1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1980, 1986, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 * 31 * @(#)if.c 8.5 (Berkeley) 1/9/95 32 * $FreeBSD$ 33 */ 34 35 #include "opt_bpf.h" 36 #include "opt_inet6.h" 37 #include "opt_inet.h" 38 39 #include <sys/param.h> 40 #include <sys/conf.h> 41 #include <sys/eventhandler.h> 42 #include <sys/malloc.h> 43 #include <sys/domainset.h> 44 #include <sys/sbuf.h> 45 #include <sys/bus.h> 46 #include <sys/epoch.h> 47 #include <sys/mbuf.h> 48 #include <sys/systm.h> 49 #include <sys/priv.h> 50 #include <sys/proc.h> 51 #include <sys/socket.h> 52 #include <sys/socketvar.h> 53 #include <sys/protosw.h> 54 #include <sys/kernel.h> 55 #include <sys/lock.h> 56 #include <sys/refcount.h> 57 #include <sys/module.h> 58 #include <sys/rwlock.h> 59 #include <sys/sockio.h> 60 #include <sys/syslog.h> 61 #include <sys/sysctl.h> 62 #include <sys/sysent.h> 63 #include <sys/taskqueue.h> 64 #include <sys/domain.h> 65 #include <sys/jail.h> 66 #include <sys/priv.h> 67 68 #include <machine/stdarg.h> 69 #include <vm/uma.h> 70 71 #include <net/bpf.h> 72 #include <net/ethernet.h> 73 #include <net/if.h> 74 #include <net/if_arp.h> 75 #include <net/if_clone.h> 76 #include <net/if_dl.h> 77 #include <net/if_types.h> 78 #include <net/if_var.h> 79 #include <net/if_media.h> 80 #include <net/if_vlan_var.h> 81 #include <net/radix.h> 82 #include <net/route.h> 83 #include <net/vnet.h> 84 85 #if defined(INET) || defined(INET6) 86 #include <net/ethernet.h> 87 #include <netinet/in.h> 88 #include <netinet/in_var.h> 89 #include <netinet/ip.h> 90 #include <netinet/ip_carp.h> 91 #ifdef INET 92 #include <net/debugnet.h> 93 #include <netinet/if_ether.h> 94 #endif /* INET */ 95 #ifdef INET6 96 #include <netinet6/in6_var.h> 97 #include <netinet6/in6_ifattach.h> 98 #endif /* INET6 */ 99 #endif /* INET || INET6 */ 100 101 #include <security/mac/mac_framework.h> 102 103 /* 104 * Consumers of struct ifreq such as tcpdump assume no pad between ifr_name 105 * and ifr_ifru when it is used in SIOCGIFCONF. 106 */ 107 _Static_assert(sizeof(((struct ifreq *)0)->ifr_name) == 108 offsetof(struct ifreq, ifr_ifru), "gap between ifr_name and ifr_ifru"); 109 110 __read_mostly epoch_t net_epoch_preempt; 111 __read_mostly epoch_t net_epoch; 112 #ifdef COMPAT_FREEBSD32 113 #include <sys/mount.h> 114 #include <compat/freebsd32/freebsd32.h> 115 116 struct ifreq_buffer32 { 117 uint32_t length; /* (size_t) */ 118 uint32_t buffer; /* (void *) */ 119 }; 120 121 /* 122 * Interface request structure used for socket 123 * ioctl's. All interface ioctl's must have parameter 124 * definitions which begin with ifr_name. The 125 * remainder may be interface specific. 126 */ 127 struct ifreq32 { 128 char ifr_name[IFNAMSIZ]; /* if name, e.g. "en0" */ 129 union { 130 struct sockaddr ifru_addr; 131 struct sockaddr ifru_dstaddr; 132 struct sockaddr ifru_broadaddr; 133 struct ifreq_buffer32 ifru_buffer; 134 short ifru_flags[2]; 135 short ifru_index; 136 int ifru_jid; 137 int ifru_metric; 138 int ifru_mtu; 139 int ifru_phys; 140 int ifru_media; 141 uint32_t ifru_data; 142 int ifru_cap[2]; 143 u_int ifru_fib; 144 u_char ifru_vlan_pcp; 145 } ifr_ifru; 146 }; 147 CTASSERT(sizeof(struct ifreq) == sizeof(struct ifreq32)); 148 CTASSERT(__offsetof(struct ifreq, ifr_ifru) == 149 __offsetof(struct ifreq32, ifr_ifru)); 150 151 struct ifgroupreq32 { 152 char ifgr_name[IFNAMSIZ]; 153 u_int ifgr_len; 154 union { 155 char ifgru_group[IFNAMSIZ]; 156 uint32_t ifgru_groups; 157 } ifgr_ifgru; 158 }; 159 160 struct ifmediareq32 { 161 char ifm_name[IFNAMSIZ]; 162 int ifm_current; 163 int ifm_mask; 164 int ifm_status; 165 int ifm_active; 166 int ifm_count; 167 uint32_t ifm_ulist; /* (int *) */ 168 }; 169 #define SIOCGIFMEDIA32 _IOC_NEWTYPE(SIOCGIFMEDIA, struct ifmediareq32) 170 #define SIOCGIFXMEDIA32 _IOC_NEWTYPE(SIOCGIFXMEDIA, struct ifmediareq32) 171 172 #define _CASE_IOC_IFGROUPREQ_32(cmd) \ 173 _IOC_NEWTYPE((cmd), struct ifgroupreq32): case 174 #else /* !COMPAT_FREEBSD32 */ 175 #define _CASE_IOC_IFGROUPREQ_32(cmd) 176 #endif /* !COMPAT_FREEBSD32 */ 177 178 #define CASE_IOC_IFGROUPREQ(cmd) \ 179 _CASE_IOC_IFGROUPREQ_32(cmd) \ 180 (cmd) 181 182 union ifreq_union { 183 struct ifreq ifr; 184 #ifdef COMPAT_FREEBSD32 185 struct ifreq32 ifr32; 186 #endif 187 }; 188 189 union ifgroupreq_union { 190 struct ifgroupreq ifgr; 191 #ifdef COMPAT_FREEBSD32 192 struct ifgroupreq32 ifgr32; 193 #endif 194 }; 195 196 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers"); 197 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management"); 198 199 SYSCTL_INT(_net_link, OID_AUTO, ifqmaxlen, CTLFLAG_RDTUN, 200 &ifqmaxlen, 0, "max send queue size"); 201 202 /* Log link state change events */ 203 static int log_link_state_change = 1; 204 205 SYSCTL_INT(_net_link, OID_AUTO, log_link_state_change, CTLFLAG_RW, 206 &log_link_state_change, 0, 207 "log interface link state change events"); 208 209 /* Log promiscuous mode change events */ 210 static int log_promisc_mode_change = 1; 211 212 SYSCTL_INT(_net_link, OID_AUTO, log_promisc_mode_change, CTLFLAG_RDTUN, 213 &log_promisc_mode_change, 1, 214 "log promiscuous mode change events"); 215 216 /* Interface description */ 217 static unsigned int ifdescr_maxlen = 1024; 218 SYSCTL_UINT(_net, OID_AUTO, ifdescr_maxlen, CTLFLAG_RW, 219 &ifdescr_maxlen, 0, 220 "administrative maximum length for interface description"); 221 222 static MALLOC_DEFINE(M_IFDESCR, "ifdescr", "ifnet descriptions"); 223 224 /* global sx for non-critical path ifdescr */ 225 static struct sx ifdescr_sx; 226 SX_SYSINIT(ifdescr_sx, &ifdescr_sx, "ifnet descr"); 227 228 void (*ng_ether_link_state_p)(struct ifnet *ifp, int state); 229 void (*lagg_linkstate_p)(struct ifnet *ifp, int state); 230 /* These are external hooks for CARP. */ 231 void (*carp_linkstate_p)(struct ifnet *ifp); 232 void (*carp_demote_adj_p)(int, char *); 233 int (*carp_master_p)(struct ifaddr *); 234 #if defined(INET) || defined(INET6) 235 int (*carp_forus_p)(struct ifnet *ifp, u_char *dhost); 236 int (*carp_output_p)(struct ifnet *ifp, struct mbuf *m, 237 const struct sockaddr *sa); 238 int (*carp_ioctl_p)(struct ifreq *, u_long, struct thread *); 239 int (*carp_attach_p)(struct ifaddr *, int); 240 void (*carp_detach_p)(struct ifaddr *, bool); 241 #endif 242 #ifdef INET 243 int (*carp_iamatch_p)(struct ifaddr *, uint8_t **); 244 #endif 245 #ifdef INET6 246 struct ifaddr *(*carp_iamatch6_p)(struct ifnet *ifp, struct in6_addr *taddr6); 247 caddr_t (*carp_macmatch6_p)(struct ifnet *ifp, struct mbuf *m, 248 const struct in6_addr *taddr); 249 #endif 250 251 struct mbuf *(*tbr_dequeue_ptr)(struct ifaltq *, int) = NULL; 252 253 /* 254 * XXX: Style; these should be sorted alphabetically, and unprototyped 255 * static functions should be prototyped. Currently they are sorted by 256 * declaration order. 257 */ 258 static void if_attachdomain(void *); 259 static void if_attachdomain1(struct ifnet *); 260 static int ifconf(u_long, caddr_t); 261 static void *if_grow(void); 262 static void if_input_default(struct ifnet *, struct mbuf *); 263 static int if_requestencap_default(struct ifnet *, struct if_encap_req *); 264 static void if_route(struct ifnet *, int flag, int fam); 265 static int if_setflag(struct ifnet *, int, int, int *, int); 266 static int if_transmit(struct ifnet *ifp, struct mbuf *m); 267 static void if_unroute(struct ifnet *, int flag, int fam); 268 static int if_delmulti_locked(struct ifnet *, struct ifmultiaddr *, int); 269 static void do_link_state_change(void *, int); 270 static int if_getgroup(struct ifgroupreq *, struct ifnet *); 271 static int if_getgroupmembers(struct ifgroupreq *); 272 static void if_delgroups(struct ifnet *); 273 static void if_attach_internal(struct ifnet *, int, struct if_clone *); 274 static int if_detach_internal(struct ifnet *, int, struct if_clone **); 275 static void if_siocaddmulti(void *, int); 276 #ifdef VIMAGE 277 static void if_vmove(struct ifnet *, struct vnet *); 278 #endif 279 280 #ifdef INET6 281 /* 282 * XXX: declare here to avoid to include many inet6 related files.. 283 * should be more generalized? 284 */ 285 extern void nd6_setmtu(struct ifnet *); 286 #endif 287 288 /* ipsec helper hooks */ 289 VNET_DEFINE(struct hhook_head *, ipsec_hhh_in[HHOOK_IPSEC_COUNT]); 290 VNET_DEFINE(struct hhook_head *, ipsec_hhh_out[HHOOK_IPSEC_COUNT]); 291 292 VNET_DEFINE(int, if_index); 293 int ifqmaxlen = IFQ_MAXLEN; 294 VNET_DEFINE(struct ifnethead, ifnet); /* depend on static init XXX */ 295 VNET_DEFINE(struct ifgrouphead, ifg_head); 296 297 VNET_DEFINE_STATIC(int, if_indexlim) = 8; 298 299 /* Table of ifnet by index. */ 300 VNET_DEFINE(struct ifnet **, ifindex_table); 301 302 #define V_if_indexlim VNET(if_indexlim) 303 #define V_ifindex_table VNET(ifindex_table) 304 305 /* 306 * The global network interface list (V_ifnet) and related state (such as 307 * if_index, if_indexlim, and ifindex_table) are protected by an sxlock and 308 * an rwlock. Either may be acquired shared to stablize the list, but both 309 * must be acquired writable to modify the list. This model allows us to 310 * both stablize the interface list during interrupt thread processing, but 311 * also to stablize it over long-running ioctls, without introducing priority 312 * inversions and deadlocks. 313 */ 314 struct rwlock ifnet_rwlock; 315 RW_SYSINIT_FLAGS(ifnet_rw, &ifnet_rwlock, "ifnet_rw", RW_RECURSE); 316 struct sx ifnet_sxlock; 317 SX_SYSINIT_FLAGS(ifnet_sx, &ifnet_sxlock, "ifnet_sx", SX_RECURSE); 318 319 /* 320 * The allocation of network interfaces is a rather non-atomic affair; we 321 * need to select an index before we are ready to expose the interface for 322 * use, so will use this pointer value to indicate reservation. 323 */ 324 #define IFNET_HOLD (void *)(uintptr_t)(-1) 325 326 static if_com_alloc_t *if_com_alloc[256]; 327 static if_com_free_t *if_com_free[256]; 328 329 static MALLOC_DEFINE(M_IFNET, "ifnet", "interface internals"); 330 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address"); 331 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address"); 332 333 struct ifnet * 334 ifnet_byindex(u_short idx) 335 { 336 struct ifnet *ifp; 337 338 if (__predict_false(idx > V_if_index)) 339 return (NULL); 340 341 ifp = *(struct ifnet * const volatile *)(V_ifindex_table + idx); 342 return (__predict_false(ifp == IFNET_HOLD) ? NULL : ifp); 343 } 344 345 struct ifnet * 346 ifnet_byindex_ref(u_short idx) 347 { 348 struct ifnet *ifp; 349 350 NET_EPOCH_ASSERT(); 351 352 ifp = ifnet_byindex(idx); 353 if (ifp == NULL || (ifp->if_flags & IFF_DYING)) 354 return (NULL); 355 if_ref(ifp); 356 return (ifp); 357 } 358 359 /* 360 * Allocate an ifindex array entry; return 0 on success or an error on 361 * failure. 362 */ 363 static u_short 364 ifindex_alloc(void **old) 365 { 366 u_short idx; 367 368 IFNET_WLOCK_ASSERT(); 369 /* 370 * Try to find an empty slot below V_if_index. If we fail, take the 371 * next slot. 372 */ 373 for (idx = 1; idx <= V_if_index; idx++) { 374 if (V_ifindex_table[idx] == NULL) 375 break; 376 } 377 378 /* Catch if_index overflow. */ 379 if (idx >= V_if_indexlim) { 380 *old = if_grow(); 381 return (USHRT_MAX); 382 } 383 if (idx > V_if_index) 384 V_if_index = idx; 385 return (idx); 386 } 387 388 static void 389 ifindex_free_locked(u_short idx) 390 { 391 392 IFNET_WLOCK_ASSERT(); 393 394 V_ifindex_table[idx] = NULL; 395 while (V_if_index > 0 && 396 V_ifindex_table[V_if_index] == NULL) 397 V_if_index--; 398 } 399 400 static void 401 ifindex_free(u_short idx) 402 { 403 404 IFNET_WLOCK(); 405 ifindex_free_locked(idx); 406 IFNET_WUNLOCK(); 407 } 408 409 static void 410 ifnet_setbyindex(u_short idx, struct ifnet *ifp) 411 { 412 413 V_ifindex_table[idx] = ifp; 414 } 415 416 struct ifaddr * 417 ifaddr_byindex(u_short idx) 418 { 419 struct ifnet *ifp; 420 struct ifaddr *ifa = NULL; 421 422 NET_EPOCH_ASSERT(); 423 424 ifp = ifnet_byindex(idx); 425 if (ifp != NULL && (ifa = ifp->if_addr) != NULL) 426 ifa_ref(ifa); 427 return (ifa); 428 } 429 430 /* 431 * Network interface utility routines. 432 * 433 * Routines with ifa_ifwith* names take sockaddr *'s as 434 * parameters. 435 */ 436 437 static void 438 vnet_if_init(const void *unused __unused) 439 { 440 void *old; 441 442 CK_STAILQ_INIT(&V_ifnet); 443 CK_STAILQ_INIT(&V_ifg_head); 444 IFNET_WLOCK(); 445 old = if_grow(); /* create initial table */ 446 IFNET_WUNLOCK(); 447 epoch_wait_preempt(net_epoch_preempt); 448 free(old, M_IFNET); 449 vnet_if_clone_init(); 450 } 451 VNET_SYSINIT(vnet_if_init, SI_SUB_INIT_IF, SI_ORDER_SECOND, vnet_if_init, 452 NULL); 453 454 #ifdef VIMAGE 455 static void 456 vnet_if_uninit(const void *unused __unused) 457 { 458 459 VNET_ASSERT(CK_STAILQ_EMPTY(&V_ifnet), ("%s:%d tailq &V_ifnet=%p " 460 "not empty", __func__, __LINE__, &V_ifnet)); 461 VNET_ASSERT(CK_STAILQ_EMPTY(&V_ifg_head), ("%s:%d tailq &V_ifg_head=%p " 462 "not empty", __func__, __LINE__, &V_ifg_head)); 463 464 free((caddr_t)V_ifindex_table, M_IFNET); 465 } 466 VNET_SYSUNINIT(vnet_if_uninit, SI_SUB_INIT_IF, SI_ORDER_FIRST, 467 vnet_if_uninit, NULL); 468 469 static void 470 vnet_if_return(const void *unused __unused) 471 { 472 struct ifnet *ifp, *nifp; 473 474 /* Return all inherited interfaces to their parent vnets. */ 475 CK_STAILQ_FOREACH_SAFE(ifp, &V_ifnet, if_link, nifp) { 476 if (ifp->if_home_vnet != ifp->if_vnet) 477 if_vmove(ifp, ifp->if_home_vnet); 478 } 479 } 480 VNET_SYSUNINIT(vnet_if_return, SI_SUB_VNET_DONE, SI_ORDER_ANY, 481 vnet_if_return, NULL); 482 #endif 483 484 485 static void * 486 if_grow(void) 487 { 488 int oldlim; 489 u_int n; 490 struct ifnet **e; 491 void *old; 492 493 old = NULL; 494 IFNET_WLOCK_ASSERT(); 495 oldlim = V_if_indexlim; 496 IFNET_WUNLOCK(); 497 n = (oldlim << 1) * sizeof(*e); 498 e = malloc(n, M_IFNET, M_WAITOK | M_ZERO); 499 IFNET_WLOCK(); 500 if (V_if_indexlim != oldlim) { 501 free(e, M_IFNET); 502 return (NULL); 503 } 504 if (V_ifindex_table != NULL) { 505 memcpy((caddr_t)e, (caddr_t)V_ifindex_table, n/2); 506 old = V_ifindex_table; 507 } 508 V_if_indexlim <<= 1; 509 V_ifindex_table = e; 510 return (old); 511 } 512 513 /* 514 * Allocate a struct ifnet and an index for an interface. A layer 2 515 * common structure will also be allocated if an allocation routine is 516 * registered for the passed type. 517 */ 518 struct ifnet * 519 if_alloc_domain(u_char type, int numa_domain) 520 { 521 struct ifnet *ifp; 522 u_short idx; 523 void *old; 524 525 KASSERT(numa_domain <= IF_NODOM, ("numa_domain too large")); 526 if (numa_domain == IF_NODOM) 527 ifp = malloc(sizeof(struct ifnet), M_IFNET, 528 M_WAITOK | M_ZERO); 529 else 530 ifp = malloc_domainset(sizeof(struct ifnet), M_IFNET, 531 DOMAINSET_PREF(numa_domain), M_WAITOK | M_ZERO); 532 restart: 533 IFNET_WLOCK(); 534 idx = ifindex_alloc(&old); 535 if (__predict_false(idx == USHRT_MAX)) { 536 IFNET_WUNLOCK(); 537 epoch_wait_preempt(net_epoch_preempt); 538 free(old, M_IFNET); 539 goto restart; 540 } 541 ifnet_setbyindex(idx, IFNET_HOLD); 542 IFNET_WUNLOCK(); 543 ifp->if_index = idx; 544 ifp->if_type = type; 545 ifp->if_alloctype = type; 546 ifp->if_numa_domain = numa_domain; 547 #ifdef VIMAGE 548 ifp->if_vnet = curvnet; 549 #endif 550 if (if_com_alloc[type] != NULL) { 551 ifp->if_l2com = if_com_alloc[type](type, ifp); 552 if (ifp->if_l2com == NULL) { 553 free(ifp, M_IFNET); 554 ifindex_free(idx); 555 return (NULL); 556 } 557 } 558 559 IF_ADDR_LOCK_INIT(ifp); 560 TASK_INIT(&ifp->if_linktask, 0, do_link_state_change, ifp); 561 TASK_INIT(&ifp->if_addmultitask, 0, if_siocaddmulti, ifp); 562 ifp->if_afdata_initialized = 0; 563 IF_AFDATA_LOCK_INIT(ifp); 564 CK_STAILQ_INIT(&ifp->if_addrhead); 565 CK_STAILQ_INIT(&ifp->if_multiaddrs); 566 CK_STAILQ_INIT(&ifp->if_groups); 567 #ifdef MAC 568 mac_ifnet_init(ifp); 569 #endif 570 ifq_init(&ifp->if_snd, ifp); 571 572 refcount_init(&ifp->if_refcount, 1); /* Index reference. */ 573 for (int i = 0; i < IFCOUNTERS; i++) 574 ifp->if_counters[i] = counter_u64_alloc(M_WAITOK); 575 ifp->if_get_counter = if_get_counter_default; 576 ifp->if_pcp = IFNET_PCP_NONE; 577 ifnet_setbyindex(ifp->if_index, ifp); 578 return (ifp); 579 } 580 581 struct ifnet * 582 if_alloc_dev(u_char type, device_t dev) 583 { 584 int numa_domain; 585 586 if (dev == NULL || bus_get_domain(dev, &numa_domain) != 0) 587 return (if_alloc_domain(type, IF_NODOM)); 588 return (if_alloc_domain(type, numa_domain)); 589 } 590 591 struct ifnet * 592 if_alloc(u_char type) 593 { 594 595 return (if_alloc_domain(type, IF_NODOM)); 596 } 597 /* 598 * Do the actual work of freeing a struct ifnet, and layer 2 common 599 * structure. This call is made when the last reference to an 600 * interface is released. 601 */ 602 static void 603 if_free_internal(struct ifnet *ifp) 604 { 605 606 KASSERT((ifp->if_flags & IFF_DYING), 607 ("if_free_internal: interface not dying")); 608 609 if (if_com_free[ifp->if_alloctype] != NULL) 610 if_com_free[ifp->if_alloctype](ifp->if_l2com, 611 ifp->if_alloctype); 612 613 #ifdef MAC 614 mac_ifnet_destroy(ifp); 615 #endif /* MAC */ 616 IF_AFDATA_DESTROY(ifp); 617 IF_ADDR_LOCK_DESTROY(ifp); 618 ifq_delete(&ifp->if_snd); 619 620 for (int i = 0; i < IFCOUNTERS; i++) 621 counter_u64_free(ifp->if_counters[i]); 622 623 free(ifp->if_description, M_IFDESCR); 624 free(ifp->if_hw_addr, M_IFADDR); 625 if (ifp->if_numa_domain == IF_NODOM) 626 free(ifp, M_IFNET); 627 else 628 free_domain(ifp, M_IFNET); 629 } 630 631 static void 632 if_destroy(epoch_context_t ctx) 633 { 634 struct ifnet *ifp; 635 636 ifp = __containerof(ctx, struct ifnet, if_epoch_ctx); 637 if_free_internal(ifp); 638 } 639 640 /* 641 * Deregister an interface and free the associated storage. 642 */ 643 void 644 if_free(struct ifnet *ifp) 645 { 646 647 ifp->if_flags |= IFF_DYING; /* XXX: Locking */ 648 649 CURVNET_SET_QUIET(ifp->if_vnet); 650 IFNET_WLOCK(); 651 KASSERT(ifp == ifnet_byindex(ifp->if_index), 652 ("%s: freeing unallocated ifnet", ifp->if_xname)); 653 654 ifindex_free_locked(ifp->if_index); 655 IFNET_WUNLOCK(); 656 657 if (refcount_release(&ifp->if_refcount)) 658 epoch_call(net_epoch_preempt, &ifp->if_epoch_ctx, if_destroy); 659 CURVNET_RESTORE(); 660 } 661 662 /* 663 * Interfaces to keep an ifnet type-stable despite the possibility of the 664 * driver calling if_free(). If there are additional references, we defer 665 * freeing the underlying data structure. 666 */ 667 void 668 if_ref(struct ifnet *ifp) 669 { 670 671 /* We don't assert the ifnet list lock here, but arguably should. */ 672 refcount_acquire(&ifp->if_refcount); 673 } 674 675 void 676 if_rele(struct ifnet *ifp) 677 { 678 679 if (!refcount_release(&ifp->if_refcount)) 680 return; 681 epoch_call(net_epoch_preempt, &ifp->if_epoch_ctx, if_destroy); 682 } 683 684 void 685 ifq_init(struct ifaltq *ifq, struct ifnet *ifp) 686 { 687 688 mtx_init(&ifq->ifq_mtx, ifp->if_xname, "if send queue", MTX_DEF); 689 690 if (ifq->ifq_maxlen == 0) 691 ifq->ifq_maxlen = ifqmaxlen; 692 693 ifq->altq_type = 0; 694 ifq->altq_disc = NULL; 695 ifq->altq_flags &= ALTQF_CANTCHANGE; 696 ifq->altq_tbr = NULL; 697 ifq->altq_ifp = ifp; 698 } 699 700 void 701 ifq_delete(struct ifaltq *ifq) 702 { 703 mtx_destroy(&ifq->ifq_mtx); 704 } 705 706 /* 707 * Perform generic interface initialization tasks and attach the interface 708 * to the list of "active" interfaces. If vmove flag is set on entry 709 * to if_attach_internal(), perform only a limited subset of initialization 710 * tasks, given that we are moving from one vnet to another an ifnet which 711 * has already been fully initialized. 712 * 713 * Note that if_detach_internal() removes group membership unconditionally 714 * even when vmove flag is set, and if_attach_internal() adds only IFG_ALL. 715 * Thus, when if_vmove() is applied to a cloned interface, group membership 716 * is lost while a cloned one always joins a group whose name is 717 * ifc->ifc_name. To recover this after if_detach_internal() and 718 * if_attach_internal(), the cloner should be specified to 719 * if_attach_internal() via ifc. If it is non-NULL, if_attach_internal() 720 * attempts to join a group whose name is ifc->ifc_name. 721 * 722 * XXX: 723 * - The decision to return void and thus require this function to 724 * succeed is questionable. 725 * - We should probably do more sanity checking. For instance we don't 726 * do anything to insure if_xname is unique or non-empty. 727 */ 728 void 729 if_attach(struct ifnet *ifp) 730 { 731 732 if_attach_internal(ifp, 0, NULL); 733 } 734 735 /* 736 * Compute the least common TSO limit. 737 */ 738 void 739 if_hw_tsomax_common(if_t ifp, struct ifnet_hw_tsomax *pmax) 740 { 741 /* 742 * 1) If there is no limit currently, take the limit from 743 * the network adapter. 744 * 745 * 2) If the network adapter has a limit below the current 746 * limit, apply it. 747 */ 748 if (pmax->tsomaxbytes == 0 || (ifp->if_hw_tsomax != 0 && 749 ifp->if_hw_tsomax < pmax->tsomaxbytes)) { 750 pmax->tsomaxbytes = ifp->if_hw_tsomax; 751 } 752 if (pmax->tsomaxsegcount == 0 || (ifp->if_hw_tsomaxsegcount != 0 && 753 ifp->if_hw_tsomaxsegcount < pmax->tsomaxsegcount)) { 754 pmax->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 755 } 756 if (pmax->tsomaxsegsize == 0 || (ifp->if_hw_tsomaxsegsize != 0 && 757 ifp->if_hw_tsomaxsegsize < pmax->tsomaxsegsize)) { 758 pmax->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 759 } 760 } 761 762 /* 763 * Update TSO limit of a network adapter. 764 * 765 * Returns zero if no change. Else non-zero. 766 */ 767 int 768 if_hw_tsomax_update(if_t ifp, struct ifnet_hw_tsomax *pmax) 769 { 770 int retval = 0; 771 if (ifp->if_hw_tsomax != pmax->tsomaxbytes) { 772 ifp->if_hw_tsomax = pmax->tsomaxbytes; 773 retval++; 774 } 775 if (ifp->if_hw_tsomaxsegsize != pmax->tsomaxsegsize) { 776 ifp->if_hw_tsomaxsegsize = pmax->tsomaxsegsize; 777 retval++; 778 } 779 if (ifp->if_hw_tsomaxsegcount != pmax->tsomaxsegcount) { 780 ifp->if_hw_tsomaxsegcount = pmax->tsomaxsegcount; 781 retval++; 782 } 783 return (retval); 784 } 785 786 static void 787 if_attach_internal(struct ifnet *ifp, int vmove, struct if_clone *ifc) 788 { 789 unsigned socksize, ifasize; 790 int namelen, masklen; 791 struct sockaddr_dl *sdl; 792 struct ifaddr *ifa; 793 794 if (ifp->if_index == 0 || ifp != ifnet_byindex(ifp->if_index)) 795 panic ("%s: BUG: if_attach called without if_alloc'd input()\n", 796 ifp->if_xname); 797 798 #ifdef VIMAGE 799 ifp->if_vnet = curvnet; 800 if (ifp->if_home_vnet == NULL) 801 ifp->if_home_vnet = curvnet; 802 #endif 803 804 if_addgroup(ifp, IFG_ALL); 805 806 /* Restore group membership for cloned interfaces. */ 807 if (vmove && ifc != NULL) 808 if_clone_addgroup(ifp, ifc); 809 810 getmicrotime(&ifp->if_lastchange); 811 ifp->if_epoch = time_uptime; 812 813 KASSERT((ifp->if_transmit == NULL && ifp->if_qflush == NULL) || 814 (ifp->if_transmit != NULL && ifp->if_qflush != NULL), 815 ("transmit and qflush must both either be set or both be NULL")); 816 if (ifp->if_transmit == NULL) { 817 ifp->if_transmit = if_transmit; 818 ifp->if_qflush = if_qflush; 819 } 820 if (ifp->if_input == NULL) 821 ifp->if_input = if_input_default; 822 823 if (ifp->if_requestencap == NULL) 824 ifp->if_requestencap = if_requestencap_default; 825 826 if (!vmove) { 827 #ifdef MAC 828 mac_ifnet_create(ifp); 829 #endif 830 831 /* 832 * Create a Link Level name for this device. 833 */ 834 namelen = strlen(ifp->if_xname); 835 /* 836 * Always save enough space for any possiable name so we 837 * can do a rename in place later. 838 */ 839 masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + IFNAMSIZ; 840 socksize = masklen + ifp->if_addrlen; 841 if (socksize < sizeof(*sdl)) 842 socksize = sizeof(*sdl); 843 socksize = roundup2(socksize, sizeof(long)); 844 ifasize = sizeof(*ifa) + 2 * socksize; 845 ifa = ifa_alloc(ifasize, M_WAITOK); 846 sdl = (struct sockaddr_dl *)(ifa + 1); 847 sdl->sdl_len = socksize; 848 sdl->sdl_family = AF_LINK; 849 bcopy(ifp->if_xname, sdl->sdl_data, namelen); 850 sdl->sdl_nlen = namelen; 851 sdl->sdl_index = ifp->if_index; 852 sdl->sdl_type = ifp->if_type; 853 ifp->if_addr = ifa; 854 ifa->ifa_ifp = ifp; 855 ifa->ifa_addr = (struct sockaddr *)sdl; 856 sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl); 857 ifa->ifa_netmask = (struct sockaddr *)sdl; 858 sdl->sdl_len = masklen; 859 while (namelen != 0) 860 sdl->sdl_data[--namelen] = 0xff; 861 CK_STAILQ_INSERT_HEAD(&ifp->if_addrhead, ifa, ifa_link); 862 /* Reliably crash if used uninitialized. */ 863 ifp->if_broadcastaddr = NULL; 864 865 if (ifp->if_type == IFT_ETHER) { 866 ifp->if_hw_addr = malloc(ifp->if_addrlen, M_IFADDR, 867 M_WAITOK | M_ZERO); 868 } 869 870 #if defined(INET) || defined(INET6) 871 /* Use defaults for TSO, if nothing is set */ 872 if (ifp->if_hw_tsomax == 0 && 873 ifp->if_hw_tsomaxsegcount == 0 && 874 ifp->if_hw_tsomaxsegsize == 0) { 875 /* 876 * The TSO defaults needs to be such that an 877 * NFS mbuf list of 35 mbufs totalling just 878 * below 64K works and that a chain of mbufs 879 * can be defragged into at most 32 segments: 880 */ 881 ifp->if_hw_tsomax = min(IP_MAXPACKET, (32 * MCLBYTES) - 882 (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN)); 883 ifp->if_hw_tsomaxsegcount = 35; 884 ifp->if_hw_tsomaxsegsize = 2048; /* 2K */ 885 886 /* XXX some drivers set IFCAP_TSO after ethernet attach */ 887 if (ifp->if_capabilities & IFCAP_TSO) { 888 if_printf(ifp, "Using defaults for TSO: %u/%u/%u\n", 889 ifp->if_hw_tsomax, 890 ifp->if_hw_tsomaxsegcount, 891 ifp->if_hw_tsomaxsegsize); 892 } 893 } 894 #endif 895 } 896 #ifdef VIMAGE 897 else { 898 /* 899 * Update the interface index in the link layer address 900 * of the interface. 901 */ 902 for (ifa = ifp->if_addr; ifa != NULL; 903 ifa = CK_STAILQ_NEXT(ifa, ifa_link)) { 904 if (ifa->ifa_addr->sa_family == AF_LINK) { 905 sdl = (struct sockaddr_dl *)ifa->ifa_addr; 906 sdl->sdl_index = ifp->if_index; 907 } 908 } 909 } 910 #endif 911 912 IFNET_WLOCK(); 913 CK_STAILQ_INSERT_TAIL(&V_ifnet, ifp, if_link); 914 #ifdef VIMAGE 915 curvnet->vnet_ifcnt++; 916 #endif 917 IFNET_WUNLOCK(); 918 919 if (domain_init_status >= 2) 920 if_attachdomain1(ifp); 921 922 EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp); 923 if (IS_DEFAULT_VNET(curvnet)) 924 devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL); 925 926 /* Announce the interface. */ 927 rt_ifannouncemsg(ifp, IFAN_ARRIVAL); 928 } 929 930 static void 931 if_epochalloc(void *dummy __unused) 932 { 933 934 net_epoch_preempt = epoch_alloc("Net preemptible", EPOCH_PREEMPT); 935 net_epoch = epoch_alloc("Net", 0); 936 } 937 SYSINIT(ifepochalloc, SI_SUB_EPOCH, SI_ORDER_ANY, if_epochalloc, NULL); 938 939 static void 940 if_attachdomain(void *dummy) 941 { 942 struct ifnet *ifp; 943 944 CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) 945 if_attachdomain1(ifp); 946 } 947 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_SECOND, 948 if_attachdomain, NULL); 949 950 static void 951 if_attachdomain1(struct ifnet *ifp) 952 { 953 struct domain *dp; 954 955 /* 956 * Since dp->dom_ifattach calls malloc() with M_WAITOK, we 957 * cannot lock ifp->if_afdata initialization, entirely. 958 */ 959 IF_AFDATA_LOCK(ifp); 960 if (ifp->if_afdata_initialized >= domain_init_status) { 961 IF_AFDATA_UNLOCK(ifp); 962 log(LOG_WARNING, "%s called more than once on %s\n", 963 __func__, ifp->if_xname); 964 return; 965 } 966 ifp->if_afdata_initialized = domain_init_status; 967 IF_AFDATA_UNLOCK(ifp); 968 969 /* address family dependent data region */ 970 bzero(ifp->if_afdata, sizeof(ifp->if_afdata)); 971 for (dp = domains; dp; dp = dp->dom_next) { 972 if (dp->dom_ifattach) 973 ifp->if_afdata[dp->dom_family] = 974 (*dp->dom_ifattach)(ifp); 975 } 976 } 977 978 /* 979 * Remove any unicast or broadcast network addresses from an interface. 980 */ 981 void 982 if_purgeaddrs(struct ifnet *ifp) 983 { 984 struct ifaddr *ifa; 985 986 while (1) { 987 struct epoch_tracker et; 988 989 NET_EPOCH_ENTER(et); 990 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 991 if (ifa->ifa_addr->sa_family != AF_LINK) 992 break; 993 } 994 NET_EPOCH_EXIT(et); 995 996 if (ifa == NULL) 997 break; 998 #ifdef INET 999 /* XXX: Ugly!! ad hoc just for INET */ 1000 if (ifa->ifa_addr->sa_family == AF_INET) { 1001 struct ifaliasreq ifr; 1002 1003 bzero(&ifr, sizeof(ifr)); 1004 ifr.ifra_addr = *ifa->ifa_addr; 1005 if (ifa->ifa_dstaddr) 1006 ifr.ifra_broadaddr = *ifa->ifa_dstaddr; 1007 if (in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp, 1008 NULL) == 0) 1009 continue; 1010 } 1011 #endif /* INET */ 1012 #ifdef INET6 1013 if (ifa->ifa_addr->sa_family == AF_INET6) { 1014 in6_purgeaddr(ifa); 1015 /* ifp_addrhead is already updated */ 1016 continue; 1017 } 1018 #endif /* INET6 */ 1019 IF_ADDR_WLOCK(ifp); 1020 CK_STAILQ_REMOVE(&ifp->if_addrhead, ifa, ifaddr, ifa_link); 1021 IF_ADDR_WUNLOCK(ifp); 1022 ifa_free(ifa); 1023 } 1024 } 1025 1026 /* 1027 * Remove any multicast network addresses from an interface when an ifnet 1028 * is going away. 1029 */ 1030 static void 1031 if_purgemaddrs(struct ifnet *ifp) 1032 { 1033 struct ifmultiaddr *ifma; 1034 1035 IF_ADDR_WLOCK(ifp); 1036 while (!CK_STAILQ_EMPTY(&ifp->if_multiaddrs)) { 1037 ifma = CK_STAILQ_FIRST(&ifp->if_multiaddrs); 1038 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifmultiaddr, ifma_link); 1039 if_delmulti_locked(ifp, ifma, 1); 1040 } 1041 IF_ADDR_WUNLOCK(ifp); 1042 } 1043 1044 /* 1045 * Detach an interface, removing it from the list of "active" interfaces. 1046 * If vmove flag is set on entry to if_detach_internal(), perform only a 1047 * limited subset of cleanup tasks, given that we are moving an ifnet from 1048 * one vnet to another, where it must be fully operational. 1049 * 1050 * XXXRW: There are some significant questions about event ordering, and 1051 * how to prevent things from starting to use the interface during detach. 1052 */ 1053 void 1054 if_detach(struct ifnet *ifp) 1055 { 1056 1057 CURVNET_SET_QUIET(ifp->if_vnet); 1058 if_detach_internal(ifp, 0, NULL); 1059 CURVNET_RESTORE(); 1060 } 1061 1062 /* 1063 * The vmove flag, if set, indicates that we are called from a callpath 1064 * that is moving an interface to a different vnet instance. 1065 * 1066 * The shutdown flag, if set, indicates that we are called in the 1067 * process of shutting down a vnet instance. Currently only the 1068 * vnet_if_return SYSUNINIT function sets it. Note: we can be called 1069 * on a vnet instance shutdown without this flag being set, e.g., when 1070 * the cloned interfaces are destoyed as first thing of teardown. 1071 */ 1072 static int 1073 if_detach_internal(struct ifnet *ifp, int vmove, struct if_clone **ifcp) 1074 { 1075 struct ifaddr *ifa; 1076 int i; 1077 struct domain *dp; 1078 struct ifnet *iter; 1079 int found = 0; 1080 #ifdef VIMAGE 1081 bool shutdown; 1082 1083 shutdown = ifp->if_vnet->vnet_shutdown; 1084 #endif 1085 IFNET_WLOCK(); 1086 CK_STAILQ_FOREACH(iter, &V_ifnet, if_link) 1087 if (iter == ifp) { 1088 CK_STAILQ_REMOVE(&V_ifnet, ifp, ifnet, if_link); 1089 if (!vmove) 1090 ifp->if_flags |= IFF_DYING; 1091 found = 1; 1092 break; 1093 } 1094 IFNET_WUNLOCK(); 1095 if (!found) { 1096 /* 1097 * While we would want to panic here, we cannot 1098 * guarantee that the interface is indeed still on 1099 * the list given we don't hold locks all the way. 1100 */ 1101 return (ENOENT); 1102 #if 0 1103 if (vmove) 1104 panic("%s: ifp=%p not on the ifnet tailq %p", 1105 __func__, ifp, &V_ifnet); 1106 else 1107 return; /* XXX this should panic as well? */ 1108 #endif 1109 } 1110 1111 /* 1112 * At this point we know the interface still was on the ifnet list 1113 * and we removed it so we are in a stable state. 1114 */ 1115 #ifdef VIMAGE 1116 curvnet->vnet_ifcnt--; 1117 #endif 1118 epoch_wait_preempt(net_epoch_preempt); 1119 1120 /* 1121 * Ensure all pending EPOCH(9) callbacks have been executed. This 1122 * fixes issues about late destruction of multicast options 1123 * which lead to leave group calls, which in turn access the 1124 * belonging ifnet structure: 1125 */ 1126 epoch_drain_callbacks(net_epoch_preempt); 1127 1128 /* 1129 * In any case (destroy or vmove) detach us from the groups 1130 * and remove/wait for pending events on the taskq. 1131 * XXX-BZ in theory an interface could still enqueue a taskq change? 1132 */ 1133 if_delgroups(ifp); 1134 1135 taskqueue_drain(taskqueue_swi, &ifp->if_linktask); 1136 taskqueue_drain(taskqueue_swi, &ifp->if_addmultitask); 1137 1138 /* 1139 * Check if this is a cloned interface or not. Must do even if 1140 * shutting down as a if_vmove_reclaim() would move the ifp and 1141 * the if_clone_addgroup() will have a corrupted string overwise 1142 * from a gibberish pointer. 1143 */ 1144 if (vmove && ifcp != NULL) 1145 *ifcp = if_clone_findifc(ifp); 1146 1147 if_down(ifp); 1148 1149 #ifdef VIMAGE 1150 /* 1151 * On VNET shutdown abort here as the stack teardown will do all 1152 * the work top-down for us. 1153 */ 1154 if (shutdown) { 1155 /* Give interface users the chance to clean up. */ 1156 EVENTHANDLER_INVOKE(ifnet_departure_event, ifp); 1157 1158 /* 1159 * In case of a vmove we are done here without error. 1160 * If we would signal an error it would lead to the same 1161 * abort as if we did not find the ifnet anymore. 1162 * if_detach() calls us in void context and does not care 1163 * about an early abort notification, so life is splendid :) 1164 */ 1165 goto finish_vnet_shutdown; 1166 } 1167 #endif 1168 1169 /* 1170 * At this point we are not tearing down a VNET and are either 1171 * going to destroy or vmove the interface and have to cleanup 1172 * accordingly. 1173 */ 1174 1175 /* 1176 * Remove routes and flush queues. 1177 */ 1178 #ifdef ALTQ 1179 if (ALTQ_IS_ENABLED(&ifp->if_snd)) 1180 altq_disable(&ifp->if_snd); 1181 if (ALTQ_IS_ATTACHED(&ifp->if_snd)) 1182 altq_detach(&ifp->if_snd); 1183 #endif 1184 1185 if_purgeaddrs(ifp); 1186 1187 #ifdef INET 1188 in_ifdetach(ifp); 1189 #endif 1190 1191 #ifdef INET6 1192 /* 1193 * Remove all IPv6 kernel structs related to ifp. This should be done 1194 * before removing routing entries below, since IPv6 interface direct 1195 * routes are expected to be removed by the IPv6-specific kernel API. 1196 * Otherwise, the kernel will detect some inconsistency and bark it. 1197 */ 1198 in6_ifdetach(ifp); 1199 #endif 1200 if_purgemaddrs(ifp); 1201 1202 /* Announce that the interface is gone. */ 1203 rt_ifannouncemsg(ifp, IFAN_DEPARTURE); 1204 EVENTHANDLER_INVOKE(ifnet_departure_event, ifp); 1205 if (IS_DEFAULT_VNET(curvnet)) 1206 devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL); 1207 1208 if (!vmove) { 1209 /* 1210 * Prevent further calls into the device driver via ifnet. 1211 */ 1212 if_dead(ifp); 1213 1214 /* 1215 * Clean up all addresses. 1216 */ 1217 IF_ADDR_WLOCK(ifp); 1218 if (!CK_STAILQ_EMPTY(&ifp->if_addrhead)) { 1219 ifa = CK_STAILQ_FIRST(&ifp->if_addrhead); 1220 CK_STAILQ_REMOVE(&ifp->if_addrhead, ifa, ifaddr, ifa_link); 1221 IF_ADDR_WUNLOCK(ifp); 1222 ifa_free(ifa); 1223 } else 1224 IF_ADDR_WUNLOCK(ifp); 1225 } 1226 1227 rt_flushifroutes(ifp); 1228 1229 #ifdef VIMAGE 1230 finish_vnet_shutdown: 1231 #endif 1232 /* 1233 * We cannot hold the lock over dom_ifdetach calls as they might 1234 * sleep, for example trying to drain a callout, thus open up the 1235 * theoretical race with re-attaching. 1236 */ 1237 IF_AFDATA_LOCK(ifp); 1238 i = ifp->if_afdata_initialized; 1239 ifp->if_afdata_initialized = 0; 1240 IF_AFDATA_UNLOCK(ifp); 1241 for (dp = domains; i > 0 && dp; dp = dp->dom_next) { 1242 if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family]) { 1243 (*dp->dom_ifdetach)(ifp, 1244 ifp->if_afdata[dp->dom_family]); 1245 ifp->if_afdata[dp->dom_family] = NULL; 1246 } 1247 } 1248 1249 return (0); 1250 } 1251 1252 #ifdef VIMAGE 1253 /* 1254 * if_vmove() performs a limited version of if_detach() in current 1255 * vnet and if_attach()es the ifnet to the vnet specified as 2nd arg. 1256 * An attempt is made to shrink if_index in current vnet, find an 1257 * unused if_index in target vnet and calls if_grow() if necessary, 1258 * and finally find an unused if_xname for the target vnet. 1259 */ 1260 static void 1261 if_vmove(struct ifnet *ifp, struct vnet *new_vnet) 1262 { 1263 struct if_clone *ifc; 1264 #ifdef DEV_BPF 1265 u_int bif_dlt, bif_hdrlen; 1266 #endif 1267 void *old; 1268 int rc; 1269 1270 #ifdef DEV_BPF 1271 /* 1272 * if_detach_internal() will call the eventhandler to notify 1273 * interface departure. That will detach if_bpf. We need to 1274 * safe the dlt and hdrlen so we can re-attach it later. 1275 */ 1276 bpf_get_bp_params(ifp->if_bpf, &bif_dlt, &bif_hdrlen); 1277 #endif 1278 1279 /* 1280 * Detach from current vnet, but preserve LLADDR info, do not 1281 * mark as dead etc. so that the ifnet can be reattached later. 1282 * If we cannot find it, we lost the race to someone else. 1283 */ 1284 rc = if_detach_internal(ifp, 1, &ifc); 1285 if (rc != 0) 1286 return; 1287 1288 /* 1289 * Unlink the ifnet from ifindex_table[] in current vnet, and shrink 1290 * the if_index for that vnet if possible. 1291 * 1292 * NOTE: IFNET_WLOCK/IFNET_WUNLOCK() are assumed to be unvirtualized, 1293 * or we'd lock on one vnet and unlock on another. 1294 */ 1295 IFNET_WLOCK(); 1296 ifindex_free_locked(ifp->if_index); 1297 IFNET_WUNLOCK(); 1298 1299 /* 1300 * Perform interface-specific reassignment tasks, if provided by 1301 * the driver. 1302 */ 1303 if (ifp->if_reassign != NULL) 1304 ifp->if_reassign(ifp, new_vnet, NULL); 1305 1306 /* 1307 * Switch to the context of the target vnet. 1308 */ 1309 CURVNET_SET_QUIET(new_vnet); 1310 restart: 1311 IFNET_WLOCK(); 1312 ifp->if_index = ifindex_alloc(&old); 1313 if (__predict_false(ifp->if_index == USHRT_MAX)) { 1314 IFNET_WUNLOCK(); 1315 epoch_wait_preempt(net_epoch_preempt); 1316 free(old, M_IFNET); 1317 goto restart; 1318 } 1319 ifnet_setbyindex(ifp->if_index, ifp); 1320 IFNET_WUNLOCK(); 1321 1322 if_attach_internal(ifp, 1, ifc); 1323 1324 #ifdef DEV_BPF 1325 if (ifp->if_bpf == NULL) 1326 bpfattach(ifp, bif_dlt, bif_hdrlen); 1327 #endif 1328 1329 CURVNET_RESTORE(); 1330 } 1331 1332 /* 1333 * Move an ifnet to or from another child prison/vnet, specified by the jail id. 1334 */ 1335 static int 1336 if_vmove_loan(struct thread *td, struct ifnet *ifp, char *ifname, int jid) 1337 { 1338 struct prison *pr; 1339 struct ifnet *difp; 1340 1341 /* Try to find the prison within our visibility. */ 1342 sx_slock(&allprison_lock); 1343 pr = prison_find_child(td->td_ucred->cr_prison, jid); 1344 sx_sunlock(&allprison_lock); 1345 if (pr == NULL) 1346 return (ENXIO); 1347 prison_hold_locked(pr); 1348 mtx_unlock(&pr->pr_mtx); 1349 1350 /* Do not try to move the iface from and to the same prison. */ 1351 if (pr->pr_vnet == ifp->if_vnet) { 1352 prison_free(pr); 1353 return (EEXIST); 1354 } 1355 1356 /* Make sure the named iface does not exists in the dst. prison/vnet. */ 1357 /* XXX Lock interfaces to avoid races. */ 1358 CURVNET_SET_QUIET(pr->pr_vnet); 1359 difp = ifunit(ifname); 1360 if (difp != NULL) { 1361 CURVNET_RESTORE(); 1362 prison_free(pr); 1363 return (EEXIST); 1364 } 1365 1366 /* Make sure the VNET is stable. */ 1367 if (ifp->if_vnet->vnet_shutdown) { 1368 CURVNET_RESTORE(); 1369 prison_free(pr); 1370 return (EBUSY); 1371 } 1372 CURVNET_RESTORE(); 1373 1374 /* Move the interface into the child jail/vnet. */ 1375 if_vmove(ifp, pr->pr_vnet); 1376 1377 /* Report the new if_xname back to the userland. */ 1378 sprintf(ifname, "%s", ifp->if_xname); 1379 1380 prison_free(pr); 1381 return (0); 1382 } 1383 1384 static int 1385 if_vmove_reclaim(struct thread *td, char *ifname, int jid) 1386 { 1387 struct prison *pr; 1388 struct vnet *vnet_dst; 1389 struct ifnet *ifp; 1390 1391 /* Try to find the prison within our visibility. */ 1392 sx_slock(&allprison_lock); 1393 pr = prison_find_child(td->td_ucred->cr_prison, jid); 1394 sx_sunlock(&allprison_lock); 1395 if (pr == NULL) 1396 return (ENXIO); 1397 prison_hold_locked(pr); 1398 mtx_unlock(&pr->pr_mtx); 1399 1400 /* Make sure the named iface exists in the source prison/vnet. */ 1401 CURVNET_SET(pr->pr_vnet); 1402 ifp = ifunit(ifname); /* XXX Lock to avoid races. */ 1403 if (ifp == NULL) { 1404 CURVNET_RESTORE(); 1405 prison_free(pr); 1406 return (ENXIO); 1407 } 1408 1409 /* Do not try to move the iface from and to the same prison. */ 1410 vnet_dst = TD_TO_VNET(td); 1411 if (vnet_dst == ifp->if_vnet) { 1412 CURVNET_RESTORE(); 1413 prison_free(pr); 1414 return (EEXIST); 1415 } 1416 1417 /* Make sure the VNET is stable. */ 1418 if (ifp->if_vnet->vnet_shutdown) { 1419 CURVNET_RESTORE(); 1420 prison_free(pr); 1421 return (EBUSY); 1422 } 1423 1424 /* Get interface back from child jail/vnet. */ 1425 if_vmove(ifp, vnet_dst); 1426 CURVNET_RESTORE(); 1427 1428 /* Report the new if_xname back to the userland. */ 1429 sprintf(ifname, "%s", ifp->if_xname); 1430 1431 prison_free(pr); 1432 return (0); 1433 } 1434 #endif /* VIMAGE */ 1435 1436 /* 1437 * Add a group to an interface 1438 */ 1439 int 1440 if_addgroup(struct ifnet *ifp, const char *groupname) 1441 { 1442 struct ifg_list *ifgl; 1443 struct ifg_group *ifg = NULL; 1444 struct ifg_member *ifgm; 1445 int new = 0; 1446 1447 if (groupname[0] && groupname[strlen(groupname) - 1] >= '0' && 1448 groupname[strlen(groupname) - 1] <= '9') 1449 return (EINVAL); 1450 1451 IFNET_WLOCK(); 1452 CK_STAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) 1453 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname)) { 1454 IFNET_WUNLOCK(); 1455 return (EEXIST); 1456 } 1457 1458 if ((ifgl = malloc(sizeof(*ifgl), M_TEMP, M_NOWAIT)) == NULL) { 1459 IFNET_WUNLOCK(); 1460 return (ENOMEM); 1461 } 1462 1463 if ((ifgm = malloc(sizeof(*ifgm), M_TEMP, M_NOWAIT)) == NULL) { 1464 free(ifgl, M_TEMP); 1465 IFNET_WUNLOCK(); 1466 return (ENOMEM); 1467 } 1468 1469 CK_STAILQ_FOREACH(ifg, &V_ifg_head, ifg_next) 1470 if (!strcmp(ifg->ifg_group, groupname)) 1471 break; 1472 1473 if (ifg == NULL) { 1474 if ((ifg = malloc(sizeof(*ifg), M_TEMP, M_NOWAIT)) == NULL) { 1475 free(ifgl, M_TEMP); 1476 free(ifgm, M_TEMP); 1477 IFNET_WUNLOCK(); 1478 return (ENOMEM); 1479 } 1480 strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group)); 1481 ifg->ifg_refcnt = 0; 1482 CK_STAILQ_INIT(&ifg->ifg_members); 1483 CK_STAILQ_INSERT_TAIL(&V_ifg_head, ifg, ifg_next); 1484 new = 1; 1485 } 1486 1487 ifg->ifg_refcnt++; 1488 ifgl->ifgl_group = ifg; 1489 ifgm->ifgm_ifp = ifp; 1490 1491 IF_ADDR_WLOCK(ifp); 1492 CK_STAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next); 1493 CK_STAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next); 1494 IF_ADDR_WUNLOCK(ifp); 1495 1496 IFNET_WUNLOCK(); 1497 1498 if (new) 1499 EVENTHANDLER_INVOKE(group_attach_event, ifg); 1500 EVENTHANDLER_INVOKE(group_change_event, groupname); 1501 1502 return (0); 1503 } 1504 1505 /* 1506 * Helper function to remove a group out of an interface. Expects the global 1507 * ifnet lock to be write-locked, and drops it before returning. 1508 */ 1509 static void 1510 _if_delgroup_locked(struct ifnet *ifp, struct ifg_list *ifgl, 1511 const char *groupname) 1512 { 1513 struct ifg_member *ifgm; 1514 bool freeifgl; 1515 1516 IFNET_WLOCK_ASSERT(); 1517 1518 IF_ADDR_WLOCK(ifp); 1519 CK_STAILQ_REMOVE(&ifp->if_groups, ifgl, ifg_list, ifgl_next); 1520 IF_ADDR_WUNLOCK(ifp); 1521 1522 CK_STAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next) { 1523 if (ifgm->ifgm_ifp == ifp) { 1524 CK_STAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, 1525 ifg_member, ifgm_next); 1526 break; 1527 } 1528 } 1529 1530 if (--ifgl->ifgl_group->ifg_refcnt == 0) { 1531 CK_STAILQ_REMOVE(&V_ifg_head, ifgl->ifgl_group, ifg_group, 1532 ifg_next); 1533 freeifgl = true; 1534 } else { 1535 freeifgl = false; 1536 } 1537 IFNET_WUNLOCK(); 1538 1539 epoch_wait_preempt(net_epoch_preempt); 1540 if (freeifgl) { 1541 EVENTHANDLER_INVOKE(group_detach_event, ifgl->ifgl_group); 1542 free(ifgl->ifgl_group, M_TEMP); 1543 } 1544 free(ifgm, M_TEMP); 1545 free(ifgl, M_TEMP); 1546 1547 EVENTHANDLER_INVOKE(group_change_event, groupname); 1548 } 1549 1550 /* 1551 * Remove a group from an interface 1552 */ 1553 int 1554 if_delgroup(struct ifnet *ifp, const char *groupname) 1555 { 1556 struct ifg_list *ifgl; 1557 1558 IFNET_WLOCK(); 1559 CK_STAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) 1560 if (strcmp(ifgl->ifgl_group->ifg_group, groupname) == 0) 1561 break; 1562 if (ifgl == NULL) { 1563 IFNET_WUNLOCK(); 1564 return (ENOENT); 1565 } 1566 1567 _if_delgroup_locked(ifp, ifgl, groupname); 1568 1569 return (0); 1570 } 1571 1572 /* 1573 * Remove an interface from all groups 1574 */ 1575 static void 1576 if_delgroups(struct ifnet *ifp) 1577 { 1578 struct ifg_list *ifgl; 1579 char groupname[IFNAMSIZ]; 1580 1581 IFNET_WLOCK(); 1582 while ((ifgl = CK_STAILQ_FIRST(&ifp->if_groups)) != NULL) { 1583 strlcpy(groupname, ifgl->ifgl_group->ifg_group, IFNAMSIZ); 1584 _if_delgroup_locked(ifp, ifgl, groupname); 1585 IFNET_WLOCK(); 1586 } 1587 IFNET_WUNLOCK(); 1588 } 1589 1590 static char * 1591 ifgr_group_get(void *ifgrp) 1592 { 1593 union ifgroupreq_union *ifgrup; 1594 1595 ifgrup = ifgrp; 1596 #ifdef COMPAT_FREEBSD32 1597 if (SV_CURPROC_FLAG(SV_ILP32)) 1598 return (&ifgrup->ifgr32.ifgr_ifgru.ifgru_group[0]); 1599 #endif 1600 return (&ifgrup->ifgr.ifgr_ifgru.ifgru_group[0]); 1601 } 1602 1603 static struct ifg_req * 1604 ifgr_groups_get(void *ifgrp) 1605 { 1606 union ifgroupreq_union *ifgrup; 1607 1608 ifgrup = ifgrp; 1609 #ifdef COMPAT_FREEBSD32 1610 if (SV_CURPROC_FLAG(SV_ILP32)) 1611 return ((struct ifg_req *)(uintptr_t) 1612 ifgrup->ifgr32.ifgr_ifgru.ifgru_groups); 1613 #endif 1614 return (ifgrup->ifgr.ifgr_ifgru.ifgru_groups); 1615 } 1616 1617 /* 1618 * Stores all groups from an interface in memory pointed to by ifgr. 1619 */ 1620 static int 1621 if_getgroup(struct ifgroupreq *ifgr, struct ifnet *ifp) 1622 { 1623 int len, error; 1624 struct ifg_list *ifgl; 1625 struct ifg_req ifgrq, *ifgp; 1626 1627 NET_EPOCH_ASSERT(); 1628 1629 if (ifgr->ifgr_len == 0) { 1630 CK_STAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) 1631 ifgr->ifgr_len += sizeof(struct ifg_req); 1632 return (0); 1633 } 1634 1635 len = ifgr->ifgr_len; 1636 ifgp = ifgr_groups_get(ifgr); 1637 /* XXX: wire */ 1638 CK_STAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) { 1639 if (len < sizeof(ifgrq)) 1640 return (EINVAL); 1641 bzero(&ifgrq, sizeof ifgrq); 1642 strlcpy(ifgrq.ifgrq_group, ifgl->ifgl_group->ifg_group, 1643 sizeof(ifgrq.ifgrq_group)); 1644 if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req)))) 1645 return (error); 1646 len -= sizeof(ifgrq); 1647 ifgp++; 1648 } 1649 1650 return (0); 1651 } 1652 1653 /* 1654 * Stores all members of a group in memory pointed to by igfr 1655 */ 1656 static int 1657 if_getgroupmembers(struct ifgroupreq *ifgr) 1658 { 1659 struct ifg_group *ifg; 1660 struct ifg_member *ifgm; 1661 struct ifg_req ifgrq, *ifgp; 1662 int len, error; 1663 1664 IFNET_RLOCK(); 1665 CK_STAILQ_FOREACH(ifg, &V_ifg_head, ifg_next) 1666 if (strcmp(ifg->ifg_group, ifgr->ifgr_name) == 0) 1667 break; 1668 if (ifg == NULL) { 1669 IFNET_RUNLOCK(); 1670 return (ENOENT); 1671 } 1672 1673 if (ifgr->ifgr_len == 0) { 1674 CK_STAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) 1675 ifgr->ifgr_len += sizeof(ifgrq); 1676 IFNET_RUNLOCK(); 1677 return (0); 1678 } 1679 1680 len = ifgr->ifgr_len; 1681 ifgp = ifgr_groups_get(ifgr); 1682 CK_STAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) { 1683 if (len < sizeof(ifgrq)) { 1684 IFNET_RUNLOCK(); 1685 return (EINVAL); 1686 } 1687 bzero(&ifgrq, sizeof ifgrq); 1688 strlcpy(ifgrq.ifgrq_member, ifgm->ifgm_ifp->if_xname, 1689 sizeof(ifgrq.ifgrq_member)); 1690 if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req)))) { 1691 IFNET_RUNLOCK(); 1692 return (error); 1693 } 1694 len -= sizeof(ifgrq); 1695 ifgp++; 1696 } 1697 IFNET_RUNLOCK(); 1698 1699 return (0); 1700 } 1701 1702 /* 1703 * Return counter values from counter(9)s stored in ifnet. 1704 */ 1705 uint64_t 1706 if_get_counter_default(struct ifnet *ifp, ift_counter cnt) 1707 { 1708 1709 KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt)); 1710 1711 return (counter_u64_fetch(ifp->if_counters[cnt])); 1712 } 1713 1714 /* 1715 * Increase an ifnet counter. Usually used for counters shared 1716 * between the stack and a driver, but function supports them all. 1717 */ 1718 void 1719 if_inc_counter(struct ifnet *ifp, ift_counter cnt, int64_t inc) 1720 { 1721 1722 KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt)); 1723 1724 counter_u64_add(ifp->if_counters[cnt], inc); 1725 } 1726 1727 /* 1728 * Copy data from ifnet to userland API structure if_data. 1729 */ 1730 void 1731 if_data_copy(struct ifnet *ifp, struct if_data *ifd) 1732 { 1733 1734 ifd->ifi_type = ifp->if_type; 1735 ifd->ifi_physical = 0; 1736 ifd->ifi_addrlen = ifp->if_addrlen; 1737 ifd->ifi_hdrlen = ifp->if_hdrlen; 1738 ifd->ifi_link_state = ifp->if_link_state; 1739 ifd->ifi_vhid = 0; 1740 ifd->ifi_datalen = sizeof(struct if_data); 1741 ifd->ifi_mtu = ifp->if_mtu; 1742 ifd->ifi_metric = ifp->if_metric; 1743 ifd->ifi_baudrate = ifp->if_baudrate; 1744 ifd->ifi_hwassist = ifp->if_hwassist; 1745 ifd->ifi_epoch = ifp->if_epoch; 1746 ifd->ifi_lastchange = ifp->if_lastchange; 1747 1748 ifd->ifi_ipackets = ifp->if_get_counter(ifp, IFCOUNTER_IPACKETS); 1749 ifd->ifi_ierrors = ifp->if_get_counter(ifp, IFCOUNTER_IERRORS); 1750 ifd->ifi_opackets = ifp->if_get_counter(ifp, IFCOUNTER_OPACKETS); 1751 ifd->ifi_oerrors = ifp->if_get_counter(ifp, IFCOUNTER_OERRORS); 1752 ifd->ifi_collisions = ifp->if_get_counter(ifp, IFCOUNTER_COLLISIONS); 1753 ifd->ifi_ibytes = ifp->if_get_counter(ifp, IFCOUNTER_IBYTES); 1754 ifd->ifi_obytes = ifp->if_get_counter(ifp, IFCOUNTER_OBYTES); 1755 ifd->ifi_imcasts = ifp->if_get_counter(ifp, IFCOUNTER_IMCASTS); 1756 ifd->ifi_omcasts = ifp->if_get_counter(ifp, IFCOUNTER_OMCASTS); 1757 ifd->ifi_iqdrops = ifp->if_get_counter(ifp, IFCOUNTER_IQDROPS); 1758 ifd->ifi_oqdrops = ifp->if_get_counter(ifp, IFCOUNTER_OQDROPS); 1759 ifd->ifi_noproto = ifp->if_get_counter(ifp, IFCOUNTER_NOPROTO); 1760 } 1761 1762 /* 1763 * Initialization, destruction and refcounting functions for ifaddrs. 1764 */ 1765 struct ifaddr * 1766 ifa_alloc(size_t size, int flags) 1767 { 1768 struct ifaddr *ifa; 1769 1770 KASSERT(size >= sizeof(struct ifaddr), 1771 ("%s: invalid size %zu", __func__, size)); 1772 1773 ifa = malloc(size, M_IFADDR, M_ZERO | flags); 1774 if (ifa == NULL) 1775 return (NULL); 1776 1777 if ((ifa->ifa_opackets = counter_u64_alloc(flags)) == NULL) 1778 goto fail; 1779 if ((ifa->ifa_ipackets = counter_u64_alloc(flags)) == NULL) 1780 goto fail; 1781 if ((ifa->ifa_obytes = counter_u64_alloc(flags)) == NULL) 1782 goto fail; 1783 if ((ifa->ifa_ibytes = counter_u64_alloc(flags)) == NULL) 1784 goto fail; 1785 1786 refcount_init(&ifa->ifa_refcnt, 1); 1787 1788 return (ifa); 1789 1790 fail: 1791 /* free(NULL) is okay */ 1792 counter_u64_free(ifa->ifa_opackets); 1793 counter_u64_free(ifa->ifa_ipackets); 1794 counter_u64_free(ifa->ifa_obytes); 1795 counter_u64_free(ifa->ifa_ibytes); 1796 free(ifa, M_IFADDR); 1797 1798 return (NULL); 1799 } 1800 1801 void 1802 ifa_ref(struct ifaddr *ifa) 1803 { 1804 1805 refcount_acquire(&ifa->ifa_refcnt); 1806 } 1807 1808 static void 1809 ifa_destroy(epoch_context_t ctx) 1810 { 1811 struct ifaddr *ifa; 1812 1813 ifa = __containerof(ctx, struct ifaddr, ifa_epoch_ctx); 1814 counter_u64_free(ifa->ifa_opackets); 1815 counter_u64_free(ifa->ifa_ipackets); 1816 counter_u64_free(ifa->ifa_obytes); 1817 counter_u64_free(ifa->ifa_ibytes); 1818 free(ifa, M_IFADDR); 1819 } 1820 1821 void 1822 ifa_free(struct ifaddr *ifa) 1823 { 1824 1825 if (refcount_release(&ifa->ifa_refcnt)) 1826 epoch_call(net_epoch_preempt, &ifa->ifa_epoch_ctx, ifa_destroy); 1827 } 1828 1829 1830 static int 1831 ifa_maintain_loopback_route(int cmd, const char *otype, struct ifaddr *ifa, 1832 struct sockaddr *ia) 1833 { 1834 struct epoch_tracker et; 1835 int error; 1836 struct rt_addrinfo info; 1837 struct sockaddr_dl null_sdl; 1838 struct ifnet *ifp; 1839 struct ifaddr *rti_ifa = NULL; 1840 1841 ifp = ifa->ifa_ifp; 1842 1843 bzero(&info, sizeof(info)); 1844 if (cmd != RTM_DELETE) 1845 info.rti_ifp = V_loif; 1846 if (cmd == RTM_ADD) { 1847 /* explicitly specify (loopback) ifa */ 1848 if (info.rti_ifp != NULL) { 1849 NET_EPOCH_ENTER(et); 1850 rti_ifa = ifaof_ifpforaddr(ifa->ifa_addr, info.rti_ifp); 1851 if (rti_ifa != NULL) 1852 ifa_ref(rti_ifa); 1853 info.rti_ifa = rti_ifa; 1854 NET_EPOCH_EXIT(et); 1855 } 1856 } 1857 info.rti_flags = ifa->ifa_flags | RTF_HOST | RTF_STATIC | RTF_PINNED; 1858 info.rti_info[RTAX_DST] = ia; 1859 info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&null_sdl; 1860 link_init_sdl(ifp, (struct sockaddr *)&null_sdl, ifp->if_type); 1861 1862 error = rtrequest1_fib(cmd, &info, NULL, ifp->if_fib); 1863 1864 if (rti_ifa != NULL) 1865 ifa_free(rti_ifa); 1866 1867 if (error != 0 && 1868 !(cmd == RTM_ADD && error == EEXIST) && 1869 !(cmd == RTM_DELETE && error == ENOENT)) 1870 if_printf(ifp, "%s failed: %d\n", otype, error); 1871 1872 return (error); 1873 } 1874 1875 int 1876 ifa_add_loopback_route(struct ifaddr *ifa, struct sockaddr *ia) 1877 { 1878 1879 return (ifa_maintain_loopback_route(RTM_ADD, "insertion", ifa, ia)); 1880 } 1881 1882 int 1883 ifa_del_loopback_route(struct ifaddr *ifa, struct sockaddr *ia) 1884 { 1885 1886 return (ifa_maintain_loopback_route(RTM_DELETE, "deletion", ifa, ia)); 1887 } 1888 1889 int 1890 ifa_switch_loopback_route(struct ifaddr *ifa, struct sockaddr *ia) 1891 { 1892 1893 return (ifa_maintain_loopback_route(RTM_CHANGE, "switch", ifa, ia)); 1894 } 1895 1896 /* 1897 * XXX: Because sockaddr_dl has deeper structure than the sockaddr 1898 * structs used to represent other address families, it is necessary 1899 * to perform a different comparison. 1900 */ 1901 1902 #define sa_dl_equal(a1, a2) \ 1903 ((((const struct sockaddr_dl *)(a1))->sdl_len == \ 1904 ((const struct sockaddr_dl *)(a2))->sdl_len) && \ 1905 (bcmp(CLLADDR((const struct sockaddr_dl *)(a1)), \ 1906 CLLADDR((const struct sockaddr_dl *)(a2)), \ 1907 ((const struct sockaddr_dl *)(a1))->sdl_alen) == 0)) 1908 1909 /* 1910 * Locate an interface based on a complete address. 1911 */ 1912 /*ARGSUSED*/ 1913 struct ifaddr * 1914 ifa_ifwithaddr(const struct sockaddr *addr) 1915 { 1916 struct ifnet *ifp; 1917 struct ifaddr *ifa; 1918 1919 NET_EPOCH_ASSERT(); 1920 1921 CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) { 1922 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1923 if (ifa->ifa_addr->sa_family != addr->sa_family) 1924 continue; 1925 if (sa_equal(addr, ifa->ifa_addr)) { 1926 goto done; 1927 } 1928 /* IP6 doesn't have broadcast */ 1929 if ((ifp->if_flags & IFF_BROADCAST) && 1930 ifa->ifa_broadaddr && 1931 ifa->ifa_broadaddr->sa_len != 0 && 1932 sa_equal(ifa->ifa_broadaddr, addr)) { 1933 goto done; 1934 } 1935 } 1936 } 1937 ifa = NULL; 1938 done: 1939 return (ifa); 1940 } 1941 1942 int 1943 ifa_ifwithaddr_check(const struct sockaddr *addr) 1944 { 1945 struct epoch_tracker et; 1946 int rc; 1947 1948 NET_EPOCH_ENTER(et); 1949 rc = (ifa_ifwithaddr(addr) != NULL); 1950 NET_EPOCH_EXIT(et); 1951 return (rc); 1952 } 1953 1954 /* 1955 * Locate an interface based on the broadcast address. 1956 */ 1957 /* ARGSUSED */ 1958 struct ifaddr * 1959 ifa_ifwithbroadaddr(const struct sockaddr *addr, int fibnum) 1960 { 1961 struct ifnet *ifp; 1962 struct ifaddr *ifa; 1963 1964 MPASS(in_epoch(net_epoch_preempt)); 1965 CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) { 1966 if ((fibnum != RT_ALL_FIBS) && (ifp->if_fib != fibnum)) 1967 continue; 1968 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1969 if (ifa->ifa_addr->sa_family != addr->sa_family) 1970 continue; 1971 if ((ifp->if_flags & IFF_BROADCAST) && 1972 ifa->ifa_broadaddr && 1973 ifa->ifa_broadaddr->sa_len != 0 && 1974 sa_equal(ifa->ifa_broadaddr, addr)) { 1975 goto done; 1976 } 1977 } 1978 } 1979 ifa = NULL; 1980 done: 1981 return (ifa); 1982 } 1983 1984 /* 1985 * Locate the point to point interface with a given destination address. 1986 */ 1987 /*ARGSUSED*/ 1988 struct ifaddr * 1989 ifa_ifwithdstaddr(const struct sockaddr *addr, int fibnum) 1990 { 1991 struct ifnet *ifp; 1992 struct ifaddr *ifa; 1993 1994 MPASS(in_epoch(net_epoch_preempt)); 1995 CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) { 1996 if ((ifp->if_flags & IFF_POINTOPOINT) == 0) 1997 continue; 1998 if ((fibnum != RT_ALL_FIBS) && (ifp->if_fib != fibnum)) 1999 continue; 2000 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 2001 if (ifa->ifa_addr->sa_family != addr->sa_family) 2002 continue; 2003 if (ifa->ifa_dstaddr != NULL && 2004 sa_equal(addr, ifa->ifa_dstaddr)) { 2005 goto done; 2006 } 2007 } 2008 } 2009 ifa = NULL; 2010 done: 2011 return (ifa); 2012 } 2013 2014 /* 2015 * Find an interface on a specific network. If many, choice 2016 * is most specific found. 2017 */ 2018 struct ifaddr * 2019 ifa_ifwithnet(const struct sockaddr *addr, int ignore_ptp, int fibnum) 2020 { 2021 struct ifnet *ifp; 2022 struct ifaddr *ifa; 2023 struct ifaddr *ifa_maybe = NULL; 2024 u_int af = addr->sa_family; 2025 const char *addr_data = addr->sa_data, *cplim; 2026 2027 MPASS(in_epoch(net_epoch_preempt)); 2028 /* 2029 * AF_LINK addresses can be looked up directly by their index number, 2030 * so do that if we can. 2031 */ 2032 if (af == AF_LINK) { 2033 const struct sockaddr_dl *sdl = (const struct sockaddr_dl *)addr; 2034 if (sdl->sdl_index && sdl->sdl_index <= V_if_index) 2035 return (ifaddr_byindex(sdl->sdl_index)); 2036 } 2037 2038 /* 2039 * Scan though each interface, looking for ones that have addresses 2040 * in this address family and the requested fib. 2041 */ 2042 CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) { 2043 if ((fibnum != RT_ALL_FIBS) && (ifp->if_fib != fibnum)) 2044 continue; 2045 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 2046 const char *cp, *cp2, *cp3; 2047 2048 if (ifa->ifa_addr->sa_family != af) 2049 next: continue; 2050 if (af == AF_INET && 2051 ifp->if_flags & IFF_POINTOPOINT && !ignore_ptp) { 2052 /* 2053 * This is a bit broken as it doesn't 2054 * take into account that the remote end may 2055 * be a single node in the network we are 2056 * looking for. 2057 * The trouble is that we don't know the 2058 * netmask for the remote end. 2059 */ 2060 if (ifa->ifa_dstaddr != NULL && 2061 sa_equal(addr, ifa->ifa_dstaddr)) { 2062 goto done; 2063 } 2064 } else { 2065 /* 2066 * Scan all the bits in the ifa's address. 2067 * If a bit dissagrees with what we are 2068 * looking for, mask it with the netmask 2069 * to see if it really matters. 2070 * (A byte at a time) 2071 */ 2072 if (ifa->ifa_netmask == 0) 2073 continue; 2074 cp = addr_data; 2075 cp2 = ifa->ifa_addr->sa_data; 2076 cp3 = ifa->ifa_netmask->sa_data; 2077 cplim = ifa->ifa_netmask->sa_len 2078 + (char *)ifa->ifa_netmask; 2079 while (cp3 < cplim) 2080 if ((*cp++ ^ *cp2++) & *cp3++) 2081 goto next; /* next address! */ 2082 /* 2083 * If the netmask of what we just found 2084 * is more specific than what we had before 2085 * (if we had one), or if the virtual status 2086 * of new prefix is better than of the old one, 2087 * then remember the new one before continuing 2088 * to search for an even better one. 2089 */ 2090 if (ifa_maybe == NULL || 2091 ifa_preferred(ifa_maybe, ifa) || 2092 rn_refines((caddr_t)ifa->ifa_netmask, 2093 (caddr_t)ifa_maybe->ifa_netmask)) { 2094 ifa_maybe = ifa; 2095 } 2096 } 2097 } 2098 } 2099 ifa = ifa_maybe; 2100 ifa_maybe = NULL; 2101 done: 2102 return (ifa); 2103 } 2104 2105 /* 2106 * Find an interface address specific to an interface best matching 2107 * a given address. 2108 */ 2109 struct ifaddr * 2110 ifaof_ifpforaddr(const struct sockaddr *addr, struct ifnet *ifp) 2111 { 2112 struct ifaddr *ifa; 2113 const char *cp, *cp2, *cp3; 2114 char *cplim; 2115 struct ifaddr *ifa_maybe = NULL; 2116 u_int af = addr->sa_family; 2117 2118 if (af >= AF_MAX) 2119 return (NULL); 2120 2121 MPASS(in_epoch(net_epoch_preempt)); 2122 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 2123 if (ifa->ifa_addr->sa_family != af) 2124 continue; 2125 if (ifa_maybe == NULL) 2126 ifa_maybe = ifa; 2127 if (ifa->ifa_netmask == 0) { 2128 if (sa_equal(addr, ifa->ifa_addr) || 2129 (ifa->ifa_dstaddr && 2130 sa_equal(addr, ifa->ifa_dstaddr))) 2131 goto done; 2132 continue; 2133 } 2134 if (ifp->if_flags & IFF_POINTOPOINT) { 2135 if (sa_equal(addr, ifa->ifa_dstaddr)) 2136 goto done; 2137 } else { 2138 cp = addr->sa_data; 2139 cp2 = ifa->ifa_addr->sa_data; 2140 cp3 = ifa->ifa_netmask->sa_data; 2141 cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask; 2142 for (; cp3 < cplim; cp3++) 2143 if ((*cp++ ^ *cp2++) & *cp3) 2144 break; 2145 if (cp3 == cplim) 2146 goto done; 2147 } 2148 } 2149 ifa = ifa_maybe; 2150 done: 2151 return (ifa); 2152 } 2153 2154 /* 2155 * See whether new ifa is better than current one: 2156 * 1) A non-virtual one is preferred over virtual. 2157 * 2) A virtual in master state preferred over any other state. 2158 * 2159 * Used in several address selecting functions. 2160 */ 2161 int 2162 ifa_preferred(struct ifaddr *cur, struct ifaddr *next) 2163 { 2164 2165 return (cur->ifa_carp && (!next->ifa_carp || 2166 ((*carp_master_p)(next) && !(*carp_master_p)(cur)))); 2167 } 2168 2169 struct sockaddr_dl * 2170 link_alloc_sdl(size_t size, int flags) 2171 { 2172 2173 return (malloc(size, M_TEMP, flags)); 2174 } 2175 2176 void 2177 link_free_sdl(struct sockaddr *sa) 2178 { 2179 free(sa, M_TEMP); 2180 } 2181 2182 /* 2183 * Fills in given sdl with interface basic info. 2184 * Returns pointer to filled sdl. 2185 */ 2186 struct sockaddr_dl * 2187 link_init_sdl(struct ifnet *ifp, struct sockaddr *paddr, u_char iftype) 2188 { 2189 struct sockaddr_dl *sdl; 2190 2191 sdl = (struct sockaddr_dl *)paddr; 2192 memset(sdl, 0, sizeof(struct sockaddr_dl)); 2193 sdl->sdl_len = sizeof(struct sockaddr_dl); 2194 sdl->sdl_family = AF_LINK; 2195 sdl->sdl_index = ifp->if_index; 2196 sdl->sdl_type = iftype; 2197 2198 return (sdl); 2199 } 2200 2201 /* 2202 * Mark an interface down and notify protocols of 2203 * the transition. 2204 */ 2205 static void 2206 if_unroute(struct ifnet *ifp, int flag, int fam) 2207 { 2208 struct ifaddr *ifa; 2209 2210 KASSERT(flag == IFF_UP, ("if_unroute: flag != IFF_UP")); 2211 2212 ifp->if_flags &= ~flag; 2213 getmicrotime(&ifp->if_lastchange); 2214 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) 2215 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family)) 2216 pfctlinput(PRC_IFDOWN, ifa->ifa_addr); 2217 ifp->if_qflush(ifp); 2218 2219 if (ifp->if_carp) 2220 (*carp_linkstate_p)(ifp); 2221 rt_ifmsg(ifp); 2222 } 2223 2224 /* 2225 * Mark an interface up and notify protocols of 2226 * the transition. 2227 */ 2228 static void 2229 if_route(struct ifnet *ifp, int flag, int fam) 2230 { 2231 struct ifaddr *ifa; 2232 2233 KASSERT(flag == IFF_UP, ("if_route: flag != IFF_UP")); 2234 2235 ifp->if_flags |= flag; 2236 getmicrotime(&ifp->if_lastchange); 2237 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) 2238 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family)) 2239 pfctlinput(PRC_IFUP, ifa->ifa_addr); 2240 if (ifp->if_carp) 2241 (*carp_linkstate_p)(ifp); 2242 rt_ifmsg(ifp); 2243 #ifdef INET6 2244 in6_if_up(ifp); 2245 #endif 2246 } 2247 2248 void (*vlan_link_state_p)(struct ifnet *); /* XXX: private from if_vlan */ 2249 void (*vlan_trunk_cap_p)(struct ifnet *); /* XXX: private from if_vlan */ 2250 struct ifnet *(*vlan_trunkdev_p)(struct ifnet *); 2251 struct ifnet *(*vlan_devat_p)(struct ifnet *, uint16_t); 2252 int (*vlan_tag_p)(struct ifnet *, uint16_t *); 2253 int (*vlan_pcp_p)(struct ifnet *, uint16_t *); 2254 int (*vlan_setcookie_p)(struct ifnet *, void *); 2255 void *(*vlan_cookie_p)(struct ifnet *); 2256 2257 /* 2258 * Handle a change in the interface link state. To avoid LORs 2259 * between driver lock and upper layer locks, as well as possible 2260 * recursions, we post event to taskqueue, and all job 2261 * is done in static do_link_state_change(). 2262 */ 2263 void 2264 if_link_state_change(struct ifnet *ifp, int link_state) 2265 { 2266 /* Return if state hasn't changed. */ 2267 if (ifp->if_link_state == link_state) 2268 return; 2269 2270 ifp->if_link_state = link_state; 2271 2272 /* XXXGL: reference ifp? */ 2273 taskqueue_enqueue(taskqueue_swi, &ifp->if_linktask); 2274 } 2275 2276 static void 2277 do_link_state_change(void *arg, int pending) 2278 { 2279 struct ifnet *ifp; 2280 int link_state; 2281 2282 ifp = arg; 2283 link_state = ifp->if_link_state; 2284 2285 CURVNET_SET(ifp->if_vnet); 2286 rt_ifmsg(ifp); 2287 if (ifp->if_vlantrunk != NULL) 2288 (*vlan_link_state_p)(ifp); 2289 2290 if ((ifp->if_type == IFT_ETHER || ifp->if_type == IFT_L2VLAN) && 2291 ifp->if_l2com != NULL) 2292 (*ng_ether_link_state_p)(ifp, link_state); 2293 if (ifp->if_carp) 2294 (*carp_linkstate_p)(ifp); 2295 if (ifp->if_bridge) 2296 ifp->if_bridge_linkstate(ifp); 2297 if (ifp->if_lagg) 2298 (*lagg_linkstate_p)(ifp, link_state); 2299 2300 if (IS_DEFAULT_VNET(curvnet)) 2301 devctl_notify("IFNET", ifp->if_xname, 2302 (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", 2303 NULL); 2304 if (pending > 1) 2305 if_printf(ifp, "%d link states coalesced\n", pending); 2306 if (log_link_state_change) 2307 if_printf(ifp, "link state changed to %s\n", 2308 (link_state == LINK_STATE_UP) ? "UP" : "DOWN" ); 2309 EVENTHANDLER_INVOKE(ifnet_link_event, ifp, link_state); 2310 CURVNET_RESTORE(); 2311 } 2312 2313 /* 2314 * Mark an interface down and notify protocols of 2315 * the transition. 2316 */ 2317 void 2318 if_down(struct ifnet *ifp) 2319 { 2320 2321 EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_DOWN); 2322 if_unroute(ifp, IFF_UP, AF_UNSPEC); 2323 } 2324 2325 /* 2326 * Mark an interface up and notify protocols of 2327 * the transition. 2328 */ 2329 void 2330 if_up(struct ifnet *ifp) 2331 { 2332 2333 if_route(ifp, IFF_UP, AF_UNSPEC); 2334 EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_UP); 2335 } 2336 2337 /* 2338 * Flush an interface queue. 2339 */ 2340 void 2341 if_qflush(struct ifnet *ifp) 2342 { 2343 struct mbuf *m, *n; 2344 struct ifaltq *ifq; 2345 2346 ifq = &ifp->if_snd; 2347 IFQ_LOCK(ifq); 2348 #ifdef ALTQ 2349 if (ALTQ_IS_ENABLED(ifq)) 2350 ALTQ_PURGE(ifq); 2351 #endif 2352 n = ifq->ifq_head; 2353 while ((m = n) != NULL) { 2354 n = m->m_nextpkt; 2355 m_freem(m); 2356 } 2357 ifq->ifq_head = 0; 2358 ifq->ifq_tail = 0; 2359 ifq->ifq_len = 0; 2360 IFQ_UNLOCK(ifq); 2361 } 2362 2363 /* 2364 * Map interface name to interface structure pointer, with or without 2365 * returning a reference. 2366 */ 2367 struct ifnet * 2368 ifunit_ref(const char *name) 2369 { 2370 struct epoch_tracker et; 2371 struct ifnet *ifp; 2372 2373 NET_EPOCH_ENTER(et); 2374 CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) { 2375 if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0 && 2376 !(ifp->if_flags & IFF_DYING)) 2377 break; 2378 } 2379 if (ifp != NULL) 2380 if_ref(ifp); 2381 NET_EPOCH_EXIT(et); 2382 return (ifp); 2383 } 2384 2385 struct ifnet * 2386 ifunit(const char *name) 2387 { 2388 struct epoch_tracker et; 2389 struct ifnet *ifp; 2390 2391 NET_EPOCH_ENTER(et); 2392 CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) { 2393 if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0) 2394 break; 2395 } 2396 NET_EPOCH_EXIT(et); 2397 return (ifp); 2398 } 2399 2400 static void * 2401 ifr_buffer_get_buffer(void *data) 2402 { 2403 union ifreq_union *ifrup; 2404 2405 ifrup = data; 2406 #ifdef COMPAT_FREEBSD32 2407 if (SV_CURPROC_FLAG(SV_ILP32)) 2408 return ((void *)(uintptr_t) 2409 ifrup->ifr32.ifr_ifru.ifru_buffer.buffer); 2410 #endif 2411 return (ifrup->ifr.ifr_ifru.ifru_buffer.buffer); 2412 } 2413 2414 static void 2415 ifr_buffer_set_buffer_null(void *data) 2416 { 2417 union ifreq_union *ifrup; 2418 2419 ifrup = data; 2420 #ifdef COMPAT_FREEBSD32 2421 if (SV_CURPROC_FLAG(SV_ILP32)) 2422 ifrup->ifr32.ifr_ifru.ifru_buffer.buffer = 0; 2423 else 2424 #endif 2425 ifrup->ifr.ifr_ifru.ifru_buffer.buffer = NULL; 2426 } 2427 2428 static size_t 2429 ifr_buffer_get_length(void *data) 2430 { 2431 union ifreq_union *ifrup; 2432 2433 ifrup = data; 2434 #ifdef COMPAT_FREEBSD32 2435 if (SV_CURPROC_FLAG(SV_ILP32)) 2436 return (ifrup->ifr32.ifr_ifru.ifru_buffer.length); 2437 #endif 2438 return (ifrup->ifr.ifr_ifru.ifru_buffer.length); 2439 } 2440 2441 static void 2442 ifr_buffer_set_length(void *data, size_t len) 2443 { 2444 union ifreq_union *ifrup; 2445 2446 ifrup = data; 2447 #ifdef COMPAT_FREEBSD32 2448 if (SV_CURPROC_FLAG(SV_ILP32)) 2449 ifrup->ifr32.ifr_ifru.ifru_buffer.length = len; 2450 else 2451 #endif 2452 ifrup->ifr.ifr_ifru.ifru_buffer.length = len; 2453 } 2454 2455 void * 2456 ifr_data_get_ptr(void *ifrp) 2457 { 2458 union ifreq_union *ifrup; 2459 2460 ifrup = ifrp; 2461 #ifdef COMPAT_FREEBSD32 2462 if (SV_CURPROC_FLAG(SV_ILP32)) 2463 return ((void *)(uintptr_t) 2464 ifrup->ifr32.ifr_ifru.ifru_data); 2465 #endif 2466 return (ifrup->ifr.ifr_ifru.ifru_data); 2467 } 2468 2469 /* 2470 * Hardware specific interface ioctls. 2471 */ 2472 int 2473 ifhwioctl(u_long cmd, struct ifnet *ifp, caddr_t data, struct thread *td) 2474 { 2475 struct ifreq *ifr; 2476 int error = 0, do_ifup = 0; 2477 int new_flags, temp_flags; 2478 size_t namelen, onamelen; 2479 size_t descrlen; 2480 char *descrbuf, *odescrbuf; 2481 char new_name[IFNAMSIZ]; 2482 struct ifaddr *ifa; 2483 struct sockaddr_dl *sdl; 2484 2485 ifr = (struct ifreq *)data; 2486 switch (cmd) { 2487 case SIOCGIFINDEX: 2488 ifr->ifr_index = ifp->if_index; 2489 break; 2490 2491 case SIOCGIFFLAGS: 2492 temp_flags = ifp->if_flags | ifp->if_drv_flags; 2493 ifr->ifr_flags = temp_flags & 0xffff; 2494 ifr->ifr_flagshigh = temp_flags >> 16; 2495 break; 2496 2497 case SIOCGIFCAP: 2498 ifr->ifr_reqcap = ifp->if_capabilities; 2499 ifr->ifr_curcap = ifp->if_capenable; 2500 break; 2501 2502 #ifdef MAC 2503 case SIOCGIFMAC: 2504 error = mac_ifnet_ioctl_get(td->td_ucred, ifr, ifp); 2505 break; 2506 #endif 2507 2508 case SIOCGIFMETRIC: 2509 ifr->ifr_metric = ifp->if_metric; 2510 break; 2511 2512 case SIOCGIFMTU: 2513 ifr->ifr_mtu = ifp->if_mtu; 2514 break; 2515 2516 case SIOCGIFPHYS: 2517 /* XXXGL: did this ever worked? */ 2518 ifr->ifr_phys = 0; 2519 break; 2520 2521 case SIOCGIFDESCR: 2522 error = 0; 2523 sx_slock(&ifdescr_sx); 2524 if (ifp->if_description == NULL) 2525 error = ENOMSG; 2526 else { 2527 /* space for terminating nul */ 2528 descrlen = strlen(ifp->if_description) + 1; 2529 if (ifr_buffer_get_length(ifr) < descrlen) 2530 ifr_buffer_set_buffer_null(ifr); 2531 else 2532 error = copyout(ifp->if_description, 2533 ifr_buffer_get_buffer(ifr), descrlen); 2534 ifr_buffer_set_length(ifr, descrlen); 2535 } 2536 sx_sunlock(&ifdescr_sx); 2537 break; 2538 2539 case SIOCSIFDESCR: 2540 error = priv_check(td, PRIV_NET_SETIFDESCR); 2541 if (error) 2542 return (error); 2543 2544 /* 2545 * Copy only (length-1) bytes to make sure that 2546 * if_description is always nul terminated. The 2547 * length parameter is supposed to count the 2548 * terminating nul in. 2549 */ 2550 if (ifr_buffer_get_length(ifr) > ifdescr_maxlen) 2551 return (ENAMETOOLONG); 2552 else if (ifr_buffer_get_length(ifr) == 0) 2553 descrbuf = NULL; 2554 else { 2555 descrbuf = malloc(ifr_buffer_get_length(ifr), 2556 M_IFDESCR, M_WAITOK | M_ZERO); 2557 error = copyin(ifr_buffer_get_buffer(ifr), descrbuf, 2558 ifr_buffer_get_length(ifr) - 1); 2559 if (error) { 2560 free(descrbuf, M_IFDESCR); 2561 break; 2562 } 2563 } 2564 2565 sx_xlock(&ifdescr_sx); 2566 odescrbuf = ifp->if_description; 2567 ifp->if_description = descrbuf; 2568 sx_xunlock(&ifdescr_sx); 2569 2570 getmicrotime(&ifp->if_lastchange); 2571 free(odescrbuf, M_IFDESCR); 2572 break; 2573 2574 case SIOCGIFFIB: 2575 ifr->ifr_fib = ifp->if_fib; 2576 break; 2577 2578 case SIOCSIFFIB: 2579 error = priv_check(td, PRIV_NET_SETIFFIB); 2580 if (error) 2581 return (error); 2582 if (ifr->ifr_fib >= rt_numfibs) 2583 return (EINVAL); 2584 2585 ifp->if_fib = ifr->ifr_fib; 2586 break; 2587 2588 case SIOCSIFFLAGS: 2589 error = priv_check(td, PRIV_NET_SETIFFLAGS); 2590 if (error) 2591 return (error); 2592 /* 2593 * Currently, no driver owned flags pass the IFF_CANTCHANGE 2594 * check, so we don't need special handling here yet. 2595 */ 2596 new_flags = (ifr->ifr_flags & 0xffff) | 2597 (ifr->ifr_flagshigh << 16); 2598 if (ifp->if_flags & IFF_UP && 2599 (new_flags & IFF_UP) == 0) { 2600 if_down(ifp); 2601 } else if (new_flags & IFF_UP && 2602 (ifp->if_flags & IFF_UP) == 0) { 2603 do_ifup = 1; 2604 } 2605 /* See if permanently promiscuous mode bit is about to flip */ 2606 if ((ifp->if_flags ^ new_flags) & IFF_PPROMISC) { 2607 if (new_flags & IFF_PPROMISC) 2608 ifp->if_flags |= IFF_PROMISC; 2609 else if (ifp->if_pcount == 0) 2610 ifp->if_flags &= ~IFF_PROMISC; 2611 if (log_promisc_mode_change) 2612 if_printf(ifp, "permanently promiscuous mode %s\n", 2613 ((new_flags & IFF_PPROMISC) ? 2614 "enabled" : "disabled")); 2615 } 2616 ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) | 2617 (new_flags &~ IFF_CANTCHANGE); 2618 if (ifp->if_ioctl) { 2619 (void) (*ifp->if_ioctl)(ifp, cmd, data); 2620 } 2621 if (do_ifup) 2622 if_up(ifp); 2623 getmicrotime(&ifp->if_lastchange); 2624 break; 2625 2626 case SIOCSIFCAP: 2627 error = priv_check(td, PRIV_NET_SETIFCAP); 2628 if (error) 2629 return (error); 2630 if (ifp->if_ioctl == NULL) 2631 return (EOPNOTSUPP); 2632 if (ifr->ifr_reqcap & ~ifp->if_capabilities) 2633 return (EINVAL); 2634 error = (*ifp->if_ioctl)(ifp, cmd, data); 2635 if (error == 0) 2636 getmicrotime(&ifp->if_lastchange); 2637 break; 2638 2639 #ifdef MAC 2640 case SIOCSIFMAC: 2641 error = mac_ifnet_ioctl_set(td->td_ucred, ifr, ifp); 2642 break; 2643 #endif 2644 2645 case SIOCSIFNAME: 2646 error = priv_check(td, PRIV_NET_SETIFNAME); 2647 if (error) 2648 return (error); 2649 error = copyinstr(ifr_data_get_ptr(ifr), new_name, IFNAMSIZ, 2650 NULL); 2651 if (error != 0) 2652 return (error); 2653 if (new_name[0] == '\0') 2654 return (EINVAL); 2655 if (new_name[IFNAMSIZ-1] != '\0') { 2656 new_name[IFNAMSIZ-1] = '\0'; 2657 if (strlen(new_name) == IFNAMSIZ-1) 2658 return (EINVAL); 2659 } 2660 if (strcmp(new_name, ifp->if_xname) == 0) 2661 break; 2662 if (ifunit(new_name) != NULL) 2663 return (EEXIST); 2664 2665 /* 2666 * XXX: Locking. Nothing else seems to lock if_flags, 2667 * and there are numerous other races with the 2668 * ifunit() checks not being atomic with namespace 2669 * changes (renames, vmoves, if_attach, etc). 2670 */ 2671 ifp->if_flags |= IFF_RENAMING; 2672 2673 /* Announce the departure of the interface. */ 2674 rt_ifannouncemsg(ifp, IFAN_DEPARTURE); 2675 EVENTHANDLER_INVOKE(ifnet_departure_event, ifp); 2676 2677 if_printf(ifp, "changing name to '%s'\n", new_name); 2678 2679 IF_ADDR_WLOCK(ifp); 2680 strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname)); 2681 ifa = ifp->if_addr; 2682 sdl = (struct sockaddr_dl *)ifa->ifa_addr; 2683 namelen = strlen(new_name); 2684 onamelen = sdl->sdl_nlen; 2685 /* 2686 * Move the address if needed. This is safe because we 2687 * allocate space for a name of length IFNAMSIZ when we 2688 * create this in if_attach(). 2689 */ 2690 if (namelen != onamelen) { 2691 bcopy(sdl->sdl_data + onamelen, 2692 sdl->sdl_data + namelen, sdl->sdl_alen); 2693 } 2694 bcopy(new_name, sdl->sdl_data, namelen); 2695 sdl->sdl_nlen = namelen; 2696 sdl = (struct sockaddr_dl *)ifa->ifa_netmask; 2697 bzero(sdl->sdl_data, onamelen); 2698 while (namelen != 0) 2699 sdl->sdl_data[--namelen] = 0xff; 2700 IF_ADDR_WUNLOCK(ifp); 2701 2702 EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp); 2703 /* Announce the return of the interface. */ 2704 rt_ifannouncemsg(ifp, IFAN_ARRIVAL); 2705 2706 ifp->if_flags &= ~IFF_RENAMING; 2707 break; 2708 2709 #ifdef VIMAGE 2710 case SIOCSIFVNET: 2711 error = priv_check(td, PRIV_NET_SETIFVNET); 2712 if (error) 2713 return (error); 2714 error = if_vmove_loan(td, ifp, ifr->ifr_name, ifr->ifr_jid); 2715 break; 2716 #endif 2717 2718 case SIOCSIFMETRIC: 2719 error = priv_check(td, PRIV_NET_SETIFMETRIC); 2720 if (error) 2721 return (error); 2722 ifp->if_metric = ifr->ifr_metric; 2723 getmicrotime(&ifp->if_lastchange); 2724 break; 2725 2726 case SIOCSIFPHYS: 2727 error = priv_check(td, PRIV_NET_SETIFPHYS); 2728 if (error) 2729 return (error); 2730 if (ifp->if_ioctl == NULL) 2731 return (EOPNOTSUPP); 2732 error = (*ifp->if_ioctl)(ifp, cmd, data); 2733 if (error == 0) 2734 getmicrotime(&ifp->if_lastchange); 2735 break; 2736 2737 case SIOCSIFMTU: 2738 { 2739 u_long oldmtu = ifp->if_mtu; 2740 2741 error = priv_check(td, PRIV_NET_SETIFMTU); 2742 if (error) 2743 return (error); 2744 if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU) 2745 return (EINVAL); 2746 if (ifp->if_ioctl == NULL) 2747 return (EOPNOTSUPP); 2748 error = (*ifp->if_ioctl)(ifp, cmd, data); 2749 if (error == 0) { 2750 getmicrotime(&ifp->if_lastchange); 2751 rt_ifmsg(ifp); 2752 #ifdef INET 2753 DEBUGNET_NOTIFY_MTU(ifp); 2754 #endif 2755 } 2756 /* 2757 * If the link MTU changed, do network layer specific procedure. 2758 */ 2759 if (ifp->if_mtu != oldmtu) { 2760 #ifdef INET6 2761 nd6_setmtu(ifp); 2762 #endif 2763 rt_updatemtu(ifp); 2764 } 2765 break; 2766 } 2767 2768 case SIOCADDMULTI: 2769 case SIOCDELMULTI: 2770 if (cmd == SIOCADDMULTI) 2771 error = priv_check(td, PRIV_NET_ADDMULTI); 2772 else 2773 error = priv_check(td, PRIV_NET_DELMULTI); 2774 if (error) 2775 return (error); 2776 2777 /* Don't allow group membership on non-multicast interfaces. */ 2778 if ((ifp->if_flags & IFF_MULTICAST) == 0) 2779 return (EOPNOTSUPP); 2780 2781 /* Don't let users screw up protocols' entries. */ 2782 if (ifr->ifr_addr.sa_family != AF_LINK) 2783 return (EINVAL); 2784 2785 if (cmd == SIOCADDMULTI) { 2786 struct epoch_tracker et; 2787 struct ifmultiaddr *ifma; 2788 2789 /* 2790 * Userland is only permitted to join groups once 2791 * via the if_addmulti() KPI, because it cannot hold 2792 * struct ifmultiaddr * between calls. It may also 2793 * lose a race while we check if the membership 2794 * already exists. 2795 */ 2796 NET_EPOCH_ENTER(et); 2797 ifma = if_findmulti(ifp, &ifr->ifr_addr); 2798 NET_EPOCH_EXIT(et); 2799 if (ifma != NULL) 2800 error = EADDRINUSE; 2801 else 2802 error = if_addmulti(ifp, &ifr->ifr_addr, &ifma); 2803 } else { 2804 error = if_delmulti(ifp, &ifr->ifr_addr); 2805 } 2806 if (error == 0) 2807 getmicrotime(&ifp->if_lastchange); 2808 break; 2809 2810 case SIOCSIFPHYADDR: 2811 case SIOCDIFPHYADDR: 2812 #ifdef INET6 2813 case SIOCSIFPHYADDR_IN6: 2814 #endif 2815 case SIOCSIFMEDIA: 2816 case SIOCSIFGENERIC: 2817 error = priv_check(td, PRIV_NET_HWIOCTL); 2818 if (error) 2819 return (error); 2820 if (ifp->if_ioctl == NULL) 2821 return (EOPNOTSUPP); 2822 error = (*ifp->if_ioctl)(ifp, cmd, data); 2823 if (error == 0) 2824 getmicrotime(&ifp->if_lastchange); 2825 break; 2826 2827 case SIOCGIFSTATUS: 2828 case SIOCGIFPSRCADDR: 2829 case SIOCGIFPDSTADDR: 2830 case SIOCGIFMEDIA: 2831 case SIOCGIFXMEDIA: 2832 case SIOCGIFGENERIC: 2833 case SIOCGIFRSSKEY: 2834 case SIOCGIFRSSHASH: 2835 case SIOCGIFDOWNREASON: 2836 if (ifp->if_ioctl == NULL) 2837 return (EOPNOTSUPP); 2838 error = (*ifp->if_ioctl)(ifp, cmd, data); 2839 break; 2840 2841 case SIOCSIFLLADDR: 2842 error = priv_check(td, PRIV_NET_SETLLADDR); 2843 if (error) 2844 return (error); 2845 error = if_setlladdr(ifp, 2846 ifr->ifr_addr.sa_data, ifr->ifr_addr.sa_len); 2847 break; 2848 2849 case SIOCGHWADDR: 2850 error = if_gethwaddr(ifp, ifr); 2851 break; 2852 2853 case CASE_IOC_IFGROUPREQ(SIOCAIFGROUP): 2854 error = priv_check(td, PRIV_NET_ADDIFGROUP); 2855 if (error) 2856 return (error); 2857 if ((error = if_addgroup(ifp, 2858 ifgr_group_get((struct ifgroupreq *)data)))) 2859 return (error); 2860 break; 2861 2862 case CASE_IOC_IFGROUPREQ(SIOCGIFGROUP): 2863 { 2864 struct epoch_tracker et; 2865 2866 NET_EPOCH_ENTER(et); 2867 error = if_getgroup((struct ifgroupreq *)data, ifp); 2868 NET_EPOCH_EXIT(et); 2869 break; 2870 } 2871 2872 case CASE_IOC_IFGROUPREQ(SIOCDIFGROUP): 2873 error = priv_check(td, PRIV_NET_DELIFGROUP); 2874 if (error) 2875 return (error); 2876 if ((error = if_delgroup(ifp, 2877 ifgr_group_get((struct ifgroupreq *)data)))) 2878 return (error); 2879 break; 2880 2881 default: 2882 error = ENOIOCTL; 2883 break; 2884 } 2885 return (error); 2886 } 2887 2888 #ifdef COMPAT_FREEBSD32 2889 struct ifconf32 { 2890 int32_t ifc_len; 2891 union { 2892 uint32_t ifcu_buf; 2893 uint32_t ifcu_req; 2894 } ifc_ifcu; 2895 }; 2896 #define SIOCGIFCONF32 _IOWR('i', 36, struct ifconf32) 2897 #endif 2898 2899 #ifdef COMPAT_FREEBSD32 2900 static void 2901 ifmr_init(struct ifmediareq *ifmr, caddr_t data) 2902 { 2903 struct ifmediareq32 *ifmr32; 2904 2905 ifmr32 = (struct ifmediareq32 *)data; 2906 memcpy(ifmr->ifm_name, ifmr32->ifm_name, 2907 sizeof(ifmr->ifm_name)); 2908 ifmr->ifm_current = ifmr32->ifm_current; 2909 ifmr->ifm_mask = ifmr32->ifm_mask; 2910 ifmr->ifm_status = ifmr32->ifm_status; 2911 ifmr->ifm_active = ifmr32->ifm_active; 2912 ifmr->ifm_count = ifmr32->ifm_count; 2913 ifmr->ifm_ulist = (int *)(uintptr_t)ifmr32->ifm_ulist; 2914 } 2915 2916 static void 2917 ifmr_update(const struct ifmediareq *ifmr, caddr_t data) 2918 { 2919 struct ifmediareq32 *ifmr32; 2920 2921 ifmr32 = (struct ifmediareq32 *)data; 2922 ifmr32->ifm_current = ifmr->ifm_current; 2923 ifmr32->ifm_mask = ifmr->ifm_mask; 2924 ifmr32->ifm_status = ifmr->ifm_status; 2925 ifmr32->ifm_active = ifmr->ifm_active; 2926 ifmr32->ifm_count = ifmr->ifm_count; 2927 } 2928 #endif 2929 2930 /* 2931 * Interface ioctls. 2932 */ 2933 int 2934 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct thread *td) 2935 { 2936 #ifdef COMPAT_FREEBSD32 2937 caddr_t saved_data = NULL; 2938 struct ifmediareq ifmr; 2939 struct ifmediareq *ifmrp = NULL; 2940 #endif 2941 struct ifnet *ifp; 2942 struct ifreq *ifr; 2943 int error; 2944 int oif_flags; 2945 2946 CURVNET_SET(so->so_vnet); 2947 #ifdef VIMAGE 2948 /* Make sure the VNET is stable. */ 2949 if (so->so_vnet->vnet_shutdown) { 2950 CURVNET_RESTORE(); 2951 return (EBUSY); 2952 } 2953 #endif 2954 2955 switch (cmd) { 2956 case SIOCGIFCONF: 2957 error = ifconf(cmd, data); 2958 goto out_noref; 2959 2960 #ifdef COMPAT_FREEBSD32 2961 case SIOCGIFCONF32: 2962 { 2963 struct ifconf32 *ifc32; 2964 struct ifconf ifc; 2965 2966 ifc32 = (struct ifconf32 *)data; 2967 ifc.ifc_len = ifc32->ifc_len; 2968 ifc.ifc_buf = PTRIN(ifc32->ifc_buf); 2969 2970 error = ifconf(SIOCGIFCONF, (void *)&ifc); 2971 if (error == 0) 2972 ifc32->ifc_len = ifc.ifc_len; 2973 goto out_noref; 2974 } 2975 #endif 2976 } 2977 2978 #ifdef COMPAT_FREEBSD32 2979 switch (cmd) { 2980 case SIOCGIFMEDIA32: 2981 case SIOCGIFXMEDIA32: 2982 ifmrp = &ifmr; 2983 ifmr_init(ifmrp, data); 2984 cmd = _IOC_NEWTYPE(cmd, struct ifmediareq); 2985 saved_data = data; 2986 data = (caddr_t)ifmrp; 2987 } 2988 #endif 2989 2990 ifr = (struct ifreq *)data; 2991 switch (cmd) { 2992 #ifdef VIMAGE 2993 case SIOCSIFRVNET: 2994 error = priv_check(td, PRIV_NET_SETIFVNET); 2995 if (error == 0) 2996 error = if_vmove_reclaim(td, ifr->ifr_name, 2997 ifr->ifr_jid); 2998 goto out_noref; 2999 #endif 3000 case SIOCIFCREATE: 3001 case SIOCIFCREATE2: 3002 error = priv_check(td, PRIV_NET_IFCREATE); 3003 if (error == 0) 3004 error = if_clone_create(ifr->ifr_name, 3005 sizeof(ifr->ifr_name), cmd == SIOCIFCREATE2 ? 3006 ifr_data_get_ptr(ifr) : NULL); 3007 goto out_noref; 3008 case SIOCIFDESTROY: 3009 error = priv_check(td, PRIV_NET_IFDESTROY); 3010 if (error == 0) 3011 error = if_clone_destroy(ifr->ifr_name); 3012 goto out_noref; 3013 3014 case SIOCIFGCLONERS: 3015 error = if_clone_list((struct if_clonereq *)data); 3016 goto out_noref; 3017 3018 case CASE_IOC_IFGROUPREQ(SIOCGIFGMEMB): 3019 error = if_getgroupmembers((struct ifgroupreq *)data); 3020 goto out_noref; 3021 3022 #if defined(INET) || defined(INET6) 3023 case SIOCSVH: 3024 case SIOCGVH: 3025 if (carp_ioctl_p == NULL) 3026 error = EPROTONOSUPPORT; 3027 else 3028 error = (*carp_ioctl_p)(ifr, cmd, td); 3029 goto out_noref; 3030 #endif 3031 } 3032 3033 ifp = ifunit_ref(ifr->ifr_name); 3034 if (ifp == NULL) { 3035 error = ENXIO; 3036 goto out_noref; 3037 } 3038 3039 error = ifhwioctl(cmd, ifp, data, td); 3040 if (error != ENOIOCTL) 3041 goto out_ref; 3042 3043 oif_flags = ifp->if_flags; 3044 if (so->so_proto == NULL) { 3045 error = EOPNOTSUPP; 3046 goto out_ref; 3047 } 3048 3049 /* 3050 * Pass the request on to the socket control method, and if the 3051 * latter returns EOPNOTSUPP, directly to the interface. 3052 * 3053 * Make an exception for the legacy SIOCSIF* requests. Drivers 3054 * trust SIOCSIFADDR et al to come from an already privileged 3055 * layer, and do not perform any credentials checks or input 3056 * validation. 3057 */ 3058 error = ((*so->so_proto->pr_usrreqs->pru_control)(so, cmd, data, 3059 ifp, td)); 3060 if (error == EOPNOTSUPP && ifp != NULL && ifp->if_ioctl != NULL && 3061 cmd != SIOCSIFADDR && cmd != SIOCSIFBRDADDR && 3062 cmd != SIOCSIFDSTADDR && cmd != SIOCSIFNETMASK) 3063 error = (*ifp->if_ioctl)(ifp, cmd, data); 3064 3065 if ((oif_flags ^ ifp->if_flags) & IFF_UP) { 3066 #ifdef INET6 3067 if (ifp->if_flags & IFF_UP) 3068 in6_if_up(ifp); 3069 #endif 3070 } 3071 3072 out_ref: 3073 if_rele(ifp); 3074 out_noref: 3075 #ifdef COMPAT_FREEBSD32 3076 if (ifmrp != NULL) { 3077 KASSERT((cmd == SIOCGIFMEDIA || cmd == SIOCGIFXMEDIA), 3078 ("ifmrp non-NULL, but cmd is not an ifmedia req 0x%lx", 3079 cmd)); 3080 data = saved_data; 3081 ifmr_update(ifmrp, data); 3082 } 3083 #endif 3084 CURVNET_RESTORE(); 3085 return (error); 3086 } 3087 3088 /* 3089 * The code common to handling reference counted flags, 3090 * e.g., in ifpromisc() and if_allmulti(). 3091 * The "pflag" argument can specify a permanent mode flag to check, 3092 * such as IFF_PPROMISC for promiscuous mode; should be 0 if none. 3093 * 3094 * Only to be used on stack-owned flags, not driver-owned flags. 3095 */ 3096 static int 3097 if_setflag(struct ifnet *ifp, int flag, int pflag, int *refcount, int onswitch) 3098 { 3099 struct ifreq ifr; 3100 int error; 3101 int oldflags, oldcount; 3102 3103 /* Sanity checks to catch programming errors */ 3104 KASSERT((flag & (IFF_DRV_OACTIVE|IFF_DRV_RUNNING)) == 0, 3105 ("%s: setting driver-owned flag %d", __func__, flag)); 3106 3107 if (onswitch) 3108 KASSERT(*refcount >= 0, 3109 ("%s: increment negative refcount %d for flag %d", 3110 __func__, *refcount, flag)); 3111 else 3112 KASSERT(*refcount > 0, 3113 ("%s: decrement non-positive refcount %d for flag %d", 3114 __func__, *refcount, flag)); 3115 3116 /* In case this mode is permanent, just touch refcount */ 3117 if (ifp->if_flags & pflag) { 3118 *refcount += onswitch ? 1 : -1; 3119 return (0); 3120 } 3121 3122 /* Save ifnet parameters for if_ioctl() may fail */ 3123 oldcount = *refcount; 3124 oldflags = ifp->if_flags; 3125 3126 /* 3127 * See if we aren't the only and touching refcount is enough. 3128 * Actually toggle interface flag if we are the first or last. 3129 */ 3130 if (onswitch) { 3131 if ((*refcount)++) 3132 return (0); 3133 ifp->if_flags |= flag; 3134 } else { 3135 if (--(*refcount)) 3136 return (0); 3137 ifp->if_flags &= ~flag; 3138 } 3139 3140 /* Call down the driver since we've changed interface flags */ 3141 if (ifp->if_ioctl == NULL) { 3142 error = EOPNOTSUPP; 3143 goto recover; 3144 } 3145 ifr.ifr_flags = ifp->if_flags & 0xffff; 3146 ifr.ifr_flagshigh = ifp->if_flags >> 16; 3147 error = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr); 3148 if (error) 3149 goto recover; 3150 /* Notify userland that interface flags have changed */ 3151 rt_ifmsg(ifp); 3152 return (0); 3153 3154 recover: 3155 /* Recover after driver error */ 3156 *refcount = oldcount; 3157 ifp->if_flags = oldflags; 3158 return (error); 3159 } 3160 3161 /* 3162 * Set/clear promiscuous mode on interface ifp based on the truth value 3163 * of pswitch. The calls are reference counted so that only the first 3164 * "on" request actually has an effect, as does the final "off" request. 3165 * Results are undefined if the "off" and "on" requests are not matched. 3166 */ 3167 int 3168 ifpromisc(struct ifnet *ifp, int pswitch) 3169 { 3170 int error; 3171 int oldflags = ifp->if_flags; 3172 3173 error = if_setflag(ifp, IFF_PROMISC, IFF_PPROMISC, 3174 &ifp->if_pcount, pswitch); 3175 /* If promiscuous mode status has changed, log a message */ 3176 if (error == 0 && ((ifp->if_flags ^ oldflags) & IFF_PROMISC) && 3177 log_promisc_mode_change) 3178 if_printf(ifp, "promiscuous mode %s\n", 3179 (ifp->if_flags & IFF_PROMISC) ? "enabled" : "disabled"); 3180 return (error); 3181 } 3182 3183 /* 3184 * Return interface configuration 3185 * of system. List may be used 3186 * in later ioctl's (above) to get 3187 * other information. 3188 */ 3189 /*ARGSUSED*/ 3190 static int 3191 ifconf(u_long cmd, caddr_t data) 3192 { 3193 struct ifconf *ifc = (struct ifconf *)data; 3194 struct ifnet *ifp; 3195 struct ifaddr *ifa; 3196 struct ifreq ifr; 3197 struct sbuf *sb; 3198 int error, full = 0, valid_len, max_len; 3199 3200 /* Limit initial buffer size to MAXPHYS to avoid DoS from userspace. */ 3201 max_len = MAXPHYS - 1; 3202 3203 /* Prevent hostile input from being able to crash the system */ 3204 if (ifc->ifc_len <= 0) 3205 return (EINVAL); 3206 3207 again: 3208 if (ifc->ifc_len <= max_len) { 3209 max_len = ifc->ifc_len; 3210 full = 1; 3211 } 3212 sb = sbuf_new(NULL, NULL, max_len + 1, SBUF_FIXEDLEN); 3213 max_len = 0; 3214 valid_len = 0; 3215 3216 IFNET_RLOCK(); 3217 CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) { 3218 struct epoch_tracker et; 3219 int addrs; 3220 3221 /* 3222 * Zero the ifr to make sure we don't disclose the contents 3223 * of the stack. 3224 */ 3225 memset(&ifr, 0, sizeof(ifr)); 3226 3227 if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name)) 3228 >= sizeof(ifr.ifr_name)) { 3229 sbuf_delete(sb); 3230 IFNET_RUNLOCK(); 3231 return (ENAMETOOLONG); 3232 } 3233 3234 addrs = 0; 3235 NET_EPOCH_ENTER(et); 3236 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 3237 struct sockaddr *sa = ifa->ifa_addr; 3238 3239 if (prison_if(curthread->td_ucred, sa) != 0) 3240 continue; 3241 addrs++; 3242 if (sa->sa_len <= sizeof(*sa)) { 3243 if (sa->sa_len < sizeof(*sa)) { 3244 memset(&ifr.ifr_ifru.ifru_addr, 0, 3245 sizeof(ifr.ifr_ifru.ifru_addr)); 3246 memcpy(&ifr.ifr_ifru.ifru_addr, sa, 3247 sa->sa_len); 3248 } else 3249 ifr.ifr_ifru.ifru_addr = *sa; 3250 sbuf_bcat(sb, &ifr, sizeof(ifr)); 3251 max_len += sizeof(ifr); 3252 } else { 3253 sbuf_bcat(sb, &ifr, 3254 offsetof(struct ifreq, ifr_addr)); 3255 max_len += offsetof(struct ifreq, ifr_addr); 3256 sbuf_bcat(sb, sa, sa->sa_len); 3257 max_len += sa->sa_len; 3258 } 3259 3260 if (sbuf_error(sb) == 0) 3261 valid_len = sbuf_len(sb); 3262 } 3263 NET_EPOCH_EXIT(et); 3264 if (addrs == 0) { 3265 sbuf_bcat(sb, &ifr, sizeof(ifr)); 3266 max_len += sizeof(ifr); 3267 3268 if (sbuf_error(sb) == 0) 3269 valid_len = sbuf_len(sb); 3270 } 3271 } 3272 IFNET_RUNLOCK(); 3273 3274 /* 3275 * If we didn't allocate enough space (uncommon), try again. If 3276 * we have already allocated as much space as we are allowed, 3277 * return what we've got. 3278 */ 3279 if (valid_len != max_len && !full) { 3280 sbuf_delete(sb); 3281 goto again; 3282 } 3283 3284 ifc->ifc_len = valid_len; 3285 sbuf_finish(sb); 3286 error = copyout(sbuf_data(sb), ifc->ifc_req, ifc->ifc_len); 3287 sbuf_delete(sb); 3288 return (error); 3289 } 3290 3291 /* 3292 * Just like ifpromisc(), but for all-multicast-reception mode. 3293 */ 3294 int 3295 if_allmulti(struct ifnet *ifp, int onswitch) 3296 { 3297 3298 return (if_setflag(ifp, IFF_ALLMULTI, 0, &ifp->if_amcount, onswitch)); 3299 } 3300 3301 struct ifmultiaddr * 3302 if_findmulti(struct ifnet *ifp, const struct sockaddr *sa) 3303 { 3304 struct ifmultiaddr *ifma; 3305 3306 IF_ADDR_LOCK_ASSERT(ifp); 3307 3308 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 3309 if (sa->sa_family == AF_LINK) { 3310 if (sa_dl_equal(ifma->ifma_addr, sa)) 3311 break; 3312 } else { 3313 if (sa_equal(ifma->ifma_addr, sa)) 3314 break; 3315 } 3316 } 3317 3318 return ifma; 3319 } 3320 3321 /* 3322 * Allocate a new ifmultiaddr and initialize based on passed arguments. We 3323 * make copies of passed sockaddrs. The ifmultiaddr will not be added to 3324 * the ifnet multicast address list here, so the caller must do that and 3325 * other setup work (such as notifying the device driver). The reference 3326 * count is initialized to 1. 3327 */ 3328 static struct ifmultiaddr * 3329 if_allocmulti(struct ifnet *ifp, struct sockaddr *sa, struct sockaddr *llsa, 3330 int mflags) 3331 { 3332 struct ifmultiaddr *ifma; 3333 struct sockaddr *dupsa; 3334 3335 ifma = malloc(sizeof *ifma, M_IFMADDR, mflags | 3336 M_ZERO); 3337 if (ifma == NULL) 3338 return (NULL); 3339 3340 dupsa = malloc(sa->sa_len, M_IFMADDR, mflags); 3341 if (dupsa == NULL) { 3342 free(ifma, M_IFMADDR); 3343 return (NULL); 3344 } 3345 bcopy(sa, dupsa, sa->sa_len); 3346 ifma->ifma_addr = dupsa; 3347 3348 ifma->ifma_ifp = ifp; 3349 ifma->ifma_refcount = 1; 3350 ifma->ifma_protospec = NULL; 3351 3352 if (llsa == NULL) { 3353 ifma->ifma_lladdr = NULL; 3354 return (ifma); 3355 } 3356 3357 dupsa = malloc(llsa->sa_len, M_IFMADDR, mflags); 3358 if (dupsa == NULL) { 3359 free(ifma->ifma_addr, M_IFMADDR); 3360 free(ifma, M_IFMADDR); 3361 return (NULL); 3362 } 3363 bcopy(llsa, dupsa, llsa->sa_len); 3364 ifma->ifma_lladdr = dupsa; 3365 3366 return (ifma); 3367 } 3368 3369 /* 3370 * if_freemulti: free ifmultiaddr structure and possibly attached related 3371 * addresses. The caller is responsible for implementing reference 3372 * counting, notifying the driver, handling routing messages, and releasing 3373 * any dependent link layer state. 3374 */ 3375 #ifdef MCAST_VERBOSE 3376 extern void kdb_backtrace(void); 3377 #endif 3378 static void 3379 if_freemulti_internal(struct ifmultiaddr *ifma) 3380 { 3381 3382 KASSERT(ifma->ifma_refcount == 0, ("if_freemulti: refcount %d", 3383 ifma->ifma_refcount)); 3384 3385 if (ifma->ifma_lladdr != NULL) 3386 free(ifma->ifma_lladdr, M_IFMADDR); 3387 #ifdef MCAST_VERBOSE 3388 kdb_backtrace(); 3389 printf("%s freeing ifma: %p\n", __func__, ifma); 3390 #endif 3391 free(ifma->ifma_addr, M_IFMADDR); 3392 free(ifma, M_IFMADDR); 3393 } 3394 3395 static void 3396 if_destroymulti(epoch_context_t ctx) 3397 { 3398 struct ifmultiaddr *ifma; 3399 3400 ifma = __containerof(ctx, struct ifmultiaddr, ifma_epoch_ctx); 3401 if_freemulti_internal(ifma); 3402 } 3403 3404 void 3405 if_freemulti(struct ifmultiaddr *ifma) 3406 { 3407 KASSERT(ifma->ifma_refcount == 0, ("if_freemulti_epoch: refcount %d", 3408 ifma->ifma_refcount)); 3409 3410 epoch_call(net_epoch_preempt, &ifma->ifma_epoch_ctx, if_destroymulti); 3411 } 3412 3413 3414 /* 3415 * Register an additional multicast address with a network interface. 3416 * 3417 * - If the address is already present, bump the reference count on the 3418 * address and return. 3419 * - If the address is not link-layer, look up a link layer address. 3420 * - Allocate address structures for one or both addresses, and attach to the 3421 * multicast address list on the interface. If automatically adding a link 3422 * layer address, the protocol address will own a reference to the link 3423 * layer address, to be freed when it is freed. 3424 * - Notify the network device driver of an addition to the multicast address 3425 * list. 3426 * 3427 * 'sa' points to caller-owned memory with the desired multicast address. 3428 * 3429 * 'retifma' will be used to return a pointer to the resulting multicast 3430 * address reference, if desired. 3431 */ 3432 int 3433 if_addmulti(struct ifnet *ifp, struct sockaddr *sa, 3434 struct ifmultiaddr **retifma) 3435 { 3436 struct ifmultiaddr *ifma, *ll_ifma; 3437 struct sockaddr *llsa; 3438 struct sockaddr_dl sdl; 3439 int error; 3440 3441 #ifdef INET 3442 IN_MULTI_LIST_UNLOCK_ASSERT(); 3443 #endif 3444 #ifdef INET6 3445 IN6_MULTI_LIST_UNLOCK_ASSERT(); 3446 #endif 3447 /* 3448 * If the address is already present, return a new reference to it; 3449 * otherwise, allocate storage and set up a new address. 3450 */ 3451 IF_ADDR_WLOCK(ifp); 3452 ifma = if_findmulti(ifp, sa); 3453 if (ifma != NULL) { 3454 ifma->ifma_refcount++; 3455 if (retifma != NULL) 3456 *retifma = ifma; 3457 IF_ADDR_WUNLOCK(ifp); 3458 return (0); 3459 } 3460 3461 /* 3462 * The address isn't already present; resolve the protocol address 3463 * into a link layer address, and then look that up, bump its 3464 * refcount or allocate an ifma for that also. 3465 * Most link layer resolving functions returns address data which 3466 * fits inside default sockaddr_dl structure. However callback 3467 * can allocate another sockaddr structure, in that case we need to 3468 * free it later. 3469 */ 3470 llsa = NULL; 3471 ll_ifma = NULL; 3472 if (ifp->if_resolvemulti != NULL) { 3473 /* Provide called function with buffer size information */ 3474 sdl.sdl_len = sizeof(sdl); 3475 llsa = (struct sockaddr *)&sdl; 3476 error = ifp->if_resolvemulti(ifp, &llsa, sa); 3477 if (error) 3478 goto unlock_out; 3479 } 3480 3481 /* 3482 * Allocate the new address. Don't hook it up yet, as we may also 3483 * need to allocate a link layer multicast address. 3484 */ 3485 ifma = if_allocmulti(ifp, sa, llsa, M_NOWAIT); 3486 if (ifma == NULL) { 3487 error = ENOMEM; 3488 goto free_llsa_out; 3489 } 3490 3491 /* 3492 * If a link layer address is found, we'll need to see if it's 3493 * already present in the address list, or allocate is as well. 3494 * When this block finishes, the link layer address will be on the 3495 * list. 3496 */ 3497 if (llsa != NULL) { 3498 ll_ifma = if_findmulti(ifp, llsa); 3499 if (ll_ifma == NULL) { 3500 ll_ifma = if_allocmulti(ifp, llsa, NULL, M_NOWAIT); 3501 if (ll_ifma == NULL) { 3502 --ifma->ifma_refcount; 3503 if_freemulti(ifma); 3504 error = ENOMEM; 3505 goto free_llsa_out; 3506 } 3507 ll_ifma->ifma_flags |= IFMA_F_ENQUEUED; 3508 CK_STAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ll_ifma, 3509 ifma_link); 3510 } else 3511 ll_ifma->ifma_refcount++; 3512 ifma->ifma_llifma = ll_ifma; 3513 } 3514 3515 /* 3516 * We now have a new multicast address, ifma, and possibly a new or 3517 * referenced link layer address. Add the primary address to the 3518 * ifnet address list. 3519 */ 3520 ifma->ifma_flags |= IFMA_F_ENQUEUED; 3521 CK_STAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link); 3522 3523 if (retifma != NULL) 3524 *retifma = ifma; 3525 3526 /* 3527 * Must generate the message while holding the lock so that 'ifma' 3528 * pointer is still valid. 3529 */ 3530 rt_newmaddrmsg(RTM_NEWMADDR, ifma); 3531 IF_ADDR_WUNLOCK(ifp); 3532 3533 /* 3534 * We are certain we have added something, so call down to the 3535 * interface to let them know about it. 3536 */ 3537 if (ifp->if_ioctl != NULL) { 3538 if (THREAD_CAN_SLEEP()) 3539 (void )(*ifp->if_ioctl)(ifp, SIOCADDMULTI, 0); 3540 else 3541 taskqueue_enqueue(taskqueue_swi, &ifp->if_addmultitask); 3542 } 3543 3544 if ((llsa != NULL) && (llsa != (struct sockaddr *)&sdl)) 3545 link_free_sdl(llsa); 3546 3547 return (0); 3548 3549 free_llsa_out: 3550 if ((llsa != NULL) && (llsa != (struct sockaddr *)&sdl)) 3551 link_free_sdl(llsa); 3552 3553 unlock_out: 3554 IF_ADDR_WUNLOCK(ifp); 3555 return (error); 3556 } 3557 3558 static void 3559 if_siocaddmulti(void *arg, int pending) 3560 { 3561 struct ifnet *ifp; 3562 3563 ifp = arg; 3564 #ifdef DIAGNOSTIC 3565 if (pending > 1) 3566 if_printf(ifp, "%d SIOCADDMULTI coalesced\n", pending); 3567 #endif 3568 CURVNET_SET(ifp->if_vnet); 3569 (void )(*ifp->if_ioctl)(ifp, SIOCADDMULTI, 0); 3570 CURVNET_RESTORE(); 3571 } 3572 3573 /* 3574 * Delete a multicast group membership by network-layer group address. 3575 * 3576 * Returns ENOENT if the entry could not be found. If ifp no longer 3577 * exists, results are undefined. This entry point should only be used 3578 * from subsystems which do appropriate locking to hold ifp for the 3579 * duration of the call. 3580 * Network-layer protocol domains must use if_delmulti_ifma(). 3581 */ 3582 int 3583 if_delmulti(struct ifnet *ifp, struct sockaddr *sa) 3584 { 3585 struct ifmultiaddr *ifma; 3586 int lastref; 3587 3588 KASSERT(ifp, ("%s: NULL ifp", __func__)); 3589 3590 IF_ADDR_WLOCK(ifp); 3591 lastref = 0; 3592 ifma = if_findmulti(ifp, sa); 3593 if (ifma != NULL) 3594 lastref = if_delmulti_locked(ifp, ifma, 0); 3595 IF_ADDR_WUNLOCK(ifp); 3596 3597 if (ifma == NULL) 3598 return (ENOENT); 3599 3600 if (lastref && ifp->if_ioctl != NULL) { 3601 (void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0); 3602 } 3603 3604 return (0); 3605 } 3606 3607 /* 3608 * Delete all multicast group membership for an interface. 3609 * Should be used to quickly flush all multicast filters. 3610 */ 3611 void 3612 if_delallmulti(struct ifnet *ifp) 3613 { 3614 struct ifmultiaddr *ifma; 3615 struct ifmultiaddr *next; 3616 3617 IF_ADDR_WLOCK(ifp); 3618 CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next) 3619 if_delmulti_locked(ifp, ifma, 0); 3620 IF_ADDR_WUNLOCK(ifp); 3621 } 3622 3623 void 3624 if_delmulti_ifma(struct ifmultiaddr *ifma) 3625 { 3626 if_delmulti_ifma_flags(ifma, 0); 3627 } 3628 3629 /* 3630 * Delete a multicast group membership by group membership pointer. 3631 * Network-layer protocol domains must use this routine. 3632 * 3633 * It is safe to call this routine if the ifp disappeared. 3634 */ 3635 void 3636 if_delmulti_ifma_flags(struct ifmultiaddr *ifma, int flags) 3637 { 3638 struct ifnet *ifp; 3639 int lastref; 3640 MCDPRINTF("%s freeing ifma: %p\n", __func__, ifma); 3641 #ifdef INET 3642 IN_MULTI_LIST_UNLOCK_ASSERT(); 3643 #endif 3644 ifp = ifma->ifma_ifp; 3645 #ifdef DIAGNOSTIC 3646 if (ifp == NULL) { 3647 printf("%s: ifma_ifp seems to be detached\n", __func__); 3648 } else { 3649 struct epoch_tracker et; 3650 struct ifnet *oifp; 3651 3652 NET_EPOCH_ENTER(et); 3653 CK_STAILQ_FOREACH(oifp, &V_ifnet, if_link) 3654 if (ifp == oifp) 3655 break; 3656 NET_EPOCH_EXIT(et); 3657 if (ifp != oifp) 3658 ifp = NULL; 3659 } 3660 #endif 3661 /* 3662 * If and only if the ifnet instance exists: Acquire the address lock. 3663 */ 3664 if (ifp != NULL) 3665 IF_ADDR_WLOCK(ifp); 3666 3667 lastref = if_delmulti_locked(ifp, ifma, flags); 3668 3669 if (ifp != NULL) { 3670 /* 3671 * If and only if the ifnet instance exists: 3672 * Release the address lock. 3673 * If the group was left: update the hardware hash filter. 3674 */ 3675 IF_ADDR_WUNLOCK(ifp); 3676 if (lastref && ifp->if_ioctl != NULL) { 3677 (void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0); 3678 } 3679 } 3680 } 3681 3682 /* 3683 * Perform deletion of network-layer and/or link-layer multicast address. 3684 * 3685 * Return 0 if the reference count was decremented. 3686 * Return 1 if the final reference was released, indicating that the 3687 * hardware hash filter should be reprogrammed. 3688 */ 3689 static int 3690 if_delmulti_locked(struct ifnet *ifp, struct ifmultiaddr *ifma, int detaching) 3691 { 3692 struct ifmultiaddr *ll_ifma; 3693 3694 if (ifp != NULL && ifma->ifma_ifp != NULL) { 3695 KASSERT(ifma->ifma_ifp == ifp, 3696 ("%s: inconsistent ifp %p", __func__, ifp)); 3697 IF_ADDR_WLOCK_ASSERT(ifp); 3698 } 3699 3700 ifp = ifma->ifma_ifp; 3701 MCDPRINTF("%s freeing %p from %s \n", __func__, ifma, ifp ? ifp->if_xname : ""); 3702 3703 /* 3704 * If the ifnet is detaching, null out references to ifnet, 3705 * so that upper protocol layers will notice, and not attempt 3706 * to obtain locks for an ifnet which no longer exists. The 3707 * routing socket announcement must happen before the ifnet 3708 * instance is detached from the system. 3709 */ 3710 if (detaching) { 3711 #ifdef DIAGNOSTIC 3712 printf("%s: detaching ifnet instance %p\n", __func__, ifp); 3713 #endif 3714 /* 3715 * ifp may already be nulled out if we are being reentered 3716 * to delete the ll_ifma. 3717 */ 3718 if (ifp != NULL) { 3719 rt_newmaddrmsg(RTM_DELMADDR, ifma); 3720 ifma->ifma_ifp = NULL; 3721 } 3722 } 3723 3724 if (--ifma->ifma_refcount > 0) 3725 return 0; 3726 3727 if (ifp != NULL && detaching == 0 && (ifma->ifma_flags & IFMA_F_ENQUEUED)) { 3728 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifmultiaddr, ifma_link); 3729 ifma->ifma_flags &= ~IFMA_F_ENQUEUED; 3730 } 3731 /* 3732 * If this ifma is a network-layer ifma, a link-layer ifma may 3733 * have been associated with it. Release it first if so. 3734 */ 3735 ll_ifma = ifma->ifma_llifma; 3736 if (ll_ifma != NULL) { 3737 KASSERT(ifma->ifma_lladdr != NULL, 3738 ("%s: llifma w/o lladdr", __func__)); 3739 if (detaching) 3740 ll_ifma->ifma_ifp = NULL; /* XXX */ 3741 if (--ll_ifma->ifma_refcount == 0) { 3742 if (ifp != NULL) { 3743 if (ll_ifma->ifma_flags & IFMA_F_ENQUEUED) { 3744 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifmultiaddr, 3745 ifma_link); 3746 ll_ifma->ifma_flags &= ~IFMA_F_ENQUEUED; 3747 } 3748 } 3749 if_freemulti(ll_ifma); 3750 } 3751 } 3752 #ifdef INVARIANTS 3753 if (ifp) { 3754 struct ifmultiaddr *ifmatmp; 3755 3756 CK_STAILQ_FOREACH(ifmatmp, &ifp->if_multiaddrs, ifma_link) 3757 MPASS(ifma != ifmatmp); 3758 } 3759 #endif 3760 if_freemulti(ifma); 3761 /* 3762 * The last reference to this instance of struct ifmultiaddr 3763 * was released; the hardware should be notified of this change. 3764 */ 3765 return 1; 3766 } 3767 3768 /* 3769 * Set the link layer address on an interface. 3770 * 3771 * At this time we only support certain types of interfaces, 3772 * and we don't allow the length of the address to change. 3773 * 3774 * Set noinline to be dtrace-friendly 3775 */ 3776 __noinline int 3777 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len) 3778 { 3779 struct sockaddr_dl *sdl; 3780 struct ifaddr *ifa; 3781 struct ifreq ifr; 3782 3783 ifa = ifp->if_addr; 3784 if (ifa == NULL) 3785 return (EINVAL); 3786 3787 sdl = (struct sockaddr_dl *)ifa->ifa_addr; 3788 if (sdl == NULL) 3789 return (EINVAL); 3790 3791 if (len != sdl->sdl_alen) /* don't allow length to change */ 3792 return (EINVAL); 3793 3794 switch (ifp->if_type) { 3795 case IFT_ETHER: 3796 case IFT_XETHER: 3797 case IFT_L2VLAN: 3798 case IFT_BRIDGE: 3799 case IFT_IEEE8023ADLAG: 3800 bcopy(lladdr, LLADDR(sdl), len); 3801 break; 3802 default: 3803 return (ENODEV); 3804 } 3805 3806 /* 3807 * If the interface is already up, we need 3808 * to re-init it in order to reprogram its 3809 * address filter. 3810 */ 3811 if ((ifp->if_flags & IFF_UP) != 0) { 3812 if (ifp->if_ioctl) { 3813 ifp->if_flags &= ~IFF_UP; 3814 ifr.ifr_flags = ifp->if_flags & 0xffff; 3815 ifr.ifr_flagshigh = ifp->if_flags >> 16; 3816 (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr); 3817 ifp->if_flags |= IFF_UP; 3818 ifr.ifr_flags = ifp->if_flags & 0xffff; 3819 ifr.ifr_flagshigh = ifp->if_flags >> 16; 3820 (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr); 3821 } 3822 } 3823 EVENTHANDLER_INVOKE(iflladdr_event, ifp); 3824 3825 return (0); 3826 } 3827 3828 /* 3829 * Compat function for handling basic encapsulation requests. 3830 * Not converted stacks (FDDI, IB, ..) supports traditional 3831 * output model: ARP (and other similar L2 protocols) are handled 3832 * inside output routine, arpresolve/nd6_resolve() returns MAC 3833 * address instead of full prepend. 3834 * 3835 * This function creates calculated header==MAC for IPv4/IPv6 and 3836 * returns EAFNOSUPPORT (which is then handled in ARP code) for other 3837 * address families. 3838 */ 3839 static int 3840 if_requestencap_default(struct ifnet *ifp, struct if_encap_req *req) 3841 { 3842 3843 if (req->rtype != IFENCAP_LL) 3844 return (EOPNOTSUPP); 3845 3846 if (req->bufsize < req->lladdr_len) 3847 return (ENOMEM); 3848 3849 switch (req->family) { 3850 case AF_INET: 3851 case AF_INET6: 3852 break; 3853 default: 3854 return (EAFNOSUPPORT); 3855 } 3856 3857 /* Copy lladdr to storage as is */ 3858 memmove(req->buf, req->lladdr, req->lladdr_len); 3859 req->bufsize = req->lladdr_len; 3860 req->lladdr_off = 0; 3861 3862 return (0); 3863 } 3864 3865 /* 3866 * Tunnel interfaces can nest, also they may cause infinite recursion 3867 * calls when misconfigured. We'll prevent this by detecting loops. 3868 * High nesting level may cause stack exhaustion. We'll prevent this 3869 * by introducing upper limit. 3870 * 3871 * Return 0, if tunnel nesting count is equal or less than limit. 3872 */ 3873 int 3874 if_tunnel_check_nesting(struct ifnet *ifp, struct mbuf *m, uint32_t cookie, 3875 int limit) 3876 { 3877 struct m_tag *mtag; 3878 int count; 3879 3880 count = 1; 3881 mtag = NULL; 3882 while ((mtag = m_tag_locate(m, cookie, 0, mtag)) != NULL) { 3883 if (*(struct ifnet **)(mtag + 1) == ifp) { 3884 log(LOG_NOTICE, "%s: loop detected\n", if_name(ifp)); 3885 return (EIO); 3886 } 3887 count++; 3888 } 3889 if (count > limit) { 3890 log(LOG_NOTICE, 3891 "%s: if_output recursively called too many times(%d)\n", 3892 if_name(ifp), count); 3893 return (EIO); 3894 } 3895 mtag = m_tag_alloc(cookie, 0, sizeof(struct ifnet *), M_NOWAIT); 3896 if (mtag == NULL) 3897 return (ENOMEM); 3898 *(struct ifnet **)(mtag + 1) = ifp; 3899 m_tag_prepend(m, mtag); 3900 return (0); 3901 } 3902 3903 /* 3904 * Get the link layer address that was read from the hardware at attach. 3905 * 3906 * This is only set by Ethernet NICs (IFT_ETHER), but laggX interfaces re-type 3907 * their component interfaces as IFT_IEEE8023ADLAG. 3908 */ 3909 int 3910 if_gethwaddr(struct ifnet *ifp, struct ifreq *ifr) 3911 { 3912 3913 if (ifp->if_hw_addr == NULL) 3914 return (ENODEV); 3915 3916 switch (ifp->if_type) { 3917 case IFT_ETHER: 3918 case IFT_IEEE8023ADLAG: 3919 bcopy(ifp->if_hw_addr, ifr->ifr_addr.sa_data, ifp->if_addrlen); 3920 return (0); 3921 default: 3922 return (ENODEV); 3923 } 3924 } 3925 3926 /* 3927 * The name argument must be a pointer to storage which will last as 3928 * long as the interface does. For physical devices, the result of 3929 * device_get_name(dev) is a good choice and for pseudo-devices a 3930 * static string works well. 3931 */ 3932 void 3933 if_initname(struct ifnet *ifp, const char *name, int unit) 3934 { 3935 ifp->if_dname = name; 3936 ifp->if_dunit = unit; 3937 if (unit != IF_DUNIT_NONE) 3938 snprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit); 3939 else 3940 strlcpy(ifp->if_xname, name, IFNAMSIZ); 3941 } 3942 3943 int 3944 if_printf(struct ifnet *ifp, const char *fmt, ...) 3945 { 3946 char if_fmt[256]; 3947 va_list ap; 3948 3949 snprintf(if_fmt, sizeof(if_fmt), "%s: %s", ifp->if_xname, fmt); 3950 va_start(ap, fmt); 3951 vlog(LOG_INFO, if_fmt, ap); 3952 va_end(ap); 3953 return (0); 3954 } 3955 3956 void 3957 if_start(struct ifnet *ifp) 3958 { 3959 3960 (*(ifp)->if_start)(ifp); 3961 } 3962 3963 /* 3964 * Backwards compatibility interface for drivers 3965 * that have not implemented it 3966 */ 3967 static int 3968 if_transmit(struct ifnet *ifp, struct mbuf *m) 3969 { 3970 int error; 3971 3972 IFQ_HANDOFF(ifp, m, error); 3973 return (error); 3974 } 3975 3976 static void 3977 if_input_default(struct ifnet *ifp __unused, struct mbuf *m) 3978 { 3979 3980 m_freem(m); 3981 } 3982 3983 int 3984 if_handoff(struct ifqueue *ifq, struct mbuf *m, struct ifnet *ifp, int adjust) 3985 { 3986 int active = 0; 3987 3988 IF_LOCK(ifq); 3989 if (_IF_QFULL(ifq)) { 3990 IF_UNLOCK(ifq); 3991 if_inc_counter(ifp, IFCOUNTER_OQDROPS, 1); 3992 m_freem(m); 3993 return (0); 3994 } 3995 if (ifp != NULL) { 3996 if_inc_counter(ifp, IFCOUNTER_OBYTES, m->m_pkthdr.len + adjust); 3997 if (m->m_flags & (M_BCAST|M_MCAST)) 3998 if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1); 3999 active = ifp->if_drv_flags & IFF_DRV_OACTIVE; 4000 } 4001 _IF_ENQUEUE(ifq, m); 4002 IF_UNLOCK(ifq); 4003 if (ifp != NULL && !active) 4004 (*(ifp)->if_start)(ifp); 4005 return (1); 4006 } 4007 4008 void 4009 if_register_com_alloc(u_char type, 4010 if_com_alloc_t *a, if_com_free_t *f) 4011 { 4012 4013 KASSERT(if_com_alloc[type] == NULL, 4014 ("if_register_com_alloc: %d already registered", type)); 4015 KASSERT(if_com_free[type] == NULL, 4016 ("if_register_com_alloc: %d free already registered", type)); 4017 4018 if_com_alloc[type] = a; 4019 if_com_free[type] = f; 4020 } 4021 4022 void 4023 if_deregister_com_alloc(u_char type) 4024 { 4025 4026 KASSERT(if_com_alloc[type] != NULL, 4027 ("if_deregister_com_alloc: %d not registered", type)); 4028 KASSERT(if_com_free[type] != NULL, 4029 ("if_deregister_com_alloc: %d free not registered", type)); 4030 if_com_alloc[type] = NULL; 4031 if_com_free[type] = NULL; 4032 } 4033 4034 /* API for driver access to network stack owned ifnet.*/ 4035 uint64_t 4036 if_setbaudrate(struct ifnet *ifp, uint64_t baudrate) 4037 { 4038 uint64_t oldbrate; 4039 4040 oldbrate = ifp->if_baudrate; 4041 ifp->if_baudrate = baudrate; 4042 return (oldbrate); 4043 } 4044 4045 uint64_t 4046 if_getbaudrate(if_t ifp) 4047 { 4048 4049 return (((struct ifnet *)ifp)->if_baudrate); 4050 } 4051 4052 int 4053 if_setcapabilities(if_t ifp, int capabilities) 4054 { 4055 ((struct ifnet *)ifp)->if_capabilities = capabilities; 4056 return (0); 4057 } 4058 4059 int 4060 if_setcapabilitiesbit(if_t ifp, int setbit, int clearbit) 4061 { 4062 ((struct ifnet *)ifp)->if_capabilities |= setbit; 4063 ((struct ifnet *)ifp)->if_capabilities &= ~clearbit; 4064 4065 return (0); 4066 } 4067 4068 int 4069 if_getcapabilities(if_t ifp) 4070 { 4071 return ((struct ifnet *)ifp)->if_capabilities; 4072 } 4073 4074 int 4075 if_setcapenable(if_t ifp, int capabilities) 4076 { 4077 ((struct ifnet *)ifp)->if_capenable = capabilities; 4078 return (0); 4079 } 4080 4081 int 4082 if_setcapenablebit(if_t ifp, int setcap, int clearcap) 4083 { 4084 if(setcap) 4085 ((struct ifnet *)ifp)->if_capenable |= setcap; 4086 if(clearcap) 4087 ((struct ifnet *)ifp)->if_capenable &= ~clearcap; 4088 4089 return (0); 4090 } 4091 4092 const char * 4093 if_getdname(if_t ifp) 4094 { 4095 return ((struct ifnet *)ifp)->if_dname; 4096 } 4097 4098 int 4099 if_togglecapenable(if_t ifp, int togglecap) 4100 { 4101 ((struct ifnet *)ifp)->if_capenable ^= togglecap; 4102 return (0); 4103 } 4104 4105 int 4106 if_getcapenable(if_t ifp) 4107 { 4108 return ((struct ifnet *)ifp)->if_capenable; 4109 } 4110 4111 /* 4112 * This is largely undesirable because it ties ifnet to a device, but does 4113 * provide flexiblity for an embedded product vendor. Should be used with 4114 * the understanding that it violates the interface boundaries, and should be 4115 * a last resort only. 4116 */ 4117 int 4118 if_setdev(if_t ifp, void *dev) 4119 { 4120 return (0); 4121 } 4122 4123 int 4124 if_setdrvflagbits(if_t ifp, int set_flags, int clear_flags) 4125 { 4126 ((struct ifnet *)ifp)->if_drv_flags |= set_flags; 4127 ((struct ifnet *)ifp)->if_drv_flags &= ~clear_flags; 4128 4129 return (0); 4130 } 4131 4132 int 4133 if_getdrvflags(if_t ifp) 4134 { 4135 return ((struct ifnet *)ifp)->if_drv_flags; 4136 } 4137 4138 int 4139 if_setdrvflags(if_t ifp, int flags) 4140 { 4141 ((struct ifnet *)ifp)->if_drv_flags = flags; 4142 return (0); 4143 } 4144 4145 4146 int 4147 if_setflags(if_t ifp, int flags) 4148 { 4149 ((struct ifnet *)ifp)->if_flags = flags; 4150 return (0); 4151 } 4152 4153 int 4154 if_setflagbits(if_t ifp, int set, int clear) 4155 { 4156 ((struct ifnet *)ifp)->if_flags |= set; 4157 ((struct ifnet *)ifp)->if_flags &= ~clear; 4158 4159 return (0); 4160 } 4161 4162 int 4163 if_getflags(if_t ifp) 4164 { 4165 return ((struct ifnet *)ifp)->if_flags; 4166 } 4167 4168 int 4169 if_clearhwassist(if_t ifp) 4170 { 4171 ((struct ifnet *)ifp)->if_hwassist = 0; 4172 return (0); 4173 } 4174 4175 int 4176 if_sethwassistbits(if_t ifp, int toset, int toclear) 4177 { 4178 ((struct ifnet *)ifp)->if_hwassist |= toset; 4179 ((struct ifnet *)ifp)->if_hwassist &= ~toclear; 4180 4181 return (0); 4182 } 4183 4184 int 4185 if_sethwassist(if_t ifp, int hwassist_bit) 4186 { 4187 ((struct ifnet *)ifp)->if_hwassist = hwassist_bit; 4188 return (0); 4189 } 4190 4191 int 4192 if_gethwassist(if_t ifp) 4193 { 4194 return ((struct ifnet *)ifp)->if_hwassist; 4195 } 4196 4197 int 4198 if_setmtu(if_t ifp, int mtu) 4199 { 4200 ((struct ifnet *)ifp)->if_mtu = mtu; 4201 return (0); 4202 } 4203 4204 int 4205 if_getmtu(if_t ifp) 4206 { 4207 return ((struct ifnet *)ifp)->if_mtu; 4208 } 4209 4210 int 4211 if_getmtu_family(if_t ifp, int family) 4212 { 4213 struct domain *dp; 4214 4215 for (dp = domains; dp; dp = dp->dom_next) { 4216 if (dp->dom_family == family && dp->dom_ifmtu != NULL) 4217 return (dp->dom_ifmtu((struct ifnet *)ifp)); 4218 } 4219 4220 return (((struct ifnet *)ifp)->if_mtu); 4221 } 4222 4223 /* 4224 * Methods for drivers to access interface unicast and multicast 4225 * link level addresses. Driver shall not know 'struct ifaddr' neither 4226 * 'struct ifmultiaddr'. 4227 */ 4228 u_int 4229 if_lladdr_count(if_t ifp) 4230 { 4231 struct epoch_tracker et; 4232 struct ifaddr *ifa; 4233 u_int count; 4234 4235 count = 0; 4236 NET_EPOCH_ENTER(et); 4237 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) 4238 if (ifa->ifa_addr->sa_family == AF_LINK) 4239 count++; 4240 NET_EPOCH_EXIT(et); 4241 4242 return (count); 4243 } 4244 4245 u_int 4246 if_foreach_lladdr(if_t ifp, iflladdr_cb_t cb, void *cb_arg) 4247 { 4248 struct epoch_tracker et; 4249 struct ifaddr *ifa; 4250 u_int count; 4251 4252 MPASS(cb); 4253 4254 count = 0; 4255 NET_EPOCH_ENTER(et); 4256 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 4257 if (ifa->ifa_addr->sa_family != AF_LINK) 4258 continue; 4259 count += (*cb)(cb_arg, (struct sockaddr_dl *)ifa->ifa_addr, 4260 count); 4261 } 4262 NET_EPOCH_EXIT(et); 4263 4264 return (count); 4265 } 4266 4267 u_int 4268 if_llmaddr_count(if_t ifp) 4269 { 4270 struct epoch_tracker et; 4271 struct ifmultiaddr *ifma; 4272 int count; 4273 4274 count = 0; 4275 NET_EPOCH_ENTER(et); 4276 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) 4277 if (ifma->ifma_addr->sa_family == AF_LINK) 4278 count++; 4279 NET_EPOCH_EXIT(et); 4280 4281 return (count); 4282 } 4283 4284 u_int 4285 if_foreach_llmaddr(if_t ifp, iflladdr_cb_t cb, void *cb_arg) 4286 { 4287 struct epoch_tracker et; 4288 struct ifmultiaddr *ifma; 4289 u_int count; 4290 4291 MPASS(cb); 4292 4293 count = 0; 4294 NET_EPOCH_ENTER(et); 4295 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 4296 if (ifma->ifma_addr->sa_family != AF_LINK) 4297 continue; 4298 count += (*cb)(cb_arg, (struct sockaddr_dl *)ifma->ifma_addr, 4299 count); 4300 } 4301 NET_EPOCH_EXIT(et); 4302 4303 return (count); 4304 } 4305 4306 int 4307 if_setsoftc(if_t ifp, void *softc) 4308 { 4309 ((struct ifnet *)ifp)->if_softc = softc; 4310 return (0); 4311 } 4312 4313 void * 4314 if_getsoftc(if_t ifp) 4315 { 4316 return ((struct ifnet *)ifp)->if_softc; 4317 } 4318 4319 void 4320 if_setrcvif(struct mbuf *m, if_t ifp) 4321 { 4322 4323 MPASS((m->m_pkthdr.csum_flags & CSUM_SND_TAG) == 0); 4324 m->m_pkthdr.rcvif = (struct ifnet *)ifp; 4325 } 4326 4327 void 4328 if_setvtag(struct mbuf *m, uint16_t tag) 4329 { 4330 m->m_pkthdr.ether_vtag = tag; 4331 } 4332 4333 uint16_t 4334 if_getvtag(struct mbuf *m) 4335 { 4336 4337 return (m->m_pkthdr.ether_vtag); 4338 } 4339 4340 int 4341 if_sendq_empty(if_t ifp) 4342 { 4343 return IFQ_DRV_IS_EMPTY(&((struct ifnet *)ifp)->if_snd); 4344 } 4345 4346 struct ifaddr * 4347 if_getifaddr(if_t ifp) 4348 { 4349 return ((struct ifnet *)ifp)->if_addr; 4350 } 4351 4352 int 4353 if_getamcount(if_t ifp) 4354 { 4355 return ((struct ifnet *)ifp)->if_amcount; 4356 } 4357 4358 4359 int 4360 if_setsendqready(if_t ifp) 4361 { 4362 IFQ_SET_READY(&((struct ifnet *)ifp)->if_snd); 4363 return (0); 4364 } 4365 4366 int 4367 if_setsendqlen(if_t ifp, int tx_desc_count) 4368 { 4369 IFQ_SET_MAXLEN(&((struct ifnet *)ifp)->if_snd, tx_desc_count); 4370 ((struct ifnet *)ifp)->if_snd.ifq_drv_maxlen = tx_desc_count; 4371 4372 return (0); 4373 } 4374 4375 int 4376 if_vlantrunkinuse(if_t ifp) 4377 { 4378 return ((struct ifnet *)ifp)->if_vlantrunk != NULL?1:0; 4379 } 4380 4381 int 4382 if_input(if_t ifp, struct mbuf* sendmp) 4383 { 4384 (*((struct ifnet *)ifp)->if_input)((struct ifnet *)ifp, sendmp); 4385 return (0); 4386 4387 } 4388 4389 struct mbuf * 4390 if_dequeue(if_t ifp) 4391 { 4392 struct mbuf *m; 4393 IFQ_DRV_DEQUEUE(&((struct ifnet *)ifp)->if_snd, m); 4394 4395 return (m); 4396 } 4397 4398 int 4399 if_sendq_prepend(if_t ifp, struct mbuf *m) 4400 { 4401 IFQ_DRV_PREPEND(&((struct ifnet *)ifp)->if_snd, m); 4402 return (0); 4403 } 4404 4405 int 4406 if_setifheaderlen(if_t ifp, int len) 4407 { 4408 ((struct ifnet *)ifp)->if_hdrlen = len; 4409 return (0); 4410 } 4411 4412 caddr_t 4413 if_getlladdr(if_t ifp) 4414 { 4415 return (IF_LLADDR((struct ifnet *)ifp)); 4416 } 4417 4418 void * 4419 if_gethandle(u_char type) 4420 { 4421 return (if_alloc(type)); 4422 } 4423 4424 void 4425 if_bpfmtap(if_t ifh, struct mbuf *m) 4426 { 4427 struct ifnet *ifp = (struct ifnet *)ifh; 4428 4429 BPF_MTAP(ifp, m); 4430 } 4431 4432 void 4433 if_etherbpfmtap(if_t ifh, struct mbuf *m) 4434 { 4435 struct ifnet *ifp = (struct ifnet *)ifh; 4436 4437 ETHER_BPF_MTAP(ifp, m); 4438 } 4439 4440 void 4441 if_vlancap(if_t ifh) 4442 { 4443 struct ifnet *ifp = (struct ifnet *)ifh; 4444 VLAN_CAPABILITIES(ifp); 4445 } 4446 4447 int 4448 if_sethwtsomax(if_t ifp, u_int if_hw_tsomax) 4449 { 4450 4451 ((struct ifnet *)ifp)->if_hw_tsomax = if_hw_tsomax; 4452 return (0); 4453 } 4454 4455 int 4456 if_sethwtsomaxsegcount(if_t ifp, u_int if_hw_tsomaxsegcount) 4457 { 4458 4459 ((struct ifnet *)ifp)->if_hw_tsomaxsegcount = if_hw_tsomaxsegcount; 4460 return (0); 4461 } 4462 4463 int 4464 if_sethwtsomaxsegsize(if_t ifp, u_int if_hw_tsomaxsegsize) 4465 { 4466 4467 ((struct ifnet *)ifp)->if_hw_tsomaxsegsize = if_hw_tsomaxsegsize; 4468 return (0); 4469 } 4470 4471 u_int 4472 if_gethwtsomax(if_t ifp) 4473 { 4474 4475 return (((struct ifnet *)ifp)->if_hw_tsomax); 4476 } 4477 4478 u_int 4479 if_gethwtsomaxsegcount(if_t ifp) 4480 { 4481 4482 return (((struct ifnet *)ifp)->if_hw_tsomaxsegcount); 4483 } 4484 4485 u_int 4486 if_gethwtsomaxsegsize(if_t ifp) 4487 { 4488 4489 return (((struct ifnet *)ifp)->if_hw_tsomaxsegsize); 4490 } 4491 4492 void 4493 if_setinitfn(if_t ifp, void (*init_fn)(void *)) 4494 { 4495 ((struct ifnet *)ifp)->if_init = init_fn; 4496 } 4497 4498 void 4499 if_setioctlfn(if_t ifp, int (*ioctl_fn)(if_t, u_long, caddr_t)) 4500 { 4501 ((struct ifnet *)ifp)->if_ioctl = (void *)ioctl_fn; 4502 } 4503 4504 void 4505 if_setstartfn(if_t ifp, void (*start_fn)(if_t)) 4506 { 4507 ((struct ifnet *)ifp)->if_start = (void *)start_fn; 4508 } 4509 4510 void 4511 if_settransmitfn(if_t ifp, if_transmit_fn_t start_fn) 4512 { 4513 ((struct ifnet *)ifp)->if_transmit = start_fn; 4514 } 4515 4516 void if_setqflushfn(if_t ifp, if_qflush_fn_t flush_fn) 4517 { 4518 ((struct ifnet *)ifp)->if_qflush = flush_fn; 4519 4520 } 4521 4522 void 4523 if_setgetcounterfn(if_t ifp, if_get_counter_t fn) 4524 { 4525 4526 ifp->if_get_counter = fn; 4527 } 4528 4529 /* Revisit these - These are inline functions originally. */ 4530 int 4531 drbr_inuse_drv(if_t ifh, struct buf_ring *br) 4532 { 4533 return drbr_inuse(ifh, br); 4534 } 4535 4536 struct mbuf* 4537 drbr_dequeue_drv(if_t ifh, struct buf_ring *br) 4538 { 4539 return drbr_dequeue(ifh, br); 4540 } 4541 4542 int 4543 drbr_needs_enqueue_drv(if_t ifh, struct buf_ring *br) 4544 { 4545 return drbr_needs_enqueue(ifh, br); 4546 } 4547 4548 int 4549 drbr_enqueue_drv(if_t ifh, struct buf_ring *br, struct mbuf *m) 4550 { 4551 return drbr_enqueue(ifh, br, m); 4552 4553 } 4554