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/tcp.h> 104 #include <net/udp.h> 105 #include <net/udplite.h> 106 #include <net/ping.h> 107 #include <linux/skbuff.h> 108 #include <net/sock.h> 109 #include <net/raw.h> 110 #include <net/icmp.h> 111 #include <net/inet_common.h> 112 #include <net/ip_tunnels.h> 113 #include <net/xfrm.h> 114 #include <net/net_namespace.h> 115 #include <net/secure_seq.h> 116 #ifdef CONFIG_IP_MROUTE 117 #include <linux/mroute.h> 118 #endif 119 #include <net/l3mdev.h> 120 #include <net/compat.h> 121 122 #include <trace/events/sock.h> 123 124 /* The inetsw table contains everything that inet_create needs to 125 * build a new socket. 126 */ 127 static struct list_head inetsw[SOCK_MAX]; 128 static DEFINE_SPINLOCK(inetsw_lock); 129 130 /* New destruction routine */ 131 132 void inet_sock_destruct(struct sock *sk) 133 { 134 struct inet_sock *inet = inet_sk(sk); 135 136 __skb_queue_purge(&sk->sk_receive_queue); 137 __skb_queue_purge(&sk->sk_error_queue); 138 139 sk_mem_reclaim_final(sk); 140 141 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) { 142 pr_err("Attempt to release TCP socket in state %d %p\n", 143 sk->sk_state, sk); 144 return; 145 } 146 if (!sock_flag(sk, SOCK_DEAD)) { 147 pr_err("Attempt to release alive inet socket %p\n", sk); 148 return; 149 } 150 151 WARN_ON(atomic_read(&sk->sk_rmem_alloc)); 152 WARN_ON(refcount_read(&sk->sk_wmem_alloc)); 153 WARN_ON(sk->sk_wmem_queued); 154 WARN_ON(sk_forward_alloc_get(sk)); 155 156 kfree(rcu_dereference_protected(inet->inet_opt, 1)); 157 dst_release(rcu_dereference_protected(sk->sk_dst_cache, 1)); 158 dst_release(sk->sk_rx_dst); 159 sk_refcnt_debug_dec(sk); 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 /* 191 * Move a socket into listening state. 192 */ 193 int inet_listen(struct socket *sock, int backlog) 194 { 195 struct sock *sk = sock->sk; 196 unsigned char old_state; 197 int err, tcp_fastopen; 198 199 lock_sock(sk); 200 201 err = -EINVAL; 202 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM) 203 goto out; 204 205 old_state = sk->sk_state; 206 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN))) 207 goto out; 208 209 WRITE_ONCE(sk->sk_max_ack_backlog, backlog); 210 /* Really, if the socket is already in listen state 211 * we can only allow the backlog to be adjusted. 212 */ 213 if (old_state != TCP_LISTEN) { 214 /* Enable TFO w/o requiring TCP_FASTOPEN socket option. 215 * Note that only TCP sockets (SOCK_STREAM) will reach here. 216 * Also fastopen backlog may already been set via the option 217 * because the socket was in TCP_LISTEN state previously but 218 * was shutdown() rather than close(). 219 */ 220 tcp_fastopen = sock_net(sk)->ipv4.sysctl_tcp_fastopen; 221 if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) && 222 (tcp_fastopen & TFO_SERVER_ENABLE) && 223 !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) { 224 fastopen_queue_tune(sk, backlog); 225 tcp_fastopen_init_key_once(sock_net(sk)); 226 } 227 228 err = inet_csk_listen_start(sk); 229 if (err) 230 goto out; 231 tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL); 232 } 233 err = 0; 234 235 out: 236 release_sock(sk); 237 return err; 238 } 239 EXPORT_SYMBOL(inet_listen); 240 241 /* 242 * Create an inet socket. 243 */ 244 245 static int inet_create(struct net *net, struct socket *sock, int protocol, 246 int kern) 247 { 248 struct sock *sk; 249 struct inet_protosw *answer; 250 struct inet_sock *inet; 251 struct proto *answer_prot; 252 unsigned char answer_flags; 253 int try_loading_module = 0; 254 int err; 255 256 if (protocol < 0 || protocol >= IPPROTO_MAX) 257 return -EINVAL; 258 259 sock->state = SS_UNCONNECTED; 260 261 /* Look for the requested type/protocol pair. */ 262 lookup_protocol: 263 err = -ESOCKTNOSUPPORT; 264 rcu_read_lock(); 265 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) { 266 267 err = 0; 268 /* Check the non-wild match. */ 269 if (protocol == answer->protocol) { 270 if (protocol != IPPROTO_IP) 271 break; 272 } else { 273 /* Check for the two wild cases. */ 274 if (IPPROTO_IP == protocol) { 275 protocol = answer->protocol; 276 break; 277 } 278 if (IPPROTO_IP == answer->protocol) 279 break; 280 } 281 err = -EPROTONOSUPPORT; 282 } 283 284 if (unlikely(err)) { 285 if (try_loading_module < 2) { 286 rcu_read_unlock(); 287 /* 288 * Be more specific, e.g. net-pf-2-proto-132-type-1 289 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM) 290 */ 291 if (++try_loading_module == 1) 292 request_module("net-pf-%d-proto-%d-type-%d", 293 PF_INET, protocol, sock->type); 294 /* 295 * Fall back to generic, e.g. net-pf-2-proto-132 296 * (net-pf-PF_INET-proto-IPPROTO_SCTP) 297 */ 298 else 299 request_module("net-pf-%d-proto-%d", 300 PF_INET, protocol); 301 goto lookup_protocol; 302 } else 303 goto out_rcu_unlock; 304 } 305 306 err = -EPERM; 307 if (sock->type == SOCK_RAW && !kern && 308 !ns_capable(net->user_ns, CAP_NET_RAW)) 309 goto out_rcu_unlock; 310 311 sock->ops = answer->ops; 312 answer_prot = answer->prot; 313 answer_flags = answer->flags; 314 rcu_read_unlock(); 315 316 WARN_ON(!answer_prot->slab); 317 318 err = -ENOMEM; 319 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern); 320 if (!sk) 321 goto out; 322 323 err = 0; 324 if (INET_PROTOSW_REUSE & answer_flags) 325 sk->sk_reuse = SK_CAN_REUSE; 326 327 inet = inet_sk(sk); 328 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0; 329 330 inet->nodefrag = 0; 331 332 if (SOCK_RAW == sock->type) { 333 inet->inet_num = protocol; 334 if (IPPROTO_RAW == protocol) 335 inet->hdrincl = 1; 336 } 337 338 if (net->ipv4.sysctl_ip_no_pmtu_disc) 339 inet->pmtudisc = IP_PMTUDISC_DONT; 340 else 341 inet->pmtudisc = IP_PMTUDISC_WANT; 342 343 inet->inet_id = 0; 344 345 sock_init_data(sock, sk); 346 347 sk->sk_destruct = inet_sock_destruct; 348 sk->sk_protocol = protocol; 349 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv; 350 351 inet->uc_ttl = -1; 352 inet->mc_loop = 1; 353 inet->mc_ttl = 1; 354 inet->mc_all = 1; 355 inet->mc_index = 0; 356 inet->mc_list = NULL; 357 inet->rcv_tos = 0; 358 359 sk_refcnt_debug_inc(sk); 360 361 if (inet->inet_num) { 362 /* It assumes that any protocol which allows 363 * the user to assign a number at socket 364 * creation time automatically 365 * shares. 366 */ 367 inet->inet_sport = htons(inet->inet_num); 368 /* Add to protocol hash chains. */ 369 err = sk->sk_prot->hash(sk); 370 if (err) { 371 sk_common_release(sk); 372 goto out; 373 } 374 } 375 376 if (sk->sk_prot->init) { 377 err = sk->sk_prot->init(sk); 378 if (err) { 379 sk_common_release(sk); 380 goto out; 381 } 382 } 383 384 if (!kern) { 385 err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk); 386 if (err) { 387 sk_common_release(sk); 388 goto out; 389 } 390 } 391 out: 392 return err; 393 out_rcu_unlock: 394 rcu_read_unlock(); 395 goto out; 396 } 397 398 399 /* 400 * The peer socket should always be NULL (or else). When we call this 401 * function we are destroying the object and from then on nobody 402 * should refer to it. 403 */ 404 int inet_release(struct socket *sock) 405 { 406 struct sock *sk = sock->sk; 407 408 if (sk) { 409 long timeout; 410 411 if (!sk->sk_kern_sock) 412 BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk); 413 414 /* Applications forget to leave groups before exiting */ 415 ip_mc_drop_socket(sk); 416 417 /* If linger is set, we don't return until the close 418 * is complete. Otherwise we return immediately. The 419 * actually closing is done the same either way. 420 * 421 * If the close is due to the process exiting, we never 422 * linger.. 423 */ 424 timeout = 0; 425 if (sock_flag(sk, SOCK_LINGER) && 426 !(current->flags & PF_EXITING)) 427 timeout = sk->sk_lingertime; 428 sk->sk_prot->close(sk, timeout); 429 sock->sk = NULL; 430 } 431 return 0; 432 } 433 EXPORT_SYMBOL(inet_release); 434 435 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) 436 { 437 struct sock *sk = sock->sk; 438 u32 flags = BIND_WITH_LOCK; 439 int err; 440 441 /* If the socket has its own bind function then use it. (RAW) */ 442 if (sk->sk_prot->bind) { 443 return sk->sk_prot->bind(sk, uaddr, addr_len); 444 } 445 if (addr_len < sizeof(struct sockaddr_in)) 446 return -EINVAL; 447 448 /* BPF prog is run before any checks are done so that if the prog 449 * changes context in a wrong way it will be caught. 450 */ 451 err = BPF_CGROUP_RUN_PROG_INET_BIND_LOCK(sk, uaddr, 452 CGROUP_INET4_BIND, &flags); 453 if (err) 454 return err; 455 456 return __inet_bind(sk, uaddr, addr_len, flags); 457 } 458 EXPORT_SYMBOL(inet_bind); 459 460 int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len, 461 u32 flags) 462 { 463 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr; 464 struct inet_sock *inet = inet_sk(sk); 465 struct net *net = sock_net(sk); 466 unsigned short snum; 467 int chk_addr_ret; 468 u32 tb_id = RT_TABLE_LOCAL; 469 int err; 470 471 if (addr->sin_family != AF_INET) { 472 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET) 473 * only if s_addr is INADDR_ANY. 474 */ 475 err = -EAFNOSUPPORT; 476 if (addr->sin_family != AF_UNSPEC || 477 addr->sin_addr.s_addr != htonl(INADDR_ANY)) 478 goto out; 479 } 480 481 tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id; 482 chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id); 483 484 /* Not specified by any standard per-se, however it breaks too 485 * many applications when removed. It is unfortunate since 486 * allowing applications to make a non-local bind solves 487 * several problems with systems using dynamic addressing. 488 * (ie. your servers still start up even if your ISDN link 489 * is temporarily down) 490 */ 491 err = -EADDRNOTAVAIL; 492 if (!inet_addr_valid_or_nonlocal(net, inet, addr->sin_addr.s_addr, 493 chk_addr_ret)) 494 goto out; 495 496 snum = ntohs(addr->sin_port); 497 err = -EACCES; 498 if (!(flags & BIND_NO_CAP_NET_BIND_SERVICE) && 499 snum && inet_port_requires_bind_service(net, snum) && 500 !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE)) 501 goto out; 502 503 /* We keep a pair of addresses. rcv_saddr is the one 504 * used by hash lookups, and saddr is used for transmit. 505 * 506 * In the BSD API these are the same except where it 507 * would be illegal to use them (multicast/broadcast) in 508 * which case the sending device address is used. 509 */ 510 if (flags & BIND_WITH_LOCK) 511 lock_sock(sk); 512 513 /* Check these errors (active socket, double bind). */ 514 err = -EINVAL; 515 if (sk->sk_state != TCP_CLOSE || inet->inet_num) 516 goto out_release_sock; 517 518 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr; 519 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST) 520 inet->inet_saddr = 0; /* Use device */ 521 522 /* Make sure we are allowed to bind here. */ 523 if (snum || !(inet->bind_address_no_port || 524 (flags & BIND_FORCE_ADDRESS_NO_PORT))) { 525 if (sk->sk_prot->get_port(sk, snum)) { 526 inet->inet_saddr = inet->inet_rcv_saddr = 0; 527 err = -EADDRINUSE; 528 goto out_release_sock; 529 } 530 if (!(flags & BIND_FROM_BPF)) { 531 err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk); 532 if (err) { 533 inet->inet_saddr = inet->inet_rcv_saddr = 0; 534 goto out_release_sock; 535 } 536 } 537 } 538 539 if (inet->inet_rcv_saddr) 540 sk->sk_userlocks |= SOCK_BINDADDR_LOCK; 541 if (snum) 542 sk->sk_userlocks |= SOCK_BINDPORT_LOCK; 543 inet->inet_sport = htons(inet->inet_num); 544 inet->inet_daddr = 0; 545 inet->inet_dport = 0; 546 sk_dst_reset(sk); 547 err = 0; 548 out_release_sock: 549 if (flags & BIND_WITH_LOCK) 550 release_sock(sk); 551 out: 552 return err; 553 } 554 555 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr, 556 int addr_len, int flags) 557 { 558 struct sock *sk = sock->sk; 559 int err; 560 561 if (addr_len < sizeof(uaddr->sa_family)) 562 return -EINVAL; 563 if (uaddr->sa_family == AF_UNSPEC) 564 return sk->sk_prot->disconnect(sk, flags); 565 566 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) { 567 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len); 568 if (err) 569 return err; 570 } 571 572 if (data_race(!inet_sk(sk)->inet_num) && inet_autobind(sk)) 573 return -EAGAIN; 574 return sk->sk_prot->connect(sk, uaddr, addr_len); 575 } 576 EXPORT_SYMBOL(inet_dgram_connect); 577 578 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias) 579 { 580 DEFINE_WAIT_FUNC(wait, woken_wake_function); 581 582 add_wait_queue(sk_sleep(sk), &wait); 583 sk->sk_write_pending += writebias; 584 585 /* Basic assumption: if someone sets sk->sk_err, he _must_ 586 * change state of the socket from TCP_SYN_*. 587 * Connect() does not allow to get error notifications 588 * without closing the socket. 589 */ 590 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 591 release_sock(sk); 592 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo); 593 lock_sock(sk); 594 if (signal_pending(current) || !timeo) 595 break; 596 } 597 remove_wait_queue(sk_sleep(sk), &wait); 598 sk->sk_write_pending -= writebias; 599 return timeo; 600 } 601 602 /* 603 * Connect to a remote host. There is regrettably still a little 604 * TCP 'magic' in here. 605 */ 606 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr, 607 int addr_len, int flags, int is_sendmsg) 608 { 609 struct sock *sk = sock->sk; 610 int err; 611 long timeo; 612 613 /* 614 * uaddr can be NULL and addr_len can be 0 if: 615 * sk is a TCP fastopen active socket and 616 * TCP_FASTOPEN_CONNECT sockopt is set and 617 * we already have a valid cookie for this socket. 618 * In this case, user can call write() after connect(). 619 * write() will invoke tcp_sendmsg_fastopen() which calls 620 * __inet_stream_connect(). 621 */ 622 if (uaddr) { 623 if (addr_len < sizeof(uaddr->sa_family)) 624 return -EINVAL; 625 626 if (uaddr->sa_family == AF_UNSPEC) { 627 err = sk->sk_prot->disconnect(sk, flags); 628 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED; 629 goto out; 630 } 631 } 632 633 switch (sock->state) { 634 default: 635 err = -EINVAL; 636 goto out; 637 case SS_CONNECTED: 638 err = -EISCONN; 639 goto out; 640 case SS_CONNECTING: 641 if (inet_sk(sk)->defer_connect) 642 err = is_sendmsg ? -EINPROGRESS : -EISCONN; 643 else 644 err = -EALREADY; 645 /* Fall out of switch with err, set for this state */ 646 break; 647 case SS_UNCONNECTED: 648 err = -EISCONN; 649 if (sk->sk_state != TCP_CLOSE) 650 goto out; 651 652 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) { 653 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len); 654 if (err) 655 goto out; 656 } 657 658 err = sk->sk_prot->connect(sk, uaddr, addr_len); 659 if (err < 0) 660 goto out; 661 662 sock->state = SS_CONNECTING; 663 664 if (!err && inet_sk(sk)->defer_connect) 665 goto out; 666 667 /* Just entered SS_CONNECTING state; the only 668 * difference is that return value in non-blocking 669 * case is EINPROGRESS, rather than EALREADY. 670 */ 671 err = -EINPROGRESS; 672 break; 673 } 674 675 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK); 676 677 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 678 int writebias = (sk->sk_protocol == IPPROTO_TCP) && 679 tcp_sk(sk)->fastopen_req && 680 tcp_sk(sk)->fastopen_req->data ? 1 : 0; 681 682 /* Error code is set above */ 683 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias)) 684 goto out; 685 686 err = sock_intr_errno(timeo); 687 if (signal_pending(current)) 688 goto out; 689 } 690 691 /* Connection was closed by RST, timeout, ICMP error 692 * or another process disconnected us. 693 */ 694 if (sk->sk_state == TCP_CLOSE) 695 goto sock_error; 696 697 /* sk->sk_err may be not zero now, if RECVERR was ordered by user 698 * and error was received after socket entered established state. 699 * Hence, it is handled normally after connect() return successfully. 700 */ 701 702 sock->state = SS_CONNECTED; 703 err = 0; 704 out: 705 return err; 706 707 sock_error: 708 err = sock_error(sk) ? : -ECONNABORTED; 709 sock->state = SS_UNCONNECTED; 710 if (sk->sk_prot->disconnect(sk, flags)) 711 sock->state = SS_DISCONNECTING; 712 goto out; 713 } 714 EXPORT_SYMBOL(__inet_stream_connect); 715 716 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr, 717 int addr_len, int flags) 718 { 719 int err; 720 721 lock_sock(sock->sk); 722 err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0); 723 release_sock(sock->sk); 724 return err; 725 } 726 EXPORT_SYMBOL(inet_stream_connect); 727 728 /* 729 * Accept a pending connection. The TCP layer now gives BSD semantics. 730 */ 731 732 int inet_accept(struct socket *sock, struct socket *newsock, int flags, 733 bool kern) 734 { 735 struct sock *sk1 = sock->sk; 736 int err = -EINVAL; 737 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err, kern); 738 739 if (!sk2) 740 goto do_err; 741 742 lock_sock(sk2); 743 744 sock_rps_record_flow(sk2); 745 WARN_ON(!((1 << sk2->sk_state) & 746 (TCPF_ESTABLISHED | TCPF_SYN_RECV | 747 TCPF_CLOSE_WAIT | TCPF_CLOSE))); 748 749 sock_graft(sk2, newsock); 750 751 newsock->state = SS_CONNECTED; 752 err = 0; 753 release_sock(sk2); 754 do_err: 755 return err; 756 } 757 EXPORT_SYMBOL(inet_accept); 758 759 /* 760 * This does both peername and sockname. 761 */ 762 int inet_getname(struct socket *sock, struct sockaddr *uaddr, 763 int peer) 764 { 765 struct sock *sk = sock->sk; 766 struct inet_sock *inet = inet_sk(sk); 767 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr); 768 769 sin->sin_family = AF_INET; 770 lock_sock(sk); 771 if (peer) { 772 if (!inet->inet_dport || 773 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) && 774 peer == 1)) { 775 release_sock(sk); 776 return -ENOTCONN; 777 } 778 sin->sin_port = inet->inet_dport; 779 sin->sin_addr.s_addr = inet->inet_daddr; 780 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin, 781 CGROUP_INET4_GETPEERNAME); 782 } else { 783 __be32 addr = inet->inet_rcv_saddr; 784 if (!addr) 785 addr = inet->inet_saddr; 786 sin->sin_port = inet->inet_sport; 787 sin->sin_addr.s_addr = addr; 788 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin, 789 CGROUP_INET4_GETSOCKNAME); 790 } 791 release_sock(sk); 792 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 793 return sizeof(*sin); 794 } 795 EXPORT_SYMBOL(inet_getname); 796 797 int inet_send_prepare(struct sock *sk) 798 { 799 sock_rps_record_flow(sk); 800 801 /* We may need to bind the socket. */ 802 if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind && 803 inet_autobind(sk)) 804 return -EAGAIN; 805 806 return 0; 807 } 808 EXPORT_SYMBOL_GPL(inet_send_prepare); 809 810 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size) 811 { 812 struct sock *sk = sock->sk; 813 814 if (unlikely(inet_send_prepare(sk))) 815 return -EAGAIN; 816 817 return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg, 818 sk, msg, size); 819 } 820 EXPORT_SYMBOL(inet_sendmsg); 821 822 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, 823 size_t size, int flags) 824 { 825 struct sock *sk = sock->sk; 826 827 if (unlikely(inet_send_prepare(sk))) 828 return -EAGAIN; 829 830 if (sk->sk_prot->sendpage) 831 return sk->sk_prot->sendpage(sk, page, offset, size, flags); 832 return sock_no_sendpage(sock, page, offset, size, flags); 833 } 834 EXPORT_SYMBOL(inet_sendpage); 835 836 INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *, 837 size_t, int, int, int *)); 838 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 839 int flags) 840 { 841 struct sock *sk = sock->sk; 842 int addr_len = 0; 843 int err; 844 845 if (likely(!(flags & MSG_ERRQUEUE))) 846 sock_rps_record_flow(sk); 847 848 err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg, 849 sk, msg, size, flags & MSG_DONTWAIT, 850 flags & ~MSG_DONTWAIT, &addr_len); 851 if (err >= 0) 852 msg->msg_namelen = addr_len; 853 return err; 854 } 855 EXPORT_SYMBOL(inet_recvmsg); 856 857 int inet_shutdown(struct socket *sock, int how) 858 { 859 struct sock *sk = sock->sk; 860 int err = 0; 861 862 /* This should really check to make sure 863 * the socket is a TCP socket. (WHY AC...) 864 */ 865 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and 866 1->2 bit 2 snds. 867 2->3 */ 868 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */ 869 return -EINVAL; 870 871 lock_sock(sk); 872 if (sock->state == SS_CONNECTING) { 873 if ((1 << sk->sk_state) & 874 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE)) 875 sock->state = SS_DISCONNECTING; 876 else 877 sock->state = SS_CONNECTED; 878 } 879 880 switch (sk->sk_state) { 881 case TCP_CLOSE: 882 err = -ENOTCONN; 883 /* Hack to wake up other listeners, who can poll for 884 EPOLLHUP, even on eg. unconnected UDP sockets -- RR */ 885 fallthrough; 886 default: 887 sk->sk_shutdown |= how; 888 if (sk->sk_prot->shutdown) 889 sk->sk_prot->shutdown(sk, how); 890 break; 891 892 /* Remaining two branches are temporary solution for missing 893 * close() in multithreaded environment. It is _not_ a good idea, 894 * but we have no choice until close() is repaired at VFS level. 895 */ 896 case TCP_LISTEN: 897 if (!(how & RCV_SHUTDOWN)) 898 break; 899 fallthrough; 900 case TCP_SYN_SENT: 901 err = sk->sk_prot->disconnect(sk, O_NONBLOCK); 902 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED; 903 break; 904 } 905 906 /* Wake up anyone sleeping in poll. */ 907 sk->sk_state_change(sk); 908 release_sock(sk); 909 return err; 910 } 911 EXPORT_SYMBOL(inet_shutdown); 912 913 /* 914 * ioctl() calls you can issue on an INET socket. Most of these are 915 * device configuration and stuff and very rarely used. Some ioctls 916 * pass on to the socket itself. 917 * 918 * NOTE: I like the idea of a module for the config stuff. ie ifconfig 919 * loads the devconfigure module does its configuring and unloads it. 920 * There's a good 20K of config code hanging around the kernel. 921 */ 922 923 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 924 { 925 struct sock *sk = sock->sk; 926 int err = 0; 927 struct net *net = sock_net(sk); 928 void __user *p = (void __user *)arg; 929 struct ifreq ifr; 930 struct rtentry rt; 931 932 switch (cmd) { 933 case SIOCADDRT: 934 case SIOCDELRT: 935 if (copy_from_user(&rt, p, sizeof(struct rtentry))) 936 return -EFAULT; 937 err = ip_rt_ioctl(net, cmd, &rt); 938 break; 939 case SIOCRTMSG: 940 err = -EINVAL; 941 break; 942 case SIOCDARP: 943 case SIOCGARP: 944 case SIOCSARP: 945 err = arp_ioctl(net, cmd, (void __user *)arg); 946 break; 947 case SIOCGIFADDR: 948 case SIOCGIFBRDADDR: 949 case SIOCGIFNETMASK: 950 case SIOCGIFDSTADDR: 951 case SIOCGIFPFLAGS: 952 if (get_user_ifreq(&ifr, NULL, p)) 953 return -EFAULT; 954 err = devinet_ioctl(net, cmd, &ifr); 955 if (!err && put_user_ifreq(&ifr, p)) 956 err = -EFAULT; 957 break; 958 959 case SIOCSIFADDR: 960 case SIOCSIFBRDADDR: 961 case SIOCSIFNETMASK: 962 case SIOCSIFDSTADDR: 963 case SIOCSIFPFLAGS: 964 case SIOCSIFFLAGS: 965 if (get_user_ifreq(&ifr, NULL, p)) 966 return -EFAULT; 967 err = devinet_ioctl(net, cmd, &ifr); 968 break; 969 default: 970 if (sk->sk_prot->ioctl) 971 err = sk->sk_prot->ioctl(sk, cmd, arg); 972 else 973 err = -ENOIOCTLCMD; 974 break; 975 } 976 return err; 977 } 978 EXPORT_SYMBOL(inet_ioctl); 979 980 #ifdef CONFIG_COMPAT 981 static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd, 982 struct compat_rtentry __user *ur) 983 { 984 compat_uptr_t rtdev; 985 struct rtentry rt; 986 987 if (copy_from_user(&rt.rt_dst, &ur->rt_dst, 988 3 * sizeof(struct sockaddr)) || 989 get_user(rt.rt_flags, &ur->rt_flags) || 990 get_user(rt.rt_metric, &ur->rt_metric) || 991 get_user(rt.rt_mtu, &ur->rt_mtu) || 992 get_user(rt.rt_window, &ur->rt_window) || 993 get_user(rt.rt_irtt, &ur->rt_irtt) || 994 get_user(rtdev, &ur->rt_dev)) 995 return -EFAULT; 996 997 rt.rt_dev = compat_ptr(rtdev); 998 return ip_rt_ioctl(sock_net(sk), cmd, &rt); 999 } 1000 1001 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 1002 { 1003 void __user *argp = compat_ptr(arg); 1004 struct sock *sk = sock->sk; 1005 1006 switch (cmd) { 1007 case SIOCADDRT: 1008 case SIOCDELRT: 1009 return inet_compat_routing_ioctl(sk, cmd, argp); 1010 default: 1011 if (!sk->sk_prot->compat_ioctl) 1012 return -ENOIOCTLCMD; 1013 return sk->sk_prot->compat_ioctl(sk, cmd, arg); 1014 } 1015 } 1016 #endif /* CONFIG_COMPAT */ 1017 1018 const struct proto_ops inet_stream_ops = { 1019 .family = PF_INET, 1020 .owner = THIS_MODULE, 1021 .release = inet_release, 1022 .bind = inet_bind, 1023 .connect = inet_stream_connect, 1024 .socketpair = sock_no_socketpair, 1025 .accept = inet_accept, 1026 .getname = inet_getname, 1027 .poll = tcp_poll, 1028 .ioctl = inet_ioctl, 1029 .gettstamp = sock_gettstamp, 1030 .listen = inet_listen, 1031 .shutdown = inet_shutdown, 1032 .setsockopt = sock_common_setsockopt, 1033 .getsockopt = sock_common_getsockopt, 1034 .sendmsg = inet_sendmsg, 1035 .recvmsg = inet_recvmsg, 1036 #ifdef CONFIG_MMU 1037 .mmap = tcp_mmap, 1038 #endif 1039 .sendpage = inet_sendpage, 1040 .splice_read = tcp_splice_read, 1041 .read_sock = tcp_read_sock, 1042 .sendmsg_locked = tcp_sendmsg_locked, 1043 .sendpage_locked = tcp_sendpage_locked, 1044 .peek_len = tcp_peek_len, 1045 #ifdef CONFIG_COMPAT 1046 .compat_ioctl = inet_compat_ioctl, 1047 #endif 1048 .set_rcvlowat = tcp_set_rcvlowat, 1049 }; 1050 EXPORT_SYMBOL(inet_stream_ops); 1051 1052 const struct proto_ops inet_dgram_ops = { 1053 .family = PF_INET, 1054 .owner = THIS_MODULE, 1055 .release = inet_release, 1056 .bind = inet_bind, 1057 .connect = inet_dgram_connect, 1058 .socketpair = sock_no_socketpair, 1059 .accept = sock_no_accept, 1060 .getname = inet_getname, 1061 .poll = udp_poll, 1062 .ioctl = inet_ioctl, 1063 .gettstamp = sock_gettstamp, 1064 .listen = sock_no_listen, 1065 .shutdown = inet_shutdown, 1066 .setsockopt = sock_common_setsockopt, 1067 .getsockopt = sock_common_getsockopt, 1068 .sendmsg = inet_sendmsg, 1069 .read_sock = udp_read_sock, 1070 .recvmsg = inet_recvmsg, 1071 .mmap = sock_no_mmap, 1072 .sendpage = inet_sendpage, 1073 .set_peek_off = sk_set_peek_off, 1074 #ifdef CONFIG_COMPAT 1075 .compat_ioctl = inet_compat_ioctl, 1076 #endif 1077 }; 1078 EXPORT_SYMBOL(inet_dgram_ops); 1079 1080 /* 1081 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without 1082 * udp_poll 1083 */ 1084 static const struct proto_ops inet_sockraw_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 = datagram_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 .recvmsg = inet_recvmsg, 1102 .mmap = sock_no_mmap, 1103 .sendpage = inet_sendpage, 1104 #ifdef CONFIG_COMPAT 1105 .compat_ioctl = inet_compat_ioctl, 1106 #endif 1107 }; 1108 1109 static const struct net_proto_family inet_family_ops = { 1110 .family = PF_INET, 1111 .create = inet_create, 1112 .owner = THIS_MODULE, 1113 }; 1114 1115 /* Upon startup we insert all the elements in inetsw_array[] into 1116 * the linked list inetsw. 1117 */ 1118 static struct inet_protosw inetsw_array[] = 1119 { 1120 { 1121 .type = SOCK_STREAM, 1122 .protocol = IPPROTO_TCP, 1123 .prot = &tcp_prot, 1124 .ops = &inet_stream_ops, 1125 .flags = INET_PROTOSW_PERMANENT | 1126 INET_PROTOSW_ICSK, 1127 }, 1128 1129 { 1130 .type = SOCK_DGRAM, 1131 .protocol = IPPROTO_UDP, 1132 .prot = &udp_prot, 1133 .ops = &inet_dgram_ops, 1134 .flags = INET_PROTOSW_PERMANENT, 1135 }, 1136 1137 { 1138 .type = SOCK_DGRAM, 1139 .protocol = IPPROTO_ICMP, 1140 .prot = &ping_prot, 1141 .ops = &inet_sockraw_ops, 1142 .flags = INET_PROTOSW_REUSE, 1143 }, 1144 1145 { 1146 .type = SOCK_RAW, 1147 .protocol = IPPROTO_IP, /* wild card */ 1148 .prot = &raw_prot, 1149 .ops = &inet_sockraw_ops, 1150 .flags = INET_PROTOSW_REUSE, 1151 } 1152 }; 1153 1154 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array) 1155 1156 void inet_register_protosw(struct inet_protosw *p) 1157 { 1158 struct list_head *lh; 1159 struct inet_protosw *answer; 1160 int protocol = p->protocol; 1161 struct list_head *last_perm; 1162 1163 spin_lock_bh(&inetsw_lock); 1164 1165 if (p->type >= SOCK_MAX) 1166 goto out_illegal; 1167 1168 /* If we are trying to override a permanent protocol, bail. */ 1169 last_perm = &inetsw[p->type]; 1170 list_for_each(lh, &inetsw[p->type]) { 1171 answer = list_entry(lh, struct inet_protosw, list); 1172 /* Check only the non-wild match. */ 1173 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0) 1174 break; 1175 if (protocol == answer->protocol) 1176 goto out_permanent; 1177 last_perm = lh; 1178 } 1179 1180 /* Add the new entry after the last permanent entry if any, so that 1181 * the new entry does not override a permanent entry when matched with 1182 * a wild-card protocol. But it is allowed to override any existing 1183 * non-permanent entry. This means that when we remove this entry, the 1184 * system automatically returns to the old behavior. 1185 */ 1186 list_add_rcu(&p->list, last_perm); 1187 out: 1188 spin_unlock_bh(&inetsw_lock); 1189 1190 return; 1191 1192 out_permanent: 1193 pr_err("Attempt to override permanent protocol %d\n", protocol); 1194 goto out; 1195 1196 out_illegal: 1197 pr_err("Ignoring attempt to register invalid socket type %d\n", 1198 p->type); 1199 goto out; 1200 } 1201 EXPORT_SYMBOL(inet_register_protosw); 1202 1203 void inet_unregister_protosw(struct inet_protosw *p) 1204 { 1205 if (INET_PROTOSW_PERMANENT & p->flags) { 1206 pr_err("Attempt to unregister permanent protocol %d\n", 1207 p->protocol); 1208 } else { 1209 spin_lock_bh(&inetsw_lock); 1210 list_del_rcu(&p->list); 1211 spin_unlock_bh(&inetsw_lock); 1212 1213 synchronize_net(); 1214 } 1215 } 1216 EXPORT_SYMBOL(inet_unregister_protosw); 1217 1218 static int inet_sk_reselect_saddr(struct sock *sk) 1219 { 1220 struct inet_sock *inet = inet_sk(sk); 1221 __be32 old_saddr = inet->inet_saddr; 1222 __be32 daddr = inet->inet_daddr; 1223 struct flowi4 *fl4; 1224 struct rtable *rt; 1225 __be32 new_saddr; 1226 struct ip_options_rcu *inet_opt; 1227 1228 inet_opt = rcu_dereference_protected(inet->inet_opt, 1229 lockdep_sock_is_held(sk)); 1230 if (inet_opt && inet_opt->opt.srr) 1231 daddr = inet_opt->opt.faddr; 1232 1233 /* Query new route. */ 1234 fl4 = &inet->cork.fl.u.ip4; 1235 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk), 1236 sk->sk_bound_dev_if, sk->sk_protocol, 1237 inet->inet_sport, inet->inet_dport, sk); 1238 if (IS_ERR(rt)) 1239 return PTR_ERR(rt); 1240 1241 sk_setup_caps(sk, &rt->dst); 1242 1243 new_saddr = fl4->saddr; 1244 1245 if (new_saddr == old_saddr) 1246 return 0; 1247 1248 if (sock_net(sk)->ipv4.sysctl_ip_dynaddr > 1) { 1249 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n", 1250 __func__, &old_saddr, &new_saddr); 1251 } 1252 1253 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr; 1254 1255 /* 1256 * XXX The only one ugly spot where we need to 1257 * XXX really change the sockets identity after 1258 * XXX it has entered the hashes. -DaveM 1259 * 1260 * Besides that, it does not check for connection 1261 * uniqueness. Wait for troubles. 1262 */ 1263 return __sk_prot_rehash(sk); 1264 } 1265 1266 int inet_sk_rebuild_header(struct sock *sk) 1267 { 1268 struct inet_sock *inet = inet_sk(sk); 1269 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0); 1270 __be32 daddr; 1271 struct ip_options_rcu *inet_opt; 1272 struct flowi4 *fl4; 1273 int err; 1274 1275 /* Route is OK, nothing to do. */ 1276 if (rt) 1277 return 0; 1278 1279 /* Reroute. */ 1280 rcu_read_lock(); 1281 inet_opt = rcu_dereference(inet->inet_opt); 1282 daddr = inet->inet_daddr; 1283 if (inet_opt && inet_opt->opt.srr) 1284 daddr = inet_opt->opt.faddr; 1285 rcu_read_unlock(); 1286 fl4 = &inet->cork.fl.u.ip4; 1287 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr, 1288 inet->inet_dport, inet->inet_sport, 1289 sk->sk_protocol, RT_CONN_FLAGS(sk), 1290 sk->sk_bound_dev_if); 1291 if (!IS_ERR(rt)) { 1292 err = 0; 1293 sk_setup_caps(sk, &rt->dst); 1294 } else { 1295 err = PTR_ERR(rt); 1296 1297 /* Routing failed... */ 1298 sk->sk_route_caps = 0; 1299 /* 1300 * Other protocols have to map its equivalent state to TCP_SYN_SENT. 1301 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme 1302 */ 1303 if (!sock_net(sk)->ipv4.sysctl_ip_dynaddr || 1304 sk->sk_state != TCP_SYN_SENT || 1305 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) || 1306 (err = inet_sk_reselect_saddr(sk)) != 0) 1307 sk->sk_err_soft = -err; 1308 } 1309 1310 return err; 1311 } 1312 EXPORT_SYMBOL(inet_sk_rebuild_header); 1313 1314 void inet_sk_set_state(struct sock *sk, int state) 1315 { 1316 trace_inet_sock_set_state(sk, sk->sk_state, state); 1317 sk->sk_state = state; 1318 } 1319 EXPORT_SYMBOL(inet_sk_set_state); 1320 1321 void inet_sk_state_store(struct sock *sk, int newstate) 1322 { 1323 trace_inet_sock_set_state(sk, sk->sk_state, newstate); 1324 smp_store_release(&sk->sk_state, newstate); 1325 } 1326 1327 struct sk_buff *inet_gso_segment(struct sk_buff *skb, 1328 netdev_features_t features) 1329 { 1330 bool udpfrag = false, fixedid = false, gso_partial, encap; 1331 struct sk_buff *segs = ERR_PTR(-EINVAL); 1332 const struct net_offload *ops; 1333 unsigned int offset = 0; 1334 struct iphdr *iph; 1335 int proto, tot_len; 1336 int nhoff; 1337 int ihl; 1338 int id; 1339 1340 skb_reset_network_header(skb); 1341 nhoff = skb_network_header(skb) - skb_mac_header(skb); 1342 if (unlikely(!pskb_may_pull(skb, sizeof(*iph)))) 1343 goto out; 1344 1345 iph = ip_hdr(skb); 1346 ihl = iph->ihl * 4; 1347 if (ihl < sizeof(*iph)) 1348 goto out; 1349 1350 id = ntohs(iph->id); 1351 proto = iph->protocol; 1352 1353 /* Warning: after this point, iph might be no longer valid */ 1354 if (unlikely(!pskb_may_pull(skb, ihl))) 1355 goto out; 1356 __skb_pull(skb, ihl); 1357 1358 encap = SKB_GSO_CB(skb)->encap_level > 0; 1359 if (encap) 1360 features &= skb->dev->hw_enc_features; 1361 SKB_GSO_CB(skb)->encap_level += ihl; 1362 1363 skb_reset_transport_header(skb); 1364 1365 segs = ERR_PTR(-EPROTONOSUPPORT); 1366 1367 if (!skb->encapsulation || encap) { 1368 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP); 1369 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID); 1370 1371 /* fixed ID is invalid if DF bit is not set */ 1372 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF))) 1373 goto out; 1374 } 1375 1376 ops = rcu_dereference(inet_offloads[proto]); 1377 if (likely(ops && ops->callbacks.gso_segment)) 1378 segs = ops->callbacks.gso_segment(skb, features); 1379 1380 if (IS_ERR_OR_NULL(segs)) 1381 goto out; 1382 1383 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL); 1384 1385 skb = segs; 1386 do { 1387 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff); 1388 if (udpfrag) { 1389 iph->frag_off = htons(offset >> 3); 1390 if (skb->next) 1391 iph->frag_off |= htons(IP_MF); 1392 offset += skb->len - nhoff - ihl; 1393 tot_len = skb->len - nhoff; 1394 } else if (skb_is_gso(skb)) { 1395 if (!fixedid) { 1396 iph->id = htons(id); 1397 id += skb_shinfo(skb)->gso_segs; 1398 } 1399 1400 if (gso_partial) 1401 tot_len = skb_shinfo(skb)->gso_size + 1402 SKB_GSO_CB(skb)->data_offset + 1403 skb->head - (unsigned char *)iph; 1404 else 1405 tot_len = skb->len - nhoff; 1406 } else { 1407 if (!fixedid) 1408 iph->id = htons(id++); 1409 tot_len = skb->len - nhoff; 1410 } 1411 iph->tot_len = htons(tot_len); 1412 ip_send_check(iph); 1413 if (encap) 1414 skb_reset_inner_headers(skb); 1415 skb->network_header = (u8 *)iph - skb->head; 1416 skb_reset_mac_len(skb); 1417 } while ((skb = skb->next)); 1418 1419 out: 1420 return segs; 1421 } 1422 1423 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb, 1424 netdev_features_t features) 1425 { 1426 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4)) 1427 return ERR_PTR(-EINVAL); 1428 1429 return inet_gso_segment(skb, features); 1430 } 1431 1432 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb) 1433 { 1434 const struct net_offload *ops; 1435 struct sk_buff *pp = NULL; 1436 const struct iphdr *iph; 1437 struct sk_buff *p; 1438 unsigned int hlen; 1439 unsigned int off; 1440 unsigned int id; 1441 int flush = 1; 1442 int proto; 1443 1444 off = skb_gro_offset(skb); 1445 hlen = off + sizeof(*iph); 1446 iph = skb_gro_header_fast(skb, off); 1447 if (skb_gro_header_hard(skb, hlen)) { 1448 iph = skb_gro_header_slow(skb, hlen, off); 1449 if (unlikely(!iph)) 1450 goto out; 1451 } 1452 1453 proto = iph->protocol; 1454 1455 ops = rcu_dereference(inet_offloads[proto]); 1456 if (!ops || !ops->callbacks.gro_receive) 1457 goto out; 1458 1459 if (*(u8 *)iph != 0x45) 1460 goto out; 1461 1462 if (ip_is_fragment(iph)) 1463 goto out; 1464 1465 if (unlikely(ip_fast_csum((u8 *)iph, 5))) 1466 goto out; 1467 1468 id = ntohl(*(__be32 *)&iph->id); 1469 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF)); 1470 id >>= 16; 1471 1472 list_for_each_entry(p, head, list) { 1473 struct iphdr *iph2; 1474 u16 flush_id; 1475 1476 if (!NAPI_GRO_CB(p)->same_flow) 1477 continue; 1478 1479 iph2 = (struct iphdr *)(p->data + off); 1480 /* The above works because, with the exception of the top 1481 * (inner most) layer, we only aggregate pkts with the same 1482 * hdr length so all the hdrs we'll need to verify will start 1483 * at the same offset. 1484 */ 1485 if ((iph->protocol ^ iph2->protocol) | 1486 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) | 1487 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) { 1488 NAPI_GRO_CB(p)->same_flow = 0; 1489 continue; 1490 } 1491 1492 /* All fields must match except length and checksum. */ 1493 NAPI_GRO_CB(p)->flush |= 1494 (iph->ttl ^ iph2->ttl) | 1495 (iph->tos ^ iph2->tos) | 1496 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF)); 1497 1498 NAPI_GRO_CB(p)->flush |= flush; 1499 1500 /* We need to store of the IP ID check to be included later 1501 * when we can verify that this packet does in fact belong 1502 * to a given flow. 1503 */ 1504 flush_id = (u16)(id - ntohs(iph2->id)); 1505 1506 /* This bit of code makes it much easier for us to identify 1507 * the cases where we are doing atomic vs non-atomic IP ID 1508 * checks. Specifically an atomic check can return IP ID 1509 * values 0 - 0xFFFF, while a non-atomic check can only 1510 * return 0 or 0xFFFF. 1511 */ 1512 if (!NAPI_GRO_CB(p)->is_atomic || 1513 !(iph->frag_off & htons(IP_DF))) { 1514 flush_id ^= NAPI_GRO_CB(p)->count; 1515 flush_id = flush_id ? 0xFFFF : 0; 1516 } 1517 1518 /* If the previous IP ID value was based on an atomic 1519 * datagram we can overwrite the value and ignore it. 1520 */ 1521 if (NAPI_GRO_CB(skb)->is_atomic) 1522 NAPI_GRO_CB(p)->flush_id = flush_id; 1523 else 1524 NAPI_GRO_CB(p)->flush_id |= flush_id; 1525 } 1526 1527 NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF)); 1528 NAPI_GRO_CB(skb)->flush |= flush; 1529 skb_set_network_header(skb, off); 1530 /* The above will be needed by the transport layer if there is one 1531 * immediately following this IP hdr. 1532 */ 1533 1534 /* Note : No need to call skb_gro_postpull_rcsum() here, 1535 * as we already checked checksum over ipv4 header was 0 1536 */ 1537 skb_gro_pull(skb, sizeof(*iph)); 1538 skb_set_transport_header(skb, skb_gro_offset(skb)); 1539 1540 pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive, 1541 ops->callbacks.gro_receive, head, skb); 1542 1543 out: 1544 skb_gro_flush_final(skb, pp, flush); 1545 1546 return pp; 1547 } 1548 1549 static struct sk_buff *ipip_gro_receive(struct list_head *head, 1550 struct sk_buff *skb) 1551 { 1552 if (NAPI_GRO_CB(skb)->encap_mark) { 1553 NAPI_GRO_CB(skb)->flush = 1; 1554 return NULL; 1555 } 1556 1557 NAPI_GRO_CB(skb)->encap_mark = 1; 1558 1559 return inet_gro_receive(head, skb); 1560 } 1561 1562 #define SECONDS_PER_DAY 86400 1563 1564 /* inet_current_timestamp - Return IP network timestamp 1565 * 1566 * Return milliseconds since midnight in network byte order. 1567 */ 1568 __be32 inet_current_timestamp(void) 1569 { 1570 u32 secs; 1571 u32 msecs; 1572 struct timespec64 ts; 1573 1574 ktime_get_real_ts64(&ts); 1575 1576 /* Get secs since midnight. */ 1577 (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs); 1578 /* Convert to msecs. */ 1579 msecs = secs * MSEC_PER_SEC; 1580 /* Convert nsec to msec. */ 1581 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC; 1582 1583 /* Convert to network byte order. */ 1584 return htonl(msecs); 1585 } 1586 EXPORT_SYMBOL(inet_current_timestamp); 1587 1588 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len) 1589 { 1590 if (sk->sk_family == AF_INET) 1591 return ip_recv_error(sk, msg, len, addr_len); 1592 #if IS_ENABLED(CONFIG_IPV6) 1593 if (sk->sk_family == AF_INET6) 1594 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len); 1595 #endif 1596 return -EINVAL; 1597 } 1598 1599 int inet_gro_complete(struct sk_buff *skb, int nhoff) 1600 { 1601 __be16 newlen = htons(skb->len - nhoff); 1602 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff); 1603 const struct net_offload *ops; 1604 int proto = iph->protocol; 1605 int err = -ENOSYS; 1606 1607 if (skb->encapsulation) { 1608 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP)); 1609 skb_set_inner_network_header(skb, nhoff); 1610 } 1611 1612 csum_replace2(&iph->check, iph->tot_len, newlen); 1613 iph->tot_len = newlen; 1614 1615 ops = rcu_dereference(inet_offloads[proto]); 1616 if (WARN_ON(!ops || !ops->callbacks.gro_complete)) 1617 goto out; 1618 1619 /* Only need to add sizeof(*iph) to get to the next hdr below 1620 * because any hdr with option will have been flushed in 1621 * inet_gro_receive(). 1622 */ 1623 err = INDIRECT_CALL_2(ops->callbacks.gro_complete, 1624 tcp4_gro_complete, udp4_gro_complete, 1625 skb, nhoff + sizeof(*iph)); 1626 1627 out: 1628 return err; 1629 } 1630 1631 static int ipip_gro_complete(struct sk_buff *skb, int nhoff) 1632 { 1633 skb->encapsulation = 1; 1634 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4; 1635 return inet_gro_complete(skb, nhoff); 1636 } 1637 1638 int inet_ctl_sock_create(struct sock **sk, unsigned short family, 1639 unsigned short type, unsigned char protocol, 1640 struct net *net) 1641 { 1642 struct socket *sock; 1643 int rc = sock_create_kern(net, family, type, protocol, &sock); 1644 1645 if (rc == 0) { 1646 *sk = sock->sk; 1647 (*sk)->sk_allocation = GFP_ATOMIC; 1648 /* 1649 * Unhash it so that IP input processing does not even see it, 1650 * we do not wish this socket to see incoming packets. 1651 */ 1652 (*sk)->sk_prot->unhash(*sk); 1653 } 1654 return rc; 1655 } 1656 EXPORT_SYMBOL_GPL(inet_ctl_sock_create); 1657 1658 unsigned long snmp_fold_field(void __percpu *mib, int offt) 1659 { 1660 unsigned long res = 0; 1661 int i; 1662 1663 for_each_possible_cpu(i) 1664 res += snmp_get_cpu_field(mib, i, offt); 1665 return res; 1666 } 1667 EXPORT_SYMBOL_GPL(snmp_fold_field); 1668 1669 #if BITS_PER_LONG==32 1670 1671 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt, 1672 size_t syncp_offset) 1673 { 1674 void *bhptr; 1675 struct u64_stats_sync *syncp; 1676 u64 v; 1677 unsigned int start; 1678 1679 bhptr = per_cpu_ptr(mib, cpu); 1680 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset); 1681 do { 1682 start = u64_stats_fetch_begin_irq(syncp); 1683 v = *(((u64 *)bhptr) + offt); 1684 } while (u64_stats_fetch_retry_irq(syncp, start)); 1685 1686 return v; 1687 } 1688 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64); 1689 1690 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset) 1691 { 1692 u64 res = 0; 1693 int cpu; 1694 1695 for_each_possible_cpu(cpu) { 1696 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset); 1697 } 1698 return res; 1699 } 1700 EXPORT_SYMBOL_GPL(snmp_fold_field64); 1701 #endif 1702 1703 #ifdef CONFIG_IP_MULTICAST 1704 static const struct net_protocol igmp_protocol = { 1705 .handler = igmp_rcv, 1706 }; 1707 #endif 1708 1709 /* thinking of making this const? Don't. 1710 * early_demux can change based on sysctl. 1711 */ 1712 static struct net_protocol tcp_protocol = { 1713 .early_demux = tcp_v4_early_demux, 1714 .early_demux_handler = tcp_v4_early_demux, 1715 .handler = tcp_v4_rcv, 1716 .err_handler = tcp_v4_err, 1717 .no_policy = 1, 1718 .icmp_strict_tag_validation = 1, 1719 }; 1720 1721 /* thinking of making this const? Don't. 1722 * early_demux can change based on sysctl. 1723 */ 1724 static struct net_protocol udp_protocol = { 1725 .early_demux = udp_v4_early_demux, 1726 .early_demux_handler = udp_v4_early_demux, 1727 .handler = udp_rcv, 1728 .err_handler = udp_err, 1729 .no_policy = 1, 1730 }; 1731 1732 static const struct net_protocol icmp_protocol = { 1733 .handler = icmp_rcv, 1734 .err_handler = icmp_err, 1735 .no_policy = 1, 1736 }; 1737 1738 static __net_init int ipv4_mib_init_net(struct net *net) 1739 { 1740 int i; 1741 1742 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib); 1743 if (!net->mib.tcp_statistics) 1744 goto err_tcp_mib; 1745 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib); 1746 if (!net->mib.ip_statistics) 1747 goto err_ip_mib; 1748 1749 for_each_possible_cpu(i) { 1750 struct ipstats_mib *af_inet_stats; 1751 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i); 1752 u64_stats_init(&af_inet_stats->syncp); 1753 } 1754 1755 net->mib.net_statistics = alloc_percpu(struct linux_mib); 1756 if (!net->mib.net_statistics) 1757 goto err_net_mib; 1758 net->mib.udp_statistics = alloc_percpu(struct udp_mib); 1759 if (!net->mib.udp_statistics) 1760 goto err_udp_mib; 1761 net->mib.udplite_statistics = alloc_percpu(struct udp_mib); 1762 if (!net->mib.udplite_statistics) 1763 goto err_udplite_mib; 1764 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib); 1765 if (!net->mib.icmp_statistics) 1766 goto err_icmp_mib; 1767 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib), 1768 GFP_KERNEL); 1769 if (!net->mib.icmpmsg_statistics) 1770 goto err_icmpmsg_mib; 1771 1772 tcp_mib_init(net); 1773 return 0; 1774 1775 err_icmpmsg_mib: 1776 free_percpu(net->mib.icmp_statistics); 1777 err_icmp_mib: 1778 free_percpu(net->mib.udplite_statistics); 1779 err_udplite_mib: 1780 free_percpu(net->mib.udp_statistics); 1781 err_udp_mib: 1782 free_percpu(net->mib.net_statistics); 1783 err_net_mib: 1784 free_percpu(net->mib.ip_statistics); 1785 err_ip_mib: 1786 free_percpu(net->mib.tcp_statistics); 1787 err_tcp_mib: 1788 return -ENOMEM; 1789 } 1790 1791 static __net_exit void ipv4_mib_exit_net(struct net *net) 1792 { 1793 kfree(net->mib.icmpmsg_statistics); 1794 free_percpu(net->mib.icmp_statistics); 1795 free_percpu(net->mib.udplite_statistics); 1796 free_percpu(net->mib.udp_statistics); 1797 free_percpu(net->mib.net_statistics); 1798 free_percpu(net->mib.ip_statistics); 1799 free_percpu(net->mib.tcp_statistics); 1800 #ifdef CONFIG_MPTCP 1801 /* allocated on demand, see mptcp_init_sock() */ 1802 free_percpu(net->mib.mptcp_statistics); 1803 #endif 1804 } 1805 1806 static __net_initdata struct pernet_operations ipv4_mib_ops = { 1807 .init = ipv4_mib_init_net, 1808 .exit = ipv4_mib_exit_net, 1809 }; 1810 1811 static int __init init_ipv4_mibs(void) 1812 { 1813 return register_pernet_subsys(&ipv4_mib_ops); 1814 } 1815 1816 static __net_init int inet_init_net(struct net *net) 1817 { 1818 /* 1819 * Set defaults for local port range 1820 */ 1821 seqlock_init(&net->ipv4.ip_local_ports.lock); 1822 net->ipv4.ip_local_ports.range[0] = 32768; 1823 net->ipv4.ip_local_ports.range[1] = 60999; 1824 1825 seqlock_init(&net->ipv4.ping_group_range.lock); 1826 /* 1827 * Sane defaults - nobody may create ping sockets. 1828 * Boot scripts should set this to distro-specific group. 1829 */ 1830 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1); 1831 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0); 1832 1833 /* Default values for sysctl-controlled parameters. 1834 * We set them here, in case sysctl is not compiled. 1835 */ 1836 net->ipv4.sysctl_ip_default_ttl = IPDEFTTL; 1837 net->ipv4.sysctl_ip_fwd_update_priority = 1; 1838 net->ipv4.sysctl_ip_dynaddr = 0; 1839 net->ipv4.sysctl_ip_early_demux = 1; 1840 net->ipv4.sysctl_udp_early_demux = 1; 1841 net->ipv4.sysctl_tcp_early_demux = 1; 1842 net->ipv4.sysctl_nexthop_compat_mode = 1; 1843 #ifdef CONFIG_SYSCTL 1844 net->ipv4.sysctl_ip_prot_sock = PROT_SOCK; 1845 #endif 1846 1847 /* Some igmp sysctl, whose values are always used */ 1848 net->ipv4.sysctl_igmp_max_memberships = 20; 1849 net->ipv4.sysctl_igmp_max_msf = 10; 1850 /* IGMP reports for link-local multicast groups are enabled by default */ 1851 net->ipv4.sysctl_igmp_llm_reports = 1; 1852 net->ipv4.sysctl_igmp_qrv = 2; 1853 1854 net->ipv4.sysctl_fib_notify_on_flag_change = 0; 1855 1856 return 0; 1857 } 1858 1859 static __net_initdata struct pernet_operations af_inet_ops = { 1860 .init = inet_init_net, 1861 }; 1862 1863 static int __init init_inet_pernet_ops(void) 1864 { 1865 return register_pernet_subsys(&af_inet_ops); 1866 } 1867 1868 static int ipv4_proc_init(void); 1869 1870 /* 1871 * IP protocol layer initialiser 1872 */ 1873 1874 static struct packet_offload ip_packet_offload __read_mostly = { 1875 .type = cpu_to_be16(ETH_P_IP), 1876 .callbacks = { 1877 .gso_segment = inet_gso_segment, 1878 .gro_receive = inet_gro_receive, 1879 .gro_complete = inet_gro_complete, 1880 }, 1881 }; 1882 1883 static const struct net_offload ipip_offload = { 1884 .callbacks = { 1885 .gso_segment = ipip_gso_segment, 1886 .gro_receive = ipip_gro_receive, 1887 .gro_complete = ipip_gro_complete, 1888 }, 1889 }; 1890 1891 static int __init ipip_offload_init(void) 1892 { 1893 return inet_add_offload(&ipip_offload, IPPROTO_IPIP); 1894 } 1895 1896 static int __init ipv4_offload_init(void) 1897 { 1898 /* 1899 * Add offloads 1900 */ 1901 if (udpv4_offload_init() < 0) 1902 pr_crit("%s: Cannot add UDP protocol offload\n", __func__); 1903 if (tcpv4_offload_init() < 0) 1904 pr_crit("%s: Cannot add TCP protocol offload\n", __func__); 1905 if (ipip_offload_init() < 0) 1906 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__); 1907 1908 dev_add_offload(&ip_packet_offload); 1909 return 0; 1910 } 1911 1912 fs_initcall(ipv4_offload_init); 1913 1914 static struct packet_type ip_packet_type __read_mostly = { 1915 .type = cpu_to_be16(ETH_P_IP), 1916 .func = ip_rcv, 1917 .list_func = ip_list_rcv, 1918 }; 1919 1920 static int __init inet_init(void) 1921 { 1922 struct inet_protosw *q; 1923 struct list_head *r; 1924 int rc; 1925 1926 sock_skb_cb_check_size(sizeof(struct inet_skb_parm)); 1927 1928 rc = proto_register(&tcp_prot, 1); 1929 if (rc) 1930 goto out; 1931 1932 rc = proto_register(&udp_prot, 1); 1933 if (rc) 1934 goto out_unregister_tcp_proto; 1935 1936 rc = proto_register(&raw_prot, 1); 1937 if (rc) 1938 goto out_unregister_udp_proto; 1939 1940 rc = proto_register(&ping_prot, 1); 1941 if (rc) 1942 goto out_unregister_raw_proto; 1943 1944 /* 1945 * Tell SOCKET that we are alive... 1946 */ 1947 1948 (void)sock_register(&inet_family_ops); 1949 1950 #ifdef CONFIG_SYSCTL 1951 ip_static_sysctl_init(); 1952 #endif 1953 1954 /* 1955 * Add all the base protocols. 1956 */ 1957 1958 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0) 1959 pr_crit("%s: Cannot add ICMP protocol\n", __func__); 1960 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0) 1961 pr_crit("%s: Cannot add UDP protocol\n", __func__); 1962 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0) 1963 pr_crit("%s: Cannot add TCP protocol\n", __func__); 1964 #ifdef CONFIG_IP_MULTICAST 1965 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0) 1966 pr_crit("%s: Cannot add IGMP protocol\n", __func__); 1967 #endif 1968 1969 /* Register the socket-side information for inet_create. */ 1970 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r) 1971 INIT_LIST_HEAD(r); 1972 1973 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q) 1974 inet_register_protosw(q); 1975 1976 /* 1977 * Set the ARP module up 1978 */ 1979 1980 arp_init(); 1981 1982 /* 1983 * Set the IP module up 1984 */ 1985 1986 ip_init(); 1987 1988 /* Setup TCP slab cache for open requests. */ 1989 tcp_init(); 1990 1991 /* Setup UDP memory threshold */ 1992 udp_init(); 1993 1994 /* Add UDP-Lite (RFC 3828) */ 1995 udplite4_register(); 1996 1997 raw_init(); 1998 1999 ping_init(); 2000 2001 /* 2002 * Set the ICMP layer up 2003 */ 2004 2005 if (icmp_init() < 0) 2006 panic("Failed to create the ICMP control socket.\n"); 2007 2008 /* 2009 * Initialise the multicast router 2010 */ 2011 #if defined(CONFIG_IP_MROUTE) 2012 if (ip_mr_init()) 2013 pr_crit("%s: Cannot init ipv4 mroute\n", __func__); 2014 #endif 2015 2016 if (init_inet_pernet_ops()) 2017 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__); 2018 /* 2019 * Initialise per-cpu ipv4 mibs 2020 */ 2021 2022 if (init_ipv4_mibs()) 2023 pr_crit("%s: Cannot init ipv4 mibs\n", __func__); 2024 2025 ipv4_proc_init(); 2026 2027 ipfrag_init(); 2028 2029 dev_add_pack(&ip_packet_type); 2030 2031 ip_tunnel_core_init(); 2032 2033 rc = 0; 2034 out: 2035 return rc; 2036 out_unregister_raw_proto: 2037 proto_unregister(&raw_prot); 2038 out_unregister_udp_proto: 2039 proto_unregister(&udp_prot); 2040 out_unregister_tcp_proto: 2041 proto_unregister(&tcp_prot); 2042 goto out; 2043 } 2044 2045 fs_initcall(inet_init); 2046 2047 /* ------------------------------------------------------------------------ */ 2048 2049 #ifdef CONFIG_PROC_FS 2050 static int __init ipv4_proc_init(void) 2051 { 2052 int rc = 0; 2053 2054 if (raw_proc_init()) 2055 goto out_raw; 2056 if (tcp4_proc_init()) 2057 goto out_tcp; 2058 if (udp4_proc_init()) 2059 goto out_udp; 2060 if (ping_proc_init()) 2061 goto out_ping; 2062 if (ip_misc_proc_init()) 2063 goto out_misc; 2064 out: 2065 return rc; 2066 out_misc: 2067 ping_proc_exit(); 2068 out_ping: 2069 udp4_proc_exit(); 2070 out_udp: 2071 tcp4_proc_exit(); 2072 out_tcp: 2073 raw_proc_exit(); 2074 out_raw: 2075 rc = -ENOMEM; 2076 goto out; 2077 } 2078 2079 #else /* CONFIG_PROC_FS */ 2080 static int __init ipv4_proc_init(void) 2081 { 2082 return 0; 2083 } 2084 #endif /* CONFIG_PROC_FS */ 2085