 { 3004 const struct sysctlnode *cnode, *rnode; 3005 3006 if (sysctl_createv(clog, 0, NULL, &rnode, 3007 CTLFLAG_PERMANENT, 3008 CTLTYPE_NODE, "interfaces", 3009 SYSCTL_DESCR("Per-interface controls"), 3010 NULL, 0, NULL, 0, 3011 CTL_NET, CTL_CREATE, CTL_EOL) != 0) 3012 goto bad; 3013 3014 if (sysctl_createv(clog, 0, &rnode, &rnode, 3015 CTLFLAG_PERMANENT, 3016 CTLTYPE_NODE, ifname, 3017 SYSCTL_DESCR("Interface controls"), 3018 NULL, 0, NULL, 0, 3019 CTL_CREATE, CTL_EOL) != 0) 3020 goto bad; 3021 3022 if (sysctl_createv(clog, 0, &rnode, &rnode, 3023 CTLFLAG_PERMANENT, 3024 CTLTYPE_NODE, "sndq", 3025 SYSCTL_DESCR("Interface output queue controls"), 3026 NULL, 0, NULL, 0, 3027 CTL_CREATE, CTL_EOL) != 0) 3028 goto bad; 3029 3030 if (sysctl_createv(clog, 0, &rnode, &cnode, 3031 CTLFLAG_PERMANENT, 3032 CTLTYPE_INT, "len", 3033 SYSCTL_DESCR("Current output queue length"), 3034 NULL, 0, &ifq->ifq_len, 0, 3035 CTL_CREATE, CTL_EOL) != 0) 3036 goto bad; 3037 3038 if (sysctl_createv(clog, 0, &rnode, &cnode, 3039 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 3040 CTLTYPE_INT, "maxlen", 3041 SYSCTL_DESCR("Maximum allowed output queue length"), 3042 NULL, 0, &ifq->ifq_maxlen, 0, 3043 CTL_CREATE, CTL_EOL) != 0) 3044 goto bad; 3045 3046 if (sysctl_createv(clog, 0, &rnode, &cnode, 3047 CTLFLAG_PERMANENT, 3048 CTLTYPE_INT, "drops", 3049 SYSCTL_DESCR("Packets dropped due to full output queue"), 3050 NULL, 0, &ifq->ifq_drops, 0, 3051 CTL_CREATE, CTL_EOL) != 0) 3052 goto bad; 3053 3054 return; 3055 bad: 3056 printf("%s: could not attach sysctl nodes\n", ifname); 3057 return; 3058 } 3059 3060 #if defined(INET) || defined(INET6) 3061 3062 #define SYSCTL_NET_PKTQ(q, cn, c) \ 3063 static int \ 3064 sysctl_net_##q##_##cn(SYSCTLFN_ARGS) \ 3065 { \ 3066 return sysctl_pktq_count(SYSCTLFN_CALL(rnode), q, c); \ 3067 } 3068 3069 #if defined(INET) 3070 static int 3071 sysctl_net_ip_pktq_maxlen(SYSCTLFN_ARGS) 3072 { 3073 return sysctl_pktq_maxlen(SYSCTLFN_CALL(rnode), ip_pktq); 3074 } 3075 SYSCTL_NET_PKTQ(ip_pktq, items, PKTQ_NITEMS) 3076 SYSCTL_NET_PKTQ(ip_pktq, drops, PKTQ_DROPS) 3077 #endif 3078 3079 #if defined(INET6) 3080 static int 3081 sysctl_net_ip6_pktq_maxlen(SYSCTLFN_ARGS) 3082 { 3083 return sysctl_pktq_maxlen(SYSCTLFN_CALL(rnode), ip6_pktq); 3084 } 3085 SYSCTL_NET_PKTQ(ip6_pktq, items, PKTQ_NITEMS) 3086 SYSCTL_NET_PKTQ(ip6_pktq, drops, PKTQ_DROPS) 3087 #endif 3088 3089 static void 3090 sysctl_net_pktq_setup(struct sysctllog **clog, int pf) 3091 { 3092 sysctlfn len_func = NULL, maxlen_func = NULL, drops_func = NULL; 3093 const char *pfname = NULL, *ipname = NULL; 3094 int ipn = 0, qid = 0; 3095 3096 switch (pf) { 3097 #if defined(INET) 3098 case PF_INET: 3099 len_func = sysctl_net_ip_pktq_items; 3100 maxlen_func = sysctl_net_ip_pktq_maxlen; 3101 drops_func = sysctl_net_ip_pktq_drops; 3102 pfname = "inet", ipn = IPPROTO_IP; 3103 ipname = "ip", qid = IPCTL_IFQ; 3104 break; 3105 #endif 3106 #if defined(INET6) 3107 case PF_INET6: 3108 len_func = sysctl_net_ip6_pktq_items; 3109 maxlen_func = sysctl_net_ip6_pktq_maxlen; 3110 drops_func = sysctl_net_ip6_pktq_drops; 3111 pfname = "inet6", ipn = IPPROTO_IPV6; 3112 ipname = "ip6", qid = IPV6CTL_IFQ; 3113 break; 3114 #endif 3115 default: 3116 KASSERT(false); 3117 } 3118 3119 sysctl_createv(clog, 0, NULL, NULL, 3120 CTLFLAG_PERMANENT, 3121 CTLTYPE_NODE, pfname, NULL, 3122 NULL, 0, NULL, 0, 3123 CTL_NET, pf, CTL_EOL); 3124 sysctl_createv(clog, 0, NULL, NULL, 3125 CTLFLAG_PERMANENT, 3126 CTLTYPE_NODE, ipname, NULL, 3127 NULL, 0, NULL, 0, 3128 CTL_NET, pf, ipn, CTL_EOL); 3129 sysctl_createv(clog, 0, NULL, NULL, 3130 CTLFLAG_PERMANENT, 3131 CTLTYPE_NODE, "ifq", 3132 SYSCTL_DESCR("Protocol input queue controls"), 3133 NULL, 0, NULL, 0, 3134 CTL_NET, pf, ipn, qid, CTL_EOL); 3135 3136 sysctl_createv(clog, 0, NULL, NULL, 3137 CTLFLAG_PERMANENT, 3138 CTLTYPE_INT, "len", 3139 SYSCTL_DESCR("Current input queue length"), 3140 len_func, 0, NULL, 0, 3141 CTL_NET, pf, ipn, qid, IFQCTL_LEN, CTL_EOL); 3142 sysctl_createv(clog, 0, NULL, NULL, 3143 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 3144 CTLTYPE_INT, "maxlen", 3145 SYSCTL_DESCR("Maximum allowed input queue length"), 3146 maxlen_func, 0, NULL, 0, 3147 CTL_NET, pf, ipn, qid, IFQCTL_MAXLEN, CTL_EOL); 3148 sysctl_createv(clog, 0, NULL, NULL, 3149 CTLFLAG_PERMANENT, 3150 CTLTYPE_INT, "drops", 3151 SYSCTL_DESCR("Packets dropped due to full input queue"), 3152 drops_func, 0, NULL, 0, 3153 CTL_NET, pf, ipn, qid, IFQCTL_DROPS, CTL_EOL); 3154 } 3155 #endif /* INET || INET6 */ 3156 3157 static int 3158 if_sdl_sysctl(SYSCTLFN_ARGS) 3159 { 3160 struct ifnet *ifp; 3161 const struct sockaddr_dl *sdl; 3162 3163 if (namelen != 1) 3164 return EINVAL; 3165 3166 ifp = if_byindex(name[0]); 3167 if (ifp == NULL) 3168 return ENODEV; 3169 3170 sdl = ifp->if_sadl; 3171 if (sdl == NULL) { 3172 *oldlenp = 0; 3173 return 0; 3174 } 3175 3176 if (oldp == NULL) { 3177 *oldlenp = sdl->sdl_alen; 3178 return 0; 3179 } 3180 3181 if (*oldlenp >= sdl->sdl_alen) 3182 *oldlenp = sdl->sdl_alen; 3183 return sysctl_copyout(l, &sdl->sdl_data[sdl->sdl_nlen], oldp, *oldlenp); 3184 } 3185 3186 SYSCTL_SETUP(sysctl_net_sdl_setup, "sysctl net.sdl subtree setup") 3187 { 3188 const struct sysctlnode *rnode = NULL; 3189 3190 sysctl_createv(clog, 0, NULL, &rnode, 3191 CTLFLAG_PERMANENT, 3192 CTLTYPE_NODE, "sdl", 3193 SYSCTL_DESCR("Get active link-layer address"), 3194 if_sdl_sysctl, 0, NULL, 0, 3195 CTL_NET, CTL_CREATE, CTL_EOL); 3196 } 3197
Indexes created Mon Sep 22 05:09:51 GMT 2025