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/gro.h> 103 #include <net/gso.h> 104 #include <net/tcp.h> 105 #include <net/psp.h> 106 #include <net/udp.h> 107 #include <net/udplite.h> 108 #include <net/ping.h> 109 #include <linux/skbuff.h> 110 #include <net/sock.h> 111 #include <net/raw.h> 112 #include <net/icmp.h> 113 #include <net/inet_common.h> 114 #include <net/ip_tunnels.h> 115 #include <net/xfrm.h> 116 #include <net/net_namespace.h> 117 #include <net/secure_seq.h> 118 #ifdef CONFIG_IP_MROUTE 119 #include <linux/mroute.h> 120 #endif 121 #include <net/l3mdev.h> 122 #include <net/compat.h> 123 #include <net/rps.h> 124 125 #include <trace/events/sock.h> 126 127 /* The inetsw table contains everything that inet_create needs to 128 * build a new socket. 129 */ 130 static struct list_head inetsw[SOCK_MAX]; 131 static DEFINE_SPINLOCK(inetsw_lock); 132 133 /* New destruction routine */ 134 135 void inet_sock_destruct(struct sock *sk) 136 { 137 struct inet_sock *inet = inet_sk(sk); 138 139 __skb_queue_purge(&sk->sk_receive_queue); 140 __skb_queue_purge(&sk->sk_error_queue); 141 142 sk_mem_reclaim_final(sk); 143 144 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) { 145 pr_err("Attempt to release TCP socket in state %d %p\n", 146 sk->sk_state, sk); 147 return; 148 } 149 if (!sock_flag(sk, SOCK_DEAD)) { 150 pr_err("Attempt to release alive inet socket %p\n", sk); 151 return; 152 } 153 154 WARN_ON_ONCE(atomic_read(&sk->sk_rmem_alloc)); 155 WARN_ON_ONCE(refcount_read(&sk->sk_wmem_alloc)); 156 WARN_ON_ONCE(sk->sk_wmem_queued); 157 WARN_ON_ONCE(sk->sk_forward_alloc); 158 159 kfree(rcu_dereference_protected(inet->inet_opt, 1)); 160 dst_release(rcu_dereference_protected(sk->sk_dst_cache, 1)); 161 dst_release(rcu_dereference_protected(sk->sk_rx_dst, 1)); 162 psp_sk_assoc_free(sk); 163 } 164 EXPORT_SYMBOL(inet_sock_destruct); 165 166 /* 167 * The routines beyond this point handle the behaviour of an AF_INET 168 * socket object. Mostly it punts to the subprotocols of IP to do 169 * the work. 170 */ 171 172 /* 173 * Automatically bind an unbound socket. 174 */ 175 176 static int inet_autobind(struct sock *sk) 177 { 178 struct inet_sock *inet; 179 /* We may need to bind the socket. */ 180 lock_sock(sk); 181 inet = inet_sk(sk); 182 if (!inet->inet_num) { 183 if (sk->sk_prot->get_port(sk, 0)) { 184 release_sock(sk); 185 return -EAGAIN; 186 } 187 inet->inet_sport = htons(inet->inet_num); 188 } 189 release_sock(sk); 190 return 0; 191 } 192 193 int __inet_listen_sk(struct sock *sk, int backlog) 194 { 195 unsigned char old_state = sk->sk_state; 196 int err, tcp_fastopen; 197 198 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN))) 199 return -EINVAL; 200 201 WRITE_ONCE(sk->sk_max_ack_backlog, backlog); 202 /* Really, if the socket is already in listen state 203 * we can only allow the backlog to be adjusted. 204 */ 205 if (old_state != TCP_LISTEN) { 206 /* Enable TFO w/o requiring TCP_FASTOPEN socket option. 207 * Note that only TCP sockets (SOCK_STREAM) will reach here. 208 * Also fastopen backlog may already been set via the option 209 * because the socket was in TCP_LISTEN state previously but 210 * was shutdown() rather than close(). 211 */ 212 tcp_fastopen = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fastopen); 213 if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) && 214 (tcp_fastopen & TFO_SERVER_ENABLE) && 215 !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) { 216 fastopen_queue_tune(sk, backlog); 217 tcp_fastopen_init_key_once(sock_net(sk)); 218 } 219 220 err = inet_csk_listen_start(sk); 221 if (err) 222 return err; 223 224 tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL); 225 } 226 return 0; 227 } 228 229 /* 230 * Move a socket into listening state. 231 */ 232 int inet_listen(struct socket *sock, int backlog) 233 { 234 struct sock *sk = sock->sk; 235 int err = -EINVAL; 236 237 lock_sock(sk); 238 239 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM) 240 goto out; 241 242 err = __inet_listen_sk(sk, backlog); 243 244 out: 245 release_sock(sk); 246 return err; 247 } 248 EXPORT_SYMBOL(inet_listen); 249 250 /* 251 * Create an inet socket. 252 */ 253 254 static int inet_create(struct net *net, struct socket *sock, int protocol, 255 int kern) 256 { 257 struct sock *sk; 258 struct inet_protosw *answer; 259 struct inet_sock *inet; 260 struct proto *answer_prot; 261 unsigned char answer_flags; 262 int try_loading_module = 0; 263 int err; 264 265 if (protocol < 0 || protocol >= IPPROTO_MAX) 266 return -EINVAL; 267 268 sock->state = SS_UNCONNECTED; 269 270 /* Look for the requested type/protocol pair. */ 271 lookup_protocol: 272 err = -ESOCKTNOSUPPORT; 273 rcu_read_lock(); 274 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) { 275 276 err = 0; 277 /* Check the non-wild match. */ 278 if (protocol == answer->protocol) { 279 if (protocol != IPPROTO_IP) 280 break; 281 } else { 282 /* Check for the two wild cases. */ 283 if (IPPROTO_IP == protocol) { 284 protocol = answer->protocol; 285 break; 286 } 287 if (IPPROTO_IP == answer->protocol) 288 break; 289 } 290 err = -EPROTONOSUPPORT; 291 } 292 293 if (unlikely(err)) { 294 if (try_loading_module < 2) { 295 rcu_read_unlock(); 296 /* 297 * Be more specific, e.g. net-pf-2-proto-132-type-1 298 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM) 299 */ 300 if (++try_loading_module == 1) 301 request_module("net-pf-%d-proto-%d-type-%d", 302 PF_INET, protocol, sock->type); 303 /* 304 * Fall back to generic, e.g. net-pf-2-proto-132 305 * (net-pf-PF_INET-proto-IPPROTO_SCTP) 306 */ 307 else 308 request_module("net-pf-%d-proto-%d", 309 PF_INET, protocol); 310 goto lookup_protocol; 311 } else 312 goto out_rcu_unlock; 313 } 314 315 err = -EPERM; 316 if (sock->type == SOCK_RAW && !kern && 317 !ns_capable(net->user_ns, CAP_NET_RAW)) 318 goto out_rcu_unlock; 319 320 sock->ops = answer->ops; 321 answer_prot = answer->prot; 322 answer_flags = answer->flags; 323 rcu_read_unlock(); 324 325 WARN_ON(!answer_prot->slab); 326 327 err = -ENOMEM; 328 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern); 329 if (!sk) 330 goto out; 331 332 err = 0; 333 if (INET_PROTOSW_REUSE & answer_flags) 334 sk->sk_reuse = SK_CAN_REUSE; 335 336 if (INET_PROTOSW_ICSK & answer_flags) 337 inet_init_csk_locks(sk); 338 339 inet = inet_sk(sk); 340 inet_assign_bit(IS_ICSK, sk, INET_PROTOSW_ICSK & answer_flags); 341 342 inet_clear_bit(NODEFRAG, sk); 343 344 if (SOCK_RAW == sock->type) { 345 inet->inet_num = protocol; 346 if (IPPROTO_RAW == protocol) 347 inet_set_bit(HDRINCL, sk); 348 } 349 350 if (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc)) 351 inet->pmtudisc = IP_PMTUDISC_DONT; 352 else 353 inet->pmtudisc = IP_PMTUDISC_WANT; 354 355 atomic_set(&inet->inet_id, 0); 356 357 sock_init_data(sock, sk); 358 359 sk->sk_destruct = inet_sock_destruct; 360 sk->sk_protocol = protocol; 361 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv; 362 sk->sk_txrehash = READ_ONCE(net->core.sysctl_txrehash); 363 364 inet->uc_ttl = -1; 365 inet_set_bit(MC_LOOP, sk); 366 inet->mc_ttl = 1; 367 inet_set_bit(MC_ALL, sk); 368 inet->mc_index = 0; 369 inet->mc_list = NULL; 370 inet->rcv_tos = 0; 371 372 if (inet->inet_num) { 373 /* It assumes that any protocol which allows 374 * the user to assign a number at socket 375 * creation time automatically 376 * shares. 377 */ 378 inet->inet_sport = htons(inet->inet_num); 379 /* Add to protocol hash chains. */ 380 err = sk->sk_prot->hash(sk); 381 if (err) 382 goto out_sk_release; 383 } 384 385 if (sk->sk_prot->init) { 386 err = sk->sk_prot->init(sk); 387 if (err) 388 goto out_sk_release; 389 } 390 391 if (!kern) { 392 err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk); 393 if (err) 394 goto out_sk_release; 395 } 396 out: 397 return err; 398 out_rcu_unlock: 399 rcu_read_unlock(); 400 goto out; 401 out_sk_release: 402 sk_common_release(sk); 403 sock->sk = NULL; 404 goto out; 405 } 406 407 408 /* 409 * The peer socket should always be NULL (or else). When we call this 410 * function we are destroying the object and from then on nobody 411 * should refer to it. 412 */ 413 int inet_release(struct socket *sock) 414 { 415 struct sock *sk = sock->sk; 416 417 if (sk) { 418 long timeout; 419 420 if (!sk->sk_kern_sock) 421 BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk); 422 423 /* Applications forget to leave groups before exiting */ 424 ip_mc_drop_socket(sk); 425 426 /* If linger is set, we don't return until the close 427 * is complete. Otherwise we return immediately. The 428 * actually closing is done the same either way. 429 * 430 * If the close is due to the process exiting, we never 431 * linger.. 432 */ 433 timeout = 0; 434 if (sock_flag(sk, SOCK_LINGER) && 435 !(current->flags & PF_EXITING)) 436 timeout = sk->sk_lingertime; 437 sk->sk_prot->close(sk, timeout); 438 sock->sk = NULL; 439 } 440 return 0; 441 } 442 EXPORT_SYMBOL(inet_release); 443 444 int inet_bind_sk(struct sock *sk, struct sockaddr *uaddr, int addr_len) 445 { 446 u32 flags = BIND_WITH_LOCK; 447 int err; 448 449 /* If the socket has its own bind function then use it. (RAW) */ 450 if (sk->sk_prot->bind) { 451 return sk->sk_prot->bind(sk, uaddr, addr_len); 452 } 453 if (addr_len < sizeof(struct sockaddr_in)) 454 return -EINVAL; 455 456 /* BPF prog is run before any checks are done so that if the prog 457 * changes context in a wrong way it will be caught. 458 */ 459 err = BPF_CGROUP_RUN_PROG_INET_BIND_LOCK(sk, uaddr, &addr_len, 460 CGROUP_INET4_BIND, &flags); 461 if (err) 462 return err; 463 464 return __inet_bind(sk, uaddr, addr_len, flags); 465 } 466 467 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) 468 { 469 return inet_bind_sk(sock->sk, uaddr, addr_len); 470 } 471 EXPORT_SYMBOL(inet_bind); 472 473 int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len, 474 u32 flags) 475 { 476 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr; 477 struct inet_sock *inet = inet_sk(sk); 478 struct net *net = sock_net(sk); 479 unsigned short snum; 480 int chk_addr_ret; 481 u32 tb_id = RT_TABLE_LOCAL; 482 int err; 483 484 if (addr->sin_family != AF_INET) { 485 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET) 486 * only if s_addr is INADDR_ANY. 487 */ 488 err = -EAFNOSUPPORT; 489 if (addr->sin_family != AF_UNSPEC || 490 addr->sin_addr.s_addr != htonl(INADDR_ANY)) 491 goto out; 492 } 493 494 tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id; 495 chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id); 496 497 /* Not specified by any standard per-se, however it breaks too 498 * many applications when removed. It is unfortunate since 499 * allowing applications to make a non-local bind solves 500 * several problems with systems using dynamic addressing. 501 * (ie. your servers still start up even if your ISDN link 502 * is temporarily down) 503 */ 504 err = -EADDRNOTAVAIL; 505 if (!inet_addr_valid_or_nonlocal(net, inet, addr->sin_addr.s_addr, 506 chk_addr_ret)) 507 goto out; 508 509 snum = ntohs(addr->sin_port); 510 err = -EACCES; 511 if (!(flags & BIND_NO_CAP_NET_BIND_SERVICE) && 512 snum && inet_port_requires_bind_service(net, snum) && 513 !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE)) 514 goto out; 515 516 /* We keep a pair of addresses. rcv_saddr is the one 517 * used by hash lookups, and saddr is used for transmit. 518 * 519 * In the BSD API these are the same except where it 520 * would be illegal to use them (multicast/broadcast) in 521 * which case the sending device address is used. 522 */ 523 if (flags & BIND_WITH_LOCK) 524 lock_sock(sk); 525 526 /* Check these errors (active socket, double bind). */ 527 err = -EINVAL; 528 if (sk->sk_state != TCP_CLOSE || inet->inet_num) 529 goto out_release_sock; 530 531 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr; 532 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST) 533 inet->inet_saddr = 0; /* Use device */ 534 535 /* Make sure we are allowed to bind here. */ 536 if (snum || !(inet_test_bit(BIND_ADDRESS_NO_PORT, sk) || 537 (flags & BIND_FORCE_ADDRESS_NO_PORT))) { 538 err = sk->sk_prot->get_port(sk, snum); 539 if (err) { 540 inet->inet_saddr = inet->inet_rcv_saddr = 0; 541 goto out_release_sock; 542 } 543 if (!(flags & BIND_FROM_BPF)) { 544 err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk); 545 if (err) { 546 inet->inet_saddr = inet->inet_rcv_saddr = 0; 547 if (sk->sk_prot->put_port) 548 sk->sk_prot->put_port(sk); 549 goto out_release_sock; 550 } 551 } 552 } 553 554 if (inet->inet_rcv_saddr) 555 sk->sk_userlocks |= SOCK_BINDADDR_LOCK; 556 if (snum) 557 sk->sk_userlocks |= SOCK_BINDPORT_LOCK; 558 inet->inet_sport = htons(inet->inet_num); 559 inet->inet_daddr = 0; 560 inet->inet_dport = 0; 561 sk_dst_reset(sk); 562 err = 0; 563 out_release_sock: 564 if (flags & BIND_WITH_LOCK) 565 release_sock(sk); 566 out: 567 return err; 568 } 569 570 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr, 571 int addr_len, int flags) 572 { 573 struct sock *sk = sock->sk; 574 const struct proto *prot; 575 int err; 576 577 if (addr_len < sizeof(uaddr->sa_family)) 578 return -EINVAL; 579 580 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 581 prot = READ_ONCE(sk->sk_prot); 582 583 if (uaddr->sa_family == AF_UNSPEC) 584 return prot->disconnect(sk, flags); 585 586 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) { 587 err = prot->pre_connect(sk, uaddr, addr_len); 588 if (err) 589 return err; 590 } 591 592 if (data_race(!inet_sk(sk)->inet_num) && inet_autobind(sk)) 593 return -EAGAIN; 594 return prot->connect(sk, uaddr, addr_len); 595 } 596 EXPORT_SYMBOL(inet_dgram_connect); 597 598 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias) 599 { 600 DEFINE_WAIT_FUNC(wait, woken_wake_function); 601 602 add_wait_queue(sk_sleep(sk), &wait); 603 sk->sk_write_pending += writebias; 604 605 /* Basic assumption: if someone sets sk->sk_err, he _must_ 606 * change state of the socket from TCP_SYN_*. 607 * Connect() does not allow to get error notifications 608 * without closing the socket. 609 */ 610 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 611 release_sock(sk); 612 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo); 613 lock_sock(sk); 614 if (signal_pending(current) || !timeo) 615 break; 616 } 617 remove_wait_queue(sk_sleep(sk), &wait); 618 sk->sk_write_pending -= writebias; 619 return timeo; 620 } 621 622 /* 623 * Connect to a remote host. There is regrettably still a little 624 * TCP 'magic' in here. 625 */ 626 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr, 627 int addr_len, int flags, int is_sendmsg) 628 { 629 struct sock *sk = sock->sk; 630 int err; 631 long timeo; 632 633 /* 634 * uaddr can be NULL and addr_len can be 0 if: 635 * sk is a TCP fastopen active socket and 636 * TCP_FASTOPEN_CONNECT sockopt is set and 637 * we already have a valid cookie for this socket. 638 * In this case, user can call write() after connect(). 639 * write() will invoke tcp_sendmsg_fastopen() which calls 640 * __inet_stream_connect(). 641 */ 642 if (uaddr) { 643 if (addr_len < sizeof(uaddr->sa_family)) 644 return -EINVAL; 645 646 if (uaddr->sa_family == AF_UNSPEC) { 647 sk->sk_disconnects++; 648 err = sk->sk_prot->disconnect(sk, flags); 649 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED; 650 goto out; 651 } 652 } 653 654 switch (sock->state) { 655 default: 656 err = -EINVAL; 657 goto out; 658 case SS_CONNECTED: 659 err = -EISCONN; 660 goto out; 661 case SS_CONNECTING: 662 if (inet_test_bit(DEFER_CONNECT, sk)) 663 err = is_sendmsg ? -EINPROGRESS : -EISCONN; 664 else 665 err = -EALREADY; 666 /* Fall out of switch with err, set for this state */ 667 break; 668 case SS_UNCONNECTED: 669 err = -EISCONN; 670 if (sk->sk_state != TCP_CLOSE) 671 goto out; 672 673 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) { 674 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len); 675 if (err) 676 goto out; 677 } 678 679 err = sk->sk_prot->connect(sk, uaddr, addr_len); 680 if (err < 0) 681 goto out; 682 683 sock->state = SS_CONNECTING; 684 685 if (!err && inet_test_bit(DEFER_CONNECT, sk)) 686 goto out; 687 688 /* Just entered SS_CONNECTING state; the only 689 * difference is that return value in non-blocking 690 * case is EINPROGRESS, rather than EALREADY. 691 */ 692 err = -EINPROGRESS; 693 break; 694 } 695 696 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK); 697 698 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 699 int writebias = (sk->sk_protocol == IPPROTO_TCP) && 700 tcp_sk(sk)->fastopen_req && 701 tcp_sk(sk)->fastopen_req->data ? 1 : 0; 702 int dis = sk->sk_disconnects; 703 704 /* Error code is set above */ 705 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias)) 706 goto out; 707 708 err = sock_intr_errno(timeo); 709 if (signal_pending(current)) 710 goto out; 711 712 if (dis != sk->sk_disconnects) { 713 err = -EPIPE; 714 goto out; 715 } 716 } 717 718 /* Connection was closed by RST, timeout, ICMP error 719 * or another process disconnected us. 720 */ 721 if (sk->sk_state == TCP_CLOSE) 722 goto sock_error; 723 724 /* sk->sk_err may be not zero now, if RECVERR was ordered by user 725 * and error was received after socket entered established state. 726 * Hence, it is handled normally after connect() return successfully. 727 */ 728 729 sock->state = SS_CONNECTED; 730 err = 0; 731 out: 732 return err; 733 734 sock_error: 735 err = sock_error(sk) ? : -ECONNABORTED; 736 sock->state = SS_UNCONNECTED; 737 sk->sk_disconnects++; 738 if (sk->sk_prot->disconnect(sk, flags)) 739 sock->state = SS_DISCONNECTING; 740 goto out; 741 } 742 EXPORT_SYMBOL(__inet_stream_connect); 743 744 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr, 745 int addr_len, int flags) 746 { 747 int err; 748 749 lock_sock(sock->sk); 750 err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0); 751 release_sock(sock->sk); 752 return err; 753 } 754 EXPORT_SYMBOL(inet_stream_connect); 755 756 void __inet_accept(struct socket *sock, struct socket *newsock, struct sock *newsk) 757 { 758 sock_rps_record_flow(newsk); 759 WARN_ON(!((1 << newsk->sk_state) & 760 (TCPF_ESTABLISHED | TCPF_SYN_RECV | 761 TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2 | 762 TCPF_CLOSING | TCPF_CLOSE_WAIT | 763 TCPF_CLOSE))); 764 765 if (test_bit(SOCK_SUPPORT_ZC, &sock->flags)) 766 set_bit(SOCK_SUPPORT_ZC, &newsock->flags); 767 sock_graft(newsk, newsock); 768 769 newsock->state = SS_CONNECTED; 770 } 771 772 /* 773 * Accept a pending connection. The TCP layer now gives BSD semantics. 774 */ 775 776 int inet_accept(struct socket *sock, struct socket *newsock, 777 struct proto_accept_arg *arg) 778 { 779 struct sock *sk1 = sock->sk, *sk2; 780 781 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 782 arg->err = -EINVAL; 783 sk2 = READ_ONCE(sk1->sk_prot)->accept(sk1, arg); 784 if (!sk2) 785 return arg->err; 786 787 lock_sock(sk2); 788 __inet_accept(sock, newsock, sk2); 789 release_sock(sk2); 790 return 0; 791 } 792 EXPORT_SYMBOL(inet_accept); 793 794 /* 795 * This does both peername and sockname. 796 */ 797 int inet_getname(struct socket *sock, struct sockaddr *uaddr, 798 int peer) 799 { 800 struct sock *sk = sock->sk; 801 struct inet_sock *inet = inet_sk(sk); 802 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr); 803 int sin_addr_len = sizeof(*sin); 804 805 sin->sin_family = AF_INET; 806 lock_sock(sk); 807 if (peer) { 808 if (!inet->inet_dport || 809 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) && 810 peer == 1)) { 811 release_sock(sk); 812 return -ENOTCONN; 813 } 814 sin->sin_port = inet->inet_dport; 815 sin->sin_addr.s_addr = inet->inet_daddr; 816 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin, &sin_addr_len, 817 CGROUP_INET4_GETPEERNAME); 818 } else { 819 __be32 addr = inet->inet_rcv_saddr; 820 if (!addr) 821 addr = inet->inet_saddr; 822 sin->sin_port = inet->inet_sport; 823 sin->sin_addr.s_addr = addr; 824 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin, &sin_addr_len, 825 CGROUP_INET4_GETSOCKNAME); 826 } 827 release_sock(sk); 828 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 829 return sin_addr_len; 830 } 831 EXPORT_SYMBOL(inet_getname); 832 833 int inet_send_prepare(struct sock *sk) 834 { 835 sock_rps_record_flow(sk); 836 837 /* We may need to bind the socket. */ 838 if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind && 839 inet_autobind(sk)) 840 return -EAGAIN; 841 842 return 0; 843 } 844 EXPORT_SYMBOL_GPL(inet_send_prepare); 845 846 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size) 847 { 848 struct sock *sk = sock->sk; 849 850 if (unlikely(inet_send_prepare(sk))) 851 return -EAGAIN; 852 853 return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg, 854 sk, msg, size); 855 } 856 EXPORT_SYMBOL(inet_sendmsg); 857 858 void inet_splice_eof(struct socket *sock) 859 { 860 const struct proto *prot; 861 struct sock *sk = sock->sk; 862 863 if (unlikely(inet_send_prepare(sk))) 864 return; 865 866 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 867 prot = READ_ONCE(sk->sk_prot); 868 if (prot->splice_eof) 869 prot->splice_eof(sock); 870 } 871 EXPORT_SYMBOL_GPL(inet_splice_eof); 872 873 INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *, 874 size_t, int, int *)); 875 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 876 int flags) 877 { 878 struct sock *sk = sock->sk; 879 int addr_len = 0; 880 int err; 881 882 if (likely(!(flags & MSG_ERRQUEUE))) 883 sock_rps_record_flow(sk); 884 885 err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg, 886 sk, msg, size, flags, &addr_len); 887 if (err >= 0) 888 msg->msg_namelen = addr_len; 889 return err; 890 } 891 EXPORT_SYMBOL(inet_recvmsg); 892 893 int inet_shutdown(struct socket *sock, int how) 894 { 895 struct sock *sk = sock->sk; 896 int err = 0; 897 898 /* This should really check to make sure 899 * the socket is a TCP socket. (WHY AC...) 900 */ 901 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and 902 1->2 bit 2 snds. 903 2->3 */ 904 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */ 905 return -EINVAL; 906 907 lock_sock(sk); 908 if (sock->state == SS_CONNECTING) { 909 if ((1 << sk->sk_state) & 910 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE)) 911 sock->state = SS_DISCONNECTING; 912 else 913 sock->state = SS_CONNECTED; 914 } 915 916 switch (sk->sk_state) { 917 case TCP_CLOSE: 918 err = -ENOTCONN; 919 /* Hack to wake up other listeners, who can poll for 920 EPOLLHUP, even on eg. unconnected UDP sockets -- RR */ 921 fallthrough; 922 default: 923 WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | how); 924 if (sk->sk_prot->shutdown) 925 sk->sk_prot->shutdown(sk, how); 926 break; 927 928 /* Remaining two branches are temporary solution for missing 929 * close() in multithreaded environment. It is _not_ a good idea, 930 * but we have no choice until close() is repaired at VFS level. 931 */ 932 case TCP_LISTEN: 933 if (!(how & RCV_SHUTDOWN)) 934 break; 935 fallthrough; 936 case TCP_SYN_SENT: 937 err = sk->sk_prot->disconnect(sk, O_NONBLOCK); 938 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED; 939 break; 940 } 941 942 /* Wake up anyone sleeping in poll. */ 943 sk->sk_state_change(sk); 944 release_sock(sk); 945 return err; 946 } 947 EXPORT_SYMBOL(inet_shutdown); 948 949 /* 950 * ioctl() calls you can issue on an INET socket. Most of these are 951 * device configuration and stuff and very rarely used. Some ioctls 952 * pass on to the socket itself. 953 * 954 * NOTE: I like the idea of a module for the config stuff. ie ifconfig 955 * loads the devconfigure module does its configuring and unloads it. 956 * There's a good 20K of config code hanging around the kernel. 957 */ 958 959 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 960 { 961 struct sock *sk = sock->sk; 962 int err = 0; 963 struct net *net = sock_net(sk); 964 void __user *p = (void __user *)arg; 965 struct ifreq ifr; 966 struct rtentry rt; 967 968 switch (cmd) { 969 case SIOCADDRT: 970 case SIOCDELRT: 971 if (copy_from_user(&rt, p, sizeof(struct rtentry))) 972 return -EFAULT; 973 err = ip_rt_ioctl(net, cmd, &rt); 974 break; 975 case SIOCRTMSG: 976 err = -EINVAL; 977 break; 978 case SIOCDARP: 979 case SIOCGARP: 980 case SIOCSARP: 981 err = arp_ioctl(net, cmd, (void __user *)arg); 982 break; 983 case SIOCGIFADDR: 984 case SIOCGIFBRDADDR: 985 case SIOCGIFNETMASK: 986 case SIOCGIFDSTADDR: 987 case SIOCGIFPFLAGS: 988 if (get_user_ifreq(&ifr, NULL, p)) 989 return -EFAULT; 990 err = devinet_ioctl(net, cmd, &ifr); 991 if (!err && put_user_ifreq(&ifr, p)) 992 err = -EFAULT; 993 break; 994 995 case SIOCSIFADDR: 996 case SIOCSIFBRDADDR: 997 case SIOCSIFNETMASK: 998 case SIOCSIFDSTADDR: 999 case SIOCSIFPFLAGS: 1000 case SIOCSIFFLAGS: 1001 if (get_user_ifreq(&ifr, NULL, p)) 1002 return -EFAULT; 1003 err = devinet_ioctl(net, cmd, &ifr); 1004 break; 1005 default: 1006 if (sk->sk_prot->ioctl) 1007 err = sk_ioctl(sk, cmd, (void __user *)arg); 1008 else 1009 err = -ENOIOCTLCMD; 1010 break; 1011 } 1012 return err; 1013 } 1014 EXPORT_SYMBOL(inet_ioctl); 1015 1016 #ifdef CONFIG_COMPAT 1017 static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd, 1018 struct compat_rtentry __user *ur) 1019 { 1020 compat_uptr_t rtdev; 1021 struct rtentry rt; 1022 1023 if (copy_from_user(&rt.rt_dst, &ur->rt_dst, 1024 3 * sizeof(struct sockaddr)) || 1025 get_user(rt.rt_flags, &ur->rt_flags) || 1026 get_user(rt.rt_metric, &ur->rt_metric) || 1027 get_user(rt.rt_mtu, &ur->rt_mtu) || 1028 get_user(rt.rt_window, &ur->rt_window) || 1029 get_user(rt.rt_irtt, &ur->rt_irtt) || 1030 get_user(rtdev, &ur->rt_dev)) 1031 return -EFAULT; 1032 1033 rt.rt_dev = compat_ptr(rtdev); 1034 return ip_rt_ioctl(sock_net(sk), cmd, &rt); 1035 } 1036 1037 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 1038 { 1039 void __user *argp = compat_ptr(arg); 1040 struct sock *sk = sock->sk; 1041 1042 switch (cmd) { 1043 case SIOCADDRT: 1044 case SIOCDELRT: 1045 return inet_compat_routing_ioctl(sk, cmd, argp); 1046 default: 1047 if (!sk->sk_prot->compat_ioctl) 1048 return -ENOIOCTLCMD; 1049 return sk->sk_prot->compat_ioctl(sk, cmd, arg); 1050 } 1051 } 1052 #endif /* CONFIG_COMPAT */ 1053 1054 const struct proto_ops inet_stream_ops = { 1055 .family = PF_INET, 1056 .owner = THIS_MODULE, 1057 .release = inet_release, 1058 .bind = inet_bind, 1059 .connect = inet_stream_connect, 1060 .socketpair = sock_no_socketpair, 1061 .accept = inet_accept, 1062 .getname = inet_getname, 1063 .poll = tcp_poll, 1064 .ioctl = inet_ioctl, 1065 .gettstamp = sock_gettstamp, 1066 .listen = inet_listen, 1067 .shutdown = inet_shutdown, 1068 .setsockopt = sock_common_setsockopt, 1069 .getsockopt = sock_common_getsockopt, 1070 .sendmsg = inet_sendmsg, 1071 .recvmsg = inet_recvmsg, 1072 #ifdef CONFIG_MMU 1073 .mmap = tcp_mmap, 1074 #endif 1075 .splice_eof = inet_splice_eof, 1076 .splice_read = tcp_splice_read, 1077 .set_peek_off = sk_set_peek_off, 1078 .read_sock = tcp_read_sock, 1079 .read_skb = tcp_read_skb, 1080 .sendmsg_locked = tcp_sendmsg_locked, 1081 .peek_len = tcp_peek_len, 1082 #ifdef CONFIG_COMPAT 1083 .compat_ioctl = inet_compat_ioctl, 1084 #endif 1085 .set_rcvlowat = tcp_set_rcvlowat, 1086 }; 1087 EXPORT_SYMBOL(inet_stream_ops); 1088 1089 const struct proto_ops inet_dgram_ops = { 1090 .family = PF_INET, 1091 .owner = THIS_MODULE, 1092 .release = inet_release, 1093 .bind = inet_bind, 1094 .connect = inet_dgram_connect, 1095 .socketpair = sock_no_socketpair, 1096 .accept = sock_no_accept, 1097 .getname = inet_getname, 1098 .poll = udp_poll, 1099 .ioctl = inet_ioctl, 1100 .gettstamp = sock_gettstamp, 1101 .listen = sock_no_listen, 1102 .shutdown = inet_shutdown, 1103 .setsockopt = sock_common_setsockopt, 1104 .getsockopt = sock_common_getsockopt, 1105 .sendmsg = inet_sendmsg, 1106 .read_skb = udp_read_skb, 1107 .recvmsg = inet_recvmsg, 1108 .mmap = sock_no_mmap, 1109 .splice_eof = inet_splice_eof, 1110 .set_peek_off = udp_set_peek_off, 1111 #ifdef CONFIG_COMPAT 1112 .compat_ioctl = inet_compat_ioctl, 1113 #endif 1114 }; 1115 EXPORT_SYMBOL(inet_dgram_ops); 1116 1117 /* 1118 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without 1119 * udp_poll 1120 */ 1121 static const struct proto_ops inet_sockraw_ops = { 1122 .family = PF_INET, 1123 .owner = THIS_MODULE, 1124 .release = inet_release, 1125 .bind = inet_bind, 1126 .connect = inet_dgram_connect, 1127 .socketpair = sock_no_socketpair, 1128 .accept = sock_no_accept, 1129 .getname = inet_getname, 1130 .poll = datagram_poll, 1131 .ioctl = inet_ioctl, 1132 .gettstamp = sock_gettstamp, 1133 .listen = sock_no_listen, 1134 .shutdown = inet_shutdown, 1135 .setsockopt = sock_common_setsockopt, 1136 .getsockopt = sock_common_getsockopt, 1137 .sendmsg = inet_sendmsg, 1138 .recvmsg = inet_recvmsg, 1139 .mmap = sock_no_mmap, 1140 .splice_eof = inet_splice_eof, 1141 #ifdef CONFIG_COMPAT 1142 .compat_ioctl = inet_compat_ioctl, 1143 #endif 1144 }; 1145 1146 static const struct net_proto_family inet_family_ops = { 1147 .family = PF_INET, 1148 .create = inet_create, 1149 .owner = THIS_MODULE, 1150 }; 1151 1152 /* Upon startup we insert all the elements in inetsw_array[] into 1153 * the linked list inetsw. 1154 */ 1155 static struct inet_protosw inetsw_array[] = 1156 { 1157 { 1158 .type = SOCK_STREAM, 1159 .protocol = IPPROTO_TCP, 1160 .prot = &tcp_prot, 1161 .ops = &inet_stream_ops, 1162 .flags = INET_PROTOSW_PERMANENT | 1163 INET_PROTOSW_ICSK, 1164 }, 1165 1166 { 1167 .type = SOCK_DGRAM, 1168 .protocol = IPPROTO_UDP, 1169 .prot = &udp_prot, 1170 .ops = &inet_dgram_ops, 1171 .flags = INET_PROTOSW_PERMANENT, 1172 }, 1173 1174 { 1175 .type = SOCK_DGRAM, 1176 .protocol = IPPROTO_ICMP, 1177 .prot = &ping_prot, 1178 .ops = &inet_sockraw_ops, 1179 .flags = INET_PROTOSW_REUSE, 1180 }, 1181 1182 { 1183 .type = SOCK_RAW, 1184 .protocol = IPPROTO_IP, /* wild card */ 1185 .prot = &raw_prot, 1186 .ops = &inet_sockraw_ops, 1187 .flags = INET_PROTOSW_REUSE, 1188 } 1189 }; 1190 1191 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array) 1192 1193 void inet_register_protosw(struct inet_protosw *p) 1194 { 1195 struct list_head *lh; 1196 struct inet_protosw *answer; 1197 int protocol = p->protocol; 1198 struct list_head *last_perm; 1199 1200 spin_lock_bh(&inetsw_lock); 1201 1202 if (p->type >= SOCK_MAX) 1203 goto out_illegal; 1204 1205 /* If we are trying to override a permanent protocol, bail. */ 1206 last_perm = &inetsw[p->type]; 1207 list_for_each(lh, &inetsw[p->type]) { 1208 answer = list_entry(lh, struct inet_protosw, list); 1209 /* Check only the non-wild match. */ 1210 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0) 1211 break; 1212 if (protocol == answer->protocol) 1213 goto out_permanent; 1214 last_perm = lh; 1215 } 1216 1217 /* Add the new entry after the last permanent entry if any, so that 1218 * the new entry does not override a permanent entry when matched with 1219 * a wild-card protocol. But it is allowed to override any existing 1220 * non-permanent entry. This means that when we remove this entry, the 1221 * system automatically returns to the old behavior. 1222 */ 1223 list_add_rcu(&p->list, last_perm); 1224 out: 1225 spin_unlock_bh(&inetsw_lock); 1226 1227 return; 1228 1229 out_permanent: 1230 pr_err("Attempt to override permanent protocol %d\n", protocol); 1231 goto out; 1232 1233 out_illegal: 1234 pr_err("Ignoring attempt to register invalid socket type %d\n", 1235 p->type); 1236 goto out; 1237 } 1238 EXPORT_SYMBOL(inet_register_protosw); 1239 1240 void inet_unregister_protosw(struct inet_protosw *p) 1241 { 1242 if (INET_PROTOSW_PERMANENT & p->flags) { 1243 pr_err("Attempt to unregister permanent protocol %d\n", 1244 p->protocol); 1245 } else { 1246 spin_lock_bh(&inetsw_lock); 1247 list_del_rcu(&p->list); 1248 spin_unlock_bh(&inetsw_lock); 1249 1250 synchronize_net(); 1251 } 1252 } 1253 EXPORT_SYMBOL(inet_unregister_protosw); 1254 1255 static int inet_sk_reselect_saddr(struct sock *sk) 1256 { 1257 struct inet_sock *inet = inet_sk(sk); 1258 __be32 old_saddr = inet->inet_saddr; 1259 __be32 daddr = inet->inet_daddr; 1260 struct flowi4 *fl4; 1261 struct rtable *rt; 1262 __be32 new_saddr; 1263 struct ip_options_rcu *inet_opt; 1264 int err; 1265 1266 inet_opt = rcu_dereference_protected(inet->inet_opt, 1267 lockdep_sock_is_held(sk)); 1268 if (inet_opt && inet_opt->opt.srr) 1269 daddr = inet_opt->opt.faddr; 1270 1271 /* Query new route. */ 1272 fl4 = &inet->cork.fl.u.ip4; 1273 rt = ip_route_connect(fl4, daddr, 0, sk->sk_bound_dev_if, 1274 sk->sk_protocol, inet->inet_sport, 1275 inet->inet_dport, sk); 1276 if (IS_ERR(rt)) 1277 return PTR_ERR(rt); 1278 1279 new_saddr = fl4->saddr; 1280 1281 if (new_saddr == old_saddr) { 1282 sk_setup_caps(sk, &rt->dst); 1283 return 0; 1284 } 1285 1286 err = inet_bhash2_update_saddr(sk, &new_saddr, AF_INET); 1287 if (err) { 1288 ip_rt_put(rt); 1289 return err; 1290 } 1291 1292 sk_setup_caps(sk, &rt->dst); 1293 1294 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) > 1) { 1295 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n", 1296 __func__, &old_saddr, &new_saddr); 1297 } 1298 1299 /* 1300 * XXX The only one ugly spot where we need to 1301 * XXX really change the sockets identity after 1302 * XXX it has entered the hashes. -DaveM 1303 * 1304 * Besides that, it does not check for connection 1305 * uniqueness. Wait for troubles. 1306 */ 1307 return __sk_prot_rehash(sk); 1308 } 1309 1310 int inet_sk_rebuild_header(struct sock *sk) 1311 { 1312 struct rtable *rt = dst_rtable(__sk_dst_check(sk, 0)); 1313 struct inet_sock *inet = inet_sk(sk); 1314 struct flowi4 *fl4; 1315 int err; 1316 1317 /* Route is OK, nothing to do. */ 1318 if (rt) 1319 return 0; 1320 1321 /* Reroute. */ 1322 fl4 = &inet->cork.fl.u.ip4; 1323 inet_sk_init_flowi4(inet, fl4); 1324 rt = ip_route_output_flow(sock_net(sk), fl4, sk); 1325 if (!IS_ERR(rt)) { 1326 err = 0; 1327 sk_setup_caps(sk, &rt->dst); 1328 } else { 1329 err = PTR_ERR(rt); 1330 1331 /* Routing failed... */ 1332 sk->sk_route_caps = 0; 1333 1334 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) || 1335 sk->sk_state != TCP_SYN_SENT || 1336 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) || 1337 (err = inet_sk_reselect_saddr(sk)) != 0) 1338 WRITE_ONCE(sk->sk_err_soft, -err); 1339 } 1340 1341 return err; 1342 } 1343 EXPORT_SYMBOL(inet_sk_rebuild_header); 1344 1345 void inet_sk_set_state(struct sock *sk, int state) 1346 { 1347 trace_inet_sock_set_state(sk, sk->sk_state, state); 1348 sk->sk_state = state; 1349 } 1350 EXPORT_SYMBOL(inet_sk_set_state); 1351 1352 void inet_sk_state_store(struct sock *sk, int newstate) 1353 { 1354 trace_inet_sock_set_state(sk, sk->sk_state, newstate); 1355 smp_store_release(&sk->sk_state, newstate); 1356 } 1357 1358 struct sk_buff *inet_gso_segment(struct sk_buff *skb, 1359 netdev_features_t features) 1360 { 1361 bool udpfrag = false, fixedid = false, gso_partial, encap; 1362 struct sk_buff *segs = ERR_PTR(-EINVAL); 1363 const struct net_offload *ops; 1364 unsigned int offset = 0; 1365 struct iphdr *iph; 1366 int proto, tot_len; 1367 int nhoff; 1368 int ihl; 1369 int id; 1370 1371 skb_reset_network_header(skb); 1372 nhoff = skb_network_header(skb) - skb_mac_header(skb); 1373 if (unlikely(!pskb_may_pull(skb, sizeof(*iph)))) 1374 goto out; 1375 1376 iph = ip_hdr(skb); 1377 ihl = iph->ihl * 4; 1378 if (ihl < sizeof(*iph)) 1379 goto out; 1380 1381 id = ntohs(iph->id); 1382 proto = iph->protocol; 1383 1384 /* Warning: after this point, iph might be no longer valid */ 1385 if (unlikely(!pskb_may_pull(skb, ihl))) 1386 goto out; 1387 __skb_pull(skb, ihl); 1388 1389 encap = SKB_GSO_CB(skb)->encap_level > 0; 1390 if (encap) 1391 features &= skb->dev->hw_enc_features; 1392 SKB_GSO_CB(skb)->encap_level += ihl; 1393 1394 skb_reset_transport_header(skb); 1395 1396 segs = ERR_PTR(-EPROTONOSUPPORT); 1397 1398 fixedid = !!(skb_shinfo(skb)->gso_type & (SKB_GSO_TCP_FIXEDID << encap)); 1399 1400 if (!skb->encapsulation || encap) 1401 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP); 1402 1403 ops = rcu_dereference(inet_offloads[proto]); 1404 if (likely(ops && ops->callbacks.gso_segment)) { 1405 segs = ops->callbacks.gso_segment(skb, features); 1406 if (!segs) 1407 skb->network_header = skb_mac_header(skb) + nhoff - skb->head; 1408 } 1409 1410 if (IS_ERR_OR_NULL(segs)) 1411 goto out; 1412 1413 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL); 1414 1415 skb = segs; 1416 do { 1417 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff); 1418 if (udpfrag) { 1419 iph->frag_off = htons(offset >> 3); 1420 if (skb->next) 1421 iph->frag_off |= htons(IP_MF); 1422 offset += skb->len - nhoff - ihl; 1423 tot_len = skb->len - nhoff; 1424 } else if (skb_is_gso(skb)) { 1425 if (!fixedid) { 1426 iph->id = htons(id); 1427 id += skb_shinfo(skb)->gso_segs; 1428 } 1429 1430 if (gso_partial) 1431 tot_len = skb_shinfo(skb)->gso_size + 1432 SKB_GSO_CB(skb)->data_offset + 1433 skb->head - (unsigned char *)iph; 1434 else 1435 tot_len = skb->len - nhoff; 1436 } else { 1437 if (!fixedid) 1438 iph->id = htons(id++); 1439 tot_len = skb->len - nhoff; 1440 } 1441 iph->tot_len = htons(tot_len); 1442 ip_send_check(iph); 1443 if (encap) 1444 skb_reset_inner_headers(skb); 1445 skb->network_header = (u8 *)iph - skb->head; 1446 skb_reset_mac_len(skb); 1447 } while ((skb = skb->next)); 1448 1449 out: 1450 return segs; 1451 } 1452 1453 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb, 1454 netdev_features_t features) 1455 { 1456 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4)) 1457 return ERR_PTR(-EINVAL); 1458 1459 return inet_gso_segment(skb, features); 1460 } 1461 1462 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb) 1463 { 1464 const struct net_offload *ops; 1465 struct sk_buff *pp = NULL; 1466 const struct iphdr *iph; 1467 struct sk_buff *p; 1468 unsigned int hlen; 1469 unsigned int off; 1470 int flush = 1; 1471 int proto; 1472 1473 off = skb_gro_offset(skb); 1474 hlen = off + sizeof(*iph); 1475 iph = skb_gro_header(skb, hlen, off); 1476 if (unlikely(!iph)) 1477 goto out; 1478 1479 proto = iph->protocol; 1480 1481 ops = rcu_dereference(inet_offloads[proto]); 1482 if (!ops || !ops->callbacks.gro_receive) 1483 goto out; 1484 1485 if (*(u8 *)iph != 0x45) 1486 goto out; 1487 1488 if (ip_is_fragment(iph)) 1489 goto out; 1490 1491 if (unlikely(ip_fast_csum((u8 *)iph, 5))) 1492 goto out; 1493 1494 NAPI_GRO_CB(skb)->proto = proto; 1495 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (ntohl(*(__be32 *)&iph->id) & ~IP_DF)); 1496 1497 list_for_each_entry(p, head, list) { 1498 struct iphdr *iph2; 1499 1500 if (!NAPI_GRO_CB(p)->same_flow) 1501 continue; 1502 1503 iph2 = (struct iphdr *)(p->data + off); 1504 /* The above works because, with the exception of the top 1505 * (inner most) layer, we only aggregate pkts with the same 1506 * hdr length so all the hdrs we'll need to verify will start 1507 * at the same offset. 1508 */ 1509 if ((iph->protocol ^ iph2->protocol) | 1510 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) | 1511 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) { 1512 NAPI_GRO_CB(p)->same_flow = 0; 1513 continue; 1514 } 1515 } 1516 1517 NAPI_GRO_CB(skb)->flush |= flush; 1518 NAPI_GRO_CB(skb)->network_offsets[NAPI_GRO_CB(skb)->encap_mark] = off; 1519 1520 /* Note : No need to call skb_gro_postpull_rcsum() here, 1521 * as we already checked checksum over ipv4 header was 0 1522 */ 1523 skb_gro_pull(skb, sizeof(*iph)); 1524 skb_set_transport_header(skb, skb_gro_offset(skb)); 1525 1526 pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive, 1527 ops->callbacks.gro_receive, head, skb); 1528 1529 out: 1530 skb_gro_flush_final(skb, pp, flush); 1531 1532 return pp; 1533 } 1534 1535 static struct sk_buff *ipip_gro_receive(struct list_head *head, 1536 struct sk_buff *skb) 1537 { 1538 if (NAPI_GRO_CB(skb)->encap_mark) { 1539 NAPI_GRO_CB(skb)->flush = 1; 1540 return NULL; 1541 } 1542 1543 NAPI_GRO_CB(skb)->encap_mark = 1; 1544 1545 return inet_gro_receive(head, skb); 1546 } 1547 1548 #define SECONDS_PER_DAY 86400 1549 1550 /* inet_current_timestamp - Return IP network timestamp 1551 * 1552 * Return milliseconds since midnight in network byte order. 1553 */ 1554 __be32 inet_current_timestamp(void) 1555 { 1556 u32 secs; 1557 u32 msecs; 1558 struct timespec64 ts; 1559 1560 ktime_get_real_ts64(&ts); 1561 1562 /* Get secs since midnight. */ 1563 (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs); 1564 /* Convert to msecs. */ 1565 msecs = secs * MSEC_PER_SEC; 1566 /* Convert nsec to msec. */ 1567 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC; 1568 1569 /* Convert to network byte order. */ 1570 return htonl(msecs); 1571 } 1572 EXPORT_SYMBOL(inet_current_timestamp); 1573 1574 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len) 1575 { 1576 unsigned int family = READ_ONCE(sk->sk_family); 1577 1578 if (family == AF_INET) 1579 return ip_recv_error(sk, msg, len, addr_len); 1580 #if IS_ENABLED(CONFIG_IPV6) 1581 if (family == AF_INET6) 1582 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len); 1583 #endif 1584 return -EINVAL; 1585 } 1586 EXPORT_SYMBOL(inet_recv_error); 1587 1588 int inet_gro_complete(struct sk_buff *skb, int nhoff) 1589 { 1590 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff); 1591 const struct net_offload *ops; 1592 __be16 totlen = iph->tot_len; 1593 int proto = iph->protocol; 1594 int err = -ENOSYS; 1595 1596 if (skb->encapsulation) { 1597 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP)); 1598 skb_set_inner_network_header(skb, nhoff); 1599 } 1600 1601 iph_set_totlen(iph, skb->len - nhoff); 1602 csum_replace2(&iph->check, totlen, iph->tot_len); 1603 1604 ops = rcu_dereference(inet_offloads[proto]); 1605 if (WARN_ON(!ops || !ops->callbacks.gro_complete)) 1606 goto out; 1607 1608 /* Only need to add sizeof(*iph) to get to the next hdr below 1609 * because any hdr with option will have been flushed in 1610 * inet_gro_receive(). 1611 */ 1612 err = INDIRECT_CALL_2(ops->callbacks.gro_complete, 1613 tcp4_gro_complete, udp4_gro_complete, 1614 skb, nhoff + sizeof(*iph)); 1615 1616 out: 1617 return err; 1618 } 1619 1620 static int ipip_gro_complete(struct sk_buff *skb, int nhoff) 1621 { 1622 skb->encapsulation = 1; 1623 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4; 1624 return inet_gro_complete(skb, nhoff); 1625 } 1626 1627 int inet_ctl_sock_create(struct sock **sk, unsigned short family, 1628 unsigned short type, unsigned char protocol, 1629 struct net *net) 1630 { 1631 struct socket *sock; 1632 int rc = sock_create_kern(net, family, type, protocol, &sock); 1633 1634 if (rc == 0) { 1635 *sk = sock->sk; 1636 (*sk)->sk_allocation = GFP_ATOMIC; 1637 (*sk)->sk_use_task_frag = false; 1638 /* 1639 * Unhash it so that IP input processing does not even see it, 1640 * we do not wish this socket to see incoming packets. 1641 */ 1642 (*sk)->sk_prot->unhash(*sk); 1643 } 1644 return rc; 1645 } 1646 EXPORT_SYMBOL_GPL(inet_ctl_sock_create); 1647 1648 unsigned long snmp_fold_field(void __percpu *mib, int offt) 1649 { 1650 unsigned long res = 0; 1651 int i; 1652 1653 for_each_possible_cpu(i) 1654 res += snmp_get_cpu_field(mib, i, offt); 1655 return res; 1656 } 1657 EXPORT_SYMBOL_GPL(snmp_fold_field); 1658 1659 #if BITS_PER_LONG==32 1660 1661 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt, 1662 size_t syncp_offset) 1663 { 1664 void *bhptr; 1665 struct u64_stats_sync *syncp; 1666 u64 v; 1667 unsigned int start; 1668 1669 bhptr = per_cpu_ptr(mib, cpu); 1670 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset); 1671 do { 1672 start = u64_stats_fetch_begin(syncp); 1673 v = *(((u64 *)bhptr) + offt); 1674 } while (u64_stats_fetch_retry(syncp, start)); 1675 1676 return v; 1677 } 1678 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64); 1679 1680 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset) 1681 { 1682 u64 res = 0; 1683 int cpu; 1684 1685 for_each_possible_cpu(cpu) { 1686 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset); 1687 } 1688 return res; 1689 } 1690 EXPORT_SYMBOL_GPL(snmp_fold_field64); 1691 #endif 1692 1693 #ifdef CONFIG_IP_MULTICAST 1694 static const struct net_protocol igmp_protocol = { 1695 .handler = igmp_rcv, 1696 }; 1697 #endif 1698 1699 static const struct net_protocol icmp_protocol = { 1700 .handler = icmp_rcv, 1701 .err_handler = icmp_err, 1702 .no_policy = 1, 1703 }; 1704 1705 static __net_init int ipv4_mib_init_net(struct net *net) 1706 { 1707 int i; 1708 1709 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib); 1710 if (!net->mib.tcp_statistics) 1711 goto err_tcp_mib; 1712 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib); 1713 if (!net->mib.ip_statistics) 1714 goto err_ip_mib; 1715 1716 for_each_possible_cpu(i) { 1717 struct ipstats_mib *af_inet_stats; 1718 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i); 1719 u64_stats_init(&af_inet_stats->syncp); 1720 } 1721 1722 net->mib.net_statistics = alloc_percpu(struct linux_mib); 1723 if (!net->mib.net_statistics) 1724 goto err_net_mib; 1725 net->mib.udp_statistics = alloc_percpu(struct udp_mib); 1726 if (!net->mib.udp_statistics) 1727 goto err_udp_mib; 1728 net->mib.udplite_statistics = alloc_percpu(struct udp_mib); 1729 if (!net->mib.udplite_statistics) 1730 goto err_udplite_mib; 1731 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib); 1732 if (!net->mib.icmp_statistics) 1733 goto err_icmp_mib; 1734 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib), 1735 GFP_KERNEL); 1736 if (!net->mib.icmpmsg_statistics) 1737 goto err_icmpmsg_mib; 1738 1739 tcp_mib_init(net); 1740 return 0; 1741 1742 err_icmpmsg_mib: 1743 free_percpu(net->mib.icmp_statistics); 1744 err_icmp_mib: 1745 free_percpu(net->mib.udplite_statistics); 1746 err_udplite_mib: 1747 free_percpu(net->mib.udp_statistics); 1748 err_udp_mib: 1749 free_percpu(net->mib.net_statistics); 1750 err_net_mib: 1751 free_percpu(net->mib.ip_statistics); 1752 err_ip_mib: 1753 free_percpu(net->mib.tcp_statistics); 1754 err_tcp_mib: 1755 return -ENOMEM; 1756 } 1757 1758 static __net_exit void ipv4_mib_exit_net(struct net *net) 1759 { 1760 kfree(net->mib.icmpmsg_statistics); 1761 free_percpu(net->mib.icmp_statistics); 1762 free_percpu(net->mib.udplite_statistics); 1763 free_percpu(net->mib.udp_statistics); 1764 free_percpu(net->mib.net_statistics); 1765 free_percpu(net->mib.ip_statistics); 1766 free_percpu(net->mib.tcp_statistics); 1767 #ifdef CONFIG_MPTCP 1768 /* allocated on demand, see mptcp_init_sock() */ 1769 free_percpu(net->mib.mptcp_statistics); 1770 #endif 1771 } 1772 1773 static __net_initdata struct pernet_operations ipv4_mib_ops = { 1774 .init = ipv4_mib_init_net, 1775 .exit = ipv4_mib_exit_net, 1776 }; 1777 1778 static int __init init_ipv4_mibs(void) 1779 { 1780 return register_pernet_subsys(&ipv4_mib_ops); 1781 } 1782 1783 static __net_init int inet_init_net(struct net *net) 1784 { 1785 /* 1786 * Set defaults for local port range 1787 */ 1788 net->ipv4.ip_local_ports.range = 60999u << 16 | 32768u; 1789 1790 seqlock_init(&net->ipv4.ping_group_range.lock); 1791 /* 1792 * Sane defaults - nobody may create ping sockets. 1793 * Boot scripts should set this to distro-specific group. 1794 */ 1795 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1); 1796 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0); 1797 1798 /* Default values for sysctl-controlled parameters. 1799 * We set them here, in case sysctl is not compiled. 1800 */ 1801 net->ipv4.sysctl_ip_default_ttl = IPDEFTTL; 1802 net->ipv4.sysctl_ip_fwd_update_priority = 1; 1803 net->ipv4.sysctl_ip_dynaddr = 0; 1804 net->ipv4.sysctl_ip_early_demux = 1; 1805 net->ipv4.sysctl_udp_early_demux = 1; 1806 net->ipv4.sysctl_tcp_early_demux = 1; 1807 net->ipv4.sysctl_nexthop_compat_mode = 1; 1808 #ifdef CONFIG_SYSCTL 1809 net->ipv4.sysctl_ip_prot_sock = PROT_SOCK; 1810 #endif 1811 1812 /* Some igmp sysctl, whose values are always used */ 1813 net->ipv4.sysctl_igmp_max_memberships = 20; 1814 net->ipv4.sysctl_igmp_max_msf = 10; 1815 /* IGMP reports for link-local multicast groups are enabled by default */ 1816 net->ipv4.sysctl_igmp_llm_reports = 1; 1817 net->ipv4.sysctl_igmp_qrv = 2; 1818 1819 net->ipv4.sysctl_fib_notify_on_flag_change = 0; 1820 1821 return 0; 1822 } 1823 1824 static __net_initdata struct pernet_operations af_inet_ops = { 1825 .init = inet_init_net, 1826 }; 1827 1828 static int __init init_inet_pernet_ops(void) 1829 { 1830 return register_pernet_subsys(&af_inet_ops); 1831 } 1832 1833 static int ipv4_proc_init(void); 1834 1835 /* 1836 * IP protocol layer initialiser 1837 */ 1838 1839 1840 static const struct net_offload ipip_offload = { 1841 .callbacks = { 1842 .gso_segment = ipip_gso_segment, 1843 .gro_receive = ipip_gro_receive, 1844 .gro_complete = ipip_gro_complete, 1845 }, 1846 }; 1847 1848 static int __init ipip_offload_init(void) 1849 { 1850 return inet_add_offload(&ipip_offload, IPPROTO_IPIP); 1851 } 1852 1853 static int __init ipv4_offload_init(void) 1854 { 1855 /* 1856 * Add offloads 1857 */ 1858 if (udpv4_offload_init() < 0) 1859 pr_crit("%s: Cannot add UDP protocol offload\n", __func__); 1860 if (tcpv4_offload_init() < 0) 1861 pr_crit("%s: Cannot add TCP protocol offload\n", __func__); 1862 if (ipip_offload_init() < 0) 1863 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__); 1864 1865 net_hotdata.ip_packet_offload = (struct packet_offload) { 1866 .type = cpu_to_be16(ETH_P_IP), 1867 .callbacks = { 1868 .gso_segment = inet_gso_segment, 1869 .gro_receive = inet_gro_receive, 1870 .gro_complete = inet_gro_complete, 1871 }, 1872 }; 1873 dev_add_offload(&net_hotdata.ip_packet_offload); 1874 return 0; 1875 } 1876 1877 fs_initcall(ipv4_offload_init); 1878 1879 static struct packet_type ip_packet_type __read_mostly = { 1880 .type = cpu_to_be16(ETH_P_IP), 1881 .func = ip_rcv, 1882 .list_func = ip_list_rcv, 1883 }; 1884 1885 static int __init inet_init(void) 1886 { 1887 struct inet_protosw *q; 1888 struct list_head *r; 1889 int rc; 1890 1891 sock_skb_cb_check_size(sizeof(struct inet_skb_parm)); 1892 1893 raw_hashinfo_init(&raw_v4_hashinfo); 1894 1895 rc = proto_register(&tcp_prot, 1); 1896 if (rc) 1897 goto out; 1898 1899 rc = proto_register(&udp_prot, 1); 1900 if (rc) 1901 goto out_unregister_tcp_proto; 1902 1903 rc = proto_register(&raw_prot, 1); 1904 if (rc) 1905 goto out_unregister_udp_proto; 1906 1907 rc = proto_register(&ping_prot, 1); 1908 if (rc) 1909 goto out_unregister_raw_proto; 1910 1911 /* 1912 * Tell SOCKET that we are alive... 1913 */ 1914 1915 (void)sock_register(&inet_family_ops); 1916 1917 #ifdef CONFIG_SYSCTL 1918 ip_static_sysctl_init(); 1919 #endif 1920 1921 /* 1922 * Add all the base protocols. 1923 */ 1924 1925 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0) 1926 pr_crit("%s: Cannot add ICMP protocol\n", __func__); 1927 1928 net_hotdata.udp_protocol = (struct net_protocol) { 1929 .handler = udp_rcv, 1930 .err_handler = udp_err, 1931 .no_policy = 1, 1932 }; 1933 if (inet_add_protocol(&net_hotdata.udp_protocol, IPPROTO_UDP) < 0) 1934 pr_crit("%s: Cannot add UDP protocol\n", __func__); 1935 1936 net_hotdata.tcp_protocol = (struct net_protocol) { 1937 .handler = tcp_v4_rcv, 1938 .err_handler = tcp_v4_err, 1939 .no_policy = 1, 1940 .icmp_strict_tag_validation = 1, 1941 }; 1942 if (inet_add_protocol(&net_hotdata.tcp_protocol, IPPROTO_TCP) < 0) 1943 pr_crit("%s: Cannot add TCP protocol\n", __func__); 1944 #ifdef CONFIG_IP_MULTICAST 1945 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0) 1946 pr_crit("%s: Cannot add IGMP protocol\n", __func__); 1947 #endif 1948 1949 /* Register the socket-side information for inet_create. */ 1950 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r) 1951 INIT_LIST_HEAD(r); 1952 1953 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q) 1954 inet_register_protosw(q); 1955 1956 /* 1957 * Set the ARP module up 1958 */ 1959 1960 arp_init(); 1961 1962 /* 1963 * Set the IP module up 1964 */ 1965 1966 ip_init(); 1967 1968 /* Initialise per-cpu ipv4 mibs */ 1969 if (init_ipv4_mibs()) 1970 panic("%s: Cannot init ipv4 mibs\n", __func__); 1971 1972 /* Setup TCP slab cache for open requests. */ 1973 tcp_init(); 1974 1975 /* Setup UDP memory threshold */ 1976 udp_init(); 1977 1978 /* Add UDP-Lite (RFC 3828) */ 1979 udplite4_register(); 1980 1981 raw_init(); 1982 1983 ping_init(); 1984 1985 /* 1986 * Set the ICMP layer up 1987 */ 1988 1989 if (icmp_init() < 0) 1990 panic("Failed to create the ICMP control socket.\n"); 1991 1992 /* 1993 * Initialise the multicast router 1994 */ 1995 #if defined(CONFIG_IP_MROUTE) 1996 if (ip_mr_init()) 1997 pr_crit("%s: Cannot init ipv4 mroute\n", __func__); 1998 #endif 1999 2000 if (init_inet_pernet_ops()) 2001 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__); 2002 2003 ipv4_proc_init(); 2004 2005 ipfrag_init(); 2006 2007 dev_add_pack(&ip_packet_type); 2008 2009 ip_tunnel_core_init(); 2010 2011 rc = 0; 2012 out: 2013 return rc; 2014 out_unregister_raw_proto: 2015 proto_unregister(&raw_prot); 2016 out_unregister_udp_proto: 2017 proto_unregister(&udp_prot); 2018 out_unregister_tcp_proto: 2019 proto_unregister(&tcp_prot); 2020 goto out; 2021 } 2022 2023 fs_initcall(inet_init); 2024 2025 /* ------------------------------------------------------------------------ */ 2026 2027 #ifdef CONFIG_PROC_FS 2028 static int __init ipv4_proc_init(void) 2029 { 2030 int rc = 0; 2031 2032 if (raw_proc_init()) 2033 goto out_raw; 2034 if (tcp4_proc_init()) 2035 goto out_tcp; 2036 if (udp4_proc_init()) 2037 goto out_udp; 2038 if (ping_proc_init()) 2039 goto out_ping; 2040 if (ip_misc_proc_init()) 2041 goto out_misc; 2042 out: 2043 return rc; 2044 out_misc: 2045 ping_proc_exit(); 2046 out_ping: 2047 udp4_proc_exit(); 2048 out_udp: 2049 tcp4_proc_exit(); 2050 out_tcp: 2051 raw_proc_exit(); 2052 out_raw: 2053 rc = -ENOMEM; 2054 goto out; 2055 } 2056 2057 #else /* CONFIG_PROC_FS */ 2058 static int __init ipv4_proc_init(void) 2059 { 2060 return 0; 2061 } 2062 #endif /* CONFIG_PROC_FS */ 2063