af_inet.c revision a9c19329eccdb145a08a4a2e969d7b40c54c9bcc
1/* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * PF_INET protocol family socket handler. 7 * 8 * Authors: Ross Biro 9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 10 * Florian La Roche, <flla@stud.uni-sb.de> 11 * Alan Cox, <A.Cox@swansea.ac.uk> 12 * 13 * Changes (see also sock.c) 14 * 15 * piggy, 16 * Karl Knutson : Socket protocol table 17 * A.N.Kuznetsov : Socket death error in accept(). 18 * John Richardson : Fix non blocking error in connect() 19 * so sockets that fail to connect 20 * don't return -EINPROGRESS. 21 * Alan Cox : Asynchronous I/O support 22 * Alan Cox : Keep correct socket pointer on sock 23 * structures 24 * when accept() ed 25 * Alan Cox : Semantics of SO_LINGER aren't state 26 * moved to close when you look carefully. 27 * With this fixed and the accept bug fixed 28 * some RPC stuff seems happier. 29 * Niibe Yutaka : 4.4BSD style write async I/O 30 * Alan Cox, 31 * Tony Gale : Fixed reuse semantics. 32 * Alan Cox : bind() shouldn't abort existing but dead 33 * sockets. Stops FTP netin:.. I hope. 34 * Alan Cox : bind() works correctly for RAW sockets. 35 * Note that FreeBSD at least was broken 36 * in this respect so be careful with 37 * compatibility tests... 38 * Alan Cox : routing cache support 39 * Alan Cox : memzero the socket structure for 40 * compactness. 41 * Matt Day : nonblock connect error handler 42 * Alan Cox : Allow large numbers of pending sockets 43 * (eg for big web sites), but only if 44 * specifically application requested. 45 * Alan Cox : New buffering throughout IP. Used 46 * dumbly. 47 * Alan Cox : New buffering now used smartly. 48 * Alan Cox : BSD rather than common sense 49 * interpretation of listen. 50 * Germano Caronni : Assorted small races. 51 * Alan Cox : sendmsg/recvmsg basic support. 52 * Alan Cox : Only sendmsg/recvmsg now supported. 53 * Alan Cox : Locked down bind (see security list). 54 * Alan Cox : Loosened bind a little. 55 * Mike McLagan : ADD/DEL DLCI Ioctls 56 * Willy Konynenberg : Transparent proxying support. 57 * David S. Miller : New socket lookup architecture. 58 * Some other random speedups. 59 * Cyrus Durgin : Cleaned up file for kmod hacks. 60 * Andi Kleen : Fix inet_stream_connect TCP race. 61 * 62 * This program is free software; you can redistribute it and/or 63 * modify it under the terms of the GNU General Public License 64 * as published by the Free Software Foundation; either version 65 * 2 of the License, or (at your option) any later version. 66 */ 67 68#include <linux/err.h> 69#include <linux/errno.h> 70#include <linux/types.h> 71#include <linux/socket.h> 72#include <linux/in.h> 73#include <linux/kernel.h> 74#include <linux/module.h> 75#include <linux/sched.h> 76#include <linux/timer.h> 77#include <linux/string.h> 78#include <linux/sockios.h> 79#include <linux/net.h> 80#include <linux/capability.h> 81#include <linux/fcntl.h> 82#include <linux/mm.h> 83#include <linux/interrupt.h> 84#include <linux/stat.h> 85#include <linux/init.h> 86#include <linux/poll.h> 87#include <linux/netfilter_ipv4.h> 88#include <linux/random.h> 89 90#include <asm/uaccess.h> 91#include <asm/system.h> 92 93#include <linux/inet.h> 94#include <linux/igmp.h> 95#include <linux/inetdevice.h> 96#include <linux/netdevice.h> 97#include <net/ip.h> 98#include <net/protocol.h> 99#include <net/arp.h> 100#include <net/route.h> 101#include <net/ip_fib.h> 102#include <net/inet_connection_sock.h> 103#include <net/tcp.h> 104#include <net/udp.h> 105#include <net/udplite.h> 106#include <linux/skbuff.h> 107#include <net/sock.h> 108#include <net/raw.h> 109#include <net/icmp.h> 110#include <net/ipip.h> 111#include <net/inet_common.h> 112#include <net/xfrm.h> 113#ifdef CONFIG_IP_MROUTE 114#include <linux/mroute.h> 115#endif 116 117DEFINE_SNMP_STAT(struct linux_mib, net_statistics) __read_mostly; 118 119extern void ip_mc_drop_socket(struct sock *sk); 120 121/* The inetsw table contains everything that inet_create needs to 122 * build a new socket. 123 */ 124static struct list_head inetsw[SOCK_MAX]; 125static DEFINE_SPINLOCK(inetsw_lock); 126 127struct ipv4_config ipv4_config; 128 129EXPORT_SYMBOL(ipv4_config); 130 131/* New destruction routine */ 132 133void inet_sock_destruct(struct sock *sk) 134{ 135 struct inet_sock *inet = inet_sk(sk); 136 137 __skb_queue_purge(&sk->sk_receive_queue); 138 __skb_queue_purge(&sk->sk_error_queue); 139 140 sk_mem_reclaim(sk); 141 142 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) { 143 printk("Attempt to release TCP socket in state %d %p\n", 144 sk->sk_state, sk); 145 return; 146 } 147 if (!sock_flag(sk, SOCK_DEAD)) { 148 printk("Attempt to release alive inet socket %p\n", sk); 149 return; 150 } 151 152 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc)); 153 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc)); 154 BUG_TRAP(!sk->sk_wmem_queued); 155 BUG_TRAP(!sk->sk_forward_alloc); 156 157 kfree(inet->opt); 158 dst_release(sk->sk_dst_cache); 159 sk_refcnt_debug_dec(sk); 160} 161 162/* 163 * The routines beyond this point handle the behaviour of an AF_INET 164 * socket object. Mostly it punts to the subprotocols of IP to do 165 * the work. 166 */ 167 168/* 169 * Automatically bind an unbound socket. 170 */ 171 172static int inet_autobind(struct sock *sk) 173{ 174 struct inet_sock *inet; 175 /* We may need to bind the socket. */ 176 lock_sock(sk); 177 inet = inet_sk(sk); 178 if (!inet->num) { 179 if (sk->sk_prot->get_port(sk, 0)) { 180 release_sock(sk); 181 return -EAGAIN; 182 } 183 inet->sport = htons(inet->num); 184 } 185 release_sock(sk); 186 return 0; 187} 188 189/* 190 * Move a socket into listening state. 191 */ 192int inet_listen(struct socket *sock, int backlog) 193{ 194 struct sock *sk = sock->sk; 195 unsigned char old_state; 196 int err; 197 198 lock_sock(sk); 199 200 err = -EINVAL; 201 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM) 202 goto out; 203 204 old_state = sk->sk_state; 205 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN))) 206 goto out; 207 208 /* Really, if the socket is already in listen state 209 * we can only allow the backlog to be adjusted. 210 */ 211 if (old_state != TCP_LISTEN) { 212 err = inet_csk_listen_start(sk, backlog); 213 if (err) 214 goto out; 215 } 216 sk->sk_max_ack_backlog = backlog; 217 err = 0; 218 219out: 220 release_sock(sk); 221 return err; 222} 223 224u32 inet_ehash_secret __read_mostly; 225EXPORT_SYMBOL(inet_ehash_secret); 226 227/* 228 * inet_ehash_secret must be set exactly once 229 * Instead of using a dedicated spinlock, we (ab)use inetsw_lock 230 */ 231void build_ehash_secret(void) 232{ 233 u32 rnd; 234 do { 235 get_random_bytes(&rnd, sizeof(rnd)); 236 } while (rnd == 0); 237 spin_lock_bh(&inetsw_lock); 238 if (!inet_ehash_secret) 239 inet_ehash_secret = rnd; 240 spin_unlock_bh(&inetsw_lock); 241} 242EXPORT_SYMBOL(build_ehash_secret); 243 244static inline int inet_netns_ok(struct net *net, int protocol) 245{ 246 int hash; 247 struct net_protocol *ipprot; 248 249 if (net == &init_net) 250 return 1; 251 252 hash = protocol & (MAX_INET_PROTOS - 1); 253 ipprot = rcu_dereference(inet_protos[hash]); 254 255 if (ipprot == NULL) 256 /* raw IP is OK */ 257 return 1; 258 return ipprot->netns_ok; 259} 260 261/* 262 * Create an inet socket. 263 */ 264 265static int inet_create(struct net *net, struct socket *sock, int protocol) 266{ 267 struct sock *sk; 268 struct list_head *p; 269 struct inet_protosw *answer; 270 struct inet_sock *inet; 271 struct proto *answer_prot; 272 unsigned char answer_flags; 273 char answer_no_check; 274 int try_loading_module = 0; 275 int err; 276 277 if (sock->type != SOCK_RAW && 278 sock->type != SOCK_DGRAM && 279 !inet_ehash_secret) 280 build_ehash_secret(); 281 282 sock->state = SS_UNCONNECTED; 283 284 /* Look for the requested type/protocol pair. */ 285 answer = NULL; 286lookup_protocol: 287 err = -ESOCKTNOSUPPORT; 288 rcu_read_lock(); 289 list_for_each_rcu(p, &inetsw[sock->type]) { 290 answer = list_entry(p, struct inet_protosw, list); 291 292 /* Check the non-wild match. */ 293 if (protocol == answer->protocol) { 294 if (protocol != IPPROTO_IP) 295 break; 296 } else { 297 /* Check for the two wild cases. */ 298 if (IPPROTO_IP == protocol) { 299 protocol = answer->protocol; 300 break; 301 } 302 if (IPPROTO_IP == answer->protocol) 303 break; 304 } 305 err = -EPROTONOSUPPORT; 306 answer = NULL; 307 } 308 309 if (unlikely(answer == NULL)) { 310 if (try_loading_module < 2) { 311 rcu_read_unlock(); 312 /* 313 * Be more specific, e.g. net-pf-2-proto-132-type-1 314 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM) 315 */ 316 if (++try_loading_module == 1) 317 request_module("net-pf-%d-proto-%d-type-%d", 318 PF_INET, protocol, sock->type); 319 /* 320 * Fall back to generic, e.g. net-pf-2-proto-132 321 * (net-pf-PF_INET-proto-IPPROTO_SCTP) 322 */ 323 else 324 request_module("net-pf-%d-proto-%d", 325 PF_INET, protocol); 326 goto lookup_protocol; 327 } else 328 goto out_rcu_unlock; 329 } 330 331 err = -EPERM; 332 if (answer->capability > 0 && !capable(answer->capability)) 333 goto out_rcu_unlock; 334 335 err = -EAFNOSUPPORT; 336 if (!inet_netns_ok(net, protocol)) 337 goto out_rcu_unlock; 338 339 sock->ops = answer->ops; 340 answer_prot = answer->prot; 341 answer_no_check = answer->no_check; 342 answer_flags = answer->flags; 343 rcu_read_unlock(); 344 345 BUG_TRAP(answer_prot->slab != NULL); 346 347 err = -ENOBUFS; 348 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot); 349 if (sk == NULL) 350 goto out; 351 352 err = 0; 353 sk->sk_no_check = answer_no_check; 354 if (INET_PROTOSW_REUSE & answer_flags) 355 sk->sk_reuse = 1; 356 357 inet = inet_sk(sk); 358 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0; 359 360 if (SOCK_RAW == sock->type) { 361 inet->num = protocol; 362 if (IPPROTO_RAW == protocol) 363 inet->hdrincl = 1; 364 } 365 366 if (ipv4_config.no_pmtu_disc) 367 inet->pmtudisc = IP_PMTUDISC_DONT; 368 else 369 inet->pmtudisc = IP_PMTUDISC_WANT; 370 371 inet->id = 0; 372 373 sock_init_data(sock, sk); 374 375 sk->sk_destruct = inet_sock_destruct; 376 sk->sk_family = PF_INET; 377 sk->sk_protocol = protocol; 378 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv; 379 380 inet->uc_ttl = -1; 381 inet->mc_loop = 1; 382 inet->mc_ttl = 1; 383 inet->mc_index = 0; 384 inet->mc_list = NULL; 385 386 sk_refcnt_debug_inc(sk); 387 388 if (inet->num) { 389 /* It assumes that any protocol which allows 390 * the user to assign a number at socket 391 * creation time automatically 392 * shares. 393 */ 394 inet->sport = htons(inet->num); 395 /* Add to protocol hash chains. */ 396 sk->sk_prot->hash(sk); 397 } 398 399 if (sk->sk_prot->init) { 400 err = sk->sk_prot->init(sk); 401 if (err) 402 sk_common_release(sk); 403 } 404out: 405 return err; 406out_rcu_unlock: 407 rcu_read_unlock(); 408 goto out; 409} 410 411 412/* 413 * The peer socket should always be NULL (or else). When we call this 414 * function we are destroying the object and from then on nobody 415 * should refer to it. 416 */ 417int inet_release(struct socket *sock) 418{ 419 struct sock *sk = sock->sk; 420 421 if (sk) { 422 long timeout; 423 424 /* Applications forget to leave groups before exiting */ 425 ip_mc_drop_socket(sk); 426 427 /* If linger is set, we don't return until the close 428 * is complete. Otherwise we return immediately. The 429 * actually closing is done the same either way. 430 * 431 * If the close is due to the process exiting, we never 432 * linger.. 433 */ 434 timeout = 0; 435 if (sock_flag(sk, SOCK_LINGER) && 436 !(current->flags & PF_EXITING)) 437 timeout = sk->sk_lingertime; 438 sock->sk = NULL; 439 sk->sk_prot->close(sk, timeout); 440 } 441 return 0; 442} 443 444/* It is off by default, see below. */ 445int sysctl_ip_nonlocal_bind __read_mostly; 446 447int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) 448{ 449 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr; 450 struct sock *sk = sock->sk; 451 struct inet_sock *inet = inet_sk(sk); 452 unsigned short snum; 453 int chk_addr_ret; 454 int err; 455 456 /* If the socket has its own bind function then use it. (RAW) */ 457 if (sk->sk_prot->bind) { 458 err = sk->sk_prot->bind(sk, uaddr, addr_len); 459 goto out; 460 } 461 err = -EINVAL; 462 if (addr_len < sizeof(struct sockaddr_in)) 463 goto out; 464 465 chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr); 466 467 /* Not specified by any standard per-se, however it breaks too 468 * many applications when removed. It is unfortunate since 469 * allowing applications to make a non-local bind solves 470 * several problems with systems using dynamic addressing. 471 * (ie. your servers still start up even if your ISDN link 472 * is temporarily down) 473 */ 474 err = -EADDRNOTAVAIL; 475 if (!sysctl_ip_nonlocal_bind && 476 !inet->freebind && 477 addr->sin_addr.s_addr != htonl(INADDR_ANY) && 478 chk_addr_ret != RTN_LOCAL && 479 chk_addr_ret != RTN_MULTICAST && 480 chk_addr_ret != RTN_BROADCAST) 481 goto out; 482 483 snum = ntohs(addr->sin_port); 484 err = -EACCES; 485 if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE)) 486 goto out; 487 488 /* We keep a pair of addresses. rcv_saddr is the one 489 * used by hash lookups, and saddr is used for transmit. 490 * 491 * In the BSD API these are the same except where it 492 * would be illegal to use them (multicast/broadcast) in 493 * which case the sending device address is used. 494 */ 495 lock_sock(sk); 496 497 /* Check these errors (active socket, double bind). */ 498 err = -EINVAL; 499 if (sk->sk_state != TCP_CLOSE || inet->num) 500 goto out_release_sock; 501 502 inet->rcv_saddr = inet->saddr = addr->sin_addr.s_addr; 503 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST) 504 inet->saddr = 0; /* Use device */ 505 506 /* Make sure we are allowed to bind here. */ 507 if (sk->sk_prot->get_port(sk, snum)) { 508 inet->saddr = inet->rcv_saddr = 0; 509 err = -EADDRINUSE; 510 goto out_release_sock; 511 } 512 513 if (inet->rcv_saddr) 514 sk->sk_userlocks |= SOCK_BINDADDR_LOCK; 515 if (snum) 516 sk->sk_userlocks |= SOCK_BINDPORT_LOCK; 517 inet->sport = htons(inet->num); 518 inet->daddr = 0; 519 inet->dport = 0; 520 sk_dst_reset(sk); 521 err = 0; 522out_release_sock: 523 release_sock(sk); 524out: 525 return err; 526} 527 528int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr, 529 int addr_len, int flags) 530{ 531 struct sock *sk = sock->sk; 532 533 if (uaddr->sa_family == AF_UNSPEC) 534 return sk->sk_prot->disconnect(sk, flags); 535 536 if (!inet_sk(sk)->num && inet_autobind(sk)) 537 return -EAGAIN; 538 return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len); 539} 540 541static long inet_wait_for_connect(struct sock *sk, long timeo) 542{ 543 DEFINE_WAIT(wait); 544 545 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE); 546 547 /* Basic assumption: if someone sets sk->sk_err, he _must_ 548 * change state of the socket from TCP_SYN_*. 549 * Connect() does not allow to get error notifications 550 * without closing the socket. 551 */ 552 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 553 release_sock(sk); 554 timeo = schedule_timeout(timeo); 555 lock_sock(sk); 556 if (signal_pending(current) || !timeo) 557 break; 558 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE); 559 } 560 finish_wait(sk->sk_sleep, &wait); 561 return timeo; 562} 563 564/* 565 * Connect to a remote host. There is regrettably still a little 566 * TCP 'magic' in here. 567 */ 568int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr, 569 int addr_len, int flags) 570{ 571 struct sock *sk = sock->sk; 572 int err; 573 long timeo; 574 575 lock_sock(sk); 576 577 if (uaddr->sa_family == AF_UNSPEC) { 578 err = sk->sk_prot->disconnect(sk, flags); 579 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED; 580 goto out; 581 } 582 583 switch (sock->state) { 584 default: 585 err = -EINVAL; 586 goto out; 587 case SS_CONNECTED: 588 err = -EISCONN; 589 goto out; 590 case SS_CONNECTING: 591 err = -EALREADY; 592 /* Fall out of switch with err, set for this state */ 593 break; 594 case SS_UNCONNECTED: 595 err = -EISCONN; 596 if (sk->sk_state != TCP_CLOSE) 597 goto out; 598 599 err = sk->sk_prot->connect(sk, uaddr, addr_len); 600 if (err < 0) 601 goto out; 602 603 sock->state = SS_CONNECTING; 604 605 /* Just entered SS_CONNECTING state; the only 606 * difference is that return value in non-blocking 607 * case is EINPROGRESS, rather than EALREADY. 608 */ 609 err = -EINPROGRESS; 610 break; 611 } 612 613 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK); 614 615 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 616 /* Error code is set above */ 617 if (!timeo || !inet_wait_for_connect(sk, timeo)) 618 goto out; 619 620 err = sock_intr_errno(timeo); 621 if (signal_pending(current)) 622 goto out; 623 } 624 625 /* Connection was closed by RST, timeout, ICMP error 626 * or another process disconnected us. 627 */ 628 if (sk->sk_state == TCP_CLOSE) 629 goto sock_error; 630 631 /* sk->sk_err may be not zero now, if RECVERR was ordered by user 632 * and error was received after socket entered established state. 633 * Hence, it is handled normally after connect() return successfully. 634 */ 635 636 sock->state = SS_CONNECTED; 637 err = 0; 638out: 639 release_sock(sk); 640 return err; 641 642sock_error: 643 err = sock_error(sk) ? : -ECONNABORTED; 644 sock->state = SS_UNCONNECTED; 645 if (sk->sk_prot->disconnect(sk, flags)) 646 sock->state = SS_DISCONNECTING; 647 goto out; 648} 649 650/* 651 * Accept a pending connection. The TCP layer now gives BSD semantics. 652 */ 653 654int inet_accept(struct socket *sock, struct socket *newsock, int flags) 655{ 656 struct sock *sk1 = sock->sk; 657 int err = -EINVAL; 658 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err); 659 660 if (!sk2) 661 goto do_err; 662 663 lock_sock(sk2); 664 665 BUG_TRAP((1 << sk2->sk_state) & 666 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE)); 667 668 sock_graft(sk2, newsock); 669 670 newsock->state = SS_CONNECTED; 671 err = 0; 672 release_sock(sk2); 673do_err: 674 return err; 675} 676 677 678/* 679 * This does both peername and sockname. 680 */ 681int inet_getname(struct socket *sock, struct sockaddr *uaddr, 682 int *uaddr_len, int peer) 683{ 684 struct sock *sk = sock->sk; 685 struct inet_sock *inet = inet_sk(sk); 686 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr; 687 688 sin->sin_family = AF_INET; 689 if (peer) { 690 if (!inet->dport || 691 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) && 692 peer == 1)) 693 return -ENOTCONN; 694 sin->sin_port = inet->dport; 695 sin->sin_addr.s_addr = inet->daddr; 696 } else { 697 __be32 addr = inet->rcv_saddr; 698 if (!addr) 699 addr = inet->saddr; 700 sin->sin_port = inet->sport; 701 sin->sin_addr.s_addr = addr; 702 } 703 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 704 *uaddr_len = sizeof(*sin); 705 return 0; 706} 707 708int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg, 709 size_t size) 710{ 711 struct sock *sk = sock->sk; 712 713 /* We may need to bind the socket. */ 714 if (!inet_sk(sk)->num && inet_autobind(sk)) 715 return -EAGAIN; 716 717 return sk->sk_prot->sendmsg(iocb, sk, msg, size); 718} 719 720 721static ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags) 722{ 723 struct sock *sk = sock->sk; 724 725 /* We may need to bind the socket. */ 726 if (!inet_sk(sk)->num && inet_autobind(sk)) 727 return -EAGAIN; 728 729 if (sk->sk_prot->sendpage) 730 return sk->sk_prot->sendpage(sk, page, offset, size, flags); 731 return sock_no_sendpage(sock, page, offset, size, flags); 732} 733 734 735int inet_shutdown(struct socket *sock, int how) 736{ 737 struct sock *sk = sock->sk; 738 int err = 0; 739 740 /* This should really check to make sure 741 * the socket is a TCP socket. (WHY AC...) 742 */ 743 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and 744 1->2 bit 2 snds. 745 2->3 */ 746 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */ 747 return -EINVAL; 748 749 lock_sock(sk); 750 if (sock->state == SS_CONNECTING) { 751 if ((1 << sk->sk_state) & 752 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE)) 753 sock->state = SS_DISCONNECTING; 754 else 755 sock->state = SS_CONNECTED; 756 } 757 758 switch (sk->sk_state) { 759 case TCP_CLOSE: 760 err = -ENOTCONN; 761 /* Hack to wake up other listeners, who can poll for 762 POLLHUP, even on eg. unconnected UDP sockets -- RR */ 763 default: 764 sk->sk_shutdown |= how; 765 if (sk->sk_prot->shutdown) 766 sk->sk_prot->shutdown(sk, how); 767 break; 768 769 /* Remaining two branches are temporary solution for missing 770 * close() in multithreaded environment. It is _not_ a good idea, 771 * but we have no choice until close() is repaired at VFS level. 772 */ 773 case TCP_LISTEN: 774 if (!(how & RCV_SHUTDOWN)) 775 break; 776 /* Fall through */ 777 case TCP_SYN_SENT: 778 err = sk->sk_prot->disconnect(sk, O_NONBLOCK); 779 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED; 780 break; 781 } 782 783 /* Wake up anyone sleeping in poll. */ 784 sk->sk_state_change(sk); 785 release_sock(sk); 786 return err; 787} 788 789/* 790 * ioctl() calls you can issue on an INET socket. Most of these are 791 * device configuration and stuff and very rarely used. Some ioctls 792 * pass on to the socket itself. 793 * 794 * NOTE: I like the idea of a module for the config stuff. ie ifconfig 795 * loads the devconfigure module does its configuring and unloads it. 796 * There's a good 20K of config code hanging around the kernel. 797 */ 798 799int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 800{ 801 struct sock *sk = sock->sk; 802 int err = 0; 803 struct net *net = sock_net(sk); 804 805 switch (cmd) { 806 case SIOCGSTAMP: 807 err = sock_get_timestamp(sk, (struct timeval __user *)arg); 808 break; 809 case SIOCGSTAMPNS: 810 err = sock_get_timestampns(sk, (struct timespec __user *)arg); 811 break; 812 case SIOCADDRT: 813 case SIOCDELRT: 814 case SIOCRTMSG: 815 err = ip_rt_ioctl(net, cmd, (void __user *)arg); 816 break; 817 case SIOCDARP: 818 case SIOCGARP: 819 case SIOCSARP: 820 err = arp_ioctl(net, cmd, (void __user *)arg); 821 break; 822 case SIOCGIFADDR: 823 case SIOCSIFADDR: 824 case SIOCGIFBRDADDR: 825 case SIOCSIFBRDADDR: 826 case SIOCGIFNETMASK: 827 case SIOCSIFNETMASK: 828 case SIOCGIFDSTADDR: 829 case SIOCSIFDSTADDR: 830 case SIOCSIFPFLAGS: 831 case SIOCGIFPFLAGS: 832 case SIOCSIFFLAGS: 833 err = devinet_ioctl(net, cmd, (void __user *)arg); 834 break; 835 default: 836 if (sk->sk_prot->ioctl) 837 err = sk->sk_prot->ioctl(sk, cmd, arg); 838 else 839 err = -ENOIOCTLCMD; 840 break; 841 } 842 return err; 843} 844 845const struct proto_ops inet_stream_ops = { 846 .family = PF_INET, 847 .owner = THIS_MODULE, 848 .release = inet_release, 849 .bind = inet_bind, 850 .connect = inet_stream_connect, 851 .socketpair = sock_no_socketpair, 852 .accept = inet_accept, 853 .getname = inet_getname, 854 .poll = tcp_poll, 855 .ioctl = inet_ioctl, 856 .listen = inet_listen, 857 .shutdown = inet_shutdown, 858 .setsockopt = sock_common_setsockopt, 859 .getsockopt = sock_common_getsockopt, 860 .sendmsg = tcp_sendmsg, 861 .recvmsg = sock_common_recvmsg, 862 .mmap = sock_no_mmap, 863 .sendpage = tcp_sendpage, 864 .splice_read = tcp_splice_read, 865#ifdef CONFIG_COMPAT 866 .compat_setsockopt = compat_sock_common_setsockopt, 867 .compat_getsockopt = compat_sock_common_getsockopt, 868#endif 869}; 870 871const struct proto_ops inet_dgram_ops = { 872 .family = PF_INET, 873 .owner = THIS_MODULE, 874 .release = inet_release, 875 .bind = inet_bind, 876 .connect = inet_dgram_connect, 877 .socketpair = sock_no_socketpair, 878 .accept = sock_no_accept, 879 .getname = inet_getname, 880 .poll = udp_poll, 881 .ioctl = inet_ioctl, 882 .listen = sock_no_listen, 883 .shutdown = inet_shutdown, 884 .setsockopt = sock_common_setsockopt, 885 .getsockopt = sock_common_getsockopt, 886 .sendmsg = inet_sendmsg, 887 .recvmsg = sock_common_recvmsg, 888 .mmap = sock_no_mmap, 889 .sendpage = inet_sendpage, 890#ifdef CONFIG_COMPAT 891 .compat_setsockopt = compat_sock_common_setsockopt, 892 .compat_getsockopt = compat_sock_common_getsockopt, 893#endif 894}; 895 896/* 897 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without 898 * udp_poll 899 */ 900static const struct proto_ops inet_sockraw_ops = { 901 .family = PF_INET, 902 .owner = THIS_MODULE, 903 .release = inet_release, 904 .bind = inet_bind, 905 .connect = inet_dgram_connect, 906 .socketpair = sock_no_socketpair, 907 .accept = sock_no_accept, 908 .getname = inet_getname, 909 .poll = datagram_poll, 910 .ioctl = inet_ioctl, 911 .listen = sock_no_listen, 912 .shutdown = inet_shutdown, 913 .setsockopt = sock_common_setsockopt, 914 .getsockopt = sock_common_getsockopt, 915 .sendmsg = inet_sendmsg, 916 .recvmsg = sock_common_recvmsg, 917 .mmap = sock_no_mmap, 918 .sendpage = inet_sendpage, 919#ifdef CONFIG_COMPAT 920 .compat_setsockopt = compat_sock_common_setsockopt, 921 .compat_getsockopt = compat_sock_common_getsockopt, 922#endif 923}; 924 925static struct net_proto_family inet_family_ops = { 926 .family = PF_INET, 927 .create = inet_create, 928 .owner = THIS_MODULE, 929}; 930 931/* Upon startup we insert all the elements in inetsw_array[] into 932 * the linked list inetsw. 933 */ 934static struct inet_protosw inetsw_array[] = 935{ 936 { 937 .type = SOCK_STREAM, 938 .protocol = IPPROTO_TCP, 939 .prot = &tcp_prot, 940 .ops = &inet_stream_ops, 941 .capability = -1, 942 .no_check = 0, 943 .flags = INET_PROTOSW_PERMANENT | 944 INET_PROTOSW_ICSK, 945 }, 946 947 { 948 .type = SOCK_DGRAM, 949 .protocol = IPPROTO_UDP, 950 .prot = &udp_prot, 951 .ops = &inet_dgram_ops, 952 .capability = -1, 953 .no_check = UDP_CSUM_DEFAULT, 954 .flags = INET_PROTOSW_PERMANENT, 955 }, 956 957 958 { 959 .type = SOCK_RAW, 960 .protocol = IPPROTO_IP, /* wild card */ 961 .prot = &raw_prot, 962 .ops = &inet_sockraw_ops, 963 .capability = CAP_NET_RAW, 964 .no_check = UDP_CSUM_DEFAULT, 965 .flags = INET_PROTOSW_REUSE, 966 } 967}; 968 969#define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array) 970 971void inet_register_protosw(struct inet_protosw *p) 972{ 973 struct list_head *lh; 974 struct inet_protosw *answer; 975 int protocol = p->protocol; 976 struct list_head *last_perm; 977 978 spin_lock_bh(&inetsw_lock); 979 980 if (p->type >= SOCK_MAX) 981 goto out_illegal; 982 983 /* If we are trying to override a permanent protocol, bail. */ 984 answer = NULL; 985 last_perm = &inetsw[p->type]; 986 list_for_each(lh, &inetsw[p->type]) { 987 answer = list_entry(lh, struct inet_protosw, list); 988 989 /* Check only the non-wild match. */ 990 if (INET_PROTOSW_PERMANENT & answer->flags) { 991 if (protocol == answer->protocol) 992 break; 993 last_perm = lh; 994 } 995 996 answer = NULL; 997 } 998 if (answer) 999 goto out_permanent; 1000 1001 /* Add the new entry after the last permanent entry if any, so that 1002 * the new entry does not override a permanent entry when matched with 1003 * a wild-card protocol. But it is allowed to override any existing 1004 * non-permanent entry. This means that when we remove this entry, the 1005 * system automatically returns to the old behavior. 1006 */ 1007 list_add_rcu(&p->list, last_perm); 1008out: 1009 spin_unlock_bh(&inetsw_lock); 1010 1011 synchronize_net(); 1012 1013 return; 1014 1015out_permanent: 1016 printk(KERN_ERR "Attempt to override permanent protocol %d.\n", 1017 protocol); 1018 goto out; 1019 1020out_illegal: 1021 printk(KERN_ERR 1022 "Ignoring attempt to register invalid socket type %d.\n", 1023 p->type); 1024 goto out; 1025} 1026 1027void inet_unregister_protosw(struct inet_protosw *p) 1028{ 1029 if (INET_PROTOSW_PERMANENT & p->flags) { 1030 printk(KERN_ERR 1031 "Attempt to unregister permanent protocol %d.\n", 1032 p->protocol); 1033 } else { 1034 spin_lock_bh(&inetsw_lock); 1035 list_del_rcu(&p->list); 1036 spin_unlock_bh(&inetsw_lock); 1037 1038 synchronize_net(); 1039 } 1040} 1041 1042/* 1043 * Shall we try to damage output packets if routing dev changes? 1044 */ 1045 1046int sysctl_ip_dynaddr __read_mostly; 1047 1048static int inet_sk_reselect_saddr(struct sock *sk) 1049{ 1050 struct inet_sock *inet = inet_sk(sk); 1051 int err; 1052 struct rtable *rt; 1053 __be32 old_saddr = inet->saddr; 1054 __be32 new_saddr; 1055 __be32 daddr = inet->daddr; 1056 1057 if (inet->opt && inet->opt->srr) 1058 daddr = inet->opt->faddr; 1059 1060 /* Query new route. */ 1061 err = ip_route_connect(&rt, daddr, 0, 1062 RT_CONN_FLAGS(sk), 1063 sk->sk_bound_dev_if, 1064 sk->sk_protocol, 1065 inet->sport, inet->dport, sk, 0); 1066 if (err) 1067 return err; 1068 1069 sk_setup_caps(sk, &rt->u.dst); 1070 1071 new_saddr = rt->rt_src; 1072 1073 if (new_saddr == old_saddr) 1074 return 0; 1075 1076 if (sysctl_ip_dynaddr > 1) { 1077 printk(KERN_INFO "%s(): shifting inet->" 1078 "saddr from " NIPQUAD_FMT " to " NIPQUAD_FMT "\n", 1079 __func__, 1080 NIPQUAD(old_saddr), 1081 NIPQUAD(new_saddr)); 1082 } 1083 1084 inet->saddr = inet->rcv_saddr = new_saddr; 1085 1086 /* 1087 * XXX The only one ugly spot where we need to 1088 * XXX really change the sockets identity after 1089 * XXX it has entered the hashes. -DaveM 1090 * 1091 * Besides that, it does not check for connection 1092 * uniqueness. Wait for troubles. 1093 */ 1094 __sk_prot_rehash(sk); 1095 return 0; 1096} 1097 1098int inet_sk_rebuild_header(struct sock *sk) 1099{ 1100 struct inet_sock *inet = inet_sk(sk); 1101 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0); 1102 __be32 daddr; 1103 int err; 1104 1105 /* Route is OK, nothing to do. */ 1106 if (rt) 1107 return 0; 1108 1109 /* Reroute. */ 1110 daddr = inet->daddr; 1111 if (inet->opt && inet->opt->srr) 1112 daddr = inet->opt->faddr; 1113{ 1114 struct flowi fl = { 1115 .oif = sk->sk_bound_dev_if, 1116 .nl_u = { 1117 .ip4_u = { 1118 .daddr = daddr, 1119 .saddr = inet->saddr, 1120 .tos = RT_CONN_FLAGS(sk), 1121 }, 1122 }, 1123 .proto = sk->sk_protocol, 1124 .uli_u = { 1125 .ports = { 1126 .sport = inet->sport, 1127 .dport = inet->dport, 1128 }, 1129 }, 1130 }; 1131 1132 security_sk_classify_flow(sk, &fl); 1133 err = ip_route_output_flow(sock_net(sk), &rt, &fl, sk, 0); 1134} 1135 if (!err) 1136 sk_setup_caps(sk, &rt->u.dst); 1137 else { 1138 /* Routing failed... */ 1139 sk->sk_route_caps = 0; 1140 /* 1141 * Other protocols have to map its equivalent state to TCP_SYN_SENT. 1142 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme 1143 */ 1144 if (!sysctl_ip_dynaddr || 1145 sk->sk_state != TCP_SYN_SENT || 1146 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) || 1147 (err = inet_sk_reselect_saddr(sk)) != 0) 1148 sk->sk_err_soft = -err; 1149 } 1150 1151 return err; 1152} 1153 1154EXPORT_SYMBOL(inet_sk_rebuild_header); 1155 1156static int inet_gso_send_check(struct sk_buff *skb) 1157{ 1158 struct iphdr *iph; 1159 struct net_protocol *ops; 1160 int proto; 1161 int ihl; 1162 int err = -EINVAL; 1163 1164 if (unlikely(!pskb_may_pull(skb, sizeof(*iph)))) 1165 goto out; 1166 1167 iph = ip_hdr(skb); 1168 ihl = iph->ihl * 4; 1169 if (ihl < sizeof(*iph)) 1170 goto out; 1171 1172 if (unlikely(!pskb_may_pull(skb, ihl))) 1173 goto out; 1174 1175 __skb_pull(skb, ihl); 1176 skb_reset_transport_header(skb); 1177 iph = ip_hdr(skb); 1178 proto = iph->protocol & (MAX_INET_PROTOS - 1); 1179 err = -EPROTONOSUPPORT; 1180 1181 rcu_read_lock(); 1182 ops = rcu_dereference(inet_protos[proto]); 1183 if (likely(ops && ops->gso_send_check)) 1184 err = ops->gso_send_check(skb); 1185 rcu_read_unlock(); 1186 1187out: 1188 return err; 1189} 1190 1191static struct sk_buff *inet_gso_segment(struct sk_buff *skb, int features) 1192{ 1193 struct sk_buff *segs = ERR_PTR(-EINVAL); 1194 struct iphdr *iph; 1195 struct net_protocol *ops; 1196 int proto; 1197 int ihl; 1198 int id; 1199 1200 if (!(features & NETIF_F_V4_CSUM)) 1201 features &= ~NETIF_F_SG; 1202 1203 if (unlikely(skb_shinfo(skb)->gso_type & 1204 ~(SKB_GSO_TCPV4 | 1205 SKB_GSO_UDP | 1206 SKB_GSO_DODGY | 1207 SKB_GSO_TCP_ECN | 1208 0))) 1209 goto out; 1210 1211 if (unlikely(!pskb_may_pull(skb, sizeof(*iph)))) 1212 goto out; 1213 1214 iph = ip_hdr(skb); 1215 ihl = iph->ihl * 4; 1216 if (ihl < sizeof(*iph)) 1217 goto out; 1218 1219 if (unlikely(!pskb_may_pull(skb, ihl))) 1220 goto out; 1221 1222 __skb_pull(skb, ihl); 1223 skb_reset_transport_header(skb); 1224 iph = ip_hdr(skb); 1225 id = ntohs(iph->id); 1226 proto = iph->protocol & (MAX_INET_PROTOS - 1); 1227 segs = ERR_PTR(-EPROTONOSUPPORT); 1228 1229 rcu_read_lock(); 1230 ops = rcu_dereference(inet_protos[proto]); 1231 if (likely(ops && ops->gso_segment)) 1232 segs = ops->gso_segment(skb, features); 1233 rcu_read_unlock(); 1234 1235 if (!segs || IS_ERR(segs)) 1236 goto out; 1237 1238 skb = segs; 1239 do { 1240 iph = ip_hdr(skb); 1241 iph->id = htons(id++); 1242 iph->tot_len = htons(skb->len - skb->mac_len); 1243 iph->check = 0; 1244 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl); 1245 } while ((skb = skb->next)); 1246 1247out: 1248 return segs; 1249} 1250 1251int inet_ctl_sock_create(struct sock **sk, unsigned short family, 1252 unsigned short type, unsigned char protocol, 1253 struct net *net) 1254{ 1255 struct socket *sock; 1256 int rc = sock_create_kern(family, type, protocol, &sock); 1257 1258 if (rc == 0) { 1259 *sk = sock->sk; 1260 (*sk)->sk_allocation = GFP_ATOMIC; 1261 /* 1262 * Unhash it so that IP input processing does not even see it, 1263 * we do not wish this socket to see incoming packets. 1264 */ 1265 (*sk)->sk_prot->unhash(*sk); 1266 1267 sk_change_net(*sk, net); 1268 } 1269 return rc; 1270} 1271 1272EXPORT_SYMBOL_GPL(inet_ctl_sock_create); 1273 1274unsigned long snmp_fold_field(void *mib[], int offt) 1275{ 1276 unsigned long res = 0; 1277 int i; 1278 1279 for_each_possible_cpu(i) { 1280 res += *(((unsigned long *) per_cpu_ptr(mib[0], i)) + offt); 1281 res += *(((unsigned long *) per_cpu_ptr(mib[1], i)) + offt); 1282 } 1283 return res; 1284} 1285EXPORT_SYMBOL_GPL(snmp_fold_field); 1286 1287int snmp_mib_init(void *ptr[2], size_t mibsize) 1288{ 1289 BUG_ON(ptr == NULL); 1290 ptr[0] = __alloc_percpu(mibsize); 1291 if (!ptr[0]) 1292 goto err0; 1293 ptr[1] = __alloc_percpu(mibsize); 1294 if (!ptr[1]) 1295 goto err1; 1296 return 0; 1297err1: 1298 free_percpu(ptr[0]); 1299 ptr[0] = NULL; 1300err0: 1301 return -ENOMEM; 1302} 1303EXPORT_SYMBOL_GPL(snmp_mib_init); 1304 1305void snmp_mib_free(void *ptr[2]) 1306{ 1307 BUG_ON(ptr == NULL); 1308 free_percpu(ptr[0]); 1309 free_percpu(ptr[1]); 1310 ptr[0] = ptr[1] = NULL; 1311} 1312EXPORT_SYMBOL_GPL(snmp_mib_free); 1313 1314#ifdef CONFIG_IP_MULTICAST 1315static struct net_protocol igmp_protocol = { 1316 .handler = igmp_rcv, 1317}; 1318#endif 1319 1320static struct net_protocol tcp_protocol = { 1321 .handler = tcp_v4_rcv, 1322 .err_handler = tcp_v4_err, 1323 .gso_send_check = tcp_v4_gso_send_check, 1324 .gso_segment = tcp_tso_segment, 1325 .no_policy = 1, 1326 .netns_ok = 1, 1327}; 1328 1329static struct net_protocol udp_protocol = { 1330 .handler = udp_rcv, 1331 .err_handler = udp_err, 1332 .no_policy = 1, 1333 .netns_ok = 1, 1334}; 1335 1336static struct net_protocol icmp_protocol = { 1337 .handler = icmp_rcv, 1338 .no_policy = 1, 1339 .netns_ok = 1, 1340}; 1341 1342static int __init init_ipv4_mibs(void) 1343{ 1344 if (snmp_mib_init((void **)net_statistics, 1345 sizeof(struct linux_mib)) < 0) 1346 goto err_net_mib; 1347 if (snmp_mib_init((void **)ip_statistics, 1348 sizeof(struct ipstats_mib)) < 0) 1349 goto err_ip_mib; 1350 if (snmp_mib_init((void **)icmp_statistics, 1351 sizeof(struct icmp_mib)) < 0) 1352 goto err_icmp_mib; 1353 if (snmp_mib_init((void **)icmpmsg_statistics, 1354 sizeof(struct icmpmsg_mib)) < 0) 1355 goto err_icmpmsg_mib; 1356 if (snmp_mib_init((void **)tcp_statistics, 1357 sizeof(struct tcp_mib)) < 0) 1358 goto err_tcp_mib; 1359 if (snmp_mib_init((void **)udp_statistics, 1360 sizeof(struct udp_mib)) < 0) 1361 goto err_udp_mib; 1362 if (snmp_mib_init((void **)udplite_statistics, 1363 sizeof(struct udp_mib)) < 0) 1364 goto err_udplite_mib; 1365 1366 tcp_mib_init(&init_net); 1367 1368 return 0; 1369 1370err_udplite_mib: 1371 snmp_mib_free((void **)udp_statistics); 1372err_udp_mib: 1373 snmp_mib_free((void **)tcp_statistics); 1374err_tcp_mib: 1375 snmp_mib_free((void **)icmpmsg_statistics); 1376err_icmpmsg_mib: 1377 snmp_mib_free((void **)icmp_statistics); 1378err_icmp_mib: 1379 snmp_mib_free((void **)ip_statistics); 1380err_ip_mib: 1381 snmp_mib_free((void **)net_statistics); 1382err_net_mib: 1383 return -ENOMEM; 1384} 1385 1386static int ipv4_proc_init(void); 1387 1388/* 1389 * IP protocol layer initialiser 1390 */ 1391 1392static struct packet_type ip_packet_type = { 1393 .type = __constant_htons(ETH_P_IP), 1394 .func = ip_rcv, 1395 .gso_send_check = inet_gso_send_check, 1396 .gso_segment = inet_gso_segment, 1397}; 1398 1399static int __init inet_init(void) 1400{ 1401 struct sk_buff *dummy_skb; 1402 struct inet_protosw *q; 1403 struct list_head *r; 1404 int rc = -EINVAL; 1405 1406 BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb)); 1407 1408 rc = proto_register(&tcp_prot, 1); 1409 if (rc) 1410 goto out; 1411 1412 rc = proto_register(&udp_prot, 1); 1413 if (rc) 1414 goto out_unregister_tcp_proto; 1415 1416 rc = proto_register(&raw_prot, 1); 1417 if (rc) 1418 goto out_unregister_udp_proto; 1419 1420 /* 1421 * Tell SOCKET that we are alive... 1422 */ 1423 1424 (void)sock_register(&inet_family_ops); 1425 1426 /* 1427 * Add all the base protocols. 1428 */ 1429 1430 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0) 1431 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n"); 1432 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0) 1433 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n"); 1434 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0) 1435 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n"); 1436#ifdef CONFIG_IP_MULTICAST 1437 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0) 1438 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n"); 1439#endif 1440 1441 /* Register the socket-side information for inet_create. */ 1442 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r) 1443 INIT_LIST_HEAD(r); 1444 1445 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q) 1446 inet_register_protosw(q); 1447 1448 /* 1449 * Set the ARP module up 1450 */ 1451 1452 arp_init(); 1453 1454 /* 1455 * Set the IP module up 1456 */ 1457 1458 ip_init(); 1459 1460 tcp_v4_init(); 1461 1462 /* Setup TCP slab cache for open requests. */ 1463 tcp_init(); 1464 1465 /* Setup UDP memory threshold */ 1466 udp_init(); 1467 1468 /* Add UDP-Lite (RFC 3828) */ 1469 udplite4_register(); 1470 1471 /* 1472 * Set the ICMP layer up 1473 */ 1474 1475 if (icmp_init() < 0) 1476 panic("Failed to create the ICMP control socket.\n"); 1477 1478 /* 1479 * Initialise the multicast router 1480 */ 1481#if defined(CONFIG_IP_MROUTE) 1482 if (ip_mr_init()) 1483 printk(KERN_CRIT "inet_init: Cannot init ipv4 mroute\n"); 1484#endif 1485 /* 1486 * Initialise per-cpu ipv4 mibs 1487 */ 1488 1489 if (init_ipv4_mibs()) 1490 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n"); 1491 1492 ipv4_proc_init(); 1493 1494 ipfrag_init(); 1495 1496 dev_add_pack(&ip_packet_type); 1497 1498 rc = 0; 1499out: 1500 return rc; 1501out_unregister_udp_proto: 1502 proto_unregister(&udp_prot); 1503out_unregister_tcp_proto: 1504 proto_unregister(&tcp_prot); 1505 goto out; 1506} 1507 1508fs_initcall(inet_init); 1509 1510/* ------------------------------------------------------------------------ */ 1511 1512#ifdef CONFIG_PROC_FS 1513static int __init ipv4_proc_init(void) 1514{ 1515 int rc = 0; 1516 1517 if (raw_proc_init()) 1518 goto out_raw; 1519 if (tcp4_proc_init()) 1520 goto out_tcp; 1521 if (udp4_proc_init()) 1522 goto out_udp; 1523 if (ip_misc_proc_init()) 1524 goto out_misc; 1525out: 1526 return rc; 1527out_misc: 1528 udp4_proc_exit(); 1529out_udp: 1530 tcp4_proc_exit(); 1531out_tcp: 1532 raw_proc_exit(); 1533out_raw: 1534 rc = -ENOMEM; 1535 goto out; 1536} 1537 1538#else /* CONFIG_PROC_FS */ 1539static int __init ipv4_proc_init(void) 1540{ 1541 return 0; 1542} 1543#endif /* CONFIG_PROC_FS */ 1544 1545MODULE_ALIAS_NETPROTO(PF_INET); 1546 1547EXPORT_SYMBOL(inet_accept); 1548EXPORT_SYMBOL(inet_bind); 1549EXPORT_SYMBOL(inet_dgram_connect); 1550EXPORT_SYMBOL(inet_dgram_ops); 1551EXPORT_SYMBOL(inet_getname); 1552EXPORT_SYMBOL(inet_ioctl); 1553EXPORT_SYMBOL(inet_listen); 1554EXPORT_SYMBOL(inet_register_protosw); 1555EXPORT_SYMBOL(inet_release); 1556EXPORT_SYMBOL(inet_sendmsg); 1557EXPORT_SYMBOL(inet_shutdown); 1558EXPORT_SYMBOL(inet_sock_destruct); 1559EXPORT_SYMBOL(inet_stream_connect); 1560EXPORT_SYMBOL(inet_stream_ops); 1561EXPORT_SYMBOL(inet_unregister_protosw); 1562EXPORT_SYMBOL(net_statistics); 1563EXPORT_SYMBOL(sysctl_ip_nonlocal_bind); 1564