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 if (mem_cgroup_sockets_enabled) { 759 gfp_t gfp = GFP_KERNEL | __GFP_NOFAIL; 760 761 mem_cgroup_sk_alloc(newsk); 762 763 if (mem_cgroup_from_sk(newsk)) { 764 int amt; 765 766 /* The socket has not been accepted yet, no need 767 * to look at newsk->sk_wmem_queued. 768 */ 769 amt = sk_mem_pages(newsk->sk_forward_alloc + 770 atomic_read(&newsk->sk_rmem_alloc)); 771 if (amt) 772 mem_cgroup_sk_charge(newsk, amt, gfp); 773 } 774 775 kmem_cache_charge(newsk, gfp); 776 } 777 778 sock_rps_record_flow(newsk); 779 WARN_ON(!((1 << newsk->sk_state) & 780 (TCPF_ESTABLISHED | TCPF_SYN_RECV | 781 TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2 | 782 TCPF_CLOSING | TCPF_CLOSE_WAIT | 783 TCPF_CLOSE))); 784 785 if (test_bit(SOCK_SUPPORT_ZC, &sock->flags)) 786 set_bit(SOCK_SUPPORT_ZC, &newsock->flags); 787 sock_graft(newsk, newsock); 788 789 newsock->state = SS_CONNECTED; 790 } 791 EXPORT_SYMBOL_GPL(__inet_accept); 792 793 /* 794 * Accept a pending connection. The TCP layer now gives BSD semantics. 795 */ 796 797 int inet_accept(struct socket *sock, struct socket *newsock, 798 struct proto_accept_arg *arg) 799 { 800 struct sock *sk1 = sock->sk, *sk2; 801 802 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 803 arg->err = -EINVAL; 804 sk2 = READ_ONCE(sk1->sk_prot)->accept(sk1, arg); 805 if (!sk2) 806 return arg->err; 807 808 lock_sock(sk2); 809 __inet_accept(sock, newsock, sk2); 810 release_sock(sk2); 811 return 0; 812 } 813 EXPORT_SYMBOL(inet_accept); 814 815 /* 816 * This does both peername and sockname. 817 */ 818 int inet_getname(struct socket *sock, struct sockaddr *uaddr, 819 int peer) 820 { 821 struct sock *sk = sock->sk; 822 struct inet_sock *inet = inet_sk(sk); 823 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr); 824 int sin_addr_len = sizeof(*sin); 825 826 sin->sin_family = AF_INET; 827 lock_sock(sk); 828 if (peer) { 829 if (!inet->inet_dport || 830 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) && 831 peer == 1)) { 832 release_sock(sk); 833 return -ENOTCONN; 834 } 835 sin->sin_port = inet->inet_dport; 836 sin->sin_addr.s_addr = inet->inet_daddr; 837 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin, &sin_addr_len, 838 CGROUP_INET4_GETPEERNAME); 839 } else { 840 __be32 addr = inet->inet_rcv_saddr; 841 if (!addr) 842 addr = inet->inet_saddr; 843 sin->sin_port = inet->inet_sport; 844 sin->sin_addr.s_addr = addr; 845 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin, &sin_addr_len, 846 CGROUP_INET4_GETSOCKNAME); 847 } 848 release_sock(sk); 849 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 850 return sin_addr_len; 851 } 852 EXPORT_SYMBOL(inet_getname); 853 854 int inet_send_prepare(struct sock *sk) 855 { 856 sock_rps_record_flow(sk); 857 858 /* We may need to bind the socket. */ 859 if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind && 860 inet_autobind(sk)) 861 return -EAGAIN; 862 863 return 0; 864 } 865 EXPORT_SYMBOL_GPL(inet_send_prepare); 866 867 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size) 868 { 869 struct sock *sk = sock->sk; 870 871 if (unlikely(inet_send_prepare(sk))) 872 return -EAGAIN; 873 874 return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg, 875 sk, msg, size); 876 } 877 EXPORT_SYMBOL(inet_sendmsg); 878 879 void inet_splice_eof(struct socket *sock) 880 { 881 const struct proto *prot; 882 struct sock *sk = sock->sk; 883 884 if (unlikely(inet_send_prepare(sk))) 885 return; 886 887 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 888 prot = READ_ONCE(sk->sk_prot); 889 if (prot->splice_eof) 890 prot->splice_eof(sock); 891 } 892 EXPORT_SYMBOL_GPL(inet_splice_eof); 893 894 INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *, 895 size_t, int, int *)); 896 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 897 int flags) 898 { 899 struct sock *sk = sock->sk; 900 int addr_len = 0; 901 int err; 902 903 if (likely(!(flags & MSG_ERRQUEUE))) 904 sock_rps_record_flow(sk); 905 906 err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg, 907 sk, msg, size, flags, &addr_len); 908 if (err >= 0) 909 msg->msg_namelen = addr_len; 910 return err; 911 } 912 EXPORT_SYMBOL(inet_recvmsg); 913 914 int inet_shutdown(struct socket *sock, int how) 915 { 916 struct sock *sk = sock->sk; 917 int err = 0; 918 919 /* This should really check to make sure 920 * the socket is a TCP socket. (WHY AC...) 921 */ 922 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and 923 1->2 bit 2 snds. 924 2->3 */ 925 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */ 926 return -EINVAL; 927 928 lock_sock(sk); 929 if (sock->state == SS_CONNECTING) { 930 if ((1 << sk->sk_state) & 931 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE)) 932 sock->state = SS_DISCONNECTING; 933 else 934 sock->state = SS_CONNECTED; 935 } 936 937 switch (sk->sk_state) { 938 case TCP_CLOSE: 939 err = -ENOTCONN; 940 /* Hack to wake up other listeners, who can poll for 941 EPOLLHUP, even on eg. unconnected UDP sockets -- RR */ 942 fallthrough; 943 default: 944 WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | how); 945 if (sk->sk_prot->shutdown) 946 sk->sk_prot->shutdown(sk, how); 947 break; 948 949 /* Remaining two branches are temporary solution for missing 950 * close() in multithreaded environment. It is _not_ a good idea, 951 * but we have no choice until close() is repaired at VFS level. 952 */ 953 case TCP_LISTEN: 954 if (!(how & RCV_SHUTDOWN)) 955 break; 956 fallthrough; 957 case TCP_SYN_SENT: 958 err = sk->sk_prot->disconnect(sk, O_NONBLOCK); 959 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED; 960 break; 961 } 962 963 /* Wake up anyone sleeping in poll. */ 964 sk->sk_state_change(sk); 965 release_sock(sk); 966 return err; 967 } 968 EXPORT_SYMBOL(inet_shutdown); 969 970 /* 971 * ioctl() calls you can issue on an INET socket. Most of these are 972 * device configuration and stuff and very rarely used. Some ioctls 973 * pass on to the socket itself. 974 * 975 * NOTE: I like the idea of a module for the config stuff. ie ifconfig 976 * loads the devconfigure module does its configuring and unloads it. 977 * There's a good 20K of config code hanging around the kernel. 978 */ 979 980 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 981 { 982 struct sock *sk = sock->sk; 983 int err = 0; 984 struct net *net = sock_net(sk); 985 void __user *p = (void __user *)arg; 986 struct ifreq ifr; 987 struct rtentry rt; 988 989 switch (cmd) { 990 case SIOCADDRT: 991 case SIOCDELRT: 992 if (copy_from_user(&rt, p, sizeof(struct rtentry))) 993 return -EFAULT; 994 err = ip_rt_ioctl(net, cmd, &rt); 995 break; 996 case SIOCRTMSG: 997 err = -EINVAL; 998 break; 999 case SIOCDARP: 1000 case SIOCGARP: 1001 case SIOCSARP: 1002 err = arp_ioctl(net, cmd, (void __user *)arg); 1003 break; 1004 case SIOCGIFADDR: 1005 case SIOCGIFBRDADDR: 1006 case SIOCGIFNETMASK: 1007 case SIOCGIFDSTADDR: 1008 case SIOCGIFPFLAGS: 1009 if (get_user_ifreq(&ifr, NULL, p)) 1010 return -EFAULT; 1011 err = devinet_ioctl(net, cmd, &ifr); 1012 if (!err && put_user_ifreq(&ifr, p)) 1013 err = -EFAULT; 1014 break; 1015 1016 case SIOCSIFADDR: 1017 case SIOCSIFBRDADDR: 1018 case SIOCSIFNETMASK: 1019 case SIOCSIFDSTADDR: 1020 case SIOCSIFPFLAGS: 1021 case SIOCSIFFLAGS: 1022 if (get_user_ifreq(&ifr, NULL, p)) 1023 return -EFAULT; 1024 err = devinet_ioctl(net, cmd, &ifr); 1025 break; 1026 default: 1027 if (sk->sk_prot->ioctl) 1028 err = sk_ioctl(sk, cmd, (void __user *)arg); 1029 else 1030 err = -ENOIOCTLCMD; 1031 break; 1032 } 1033 return err; 1034 } 1035 EXPORT_SYMBOL(inet_ioctl); 1036 1037 #ifdef CONFIG_COMPAT 1038 static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd, 1039 struct compat_rtentry __user *ur) 1040 { 1041 compat_uptr_t rtdev; 1042 struct rtentry rt; 1043 1044 if (copy_from_user(&rt.rt_dst, &ur->rt_dst, 1045 3 * sizeof(struct sockaddr)) || 1046 get_user(rt.rt_flags, &ur->rt_flags) || 1047 get_user(rt.rt_metric, &ur->rt_metric) || 1048 get_user(rt.rt_mtu, &ur->rt_mtu) || 1049 get_user(rt.rt_window, &ur->rt_window) || 1050 get_user(rt.rt_irtt, &ur->rt_irtt) || 1051 get_user(rtdev, &ur->rt_dev)) 1052 return -EFAULT; 1053 1054 rt.rt_dev = compat_ptr(rtdev); 1055 return ip_rt_ioctl(sock_net(sk), cmd, &rt); 1056 } 1057 1058 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 1059 { 1060 void __user *argp = compat_ptr(arg); 1061 struct sock *sk = sock->sk; 1062 1063 switch (cmd) { 1064 case SIOCADDRT: 1065 case SIOCDELRT: 1066 return inet_compat_routing_ioctl(sk, cmd, argp); 1067 default: 1068 if (!sk->sk_prot->compat_ioctl) 1069 return -ENOIOCTLCMD; 1070 return sk->sk_prot->compat_ioctl(sk, cmd, arg); 1071 } 1072 } 1073 #endif /* CONFIG_COMPAT */ 1074 1075 const struct proto_ops inet_stream_ops = { 1076 .family = PF_INET, 1077 .owner = THIS_MODULE, 1078 .release = inet_release, 1079 .bind = inet_bind, 1080 .connect = inet_stream_connect, 1081 .socketpair = sock_no_socketpair, 1082 .accept = inet_accept, 1083 .getname = inet_getname, 1084 .poll = tcp_poll, 1085 .ioctl = inet_ioctl, 1086 .gettstamp = sock_gettstamp, 1087 .listen = inet_listen, 1088 .shutdown = inet_shutdown, 1089 .setsockopt = sock_common_setsockopt, 1090 .getsockopt = sock_common_getsockopt, 1091 .sendmsg = inet_sendmsg, 1092 .recvmsg = inet_recvmsg, 1093 #ifdef CONFIG_MMU 1094 .mmap = tcp_mmap, 1095 #endif 1096 .splice_eof = inet_splice_eof, 1097 .splice_read = tcp_splice_read, 1098 .set_peek_off = sk_set_peek_off, 1099 .read_sock = tcp_read_sock, 1100 .read_skb = tcp_read_skb, 1101 .sendmsg_locked = tcp_sendmsg_locked, 1102 .peek_len = tcp_peek_len, 1103 #ifdef CONFIG_COMPAT 1104 .compat_ioctl = inet_compat_ioctl, 1105 #endif 1106 .set_rcvlowat = tcp_set_rcvlowat, 1107 }; 1108 EXPORT_SYMBOL(inet_stream_ops); 1109 1110 const struct proto_ops inet_dgram_ops = { 1111 .family = PF_INET, 1112 .owner = THIS_MODULE, 1113 .release = inet_release, 1114 .bind = inet_bind, 1115 .connect = inet_dgram_connect, 1116 .socketpair = sock_no_socketpair, 1117 .accept = sock_no_accept, 1118 .getname = inet_getname, 1119 .poll = udp_poll, 1120 .ioctl = inet_ioctl, 1121 .gettstamp = sock_gettstamp, 1122 .listen = sock_no_listen, 1123 .shutdown = inet_shutdown, 1124 .setsockopt = sock_common_setsockopt, 1125 .getsockopt = sock_common_getsockopt, 1126 .sendmsg = inet_sendmsg, 1127 .read_skb = udp_read_skb, 1128 .recvmsg = inet_recvmsg, 1129 .mmap = sock_no_mmap, 1130 .splice_eof = inet_splice_eof, 1131 .set_peek_off = udp_set_peek_off, 1132 #ifdef CONFIG_COMPAT 1133 .compat_ioctl = inet_compat_ioctl, 1134 #endif 1135 }; 1136 EXPORT_SYMBOL(inet_dgram_ops); 1137 1138 /* 1139 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without 1140 * udp_poll 1141 */ 1142 static const struct proto_ops inet_sockraw_ops = { 1143 .family = PF_INET, 1144 .owner = THIS_MODULE, 1145 .release = inet_release, 1146 .bind = inet_bind, 1147 .connect = inet_dgram_connect, 1148 .socketpair = sock_no_socketpair, 1149 .accept = sock_no_accept, 1150 .getname = inet_getname, 1151 .poll = datagram_poll, 1152 .ioctl = inet_ioctl, 1153 .gettstamp = sock_gettstamp, 1154 .listen = sock_no_listen, 1155 .shutdown = inet_shutdown, 1156 .setsockopt = sock_common_setsockopt, 1157 .getsockopt = sock_common_getsockopt, 1158 .sendmsg = inet_sendmsg, 1159 .recvmsg = inet_recvmsg, 1160 .mmap = sock_no_mmap, 1161 .splice_eof = inet_splice_eof, 1162 #ifdef CONFIG_COMPAT 1163 .compat_ioctl = inet_compat_ioctl, 1164 #endif 1165 }; 1166 1167 static const struct net_proto_family inet_family_ops = { 1168 .family = PF_INET, 1169 .create = inet_create, 1170 .owner = THIS_MODULE, 1171 }; 1172 1173 /* Upon startup we insert all the elements in inetsw_array[] into 1174 * the linked list inetsw. 1175 */ 1176 static struct inet_protosw inetsw_array[] = 1177 { 1178 { 1179 .type = SOCK_STREAM, 1180 .protocol = IPPROTO_TCP, 1181 .prot = &tcp_prot, 1182 .ops = &inet_stream_ops, 1183 .flags = INET_PROTOSW_PERMANENT | 1184 INET_PROTOSW_ICSK, 1185 }, 1186 1187 { 1188 .type = SOCK_DGRAM, 1189 .protocol = IPPROTO_UDP, 1190 .prot = &udp_prot, 1191 .ops = &inet_dgram_ops, 1192 .flags = INET_PROTOSW_PERMANENT, 1193 }, 1194 1195 { 1196 .type = SOCK_DGRAM, 1197 .protocol = IPPROTO_ICMP, 1198 .prot = &ping_prot, 1199 .ops = &inet_sockraw_ops, 1200 .flags = INET_PROTOSW_REUSE, 1201 }, 1202 1203 { 1204 .type = SOCK_RAW, 1205 .protocol = IPPROTO_IP, /* wild card */ 1206 .prot = &raw_prot, 1207 .ops = &inet_sockraw_ops, 1208 .flags = INET_PROTOSW_REUSE, 1209 } 1210 }; 1211 1212 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array) 1213 1214 void inet_register_protosw(struct inet_protosw *p) 1215 { 1216 struct list_head *lh; 1217 struct inet_protosw *answer; 1218 int protocol = p->protocol; 1219 struct list_head *last_perm; 1220 1221 spin_lock_bh(&inetsw_lock); 1222 1223 if (p->type >= SOCK_MAX) 1224 goto out_illegal; 1225 1226 /* If we are trying to override a permanent protocol, bail. */ 1227 last_perm = &inetsw[p->type]; 1228 list_for_each(lh, &inetsw[p->type]) { 1229 answer = list_entry(lh, struct inet_protosw, list); 1230 /* Check only the non-wild match. */ 1231 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0) 1232 break; 1233 if (protocol == answer->protocol) 1234 goto out_permanent; 1235 last_perm = lh; 1236 } 1237 1238 /* Add the new entry after the last permanent entry if any, so that 1239 * the new entry does not override a permanent entry when matched with 1240 * a wild-card protocol. But it is allowed to override any existing 1241 * non-permanent entry. This means that when we remove this entry, the 1242 * system automatically returns to the old behavior. 1243 */ 1244 list_add_rcu(&p->list, last_perm); 1245 out: 1246 spin_unlock_bh(&inetsw_lock); 1247 1248 return; 1249 1250 out_permanent: 1251 pr_err("Attempt to override permanent protocol %d\n", protocol); 1252 goto out; 1253 1254 out_illegal: 1255 pr_err("Ignoring attempt to register invalid socket type %d\n", 1256 p->type); 1257 goto out; 1258 } 1259 EXPORT_SYMBOL(inet_register_protosw); 1260 1261 void inet_unregister_protosw(struct inet_protosw *p) 1262 { 1263 if (INET_PROTOSW_PERMANENT & p->flags) { 1264 pr_err("Attempt to unregister permanent protocol %d\n", 1265 p->protocol); 1266 } else { 1267 spin_lock_bh(&inetsw_lock); 1268 list_del_rcu(&p->list); 1269 spin_unlock_bh(&inetsw_lock); 1270 1271 synchronize_net(); 1272 } 1273 } 1274 EXPORT_SYMBOL(inet_unregister_protosw); 1275 1276 static int inet_sk_reselect_saddr(struct sock *sk) 1277 { 1278 struct inet_sock *inet = inet_sk(sk); 1279 __be32 old_saddr = inet->inet_saddr; 1280 __be32 daddr = inet->inet_daddr; 1281 struct flowi4 *fl4; 1282 struct rtable *rt; 1283 __be32 new_saddr; 1284 struct ip_options_rcu *inet_opt; 1285 int err; 1286 1287 inet_opt = rcu_dereference_protected(inet->inet_opt, 1288 lockdep_sock_is_held(sk)); 1289 if (inet_opt && inet_opt->opt.srr) 1290 daddr = inet_opt->opt.faddr; 1291 1292 /* Query new route. */ 1293 fl4 = &inet->cork.fl.u.ip4; 1294 rt = ip_route_connect(fl4, daddr, 0, sk->sk_bound_dev_if, 1295 sk->sk_protocol, inet->inet_sport, 1296 inet->inet_dport, sk); 1297 if (IS_ERR(rt)) 1298 return PTR_ERR(rt); 1299 1300 new_saddr = fl4->saddr; 1301 1302 if (new_saddr == old_saddr) { 1303 sk_setup_caps(sk, &rt->dst); 1304 return 0; 1305 } 1306 1307 err = inet_bhash2_update_saddr(sk, &new_saddr, AF_INET); 1308 if (err) { 1309 ip_rt_put(rt); 1310 return err; 1311 } 1312 1313 sk_setup_caps(sk, &rt->dst); 1314 1315 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) > 1) { 1316 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n", 1317 __func__, &old_saddr, &new_saddr); 1318 } 1319 1320 /* 1321 * XXX The only one ugly spot where we need to 1322 * XXX really change the sockets identity after 1323 * XXX it has entered the hashes. -DaveM 1324 * 1325 * Besides that, it does not check for connection 1326 * uniqueness. Wait for troubles. 1327 */ 1328 return __sk_prot_rehash(sk); 1329 } 1330 1331 int inet_sk_rebuild_header(struct sock *sk) 1332 { 1333 struct rtable *rt = dst_rtable(__sk_dst_check(sk, 0)); 1334 struct inet_sock *inet = inet_sk(sk); 1335 struct flowi4 *fl4; 1336 int err; 1337 1338 /* Route is OK, nothing to do. */ 1339 if (rt) 1340 return 0; 1341 1342 /* Reroute. */ 1343 fl4 = &inet->cork.fl.u.ip4; 1344 inet_sk_init_flowi4(inet, fl4); 1345 rt = ip_route_output_flow(sock_net(sk), fl4, sk); 1346 if (!IS_ERR(rt)) { 1347 err = 0; 1348 sk_setup_caps(sk, &rt->dst); 1349 } else { 1350 err = PTR_ERR(rt); 1351 1352 /* Routing failed... */ 1353 sk->sk_route_caps = 0; 1354 1355 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) || 1356 sk->sk_state != TCP_SYN_SENT || 1357 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) || 1358 (err = inet_sk_reselect_saddr(sk)) != 0) 1359 WRITE_ONCE(sk->sk_err_soft, -err); 1360 } 1361 1362 return err; 1363 } 1364 EXPORT_SYMBOL(inet_sk_rebuild_header); 1365 1366 void inet_sk_set_state(struct sock *sk, int state) 1367 { 1368 trace_inet_sock_set_state(sk, sk->sk_state, state); 1369 sk->sk_state = state; 1370 } 1371 EXPORT_SYMBOL(inet_sk_set_state); 1372 1373 void inet_sk_state_store(struct sock *sk, int newstate) 1374 { 1375 trace_inet_sock_set_state(sk, sk->sk_state, newstate); 1376 smp_store_release(&sk->sk_state, newstate); 1377 } 1378 1379 struct sk_buff *inet_gso_segment(struct sk_buff *skb, 1380 netdev_features_t features) 1381 { 1382 bool udpfrag = false, fixedid = false, gso_partial, encap; 1383 struct sk_buff *segs = ERR_PTR(-EINVAL); 1384 const struct net_offload *ops; 1385 unsigned int offset = 0; 1386 struct iphdr *iph; 1387 int proto, tot_len; 1388 int nhoff; 1389 int ihl; 1390 int id; 1391 1392 skb_reset_network_header(skb); 1393 nhoff = skb_network_header(skb) - skb_mac_header(skb); 1394 if (unlikely(!pskb_may_pull(skb, sizeof(*iph)))) 1395 goto out; 1396 1397 iph = ip_hdr(skb); 1398 ihl = iph->ihl * 4; 1399 if (ihl < sizeof(*iph)) 1400 goto out; 1401 1402 id = ntohs(iph->id); 1403 proto = iph->protocol; 1404 1405 /* Warning: after this point, iph might be no longer valid */ 1406 if (unlikely(!pskb_may_pull(skb, ihl))) 1407 goto out; 1408 __skb_pull(skb, ihl); 1409 1410 encap = SKB_GSO_CB(skb)->encap_level > 0; 1411 if (encap) 1412 features &= skb->dev->hw_enc_features; 1413 SKB_GSO_CB(skb)->encap_level += ihl; 1414 1415 skb_reset_transport_header(skb); 1416 1417 segs = ERR_PTR(-EPROTONOSUPPORT); 1418 1419 fixedid = !!(skb_shinfo(skb)->gso_type & (SKB_GSO_TCP_FIXEDID << encap)); 1420 1421 if (!skb->encapsulation || encap) 1422 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP); 1423 1424 ops = rcu_dereference(inet_offloads[proto]); 1425 if (likely(ops && ops->callbacks.gso_segment)) { 1426 segs = ops->callbacks.gso_segment(skb, features); 1427 if (!segs) 1428 skb->network_header = skb_mac_header(skb) + nhoff - skb->head; 1429 } 1430 1431 if (IS_ERR_OR_NULL(segs)) 1432 goto out; 1433 1434 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL); 1435 1436 skb = segs; 1437 do { 1438 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff); 1439 if (udpfrag) { 1440 iph->frag_off = htons(offset >> 3); 1441 if (skb->next) 1442 iph->frag_off |= htons(IP_MF); 1443 offset += skb->len - nhoff - ihl; 1444 tot_len = skb->len - nhoff; 1445 } else if (skb_is_gso(skb)) { 1446 if (!fixedid) { 1447 iph->id = htons(id); 1448 id += skb_shinfo(skb)->gso_segs; 1449 } 1450 1451 if (gso_partial) 1452 tot_len = skb_shinfo(skb)->gso_size + 1453 SKB_GSO_CB(skb)->data_offset + 1454 skb->head - (unsigned char *)iph; 1455 else 1456 tot_len = skb->len - nhoff; 1457 } else { 1458 if (!fixedid) 1459 iph->id = htons(id++); 1460 tot_len = skb->len - nhoff; 1461 } 1462 iph->tot_len = htons(tot_len); 1463 ip_send_check(iph); 1464 if (encap) 1465 skb_reset_inner_headers(skb); 1466 skb->network_header = (u8 *)iph - skb->head; 1467 skb_reset_mac_len(skb); 1468 } while ((skb = skb->next)); 1469 1470 out: 1471 return segs; 1472 } 1473 1474 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb, 1475 netdev_features_t features) 1476 { 1477 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4)) 1478 return ERR_PTR(-EINVAL); 1479 1480 return inet_gso_segment(skb, features); 1481 } 1482 1483 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb) 1484 { 1485 const struct net_offload *ops; 1486 struct sk_buff *pp = NULL; 1487 const struct iphdr *iph; 1488 struct sk_buff *p; 1489 unsigned int hlen; 1490 unsigned int off; 1491 int flush = 1; 1492 int proto; 1493 1494 off = skb_gro_offset(skb); 1495 hlen = off + sizeof(*iph); 1496 iph = skb_gro_header(skb, hlen, off); 1497 if (unlikely(!iph)) 1498 goto out; 1499 1500 proto = iph->protocol; 1501 1502 ops = rcu_dereference(inet_offloads[proto]); 1503 if (!ops || !ops->callbacks.gro_receive) 1504 goto out; 1505 1506 if (*(u8 *)iph != 0x45) 1507 goto out; 1508 1509 if (ip_is_fragment(iph)) 1510 goto out; 1511 1512 if (unlikely(ip_fast_csum((u8 *)iph, 5))) 1513 goto out; 1514 1515 NAPI_GRO_CB(skb)->proto = proto; 1516 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (ntohl(*(__be32 *)&iph->id) & ~IP_DF)); 1517 1518 list_for_each_entry(p, head, list) { 1519 struct iphdr *iph2; 1520 1521 if (!NAPI_GRO_CB(p)->same_flow) 1522 continue; 1523 1524 iph2 = (struct iphdr *)(p->data + off); 1525 /* The above works because, with the exception of the top 1526 * (inner most) layer, we only aggregate pkts with the same 1527 * hdr length so all the hdrs we'll need to verify will start 1528 * at the same offset. 1529 */ 1530 if ((iph->protocol ^ iph2->protocol) | 1531 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) | 1532 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) { 1533 NAPI_GRO_CB(p)->same_flow = 0; 1534 continue; 1535 } 1536 } 1537 1538 NAPI_GRO_CB(skb)->flush |= flush; 1539 NAPI_GRO_CB(skb)->network_offsets[NAPI_GRO_CB(skb)->encap_mark] = off; 1540 1541 /* Note : No need to call skb_gro_postpull_rcsum() here, 1542 * as we already checked checksum over ipv4 header was 0 1543 */ 1544 skb_gro_pull(skb, sizeof(*iph)); 1545 skb_set_transport_header(skb, skb_gro_offset(skb)); 1546 1547 pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive, 1548 ops->callbacks.gro_receive, head, skb); 1549 1550 out: 1551 skb_gro_flush_final(skb, pp, flush); 1552 1553 return pp; 1554 } 1555 1556 static struct sk_buff *ipip_gro_receive(struct list_head *head, 1557 struct sk_buff *skb) 1558 { 1559 if (NAPI_GRO_CB(skb)->encap_mark) { 1560 NAPI_GRO_CB(skb)->flush = 1; 1561 return NULL; 1562 } 1563 1564 NAPI_GRO_CB(skb)->encap_mark = 1; 1565 1566 return inet_gro_receive(head, skb); 1567 } 1568 1569 #define SECONDS_PER_DAY 86400 1570 1571 /* inet_current_timestamp - Return IP network timestamp 1572 * 1573 * Return milliseconds since midnight in network byte order. 1574 */ 1575 __be32 inet_current_timestamp(void) 1576 { 1577 u32 secs; 1578 u32 msecs; 1579 struct timespec64 ts; 1580 1581 ktime_get_real_ts64(&ts); 1582 1583 /* Get secs since midnight. */ 1584 (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs); 1585 /* Convert to msecs. */ 1586 msecs = secs * MSEC_PER_SEC; 1587 /* Convert nsec to msec. */ 1588 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC; 1589 1590 /* Convert to network byte order. */ 1591 return htonl(msecs); 1592 } 1593 EXPORT_SYMBOL(inet_current_timestamp); 1594 1595 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len) 1596 { 1597 unsigned int family = READ_ONCE(sk->sk_family); 1598 1599 if (family == AF_INET) 1600 return ip_recv_error(sk, msg, len, addr_len); 1601 #if IS_ENABLED(CONFIG_IPV6) 1602 if (family == AF_INET6) 1603 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len); 1604 #endif 1605 return -EINVAL; 1606 } 1607 EXPORT_SYMBOL(inet_recv_error); 1608 1609 int inet_gro_complete(struct sk_buff *skb, int nhoff) 1610 { 1611 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff); 1612 const struct net_offload *ops; 1613 __be16 totlen = iph->tot_len; 1614 int proto = iph->protocol; 1615 int err = -ENOSYS; 1616 1617 if (skb->encapsulation) { 1618 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP)); 1619 skb_set_inner_network_header(skb, nhoff); 1620 } 1621 1622 iph_set_totlen(iph, skb->len - nhoff); 1623 csum_replace2(&iph->check, totlen, iph->tot_len); 1624 1625 ops = rcu_dereference(inet_offloads[proto]); 1626 if (WARN_ON(!ops || !ops->callbacks.gro_complete)) 1627 goto out; 1628 1629 /* Only need to add sizeof(*iph) to get to the next hdr below 1630 * because any hdr with option will have been flushed in 1631 * inet_gro_receive(). 1632 */ 1633 err = INDIRECT_CALL_2(ops->callbacks.gro_complete, 1634 tcp4_gro_complete, udp4_gro_complete, 1635 skb, nhoff + sizeof(*iph)); 1636 1637 out: 1638 return err; 1639 } 1640 1641 static int ipip_gro_complete(struct sk_buff *skb, int nhoff) 1642 { 1643 skb->encapsulation = 1; 1644 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4; 1645 return inet_gro_complete(skb, nhoff); 1646 } 1647 1648 int inet_ctl_sock_create(struct sock **sk, unsigned short family, 1649 unsigned short type, unsigned char protocol, 1650 struct net *net) 1651 { 1652 struct socket *sock; 1653 int rc = sock_create_kern(net, family, type, protocol, &sock); 1654 1655 if (rc == 0) { 1656 *sk = sock->sk; 1657 (*sk)->sk_allocation = GFP_ATOMIC; 1658 (*sk)->sk_use_task_frag = false; 1659 /* 1660 * Unhash it so that IP input processing does not even see it, 1661 * we do not wish this socket to see incoming packets. 1662 */ 1663 (*sk)->sk_prot->unhash(*sk); 1664 } 1665 return rc; 1666 } 1667 EXPORT_SYMBOL_GPL(inet_ctl_sock_create); 1668 1669 unsigned long snmp_fold_field(void __percpu *mib, int offt) 1670 { 1671 unsigned long res = 0; 1672 int i; 1673 1674 for_each_possible_cpu(i) 1675 res += snmp_get_cpu_field(mib, i, offt); 1676 return res; 1677 } 1678 EXPORT_SYMBOL_GPL(snmp_fold_field); 1679 1680 #if BITS_PER_LONG==32 1681 1682 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt, 1683 size_t syncp_offset) 1684 { 1685 void *bhptr; 1686 struct u64_stats_sync *syncp; 1687 u64 v; 1688 unsigned int start; 1689 1690 bhptr = per_cpu_ptr(mib, cpu); 1691 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset); 1692 do { 1693 start = u64_stats_fetch_begin(syncp); 1694 v = *(((u64 *)bhptr) + offt); 1695 } while (u64_stats_fetch_retry(syncp, start)); 1696 1697 return v; 1698 } 1699 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64); 1700 1701 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset) 1702 { 1703 u64 res = 0; 1704 int cpu; 1705 1706 for_each_possible_cpu(cpu) { 1707 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset); 1708 } 1709 return res; 1710 } 1711 EXPORT_SYMBOL_GPL(snmp_fold_field64); 1712 #endif 1713 1714 #ifdef CONFIG_IP_MULTICAST 1715 static const struct net_protocol igmp_protocol = { 1716 .handler = igmp_rcv, 1717 }; 1718 #endif 1719 1720 static const struct net_protocol icmp_protocol = { 1721 .handler = icmp_rcv, 1722 .err_handler = icmp_err, 1723 .no_policy = 1, 1724 }; 1725 1726 static __net_init int ipv4_mib_init_net(struct net *net) 1727 { 1728 int i; 1729 1730 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib); 1731 if (!net->mib.tcp_statistics) 1732 goto err_tcp_mib; 1733 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib); 1734 if (!net->mib.ip_statistics) 1735 goto err_ip_mib; 1736 1737 for_each_possible_cpu(i) { 1738 struct ipstats_mib *af_inet_stats; 1739 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i); 1740 u64_stats_init(&af_inet_stats->syncp); 1741 } 1742 1743 net->mib.net_statistics = alloc_percpu(struct linux_mib); 1744 if (!net->mib.net_statistics) 1745 goto err_net_mib; 1746 net->mib.udp_statistics = alloc_percpu(struct udp_mib); 1747 if (!net->mib.udp_statistics) 1748 goto err_udp_mib; 1749 net->mib.udplite_statistics = alloc_percpu(struct udp_mib); 1750 if (!net->mib.udplite_statistics) 1751 goto err_udplite_mib; 1752 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib); 1753 if (!net->mib.icmp_statistics) 1754 goto err_icmp_mib; 1755 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib), 1756 GFP_KERNEL); 1757 if (!net->mib.icmpmsg_statistics) 1758 goto err_icmpmsg_mib; 1759 1760 tcp_mib_init(net); 1761 return 0; 1762 1763 err_icmpmsg_mib: 1764 free_percpu(net->mib.icmp_statistics); 1765 err_icmp_mib: 1766 free_percpu(net->mib.udplite_statistics); 1767 err_udplite_mib: 1768 free_percpu(net->mib.udp_statistics); 1769 err_udp_mib: 1770 free_percpu(net->mib.net_statistics); 1771 err_net_mib: 1772 free_percpu(net->mib.ip_statistics); 1773 err_ip_mib: 1774 free_percpu(net->mib.tcp_statistics); 1775 err_tcp_mib: 1776 return -ENOMEM; 1777 } 1778 1779 static __net_exit void ipv4_mib_exit_net(struct net *net) 1780 { 1781 kfree(net->mib.icmpmsg_statistics); 1782 free_percpu(net->mib.icmp_statistics); 1783 free_percpu(net->mib.udplite_statistics); 1784 free_percpu(net->mib.udp_statistics); 1785 free_percpu(net->mib.net_statistics); 1786 free_percpu(net->mib.ip_statistics); 1787 free_percpu(net->mib.tcp_statistics); 1788 #ifdef CONFIG_MPTCP 1789 /* allocated on demand, see mptcp_init_sock() */ 1790 free_percpu(net->mib.mptcp_statistics); 1791 #endif 1792 } 1793 1794 static __net_initdata struct pernet_operations ipv4_mib_ops = { 1795 .init = ipv4_mib_init_net, 1796 .exit = ipv4_mib_exit_net, 1797 }; 1798 1799 static int __init init_ipv4_mibs(void) 1800 { 1801 return register_pernet_subsys(&ipv4_mib_ops); 1802 } 1803 1804 static __net_init int inet_init_net(struct net *net) 1805 { 1806 /* 1807 * Set defaults for local port range 1808 */ 1809 net->ipv4.ip_local_ports.range = 60999u << 16 | 32768u; 1810 1811 seqlock_init(&net->ipv4.ping_group_range.lock); 1812 /* 1813 * Sane defaults - nobody may create ping sockets. 1814 * Boot scripts should set this to distro-specific group. 1815 */ 1816 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1); 1817 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0); 1818 1819 /* Default values for sysctl-controlled parameters. 1820 * We set them here, in case sysctl is not compiled. 1821 */ 1822 net->ipv4.sysctl_ip_default_ttl = IPDEFTTL; 1823 net->ipv4.sysctl_ip_fwd_update_priority = 1; 1824 net->ipv4.sysctl_ip_dynaddr = 0; 1825 net->ipv4.sysctl_ip_early_demux = 1; 1826 net->ipv4.sysctl_udp_early_demux = 1; 1827 net->ipv4.sysctl_tcp_early_demux = 1; 1828 net->ipv4.sysctl_nexthop_compat_mode = 1; 1829 #ifdef CONFIG_SYSCTL 1830 net->ipv4.sysctl_ip_prot_sock = PROT_SOCK; 1831 #endif 1832 1833 /* Some igmp sysctl, whose values are always used */ 1834 net->ipv4.sysctl_igmp_max_memberships = 20; 1835 net->ipv4.sysctl_igmp_max_msf = 10; 1836 /* IGMP reports for link-local multicast groups are enabled by default */ 1837 net->ipv4.sysctl_igmp_llm_reports = 1; 1838 net->ipv4.sysctl_igmp_qrv = 2; 1839 1840 net->ipv4.sysctl_fib_notify_on_flag_change = 0; 1841 1842 return 0; 1843 } 1844 1845 static __net_initdata struct pernet_operations af_inet_ops = { 1846 .init = inet_init_net, 1847 }; 1848 1849 static int __init init_inet_pernet_ops(void) 1850 { 1851 return register_pernet_subsys(&af_inet_ops); 1852 } 1853 1854 static int ipv4_proc_init(void); 1855 1856 /* 1857 * IP protocol layer initialiser 1858 */ 1859 1860 1861 static const struct net_offload ipip_offload = { 1862 .callbacks = { 1863 .gso_segment = ipip_gso_segment, 1864 .gro_receive = ipip_gro_receive, 1865 .gro_complete = ipip_gro_complete, 1866 }, 1867 }; 1868 1869 static int __init ipip_offload_init(void) 1870 { 1871 return inet_add_offload(&ipip_offload, IPPROTO_IPIP); 1872 } 1873 1874 static int __init ipv4_offload_init(void) 1875 { 1876 /* 1877 * Add offloads 1878 */ 1879 if (udpv4_offload_init() < 0) 1880 pr_crit("%s: Cannot add UDP protocol offload\n", __func__); 1881 if (tcpv4_offload_init() < 0) 1882 pr_crit("%s: Cannot add TCP protocol offload\n", __func__); 1883 if (ipip_offload_init() < 0) 1884 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__); 1885 1886 net_hotdata.ip_packet_offload = (struct packet_offload) { 1887 .type = cpu_to_be16(ETH_P_IP), 1888 .callbacks = { 1889 .gso_segment = inet_gso_segment, 1890 .gro_receive = inet_gro_receive, 1891 .gro_complete = inet_gro_complete, 1892 }, 1893 }; 1894 dev_add_offload(&net_hotdata.ip_packet_offload); 1895 return 0; 1896 } 1897 1898 fs_initcall(ipv4_offload_init); 1899 1900 static struct packet_type ip_packet_type __read_mostly = { 1901 .type = cpu_to_be16(ETH_P_IP), 1902 .func = ip_rcv, 1903 .list_func = ip_list_rcv, 1904 }; 1905 1906 static int __init inet_init(void) 1907 { 1908 struct inet_protosw *q; 1909 struct list_head *r; 1910 int rc; 1911 1912 sock_skb_cb_check_size(sizeof(struct inet_skb_parm)); 1913 1914 raw_hashinfo_init(&raw_v4_hashinfo); 1915 1916 rc = proto_register(&tcp_prot, 1); 1917 if (rc) 1918 goto out; 1919 1920 rc = proto_register(&udp_prot, 1); 1921 if (rc) 1922 goto out_unregister_tcp_proto; 1923 1924 rc = proto_register(&raw_prot, 1); 1925 if (rc) 1926 goto out_unregister_udp_proto; 1927 1928 rc = proto_register(&ping_prot, 1); 1929 if (rc) 1930 goto out_unregister_raw_proto; 1931 1932 /* 1933 * Tell SOCKET that we are alive... 1934 */ 1935 1936 (void)sock_register(&inet_family_ops); 1937 1938 #ifdef CONFIG_SYSCTL 1939 ip_static_sysctl_init(); 1940 #endif 1941 1942 /* 1943 * Add all the base protocols. 1944 */ 1945 1946 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0) 1947 pr_crit("%s: Cannot add ICMP protocol\n", __func__); 1948 1949 net_hotdata.udp_protocol = (struct net_protocol) { 1950 .handler = udp_rcv, 1951 .err_handler = udp_err, 1952 .no_policy = 1, 1953 }; 1954 if (inet_add_protocol(&net_hotdata.udp_protocol, IPPROTO_UDP) < 0) 1955 pr_crit("%s: Cannot add UDP protocol\n", __func__); 1956 1957 net_hotdata.tcp_protocol = (struct net_protocol) { 1958 .handler = tcp_v4_rcv, 1959 .err_handler = tcp_v4_err, 1960 .no_policy = 1, 1961 .icmp_strict_tag_validation = 1, 1962 }; 1963 if (inet_add_protocol(&net_hotdata.tcp_protocol, IPPROTO_TCP) < 0) 1964 pr_crit("%s: Cannot add TCP protocol\n", __func__); 1965 #ifdef CONFIG_IP_MULTICAST 1966 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0) 1967 pr_crit("%s: Cannot add IGMP protocol\n", __func__); 1968 #endif 1969 1970 /* Register the socket-side information for inet_create. */ 1971 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r) 1972 INIT_LIST_HEAD(r); 1973 1974 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q) 1975 inet_register_protosw(q); 1976 1977 /* 1978 * Set the ARP module up 1979 */ 1980 1981 arp_init(); 1982 1983 /* 1984 * Set the IP module up 1985 */ 1986 1987 ip_init(); 1988 1989 /* Initialise per-cpu ipv4 mibs */ 1990 if (init_ipv4_mibs()) 1991 panic("%s: Cannot init ipv4 mibs\n", __func__); 1992 1993 /* Setup TCP slab cache for open requests. */ 1994 tcp_init(); 1995 1996 /* Setup UDP memory threshold */ 1997 udp_init(); 1998 1999 /* Add UDP-Lite (RFC 3828) */ 2000 udplite4_register(); 2001 2002 raw_init(); 2003 2004 ping_init(); 2005 2006 /* 2007 * Set the ICMP layer up 2008 */ 2009 2010 if (icmp_init() < 0) 2011 panic("Failed to create the ICMP control socket.\n"); 2012 2013 /* 2014 * Initialise the multicast router 2015 */ 2016 #if defined(CONFIG_IP_MROUTE) 2017 if (ip_mr_init()) 2018 pr_crit("%s: Cannot init ipv4 mroute\n", __func__); 2019 #endif 2020 2021 if (init_inet_pernet_ops()) 2022 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__); 2023 2024 ipv4_proc_init(); 2025 2026 ipfrag_init(); 2027 2028 dev_add_pack(&ip_packet_type); 2029 2030 ip_tunnel_core_init(); 2031 2032 rc = 0; 2033 out: 2034 return rc; 2035 out_unregister_raw_proto: 2036 proto_unregister(&raw_prot); 2037 out_unregister_udp_proto: 2038 proto_unregister(&udp_prot); 2039 out_unregister_tcp_proto: 2040 proto_unregister(&tcp_prot); 2041 goto out; 2042 } 2043 2044 fs_initcall(inet_init); 2045 2046 /* ------------------------------------------------------------------------ */ 2047 2048 #ifdef CONFIG_PROC_FS 2049 static int __init ipv4_proc_init(void) 2050 { 2051 int rc = 0; 2052 2053 if (raw_proc_init()) 2054 goto out_raw; 2055 if (tcp4_proc_init()) 2056 goto out_tcp; 2057 if (udp4_proc_init()) 2058 goto out_udp; 2059 if (ping_proc_init()) 2060 goto out_ping; 2061 if (ip_misc_proc_init()) 2062 goto out_misc; 2063 out: 2064 return rc; 2065 out_misc: 2066 ping_proc_exit(); 2067 out_ping: 2068 udp4_proc_exit(); 2069 out_udp: 2070 tcp4_proc_exit(); 2071 out_tcp: 2072 raw_proc_exit(); 2073 out_raw: 2074 rc = -ENOMEM; 2075 goto out; 2076 } 2077 2078 #else /* CONFIG_PROC_FS */ 2079 static int __init ipv4_proc_init(void) 2080 { 2081 return 0; 2082 } 2083 #endif /* CONFIG_PROC_FS */ 2084