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/udp.h> 106 #include <net/udplite.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> 122 123 #include <trace/events/sock.h> 124 125 /* The inetsw table contains everything that inet_create needs to 126 * build a new socket. 127 */ 128 static struct list_head inetsw[SOCK_MAX]; 129 static DEFINE_SPINLOCK(inetsw_lock); 130 131 /* New destruction routine */ 132 133 void inet_sock_destruct(struct sock *sk) 134 { 135 struct inet_sock *inet = inet_sk(sk); 136 137 __skb_queue_purge(&sk->sk_receive_queue); 138 __skb_queue_purge(&sk->sk_error_queue); 139 140 sk_mem_reclaim_final(sk); 141 142 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) { 143 pr_err("Attempt to release TCP socket in state %d %p\n", 144 sk->sk_state, sk); 145 return; 146 } 147 if (!sock_flag(sk, SOCK_DEAD)) { 148 pr_err("Attempt to release alive inet socket %p\n", sk); 149 return; 150 } 151 152 WARN_ON_ONCE(atomic_read(&sk->sk_rmem_alloc)); 153 WARN_ON_ONCE(refcount_read(&sk->sk_wmem_alloc)); 154 WARN_ON_ONCE(sk->sk_wmem_queued); 155 WARN_ON_ONCE(sk_forward_alloc_get(sk)); 156 157 kfree(rcu_dereference_protected(inet->inet_opt, 1)); 158 dst_release(rcu_dereference_protected(sk->sk_dst_cache, 1)); 159 dst_release(rcu_dereference_protected(sk->sk_rx_dst, 1)); 160 } 161 EXPORT_SYMBOL(inet_sock_destruct); 162 163 /* 164 * The routines beyond this point handle the behaviour of an AF_INET 165 * socket object. Mostly it punts to the subprotocols of IP to do 166 * the work. 167 */ 168 169 /* 170 * Automatically bind an unbound socket. 171 */ 172 173 static int inet_autobind(struct sock *sk) 174 { 175 struct inet_sock *inet; 176 /* We may need to bind the socket. */ 177 lock_sock(sk); 178 inet = inet_sk(sk); 179 if (!inet->inet_num) { 180 if (sk->sk_prot->get_port(sk, 0)) { 181 release_sock(sk); 182 return -EAGAIN; 183 } 184 inet->inet_sport = htons(inet->inet_num); 185 } 186 release_sock(sk); 187 return 0; 188 } 189 190 int __inet_listen_sk(struct sock *sk, int backlog) 191 { 192 unsigned char old_state = sk->sk_state; 193 int err, tcp_fastopen; 194 195 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN))) 196 return -EINVAL; 197 198 WRITE_ONCE(sk->sk_max_ack_backlog, backlog); 199 /* Really, if the socket is already in listen state 200 * we can only allow the backlog to be adjusted. 201 */ 202 if (old_state != TCP_LISTEN) { 203 /* Enable TFO w/o requiring TCP_FASTOPEN socket option. 204 * Note that only TCP sockets (SOCK_STREAM) will reach here. 205 * Also fastopen backlog may already been set via the option 206 * because the socket was in TCP_LISTEN state previously but 207 * was shutdown() rather than close(). 208 */ 209 tcp_fastopen = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fastopen); 210 if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) && 211 (tcp_fastopen & TFO_SERVER_ENABLE) && 212 !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) { 213 fastopen_queue_tune(sk, backlog); 214 tcp_fastopen_init_key_once(sock_net(sk)); 215 } 216 217 err = inet_csk_listen_start(sk); 218 if (err) 219 return err; 220 221 tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL); 222 } 223 return 0; 224 } 225 226 /* 227 * Move a socket into listening state. 228 */ 229 int inet_listen(struct socket *sock, int backlog) 230 { 231 struct sock *sk = sock->sk; 232 int err = -EINVAL; 233 234 lock_sock(sk); 235 236 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM) 237 goto out; 238 239 err = __inet_listen_sk(sk, backlog); 240 241 out: 242 release_sock(sk); 243 return err; 244 } 245 EXPORT_SYMBOL(inet_listen); 246 247 /* 248 * Create an inet socket. 249 */ 250 251 static int inet_create(struct net *net, struct socket *sock, int protocol, 252 int kern) 253 { 254 struct sock *sk; 255 struct inet_protosw *answer; 256 struct inet_sock *inet; 257 struct proto *answer_prot; 258 unsigned char answer_flags; 259 int try_loading_module = 0; 260 int err; 261 262 if (protocol < 0 || protocol >= IPPROTO_MAX) 263 return -EINVAL; 264 265 sock->state = SS_UNCONNECTED; 266 267 /* Look for the requested type/protocol pair. */ 268 lookup_protocol: 269 err = -ESOCKTNOSUPPORT; 270 rcu_read_lock(); 271 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) { 272 273 err = 0; 274 /* Check the non-wild match. */ 275 if (protocol == answer->protocol) { 276 if (protocol != IPPROTO_IP) 277 break; 278 } else { 279 /* Check for the two wild cases. */ 280 if (IPPROTO_IP == protocol) { 281 protocol = answer->protocol; 282 break; 283 } 284 if (IPPROTO_IP == answer->protocol) 285 break; 286 } 287 err = -EPROTONOSUPPORT; 288 } 289 290 if (unlikely(err)) { 291 if (try_loading_module < 2) { 292 rcu_read_unlock(); 293 /* 294 * Be more specific, e.g. net-pf-2-proto-132-type-1 295 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM) 296 */ 297 if (++try_loading_module == 1) 298 request_module("net-pf-%d-proto-%d-type-%d", 299 PF_INET, protocol, sock->type); 300 /* 301 * Fall back to generic, e.g. net-pf-2-proto-132 302 * (net-pf-PF_INET-proto-IPPROTO_SCTP) 303 */ 304 else 305 request_module("net-pf-%d-proto-%d", 306 PF_INET, protocol); 307 goto lookup_protocol; 308 } else 309 goto out_rcu_unlock; 310 } 311 312 err = -EPERM; 313 if (sock->type == SOCK_RAW && !kern && 314 !ns_capable(net->user_ns, CAP_NET_RAW)) 315 goto out_rcu_unlock; 316 317 sock->ops = answer->ops; 318 answer_prot = answer->prot; 319 answer_flags = answer->flags; 320 rcu_read_unlock(); 321 322 WARN_ON(!answer_prot->slab); 323 324 err = -ENOMEM; 325 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern); 326 if (!sk) 327 goto out; 328 329 err = 0; 330 if (INET_PROTOSW_REUSE & answer_flags) 331 sk->sk_reuse = SK_CAN_REUSE; 332 333 if (INET_PROTOSW_ICSK & answer_flags) 334 inet_init_csk_locks(sk); 335 336 inet = inet_sk(sk); 337 inet_assign_bit(IS_ICSK, sk, INET_PROTOSW_ICSK & answer_flags); 338 339 inet_clear_bit(NODEFRAG, sk); 340 341 if (SOCK_RAW == sock->type) { 342 inet->inet_num = protocol; 343 if (IPPROTO_RAW == protocol) 344 inet_set_bit(HDRINCL, sk); 345 } 346 347 if (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc)) 348 inet->pmtudisc = IP_PMTUDISC_DONT; 349 else 350 inet->pmtudisc = IP_PMTUDISC_WANT; 351 352 atomic_set(&inet->inet_id, 0); 353 354 sock_init_data(sock, sk); 355 356 sk->sk_destruct = inet_sock_destruct; 357 sk->sk_protocol = protocol; 358 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv; 359 sk->sk_txrehash = READ_ONCE(net->core.sysctl_txrehash); 360 361 inet->uc_ttl = -1; 362 inet_set_bit(MC_LOOP, sk); 363 inet->mc_ttl = 1; 364 inet_set_bit(MC_ALL, sk); 365 inet->mc_index = 0; 366 inet->mc_list = NULL; 367 inet->rcv_tos = 0; 368 369 if (inet->inet_num) { 370 /* It assumes that any protocol which allows 371 * the user to assign a number at socket 372 * creation time automatically 373 * shares. 374 */ 375 inet->inet_sport = htons(inet->inet_num); 376 /* Add to protocol hash chains. */ 377 err = sk->sk_prot->hash(sk); 378 if (err) { 379 sk_common_release(sk); 380 goto out; 381 } 382 } 383 384 if (sk->sk_prot->init) { 385 err = sk->sk_prot->init(sk); 386 if (err) { 387 sk_common_release(sk); 388 goto out; 389 } 390 } 391 392 if (!kern) { 393 err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk); 394 if (err) { 395 sk_common_release(sk); 396 goto out; 397 } 398 } 399 out: 400 return err; 401 out_rcu_unlock: 402 rcu_read_unlock(); 403 goto out; 404 } 405 406 407 /* 408 * The peer socket should always be NULL (or else). When we call this 409 * function we are destroying the object and from then on nobody 410 * should refer to it. 411 */ 412 int inet_release(struct socket *sock) 413 { 414 struct sock *sk = sock->sk; 415 416 if (sk) { 417 long timeout; 418 419 if (!sk->sk_kern_sock) 420 BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk); 421 422 /* Applications forget to leave groups before exiting */ 423 ip_mc_drop_socket(sk); 424 425 /* If linger is set, we don't return until the close 426 * is complete. Otherwise we return immediately. The 427 * actually closing is done the same either way. 428 * 429 * If the close is due to the process exiting, we never 430 * linger.. 431 */ 432 timeout = 0; 433 if (sock_flag(sk, SOCK_LINGER) && 434 !(current->flags & PF_EXITING)) 435 timeout = sk->sk_lingertime; 436 sk->sk_prot->close(sk, timeout); 437 sock->sk = NULL; 438 } 439 return 0; 440 } 441 EXPORT_SYMBOL(inet_release); 442 443 int inet_bind_sk(struct sock *sk, struct sockaddr *uaddr, int addr_len) 444 { 445 u32 flags = BIND_WITH_LOCK; 446 int err; 447 448 /* If the socket has its own bind function then use it. (RAW) */ 449 if (sk->sk_prot->bind) { 450 return sk->sk_prot->bind(sk, uaddr, addr_len); 451 } 452 if (addr_len < sizeof(struct sockaddr_in)) 453 return -EINVAL; 454 455 /* BPF prog is run before any checks are done so that if the prog 456 * changes context in a wrong way it will be caught. 457 */ 458 err = BPF_CGROUP_RUN_PROG_INET_BIND_LOCK(sk, uaddr, 459 CGROUP_INET4_BIND, &flags); 460 if (err) 461 return err; 462 463 return __inet_bind(sk, uaddr, addr_len, flags); 464 } 465 466 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) 467 { 468 return inet_bind_sk(sock->sk, uaddr, addr_len); 469 } 470 EXPORT_SYMBOL(inet_bind); 471 472 int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len, 473 u32 flags) 474 { 475 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr; 476 struct inet_sock *inet = inet_sk(sk); 477 struct net *net = sock_net(sk); 478 unsigned short snum; 479 int chk_addr_ret; 480 u32 tb_id = RT_TABLE_LOCAL; 481 int err; 482 483 if (addr->sin_family != AF_INET) { 484 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET) 485 * only if s_addr is INADDR_ANY. 486 */ 487 err = -EAFNOSUPPORT; 488 if (addr->sin_family != AF_UNSPEC || 489 addr->sin_addr.s_addr != htonl(INADDR_ANY)) 490 goto out; 491 } 492 493 tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id; 494 chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id); 495 496 /* Not specified by any standard per-se, however it breaks too 497 * many applications when removed. It is unfortunate since 498 * allowing applications to make a non-local bind solves 499 * several problems with systems using dynamic addressing. 500 * (ie. your servers still start up even if your ISDN link 501 * is temporarily down) 502 */ 503 err = -EADDRNOTAVAIL; 504 if (!inet_addr_valid_or_nonlocal(net, inet, addr->sin_addr.s_addr, 505 chk_addr_ret)) 506 goto out; 507 508 snum = ntohs(addr->sin_port); 509 err = -EACCES; 510 if (!(flags & BIND_NO_CAP_NET_BIND_SERVICE) && 511 snum && inet_port_requires_bind_service(net, snum) && 512 !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE)) 513 goto out; 514 515 /* We keep a pair of addresses. rcv_saddr is the one 516 * used by hash lookups, and saddr is used for transmit. 517 * 518 * In the BSD API these are the same except where it 519 * would be illegal to use them (multicast/broadcast) in 520 * which case the sending device address is used. 521 */ 522 if (flags & BIND_WITH_LOCK) 523 lock_sock(sk); 524 525 /* Check these errors (active socket, double bind). */ 526 err = -EINVAL; 527 if (sk->sk_state != TCP_CLOSE || inet->inet_num) 528 goto out_release_sock; 529 530 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr; 531 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST) 532 inet->inet_saddr = 0; /* Use device */ 533 534 /* Make sure we are allowed to bind here. */ 535 if (snum || !(inet_test_bit(BIND_ADDRESS_NO_PORT, sk) || 536 (flags & BIND_FORCE_ADDRESS_NO_PORT))) { 537 err = sk->sk_prot->get_port(sk, snum); 538 if (err) { 539 inet->inet_saddr = inet->inet_rcv_saddr = 0; 540 goto out_release_sock; 541 } 542 if (!(flags & BIND_FROM_BPF)) { 543 err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk); 544 if (err) { 545 inet->inet_saddr = inet->inet_rcv_saddr = 0; 546 if (sk->sk_prot->put_port) 547 sk->sk_prot->put_port(sk); 548 goto out_release_sock; 549 } 550 } 551 } 552 553 if (inet->inet_rcv_saddr) 554 sk->sk_userlocks |= SOCK_BINDADDR_LOCK; 555 if (snum) 556 sk->sk_userlocks |= SOCK_BINDPORT_LOCK; 557 inet->inet_sport = htons(inet->inet_num); 558 inet->inet_daddr = 0; 559 inet->inet_dport = 0; 560 sk_dst_reset(sk); 561 err = 0; 562 out_release_sock: 563 if (flags & BIND_WITH_LOCK) 564 release_sock(sk); 565 out: 566 return err; 567 } 568 569 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr, 570 int addr_len, int flags) 571 { 572 struct sock *sk = sock->sk; 573 const struct proto *prot; 574 int err; 575 576 if (addr_len < sizeof(uaddr->sa_family)) 577 return -EINVAL; 578 579 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 580 prot = READ_ONCE(sk->sk_prot); 581 582 if (uaddr->sa_family == AF_UNSPEC) 583 return prot->disconnect(sk, flags); 584 585 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) { 586 err = prot->pre_connect(sk, uaddr, addr_len); 587 if (err) 588 return err; 589 } 590 591 if (data_race(!inet_sk(sk)->inet_num) && inet_autobind(sk)) 592 return -EAGAIN; 593 return prot->connect(sk, uaddr, addr_len); 594 } 595 EXPORT_SYMBOL(inet_dgram_connect); 596 597 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias) 598 { 599 DEFINE_WAIT_FUNC(wait, woken_wake_function); 600 601 add_wait_queue(sk_sleep(sk), &wait); 602 sk->sk_write_pending += writebias; 603 604 /* Basic assumption: if someone sets sk->sk_err, he _must_ 605 * change state of the socket from TCP_SYN_*. 606 * Connect() does not allow to get error notifications 607 * without closing the socket. 608 */ 609 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 610 release_sock(sk); 611 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo); 612 lock_sock(sk); 613 if (signal_pending(current) || !timeo) 614 break; 615 } 616 remove_wait_queue(sk_sleep(sk), &wait); 617 sk->sk_write_pending -= writebias; 618 return timeo; 619 } 620 621 /* 622 * Connect to a remote host. There is regrettably still a little 623 * TCP 'magic' in here. 624 */ 625 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr, 626 int addr_len, int flags, int is_sendmsg) 627 { 628 struct sock *sk = sock->sk; 629 int err; 630 long timeo; 631 632 /* 633 * uaddr can be NULL and addr_len can be 0 if: 634 * sk is a TCP fastopen active socket and 635 * TCP_FASTOPEN_CONNECT sockopt is set and 636 * we already have a valid cookie for this socket. 637 * In this case, user can call write() after connect(). 638 * write() will invoke tcp_sendmsg_fastopen() which calls 639 * __inet_stream_connect(). 640 */ 641 if (uaddr) { 642 if (addr_len < sizeof(uaddr->sa_family)) 643 return -EINVAL; 644 645 if (uaddr->sa_family == AF_UNSPEC) { 646 sk->sk_disconnects++; 647 err = sk->sk_prot->disconnect(sk, flags); 648 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED; 649 goto out; 650 } 651 } 652 653 switch (sock->state) { 654 default: 655 err = -EINVAL; 656 goto out; 657 case SS_CONNECTED: 658 err = -EISCONN; 659 goto out; 660 case SS_CONNECTING: 661 if (inet_test_bit(DEFER_CONNECT, sk)) 662 err = is_sendmsg ? -EINPROGRESS : -EISCONN; 663 else 664 err = -EALREADY; 665 /* Fall out of switch with err, set for this state */ 666 break; 667 case SS_UNCONNECTED: 668 err = -EISCONN; 669 if (sk->sk_state != TCP_CLOSE) 670 goto out; 671 672 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) { 673 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len); 674 if (err) 675 goto out; 676 } 677 678 err = sk->sk_prot->connect(sk, uaddr, addr_len); 679 if (err < 0) 680 goto out; 681 682 sock->state = SS_CONNECTING; 683 684 if (!err && inet_test_bit(DEFER_CONNECT, sk)) 685 goto out; 686 687 /* Just entered SS_CONNECTING state; the only 688 * difference is that return value in non-blocking 689 * case is EINPROGRESS, rather than EALREADY. 690 */ 691 err = -EINPROGRESS; 692 break; 693 } 694 695 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK); 696 697 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 698 int writebias = (sk->sk_protocol == IPPROTO_TCP) && 699 tcp_sk(sk)->fastopen_req && 700 tcp_sk(sk)->fastopen_req->data ? 1 : 0; 701 int dis = sk->sk_disconnects; 702 703 /* Error code is set above */ 704 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias)) 705 goto out; 706 707 err = sock_intr_errno(timeo); 708 if (signal_pending(current)) 709 goto out; 710 711 if (dis != sk->sk_disconnects) { 712 err = -EPIPE; 713 goto out; 714 } 715 } 716 717 /* Connection was closed by RST, timeout, ICMP error 718 * or another process disconnected us. 719 */ 720 if (sk->sk_state == TCP_CLOSE) 721 goto sock_error; 722 723 /* sk->sk_err may be not zero now, if RECVERR was ordered by user 724 * and error was received after socket entered established state. 725 * Hence, it is handled normally after connect() return successfully. 726 */ 727 728 sock->state = SS_CONNECTED; 729 err = 0; 730 out: 731 return err; 732 733 sock_error: 734 err = sock_error(sk) ? : -ECONNABORTED; 735 sock->state = SS_UNCONNECTED; 736 sk->sk_disconnects++; 737 if (sk->sk_prot->disconnect(sk, flags)) 738 sock->state = SS_DISCONNECTING; 739 goto out; 740 } 741 EXPORT_SYMBOL(__inet_stream_connect); 742 743 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr, 744 int addr_len, int flags) 745 { 746 int err; 747 748 lock_sock(sock->sk); 749 err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0); 750 release_sock(sock->sk); 751 return err; 752 } 753 EXPORT_SYMBOL(inet_stream_connect); 754 755 void __inet_accept(struct socket *sock, struct socket *newsock, struct sock *newsk) 756 { 757 sock_rps_record_flow(newsk); 758 WARN_ON(!((1 << newsk->sk_state) & 759 (TCPF_ESTABLISHED | TCPF_SYN_RECV | 760 TCPF_CLOSE_WAIT | TCPF_CLOSE))); 761 762 if (test_bit(SOCK_SUPPORT_ZC, &sock->flags)) 763 set_bit(SOCK_SUPPORT_ZC, &newsock->flags); 764 sock_graft(newsk, newsock); 765 766 newsock->state = SS_CONNECTED; 767 } 768 769 /* 770 * Accept a pending connection. The TCP layer now gives BSD semantics. 771 */ 772 773 int inet_accept(struct socket *sock, struct socket *newsock, int flags, 774 bool kern) 775 { 776 struct sock *sk1 = sock->sk, *sk2; 777 int err = -EINVAL; 778 779 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 780 sk2 = READ_ONCE(sk1->sk_prot)->accept(sk1, flags, &err, kern); 781 if (!sk2) 782 return err; 783 784 lock_sock(sk2); 785 __inet_accept(sock, newsock, sk2); 786 release_sock(sk2); 787 return 0; 788 } 789 EXPORT_SYMBOL(inet_accept); 790 791 /* 792 * This does both peername and sockname. 793 */ 794 int inet_getname(struct socket *sock, struct sockaddr *uaddr, 795 int peer) 796 { 797 struct sock *sk = sock->sk; 798 struct inet_sock *inet = inet_sk(sk); 799 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr); 800 801 sin->sin_family = AF_INET; 802 lock_sock(sk); 803 if (peer) { 804 if (!inet->inet_dport || 805 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) && 806 peer == 1)) { 807 release_sock(sk); 808 return -ENOTCONN; 809 } 810 sin->sin_port = inet->inet_dport; 811 sin->sin_addr.s_addr = inet->inet_daddr; 812 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin, 813 CGROUP_INET4_GETPEERNAME); 814 } else { 815 __be32 addr = inet->inet_rcv_saddr; 816 if (!addr) 817 addr = inet->inet_saddr; 818 sin->sin_port = inet->inet_sport; 819 sin->sin_addr.s_addr = addr; 820 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin, 821 CGROUP_INET4_GETSOCKNAME); 822 } 823 release_sock(sk); 824 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 825 return sizeof(*sin); 826 } 827 EXPORT_SYMBOL(inet_getname); 828 829 int inet_send_prepare(struct sock *sk) 830 { 831 sock_rps_record_flow(sk); 832 833 /* We may need to bind the socket. */ 834 if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind && 835 inet_autobind(sk)) 836 return -EAGAIN; 837 838 return 0; 839 } 840 EXPORT_SYMBOL_GPL(inet_send_prepare); 841 842 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size) 843 { 844 struct sock *sk = sock->sk; 845 846 if (unlikely(inet_send_prepare(sk))) 847 return -EAGAIN; 848 849 return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg, 850 sk, msg, size); 851 } 852 EXPORT_SYMBOL(inet_sendmsg); 853 854 void inet_splice_eof(struct socket *sock) 855 { 856 const struct proto *prot; 857 struct sock *sk = sock->sk; 858 859 if (unlikely(inet_send_prepare(sk))) 860 return; 861 862 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 863 prot = READ_ONCE(sk->sk_prot); 864 if (prot->splice_eof) 865 prot->splice_eof(sock); 866 } 867 EXPORT_SYMBOL_GPL(inet_splice_eof); 868 869 INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *, 870 size_t, int, int *)); 871 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 872 int flags) 873 { 874 struct sock *sk = sock->sk; 875 int addr_len = 0; 876 int err; 877 878 if (likely(!(flags & MSG_ERRQUEUE))) 879 sock_rps_record_flow(sk); 880 881 err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg, 882 sk, msg, size, flags, &addr_len); 883 if (err >= 0) 884 msg->msg_namelen = addr_len; 885 return err; 886 } 887 EXPORT_SYMBOL(inet_recvmsg); 888 889 int inet_shutdown(struct socket *sock, int how) 890 { 891 struct sock *sk = sock->sk; 892 int err = 0; 893 894 /* This should really check to make sure 895 * the socket is a TCP socket. (WHY AC...) 896 */ 897 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and 898 1->2 bit 2 snds. 899 2->3 */ 900 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */ 901 return -EINVAL; 902 903 lock_sock(sk); 904 if (sock->state == SS_CONNECTING) { 905 if ((1 << sk->sk_state) & 906 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE)) 907 sock->state = SS_DISCONNECTING; 908 else 909 sock->state = SS_CONNECTED; 910 } 911 912 switch (sk->sk_state) { 913 case TCP_CLOSE: 914 err = -ENOTCONN; 915 /* Hack to wake up other listeners, who can poll for 916 EPOLLHUP, even on eg. unconnected UDP sockets -- RR */ 917 fallthrough; 918 default: 919 WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | how); 920 if (sk->sk_prot->shutdown) 921 sk->sk_prot->shutdown(sk, how); 922 break; 923 924 /* Remaining two branches are temporary solution for missing 925 * close() in multithreaded environment. It is _not_ a good idea, 926 * but we have no choice until close() is repaired at VFS level. 927 */ 928 case TCP_LISTEN: 929 if (!(how & RCV_SHUTDOWN)) 930 break; 931 fallthrough; 932 case TCP_SYN_SENT: 933 err = sk->sk_prot->disconnect(sk, O_NONBLOCK); 934 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED; 935 break; 936 } 937 938 /* Wake up anyone sleeping in poll. */ 939 sk->sk_state_change(sk); 940 release_sock(sk); 941 return err; 942 } 943 EXPORT_SYMBOL(inet_shutdown); 944 945 /* 946 * ioctl() calls you can issue on an INET socket. Most of these are 947 * device configuration and stuff and very rarely used. Some ioctls 948 * pass on to the socket itself. 949 * 950 * NOTE: I like the idea of a module for the config stuff. ie ifconfig 951 * loads the devconfigure module does its configuring and unloads it. 952 * There's a good 20K of config code hanging around the kernel. 953 */ 954 955 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 956 { 957 struct sock *sk = sock->sk; 958 int err = 0; 959 struct net *net = sock_net(sk); 960 void __user *p = (void __user *)arg; 961 struct ifreq ifr; 962 struct rtentry rt; 963 964 switch (cmd) { 965 case SIOCADDRT: 966 case SIOCDELRT: 967 if (copy_from_user(&rt, p, sizeof(struct rtentry))) 968 return -EFAULT; 969 err = ip_rt_ioctl(net, cmd, &rt); 970 break; 971 case SIOCRTMSG: 972 err = -EINVAL; 973 break; 974 case SIOCDARP: 975 case SIOCGARP: 976 case SIOCSARP: 977 err = arp_ioctl(net, cmd, (void __user *)arg); 978 break; 979 case SIOCGIFADDR: 980 case SIOCGIFBRDADDR: 981 case SIOCGIFNETMASK: 982 case SIOCGIFDSTADDR: 983 case SIOCGIFPFLAGS: 984 if (get_user_ifreq(&ifr, NULL, p)) 985 return -EFAULT; 986 err = devinet_ioctl(net, cmd, &ifr); 987 if (!err && put_user_ifreq(&ifr, p)) 988 err = -EFAULT; 989 break; 990 991 case SIOCSIFADDR: 992 case SIOCSIFBRDADDR: 993 case SIOCSIFNETMASK: 994 case SIOCSIFDSTADDR: 995 case SIOCSIFPFLAGS: 996 case SIOCSIFFLAGS: 997 if (get_user_ifreq(&ifr, NULL, p)) 998 return -EFAULT; 999 err = devinet_ioctl(net, cmd, &ifr); 1000 break; 1001 default: 1002 if (sk->sk_prot->ioctl) 1003 err = sk_ioctl(sk, cmd, (void __user *)arg); 1004 else 1005 err = -ENOIOCTLCMD; 1006 break; 1007 } 1008 return err; 1009 } 1010 EXPORT_SYMBOL(inet_ioctl); 1011 1012 #ifdef CONFIG_COMPAT 1013 static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd, 1014 struct compat_rtentry __user *ur) 1015 { 1016 compat_uptr_t rtdev; 1017 struct rtentry rt; 1018 1019 if (copy_from_user(&rt.rt_dst, &ur->rt_dst, 1020 3 * sizeof(struct sockaddr)) || 1021 get_user(rt.rt_flags, &ur->rt_flags) || 1022 get_user(rt.rt_metric, &ur->rt_metric) || 1023 get_user(rt.rt_mtu, &ur->rt_mtu) || 1024 get_user(rt.rt_window, &ur->rt_window) || 1025 get_user(rt.rt_irtt, &ur->rt_irtt) || 1026 get_user(rtdev, &ur->rt_dev)) 1027 return -EFAULT; 1028 1029 rt.rt_dev = compat_ptr(rtdev); 1030 return ip_rt_ioctl(sock_net(sk), cmd, &rt); 1031 } 1032 1033 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 1034 { 1035 void __user *argp = compat_ptr(arg); 1036 struct sock *sk = sock->sk; 1037 1038 switch (cmd) { 1039 case SIOCADDRT: 1040 case SIOCDELRT: 1041 return inet_compat_routing_ioctl(sk, cmd, argp); 1042 default: 1043 if (!sk->sk_prot->compat_ioctl) 1044 return -ENOIOCTLCMD; 1045 return sk->sk_prot->compat_ioctl(sk, cmd, arg); 1046 } 1047 } 1048 #endif /* CONFIG_COMPAT */ 1049 1050 const struct proto_ops inet_stream_ops = { 1051 .family = PF_INET, 1052 .owner = THIS_MODULE, 1053 .release = inet_release, 1054 .bind = inet_bind, 1055 .connect = inet_stream_connect, 1056 .socketpair = sock_no_socketpair, 1057 .accept = inet_accept, 1058 .getname = inet_getname, 1059 .poll = tcp_poll, 1060 .ioctl = inet_ioctl, 1061 .gettstamp = sock_gettstamp, 1062 .listen = inet_listen, 1063 .shutdown = inet_shutdown, 1064 .setsockopt = sock_common_setsockopt, 1065 .getsockopt = sock_common_getsockopt, 1066 .sendmsg = inet_sendmsg, 1067 .recvmsg = inet_recvmsg, 1068 #ifdef CONFIG_MMU 1069 .mmap = tcp_mmap, 1070 #endif 1071 .splice_eof = inet_splice_eof, 1072 .splice_read = tcp_splice_read, 1073 .read_sock = tcp_read_sock, 1074 .read_skb = tcp_read_skb, 1075 .sendmsg_locked = tcp_sendmsg_locked, 1076 .peek_len = tcp_peek_len, 1077 #ifdef CONFIG_COMPAT 1078 .compat_ioctl = inet_compat_ioctl, 1079 #endif 1080 .set_rcvlowat = tcp_set_rcvlowat, 1081 }; 1082 EXPORT_SYMBOL(inet_stream_ops); 1083 1084 const struct proto_ops inet_dgram_ops = { 1085 .family = PF_INET, 1086 .owner = THIS_MODULE, 1087 .release = inet_release, 1088 .bind = inet_bind, 1089 .connect = inet_dgram_connect, 1090 .socketpair = sock_no_socketpair, 1091 .accept = sock_no_accept, 1092 .getname = inet_getname, 1093 .poll = udp_poll, 1094 .ioctl = inet_ioctl, 1095 .gettstamp = sock_gettstamp, 1096 .listen = sock_no_listen, 1097 .shutdown = inet_shutdown, 1098 .setsockopt = sock_common_setsockopt, 1099 .getsockopt = sock_common_getsockopt, 1100 .sendmsg = inet_sendmsg, 1101 .read_skb = udp_read_skb, 1102 .recvmsg = inet_recvmsg, 1103 .mmap = sock_no_mmap, 1104 .splice_eof = inet_splice_eof, 1105 .set_peek_off = sk_set_peek_off, 1106 #ifdef CONFIG_COMPAT 1107 .compat_ioctl = inet_compat_ioctl, 1108 #endif 1109 }; 1110 EXPORT_SYMBOL(inet_dgram_ops); 1111 1112 /* 1113 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without 1114 * udp_poll 1115 */ 1116 static const struct proto_ops inet_sockraw_ops = { 1117 .family = PF_INET, 1118 .owner = THIS_MODULE, 1119 .release = inet_release, 1120 .bind = inet_bind, 1121 .connect = inet_dgram_connect, 1122 .socketpair = sock_no_socketpair, 1123 .accept = sock_no_accept, 1124 .getname = inet_getname, 1125 .poll = datagram_poll, 1126 .ioctl = inet_ioctl, 1127 .gettstamp = sock_gettstamp, 1128 .listen = sock_no_listen, 1129 .shutdown = inet_shutdown, 1130 .setsockopt = sock_common_setsockopt, 1131 .getsockopt = sock_common_getsockopt, 1132 .sendmsg = inet_sendmsg, 1133 .recvmsg = inet_recvmsg, 1134 .mmap = sock_no_mmap, 1135 .splice_eof = inet_splice_eof, 1136 #ifdef CONFIG_COMPAT 1137 .compat_ioctl = inet_compat_ioctl, 1138 #endif 1139 }; 1140 1141 static const struct net_proto_family inet_family_ops = { 1142 .family = PF_INET, 1143 .create = inet_create, 1144 .owner = THIS_MODULE, 1145 }; 1146 1147 /* Upon startup we insert all the elements in inetsw_array[] into 1148 * the linked list inetsw. 1149 */ 1150 static struct inet_protosw inetsw_array[] = 1151 { 1152 { 1153 .type = SOCK_STREAM, 1154 .protocol = IPPROTO_TCP, 1155 .prot = &tcp_prot, 1156 .ops = &inet_stream_ops, 1157 .flags = INET_PROTOSW_PERMANENT | 1158 INET_PROTOSW_ICSK, 1159 }, 1160 1161 { 1162 .type = SOCK_DGRAM, 1163 .protocol = IPPROTO_UDP, 1164 .prot = &udp_prot, 1165 .ops = &inet_dgram_ops, 1166 .flags = INET_PROTOSW_PERMANENT, 1167 }, 1168 1169 { 1170 .type = SOCK_DGRAM, 1171 .protocol = IPPROTO_ICMP, 1172 .prot = &ping_prot, 1173 .ops = &inet_sockraw_ops, 1174 .flags = INET_PROTOSW_REUSE, 1175 }, 1176 1177 { 1178 .type = SOCK_RAW, 1179 .protocol = IPPROTO_IP, /* wild card */ 1180 .prot = &raw_prot, 1181 .ops = &inet_sockraw_ops, 1182 .flags = INET_PROTOSW_REUSE, 1183 } 1184 }; 1185 1186 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array) 1187 1188 void inet_register_protosw(struct inet_protosw *p) 1189 { 1190 struct list_head *lh; 1191 struct inet_protosw *answer; 1192 int protocol = p->protocol; 1193 struct list_head *last_perm; 1194 1195 spin_lock_bh(&inetsw_lock); 1196 1197 if (p->type >= SOCK_MAX) 1198 goto out_illegal; 1199 1200 /* If we are trying to override a permanent protocol, bail. */ 1201 last_perm = &inetsw[p->type]; 1202 list_for_each(lh, &inetsw[p->type]) { 1203 answer = list_entry(lh, struct inet_protosw, list); 1204 /* Check only the non-wild match. */ 1205 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0) 1206 break; 1207 if (protocol == answer->protocol) 1208 goto out_permanent; 1209 last_perm = lh; 1210 } 1211 1212 /* Add the new entry after the last permanent entry if any, so that 1213 * the new entry does not override a permanent entry when matched with 1214 * a wild-card protocol. But it is allowed to override any existing 1215 * non-permanent entry. This means that when we remove this entry, the 1216 * system automatically returns to the old behavior. 1217 */ 1218 list_add_rcu(&p->list, last_perm); 1219 out: 1220 spin_unlock_bh(&inetsw_lock); 1221 1222 return; 1223 1224 out_permanent: 1225 pr_err("Attempt to override permanent protocol %d\n", protocol); 1226 goto out; 1227 1228 out_illegal: 1229 pr_err("Ignoring attempt to register invalid socket type %d\n", 1230 p->type); 1231 goto out; 1232 } 1233 EXPORT_SYMBOL(inet_register_protosw); 1234 1235 void inet_unregister_protosw(struct inet_protosw *p) 1236 { 1237 if (INET_PROTOSW_PERMANENT & p->flags) { 1238 pr_err("Attempt to unregister permanent protocol %d\n", 1239 p->protocol); 1240 } else { 1241 spin_lock_bh(&inetsw_lock); 1242 list_del_rcu(&p->list); 1243 spin_unlock_bh(&inetsw_lock); 1244 1245 synchronize_net(); 1246 } 1247 } 1248 EXPORT_SYMBOL(inet_unregister_protosw); 1249 1250 static int inet_sk_reselect_saddr(struct sock *sk) 1251 { 1252 struct inet_sock *inet = inet_sk(sk); 1253 __be32 old_saddr = inet->inet_saddr; 1254 __be32 daddr = inet->inet_daddr; 1255 struct flowi4 *fl4; 1256 struct rtable *rt; 1257 __be32 new_saddr; 1258 struct ip_options_rcu *inet_opt; 1259 int err; 1260 1261 inet_opt = rcu_dereference_protected(inet->inet_opt, 1262 lockdep_sock_is_held(sk)); 1263 if (inet_opt && inet_opt->opt.srr) 1264 daddr = inet_opt->opt.faddr; 1265 1266 /* Query new route. */ 1267 fl4 = &inet->cork.fl.u.ip4; 1268 rt = ip_route_connect(fl4, daddr, 0, sk->sk_bound_dev_if, 1269 sk->sk_protocol, inet->inet_sport, 1270 inet->inet_dport, sk); 1271 if (IS_ERR(rt)) 1272 return PTR_ERR(rt); 1273 1274 new_saddr = fl4->saddr; 1275 1276 if (new_saddr == old_saddr) { 1277 sk_setup_caps(sk, &rt->dst); 1278 return 0; 1279 } 1280 1281 err = inet_bhash2_update_saddr(sk, &new_saddr, AF_INET); 1282 if (err) { 1283 ip_rt_put(rt); 1284 return err; 1285 } 1286 1287 sk_setup_caps(sk, &rt->dst); 1288 1289 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) > 1) { 1290 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n", 1291 __func__, &old_saddr, &new_saddr); 1292 } 1293 1294 /* 1295 * XXX The only one ugly spot where we need to 1296 * XXX really change the sockets identity after 1297 * XXX it has entered the hashes. -DaveM 1298 * 1299 * Besides that, it does not check for connection 1300 * uniqueness. Wait for troubles. 1301 */ 1302 return __sk_prot_rehash(sk); 1303 } 1304 1305 int inet_sk_rebuild_header(struct sock *sk) 1306 { 1307 struct inet_sock *inet = inet_sk(sk); 1308 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0); 1309 __be32 daddr; 1310 struct ip_options_rcu *inet_opt; 1311 struct flowi4 *fl4; 1312 int err; 1313 1314 /* Route is OK, nothing to do. */ 1315 if (rt) 1316 return 0; 1317 1318 /* Reroute. */ 1319 rcu_read_lock(); 1320 inet_opt = rcu_dereference(inet->inet_opt); 1321 daddr = inet->inet_daddr; 1322 if (inet_opt && inet_opt->opt.srr) 1323 daddr = inet_opt->opt.faddr; 1324 rcu_read_unlock(); 1325 fl4 = &inet->cork.fl.u.ip4; 1326 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr, 1327 inet->inet_dport, inet->inet_sport, 1328 sk->sk_protocol, RT_CONN_FLAGS(sk), 1329 sk->sk_bound_dev_if); 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 * Other protocols have to map its equivalent state to TCP_SYN_SENT. 1340 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme 1341 */ 1342 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) || 1343 sk->sk_state != TCP_SYN_SENT || 1344 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) || 1345 (err = inet_sk_reselect_saddr(sk)) != 0) 1346 WRITE_ONCE(sk->sk_err_soft, -err); 1347 } 1348 1349 return err; 1350 } 1351 EXPORT_SYMBOL(inet_sk_rebuild_header); 1352 1353 void inet_sk_set_state(struct sock *sk, int state) 1354 { 1355 trace_inet_sock_set_state(sk, sk->sk_state, state); 1356 sk->sk_state = state; 1357 } 1358 EXPORT_SYMBOL(inet_sk_set_state); 1359 1360 void inet_sk_state_store(struct sock *sk, int newstate) 1361 { 1362 trace_inet_sock_set_state(sk, sk->sk_state, newstate); 1363 smp_store_release(&sk->sk_state, newstate); 1364 } 1365 1366 struct sk_buff *inet_gso_segment(struct sk_buff *skb, 1367 netdev_features_t features) 1368 { 1369 bool udpfrag = false, fixedid = false, gso_partial, encap; 1370 struct sk_buff *segs = ERR_PTR(-EINVAL); 1371 const struct net_offload *ops; 1372 unsigned int offset = 0; 1373 struct iphdr *iph; 1374 int proto, tot_len; 1375 int nhoff; 1376 int ihl; 1377 int id; 1378 1379 skb_reset_network_header(skb); 1380 nhoff = skb_network_header(skb) - skb_mac_header(skb); 1381 if (unlikely(!pskb_may_pull(skb, sizeof(*iph)))) 1382 goto out; 1383 1384 iph = ip_hdr(skb); 1385 ihl = iph->ihl * 4; 1386 if (ihl < sizeof(*iph)) 1387 goto out; 1388 1389 id = ntohs(iph->id); 1390 proto = iph->protocol; 1391 1392 /* Warning: after this point, iph might be no longer valid */ 1393 if (unlikely(!pskb_may_pull(skb, ihl))) 1394 goto out; 1395 __skb_pull(skb, ihl); 1396 1397 encap = SKB_GSO_CB(skb)->encap_level > 0; 1398 if (encap) 1399 features &= skb->dev->hw_enc_features; 1400 SKB_GSO_CB(skb)->encap_level += ihl; 1401 1402 skb_reset_transport_header(skb); 1403 1404 segs = ERR_PTR(-EPROTONOSUPPORT); 1405 1406 if (!skb->encapsulation || encap) { 1407 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP); 1408 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID); 1409 1410 /* fixed ID is invalid if DF bit is not set */ 1411 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF))) 1412 goto out; 1413 } 1414 1415 ops = rcu_dereference(inet_offloads[proto]); 1416 if (likely(ops && ops->callbacks.gso_segment)) { 1417 segs = ops->callbacks.gso_segment(skb, features); 1418 if (!segs) 1419 skb->network_header = skb_mac_header(skb) + nhoff - skb->head; 1420 } 1421 1422 if (IS_ERR_OR_NULL(segs)) 1423 goto out; 1424 1425 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL); 1426 1427 skb = segs; 1428 do { 1429 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff); 1430 if (udpfrag) { 1431 iph->frag_off = htons(offset >> 3); 1432 if (skb->next) 1433 iph->frag_off |= htons(IP_MF); 1434 offset += skb->len - nhoff - ihl; 1435 tot_len = skb->len - nhoff; 1436 } else if (skb_is_gso(skb)) { 1437 if (!fixedid) { 1438 iph->id = htons(id); 1439 id += skb_shinfo(skb)->gso_segs; 1440 } 1441 1442 if (gso_partial) 1443 tot_len = skb_shinfo(skb)->gso_size + 1444 SKB_GSO_CB(skb)->data_offset + 1445 skb->head - (unsigned char *)iph; 1446 else 1447 tot_len = skb->len - nhoff; 1448 } else { 1449 if (!fixedid) 1450 iph->id = htons(id++); 1451 tot_len = skb->len - nhoff; 1452 } 1453 iph->tot_len = htons(tot_len); 1454 ip_send_check(iph); 1455 if (encap) 1456 skb_reset_inner_headers(skb); 1457 skb->network_header = (u8 *)iph - skb->head; 1458 skb_reset_mac_len(skb); 1459 } while ((skb = skb->next)); 1460 1461 out: 1462 return segs; 1463 } 1464 1465 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb, 1466 netdev_features_t features) 1467 { 1468 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4)) 1469 return ERR_PTR(-EINVAL); 1470 1471 return inet_gso_segment(skb, features); 1472 } 1473 1474 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb) 1475 { 1476 const struct net_offload *ops; 1477 struct sk_buff *pp = NULL; 1478 const struct iphdr *iph; 1479 struct sk_buff *p; 1480 unsigned int hlen; 1481 unsigned int off; 1482 unsigned int id; 1483 int flush = 1; 1484 int proto; 1485 1486 off = skb_gro_offset(skb); 1487 hlen = off + sizeof(*iph); 1488 iph = skb_gro_header(skb, hlen, off); 1489 if (unlikely(!iph)) 1490 goto out; 1491 1492 proto = iph->protocol; 1493 1494 ops = rcu_dereference(inet_offloads[proto]); 1495 if (!ops || !ops->callbacks.gro_receive) 1496 goto out; 1497 1498 if (*(u8 *)iph != 0x45) 1499 goto out; 1500 1501 if (ip_is_fragment(iph)) 1502 goto out; 1503 1504 if (unlikely(ip_fast_csum((u8 *)iph, 5))) 1505 goto out; 1506 1507 NAPI_GRO_CB(skb)->proto = proto; 1508 id = ntohl(*(__be32 *)&iph->id); 1509 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF)); 1510 id >>= 16; 1511 1512 list_for_each_entry(p, head, list) { 1513 struct iphdr *iph2; 1514 u16 flush_id; 1515 1516 if (!NAPI_GRO_CB(p)->same_flow) 1517 continue; 1518 1519 iph2 = (struct iphdr *)(p->data + off); 1520 /* The above works because, with the exception of the top 1521 * (inner most) layer, we only aggregate pkts with the same 1522 * hdr length so all the hdrs we'll need to verify will start 1523 * at the same offset. 1524 */ 1525 if ((iph->protocol ^ iph2->protocol) | 1526 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) | 1527 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) { 1528 NAPI_GRO_CB(p)->same_flow = 0; 1529 continue; 1530 } 1531 1532 /* All fields must match except length and checksum. */ 1533 NAPI_GRO_CB(p)->flush |= 1534 (iph->ttl ^ iph2->ttl) | 1535 (iph->tos ^ iph2->tos) | 1536 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF)); 1537 1538 NAPI_GRO_CB(p)->flush |= flush; 1539 1540 /* We need to store of the IP ID check to be included later 1541 * when we can verify that this packet does in fact belong 1542 * to a given flow. 1543 */ 1544 flush_id = (u16)(id - ntohs(iph2->id)); 1545 1546 /* This bit of code makes it much easier for us to identify 1547 * the cases where we are doing atomic vs non-atomic IP ID 1548 * checks. Specifically an atomic check can return IP ID 1549 * values 0 - 0xFFFF, while a non-atomic check can only 1550 * return 0 or 0xFFFF. 1551 */ 1552 if (!NAPI_GRO_CB(p)->is_atomic || 1553 !(iph->frag_off & htons(IP_DF))) { 1554 flush_id ^= NAPI_GRO_CB(p)->count; 1555 flush_id = flush_id ? 0xFFFF : 0; 1556 } 1557 1558 /* If the previous IP ID value was based on an atomic 1559 * datagram we can overwrite the value and ignore it. 1560 */ 1561 if (NAPI_GRO_CB(skb)->is_atomic) 1562 NAPI_GRO_CB(p)->flush_id = flush_id; 1563 else 1564 NAPI_GRO_CB(p)->flush_id |= flush_id; 1565 } 1566 1567 NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF)); 1568 NAPI_GRO_CB(skb)->flush |= flush; 1569 skb_set_network_header(skb, off); 1570 /* The above will be needed by the transport layer if there is one 1571 * immediately following this IP hdr. 1572 */ 1573 1574 /* Note : No need to call skb_gro_postpull_rcsum() here, 1575 * as we already checked checksum over ipv4 header was 0 1576 */ 1577 skb_gro_pull(skb, sizeof(*iph)); 1578 skb_set_transport_header(skb, skb_gro_offset(skb)); 1579 1580 pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive, 1581 ops->callbacks.gro_receive, head, skb); 1582 1583 out: 1584 skb_gro_flush_final(skb, pp, flush); 1585 1586 return pp; 1587 } 1588 1589 static struct sk_buff *ipip_gro_receive(struct list_head *head, 1590 struct sk_buff *skb) 1591 { 1592 if (NAPI_GRO_CB(skb)->encap_mark) { 1593 NAPI_GRO_CB(skb)->flush = 1; 1594 return NULL; 1595 } 1596 1597 NAPI_GRO_CB(skb)->encap_mark = 1; 1598 1599 return inet_gro_receive(head, skb); 1600 } 1601 1602 #define SECONDS_PER_DAY 86400 1603 1604 /* inet_current_timestamp - Return IP network timestamp 1605 * 1606 * Return milliseconds since midnight in network byte order. 1607 */ 1608 __be32 inet_current_timestamp(void) 1609 { 1610 u32 secs; 1611 u32 msecs; 1612 struct timespec64 ts; 1613 1614 ktime_get_real_ts64(&ts); 1615 1616 /* Get secs since midnight. */ 1617 (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs); 1618 /* Convert to msecs. */ 1619 msecs = secs * MSEC_PER_SEC; 1620 /* Convert nsec to msec. */ 1621 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC; 1622 1623 /* Convert to network byte order. */ 1624 return htonl(msecs); 1625 } 1626 EXPORT_SYMBOL(inet_current_timestamp); 1627 1628 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len) 1629 { 1630 if (sk->sk_family == AF_INET) 1631 return ip_recv_error(sk, msg, len, addr_len); 1632 #if IS_ENABLED(CONFIG_IPV6) 1633 if (sk->sk_family == AF_INET6) 1634 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len); 1635 #endif 1636 return -EINVAL; 1637 } 1638 EXPORT_SYMBOL(inet_recv_error); 1639 1640 int inet_gro_complete(struct sk_buff *skb, int nhoff) 1641 { 1642 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff); 1643 const struct net_offload *ops; 1644 __be16 totlen = iph->tot_len; 1645 int proto = iph->protocol; 1646 int err = -ENOSYS; 1647 1648 if (skb->encapsulation) { 1649 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP)); 1650 skb_set_inner_network_header(skb, nhoff); 1651 } 1652 1653 iph_set_totlen(iph, skb->len - nhoff); 1654 csum_replace2(&iph->check, totlen, iph->tot_len); 1655 1656 ops = rcu_dereference(inet_offloads[proto]); 1657 if (WARN_ON(!ops || !ops->callbacks.gro_complete)) 1658 goto out; 1659 1660 /* Only need to add sizeof(*iph) to get to the next hdr below 1661 * because any hdr with option will have been flushed in 1662 * inet_gro_receive(). 1663 */ 1664 err = INDIRECT_CALL_2(ops->callbacks.gro_complete, 1665 tcp4_gro_complete, udp4_gro_complete, 1666 skb, nhoff + sizeof(*iph)); 1667 1668 out: 1669 return err; 1670 } 1671 1672 static int ipip_gro_complete(struct sk_buff *skb, int nhoff) 1673 { 1674 skb->encapsulation = 1; 1675 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4; 1676 return inet_gro_complete(skb, nhoff); 1677 } 1678 1679 int inet_ctl_sock_create(struct sock **sk, unsigned short family, 1680 unsigned short type, unsigned char protocol, 1681 struct net *net) 1682 { 1683 struct socket *sock; 1684 int rc = sock_create_kern(net, family, type, protocol, &sock); 1685 1686 if (rc == 0) { 1687 *sk = sock->sk; 1688 (*sk)->sk_allocation = GFP_ATOMIC; 1689 (*sk)->sk_use_task_frag = false; 1690 /* 1691 * Unhash it so that IP input processing does not even see it, 1692 * we do not wish this socket to see incoming packets. 1693 */ 1694 (*sk)->sk_prot->unhash(*sk); 1695 } 1696 return rc; 1697 } 1698 EXPORT_SYMBOL_GPL(inet_ctl_sock_create); 1699 1700 unsigned long snmp_fold_field(void __percpu *mib, int offt) 1701 { 1702 unsigned long res = 0; 1703 int i; 1704 1705 for_each_possible_cpu(i) 1706 res += snmp_get_cpu_field(mib, i, offt); 1707 return res; 1708 } 1709 EXPORT_SYMBOL_GPL(snmp_fold_field); 1710 1711 #if BITS_PER_LONG==32 1712 1713 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt, 1714 size_t syncp_offset) 1715 { 1716 void *bhptr; 1717 struct u64_stats_sync *syncp; 1718 u64 v; 1719 unsigned int start; 1720 1721 bhptr = per_cpu_ptr(mib, cpu); 1722 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset); 1723 do { 1724 start = u64_stats_fetch_begin(syncp); 1725 v = *(((u64 *)bhptr) + offt); 1726 } while (u64_stats_fetch_retry(syncp, start)); 1727 1728 return v; 1729 } 1730 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64); 1731 1732 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset) 1733 { 1734 u64 res = 0; 1735 int cpu; 1736 1737 for_each_possible_cpu(cpu) { 1738 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset); 1739 } 1740 return res; 1741 } 1742 EXPORT_SYMBOL_GPL(snmp_fold_field64); 1743 #endif 1744 1745 #ifdef CONFIG_IP_MULTICAST 1746 static const struct net_protocol igmp_protocol = { 1747 .handler = igmp_rcv, 1748 }; 1749 #endif 1750 1751 static const struct net_protocol tcp_protocol = { 1752 .handler = tcp_v4_rcv, 1753 .err_handler = tcp_v4_err, 1754 .no_policy = 1, 1755 .icmp_strict_tag_validation = 1, 1756 }; 1757 1758 static const struct net_protocol udp_protocol = { 1759 .handler = udp_rcv, 1760 .err_handler = udp_err, 1761 .no_policy = 1, 1762 }; 1763 1764 static const struct net_protocol icmp_protocol = { 1765 .handler = icmp_rcv, 1766 .err_handler = icmp_err, 1767 .no_policy = 1, 1768 }; 1769 1770 static __net_init int ipv4_mib_init_net(struct net *net) 1771 { 1772 int i; 1773 1774 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib); 1775 if (!net->mib.tcp_statistics) 1776 goto err_tcp_mib; 1777 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib); 1778 if (!net->mib.ip_statistics) 1779 goto err_ip_mib; 1780 1781 for_each_possible_cpu(i) { 1782 struct ipstats_mib *af_inet_stats; 1783 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i); 1784 u64_stats_init(&af_inet_stats->syncp); 1785 } 1786 1787 net->mib.net_statistics = alloc_percpu(struct linux_mib); 1788 if (!net->mib.net_statistics) 1789 goto err_net_mib; 1790 net->mib.udp_statistics = alloc_percpu(struct udp_mib); 1791 if (!net->mib.udp_statistics) 1792 goto err_udp_mib; 1793 net->mib.udplite_statistics = alloc_percpu(struct udp_mib); 1794 if (!net->mib.udplite_statistics) 1795 goto err_udplite_mib; 1796 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib); 1797 if (!net->mib.icmp_statistics) 1798 goto err_icmp_mib; 1799 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib), 1800 GFP_KERNEL); 1801 if (!net->mib.icmpmsg_statistics) 1802 goto err_icmpmsg_mib; 1803 1804 tcp_mib_init(net); 1805 return 0; 1806 1807 err_icmpmsg_mib: 1808 free_percpu(net->mib.icmp_statistics); 1809 err_icmp_mib: 1810 free_percpu(net->mib.udplite_statistics); 1811 err_udplite_mib: 1812 free_percpu(net->mib.udp_statistics); 1813 err_udp_mib: 1814 free_percpu(net->mib.net_statistics); 1815 err_net_mib: 1816 free_percpu(net->mib.ip_statistics); 1817 err_ip_mib: 1818 free_percpu(net->mib.tcp_statistics); 1819 err_tcp_mib: 1820 return -ENOMEM; 1821 } 1822 1823 static __net_exit void ipv4_mib_exit_net(struct net *net) 1824 { 1825 kfree(net->mib.icmpmsg_statistics); 1826 free_percpu(net->mib.icmp_statistics); 1827 free_percpu(net->mib.udplite_statistics); 1828 free_percpu(net->mib.udp_statistics); 1829 free_percpu(net->mib.net_statistics); 1830 free_percpu(net->mib.ip_statistics); 1831 free_percpu(net->mib.tcp_statistics); 1832 #ifdef CONFIG_MPTCP 1833 /* allocated on demand, see mptcp_init_sock() */ 1834 free_percpu(net->mib.mptcp_statistics); 1835 #endif 1836 } 1837 1838 static __net_initdata struct pernet_operations ipv4_mib_ops = { 1839 .init = ipv4_mib_init_net, 1840 .exit = ipv4_mib_exit_net, 1841 }; 1842 1843 static int __init init_ipv4_mibs(void) 1844 { 1845 return register_pernet_subsys(&ipv4_mib_ops); 1846 } 1847 1848 static __net_init int inet_init_net(struct net *net) 1849 { 1850 /* 1851 * Set defaults for local port range 1852 */ 1853 seqlock_init(&net->ipv4.ip_local_ports.lock); 1854 net->ipv4.ip_local_ports.range[0] = 32768; 1855 net->ipv4.ip_local_ports.range[1] = 60999; 1856 1857 seqlock_init(&net->ipv4.ping_group_range.lock); 1858 /* 1859 * Sane defaults - nobody may create ping sockets. 1860 * Boot scripts should set this to distro-specific group. 1861 */ 1862 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1); 1863 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0); 1864 1865 /* Default values for sysctl-controlled parameters. 1866 * We set them here, in case sysctl is not compiled. 1867 */ 1868 net->ipv4.sysctl_ip_default_ttl = IPDEFTTL; 1869 net->ipv4.sysctl_ip_fwd_update_priority = 1; 1870 net->ipv4.sysctl_ip_dynaddr = 0; 1871 net->ipv4.sysctl_ip_early_demux = 1; 1872 net->ipv4.sysctl_udp_early_demux = 1; 1873 net->ipv4.sysctl_tcp_early_demux = 1; 1874 net->ipv4.sysctl_nexthop_compat_mode = 1; 1875 #ifdef CONFIG_SYSCTL 1876 net->ipv4.sysctl_ip_prot_sock = PROT_SOCK; 1877 #endif 1878 1879 /* Some igmp sysctl, whose values are always used */ 1880 net->ipv4.sysctl_igmp_max_memberships = 20; 1881 net->ipv4.sysctl_igmp_max_msf = 10; 1882 /* IGMP reports for link-local multicast groups are enabled by default */ 1883 net->ipv4.sysctl_igmp_llm_reports = 1; 1884 net->ipv4.sysctl_igmp_qrv = 2; 1885 1886 net->ipv4.sysctl_fib_notify_on_flag_change = 0; 1887 1888 return 0; 1889 } 1890 1891 static __net_initdata struct pernet_operations af_inet_ops = { 1892 .init = inet_init_net, 1893 }; 1894 1895 static int __init init_inet_pernet_ops(void) 1896 { 1897 return register_pernet_subsys(&af_inet_ops); 1898 } 1899 1900 static int ipv4_proc_init(void); 1901 1902 /* 1903 * IP protocol layer initialiser 1904 */ 1905 1906 static struct packet_offload ip_packet_offload __read_mostly = { 1907 .type = cpu_to_be16(ETH_P_IP), 1908 .callbacks = { 1909 .gso_segment = inet_gso_segment, 1910 .gro_receive = inet_gro_receive, 1911 .gro_complete = inet_gro_complete, 1912 }, 1913 }; 1914 1915 static const struct net_offload ipip_offload = { 1916 .callbacks = { 1917 .gso_segment = ipip_gso_segment, 1918 .gro_receive = ipip_gro_receive, 1919 .gro_complete = ipip_gro_complete, 1920 }, 1921 }; 1922 1923 static int __init ipip_offload_init(void) 1924 { 1925 return inet_add_offload(&ipip_offload, IPPROTO_IPIP); 1926 } 1927 1928 static int __init ipv4_offload_init(void) 1929 { 1930 /* 1931 * Add offloads 1932 */ 1933 if (udpv4_offload_init() < 0) 1934 pr_crit("%s: Cannot add UDP protocol offload\n", __func__); 1935 if (tcpv4_offload_init() < 0) 1936 pr_crit("%s: Cannot add TCP protocol offload\n", __func__); 1937 if (ipip_offload_init() < 0) 1938 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__); 1939 1940 dev_add_offload(&ip_packet_offload); 1941 return 0; 1942 } 1943 1944 fs_initcall(ipv4_offload_init); 1945 1946 static struct packet_type ip_packet_type __read_mostly = { 1947 .type = cpu_to_be16(ETH_P_IP), 1948 .func = ip_rcv, 1949 .list_func = ip_list_rcv, 1950 }; 1951 1952 static int __init inet_init(void) 1953 { 1954 struct inet_protosw *q; 1955 struct list_head *r; 1956 int rc; 1957 1958 sock_skb_cb_check_size(sizeof(struct inet_skb_parm)); 1959 1960 raw_hashinfo_init(&raw_v4_hashinfo); 1961 1962 rc = proto_register(&tcp_prot, 1); 1963 if (rc) 1964 goto out; 1965 1966 rc = proto_register(&udp_prot, 1); 1967 if (rc) 1968 goto out_unregister_tcp_proto; 1969 1970 rc = proto_register(&raw_prot, 1); 1971 if (rc) 1972 goto out_unregister_udp_proto; 1973 1974 rc = proto_register(&ping_prot, 1); 1975 if (rc) 1976 goto out_unregister_raw_proto; 1977 1978 /* 1979 * Tell SOCKET that we are alive... 1980 */ 1981 1982 (void)sock_register(&inet_family_ops); 1983 1984 #ifdef CONFIG_SYSCTL 1985 ip_static_sysctl_init(); 1986 #endif 1987 1988 /* 1989 * Add all the base protocols. 1990 */ 1991 1992 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0) 1993 pr_crit("%s: Cannot add ICMP protocol\n", __func__); 1994 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0) 1995 pr_crit("%s: Cannot add UDP protocol\n", __func__); 1996 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0) 1997 pr_crit("%s: Cannot add TCP protocol\n", __func__); 1998 #ifdef CONFIG_IP_MULTICAST 1999 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0) 2000 pr_crit("%s: Cannot add IGMP protocol\n", __func__); 2001 #endif 2002 2003 /* Register the socket-side information for inet_create. */ 2004 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r) 2005 INIT_LIST_HEAD(r); 2006 2007 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q) 2008 inet_register_protosw(q); 2009 2010 /* 2011 * Set the ARP module up 2012 */ 2013 2014 arp_init(); 2015 2016 /* 2017 * Set the IP module up 2018 */ 2019 2020 ip_init(); 2021 2022 /* Initialise per-cpu ipv4 mibs */ 2023 if (init_ipv4_mibs()) 2024 panic("%s: Cannot init ipv4 mibs\n", __func__); 2025 2026 /* Setup TCP slab cache for open requests. */ 2027 tcp_init(); 2028 2029 /* Setup UDP memory threshold */ 2030 udp_init(); 2031 2032 /* Add UDP-Lite (RFC 3828) */ 2033 udplite4_register(); 2034 2035 raw_init(); 2036 2037 ping_init(); 2038 2039 /* 2040 * Set the ICMP layer up 2041 */ 2042 2043 if (icmp_init() < 0) 2044 panic("Failed to create the ICMP control socket.\n"); 2045 2046 /* 2047 * Initialise the multicast router 2048 */ 2049 #if defined(CONFIG_IP_MROUTE) 2050 if (ip_mr_init()) 2051 pr_crit("%s: Cannot init ipv4 mroute\n", __func__); 2052 #endif 2053 2054 if (init_inet_pernet_ops()) 2055 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__); 2056 2057 ipv4_proc_init(); 2058 2059 ipfrag_init(); 2060 2061 dev_add_pack(&ip_packet_type); 2062 2063 ip_tunnel_core_init(); 2064 2065 rc = 0; 2066 out: 2067 return rc; 2068 out_unregister_raw_proto: 2069 proto_unregister(&raw_prot); 2070 out_unregister_udp_proto: 2071 proto_unregister(&udp_prot); 2072 out_unregister_tcp_proto: 2073 proto_unregister(&tcp_prot); 2074 goto out; 2075 } 2076 2077 fs_initcall(inet_init); 2078 2079 /* ------------------------------------------------------------------------ */ 2080 2081 #ifdef CONFIG_PROC_FS 2082 static int __init ipv4_proc_init(void) 2083 { 2084 int rc = 0; 2085 2086 if (raw_proc_init()) 2087 goto out_raw; 2088 if (tcp4_proc_init()) 2089 goto out_tcp; 2090 if (udp4_proc_init()) 2091 goto out_udp; 2092 if (ping_proc_init()) 2093 goto out_ping; 2094 if (ip_misc_proc_init()) 2095 goto out_misc; 2096 out: 2097 return rc; 2098 out_misc: 2099 ping_proc_exit(); 2100 out_ping: 2101 udp4_proc_exit(); 2102 out_udp: 2103 tcp4_proc_exit(); 2104 out_tcp: 2105 raw_proc_exit(); 2106 out_raw: 2107 rc = -ENOMEM; 2108 goto out; 2109 } 2110 2111 #else /* CONFIG_PROC_FS */ 2112 static int __init ipv4_proc_init(void) 2113 { 2114 return 0; 2115 } 2116 #endif /* CONFIG_PROC_FS */ 2117