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/ping.h> 108 #include <linux/skbuff.h> 109 #include <net/sock.h> 110 #include <net/raw.h> 111 #include <net/icmp.h> 112 #include <net/inet_common.h> 113 #include <net/ip_tunnels.h> 114 #include <net/xfrm.h> 115 #include <net/net_namespace.h> 116 #include <net/secure_seq.h> 117 #ifdef CONFIG_IP_MROUTE 118 #include <linux/mroute.h> 119 #endif 120 #include <net/l3mdev.h> 121 #include <net/compat.h> 122 #include <net/rps.h> 123 124 #include <trace/events/sock.h> 125 126 /* Keep the definition of IPv6 disable here for now, to avoid annoying linker 127 * issues in case IPv6=m 128 */ 129 int disable_ipv6_mod; 130 EXPORT_SYMBOL(disable_ipv6_mod); 131 132 /* The inetsw table contains everything that inet_create needs to 133 * build a new socket. 134 */ 135 static struct list_head inetsw[SOCK_MAX]; 136 static DEFINE_SPINLOCK(inetsw_lock); 137 138 /* New destruction routine */ 139 140 void inet_sock_destruct(struct sock *sk) 141 { 142 struct inet_sock *inet = inet_sk(sk); 143 144 __skb_queue_purge(&sk->sk_receive_queue); 145 __skb_queue_purge(&sk->sk_error_queue); 146 147 sk_mem_reclaim_final(sk); 148 149 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) { 150 pr_err("Attempt to release TCP socket in state %d %p\n", 151 sk->sk_state, sk); 152 return; 153 } 154 if (!sock_flag(sk, SOCK_DEAD)) { 155 pr_err("Attempt to release alive inet socket %p\n", sk); 156 return; 157 } 158 159 WARN_ON_ONCE(atomic_read(&sk->sk_rmem_alloc)); 160 WARN_ON_ONCE(refcount_read(&sk->sk_wmem_alloc)); 161 WARN_ON_ONCE(sk->sk_wmem_queued); 162 WARN_ON_ONCE(sk->sk_forward_alloc); 163 164 kfree_rcu(rcu_dereference_protected(inet->inet_opt, 1), rcu); 165 dst_release(rcu_dereference_protected(sk->sk_dst_cache, 1)); 166 dst_release(rcu_dereference_protected(sk->sk_rx_dst, 1)); 167 psp_sk_assoc_free(sk); 168 } 169 EXPORT_SYMBOL(inet_sock_destruct); 170 171 /* 172 * The routines beyond this point handle the behaviour of an AF_INET 173 * socket object. Mostly it punts to the subprotocols of IP to do 174 * the work. 175 */ 176 177 /* 178 * Automatically bind an unbound socket. 179 */ 180 181 static int inet_autobind(struct sock *sk) 182 { 183 struct inet_sock *inet; 184 /* We may need to bind the socket. */ 185 lock_sock(sk); 186 inet = inet_sk(sk); 187 if (!inet->inet_num) { 188 if (sk->sk_prot->get_port(sk, 0)) { 189 release_sock(sk); 190 return -EAGAIN; 191 } 192 inet->inet_sport = htons(inet->inet_num); 193 } 194 release_sock(sk); 195 return 0; 196 } 197 198 int __inet_listen_sk(struct sock *sk, int backlog) 199 { 200 unsigned char old_state = sk->sk_state; 201 int err, tcp_fastopen; 202 203 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN))) 204 return -EINVAL; 205 206 WRITE_ONCE(sk->sk_max_ack_backlog, backlog); 207 /* Really, if the socket is already in listen state 208 * we can only allow the backlog to be adjusted. 209 */ 210 if (old_state != TCP_LISTEN) { 211 /* Enable TFO w/o requiring TCP_FASTOPEN socket option. 212 * Note that only TCP sockets (SOCK_STREAM) will reach here. 213 * Also fastopen backlog may already been set via the option 214 * because the socket was in TCP_LISTEN state previously but 215 * was shutdown() rather than close(). 216 */ 217 tcp_fastopen = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fastopen); 218 if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) && 219 (tcp_fastopen & TFO_SERVER_ENABLE) && 220 !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) { 221 fastopen_queue_tune(sk, backlog); 222 tcp_fastopen_init_key_once(sock_net(sk)); 223 } 224 225 err = inet_csk_listen_start(sk); 226 if (err) 227 return err; 228 229 tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB); 230 bpf_tcp_ops_call(listen, sk); 231 } 232 return 0; 233 } 234 235 /* 236 * Move a socket into listening state. 237 */ 238 int inet_listen(struct socket *sock, int backlog) 239 { 240 struct sock *sk = sock->sk; 241 int err = -EINVAL; 242 243 lock_sock(sk); 244 245 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM) 246 goto out; 247 248 err = __inet_listen_sk(sk, backlog); 249 250 out: 251 release_sock(sk); 252 return err; 253 } 254 EXPORT_SYMBOL(inet_listen); 255 256 /* 257 * Create an inet socket. 258 */ 259 260 static int inet_create(struct net *net, struct socket *sock, int protocol, 261 int kern) 262 { 263 struct sock *sk; 264 struct inet_protosw *answer; 265 struct inet_sock *inet; 266 struct proto *answer_prot; 267 unsigned char answer_flags; 268 int try_loading_module = 0; 269 int err; 270 271 if (protocol < 0 || protocol >= IPPROTO_MAX) 272 return -EINVAL; 273 274 sock->state = SS_UNCONNECTED; 275 276 /* Look for the requested type/protocol pair. */ 277 lookup_protocol: 278 err = -ESOCKTNOSUPPORT; 279 rcu_read_lock(); 280 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) { 281 282 err = 0; 283 /* Check the non-wild match. */ 284 if (protocol == answer->protocol) { 285 if (protocol != IPPROTO_IP) 286 break; 287 } else { 288 /* Check for the two wild cases. */ 289 if (IPPROTO_IP == protocol) { 290 protocol = answer->protocol; 291 break; 292 } 293 if (IPPROTO_IP == answer->protocol) 294 break; 295 } 296 err = -EPROTONOSUPPORT; 297 } 298 299 if (unlikely(err)) { 300 if (try_loading_module < 2) { 301 rcu_read_unlock(); 302 /* 303 * Be more specific, e.g. net-pf-2-proto-132-type-1 304 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM) 305 */ 306 if (++try_loading_module == 1) 307 request_module("net-pf-%d-proto-%d-type-%d", 308 PF_INET, protocol, sock->type); 309 /* 310 * Fall back to generic, e.g. net-pf-2-proto-132 311 * (net-pf-PF_INET-proto-IPPROTO_SCTP) 312 */ 313 else 314 request_module("net-pf-%d-proto-%d", 315 PF_INET, protocol); 316 goto lookup_protocol; 317 } else 318 goto out_rcu_unlock; 319 } 320 321 err = -EPERM; 322 if (sock->type == SOCK_RAW && !kern && 323 !ns_capable(net->user_ns, CAP_NET_RAW)) 324 goto out_rcu_unlock; 325 326 sock->ops = answer->ops; 327 answer_prot = answer->prot; 328 answer_flags = answer->flags; 329 rcu_read_unlock(); 330 331 WARN_ON(!answer_prot->slab); 332 333 err = -ENOMEM; 334 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern); 335 if (!sk) 336 goto out; 337 338 err = 0; 339 if (INET_PROTOSW_REUSE & answer_flags) 340 sk->sk_reuse = SK_CAN_REUSE; 341 342 if (INET_PROTOSW_ICSK & answer_flags) 343 inet_init_csk_locks(sk); 344 345 inet = inet_sk(sk); 346 inet_assign_bit(IS_ICSK, sk, INET_PROTOSW_ICSK & answer_flags); 347 348 inet_clear_bit(NODEFRAG, sk); 349 350 if (SOCK_RAW == sock->type) { 351 inet->inet_num = protocol; 352 if (IPPROTO_RAW == protocol) 353 inet_set_bit(HDRINCL, sk); 354 } 355 356 if (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc)) 357 inet->pmtudisc = IP_PMTUDISC_DONT; 358 else 359 inet->pmtudisc = IP_PMTUDISC_WANT; 360 361 atomic_set(&inet->inet_id, 0); 362 363 sock_init_data(sock, sk); 364 365 sk->sk_destruct = inet_sock_destruct; 366 sk->sk_protocol = protocol; 367 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv; 368 sk->sk_txrehash = READ_ONCE(net->core.sysctl_txrehash); 369 370 inet->uc_ttl = -1; 371 inet_set_bit(MC_LOOP, sk); 372 inet->mc_ttl = 1; 373 inet_set_bit(MC_ALL, sk); 374 inet->mc_index = 0; 375 inet->mc_list = NULL; 376 inet->rcv_tos = 0; 377 378 if (inet->inet_num) { 379 /* It assumes that any protocol which allows 380 * the user to assign a number at socket 381 * creation time automatically 382 * shares. 383 */ 384 inet->inet_sport = htons(inet->inet_num); 385 /* Add to protocol hash chains. */ 386 err = sk->sk_prot->hash(sk); 387 if (err) 388 goto out_sk_release; 389 } 390 391 if (sk->sk_prot->init) { 392 err = sk->sk_prot->init(sk); 393 if (err) 394 goto out_sk_release; 395 } 396 397 if (!kern) { 398 err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk); 399 if (err) 400 goto out_sk_release; 401 } 402 out: 403 return err; 404 out_rcu_unlock: 405 rcu_read_unlock(); 406 goto out; 407 out_sk_release: 408 sk_common_release(sk); 409 sock->sk = NULL; 410 goto out; 411 } 412 413 414 /* 415 * The peer socket should always be NULL (or else). When we call this 416 * function we are destroying the object and from then on nobody 417 * should refer to it. 418 */ 419 int inet_release(struct socket *sock) 420 { 421 struct sock *sk = sock->sk; 422 423 if (sk) { 424 long timeout; 425 426 if (!sk->sk_kern_sock) 427 BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk); 428 429 /* Applications forget to leave groups before exiting */ 430 ip_mc_drop_socket(sk); 431 432 /* If linger is set, we don't return until the close 433 * is complete. Otherwise we return immediately. The 434 * actually closing is done the same either way. 435 * 436 * If the close is due to the process exiting, we never 437 * linger.. 438 */ 439 timeout = 0; 440 if (sock_flag(sk, SOCK_LINGER) && 441 !(current->flags & PF_EXITING)) 442 timeout = sk->sk_lingertime; 443 sk->sk_prot->close(sk, timeout); 444 sock->sk = NULL; 445 } 446 return 0; 447 } 448 EXPORT_SYMBOL(inet_release); 449 450 int inet_bind_sk(struct sock *sk, struct sockaddr_unsized *uaddr, int addr_len) 451 { 452 u32 flags = BIND_WITH_LOCK; 453 int err; 454 455 /* If the socket has its own bind function then use it. (RAW) */ 456 if (sk->sk_prot->bind) { 457 return sk->sk_prot->bind(sk, uaddr, addr_len); 458 } 459 if (addr_len < sizeof(struct sockaddr_in)) 460 return -EINVAL; 461 462 /* BPF prog is run before any checks are done so that if the prog 463 * changes context in a wrong way it will be caught. 464 */ 465 err = BPF_CGROUP_RUN_PROG_INET_BIND_LOCK(sk, uaddr, &addr_len, 466 CGROUP_INET4_BIND, &flags); 467 if (err) 468 return err; 469 470 return __inet_bind(sk, uaddr, addr_len, flags); 471 } 472 473 int inet_bind(struct socket *sock, struct sockaddr_unsized *uaddr, int addr_len) 474 { 475 return inet_bind_sk(sock->sk, uaddr, addr_len); 476 } 477 EXPORT_SYMBOL(inet_bind); 478 479 int __inet_bind(struct sock *sk, struct sockaddr_unsized *uaddr, int addr_len, 480 u32 flags) 481 { 482 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr; 483 struct inet_sock *inet = inet_sk(sk); 484 struct net *net = sock_net(sk); 485 unsigned short snum; 486 int chk_addr_ret; 487 u32 tb_id = RT_TABLE_LOCAL; 488 int err; 489 490 if (addr->sin_family != AF_INET) { 491 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET) 492 * only if s_addr is INADDR_ANY. 493 */ 494 err = -EAFNOSUPPORT; 495 if (addr->sin_family != AF_UNSPEC || 496 addr->sin_addr.s_addr != htonl(INADDR_ANY)) 497 goto out; 498 } 499 500 tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id; 501 chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id); 502 503 /* Not specified by any standard per-se, however it breaks too 504 * many applications when removed. It is unfortunate since 505 * allowing applications to make a non-local bind solves 506 * several problems with systems using dynamic addressing. 507 * (ie. your servers still start up even if your ISDN link 508 * is temporarily down) 509 */ 510 err = -EADDRNOTAVAIL; 511 if (!inet_addr_valid_or_nonlocal(net, inet, addr->sin_addr.s_addr, 512 chk_addr_ret)) 513 goto out; 514 515 snum = ntohs(addr->sin_port); 516 err = -EACCES; 517 if (!(flags & BIND_NO_CAP_NET_BIND_SERVICE) && 518 snum && inet_port_requires_bind_service(net, snum) && 519 !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE)) 520 goto out; 521 522 /* We keep a pair of addresses. rcv_saddr is the one 523 * used by hash lookups, and saddr is used for transmit. 524 * 525 * In the BSD API these are the same except where it 526 * would be illegal to use them (multicast/broadcast) in 527 * which case the sending device address is used. 528 */ 529 if (flags & BIND_WITH_LOCK) 530 lock_sock(sk); 531 532 /* Check these errors (active socket, double bind). */ 533 err = -EINVAL; 534 if (sk->sk_state != TCP_CLOSE || inet->inet_num) 535 goto out_release_sock; 536 537 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr; 538 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST) 539 inet->inet_saddr = 0; /* Use device */ 540 541 /* Make sure we are allowed to bind here. */ 542 if (snum || !(inet_test_bit(BIND_ADDRESS_NO_PORT, sk) || 543 (flags & BIND_FORCE_ADDRESS_NO_PORT))) { 544 err = sk->sk_prot->get_port(sk, snum); 545 if (err) { 546 inet->inet_saddr = inet->inet_rcv_saddr = 0; 547 goto out_release_sock; 548 } 549 if (!(flags & BIND_FROM_BPF)) { 550 err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk); 551 if (err) { 552 inet->inet_saddr = inet->inet_rcv_saddr = 0; 553 if (sk->sk_prot->put_port) 554 sk->sk_prot->put_port(sk); 555 goto out_release_sock; 556 } 557 } 558 } 559 560 if (inet->inet_rcv_saddr) 561 sk->sk_userlocks |= SOCK_BINDADDR_LOCK; 562 if (snum) 563 sk->sk_userlocks |= SOCK_BINDPORT_LOCK; 564 inet->inet_sport = htons(inet->inet_num); 565 inet->inet_daddr = 0; 566 inet->inet_dport = 0; 567 sk_dst_reset(sk); 568 err = 0; 569 out_release_sock: 570 if (flags & BIND_WITH_LOCK) 571 release_sock(sk); 572 out: 573 return err; 574 } 575 576 int inet_dgram_connect(struct socket *sock, struct sockaddr_unsized *uaddr, 577 int addr_len, int flags) 578 { 579 struct sock *sk = sock->sk; 580 const struct proto *prot; 581 int err; 582 583 if (addr_len < sizeof(uaddr->sa_family)) 584 return -EINVAL; 585 586 prot = READ_ONCE(sk->sk_prot); 587 588 if (uaddr->sa_family == AF_UNSPEC) 589 return prot->disconnect(sk, flags); 590 591 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) { 592 err = prot->pre_connect(sk, uaddr, addr_len); 593 if (err) 594 return err; 595 } 596 597 if (data_race(!inet_sk(sk)->inet_num) && inet_autobind(sk)) 598 return -EAGAIN; 599 return prot->connect(sk, uaddr, addr_len); 600 } 601 EXPORT_SYMBOL(inet_dgram_connect); 602 603 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias) 604 { 605 DEFINE_WAIT_FUNC(wait, woken_wake_function); 606 607 add_wait_queue(sk_sleep(sk), &wait); 608 sk->sk_write_pending += writebias; 609 610 /* Basic assumption: if someone sets sk->sk_err, he _must_ 611 * change state of the socket from TCP_SYN_*. 612 * Connect() does not allow to get error notifications 613 * without closing the socket. 614 */ 615 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 616 release_sock(sk); 617 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo); 618 lock_sock(sk); 619 if (signal_pending(current) || !timeo) 620 break; 621 } 622 remove_wait_queue(sk_sleep(sk), &wait); 623 sk->sk_write_pending -= writebias; 624 return timeo; 625 } 626 627 /* 628 * Connect to a remote host. There is regrettably still a little 629 * TCP 'magic' in here. 630 */ 631 int __inet_stream_connect(struct socket *sock, struct sockaddr_unsized *uaddr, 632 int addr_len, int flags, int is_sendmsg) 633 { 634 struct sock *sk = sock->sk; 635 int err; 636 long timeo; 637 638 /* 639 * uaddr can be NULL and addr_len can be 0 if: 640 * sk is a TCP fastopen active socket and 641 * TCP_FASTOPEN_CONNECT sockopt is set and 642 * we already have a valid cookie for this socket. 643 * In this case, user can call write() after connect(). 644 * write() will invoke tcp_sendmsg_fastopen() which calls 645 * __inet_stream_connect(). 646 */ 647 if (uaddr) { 648 if (addr_len < sizeof(uaddr->sa_family)) 649 return -EINVAL; 650 651 if (uaddr->sa_family == AF_UNSPEC) { 652 sk->sk_disconnects++; 653 err = sk->sk_prot->disconnect(sk, flags); 654 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED; 655 goto out; 656 } 657 } 658 659 switch (sock->state) { 660 default: 661 err = -EINVAL; 662 goto out; 663 case SS_CONNECTED: 664 err = -EISCONN; 665 goto out; 666 case SS_CONNECTING: 667 if (inet_test_bit(DEFER_CONNECT, sk)) 668 err = is_sendmsg ? -EINPROGRESS : -EISCONN; 669 else 670 err = -EALREADY; 671 /* Fall out of switch with err, set for this state */ 672 break; 673 case SS_UNCONNECTED: 674 err = -EISCONN; 675 if (sk->sk_state != TCP_CLOSE) 676 goto out; 677 678 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) { 679 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len); 680 if (err) 681 goto out; 682 } 683 684 err = sk->sk_prot->connect(sk, uaddr, addr_len); 685 if (err < 0) 686 goto out; 687 688 sock->state = SS_CONNECTING; 689 690 if (!err && inet_test_bit(DEFER_CONNECT, sk)) 691 goto out; 692 693 /* Just entered SS_CONNECTING state; the only 694 * difference is that return value in non-blocking 695 * case is EINPROGRESS, rather than EALREADY. 696 */ 697 err = -EINPROGRESS; 698 break; 699 } 700 701 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK); 702 703 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 704 int writebias = (sk->sk_protocol == IPPROTO_TCP) && 705 tcp_sk(sk)->fastopen_req && 706 tcp_sk(sk)->fastopen_req->data ? 1 : 0; 707 int dis = sk->sk_disconnects; 708 709 /* Error code is set above */ 710 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias)) 711 goto out; 712 713 err = sock_intr_errno(timeo); 714 if (signal_pending(current)) 715 goto out; 716 717 if (dis != sk->sk_disconnects) { 718 err = -EPIPE; 719 goto out; 720 } 721 } 722 723 /* Connection was closed by RST, timeout, ICMP error 724 * or another process disconnected us. 725 */ 726 if (sk->sk_state == TCP_CLOSE) 727 goto sock_error; 728 729 /* sk->sk_err may be not zero now, if RECVERR was ordered by user 730 * and error was received after socket entered established state. 731 * Hence, it is handled normally after connect() return successfully. 732 */ 733 734 sock->state = SS_CONNECTED; 735 err = 0; 736 out: 737 return err; 738 739 sock_error: 740 err = sock_error(sk) ? : -ECONNABORTED; 741 sock->state = SS_UNCONNECTED; 742 sk->sk_disconnects++; 743 if (sk->sk_prot->disconnect(sk, flags)) 744 sock->state = SS_DISCONNECTING; 745 goto out; 746 } 747 EXPORT_SYMBOL(__inet_stream_connect); 748 749 int inet_stream_connect(struct socket *sock, struct sockaddr_unsized *uaddr, 750 int addr_len, int flags) 751 { 752 int err; 753 754 lock_sock(sock->sk); 755 err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0); 756 release_sock(sock->sk); 757 return err; 758 } 759 EXPORT_SYMBOL(inet_stream_connect); 760 761 void __inet_accept(struct socket *sock, struct socket *newsock, struct sock *newsk) 762 { 763 if (mem_cgroup_sockets_enabled) { 764 mem_cgroup_sk_alloc(newsk); 765 __sk_charge(newsk, GFP_KERNEL); 766 } 767 768 sock_rps_record_flow(newsk); 769 WARN_ON(!((1 << newsk->sk_state) & 770 (TCPF_ESTABLISHED | TCPF_SYN_RECV | 771 TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2 | 772 TCPF_CLOSING | TCPF_CLOSE_WAIT | 773 TCPF_CLOSE))); 774 775 if (test_bit(SOCK_SUPPORT_ZC, &sock->flags)) 776 set_bit(SOCK_SUPPORT_ZC, &newsock->flags); 777 sock_graft(newsk, newsock); 778 779 newsock->state = SS_CONNECTED; 780 } 781 EXPORT_SYMBOL_GPL(__inet_accept); 782 783 /* 784 * Accept a pending connection. The TCP layer now gives BSD semantics. 785 */ 786 787 int inet_accept(struct socket *sock, struct socket *newsock, 788 struct proto_accept_arg *arg) 789 { 790 struct sock *sk1 = sock->sk, *sk2; 791 792 arg->err = -EINVAL; 793 sk2 = READ_ONCE(sk1->sk_prot)->accept(sk1, arg); 794 if (!sk2) 795 return arg->err; 796 797 lock_sock(sk2); 798 __inet_accept(sock, newsock, sk2); 799 release_sock(sk2); 800 return 0; 801 } 802 EXPORT_SYMBOL(inet_accept); 803 804 /* 805 * This does both peername and sockname. 806 */ 807 int inet_getname(struct socket *sock, struct sockaddr *uaddr, 808 int peer) 809 { 810 struct sock *sk = sock->sk; 811 struct inet_sock *inet = inet_sk(sk); 812 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr); 813 int sin_addr_len = sizeof(*sin); 814 815 sin->sin_family = AF_INET; 816 lock_sock(sk); 817 if (peer) { 818 if (!inet->inet_dport || 819 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) && 820 peer == 1)) { 821 release_sock(sk); 822 return -ENOTCONN; 823 } 824 sin->sin_port = inet->inet_dport; 825 sin->sin_addr.s_addr = inet->inet_daddr; 826 BPF_CGROUP_RUN_SA_PROG(sk, sin, &sin_addr_len, 827 CGROUP_INET4_GETPEERNAME); 828 } else { 829 __be32 addr = inet->inet_rcv_saddr; 830 if (!addr) 831 addr = inet->inet_saddr; 832 sin->sin_port = inet->inet_sport; 833 sin->sin_addr.s_addr = addr; 834 BPF_CGROUP_RUN_SA_PROG(sk, sin, &sin_addr_len, 835 CGROUP_INET4_GETSOCKNAME); 836 } 837 release_sock(sk); 838 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 839 return sin_addr_len; 840 } 841 EXPORT_SYMBOL(inet_getname); 842 843 int inet_send_prepare(struct sock *sk) 844 { 845 sock_rps_record_flow(sk); 846 847 /* We may need to bind the socket. */ 848 if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind && 849 inet_autobind(sk)) 850 return -EAGAIN; 851 852 return 0; 853 } 854 855 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size) 856 { 857 struct sock *sk = sock->sk; 858 const struct proto *prot; 859 860 if (unlikely(inet_send_prepare(sk))) 861 return -EAGAIN; 862 863 prot = READ_ONCE(sk->sk_prot); 864 return INDIRECT_CALL_2(prot->sendmsg, tcp_sendmsg, udp_sendmsg, 865 sk, msg, size); 866 } 867 EXPORT_SYMBOL(inet_sendmsg); 868 869 void inet_splice_eof(struct socket *sock) 870 { 871 const struct proto *prot; 872 struct sock *sk = sock->sk; 873 874 if (unlikely(inet_send_prepare(sk))) 875 return; 876 877 prot = READ_ONCE(sk->sk_prot); 878 if (prot->splice_eof) 879 prot->splice_eof(sock); 880 } 881 882 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 883 int flags) 884 { 885 struct sock *sk = sock->sk; 886 const struct proto *prot; 887 888 if (likely(!(flags & MSG_ERRQUEUE))) 889 sock_rps_record_flow(sk); 890 891 prot = READ_ONCE(sk->sk_prot); 892 return INDIRECT_CALL_2(prot->recvmsg, tcp_recvmsg, udp_recvmsg, 893 sk, msg, size, flags); 894 } 895 EXPORT_SYMBOL(inet_recvmsg); 896 897 int inet_shutdown(struct socket *sock, int how) 898 { 899 struct sock *sk = sock->sk; 900 int err = 0; 901 902 /* This should really check to make sure 903 * the socket is a TCP socket. (WHY AC...) 904 */ 905 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and 906 1->2 bit 2 snds. 907 2->3 */ 908 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */ 909 return -EINVAL; 910 911 lock_sock(sk); 912 if (sock->state == SS_CONNECTING) { 913 if ((1 << sk->sk_state) & 914 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE)) 915 sock->state = SS_DISCONNECTING; 916 else 917 sock->state = SS_CONNECTED; 918 } 919 920 switch (sk->sk_state) { 921 case TCP_CLOSE: 922 err = -ENOTCONN; 923 /* Hack to wake up other listeners, who can poll for 924 EPOLLHUP, even on eg. unconnected UDP sockets -- RR */ 925 fallthrough; 926 default: 927 WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | how); 928 if (sk->sk_prot->shutdown) 929 sk->sk_prot->shutdown(sk, how); 930 break; 931 932 /* Remaining two branches are temporary solution for missing 933 * close() in multithreaded environment. It is _not_ a good idea, 934 * but we have no choice until close() is repaired at VFS level. 935 */ 936 case TCP_LISTEN: 937 if (!(how & RCV_SHUTDOWN)) 938 break; 939 fallthrough; 940 case TCP_SYN_SENT: 941 err = sk->sk_prot->disconnect(sk, O_NONBLOCK); 942 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED; 943 break; 944 } 945 946 /* Wake up anyone sleeping in poll. */ 947 sk->sk_state_change(sk); 948 release_sock(sk); 949 return err; 950 } 951 EXPORT_SYMBOL(inet_shutdown); 952 953 /* 954 * ioctl() calls you can issue on an INET socket. Most of these are 955 * device configuration and stuff and very rarely used. Some ioctls 956 * pass on to the socket itself. 957 * 958 * NOTE: I like the idea of a module for the config stuff. ie ifconfig 959 * loads the devconfigure module does its configuring and unloads it. 960 * There's a good 20K of config code hanging around the kernel. 961 */ 962 963 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 964 { 965 struct sock *sk = sock->sk; 966 int err = 0; 967 struct net *net = sock_net(sk); 968 void __user *p = (void __user *)arg; 969 struct ifreq ifr; 970 struct rtentry rt; 971 972 switch (cmd) { 973 case SIOCADDRT: 974 case SIOCDELRT: 975 if (copy_from_user(&rt, p, sizeof(struct rtentry))) 976 return -EFAULT; 977 err = ip_rt_ioctl(net, cmd, &rt); 978 break; 979 case SIOCRTMSG: 980 err = -EINVAL; 981 break; 982 case SIOCDARP: 983 case SIOCGARP: 984 case SIOCSARP: 985 err = arp_ioctl(net, cmd, (void __user *)arg); 986 break; 987 case SIOCGIFADDR: 988 case SIOCGIFBRDADDR: 989 case SIOCGIFNETMASK: 990 case SIOCGIFDSTADDR: 991 case SIOCGIFPFLAGS: 992 if (get_user_ifreq(&ifr, NULL, p)) 993 return -EFAULT; 994 err = devinet_ioctl(net, cmd, &ifr); 995 if (!err && put_user_ifreq(&ifr, p)) 996 err = -EFAULT; 997 break; 998 999 case SIOCSIFADDR: 1000 case SIOCSIFBRDADDR: 1001 case SIOCSIFNETMASK: 1002 case SIOCSIFDSTADDR: 1003 case SIOCSIFPFLAGS: 1004 case SIOCSIFFLAGS: 1005 if (get_user_ifreq(&ifr, NULL, p)) 1006 return -EFAULT; 1007 err = devinet_ioctl(net, cmd, &ifr); 1008 break; 1009 default: 1010 if (sk->sk_prot->ioctl) 1011 err = sk_ioctl(sk, cmd, (void __user *)arg); 1012 else 1013 err = -ENOIOCTLCMD; 1014 break; 1015 } 1016 return err; 1017 } 1018 EXPORT_SYMBOL(inet_ioctl); 1019 1020 #ifdef CONFIG_COMPAT 1021 static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd, 1022 struct compat_rtentry __user *ur) 1023 { 1024 compat_uptr_t rtdev; 1025 struct rtentry rt; 1026 1027 if (copy_from_user(&rt.rt_dst, &ur->rt_dst, 1028 3 * sizeof(struct sockaddr)) || 1029 get_user(rt.rt_flags, &ur->rt_flags) || 1030 get_user(rt.rt_metric, &ur->rt_metric) || 1031 get_user(rt.rt_mtu, &ur->rt_mtu) || 1032 get_user(rt.rt_window, &ur->rt_window) || 1033 get_user(rt.rt_irtt, &ur->rt_irtt) || 1034 get_user(rtdev, &ur->rt_dev)) 1035 return -EFAULT; 1036 1037 rt.rt_dev = compat_ptr(rtdev); 1038 return ip_rt_ioctl(sock_net(sk), cmd, &rt); 1039 } 1040 1041 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 1042 { 1043 void __user *argp = compat_ptr(arg); 1044 struct sock *sk = sock->sk; 1045 1046 switch (cmd) { 1047 case SIOCADDRT: 1048 case SIOCDELRT: 1049 return inet_compat_routing_ioctl(sk, cmd, argp); 1050 default: 1051 if (!sk->sk_prot->compat_ioctl) 1052 return -ENOIOCTLCMD; 1053 return sk->sk_prot->compat_ioctl(sk, cmd, arg); 1054 } 1055 } 1056 #endif /* CONFIG_COMPAT */ 1057 1058 const struct proto_ops inet_stream_ops = { 1059 .family = PF_INET, 1060 .owner = THIS_MODULE, 1061 .release = inet_release, 1062 .bind = inet_bind, 1063 .connect = inet_stream_connect, 1064 .socketpair = sock_no_socketpair, 1065 .accept = inet_accept, 1066 .getname = inet_getname, 1067 .poll = tcp_poll, 1068 .ioctl = inet_ioctl, 1069 .gettstamp = sock_gettstamp, 1070 .listen = inet_listen, 1071 .shutdown = inet_shutdown, 1072 .setsockopt = sock_common_setsockopt, 1073 .getsockopt = sock_common_getsockopt, 1074 .sendmsg = inet_sendmsg, 1075 .recvmsg = inet_recvmsg, 1076 #ifdef CONFIG_MMU 1077 .mmap = tcp_mmap, 1078 #endif 1079 .splice_eof = inet_splice_eof, 1080 .splice_read = tcp_splice_read, 1081 .set_peek_off = sk_set_peek_off, 1082 .read_sock = tcp_read_sock, 1083 .read_skb = tcp_read_skb, 1084 .sendmsg_locked = tcp_sendmsg_locked, 1085 .peek_len = tcp_peek_len, 1086 #ifdef CONFIG_COMPAT 1087 .compat_ioctl = inet_compat_ioctl, 1088 #endif 1089 .set_rcvlowat = tcp_set_rcvlowat, 1090 .set_rcvbuf = tcp_set_rcvbuf, 1091 }; 1092 EXPORT_SYMBOL(inet_stream_ops); 1093 1094 const struct proto_ops inet_dgram_ops = { 1095 .family = PF_INET, 1096 .owner = THIS_MODULE, 1097 .release = inet_release, 1098 .bind = inet_bind, 1099 .connect = inet_dgram_connect, 1100 .socketpair = sock_no_socketpair, 1101 .accept = sock_no_accept, 1102 .getname = inet_getname, 1103 .poll = udp_poll, 1104 .ioctl = inet_ioctl, 1105 .gettstamp = sock_gettstamp, 1106 .listen = sock_no_listen, 1107 .shutdown = inet_shutdown, 1108 .setsockopt = sock_common_setsockopt, 1109 .getsockopt = sock_common_getsockopt, 1110 .sendmsg = inet_sendmsg, 1111 .read_skb = udp_read_skb, 1112 .recvmsg = inet_recvmsg, 1113 .mmap = sock_no_mmap, 1114 .splice_eof = inet_splice_eof, 1115 .set_peek_off = udp_set_peek_off, 1116 #ifdef CONFIG_COMPAT 1117 .compat_ioctl = inet_compat_ioctl, 1118 #endif 1119 }; 1120 EXPORT_SYMBOL(inet_dgram_ops); 1121 1122 /* 1123 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without 1124 * udp_poll 1125 */ 1126 static const struct proto_ops inet_sockraw_ops = { 1127 .family = PF_INET, 1128 .owner = THIS_MODULE, 1129 .release = inet_release, 1130 .bind = inet_bind, 1131 .connect = inet_dgram_connect, 1132 .socketpair = sock_no_socketpair, 1133 .accept = sock_no_accept, 1134 .getname = inet_getname, 1135 .poll = datagram_poll, 1136 .ioctl = inet_ioctl, 1137 .gettstamp = sock_gettstamp, 1138 .listen = sock_no_listen, 1139 .shutdown = inet_shutdown, 1140 .setsockopt = sock_common_setsockopt, 1141 .getsockopt = sock_common_getsockopt, 1142 .sendmsg = inet_sendmsg, 1143 .recvmsg = inet_recvmsg, 1144 .mmap = sock_no_mmap, 1145 .splice_eof = inet_splice_eof, 1146 #ifdef CONFIG_COMPAT 1147 .compat_ioctl = inet_compat_ioctl, 1148 #endif 1149 }; 1150 1151 static const struct net_proto_family inet_family_ops = { 1152 .family = PF_INET, 1153 .create = inet_create, 1154 .owner = THIS_MODULE, 1155 }; 1156 1157 /* Upon startup we insert all the elements in inetsw_array[] into 1158 * the linked list inetsw. 1159 */ 1160 static struct inet_protosw inetsw_array[] = 1161 { 1162 { 1163 .type = SOCK_STREAM, 1164 .protocol = IPPROTO_TCP, 1165 .prot = &tcp_prot, 1166 .ops = &inet_stream_ops, 1167 .flags = INET_PROTOSW_PERMANENT | 1168 INET_PROTOSW_ICSK, 1169 }, 1170 1171 { 1172 .type = SOCK_DGRAM, 1173 .protocol = IPPROTO_UDP, 1174 .prot = &udp_prot, 1175 .ops = &inet_dgram_ops, 1176 .flags = INET_PROTOSW_PERMANENT, 1177 }, 1178 1179 { 1180 .type = SOCK_DGRAM, 1181 .protocol = IPPROTO_ICMP, 1182 .prot = &ping_prot, 1183 .ops = &inet_sockraw_ops, 1184 .flags = INET_PROTOSW_REUSE, 1185 }, 1186 1187 { 1188 .type = SOCK_RAW, 1189 .protocol = IPPROTO_IP, /* wild card */ 1190 .prot = &raw_prot, 1191 .ops = &inet_sockraw_ops, 1192 .flags = INET_PROTOSW_REUSE, 1193 } 1194 }; 1195 1196 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array) 1197 1198 void inet_register_protosw(struct inet_protosw *p) 1199 { 1200 struct list_head *lh; 1201 struct inet_protosw *answer; 1202 int protocol = p->protocol; 1203 struct list_head *last_perm; 1204 1205 spin_lock_bh(&inetsw_lock); 1206 1207 if (p->type >= SOCK_MAX) 1208 goto out_illegal; 1209 1210 /* If we are trying to override a permanent protocol, bail. */ 1211 last_perm = &inetsw[p->type]; 1212 list_for_each(lh, &inetsw[p->type]) { 1213 answer = list_entry(lh, struct inet_protosw, list); 1214 /* Check only the non-wild match. */ 1215 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0) 1216 break; 1217 if (protocol == answer->protocol) 1218 goto out_permanent; 1219 last_perm = lh; 1220 } 1221 1222 /* Add the new entry after the last permanent entry if any, so that 1223 * the new entry does not override a permanent entry when matched with 1224 * a wild-card protocol. But it is allowed to override any existing 1225 * non-permanent entry. This means that when we remove this entry, the 1226 * system automatically returns to the old behavior. 1227 */ 1228 list_add_rcu(&p->list, last_perm); 1229 out: 1230 spin_unlock_bh(&inetsw_lock); 1231 1232 return; 1233 1234 out_permanent: 1235 pr_err("Attempt to override permanent protocol %d\n", protocol); 1236 goto out; 1237 1238 out_illegal: 1239 pr_err("Ignoring attempt to register invalid socket type %d\n", 1240 p->type); 1241 goto out; 1242 } 1243 EXPORT_SYMBOL(inet_register_protosw); 1244 1245 void inet_unregister_protosw(struct inet_protosw *p) 1246 { 1247 if (INET_PROTOSW_PERMANENT & p->flags) { 1248 pr_err("Attempt to unregister permanent protocol %d\n", 1249 p->protocol); 1250 } else { 1251 spin_lock_bh(&inetsw_lock); 1252 list_del_rcu(&p->list); 1253 spin_unlock_bh(&inetsw_lock); 1254 1255 synchronize_net(); 1256 } 1257 } 1258 EXPORT_SYMBOL(inet_unregister_protosw); 1259 1260 static int inet_sk_reselect_saddr(struct sock *sk) 1261 { 1262 struct inet_sock *inet = inet_sk(sk); 1263 __be32 old_saddr = inet->inet_saddr; 1264 __be32 daddr = inet->inet_daddr; 1265 struct flowi4 *fl4; 1266 struct rtable *rt; 1267 __be32 new_saddr; 1268 struct ip_options_rcu *inet_opt; 1269 int err; 1270 1271 inet_opt = rcu_dereference_protected(inet->inet_opt, 1272 lockdep_sock_is_held(sk)); 1273 if (inet_opt && inet_opt->opt.srr) 1274 daddr = inet_opt->opt.faddr; 1275 1276 /* Query new route. */ 1277 fl4 = &inet->cork.fl.u.ip4; 1278 rt = ip_route_connect(fl4, daddr, 0, sk->sk_bound_dev_if, 1279 sk->sk_protocol, inet->inet_sport, 1280 inet->inet_dport, sk); 1281 if (IS_ERR(rt)) 1282 return PTR_ERR(rt); 1283 1284 new_saddr = fl4->saddr; 1285 1286 if (new_saddr == old_saddr) { 1287 sk_setup_caps(sk, &rt->dst); 1288 return 0; 1289 } 1290 1291 err = inet_bhash2_update_saddr(sk, &new_saddr, AF_INET); 1292 if (err) { 1293 ip_rt_put(rt); 1294 return err; 1295 } 1296 1297 sk_setup_caps(sk, &rt->dst); 1298 1299 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) > 1) { 1300 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n", 1301 __func__, &old_saddr, &new_saddr); 1302 } 1303 1304 /* 1305 * XXX The only one ugly spot where we need to 1306 * XXX really change the sockets identity after 1307 * XXX it has entered the hashes. -DaveM 1308 * 1309 * Besides that, it does not check for connection 1310 * uniqueness. Wait for troubles. 1311 */ 1312 return __sk_prot_rehash(sk); 1313 } 1314 1315 int inet_sk_rebuild_header(struct sock *sk) 1316 { 1317 struct rtable *rt = dst_rtable(__sk_dst_check(sk, 0)); 1318 struct inet_sock *inet = inet_sk(sk); 1319 struct flowi4 *fl4; 1320 int err; 1321 1322 /* Route is OK, nothing to do. */ 1323 if (rt) 1324 return 0; 1325 1326 /* Reroute. */ 1327 fl4 = &inet->cork.fl.u.ip4; 1328 inet_sk_init_flowi4(inet, fl4); 1329 rt = ip_route_output_flow(sock_net(sk), fl4, sk); 1330 if (!IS_ERR(rt)) { 1331 err = 0; 1332 sk_setup_caps(sk, &rt->dst); 1333 } else { 1334 err = PTR_ERR(rt); 1335 1336 /* Routing failed... */ 1337 sk->sk_route_caps = 0; 1338 1339 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) || 1340 sk->sk_state != TCP_SYN_SENT || 1341 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) || 1342 (err = inet_sk_reselect_saddr(sk)) != 0) 1343 WRITE_ONCE(sk->sk_err_soft, -err); 1344 } 1345 1346 return err; 1347 } 1348 1349 void inet_sk_set_state(struct sock *sk, int state) 1350 { 1351 trace_inet_sock_set_state(sk, sk->sk_state, state); 1352 sk->sk_state = state; 1353 } 1354 EXPORT_SYMBOL(inet_sk_set_state); 1355 1356 void inet_sk_state_store(struct sock *sk, int newstate) 1357 { 1358 trace_inet_sock_set_state(sk, sk->sk_state, newstate); 1359 smp_store_release(&sk->sk_state, newstate); 1360 } 1361 1362 struct sk_buff *inet_gso_segment(struct sk_buff *skb, 1363 netdev_features_t features) 1364 { 1365 bool udpfrag = false, fixedid = false, gso_partial, encap; 1366 struct sk_buff *segs = ERR_PTR(-EINVAL); 1367 const struct net_offload *ops; 1368 unsigned int offset = 0; 1369 struct iphdr *iph; 1370 int proto, tot_len; 1371 int nhoff; 1372 int ihl; 1373 int id; 1374 1375 skb_reset_network_header(skb); 1376 if (unlikely(gso_recursion_inc_test(skb, 1377 IP_TUNNEL_RECURSION_LIMIT))) 1378 goto out; 1379 nhoff = skb_network_header(skb) - skb_mac_header(skb); 1380 if (unlikely(!pskb_may_pull(skb, sizeof(*iph)))) 1381 goto out; 1382 1383 iph = ip_hdr(skb); 1384 ihl = iph->ihl * 4; 1385 if (ihl < sizeof(*iph)) 1386 goto out; 1387 1388 id = ntohs(iph->id); 1389 proto = iph->protocol; 1390 1391 /* Warning: after this point, iph might be no longer valid */ 1392 if (unlikely(!pskb_may_pull(skb, ihl))) 1393 goto out; 1394 __skb_pull(skb, ihl); 1395 1396 encap = SKB_GSO_CB(skb)->encap_level > 0; 1397 if (encap) 1398 features &= skb->dev->hw_enc_features; 1399 SKB_GSO_CB(skb)->encap_level += ihl; 1400 1401 skb_reset_transport_header(skb); 1402 1403 segs = ERR_PTR(-EPROTONOSUPPORT); 1404 1405 fixedid = !!(skb_shinfo(skb)->gso_type & (SKB_GSO_TCP_FIXEDID << encap)); 1406 1407 if (!skb->encapsulation || encap) 1408 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP); 1409 1410 ops = rcu_dereference(inet_offloads[proto]); 1411 if (likely(ops && ops->callbacks.gso_segment)) { 1412 segs = ops->callbacks.gso_segment(skb, features); 1413 if (!segs) 1414 skb->network_header = skb_mac_header(skb) + nhoff - skb->head; 1415 } 1416 1417 if (IS_ERR_OR_NULL(segs)) 1418 goto out; 1419 1420 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL); 1421 1422 skb = segs; 1423 do { 1424 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff); 1425 if (udpfrag) { 1426 iph->frag_off = htons(offset >> 3); 1427 if (skb->next) 1428 iph->frag_off |= htons(IP_MF); 1429 offset += skb->len - nhoff - ihl; 1430 tot_len = skb->len - nhoff; 1431 } else if (skb_is_gso(skb)) { 1432 if (!fixedid) { 1433 iph->id = htons(id); 1434 id += skb_shinfo(skb)->gso_segs; 1435 } 1436 1437 if (gso_partial) 1438 tot_len = skb_shinfo(skb)->gso_size + 1439 SKB_GSO_CB(skb)->data_offset + 1440 skb->head - (unsigned char *)iph; 1441 else 1442 tot_len = skb->len - nhoff; 1443 } else { 1444 if (!fixedid) 1445 iph->id = htons(id++); 1446 tot_len = skb->len - nhoff; 1447 } 1448 iph->tot_len = htons(tot_len); 1449 ip_send_check(iph); 1450 if (encap) 1451 skb_reset_inner_headers(skb); 1452 skb->network_header = (u8 *)iph - skb->head; 1453 skb_reset_mac_len(skb); 1454 } while ((skb = skb->next)); 1455 1456 out: 1457 return segs; 1458 } 1459 1460 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb, 1461 netdev_features_t features) 1462 { 1463 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4)) 1464 return ERR_PTR(-EINVAL); 1465 1466 return inet_gso_segment(skb, features); 1467 } 1468 1469 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb) 1470 { 1471 const struct net_offload *ops; 1472 struct sk_buff *pp = NULL; 1473 const struct iphdr *iph; 1474 struct sk_buff *p; 1475 unsigned int hlen; 1476 unsigned int off; 1477 int flush = 1; 1478 int proto; 1479 1480 off = skb_gro_offset(skb); 1481 hlen = off + sizeof(*iph); 1482 iph = skb_gro_header(skb, hlen, off); 1483 if (unlikely(!iph)) 1484 goto out; 1485 1486 proto = iph->protocol; 1487 1488 ops = rcu_dereference(inet_offloads[proto]); 1489 if (!ops || !ops->callbacks.gro_receive) 1490 goto out; 1491 1492 if (*(u8 *)iph != 0x45) 1493 goto out; 1494 1495 if (ip_is_fragment(iph)) 1496 goto out; 1497 1498 if (unlikely(ip_fast_csum((u8 *)iph, 5))) 1499 goto out; 1500 1501 NAPI_GRO_CB(skb)->proto = proto; 1502 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (ntohl(*(__be32 *)&iph->id) & ~IP_DF)); 1503 1504 list_for_each_entry(p, head, list) { 1505 struct iphdr *iph2; 1506 1507 if (!NAPI_GRO_CB(p)->same_flow) 1508 continue; 1509 1510 iph2 = (struct iphdr *)(p->data + off); 1511 /* The above works because, with the exception of the top 1512 * (inner most) layer, we only aggregate pkts with the same 1513 * hdr length so all the hdrs we'll need to verify will start 1514 * at the same offset. 1515 */ 1516 if ((iph->protocol ^ iph2->protocol) | 1517 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) | 1518 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) { 1519 NAPI_GRO_CB(p)->same_flow = 0; 1520 continue; 1521 } 1522 } 1523 1524 NAPI_GRO_CB(skb)->flush |= flush; 1525 NAPI_GRO_CB(skb)->network_offsets[NAPI_GRO_CB(skb)->encap_mark] = off; 1526 1527 /* Note : No need to call skb_gro_postpull_rcsum() here, 1528 * as we already checked checksum over ipv4 header was 0 1529 */ 1530 skb_gro_pull(skb, sizeof(*iph)); 1531 skb_set_transport_header(skb, skb_gro_offset(skb)); 1532 1533 pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive, 1534 ops->callbacks.gro_receive, head, skb); 1535 1536 out: 1537 skb_gro_flush_final(skb, pp, flush); 1538 1539 return pp; 1540 } 1541 EXPORT_INDIRECT_CALLABLE(inet_gro_receive); 1542 1543 static struct sk_buff *ipip_gro_receive(struct list_head *head, 1544 struct sk_buff *skb) 1545 { 1546 if (NAPI_GRO_CB(skb)->encap_mark) { 1547 NAPI_GRO_CB(skb)->flush = 1; 1548 return NULL; 1549 } 1550 1551 NAPI_GRO_CB(skb)->encap_mark = 1; 1552 1553 return inet_gro_receive(head, skb); 1554 } 1555 1556 #define SECONDS_PER_DAY 86400 1557 1558 /* inet_current_timestamp - Return IP network timestamp 1559 * 1560 * Return milliseconds since midnight in network byte order. 1561 */ 1562 __be32 inet_current_timestamp(void) 1563 { 1564 u32 secs; 1565 u32 msecs; 1566 struct timespec64 ts; 1567 1568 ktime_get_real_ts64(&ts); 1569 1570 /* Get secs since midnight. */ 1571 (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs); 1572 /* Convert to msecs. */ 1573 msecs = secs * MSEC_PER_SEC; 1574 /* Convert nsec to msec. */ 1575 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC; 1576 1577 /* Convert to network byte order. */ 1578 return htonl(msecs); 1579 } 1580 1581 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len) 1582 { 1583 unsigned int family = READ_ONCE(sk->sk_family); 1584 1585 if (family == AF_INET) 1586 return ip_recv_error(sk, msg, len); 1587 #if IS_ENABLED(CONFIG_IPV6) 1588 if (family == AF_INET6) 1589 return pingv6_ops.ipv6_recv_error(sk, msg, len); 1590 #endif 1591 return -EINVAL; 1592 } 1593 EXPORT_SYMBOL(inet_recv_error); 1594 1595 int inet_gro_complete(struct sk_buff *skb, int nhoff) 1596 { 1597 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff); 1598 const struct net_offload *ops; 1599 __be16 totlen = iph->tot_len; 1600 int proto = iph->protocol; 1601 int err = -ENOSYS; 1602 1603 if (skb->encapsulation) { 1604 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP)); 1605 skb_set_inner_network_header(skb, nhoff); 1606 } 1607 1608 iph_set_totlen(iph, skb->len - nhoff); 1609 csum_replace2(&iph->check, totlen, iph->tot_len); 1610 1611 ops = rcu_dereference(inet_offloads[proto]); 1612 if (WARN_ON(!ops || !ops->callbacks.gro_complete)) 1613 goto out; 1614 1615 /* Only need to add sizeof(*iph) to get to the next hdr below 1616 * because any hdr with option will have been flushed in 1617 * inet_gro_receive(). 1618 */ 1619 err = INDIRECT_CALL_2(ops->callbacks.gro_complete, 1620 tcp4_gro_complete, udp4_gro_complete, 1621 skb, nhoff + sizeof(*iph)); 1622 1623 out: 1624 return err; 1625 } 1626 EXPORT_INDIRECT_CALLABLE(inet_gro_complete); 1627 1628 static int ipip_gro_complete(struct sk_buff *skb, int nhoff) 1629 { 1630 skb->encapsulation = 1; 1631 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4; 1632 return inet_gro_complete(skb, nhoff); 1633 } 1634 1635 int inet_ctl_sock_create(struct sock **sk, unsigned short family, 1636 unsigned short type, unsigned char protocol, 1637 struct net *net) 1638 { 1639 struct socket *sock; 1640 int rc = sock_create_kern(net, family, type, protocol, &sock); 1641 1642 if (rc == 0) { 1643 *sk = sock->sk; 1644 (*sk)->sk_allocation = GFP_ATOMIC; 1645 (*sk)->sk_use_task_frag = false; 1646 /* 1647 * Unhash it so that IP input processing does not even see it, 1648 * we do not wish this socket to see incoming packets. 1649 */ 1650 (*sk)->sk_prot->unhash(*sk); 1651 } 1652 return rc; 1653 } 1654 EXPORT_SYMBOL_GPL(inet_ctl_sock_create); 1655 1656 unsigned long snmp_fold_field(void __percpu *mib, int offt) 1657 { 1658 unsigned long res = 0; 1659 int i; 1660 1661 for_each_possible_cpu(i) 1662 res += snmp_get_cpu_field(mib, i, offt); 1663 return res; 1664 } 1665 1666 #if BITS_PER_LONG==32 1667 1668 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt, 1669 size_t syncp_offset) 1670 { 1671 void *bhptr; 1672 struct u64_stats_sync *syncp; 1673 u64 v; 1674 unsigned int start; 1675 1676 bhptr = per_cpu_ptr(mib, cpu); 1677 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset); 1678 do { 1679 start = u64_stats_fetch_begin(syncp); 1680 v = *(((u64 *)bhptr) + offt); 1681 } while (u64_stats_fetch_retry(syncp, start)); 1682 1683 return v; 1684 } 1685 1686 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset) 1687 { 1688 u64 res = 0; 1689 int cpu; 1690 1691 for_each_possible_cpu(cpu) { 1692 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset); 1693 } 1694 return res; 1695 } 1696 #endif 1697 1698 #ifdef CONFIG_IP_MULTICAST 1699 static const struct net_protocol igmp_protocol = { 1700 .handler = igmp_rcv, 1701 }; 1702 #endif 1703 1704 static const struct net_protocol icmp_protocol = { 1705 .handler = icmp_rcv, 1706 .err_handler = icmp_err, 1707 .no_policy = 1, 1708 }; 1709 1710 static __net_init int ipv4_mib_init_net(struct net *net) 1711 { 1712 int i; 1713 1714 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib); 1715 if (!net->mib.tcp_statistics) 1716 goto err_tcp_mib; 1717 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib); 1718 if (!net->mib.ip_statistics) 1719 goto err_ip_mib; 1720 1721 for_each_possible_cpu(i) { 1722 struct ipstats_mib *af_inet_stats; 1723 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i); 1724 u64_stats_init(&af_inet_stats->syncp); 1725 } 1726 1727 net->mib.net_statistics = alloc_percpu(struct linux_mib); 1728 if (!net->mib.net_statistics) 1729 goto err_net_mib; 1730 net->mib.udp_statistics = alloc_percpu(struct udp_mib); 1731 if (!net->mib.udp_statistics) 1732 goto err_udp_mib; 1733 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib); 1734 if (!net->mib.icmp_statistics) 1735 goto err_icmp_mib; 1736 net->mib.icmpmsg_statistics = kzalloc_obj(struct icmpmsg_mib); 1737 if (!net->mib.icmpmsg_statistics) 1738 goto err_icmpmsg_mib; 1739 1740 tcp_mib_init(net); 1741 return 0; 1742 1743 err_icmpmsg_mib: 1744 free_percpu(net->mib.icmp_statistics); 1745 err_icmp_mib: 1746 free_percpu(net->mib.udp_statistics); 1747 err_udp_mib: 1748 free_percpu(net->mib.net_statistics); 1749 err_net_mib: 1750 free_percpu(net->mib.ip_statistics); 1751 err_ip_mib: 1752 free_percpu(net->mib.tcp_statistics); 1753 err_tcp_mib: 1754 return -ENOMEM; 1755 } 1756 1757 static __net_exit void ipv4_mib_exit_net(struct net *net) 1758 { 1759 kfree(net->mib.icmpmsg_statistics); 1760 free_percpu(net->mib.icmp_statistics); 1761 free_percpu(net->mib.udp_statistics); 1762 free_percpu(net->mib.net_statistics); 1763 free_percpu(net->mib.ip_statistics); 1764 free_percpu(net->mib.tcp_statistics); 1765 #ifdef CONFIG_MPTCP 1766 /* allocated on demand, see mptcp_init_sock() */ 1767 free_percpu(net->mib.mptcp_statistics); 1768 #endif 1769 } 1770 1771 static __net_initdata struct pernet_operations ipv4_mib_ops = { 1772 .init = ipv4_mib_init_net, 1773 .exit = ipv4_mib_exit_net, 1774 }; 1775 1776 static int __init init_ipv4_mibs(void) 1777 { 1778 return register_pernet_subsys(&ipv4_mib_ops); 1779 } 1780 1781 static __net_init int inet_init_net(struct net *net) 1782 { 1783 /* 1784 * Set defaults for local port range 1785 */ 1786 net->ipv4.ip_local_ports.range = 60999u << 16 | 32768u; 1787 1788 seqlock_init(&net->ipv4.ping_group_range.lock); 1789 /* 1790 * Sane defaults - nobody may create ping sockets. 1791 * Boot scripts should set this to distro-specific group. 1792 */ 1793 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1); 1794 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0); 1795 1796 /* Default values for sysctl-controlled parameters. 1797 * We set them here, in case sysctl is not compiled. 1798 */ 1799 net->ipv4.sysctl_ip_default_ttl = IPDEFTTL; 1800 net->ipv4.sysctl_ip_fwd_update_priority = 1; 1801 net->ipv4.sysctl_ip_dynaddr = 0; 1802 net->ipv4.sysctl_ip_early_demux = 1; 1803 net->ipv4.sysctl_udp_early_demux = 1; 1804 net->ipv4.sysctl_tcp_early_demux = 1; 1805 net->ipv4.sysctl_nexthop_compat_mode = 1; 1806 #ifdef CONFIG_SYSCTL 1807 net->ipv4.sysctl_ip_prot_sock = PROT_SOCK; 1808 #endif 1809 1810 /* Some igmp sysctl, whose values are always used */ 1811 net->ipv4.sysctl_igmp_max_memberships = 20; 1812 net->ipv4.sysctl_igmp_max_msf = 10; 1813 /* IGMP reports for link-local multicast groups are enabled by default */ 1814 net->ipv4.sysctl_igmp_llm_reports = 1; 1815 net->ipv4.sysctl_igmp_qrv = 2; 1816 1817 net->ipv4.sysctl_fib_notify_on_flag_change = 0; 1818 1819 return 0; 1820 } 1821 1822 static __net_initdata struct pernet_operations af_inet_ops = { 1823 .init = inet_init_net, 1824 }; 1825 1826 static int __init init_inet_pernet_ops(void) 1827 { 1828 return register_pernet_subsys(&af_inet_ops); 1829 } 1830 1831 static int ipv4_proc_init(void); 1832 1833 /* 1834 * IP protocol layer initialiser 1835 */ 1836 1837 1838 static const struct net_offload ipip_offload = { 1839 .callbacks = { 1840 .gso_segment = ipip_gso_segment, 1841 .gro_receive = ipip_gro_receive, 1842 .gro_complete = ipip_gro_complete, 1843 }, 1844 }; 1845 1846 static int __init ipip_offload_init(void) 1847 { 1848 return inet_add_offload(&ipip_offload, IPPROTO_IPIP); 1849 } 1850 1851 static int __init ipv4_offload_init(void) 1852 { 1853 /* 1854 * Add offloads 1855 */ 1856 if (udpv4_offload_init() < 0) 1857 pr_crit("%s: Cannot add UDP protocol offload\n", __func__); 1858 if (tcpv4_offload_init() < 0) 1859 pr_crit("%s: Cannot add TCP protocol offload\n", __func__); 1860 if (ipip_offload_init() < 0) 1861 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__); 1862 1863 net_hotdata.ip_packet_offload = (struct packet_offload) { 1864 .type = cpu_to_be16(ETH_P_IP), 1865 .callbacks = { 1866 .gso_segment = inet_gso_segment, 1867 .gro_receive = inet_gro_receive, 1868 .gro_complete = inet_gro_complete, 1869 }, 1870 }; 1871 dev_add_offload(&net_hotdata.ip_packet_offload); 1872 return 0; 1873 } 1874 1875 fs_initcall(ipv4_offload_init); 1876 1877 static struct packet_type ip_packet_type __read_mostly = { 1878 .type = cpu_to_be16(ETH_P_IP), 1879 .func = ip_rcv, 1880 .list_func = ip_list_rcv, 1881 }; 1882 1883 static int __init inet_init(void) 1884 { 1885 struct inet_protosw *q; 1886 struct list_head *r; 1887 int rc; 1888 1889 sock_skb_cb_check_size(sizeof(struct inet_skb_parm)); 1890 1891 raw_hashinfo_init(&raw_v4_hashinfo); 1892 1893 rc = proto_register(&tcp_prot, 1); 1894 if (rc) 1895 goto out; 1896 1897 rc = proto_register(&udp_prot, 1); 1898 if (rc) 1899 goto out_unregister_tcp_proto; 1900 1901 rc = proto_register(&raw_prot, 1); 1902 if (rc) 1903 goto out_unregister_udp_proto; 1904 1905 rc = proto_register(&ping_prot, 1); 1906 if (rc) 1907 goto out_unregister_raw_proto; 1908 1909 /* 1910 * Tell SOCKET that we are alive... 1911 */ 1912 1913 (void)sock_register(&inet_family_ops); 1914 1915 #ifdef CONFIG_SYSCTL 1916 ip_static_sysctl_init(); 1917 #endif 1918 1919 /* 1920 * Add all the base protocols. 1921 */ 1922 1923 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0) 1924 pr_crit("%s: Cannot add ICMP protocol\n", __func__); 1925 1926 net_hotdata.udp_protocol = (struct net_protocol) { 1927 .handler = udp_rcv, 1928 .err_handler = udp_err, 1929 .no_policy = 1, 1930 }; 1931 if (inet_add_protocol(&net_hotdata.udp_protocol, IPPROTO_UDP) < 0) 1932 pr_crit("%s: Cannot add UDP protocol\n", __func__); 1933 1934 net_hotdata.tcp_protocol = (struct net_protocol) { 1935 .handler = tcp_v4_rcv, 1936 .err_handler = tcp_v4_err, 1937 .no_policy = 1, 1938 .icmp_strict_tag_validation = 1, 1939 }; 1940 if (inet_add_protocol(&net_hotdata.tcp_protocol, IPPROTO_TCP) < 0) 1941 pr_crit("%s: Cannot add TCP protocol\n", __func__); 1942 #ifdef CONFIG_IP_MULTICAST 1943 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0) 1944 pr_crit("%s: Cannot add IGMP protocol\n", __func__); 1945 #endif 1946 1947 /* Register the socket-side information for inet_create. */ 1948 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r) 1949 INIT_LIST_HEAD(r); 1950 1951 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q) 1952 inet_register_protosw(q); 1953 1954 /* 1955 * Set the ARP module up 1956 */ 1957 1958 arp_init(); 1959 1960 /* 1961 * Set the IP module up 1962 */ 1963 1964 ip_init(); 1965 1966 /* Initialise per-cpu ipv4 mibs */ 1967 if (init_ipv4_mibs()) 1968 panic("%s: Cannot init ipv4 mibs\n", __func__); 1969 1970 /* Setup TCP slab cache for open requests. */ 1971 tcp_init(); 1972 1973 /* Setup UDP memory threshold */ 1974 udp_init(); 1975 1976 raw_init(); 1977 1978 ping_init(); 1979 1980 /* 1981 * Set the ICMP layer up 1982 */ 1983 1984 if (icmp_init() < 0) 1985 panic("Failed to create the ICMP control socket.\n"); 1986 1987 /* 1988 * Initialise the multicast router 1989 */ 1990 #if defined(CONFIG_IP_MROUTE) 1991 if (ip_mr_init()) 1992 pr_crit("%s: Cannot init ipv4 mroute\n", __func__); 1993 #endif 1994 1995 if (init_inet_pernet_ops()) 1996 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__); 1997 1998 ipv4_proc_init(); 1999 2000 ipfrag_init(); 2001 2002 dev_add_pack(&ip_packet_type); 2003 2004 ip_tunnel_core_init(); 2005 2006 rc = 0; 2007 out: 2008 return rc; 2009 out_unregister_raw_proto: 2010 proto_unregister(&raw_prot); 2011 out_unregister_udp_proto: 2012 proto_unregister(&udp_prot); 2013 out_unregister_tcp_proto: 2014 proto_unregister(&tcp_prot); 2015 goto out; 2016 } 2017 2018 fs_initcall(inet_init); 2019 2020 /* ------------------------------------------------------------------------ */ 2021 2022 #ifdef CONFIG_PROC_FS 2023 static int __init ipv4_proc_init(void) 2024 { 2025 int rc = 0; 2026 2027 if (raw_proc_init()) 2028 goto out_raw; 2029 if (tcp4_proc_init()) 2030 goto out_tcp; 2031 if (udp4_proc_init()) 2032 goto out_udp; 2033 if (ping_proc_init()) 2034 goto out_ping; 2035 if (ip_misc_proc_init()) 2036 goto out_misc; 2037 out: 2038 return rc; 2039 out_misc: 2040 ping_proc_exit(); 2041 out_ping: 2042 udp4_proc_exit(); 2043 out_udp: 2044 tcp4_proc_exit(); 2045 out_tcp: 2046 raw_proc_exit(); 2047 out_raw: 2048 rc = -ENOMEM; 2049 goto out; 2050 } 2051 2052 #else /* CONFIG_PROC_FS */ 2053 static int __init ipv4_proc_init(void) 2054 { 2055 return 0; 2056 } 2057 #endif /* CONFIG_PROC_FS */ 2058