1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * PF_INET protocol family socket handler. 8 * 9 * Authors: Ross Biro 10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 11 * Florian La Roche, <flla@stud.uni-sb.de> 12 * Alan Cox, <A.Cox@swansea.ac.uk> 13 * 14 * Changes (see also sock.c) 15 * 16 * piggy, 17 * Karl Knutson : Socket protocol table 18 * A.N.Kuznetsov : Socket death error in accept(). 19 * John Richardson : Fix non blocking error in connect() 20 * so sockets that fail to connect 21 * don't return -EINPROGRESS. 22 * Alan Cox : Asynchronous I/O support 23 * Alan Cox : Keep correct socket pointer on sock 24 * structures 25 * when accept() ed 26 * Alan Cox : Semantics of SO_LINGER aren't state 27 * moved to close when you look carefully. 28 * With this fixed and the accept bug fixed 29 * some RPC stuff seems happier. 30 * Niibe Yutaka : 4.4BSD style write async I/O 31 * Alan Cox, 32 * Tony Gale : Fixed reuse semantics. 33 * Alan Cox : bind() shouldn't abort existing but dead 34 * sockets. Stops FTP netin:.. I hope. 35 * Alan Cox : bind() works correctly for RAW sockets. 36 * Note that FreeBSD at least was broken 37 * in this respect so be careful with 38 * compatibility tests... 39 * Alan Cox : routing cache support 40 * Alan Cox : memzero the socket structure for 41 * compactness. 42 * Matt Day : nonblock connect error handler 43 * Alan Cox : Allow large numbers of pending sockets 44 * (eg for big web sites), but only if 45 * specifically application requested. 46 * Alan Cox : New buffering throughout IP. Used 47 * dumbly. 48 * Alan Cox : New buffering now used smartly. 49 * Alan Cox : BSD rather than common sense 50 * interpretation of listen. 51 * Germano Caronni : Assorted small races. 52 * Alan Cox : sendmsg/recvmsg basic support. 53 * Alan Cox : Only sendmsg/recvmsg now supported. 54 * Alan Cox : Locked down bind (see security list). 55 * Alan Cox : Loosened bind a little. 56 * Mike McLagan : ADD/DEL DLCI Ioctls 57 * Willy Konynenberg : Transparent proxying support. 58 * David S. Miller : New socket lookup architecture. 59 * Some other random speedups. 60 * Cyrus Durgin : Cleaned up file for kmod hacks. 61 * Andi Kleen : Fix inet_stream_connect TCP race. 62 */ 63 64 #define pr_fmt(fmt) "IPv4: " fmt 65 66 #include <linux/err.h> 67 #include <linux/errno.h> 68 #include <linux/types.h> 69 #include <linux/socket.h> 70 #include <linux/in.h> 71 #include <linux/kernel.h> 72 #include <linux/kmod.h> 73 #include <linux/sched.h> 74 #include <linux/timer.h> 75 #include <linux/string.h> 76 #include <linux/sockios.h> 77 #include <linux/net.h> 78 #include <linux/capability.h> 79 #include <linux/fcntl.h> 80 #include <linux/mm.h> 81 #include <linux/interrupt.h> 82 #include <linux/stat.h> 83 #include <linux/init.h> 84 #include <linux/poll.h> 85 #include <linux/netfilter_ipv4.h> 86 #include <linux/random.h> 87 #include <linux/slab.h> 88 89 #include <linux/uaccess.h> 90 91 #include <linux/inet.h> 92 #include <linux/igmp.h> 93 #include <linux/inetdevice.h> 94 #include <linux/netdevice.h> 95 #include <net/checksum.h> 96 #include <net/ip.h> 97 #include <net/protocol.h> 98 #include <net/arp.h> 99 #include <net/route.h> 100 #include <net/ip_fib.h> 101 #include <net/inet_connection_sock.h> 102 #include <net/tcp.h> 103 #include <net/udp.h> 104 #include <net/udplite.h> 105 #include <net/ping.h> 106 #include <linux/skbuff.h> 107 #include <net/sock.h> 108 #include <net/raw.h> 109 #include <net/icmp.h> 110 #include <net/inet_common.h> 111 #include <net/ip_tunnels.h> 112 #include <net/xfrm.h> 113 #include <net/net_namespace.h> 114 #include <net/secure_seq.h> 115 #ifdef CONFIG_IP_MROUTE 116 #include <linux/mroute.h> 117 #endif 118 #include <net/l3mdev.h> 119 120 #include <trace/events/sock.h> 121 122 /* The inetsw table contains everything that inet_create needs to 123 * build a new socket. 124 */ 125 static struct list_head inetsw[SOCK_MAX]; 126 static DEFINE_SPINLOCK(inetsw_lock); 127 128 /* New destruction routine */ 129 130 void inet_sock_destruct(struct sock *sk) 131 { 132 struct inet_sock *inet = inet_sk(sk); 133 134 __skb_queue_purge(&sk->sk_receive_queue); 135 if (sk->sk_rx_skb_cache) { 136 __kfree_skb(sk->sk_rx_skb_cache); 137 sk->sk_rx_skb_cache = NULL; 138 } 139 __skb_queue_purge(&sk->sk_error_queue); 140 141 sk_mem_reclaim(sk); 142 143 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) { 144 pr_err("Attempt to release TCP socket in state %d %p\n", 145 sk->sk_state, sk); 146 return; 147 } 148 if (!sock_flag(sk, SOCK_DEAD)) { 149 pr_err("Attempt to release alive inet socket %p\n", sk); 150 return; 151 } 152 153 WARN_ON(atomic_read(&sk->sk_rmem_alloc)); 154 WARN_ON(refcount_read(&sk->sk_wmem_alloc)); 155 WARN_ON(sk->sk_wmem_queued); 156 WARN_ON(sk->sk_forward_alloc); 157 158 kfree(rcu_dereference_protected(inet->inet_opt, 1)); 159 dst_release(rcu_dereference_protected(sk->sk_dst_cache, 1)); 160 dst_release(sk->sk_rx_dst); 161 sk_refcnt_debug_dec(sk); 162 } 163 EXPORT_SYMBOL(inet_sock_destruct); 164 165 /* 166 * The routines beyond this point handle the behaviour of an AF_INET 167 * socket object. Mostly it punts to the subprotocols of IP to do 168 * the work. 169 */ 170 171 /* 172 * Automatically bind an unbound socket. 173 */ 174 175 static int inet_autobind(struct sock *sk) 176 { 177 struct inet_sock *inet; 178 /* We may need to bind the socket. */ 179 lock_sock(sk); 180 inet = inet_sk(sk); 181 if (!inet->inet_num) { 182 if (sk->sk_prot->get_port(sk, 0)) { 183 release_sock(sk); 184 return -EAGAIN; 185 } 186 inet->inet_sport = htons(inet->inet_num); 187 } 188 release_sock(sk); 189 return 0; 190 } 191 192 /* 193 * Move a socket into listening state. 194 */ 195 int inet_listen(struct socket *sock, int backlog) 196 { 197 struct sock *sk = sock->sk; 198 unsigned char old_state; 199 int err, tcp_fastopen; 200 201 lock_sock(sk); 202 203 err = -EINVAL; 204 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM) 205 goto out; 206 207 old_state = sk->sk_state; 208 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN))) 209 goto out; 210 211 sk->sk_max_ack_backlog = backlog; 212 /* Really, if the socket is already in listen state 213 * we can only allow the backlog to be adjusted. 214 */ 215 if (old_state != TCP_LISTEN) { 216 /* Enable TFO w/o requiring TCP_FASTOPEN socket option. 217 * Note that only TCP sockets (SOCK_STREAM) will reach here. 218 * Also fastopen backlog may already been set via the option 219 * because the socket was in TCP_LISTEN state previously but 220 * was shutdown() rather than close(). 221 */ 222 tcp_fastopen = sock_net(sk)->ipv4.sysctl_tcp_fastopen; 223 if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) && 224 (tcp_fastopen & TFO_SERVER_ENABLE) && 225 !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) { 226 fastopen_queue_tune(sk, backlog); 227 tcp_fastopen_init_key_once(sock_net(sk)); 228 } 229 230 err = inet_csk_listen_start(sk, backlog); 231 if (err) 232 goto out; 233 tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL); 234 } 235 err = 0; 236 237 out: 238 release_sock(sk); 239 return err; 240 } 241 EXPORT_SYMBOL(inet_listen); 242 243 /* 244 * Create an inet socket. 245 */ 246 247 static int inet_create(struct net *net, struct socket *sock, int protocol, 248 int kern) 249 { 250 struct sock *sk; 251 struct inet_protosw *answer; 252 struct inet_sock *inet; 253 struct proto *answer_prot; 254 unsigned char answer_flags; 255 int try_loading_module = 0; 256 int err; 257 258 if (protocol < 0 || protocol >= IPPROTO_MAX) 259 return -EINVAL; 260 261 sock->state = SS_UNCONNECTED; 262 263 /* Look for the requested type/protocol pair. */ 264 lookup_protocol: 265 err = -ESOCKTNOSUPPORT; 266 rcu_read_lock(); 267 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) { 268 269 err = 0; 270 /* Check the non-wild match. */ 271 if (protocol == answer->protocol) { 272 if (protocol != IPPROTO_IP) 273 break; 274 } else { 275 /* Check for the two wild cases. */ 276 if (IPPROTO_IP == protocol) { 277 protocol = answer->protocol; 278 break; 279 } 280 if (IPPROTO_IP == answer->protocol) 281 break; 282 } 283 err = -EPROTONOSUPPORT; 284 } 285 286 if (unlikely(err)) { 287 if (try_loading_module < 2) { 288 rcu_read_unlock(); 289 /* 290 * Be more specific, e.g. net-pf-2-proto-132-type-1 291 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM) 292 */ 293 if (++try_loading_module == 1) 294 request_module("net-pf-%d-proto-%d-type-%d", 295 PF_INET, protocol, sock->type); 296 /* 297 * Fall back to generic, e.g. net-pf-2-proto-132 298 * (net-pf-PF_INET-proto-IPPROTO_SCTP) 299 */ 300 else 301 request_module("net-pf-%d-proto-%d", 302 PF_INET, protocol); 303 goto lookup_protocol; 304 } else 305 goto out_rcu_unlock; 306 } 307 308 err = -EPERM; 309 if (sock->type == SOCK_RAW && !kern && 310 !ns_capable(net->user_ns, CAP_NET_RAW)) 311 goto out_rcu_unlock; 312 313 sock->ops = answer->ops; 314 answer_prot = answer->prot; 315 answer_flags = answer->flags; 316 rcu_read_unlock(); 317 318 WARN_ON(!answer_prot->slab); 319 320 err = -ENOBUFS; 321 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern); 322 if (!sk) 323 goto out; 324 325 err = 0; 326 if (INET_PROTOSW_REUSE & answer_flags) 327 sk->sk_reuse = SK_CAN_REUSE; 328 329 inet = inet_sk(sk); 330 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0; 331 332 inet->nodefrag = 0; 333 334 if (SOCK_RAW == sock->type) { 335 inet->inet_num = protocol; 336 if (IPPROTO_RAW == protocol) 337 inet->hdrincl = 1; 338 } 339 340 if (net->ipv4.sysctl_ip_no_pmtu_disc) 341 inet->pmtudisc = IP_PMTUDISC_DONT; 342 else 343 inet->pmtudisc = IP_PMTUDISC_WANT; 344 345 inet->inet_id = 0; 346 347 sock_init_data(sock, sk); 348 349 sk->sk_destruct = inet_sock_destruct; 350 sk->sk_protocol = protocol; 351 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv; 352 353 inet->uc_ttl = -1; 354 inet->mc_loop = 1; 355 inet->mc_ttl = 1; 356 inet->mc_all = 1; 357 inet->mc_index = 0; 358 inet->mc_list = NULL; 359 inet->rcv_tos = 0; 360 361 sk_refcnt_debug_inc(sk); 362 363 if (inet->inet_num) { 364 /* It assumes that any protocol which allows 365 * the user to assign a number at socket 366 * creation time automatically 367 * shares. 368 */ 369 inet->inet_sport = htons(inet->inet_num); 370 /* Add to protocol hash chains. */ 371 err = sk->sk_prot->hash(sk); 372 if (err) { 373 sk_common_release(sk); 374 goto out; 375 } 376 } 377 378 if (sk->sk_prot->init) { 379 err = sk->sk_prot->init(sk); 380 if (err) { 381 sk_common_release(sk); 382 goto out; 383 } 384 } 385 386 if (!kern) { 387 err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk); 388 if (err) { 389 sk_common_release(sk); 390 goto out; 391 } 392 } 393 out: 394 return err; 395 out_rcu_unlock: 396 rcu_read_unlock(); 397 goto out; 398 } 399 400 401 /* 402 * The peer socket should always be NULL (or else). When we call this 403 * function we are destroying the object and from then on nobody 404 * should refer to it. 405 */ 406 int inet_release(struct socket *sock) 407 { 408 struct sock *sk = sock->sk; 409 410 if (sk) { 411 long timeout; 412 413 /* Applications forget to leave groups before exiting */ 414 ip_mc_drop_socket(sk); 415 416 /* If linger is set, we don't return until the close 417 * is complete. Otherwise we return immediately. The 418 * actually closing is done the same either way. 419 * 420 * If the close is due to the process exiting, we never 421 * linger.. 422 */ 423 timeout = 0; 424 if (sock_flag(sk, SOCK_LINGER) && 425 !(current->flags & PF_EXITING)) 426 timeout = sk->sk_lingertime; 427 sk->sk_prot->close(sk, timeout); 428 sock->sk = NULL; 429 } 430 return 0; 431 } 432 EXPORT_SYMBOL(inet_release); 433 434 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) 435 { 436 struct sock *sk = sock->sk; 437 int err; 438 439 /* If the socket has its own bind function then use it. (RAW) */ 440 if (sk->sk_prot->bind) { 441 return sk->sk_prot->bind(sk, uaddr, addr_len); 442 } 443 if (addr_len < sizeof(struct sockaddr_in)) 444 return -EINVAL; 445 446 /* BPF prog is run before any checks are done so that if the prog 447 * changes context in a wrong way it will be caught. 448 */ 449 err = BPF_CGROUP_RUN_PROG_INET4_BIND(sk, uaddr); 450 if (err) 451 return err; 452 453 return __inet_bind(sk, uaddr, addr_len, false, true); 454 } 455 EXPORT_SYMBOL(inet_bind); 456 457 int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len, 458 bool force_bind_address_no_port, bool with_lock) 459 { 460 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr; 461 struct inet_sock *inet = inet_sk(sk); 462 struct net *net = sock_net(sk); 463 unsigned short snum; 464 int chk_addr_ret; 465 u32 tb_id = RT_TABLE_LOCAL; 466 int err; 467 468 if (addr->sin_family != AF_INET) { 469 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET) 470 * only if s_addr is INADDR_ANY. 471 */ 472 err = -EAFNOSUPPORT; 473 if (addr->sin_family != AF_UNSPEC || 474 addr->sin_addr.s_addr != htonl(INADDR_ANY)) 475 goto out; 476 } 477 478 tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id; 479 chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id); 480 481 /* Not specified by any standard per-se, however it breaks too 482 * many applications when removed. It is unfortunate since 483 * allowing applications to make a non-local bind solves 484 * several problems with systems using dynamic addressing. 485 * (ie. your servers still start up even if your ISDN link 486 * is temporarily down) 487 */ 488 err = -EADDRNOTAVAIL; 489 if (!inet_can_nonlocal_bind(net, inet) && 490 addr->sin_addr.s_addr != htonl(INADDR_ANY) && 491 chk_addr_ret != RTN_LOCAL && 492 chk_addr_ret != RTN_MULTICAST && 493 chk_addr_ret != RTN_BROADCAST) 494 goto out; 495 496 snum = ntohs(addr->sin_port); 497 err = -EACCES; 498 if (snum && snum < inet_prot_sock(net) && 499 !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE)) 500 goto out; 501 502 /* We keep a pair of addresses. rcv_saddr is the one 503 * used by hash lookups, and saddr is used for transmit. 504 * 505 * In the BSD API these are the same except where it 506 * would be illegal to use them (multicast/broadcast) in 507 * which case the sending device address is used. 508 */ 509 if (with_lock) 510 lock_sock(sk); 511 512 /* Check these errors (active socket, double bind). */ 513 err = -EINVAL; 514 if (sk->sk_state != TCP_CLOSE || inet->inet_num) 515 goto out_release_sock; 516 517 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr; 518 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST) 519 inet->inet_saddr = 0; /* Use device */ 520 521 /* Make sure we are allowed to bind here. */ 522 if (snum || !(inet->bind_address_no_port || 523 force_bind_address_no_port)) { 524 if (sk->sk_prot->get_port(sk, snum)) { 525 inet->inet_saddr = inet->inet_rcv_saddr = 0; 526 err = -EADDRINUSE; 527 goto out_release_sock; 528 } 529 err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk); 530 if (err) { 531 inet->inet_saddr = inet->inet_rcv_saddr = 0; 532 goto out_release_sock; 533 } 534 } 535 536 if (inet->inet_rcv_saddr) 537 sk->sk_userlocks |= SOCK_BINDADDR_LOCK; 538 if (snum) 539 sk->sk_userlocks |= SOCK_BINDPORT_LOCK; 540 inet->inet_sport = htons(inet->inet_num); 541 inet->inet_daddr = 0; 542 inet->inet_dport = 0; 543 sk_dst_reset(sk); 544 err = 0; 545 out_release_sock: 546 if (with_lock) 547 release_sock(sk); 548 out: 549 return err; 550 } 551 552 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr, 553 int addr_len, int flags) 554 { 555 struct sock *sk = sock->sk; 556 int err; 557 558 if (addr_len < sizeof(uaddr->sa_family)) 559 return -EINVAL; 560 if (uaddr->sa_family == AF_UNSPEC) 561 return sk->sk_prot->disconnect(sk, flags); 562 563 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) { 564 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len); 565 if (err) 566 return err; 567 } 568 569 if (!inet_sk(sk)->inet_num && inet_autobind(sk)) 570 return -EAGAIN; 571 return sk->sk_prot->connect(sk, uaddr, addr_len); 572 } 573 EXPORT_SYMBOL(inet_dgram_connect); 574 575 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias) 576 { 577 DEFINE_WAIT_FUNC(wait, woken_wake_function); 578 579 add_wait_queue(sk_sleep(sk), &wait); 580 sk->sk_write_pending += writebias; 581 582 /* Basic assumption: if someone sets sk->sk_err, he _must_ 583 * change state of the socket from TCP_SYN_*. 584 * Connect() does not allow to get error notifications 585 * without closing the socket. 586 */ 587 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 588 release_sock(sk); 589 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo); 590 lock_sock(sk); 591 if (signal_pending(current) || !timeo) 592 break; 593 } 594 remove_wait_queue(sk_sleep(sk), &wait); 595 sk->sk_write_pending -= writebias; 596 return timeo; 597 } 598 599 /* 600 * Connect to a remote host. There is regrettably still a little 601 * TCP 'magic' in here. 602 */ 603 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr, 604 int addr_len, int flags, int is_sendmsg) 605 { 606 struct sock *sk = sock->sk; 607 int err; 608 long timeo; 609 610 /* 611 * uaddr can be NULL and addr_len can be 0 if: 612 * sk is a TCP fastopen active socket and 613 * TCP_FASTOPEN_CONNECT sockopt is set and 614 * we already have a valid cookie for this socket. 615 * In this case, user can call write() after connect(). 616 * write() will invoke tcp_sendmsg_fastopen() which calls 617 * __inet_stream_connect(). 618 */ 619 if (uaddr) { 620 if (addr_len < sizeof(uaddr->sa_family)) 621 return -EINVAL; 622 623 if (uaddr->sa_family == AF_UNSPEC) { 624 err = sk->sk_prot->disconnect(sk, flags); 625 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED; 626 goto out; 627 } 628 } 629 630 switch (sock->state) { 631 default: 632 err = -EINVAL; 633 goto out; 634 case SS_CONNECTED: 635 err = -EISCONN; 636 goto out; 637 case SS_CONNECTING: 638 if (inet_sk(sk)->defer_connect) 639 err = is_sendmsg ? -EINPROGRESS : -EISCONN; 640 else 641 err = -EALREADY; 642 /* Fall out of switch with err, set for this state */ 643 break; 644 case SS_UNCONNECTED: 645 err = -EISCONN; 646 if (sk->sk_state != TCP_CLOSE) 647 goto out; 648 649 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) { 650 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len); 651 if (err) 652 goto out; 653 } 654 655 err = sk->sk_prot->connect(sk, uaddr, addr_len); 656 if (err < 0) 657 goto out; 658 659 sock->state = SS_CONNECTING; 660 661 if (!err && inet_sk(sk)->defer_connect) 662 goto out; 663 664 /* Just entered SS_CONNECTING state; the only 665 * difference is that return value in non-blocking 666 * case is EINPROGRESS, rather than EALREADY. 667 */ 668 err = -EINPROGRESS; 669 break; 670 } 671 672 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK); 673 674 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 675 int writebias = (sk->sk_protocol == IPPROTO_TCP) && 676 tcp_sk(sk)->fastopen_req && 677 tcp_sk(sk)->fastopen_req->data ? 1 : 0; 678 679 /* Error code is set above */ 680 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias)) 681 goto out; 682 683 err = sock_intr_errno(timeo); 684 if (signal_pending(current)) 685 goto out; 686 } 687 688 /* Connection was closed by RST, timeout, ICMP error 689 * or another process disconnected us. 690 */ 691 if (sk->sk_state == TCP_CLOSE) 692 goto sock_error; 693 694 /* sk->sk_err may be not zero now, if RECVERR was ordered by user 695 * and error was received after socket entered established state. 696 * Hence, it is handled normally after connect() return successfully. 697 */ 698 699 sock->state = SS_CONNECTED; 700 err = 0; 701 out: 702 return err; 703 704 sock_error: 705 err = sock_error(sk) ? : -ECONNABORTED; 706 sock->state = SS_UNCONNECTED; 707 if (sk->sk_prot->disconnect(sk, flags)) 708 sock->state = SS_DISCONNECTING; 709 goto out; 710 } 711 EXPORT_SYMBOL(__inet_stream_connect); 712 713 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr, 714 int addr_len, int flags) 715 { 716 int err; 717 718 lock_sock(sock->sk); 719 err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0); 720 release_sock(sock->sk); 721 return err; 722 } 723 EXPORT_SYMBOL(inet_stream_connect); 724 725 /* 726 * Accept a pending connection. The TCP layer now gives BSD semantics. 727 */ 728 729 int inet_accept(struct socket *sock, struct socket *newsock, int flags, 730 bool kern) 731 { 732 struct sock *sk1 = sock->sk; 733 int err = -EINVAL; 734 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err, kern); 735 736 if (!sk2) 737 goto do_err; 738 739 lock_sock(sk2); 740 741 sock_rps_record_flow(sk2); 742 WARN_ON(!((1 << sk2->sk_state) & 743 (TCPF_ESTABLISHED | TCPF_SYN_RECV | 744 TCPF_CLOSE_WAIT | TCPF_CLOSE))); 745 746 sock_graft(sk2, newsock); 747 748 newsock->state = SS_CONNECTED; 749 err = 0; 750 release_sock(sk2); 751 do_err: 752 return err; 753 } 754 EXPORT_SYMBOL(inet_accept); 755 756 757 /* 758 * This does both peername and sockname. 759 */ 760 int inet_getname(struct socket *sock, struct sockaddr *uaddr, 761 int peer) 762 { 763 struct sock *sk = sock->sk; 764 struct inet_sock *inet = inet_sk(sk); 765 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr); 766 767 sin->sin_family = AF_INET; 768 if (peer) { 769 if (!inet->inet_dport || 770 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) && 771 peer == 1)) 772 return -ENOTCONN; 773 sin->sin_port = inet->inet_dport; 774 sin->sin_addr.s_addr = inet->inet_daddr; 775 } else { 776 __be32 addr = inet->inet_rcv_saddr; 777 if (!addr) 778 addr = inet->inet_saddr; 779 sin->sin_port = inet->inet_sport; 780 sin->sin_addr.s_addr = addr; 781 } 782 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 783 return sizeof(*sin); 784 } 785 EXPORT_SYMBOL(inet_getname); 786 787 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size) 788 { 789 struct sock *sk = sock->sk; 790 791 sock_rps_record_flow(sk); 792 793 /* We may need to bind the socket. */ 794 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind && 795 inet_autobind(sk)) 796 return -EAGAIN; 797 798 return sk->sk_prot->sendmsg(sk, msg, size); 799 } 800 EXPORT_SYMBOL(inet_sendmsg); 801 802 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, 803 size_t size, int flags) 804 { 805 struct sock *sk = sock->sk; 806 807 sock_rps_record_flow(sk); 808 809 /* We may need to bind the socket. */ 810 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind && 811 inet_autobind(sk)) 812 return -EAGAIN; 813 814 if (sk->sk_prot->sendpage) 815 return sk->sk_prot->sendpage(sk, page, offset, size, flags); 816 return sock_no_sendpage(sock, page, offset, size, flags); 817 } 818 EXPORT_SYMBOL(inet_sendpage); 819 820 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 821 int flags) 822 { 823 struct sock *sk = sock->sk; 824 int addr_len = 0; 825 int err; 826 827 if (likely(!(flags & MSG_ERRQUEUE))) 828 sock_rps_record_flow(sk); 829 830 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT, 831 flags & ~MSG_DONTWAIT, &addr_len); 832 if (err >= 0) 833 msg->msg_namelen = addr_len; 834 return err; 835 } 836 EXPORT_SYMBOL(inet_recvmsg); 837 838 int inet_shutdown(struct socket *sock, int how) 839 { 840 struct sock *sk = sock->sk; 841 int err = 0; 842 843 /* This should really check to make sure 844 * the socket is a TCP socket. (WHY AC...) 845 */ 846 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and 847 1->2 bit 2 snds. 848 2->3 */ 849 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */ 850 return -EINVAL; 851 852 lock_sock(sk); 853 if (sock->state == SS_CONNECTING) { 854 if ((1 << sk->sk_state) & 855 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE)) 856 sock->state = SS_DISCONNECTING; 857 else 858 sock->state = SS_CONNECTED; 859 } 860 861 switch (sk->sk_state) { 862 case TCP_CLOSE: 863 err = -ENOTCONN; 864 /* Hack to wake up other listeners, who can poll for 865 EPOLLHUP, even on eg. unconnected UDP sockets -- RR */ 866 /* fall through */ 867 default: 868 sk->sk_shutdown |= how; 869 if (sk->sk_prot->shutdown) 870 sk->sk_prot->shutdown(sk, how); 871 break; 872 873 /* Remaining two branches are temporary solution for missing 874 * close() in multithreaded environment. It is _not_ a good idea, 875 * but we have no choice until close() is repaired at VFS level. 876 */ 877 case TCP_LISTEN: 878 if (!(how & RCV_SHUTDOWN)) 879 break; 880 /* fall through */ 881 case TCP_SYN_SENT: 882 err = sk->sk_prot->disconnect(sk, O_NONBLOCK); 883 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED; 884 break; 885 } 886 887 /* Wake up anyone sleeping in poll. */ 888 sk->sk_state_change(sk); 889 release_sock(sk); 890 return err; 891 } 892 EXPORT_SYMBOL(inet_shutdown); 893 894 /* 895 * ioctl() calls you can issue on an INET socket. Most of these are 896 * device configuration and stuff and very rarely used. Some ioctls 897 * pass on to the socket itself. 898 * 899 * NOTE: I like the idea of a module for the config stuff. ie ifconfig 900 * loads the devconfigure module does its configuring and unloads it. 901 * There's a good 20K of config code hanging around the kernel. 902 */ 903 904 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 905 { 906 struct sock *sk = sock->sk; 907 int err = 0; 908 struct net *net = sock_net(sk); 909 void __user *p = (void __user *)arg; 910 struct ifreq ifr; 911 struct rtentry rt; 912 913 switch (cmd) { 914 case SIOCADDRT: 915 case SIOCDELRT: 916 if (copy_from_user(&rt, p, sizeof(struct rtentry))) 917 return -EFAULT; 918 err = ip_rt_ioctl(net, cmd, &rt); 919 break; 920 case SIOCRTMSG: 921 err = -EINVAL; 922 break; 923 case SIOCDARP: 924 case SIOCGARP: 925 case SIOCSARP: 926 err = arp_ioctl(net, cmd, (void __user *)arg); 927 break; 928 case SIOCGIFADDR: 929 case SIOCGIFBRDADDR: 930 case SIOCGIFNETMASK: 931 case SIOCGIFDSTADDR: 932 case SIOCGIFPFLAGS: 933 if (copy_from_user(&ifr, p, sizeof(struct ifreq))) 934 return -EFAULT; 935 err = devinet_ioctl(net, cmd, &ifr); 936 if (!err && copy_to_user(p, &ifr, sizeof(struct ifreq))) 937 err = -EFAULT; 938 break; 939 940 case SIOCSIFADDR: 941 case SIOCSIFBRDADDR: 942 case SIOCSIFNETMASK: 943 case SIOCSIFDSTADDR: 944 case SIOCSIFPFLAGS: 945 case SIOCSIFFLAGS: 946 if (copy_from_user(&ifr, p, sizeof(struct ifreq))) 947 return -EFAULT; 948 err = devinet_ioctl(net, cmd, &ifr); 949 break; 950 default: 951 if (sk->sk_prot->ioctl) 952 err = sk->sk_prot->ioctl(sk, cmd, arg); 953 else 954 err = -ENOIOCTLCMD; 955 break; 956 } 957 return err; 958 } 959 EXPORT_SYMBOL(inet_ioctl); 960 961 #ifdef CONFIG_COMPAT 962 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 963 { 964 struct sock *sk = sock->sk; 965 int err = -ENOIOCTLCMD; 966 967 if (sk->sk_prot->compat_ioctl) 968 err = sk->sk_prot->compat_ioctl(sk, cmd, arg); 969 970 return err; 971 } 972 #endif 973 974 const struct proto_ops inet_stream_ops = { 975 .family = PF_INET, 976 .owner = THIS_MODULE, 977 .release = inet_release, 978 .bind = inet_bind, 979 .connect = inet_stream_connect, 980 .socketpair = sock_no_socketpair, 981 .accept = inet_accept, 982 .getname = inet_getname, 983 .poll = tcp_poll, 984 .ioctl = inet_ioctl, 985 .gettstamp = sock_gettstamp, 986 .listen = inet_listen, 987 .shutdown = inet_shutdown, 988 .setsockopt = sock_common_setsockopt, 989 .getsockopt = sock_common_getsockopt, 990 .sendmsg = inet_sendmsg, 991 .recvmsg = inet_recvmsg, 992 #ifdef CONFIG_MMU 993 .mmap = tcp_mmap, 994 #endif 995 .sendpage = inet_sendpage, 996 .splice_read = tcp_splice_read, 997 .read_sock = tcp_read_sock, 998 .sendmsg_locked = tcp_sendmsg_locked, 999 .sendpage_locked = tcp_sendpage_locked, 1000 .peek_len = tcp_peek_len, 1001 #ifdef CONFIG_COMPAT 1002 .compat_setsockopt = compat_sock_common_setsockopt, 1003 .compat_getsockopt = compat_sock_common_getsockopt, 1004 .compat_ioctl = inet_compat_ioctl, 1005 #endif 1006 .set_rcvlowat = tcp_set_rcvlowat, 1007 }; 1008 EXPORT_SYMBOL(inet_stream_ops); 1009 1010 const struct proto_ops inet_dgram_ops = { 1011 .family = PF_INET, 1012 .owner = THIS_MODULE, 1013 .release = inet_release, 1014 .bind = inet_bind, 1015 .connect = inet_dgram_connect, 1016 .socketpair = sock_no_socketpair, 1017 .accept = sock_no_accept, 1018 .getname = inet_getname, 1019 .poll = udp_poll, 1020 .ioctl = inet_ioctl, 1021 .gettstamp = sock_gettstamp, 1022 .listen = sock_no_listen, 1023 .shutdown = inet_shutdown, 1024 .setsockopt = sock_common_setsockopt, 1025 .getsockopt = sock_common_getsockopt, 1026 .sendmsg = inet_sendmsg, 1027 .recvmsg = inet_recvmsg, 1028 .mmap = sock_no_mmap, 1029 .sendpage = inet_sendpage, 1030 .set_peek_off = sk_set_peek_off, 1031 #ifdef CONFIG_COMPAT 1032 .compat_setsockopt = compat_sock_common_setsockopt, 1033 .compat_getsockopt = compat_sock_common_getsockopt, 1034 .compat_ioctl = inet_compat_ioctl, 1035 #endif 1036 }; 1037 EXPORT_SYMBOL(inet_dgram_ops); 1038 1039 /* 1040 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without 1041 * udp_poll 1042 */ 1043 static const struct proto_ops inet_sockraw_ops = { 1044 .family = PF_INET, 1045 .owner = THIS_MODULE, 1046 .release = inet_release, 1047 .bind = inet_bind, 1048 .connect = inet_dgram_connect, 1049 .socketpair = sock_no_socketpair, 1050 .accept = sock_no_accept, 1051 .getname = inet_getname, 1052 .poll = datagram_poll, 1053 .ioctl = inet_ioctl, 1054 .gettstamp = sock_gettstamp, 1055 .listen = sock_no_listen, 1056 .shutdown = inet_shutdown, 1057 .setsockopt = sock_common_setsockopt, 1058 .getsockopt = sock_common_getsockopt, 1059 .sendmsg = inet_sendmsg, 1060 .recvmsg = inet_recvmsg, 1061 .mmap = sock_no_mmap, 1062 .sendpage = inet_sendpage, 1063 #ifdef CONFIG_COMPAT 1064 .compat_setsockopt = compat_sock_common_setsockopt, 1065 .compat_getsockopt = compat_sock_common_getsockopt, 1066 .compat_ioctl = inet_compat_ioctl, 1067 #endif 1068 }; 1069 1070 static const struct net_proto_family inet_family_ops = { 1071 .family = PF_INET, 1072 .create = inet_create, 1073 .owner = THIS_MODULE, 1074 }; 1075 1076 /* Upon startup we insert all the elements in inetsw_array[] into 1077 * the linked list inetsw. 1078 */ 1079 static struct inet_protosw inetsw_array[] = 1080 { 1081 { 1082 .type = SOCK_STREAM, 1083 .protocol = IPPROTO_TCP, 1084 .prot = &tcp_prot, 1085 .ops = &inet_stream_ops, 1086 .flags = INET_PROTOSW_PERMANENT | 1087 INET_PROTOSW_ICSK, 1088 }, 1089 1090 { 1091 .type = SOCK_DGRAM, 1092 .protocol = IPPROTO_UDP, 1093 .prot = &udp_prot, 1094 .ops = &inet_dgram_ops, 1095 .flags = INET_PROTOSW_PERMANENT, 1096 }, 1097 1098 { 1099 .type = SOCK_DGRAM, 1100 .protocol = IPPROTO_ICMP, 1101 .prot = &ping_prot, 1102 .ops = &inet_sockraw_ops, 1103 .flags = INET_PROTOSW_REUSE, 1104 }, 1105 1106 { 1107 .type = SOCK_RAW, 1108 .protocol = IPPROTO_IP, /* wild card */ 1109 .prot = &raw_prot, 1110 .ops = &inet_sockraw_ops, 1111 .flags = INET_PROTOSW_REUSE, 1112 } 1113 }; 1114 1115 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array) 1116 1117 void inet_register_protosw(struct inet_protosw *p) 1118 { 1119 struct list_head *lh; 1120 struct inet_protosw *answer; 1121 int protocol = p->protocol; 1122 struct list_head *last_perm; 1123 1124 spin_lock_bh(&inetsw_lock); 1125 1126 if (p->type >= SOCK_MAX) 1127 goto out_illegal; 1128 1129 /* If we are trying to override a permanent protocol, bail. */ 1130 last_perm = &inetsw[p->type]; 1131 list_for_each(lh, &inetsw[p->type]) { 1132 answer = list_entry(lh, struct inet_protosw, list); 1133 /* Check only the non-wild match. */ 1134 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0) 1135 break; 1136 if (protocol == answer->protocol) 1137 goto out_permanent; 1138 last_perm = lh; 1139 } 1140 1141 /* Add the new entry after the last permanent entry if any, so that 1142 * the new entry does not override a permanent entry when matched with 1143 * a wild-card protocol. But it is allowed to override any existing 1144 * non-permanent entry. This means that when we remove this entry, the 1145 * system automatically returns to the old behavior. 1146 */ 1147 list_add_rcu(&p->list, last_perm); 1148 out: 1149 spin_unlock_bh(&inetsw_lock); 1150 1151 return; 1152 1153 out_permanent: 1154 pr_err("Attempt to override permanent protocol %d\n", protocol); 1155 goto out; 1156 1157 out_illegal: 1158 pr_err("Ignoring attempt to register invalid socket type %d\n", 1159 p->type); 1160 goto out; 1161 } 1162 EXPORT_SYMBOL(inet_register_protosw); 1163 1164 void inet_unregister_protosw(struct inet_protosw *p) 1165 { 1166 if (INET_PROTOSW_PERMANENT & p->flags) { 1167 pr_err("Attempt to unregister permanent protocol %d\n", 1168 p->protocol); 1169 } else { 1170 spin_lock_bh(&inetsw_lock); 1171 list_del_rcu(&p->list); 1172 spin_unlock_bh(&inetsw_lock); 1173 1174 synchronize_net(); 1175 } 1176 } 1177 EXPORT_SYMBOL(inet_unregister_protosw); 1178 1179 static int inet_sk_reselect_saddr(struct sock *sk) 1180 { 1181 struct inet_sock *inet = inet_sk(sk); 1182 __be32 old_saddr = inet->inet_saddr; 1183 __be32 daddr = inet->inet_daddr; 1184 struct flowi4 *fl4; 1185 struct rtable *rt; 1186 __be32 new_saddr; 1187 struct ip_options_rcu *inet_opt; 1188 1189 inet_opt = rcu_dereference_protected(inet->inet_opt, 1190 lockdep_sock_is_held(sk)); 1191 if (inet_opt && inet_opt->opt.srr) 1192 daddr = inet_opt->opt.faddr; 1193 1194 /* Query new route. */ 1195 fl4 = &inet->cork.fl.u.ip4; 1196 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk), 1197 sk->sk_bound_dev_if, sk->sk_protocol, 1198 inet->inet_sport, inet->inet_dport, sk); 1199 if (IS_ERR(rt)) 1200 return PTR_ERR(rt); 1201 1202 sk_setup_caps(sk, &rt->dst); 1203 1204 new_saddr = fl4->saddr; 1205 1206 if (new_saddr == old_saddr) 1207 return 0; 1208 1209 if (sock_net(sk)->ipv4.sysctl_ip_dynaddr > 1) { 1210 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n", 1211 __func__, &old_saddr, &new_saddr); 1212 } 1213 1214 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr; 1215 1216 /* 1217 * XXX The only one ugly spot where we need to 1218 * XXX really change the sockets identity after 1219 * XXX it has entered the hashes. -DaveM 1220 * 1221 * Besides that, it does not check for connection 1222 * uniqueness. Wait for troubles. 1223 */ 1224 return __sk_prot_rehash(sk); 1225 } 1226 1227 int inet_sk_rebuild_header(struct sock *sk) 1228 { 1229 struct inet_sock *inet = inet_sk(sk); 1230 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0); 1231 __be32 daddr; 1232 struct ip_options_rcu *inet_opt; 1233 struct flowi4 *fl4; 1234 int err; 1235 1236 /* Route is OK, nothing to do. */ 1237 if (rt) 1238 return 0; 1239 1240 /* Reroute. */ 1241 rcu_read_lock(); 1242 inet_opt = rcu_dereference(inet->inet_opt); 1243 daddr = inet->inet_daddr; 1244 if (inet_opt && inet_opt->opt.srr) 1245 daddr = inet_opt->opt.faddr; 1246 rcu_read_unlock(); 1247 fl4 = &inet->cork.fl.u.ip4; 1248 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr, 1249 inet->inet_dport, inet->inet_sport, 1250 sk->sk_protocol, RT_CONN_FLAGS(sk), 1251 sk->sk_bound_dev_if); 1252 if (!IS_ERR(rt)) { 1253 err = 0; 1254 sk_setup_caps(sk, &rt->dst); 1255 } else { 1256 err = PTR_ERR(rt); 1257 1258 /* Routing failed... */ 1259 sk->sk_route_caps = 0; 1260 /* 1261 * Other protocols have to map its equivalent state to TCP_SYN_SENT. 1262 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme 1263 */ 1264 if (!sock_net(sk)->ipv4.sysctl_ip_dynaddr || 1265 sk->sk_state != TCP_SYN_SENT || 1266 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) || 1267 (err = inet_sk_reselect_saddr(sk)) != 0) 1268 sk->sk_err_soft = -err; 1269 } 1270 1271 return err; 1272 } 1273 EXPORT_SYMBOL(inet_sk_rebuild_header); 1274 1275 void inet_sk_set_state(struct sock *sk, int state) 1276 { 1277 trace_inet_sock_set_state(sk, sk->sk_state, state); 1278 sk->sk_state = state; 1279 } 1280 EXPORT_SYMBOL(inet_sk_set_state); 1281 1282 void inet_sk_state_store(struct sock *sk, int newstate) 1283 { 1284 trace_inet_sock_set_state(sk, sk->sk_state, newstate); 1285 smp_store_release(&sk->sk_state, newstate); 1286 } 1287 1288 struct sk_buff *inet_gso_segment(struct sk_buff *skb, 1289 netdev_features_t features) 1290 { 1291 bool udpfrag = false, fixedid = false, gso_partial, encap; 1292 struct sk_buff *segs = ERR_PTR(-EINVAL); 1293 const struct net_offload *ops; 1294 unsigned int offset = 0; 1295 struct iphdr *iph; 1296 int proto, tot_len; 1297 int nhoff; 1298 int ihl; 1299 int id; 1300 1301 skb_reset_network_header(skb); 1302 nhoff = skb_network_header(skb) - skb_mac_header(skb); 1303 if (unlikely(!pskb_may_pull(skb, sizeof(*iph)))) 1304 goto out; 1305 1306 iph = ip_hdr(skb); 1307 ihl = iph->ihl * 4; 1308 if (ihl < sizeof(*iph)) 1309 goto out; 1310 1311 id = ntohs(iph->id); 1312 proto = iph->protocol; 1313 1314 /* Warning: after this point, iph might be no longer valid */ 1315 if (unlikely(!pskb_may_pull(skb, ihl))) 1316 goto out; 1317 __skb_pull(skb, ihl); 1318 1319 encap = SKB_GSO_CB(skb)->encap_level > 0; 1320 if (encap) 1321 features &= skb->dev->hw_enc_features; 1322 SKB_GSO_CB(skb)->encap_level += ihl; 1323 1324 skb_reset_transport_header(skb); 1325 1326 segs = ERR_PTR(-EPROTONOSUPPORT); 1327 1328 if (!skb->encapsulation || encap) { 1329 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP); 1330 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID); 1331 1332 /* fixed ID is invalid if DF bit is not set */ 1333 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF))) 1334 goto out; 1335 } 1336 1337 ops = rcu_dereference(inet_offloads[proto]); 1338 if (likely(ops && ops->callbacks.gso_segment)) 1339 segs = ops->callbacks.gso_segment(skb, features); 1340 1341 if (IS_ERR_OR_NULL(segs)) 1342 goto out; 1343 1344 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL); 1345 1346 skb = segs; 1347 do { 1348 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff); 1349 if (udpfrag) { 1350 iph->frag_off = htons(offset >> 3); 1351 if (skb->next) 1352 iph->frag_off |= htons(IP_MF); 1353 offset += skb->len - nhoff - ihl; 1354 tot_len = skb->len - nhoff; 1355 } else if (skb_is_gso(skb)) { 1356 if (!fixedid) { 1357 iph->id = htons(id); 1358 id += skb_shinfo(skb)->gso_segs; 1359 } 1360 1361 if (gso_partial) 1362 tot_len = skb_shinfo(skb)->gso_size + 1363 SKB_GSO_CB(skb)->data_offset + 1364 skb->head - (unsigned char *)iph; 1365 else 1366 tot_len = skb->len - nhoff; 1367 } else { 1368 if (!fixedid) 1369 iph->id = htons(id++); 1370 tot_len = skb->len - nhoff; 1371 } 1372 iph->tot_len = htons(tot_len); 1373 ip_send_check(iph); 1374 if (encap) 1375 skb_reset_inner_headers(skb); 1376 skb->network_header = (u8 *)iph - skb->head; 1377 skb_reset_mac_len(skb); 1378 } while ((skb = skb->next)); 1379 1380 out: 1381 return segs; 1382 } 1383 EXPORT_SYMBOL(inet_gso_segment); 1384 1385 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb, 1386 netdev_features_t features) 1387 { 1388 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4)) 1389 return ERR_PTR(-EINVAL); 1390 1391 return inet_gso_segment(skb, features); 1392 } 1393 1394 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *, 1395 struct sk_buff *)); 1396 INDIRECT_CALLABLE_DECLARE(struct sk_buff *udp4_gro_receive(struct list_head *, 1397 struct sk_buff *)); 1398 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb) 1399 { 1400 const struct net_offload *ops; 1401 struct sk_buff *pp = NULL; 1402 const struct iphdr *iph; 1403 struct sk_buff *p; 1404 unsigned int hlen; 1405 unsigned int off; 1406 unsigned int id; 1407 int flush = 1; 1408 int proto; 1409 1410 off = skb_gro_offset(skb); 1411 hlen = off + sizeof(*iph); 1412 iph = skb_gro_header_fast(skb, off); 1413 if (skb_gro_header_hard(skb, hlen)) { 1414 iph = skb_gro_header_slow(skb, hlen, off); 1415 if (unlikely(!iph)) 1416 goto out; 1417 } 1418 1419 proto = iph->protocol; 1420 1421 rcu_read_lock(); 1422 ops = rcu_dereference(inet_offloads[proto]); 1423 if (!ops || !ops->callbacks.gro_receive) 1424 goto out_unlock; 1425 1426 if (*(u8 *)iph != 0x45) 1427 goto out_unlock; 1428 1429 if (ip_is_fragment(iph)) 1430 goto out_unlock; 1431 1432 if (unlikely(ip_fast_csum((u8 *)iph, 5))) 1433 goto out_unlock; 1434 1435 id = ntohl(*(__be32 *)&iph->id); 1436 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF)); 1437 id >>= 16; 1438 1439 list_for_each_entry(p, head, list) { 1440 struct iphdr *iph2; 1441 u16 flush_id; 1442 1443 if (!NAPI_GRO_CB(p)->same_flow) 1444 continue; 1445 1446 iph2 = (struct iphdr *)(p->data + off); 1447 /* The above works because, with the exception of the top 1448 * (inner most) layer, we only aggregate pkts with the same 1449 * hdr length so all the hdrs we'll need to verify will start 1450 * at the same offset. 1451 */ 1452 if ((iph->protocol ^ iph2->protocol) | 1453 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) | 1454 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) { 1455 NAPI_GRO_CB(p)->same_flow = 0; 1456 continue; 1457 } 1458 1459 /* All fields must match except length and checksum. */ 1460 NAPI_GRO_CB(p)->flush |= 1461 (iph->ttl ^ iph2->ttl) | 1462 (iph->tos ^ iph2->tos) | 1463 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF)); 1464 1465 NAPI_GRO_CB(p)->flush |= flush; 1466 1467 /* We need to store of the IP ID check to be included later 1468 * when we can verify that this packet does in fact belong 1469 * to a given flow. 1470 */ 1471 flush_id = (u16)(id - ntohs(iph2->id)); 1472 1473 /* This bit of code makes it much easier for us to identify 1474 * the cases where we are doing atomic vs non-atomic IP ID 1475 * checks. Specifically an atomic check can return IP ID 1476 * values 0 - 0xFFFF, while a non-atomic check can only 1477 * return 0 or 0xFFFF. 1478 */ 1479 if (!NAPI_GRO_CB(p)->is_atomic || 1480 !(iph->frag_off & htons(IP_DF))) { 1481 flush_id ^= NAPI_GRO_CB(p)->count; 1482 flush_id = flush_id ? 0xFFFF : 0; 1483 } 1484 1485 /* If the previous IP ID value was based on an atomic 1486 * datagram we can overwrite the value and ignore it. 1487 */ 1488 if (NAPI_GRO_CB(skb)->is_atomic) 1489 NAPI_GRO_CB(p)->flush_id = flush_id; 1490 else 1491 NAPI_GRO_CB(p)->flush_id |= flush_id; 1492 } 1493 1494 NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF)); 1495 NAPI_GRO_CB(skb)->flush |= flush; 1496 skb_set_network_header(skb, off); 1497 /* The above will be needed by the transport layer if there is one 1498 * immediately following this IP hdr. 1499 */ 1500 1501 /* Note : No need to call skb_gro_postpull_rcsum() here, 1502 * as we already checked checksum over ipv4 header was 0 1503 */ 1504 skb_gro_pull(skb, sizeof(*iph)); 1505 skb_set_transport_header(skb, skb_gro_offset(skb)); 1506 1507 pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive, 1508 ops->callbacks.gro_receive, head, skb); 1509 1510 out_unlock: 1511 rcu_read_unlock(); 1512 1513 out: 1514 skb_gro_flush_final(skb, pp, flush); 1515 1516 return pp; 1517 } 1518 EXPORT_SYMBOL(inet_gro_receive); 1519 1520 static struct sk_buff *ipip_gro_receive(struct list_head *head, 1521 struct sk_buff *skb) 1522 { 1523 if (NAPI_GRO_CB(skb)->encap_mark) { 1524 NAPI_GRO_CB(skb)->flush = 1; 1525 return NULL; 1526 } 1527 1528 NAPI_GRO_CB(skb)->encap_mark = 1; 1529 1530 return inet_gro_receive(head, skb); 1531 } 1532 1533 #define SECONDS_PER_DAY 86400 1534 1535 /* inet_current_timestamp - Return IP network timestamp 1536 * 1537 * Return milliseconds since midnight in network byte order. 1538 */ 1539 __be32 inet_current_timestamp(void) 1540 { 1541 u32 secs; 1542 u32 msecs; 1543 struct timespec64 ts; 1544 1545 ktime_get_real_ts64(&ts); 1546 1547 /* Get secs since midnight. */ 1548 (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs); 1549 /* Convert to msecs. */ 1550 msecs = secs * MSEC_PER_SEC; 1551 /* Convert nsec to msec. */ 1552 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC; 1553 1554 /* Convert to network byte order. */ 1555 return htonl(msecs); 1556 } 1557 EXPORT_SYMBOL(inet_current_timestamp); 1558 1559 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len) 1560 { 1561 if (sk->sk_family == AF_INET) 1562 return ip_recv_error(sk, msg, len, addr_len); 1563 #if IS_ENABLED(CONFIG_IPV6) 1564 if (sk->sk_family == AF_INET6) 1565 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len); 1566 #endif 1567 return -EINVAL; 1568 } 1569 1570 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *, int)); 1571 INDIRECT_CALLABLE_DECLARE(int udp4_gro_complete(struct sk_buff *, int)); 1572 int inet_gro_complete(struct sk_buff *skb, int nhoff) 1573 { 1574 __be16 newlen = htons(skb->len - nhoff); 1575 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff); 1576 const struct net_offload *ops; 1577 int proto = iph->protocol; 1578 int err = -ENOSYS; 1579 1580 if (skb->encapsulation) { 1581 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP)); 1582 skb_set_inner_network_header(skb, nhoff); 1583 } 1584 1585 csum_replace2(&iph->check, iph->tot_len, newlen); 1586 iph->tot_len = newlen; 1587 1588 rcu_read_lock(); 1589 ops = rcu_dereference(inet_offloads[proto]); 1590 if (WARN_ON(!ops || !ops->callbacks.gro_complete)) 1591 goto out_unlock; 1592 1593 /* Only need to add sizeof(*iph) to get to the next hdr below 1594 * because any hdr with option will have been flushed in 1595 * inet_gro_receive(). 1596 */ 1597 err = INDIRECT_CALL_2(ops->callbacks.gro_complete, 1598 tcp4_gro_complete, udp4_gro_complete, 1599 skb, nhoff + sizeof(*iph)); 1600 1601 out_unlock: 1602 rcu_read_unlock(); 1603 1604 return err; 1605 } 1606 EXPORT_SYMBOL(inet_gro_complete); 1607 1608 static int ipip_gro_complete(struct sk_buff *skb, int nhoff) 1609 { 1610 skb->encapsulation = 1; 1611 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4; 1612 return inet_gro_complete(skb, nhoff); 1613 } 1614 1615 int inet_ctl_sock_create(struct sock **sk, unsigned short family, 1616 unsigned short type, unsigned char protocol, 1617 struct net *net) 1618 { 1619 struct socket *sock; 1620 int rc = sock_create_kern(net, family, type, protocol, &sock); 1621 1622 if (rc == 0) { 1623 *sk = sock->sk; 1624 (*sk)->sk_allocation = GFP_ATOMIC; 1625 /* 1626 * Unhash it so that IP input processing does not even see it, 1627 * we do not wish this socket to see incoming packets. 1628 */ 1629 (*sk)->sk_prot->unhash(*sk); 1630 } 1631 return rc; 1632 } 1633 EXPORT_SYMBOL_GPL(inet_ctl_sock_create); 1634 1635 u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt) 1636 { 1637 return *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt); 1638 } 1639 EXPORT_SYMBOL_GPL(snmp_get_cpu_field); 1640 1641 unsigned long snmp_fold_field(void __percpu *mib, int offt) 1642 { 1643 unsigned long res = 0; 1644 int i; 1645 1646 for_each_possible_cpu(i) 1647 res += snmp_get_cpu_field(mib, i, offt); 1648 return res; 1649 } 1650 EXPORT_SYMBOL_GPL(snmp_fold_field); 1651 1652 #if BITS_PER_LONG==32 1653 1654 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt, 1655 size_t syncp_offset) 1656 { 1657 void *bhptr; 1658 struct u64_stats_sync *syncp; 1659 u64 v; 1660 unsigned int start; 1661 1662 bhptr = per_cpu_ptr(mib, cpu); 1663 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset); 1664 do { 1665 start = u64_stats_fetch_begin_irq(syncp); 1666 v = *(((u64 *)bhptr) + offt); 1667 } while (u64_stats_fetch_retry_irq(syncp, start)); 1668 1669 return v; 1670 } 1671 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64); 1672 1673 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset) 1674 { 1675 u64 res = 0; 1676 int cpu; 1677 1678 for_each_possible_cpu(cpu) { 1679 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset); 1680 } 1681 return res; 1682 } 1683 EXPORT_SYMBOL_GPL(snmp_fold_field64); 1684 #endif 1685 1686 #ifdef CONFIG_IP_MULTICAST 1687 static const struct net_protocol igmp_protocol = { 1688 .handler = igmp_rcv, 1689 .netns_ok = 1, 1690 }; 1691 #endif 1692 1693 /* thinking of making this const? Don't. 1694 * early_demux can change based on sysctl. 1695 */ 1696 static struct net_protocol tcp_protocol = { 1697 .early_demux = tcp_v4_early_demux, 1698 .early_demux_handler = tcp_v4_early_demux, 1699 .handler = tcp_v4_rcv, 1700 .err_handler = tcp_v4_err, 1701 .no_policy = 1, 1702 .netns_ok = 1, 1703 .icmp_strict_tag_validation = 1, 1704 }; 1705 1706 /* thinking of making this const? Don't. 1707 * early_demux can change based on sysctl. 1708 */ 1709 static struct net_protocol udp_protocol = { 1710 .early_demux = udp_v4_early_demux, 1711 .early_demux_handler = udp_v4_early_demux, 1712 .handler = udp_rcv, 1713 .err_handler = udp_err, 1714 .no_policy = 1, 1715 .netns_ok = 1, 1716 }; 1717 1718 static const struct net_protocol icmp_protocol = { 1719 .handler = icmp_rcv, 1720 .err_handler = icmp_err, 1721 .no_policy = 1, 1722 .netns_ok = 1, 1723 }; 1724 1725 static __net_init int ipv4_mib_init_net(struct net *net) 1726 { 1727 int i; 1728 1729 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib); 1730 if (!net->mib.tcp_statistics) 1731 goto err_tcp_mib; 1732 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib); 1733 if (!net->mib.ip_statistics) 1734 goto err_ip_mib; 1735 1736 for_each_possible_cpu(i) { 1737 struct ipstats_mib *af_inet_stats; 1738 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i); 1739 u64_stats_init(&af_inet_stats->syncp); 1740 } 1741 1742 net->mib.net_statistics = alloc_percpu(struct linux_mib); 1743 if (!net->mib.net_statistics) 1744 goto err_net_mib; 1745 net->mib.udp_statistics = alloc_percpu(struct udp_mib); 1746 if (!net->mib.udp_statistics) 1747 goto err_udp_mib; 1748 net->mib.udplite_statistics = alloc_percpu(struct udp_mib); 1749 if (!net->mib.udplite_statistics) 1750 goto err_udplite_mib; 1751 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib); 1752 if (!net->mib.icmp_statistics) 1753 goto err_icmp_mib; 1754 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib), 1755 GFP_KERNEL); 1756 if (!net->mib.icmpmsg_statistics) 1757 goto err_icmpmsg_mib; 1758 1759 tcp_mib_init(net); 1760 return 0; 1761 1762 err_icmpmsg_mib: 1763 free_percpu(net->mib.icmp_statistics); 1764 err_icmp_mib: 1765 free_percpu(net->mib.udplite_statistics); 1766 err_udplite_mib: 1767 free_percpu(net->mib.udp_statistics); 1768 err_udp_mib: 1769 free_percpu(net->mib.net_statistics); 1770 err_net_mib: 1771 free_percpu(net->mib.ip_statistics); 1772 err_ip_mib: 1773 free_percpu(net->mib.tcp_statistics); 1774 err_tcp_mib: 1775 return -ENOMEM; 1776 } 1777 1778 static __net_exit void ipv4_mib_exit_net(struct net *net) 1779 { 1780 kfree(net->mib.icmpmsg_statistics); 1781 free_percpu(net->mib.icmp_statistics); 1782 free_percpu(net->mib.udplite_statistics); 1783 free_percpu(net->mib.udp_statistics); 1784 free_percpu(net->mib.net_statistics); 1785 free_percpu(net->mib.ip_statistics); 1786 free_percpu(net->mib.tcp_statistics); 1787 } 1788 1789 static __net_initdata struct pernet_operations ipv4_mib_ops = { 1790 .init = ipv4_mib_init_net, 1791 .exit = ipv4_mib_exit_net, 1792 }; 1793 1794 static int __init init_ipv4_mibs(void) 1795 { 1796 return register_pernet_subsys(&ipv4_mib_ops); 1797 } 1798 1799 static __net_init int inet_init_net(struct net *net) 1800 { 1801 /* 1802 * Set defaults for local port range 1803 */ 1804 seqlock_init(&net->ipv4.ip_local_ports.lock); 1805 net->ipv4.ip_local_ports.range[0] = 32768; 1806 net->ipv4.ip_local_ports.range[1] = 60999; 1807 1808 seqlock_init(&net->ipv4.ping_group_range.lock); 1809 /* 1810 * Sane defaults - nobody may create ping sockets. 1811 * Boot scripts should set this to distro-specific group. 1812 */ 1813 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1); 1814 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0); 1815 1816 /* Default values for sysctl-controlled parameters. 1817 * We set them here, in case sysctl is not compiled. 1818 */ 1819 net->ipv4.sysctl_ip_default_ttl = IPDEFTTL; 1820 net->ipv4.sysctl_ip_fwd_update_priority = 1; 1821 net->ipv4.sysctl_ip_dynaddr = 0; 1822 net->ipv4.sysctl_ip_early_demux = 1; 1823 net->ipv4.sysctl_udp_early_demux = 1; 1824 net->ipv4.sysctl_tcp_early_demux = 1; 1825 #ifdef CONFIG_SYSCTL 1826 net->ipv4.sysctl_ip_prot_sock = PROT_SOCK; 1827 #endif 1828 1829 /* Some igmp sysctl, whose values are always used */ 1830 net->ipv4.sysctl_igmp_max_memberships = 20; 1831 net->ipv4.sysctl_igmp_max_msf = 10; 1832 /* IGMP reports for link-local multicast groups are enabled by default */ 1833 net->ipv4.sysctl_igmp_llm_reports = 1; 1834 net->ipv4.sysctl_igmp_qrv = 2; 1835 1836 return 0; 1837 } 1838 1839 static __net_exit void inet_exit_net(struct net *net) 1840 { 1841 } 1842 1843 static __net_initdata struct pernet_operations af_inet_ops = { 1844 .init = inet_init_net, 1845 .exit = inet_exit_net, 1846 }; 1847 1848 static int __init init_inet_pernet_ops(void) 1849 { 1850 return register_pernet_subsys(&af_inet_ops); 1851 } 1852 1853 static int ipv4_proc_init(void); 1854 1855 /* 1856 * IP protocol layer initialiser 1857 */ 1858 1859 static struct packet_offload ip_packet_offload __read_mostly = { 1860 .type = cpu_to_be16(ETH_P_IP), 1861 .callbacks = { 1862 .gso_segment = inet_gso_segment, 1863 .gro_receive = inet_gro_receive, 1864 .gro_complete = inet_gro_complete, 1865 }, 1866 }; 1867 1868 static const struct net_offload ipip_offload = { 1869 .callbacks = { 1870 .gso_segment = ipip_gso_segment, 1871 .gro_receive = ipip_gro_receive, 1872 .gro_complete = ipip_gro_complete, 1873 }, 1874 }; 1875 1876 static int __init ipip_offload_init(void) 1877 { 1878 return inet_add_offload(&ipip_offload, IPPROTO_IPIP); 1879 } 1880 1881 static int __init ipv4_offload_init(void) 1882 { 1883 /* 1884 * Add offloads 1885 */ 1886 if (udpv4_offload_init() < 0) 1887 pr_crit("%s: Cannot add UDP protocol offload\n", __func__); 1888 if (tcpv4_offload_init() < 0) 1889 pr_crit("%s: Cannot add TCP protocol offload\n", __func__); 1890 if (ipip_offload_init() < 0) 1891 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__); 1892 1893 dev_add_offload(&ip_packet_offload); 1894 return 0; 1895 } 1896 1897 fs_initcall(ipv4_offload_init); 1898 1899 static struct packet_type ip_packet_type __read_mostly = { 1900 .type = cpu_to_be16(ETH_P_IP), 1901 .func = ip_rcv, 1902 .list_func = ip_list_rcv, 1903 }; 1904 1905 static int __init inet_init(void) 1906 { 1907 struct inet_protosw *q; 1908 struct list_head *r; 1909 int rc = -EINVAL; 1910 1911 sock_skb_cb_check_size(sizeof(struct inet_skb_parm)); 1912 1913 rc = proto_register(&tcp_prot, 1); 1914 if (rc) 1915 goto out; 1916 1917 rc = proto_register(&udp_prot, 1); 1918 if (rc) 1919 goto out_unregister_tcp_proto; 1920 1921 rc = proto_register(&raw_prot, 1); 1922 if (rc) 1923 goto out_unregister_udp_proto; 1924 1925 rc = proto_register(&ping_prot, 1); 1926 if (rc) 1927 goto out_unregister_raw_proto; 1928 1929 /* 1930 * Tell SOCKET that we are alive... 1931 */ 1932 1933 (void)sock_register(&inet_family_ops); 1934 1935 #ifdef CONFIG_SYSCTL 1936 ip_static_sysctl_init(); 1937 #endif 1938 1939 /* 1940 * Add all the base protocols. 1941 */ 1942 1943 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0) 1944 pr_crit("%s: Cannot add ICMP protocol\n", __func__); 1945 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0) 1946 pr_crit("%s: Cannot add UDP protocol\n", __func__); 1947 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0) 1948 pr_crit("%s: Cannot add TCP protocol\n", __func__); 1949 #ifdef CONFIG_IP_MULTICAST 1950 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0) 1951 pr_crit("%s: Cannot add IGMP protocol\n", __func__); 1952 #endif 1953 1954 /* Register the socket-side information for inet_create. */ 1955 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r) 1956 INIT_LIST_HEAD(r); 1957 1958 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q) 1959 inet_register_protosw(q); 1960 1961 /* 1962 * Set the ARP module up 1963 */ 1964 1965 arp_init(); 1966 1967 /* 1968 * Set the IP module up 1969 */ 1970 1971 ip_init(); 1972 1973 /* Setup TCP slab cache for open requests. */ 1974 tcp_init(); 1975 1976 /* Setup UDP memory threshold */ 1977 udp_init(); 1978 1979 /* Add UDP-Lite (RFC 3828) */ 1980 udplite4_register(); 1981 1982 raw_init(); 1983 1984 ping_init(); 1985 1986 /* 1987 * Set the ICMP layer up 1988 */ 1989 1990 if (icmp_init() < 0) 1991 panic("Failed to create the ICMP control socket.\n"); 1992 1993 /* 1994 * Initialise the multicast router 1995 */ 1996 #if defined(CONFIG_IP_MROUTE) 1997 if (ip_mr_init()) 1998 pr_crit("%s: Cannot init ipv4 mroute\n", __func__); 1999 #endif 2000 2001 if (init_inet_pernet_ops()) 2002 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__); 2003 /* 2004 * Initialise per-cpu ipv4 mibs 2005 */ 2006 2007 if (init_ipv4_mibs()) 2008 pr_crit("%s: Cannot init ipv4 mibs\n", __func__); 2009 2010 ipv4_proc_init(); 2011 2012 ipfrag_init(); 2013 2014 dev_add_pack(&ip_packet_type); 2015 2016 ip_tunnel_core_init(); 2017 2018 rc = 0; 2019 out: 2020 return rc; 2021 out_unregister_raw_proto: 2022 proto_unregister(&raw_prot); 2023 out_unregister_udp_proto: 2024 proto_unregister(&udp_prot); 2025 out_unregister_tcp_proto: 2026 proto_unregister(&tcp_prot); 2027 goto out; 2028 } 2029 2030 fs_initcall(inet_init); 2031 2032 /* ------------------------------------------------------------------------ */ 2033 2034 #ifdef CONFIG_PROC_FS 2035 static int __init ipv4_proc_init(void) 2036 { 2037 int rc = 0; 2038 2039 if (raw_proc_init()) 2040 goto out_raw; 2041 if (tcp4_proc_init()) 2042 goto out_tcp; 2043 if (udp4_proc_init()) 2044 goto out_udp; 2045 if (ping_proc_init()) 2046 goto out_ping; 2047 if (ip_misc_proc_init()) 2048 goto out_misc; 2049 out: 2050 return rc; 2051 out_misc: 2052 ping_proc_exit(); 2053 out_ping: 2054 udp4_proc_exit(); 2055 out_udp: 2056 tcp4_proc_exit(); 2057 out_tcp: 2058 raw_proc_exit(); 2059 out_raw: 2060 rc = -ENOMEM; 2061 goto out; 2062 } 2063 2064 #else /* CONFIG_PROC_FS */ 2065 static int __init ipv4_proc_init(void) 2066 { 2067 return 0; 2068 } 2069 #endif /* CONFIG_PROC_FS */ 2070