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/tcp.h> 103 #include <net/udp.h> 104 #include <net/udplite.h> 105 #include <net/ping.h> 106 #include <linux/skbuff.h> 107 #include <net/sock.h> 108 #include <net/raw.h> 109 #include <net/icmp.h> 110 #include <net/inet_common.h> 111 #include <net/ip_tunnels.h> 112 #include <net/xfrm.h> 113 #include <net/net_namespace.h> 114 #include <net/secure_seq.h> 115 #ifdef CONFIG_IP_MROUTE 116 #include <linux/mroute.h> 117 #endif 118 #include <net/l3mdev.h> 119 120 #include <trace/events/sock.h> 121 122 /* The inetsw table contains everything that inet_create needs to 123 * build a new socket. 124 */ 125 static struct list_head inetsw[SOCK_MAX]; 126 static DEFINE_SPINLOCK(inetsw_lock); 127 128 /* New destruction routine */ 129 130 void inet_sock_destruct(struct sock *sk) 131 { 132 struct inet_sock *inet = inet_sk(sk); 133 134 __skb_queue_purge(&sk->sk_receive_queue); 135 if (sk->sk_rx_skb_cache) { 136 __kfree_skb(sk->sk_rx_skb_cache); 137 sk->sk_rx_skb_cache = NULL; 138 } 139 __skb_queue_purge(&sk->sk_error_queue); 140 141 sk_mem_reclaim(sk); 142 143 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) { 144 pr_err("Attempt to release TCP socket in state %d %p\n", 145 sk->sk_state, sk); 146 return; 147 } 148 if (!sock_flag(sk, SOCK_DEAD)) { 149 pr_err("Attempt to release alive inet socket %p\n", sk); 150 return; 151 } 152 153 WARN_ON(atomic_read(&sk->sk_rmem_alloc)); 154 WARN_ON(refcount_read(&sk->sk_wmem_alloc)); 155 WARN_ON(sk->sk_wmem_queued); 156 WARN_ON(sk->sk_forward_alloc); 157 158 kfree(rcu_dereference_protected(inet->inet_opt, 1)); 159 dst_release(rcu_dereference_protected(sk->sk_dst_cache, 1)); 160 dst_release(sk->sk_rx_dst); 161 sk_refcnt_debug_dec(sk); 162 } 163 EXPORT_SYMBOL(inet_sock_destruct); 164 165 /* 166 * The routines beyond this point handle the behaviour of an AF_INET 167 * socket object. Mostly it punts to the subprotocols of IP to do 168 * the work. 169 */ 170 171 /* 172 * Automatically bind an unbound socket. 173 */ 174 175 static int inet_autobind(struct sock *sk) 176 { 177 struct inet_sock *inet; 178 /* We may need to bind the socket. */ 179 lock_sock(sk); 180 inet = inet_sk(sk); 181 if (!inet->inet_num) { 182 if (sk->sk_prot->get_port(sk, 0)) { 183 release_sock(sk); 184 return -EAGAIN; 185 } 186 inet->inet_sport = htons(inet->inet_num); 187 } 188 release_sock(sk); 189 return 0; 190 } 191 192 /* 193 * Move a socket into listening state. 194 */ 195 int inet_listen(struct socket *sock, int backlog) 196 { 197 struct sock *sk = sock->sk; 198 unsigned char old_state; 199 int err, tcp_fastopen; 200 201 lock_sock(sk); 202 203 err = -EINVAL; 204 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM) 205 goto out; 206 207 old_state = sk->sk_state; 208 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN))) 209 goto out; 210 211 WRITE_ONCE(sk->sk_max_ack_backlog, backlog); 212 /* Really, if the socket is already in listen state 213 * we can only allow the backlog to be adjusted. 214 */ 215 if (old_state != TCP_LISTEN) { 216 /* Enable TFO w/o requiring TCP_FASTOPEN socket option. 217 * Note that only TCP sockets (SOCK_STREAM) will reach here. 218 * Also fastopen backlog may already been set via the option 219 * because the socket was in TCP_LISTEN state previously but 220 * was shutdown() rather than close(). 221 */ 222 tcp_fastopen = sock_net(sk)->ipv4.sysctl_tcp_fastopen; 223 if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) && 224 (tcp_fastopen & TFO_SERVER_ENABLE) && 225 !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) { 226 fastopen_queue_tune(sk, backlog); 227 tcp_fastopen_init_key_once(sock_net(sk)); 228 } 229 230 err = inet_csk_listen_start(sk, backlog); 231 if (err) 232 goto out; 233 tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL); 234 } 235 err = 0; 236 237 out: 238 release_sock(sk); 239 return err; 240 } 241 EXPORT_SYMBOL(inet_listen); 242 243 /* 244 * Create an inet socket. 245 */ 246 247 static int inet_create(struct net *net, struct socket *sock, int protocol, 248 int kern) 249 { 250 struct sock *sk; 251 struct inet_protosw *answer; 252 struct inet_sock *inet; 253 struct proto *answer_prot; 254 unsigned char answer_flags; 255 int try_loading_module = 0; 256 int err; 257 258 if (protocol < 0 || protocol >= IPPROTO_MAX) 259 return -EINVAL; 260 261 sock->state = SS_UNCONNECTED; 262 263 /* Look for the requested type/protocol pair. */ 264 lookup_protocol: 265 err = -ESOCKTNOSUPPORT; 266 rcu_read_lock(); 267 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) { 268 269 err = 0; 270 /* Check the non-wild match. */ 271 if (protocol == answer->protocol) { 272 if (protocol != IPPROTO_IP) 273 break; 274 } else { 275 /* Check for the two wild cases. */ 276 if (IPPROTO_IP == protocol) { 277 protocol = answer->protocol; 278 break; 279 } 280 if (IPPROTO_IP == answer->protocol) 281 break; 282 } 283 err = -EPROTONOSUPPORT; 284 } 285 286 if (unlikely(err)) { 287 if (try_loading_module < 2) { 288 rcu_read_unlock(); 289 /* 290 * Be more specific, e.g. net-pf-2-proto-132-type-1 291 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM) 292 */ 293 if (++try_loading_module == 1) 294 request_module("net-pf-%d-proto-%d-type-%d", 295 PF_INET, protocol, sock->type); 296 /* 297 * Fall back to generic, e.g. net-pf-2-proto-132 298 * (net-pf-PF_INET-proto-IPPROTO_SCTP) 299 */ 300 else 301 request_module("net-pf-%d-proto-%d", 302 PF_INET, protocol); 303 goto lookup_protocol; 304 } else 305 goto out_rcu_unlock; 306 } 307 308 err = -EPERM; 309 if (sock->type == SOCK_RAW && !kern && 310 !ns_capable(net->user_ns, CAP_NET_RAW)) 311 goto out_rcu_unlock; 312 313 sock->ops = answer->ops; 314 answer_prot = answer->prot; 315 answer_flags = answer->flags; 316 rcu_read_unlock(); 317 318 WARN_ON(!answer_prot->slab); 319 320 err = -ENOBUFS; 321 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern); 322 if (!sk) 323 goto out; 324 325 err = 0; 326 if (INET_PROTOSW_REUSE & answer_flags) 327 sk->sk_reuse = SK_CAN_REUSE; 328 329 inet = inet_sk(sk); 330 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0; 331 332 inet->nodefrag = 0; 333 334 if (SOCK_RAW == sock->type) { 335 inet->inet_num = protocol; 336 if (IPPROTO_RAW == protocol) 337 inet->hdrincl = 1; 338 } 339 340 if (net->ipv4.sysctl_ip_no_pmtu_disc) 341 inet->pmtudisc = IP_PMTUDISC_DONT; 342 else 343 inet->pmtudisc = IP_PMTUDISC_WANT; 344 345 inet->inet_id = 0; 346 347 sock_init_data(sock, sk); 348 349 sk->sk_destruct = inet_sock_destruct; 350 sk->sk_protocol = protocol; 351 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv; 352 353 inet->uc_ttl = -1; 354 inet->mc_loop = 1; 355 inet->mc_ttl = 1; 356 inet->mc_all = 1; 357 inet->mc_index = 0; 358 inet->mc_list = NULL; 359 inet->rcv_tos = 0; 360 361 sk_refcnt_debug_inc(sk); 362 363 if (inet->inet_num) { 364 /* It assumes that any protocol which allows 365 * the user to assign a number at socket 366 * creation time automatically 367 * shares. 368 */ 369 inet->inet_sport = htons(inet->inet_num); 370 /* Add to protocol hash chains. */ 371 err = sk->sk_prot->hash(sk); 372 if (err) { 373 sk_common_release(sk); 374 goto out; 375 } 376 } 377 378 if (sk->sk_prot->init) { 379 err = sk->sk_prot->init(sk); 380 if (err) { 381 sk_common_release(sk); 382 goto out; 383 } 384 } 385 386 if (!kern) { 387 err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk); 388 if (err) { 389 sk_common_release(sk); 390 goto out; 391 } 392 } 393 out: 394 return err; 395 out_rcu_unlock: 396 rcu_read_unlock(); 397 goto out; 398 } 399 400 401 /* 402 * The peer socket should always be NULL (or else). When we call this 403 * function we are destroying the object and from then on nobody 404 * should refer to it. 405 */ 406 int inet_release(struct socket *sock) 407 { 408 struct sock *sk = sock->sk; 409 410 if (sk) { 411 long timeout; 412 413 /* Applications forget to leave groups before exiting */ 414 ip_mc_drop_socket(sk); 415 416 /* If linger is set, we don't return until the close 417 * is complete. Otherwise we return immediately. The 418 * actually closing is done the same either way. 419 * 420 * If the close is due to the process exiting, we never 421 * linger.. 422 */ 423 timeout = 0; 424 if (sock_flag(sk, SOCK_LINGER) && 425 !(current->flags & PF_EXITING)) 426 timeout = sk->sk_lingertime; 427 sk->sk_prot->close(sk, timeout); 428 sock->sk = NULL; 429 } 430 return 0; 431 } 432 EXPORT_SYMBOL(inet_release); 433 434 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) 435 { 436 struct sock *sk = sock->sk; 437 int err; 438 439 /* If the socket has its own bind function then use it. (RAW) */ 440 if (sk->sk_prot->bind) { 441 return sk->sk_prot->bind(sk, uaddr, addr_len); 442 } 443 if (addr_len < sizeof(struct sockaddr_in)) 444 return -EINVAL; 445 446 /* BPF prog is run before any checks are done so that if the prog 447 * changes context in a wrong way it will be caught. 448 */ 449 err = BPF_CGROUP_RUN_PROG_INET4_BIND(sk, uaddr); 450 if (err) 451 return err; 452 453 return __inet_bind(sk, uaddr, addr_len, BIND_WITH_LOCK); 454 } 455 EXPORT_SYMBOL(inet_bind); 456 457 int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len, 458 u32 flags) 459 { 460 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr; 461 struct inet_sock *inet = inet_sk(sk); 462 struct net *net = sock_net(sk); 463 unsigned short snum; 464 int chk_addr_ret; 465 u32 tb_id = RT_TABLE_LOCAL; 466 int err; 467 468 if (addr->sin_family != AF_INET) { 469 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET) 470 * only if s_addr is INADDR_ANY. 471 */ 472 err = -EAFNOSUPPORT; 473 if (addr->sin_family != AF_UNSPEC || 474 addr->sin_addr.s_addr != htonl(INADDR_ANY)) 475 goto out; 476 } 477 478 tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id; 479 chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id); 480 481 /* Not specified by any standard per-se, however it breaks too 482 * many applications when removed. It is unfortunate since 483 * allowing applications to make a non-local bind solves 484 * several problems with systems using dynamic addressing. 485 * (ie. your servers still start up even if your ISDN link 486 * is temporarily down) 487 */ 488 err = -EADDRNOTAVAIL; 489 if (!inet_can_nonlocal_bind(net, inet) && 490 addr->sin_addr.s_addr != htonl(INADDR_ANY) && 491 chk_addr_ret != RTN_LOCAL && 492 chk_addr_ret != RTN_MULTICAST && 493 chk_addr_ret != RTN_BROADCAST) 494 goto out; 495 496 snum = ntohs(addr->sin_port); 497 err = -EACCES; 498 if (snum && inet_port_requires_bind_service(net, snum) && 499 !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE)) 500 goto out; 501 502 /* We keep a pair of addresses. rcv_saddr is the one 503 * used by hash lookups, and saddr is used for transmit. 504 * 505 * In the BSD API these are the same except where it 506 * would be illegal to use them (multicast/broadcast) in 507 * which case the sending device address is used. 508 */ 509 if (flags & BIND_WITH_LOCK) 510 lock_sock(sk); 511 512 /* Check these errors (active socket, double bind). */ 513 err = -EINVAL; 514 if (sk->sk_state != TCP_CLOSE || inet->inet_num) 515 goto out_release_sock; 516 517 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr; 518 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST) 519 inet->inet_saddr = 0; /* Use device */ 520 521 /* Make sure we are allowed to bind here. */ 522 if (snum || !(inet->bind_address_no_port || 523 (flags & BIND_FORCE_ADDRESS_NO_PORT))) { 524 if (sk->sk_prot->get_port(sk, snum)) { 525 inet->inet_saddr = inet->inet_rcv_saddr = 0; 526 err = -EADDRINUSE; 527 goto out_release_sock; 528 } 529 if (!(flags & BIND_FROM_BPF)) { 530 err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk); 531 if (err) { 532 inet->inet_saddr = inet->inet_rcv_saddr = 0; 533 goto out_release_sock; 534 } 535 } 536 } 537 538 if (inet->inet_rcv_saddr) 539 sk->sk_userlocks |= SOCK_BINDADDR_LOCK; 540 if (snum) 541 sk->sk_userlocks |= SOCK_BINDPORT_LOCK; 542 inet->inet_sport = htons(inet->inet_num); 543 inet->inet_daddr = 0; 544 inet->inet_dport = 0; 545 sk_dst_reset(sk); 546 err = 0; 547 out_release_sock: 548 if (flags & BIND_WITH_LOCK) 549 release_sock(sk); 550 out: 551 return err; 552 } 553 554 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr, 555 int addr_len, int flags) 556 { 557 struct sock *sk = sock->sk; 558 int err; 559 560 if (addr_len < sizeof(uaddr->sa_family)) 561 return -EINVAL; 562 if (uaddr->sa_family == AF_UNSPEC) 563 return sk->sk_prot->disconnect(sk, flags); 564 565 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) { 566 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len); 567 if (err) 568 return err; 569 } 570 571 if (!inet_sk(sk)->inet_num && inet_autobind(sk)) 572 return -EAGAIN; 573 return sk->sk_prot->connect(sk, uaddr, addr_len); 574 } 575 EXPORT_SYMBOL(inet_dgram_connect); 576 577 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias) 578 { 579 DEFINE_WAIT_FUNC(wait, woken_wake_function); 580 581 add_wait_queue(sk_sleep(sk), &wait); 582 sk->sk_write_pending += writebias; 583 584 /* Basic assumption: if someone sets sk->sk_err, he _must_ 585 * change state of the socket from TCP_SYN_*. 586 * Connect() does not allow to get error notifications 587 * without closing the socket. 588 */ 589 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 590 release_sock(sk); 591 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo); 592 lock_sock(sk); 593 if (signal_pending(current) || !timeo) 594 break; 595 } 596 remove_wait_queue(sk_sleep(sk), &wait); 597 sk->sk_write_pending -= writebias; 598 return timeo; 599 } 600 601 /* 602 * Connect to a remote host. There is regrettably still a little 603 * TCP 'magic' in here. 604 */ 605 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr, 606 int addr_len, int flags, int is_sendmsg) 607 { 608 struct sock *sk = sock->sk; 609 int err; 610 long timeo; 611 612 /* 613 * uaddr can be NULL and addr_len can be 0 if: 614 * sk is a TCP fastopen active socket and 615 * TCP_FASTOPEN_CONNECT sockopt is set and 616 * we already have a valid cookie for this socket. 617 * In this case, user can call write() after connect(). 618 * write() will invoke tcp_sendmsg_fastopen() which calls 619 * __inet_stream_connect(). 620 */ 621 if (uaddr) { 622 if (addr_len < sizeof(uaddr->sa_family)) 623 return -EINVAL; 624 625 if (uaddr->sa_family == AF_UNSPEC) { 626 err = sk->sk_prot->disconnect(sk, flags); 627 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED; 628 goto out; 629 } 630 } 631 632 switch (sock->state) { 633 default: 634 err = -EINVAL; 635 goto out; 636 case SS_CONNECTED: 637 err = -EISCONN; 638 goto out; 639 case SS_CONNECTING: 640 if (inet_sk(sk)->defer_connect) 641 err = is_sendmsg ? -EINPROGRESS : -EISCONN; 642 else 643 err = -EALREADY; 644 /* Fall out of switch with err, set for this state */ 645 break; 646 case SS_UNCONNECTED: 647 err = -EISCONN; 648 if (sk->sk_state != TCP_CLOSE) 649 goto out; 650 651 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) { 652 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len); 653 if (err) 654 goto out; 655 } 656 657 err = sk->sk_prot->connect(sk, uaddr, addr_len); 658 if (err < 0) 659 goto out; 660 661 sock->state = SS_CONNECTING; 662 663 if (!err && inet_sk(sk)->defer_connect) 664 goto out; 665 666 /* Just entered SS_CONNECTING state; the only 667 * difference is that return value in non-blocking 668 * case is EINPROGRESS, rather than EALREADY. 669 */ 670 err = -EINPROGRESS; 671 break; 672 } 673 674 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK); 675 676 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 677 int writebias = (sk->sk_protocol == IPPROTO_TCP) && 678 tcp_sk(sk)->fastopen_req && 679 tcp_sk(sk)->fastopen_req->data ? 1 : 0; 680 681 /* Error code is set above */ 682 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias)) 683 goto out; 684 685 err = sock_intr_errno(timeo); 686 if (signal_pending(current)) 687 goto out; 688 } 689 690 /* Connection was closed by RST, timeout, ICMP error 691 * or another process disconnected us. 692 */ 693 if (sk->sk_state == TCP_CLOSE) 694 goto sock_error; 695 696 /* sk->sk_err may be not zero now, if RECVERR was ordered by user 697 * and error was received after socket entered established state. 698 * Hence, it is handled normally after connect() return successfully. 699 */ 700 701 sock->state = SS_CONNECTED; 702 err = 0; 703 out: 704 return err; 705 706 sock_error: 707 err = sock_error(sk) ? : -ECONNABORTED; 708 sock->state = SS_UNCONNECTED; 709 if (sk->sk_prot->disconnect(sk, flags)) 710 sock->state = SS_DISCONNECTING; 711 goto out; 712 } 713 EXPORT_SYMBOL(__inet_stream_connect); 714 715 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr, 716 int addr_len, int flags) 717 { 718 int err; 719 720 lock_sock(sock->sk); 721 err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0); 722 release_sock(sock->sk); 723 return err; 724 } 725 EXPORT_SYMBOL(inet_stream_connect); 726 727 /* 728 * Accept a pending connection. The TCP layer now gives BSD semantics. 729 */ 730 731 int inet_accept(struct socket *sock, struct socket *newsock, int flags, 732 bool kern) 733 { 734 struct sock *sk1 = sock->sk; 735 int err = -EINVAL; 736 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err, kern); 737 738 if (!sk2) 739 goto do_err; 740 741 lock_sock(sk2); 742 743 sock_rps_record_flow(sk2); 744 WARN_ON(!((1 << sk2->sk_state) & 745 (TCPF_ESTABLISHED | TCPF_SYN_RECV | 746 TCPF_CLOSE_WAIT | TCPF_CLOSE))); 747 748 sock_graft(sk2, newsock); 749 750 newsock->state = SS_CONNECTED; 751 err = 0; 752 release_sock(sk2); 753 do_err: 754 return err; 755 } 756 EXPORT_SYMBOL(inet_accept); 757 758 /* 759 * This does both peername and sockname. 760 */ 761 int inet_getname(struct socket *sock, struct sockaddr *uaddr, 762 int peer) 763 { 764 struct sock *sk = sock->sk; 765 struct inet_sock *inet = inet_sk(sk); 766 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr); 767 768 sin->sin_family = AF_INET; 769 if (peer) { 770 if (!inet->inet_dport || 771 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) && 772 peer == 1)) 773 return -ENOTCONN; 774 sin->sin_port = inet->inet_dport; 775 sin->sin_addr.s_addr = inet->inet_daddr; 776 } else { 777 __be32 addr = inet->inet_rcv_saddr; 778 if (!addr) 779 addr = inet->inet_saddr; 780 sin->sin_port = inet->inet_sport; 781 sin->sin_addr.s_addr = addr; 782 } 783 if (cgroup_bpf_enabled) 784 BPF_CGROUP_RUN_SA_PROG_LOCK(sk, (struct sockaddr *)sin, 785 peer ? BPF_CGROUP_INET4_GETPEERNAME : 786 BPF_CGROUP_INET4_GETSOCKNAME, 787 NULL); 788 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 789 return sizeof(*sin); 790 } 791 EXPORT_SYMBOL(inet_getname); 792 793 int inet_send_prepare(struct sock *sk) 794 { 795 sock_rps_record_flow(sk); 796 797 /* We may need to bind the socket. */ 798 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind && 799 inet_autobind(sk)) 800 return -EAGAIN; 801 802 return 0; 803 } 804 EXPORT_SYMBOL_GPL(inet_send_prepare); 805 806 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size) 807 { 808 struct sock *sk = sock->sk; 809 810 if (unlikely(inet_send_prepare(sk))) 811 return -EAGAIN; 812 813 return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg, 814 sk, msg, size); 815 } 816 EXPORT_SYMBOL(inet_sendmsg); 817 818 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, 819 size_t size, int flags) 820 { 821 struct sock *sk = sock->sk; 822 823 if (unlikely(inet_send_prepare(sk))) 824 return -EAGAIN; 825 826 if (sk->sk_prot->sendpage) 827 return sk->sk_prot->sendpage(sk, page, offset, size, flags); 828 return sock_no_sendpage(sock, page, offset, size, flags); 829 } 830 EXPORT_SYMBOL(inet_sendpage); 831 832 INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *, 833 size_t, int, int, int *)); 834 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 835 int flags) 836 { 837 struct sock *sk = sock->sk; 838 int addr_len = 0; 839 int err; 840 841 if (likely(!(flags & MSG_ERRQUEUE))) 842 sock_rps_record_flow(sk); 843 844 err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg, 845 sk, msg, size, flags & MSG_DONTWAIT, 846 flags & ~MSG_DONTWAIT, &addr_len); 847 if (err >= 0) 848 msg->msg_namelen = addr_len; 849 return err; 850 } 851 EXPORT_SYMBOL(inet_recvmsg); 852 853 int inet_shutdown(struct socket *sock, int how) 854 { 855 struct sock *sk = sock->sk; 856 int err = 0; 857 858 /* This should really check to make sure 859 * the socket is a TCP socket. (WHY AC...) 860 */ 861 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and 862 1->2 bit 2 snds. 863 2->3 */ 864 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */ 865 return -EINVAL; 866 867 lock_sock(sk); 868 if (sock->state == SS_CONNECTING) { 869 if ((1 << sk->sk_state) & 870 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE)) 871 sock->state = SS_DISCONNECTING; 872 else 873 sock->state = SS_CONNECTED; 874 } 875 876 switch (sk->sk_state) { 877 case TCP_CLOSE: 878 err = -ENOTCONN; 879 /* Hack to wake up other listeners, who can poll for 880 EPOLLHUP, even on eg. unconnected UDP sockets -- RR */ 881 fallthrough; 882 default: 883 sk->sk_shutdown |= how; 884 if (sk->sk_prot->shutdown) 885 sk->sk_prot->shutdown(sk, how); 886 break; 887 888 /* Remaining two branches are temporary solution for missing 889 * close() in multithreaded environment. It is _not_ a good idea, 890 * but we have no choice until close() is repaired at VFS level. 891 */ 892 case TCP_LISTEN: 893 if (!(how & RCV_SHUTDOWN)) 894 break; 895 fallthrough; 896 case TCP_SYN_SENT: 897 err = sk->sk_prot->disconnect(sk, O_NONBLOCK); 898 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED; 899 break; 900 } 901 902 /* Wake up anyone sleeping in poll. */ 903 sk->sk_state_change(sk); 904 release_sock(sk); 905 return err; 906 } 907 EXPORT_SYMBOL(inet_shutdown); 908 909 /* 910 * ioctl() calls you can issue on an INET socket. Most of these are 911 * device configuration and stuff and very rarely used. Some ioctls 912 * pass on to the socket itself. 913 * 914 * NOTE: I like the idea of a module for the config stuff. ie ifconfig 915 * loads the devconfigure module does its configuring and unloads it. 916 * There's a good 20K of config code hanging around the kernel. 917 */ 918 919 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 920 { 921 struct sock *sk = sock->sk; 922 int err = 0; 923 struct net *net = sock_net(sk); 924 void __user *p = (void __user *)arg; 925 struct ifreq ifr; 926 struct rtentry rt; 927 928 switch (cmd) { 929 case SIOCADDRT: 930 case SIOCDELRT: 931 if (copy_from_user(&rt, p, sizeof(struct rtentry))) 932 return -EFAULT; 933 err = ip_rt_ioctl(net, cmd, &rt); 934 break; 935 case SIOCRTMSG: 936 err = -EINVAL; 937 break; 938 case SIOCDARP: 939 case SIOCGARP: 940 case SIOCSARP: 941 err = arp_ioctl(net, cmd, (void __user *)arg); 942 break; 943 case SIOCGIFADDR: 944 case SIOCGIFBRDADDR: 945 case SIOCGIFNETMASK: 946 case SIOCGIFDSTADDR: 947 case SIOCGIFPFLAGS: 948 if (copy_from_user(&ifr, p, sizeof(struct ifreq))) 949 return -EFAULT; 950 err = devinet_ioctl(net, cmd, &ifr); 951 if (!err && copy_to_user(p, &ifr, sizeof(struct ifreq))) 952 err = -EFAULT; 953 break; 954 955 case SIOCSIFADDR: 956 case SIOCSIFBRDADDR: 957 case SIOCSIFNETMASK: 958 case SIOCSIFDSTADDR: 959 case SIOCSIFPFLAGS: 960 case SIOCSIFFLAGS: 961 if (copy_from_user(&ifr, p, sizeof(struct ifreq))) 962 return -EFAULT; 963 err = devinet_ioctl(net, cmd, &ifr); 964 break; 965 default: 966 if (sk->sk_prot->ioctl) 967 err = sk->sk_prot->ioctl(sk, cmd, arg); 968 else 969 err = -ENOIOCTLCMD; 970 break; 971 } 972 return err; 973 } 974 EXPORT_SYMBOL(inet_ioctl); 975 976 #ifdef CONFIG_COMPAT 977 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 978 { 979 struct sock *sk = sock->sk; 980 int err = -ENOIOCTLCMD; 981 982 if (sk->sk_prot->compat_ioctl) 983 err = sk->sk_prot->compat_ioctl(sk, cmd, arg); 984 985 return err; 986 } 987 #endif 988 989 const struct proto_ops inet_stream_ops = { 990 .family = PF_INET, 991 .owner = THIS_MODULE, 992 .release = inet_release, 993 .bind = inet_bind, 994 .connect = inet_stream_connect, 995 .socketpair = sock_no_socketpair, 996 .accept = inet_accept, 997 .getname = inet_getname, 998 .poll = tcp_poll, 999 .ioctl = inet_ioctl, 1000 .gettstamp = sock_gettstamp, 1001 .listen = inet_listen, 1002 .shutdown = inet_shutdown, 1003 .setsockopt = sock_common_setsockopt, 1004 .getsockopt = sock_common_getsockopt, 1005 .sendmsg = inet_sendmsg, 1006 .recvmsg = inet_recvmsg, 1007 #ifdef CONFIG_MMU 1008 .mmap = tcp_mmap, 1009 #endif 1010 .sendpage = inet_sendpage, 1011 .splice_read = tcp_splice_read, 1012 .read_sock = tcp_read_sock, 1013 .sendmsg_locked = tcp_sendmsg_locked, 1014 .sendpage_locked = tcp_sendpage_locked, 1015 .peek_len = tcp_peek_len, 1016 #ifdef CONFIG_COMPAT 1017 .compat_setsockopt = compat_sock_common_setsockopt, 1018 .compat_getsockopt = compat_sock_common_getsockopt, 1019 .compat_ioctl = inet_compat_ioctl, 1020 #endif 1021 .set_rcvlowat = tcp_set_rcvlowat, 1022 }; 1023 EXPORT_SYMBOL(inet_stream_ops); 1024 1025 const struct proto_ops inet_dgram_ops = { 1026 .family = PF_INET, 1027 .owner = THIS_MODULE, 1028 .release = inet_release, 1029 .bind = inet_bind, 1030 .connect = inet_dgram_connect, 1031 .socketpair = sock_no_socketpair, 1032 .accept = sock_no_accept, 1033 .getname = inet_getname, 1034 .poll = udp_poll, 1035 .ioctl = inet_ioctl, 1036 .gettstamp = sock_gettstamp, 1037 .listen = sock_no_listen, 1038 .shutdown = inet_shutdown, 1039 .setsockopt = sock_common_setsockopt, 1040 .getsockopt = sock_common_getsockopt, 1041 .sendmsg = inet_sendmsg, 1042 .recvmsg = inet_recvmsg, 1043 .mmap = sock_no_mmap, 1044 .sendpage = inet_sendpage, 1045 .set_peek_off = sk_set_peek_off, 1046 #ifdef CONFIG_COMPAT 1047 .compat_setsockopt = compat_sock_common_setsockopt, 1048 .compat_getsockopt = compat_sock_common_getsockopt, 1049 .compat_ioctl = inet_compat_ioctl, 1050 #endif 1051 }; 1052 EXPORT_SYMBOL(inet_dgram_ops); 1053 1054 /* 1055 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without 1056 * udp_poll 1057 */ 1058 static const struct proto_ops inet_sockraw_ops = { 1059 .family = PF_INET, 1060 .owner = THIS_MODULE, 1061 .release = inet_release, 1062 .bind = inet_bind, 1063 .connect = inet_dgram_connect, 1064 .socketpair = sock_no_socketpair, 1065 .accept = sock_no_accept, 1066 .getname = inet_getname, 1067 .poll = datagram_poll, 1068 .ioctl = inet_ioctl, 1069 .gettstamp = sock_gettstamp, 1070 .listen = sock_no_listen, 1071 .shutdown = inet_shutdown, 1072 .setsockopt = sock_common_setsockopt, 1073 .getsockopt = sock_common_getsockopt, 1074 .sendmsg = inet_sendmsg, 1075 .recvmsg = inet_recvmsg, 1076 .mmap = sock_no_mmap, 1077 .sendpage = inet_sendpage, 1078 #ifdef CONFIG_COMPAT 1079 .compat_setsockopt = compat_sock_common_setsockopt, 1080 .compat_getsockopt = compat_sock_common_getsockopt, 1081 .compat_ioctl = inet_compat_ioctl, 1082 #endif 1083 }; 1084 1085 static const struct net_proto_family inet_family_ops = { 1086 .family = PF_INET, 1087 .create = inet_create, 1088 .owner = THIS_MODULE, 1089 }; 1090 1091 /* Upon startup we insert all the elements in inetsw_array[] into 1092 * the linked list inetsw. 1093 */ 1094 static struct inet_protosw inetsw_array[] = 1095 { 1096 { 1097 .type = SOCK_STREAM, 1098 .protocol = IPPROTO_TCP, 1099 .prot = &tcp_prot, 1100 .ops = &inet_stream_ops, 1101 .flags = INET_PROTOSW_PERMANENT | 1102 INET_PROTOSW_ICSK, 1103 }, 1104 1105 { 1106 .type = SOCK_DGRAM, 1107 .protocol = IPPROTO_UDP, 1108 .prot = &udp_prot, 1109 .ops = &inet_dgram_ops, 1110 .flags = INET_PROTOSW_PERMANENT, 1111 }, 1112 1113 { 1114 .type = SOCK_DGRAM, 1115 .protocol = IPPROTO_ICMP, 1116 .prot = &ping_prot, 1117 .ops = &inet_sockraw_ops, 1118 .flags = INET_PROTOSW_REUSE, 1119 }, 1120 1121 { 1122 .type = SOCK_RAW, 1123 .protocol = IPPROTO_IP, /* wild card */ 1124 .prot = &raw_prot, 1125 .ops = &inet_sockraw_ops, 1126 .flags = INET_PROTOSW_REUSE, 1127 } 1128 }; 1129 1130 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array) 1131 1132 void inet_register_protosw(struct inet_protosw *p) 1133 { 1134 struct list_head *lh; 1135 struct inet_protosw *answer; 1136 int protocol = p->protocol; 1137 struct list_head *last_perm; 1138 1139 spin_lock_bh(&inetsw_lock); 1140 1141 if (p->type >= SOCK_MAX) 1142 goto out_illegal; 1143 1144 /* If we are trying to override a permanent protocol, bail. */ 1145 last_perm = &inetsw[p->type]; 1146 list_for_each(lh, &inetsw[p->type]) { 1147 answer = list_entry(lh, struct inet_protosw, list); 1148 /* Check only the non-wild match. */ 1149 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0) 1150 break; 1151 if (protocol == answer->protocol) 1152 goto out_permanent; 1153 last_perm = lh; 1154 } 1155 1156 /* Add the new entry after the last permanent entry if any, so that 1157 * the new entry does not override a permanent entry when matched with 1158 * a wild-card protocol. But it is allowed to override any existing 1159 * non-permanent entry. This means that when we remove this entry, the 1160 * system automatically returns to the old behavior. 1161 */ 1162 list_add_rcu(&p->list, last_perm); 1163 out: 1164 spin_unlock_bh(&inetsw_lock); 1165 1166 return; 1167 1168 out_permanent: 1169 pr_err("Attempt to override permanent protocol %d\n", protocol); 1170 goto out; 1171 1172 out_illegal: 1173 pr_err("Ignoring attempt to register invalid socket type %d\n", 1174 p->type); 1175 goto out; 1176 } 1177 EXPORT_SYMBOL(inet_register_protosw); 1178 1179 void inet_unregister_protosw(struct inet_protosw *p) 1180 { 1181 if (INET_PROTOSW_PERMANENT & p->flags) { 1182 pr_err("Attempt to unregister permanent protocol %d\n", 1183 p->protocol); 1184 } else { 1185 spin_lock_bh(&inetsw_lock); 1186 list_del_rcu(&p->list); 1187 spin_unlock_bh(&inetsw_lock); 1188 1189 synchronize_net(); 1190 } 1191 } 1192 EXPORT_SYMBOL(inet_unregister_protosw); 1193 1194 static int inet_sk_reselect_saddr(struct sock *sk) 1195 { 1196 struct inet_sock *inet = inet_sk(sk); 1197 __be32 old_saddr = inet->inet_saddr; 1198 __be32 daddr = inet->inet_daddr; 1199 struct flowi4 *fl4; 1200 struct rtable *rt; 1201 __be32 new_saddr; 1202 struct ip_options_rcu *inet_opt; 1203 1204 inet_opt = rcu_dereference_protected(inet->inet_opt, 1205 lockdep_sock_is_held(sk)); 1206 if (inet_opt && inet_opt->opt.srr) 1207 daddr = inet_opt->opt.faddr; 1208 1209 /* Query new route. */ 1210 fl4 = &inet->cork.fl.u.ip4; 1211 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk), 1212 sk->sk_bound_dev_if, sk->sk_protocol, 1213 inet->inet_sport, inet->inet_dport, sk); 1214 if (IS_ERR(rt)) 1215 return PTR_ERR(rt); 1216 1217 sk_setup_caps(sk, &rt->dst); 1218 1219 new_saddr = fl4->saddr; 1220 1221 if (new_saddr == old_saddr) 1222 return 0; 1223 1224 if (sock_net(sk)->ipv4.sysctl_ip_dynaddr > 1) { 1225 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n", 1226 __func__, &old_saddr, &new_saddr); 1227 } 1228 1229 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr; 1230 1231 /* 1232 * XXX The only one ugly spot where we need to 1233 * XXX really change the sockets identity after 1234 * XXX it has entered the hashes. -DaveM 1235 * 1236 * Besides that, it does not check for connection 1237 * uniqueness. Wait for troubles. 1238 */ 1239 return __sk_prot_rehash(sk); 1240 } 1241 1242 int inet_sk_rebuild_header(struct sock *sk) 1243 { 1244 struct inet_sock *inet = inet_sk(sk); 1245 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0); 1246 __be32 daddr; 1247 struct ip_options_rcu *inet_opt; 1248 struct flowi4 *fl4; 1249 int err; 1250 1251 /* Route is OK, nothing to do. */ 1252 if (rt) 1253 return 0; 1254 1255 /* Reroute. */ 1256 rcu_read_lock(); 1257 inet_opt = rcu_dereference(inet->inet_opt); 1258 daddr = inet->inet_daddr; 1259 if (inet_opt && inet_opt->opt.srr) 1260 daddr = inet_opt->opt.faddr; 1261 rcu_read_unlock(); 1262 fl4 = &inet->cork.fl.u.ip4; 1263 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr, 1264 inet->inet_dport, inet->inet_sport, 1265 sk->sk_protocol, RT_CONN_FLAGS(sk), 1266 sk->sk_bound_dev_if); 1267 if (!IS_ERR(rt)) { 1268 err = 0; 1269 sk_setup_caps(sk, &rt->dst); 1270 } else { 1271 err = PTR_ERR(rt); 1272 1273 /* Routing failed... */ 1274 sk->sk_route_caps = 0; 1275 /* 1276 * Other protocols have to map its equivalent state to TCP_SYN_SENT. 1277 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme 1278 */ 1279 if (!sock_net(sk)->ipv4.sysctl_ip_dynaddr || 1280 sk->sk_state != TCP_SYN_SENT || 1281 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) || 1282 (err = inet_sk_reselect_saddr(sk)) != 0) 1283 sk->sk_err_soft = -err; 1284 } 1285 1286 return err; 1287 } 1288 EXPORT_SYMBOL(inet_sk_rebuild_header); 1289 1290 void inet_sk_set_state(struct sock *sk, int state) 1291 { 1292 trace_inet_sock_set_state(sk, sk->sk_state, state); 1293 sk->sk_state = state; 1294 } 1295 EXPORT_SYMBOL(inet_sk_set_state); 1296 1297 void inet_sk_state_store(struct sock *sk, int newstate) 1298 { 1299 trace_inet_sock_set_state(sk, sk->sk_state, newstate); 1300 smp_store_release(&sk->sk_state, newstate); 1301 } 1302 1303 struct sk_buff *inet_gso_segment(struct sk_buff *skb, 1304 netdev_features_t features) 1305 { 1306 bool udpfrag = false, fixedid = false, gso_partial, encap; 1307 struct sk_buff *segs = ERR_PTR(-EINVAL); 1308 const struct net_offload *ops; 1309 unsigned int offset = 0; 1310 struct iphdr *iph; 1311 int proto, tot_len; 1312 int nhoff; 1313 int ihl; 1314 int id; 1315 1316 skb_reset_network_header(skb); 1317 nhoff = skb_network_header(skb) - skb_mac_header(skb); 1318 if (unlikely(!pskb_may_pull(skb, sizeof(*iph)))) 1319 goto out; 1320 1321 iph = ip_hdr(skb); 1322 ihl = iph->ihl * 4; 1323 if (ihl < sizeof(*iph)) 1324 goto out; 1325 1326 id = ntohs(iph->id); 1327 proto = iph->protocol; 1328 1329 /* Warning: after this point, iph might be no longer valid */ 1330 if (unlikely(!pskb_may_pull(skb, ihl))) 1331 goto out; 1332 __skb_pull(skb, ihl); 1333 1334 encap = SKB_GSO_CB(skb)->encap_level > 0; 1335 if (encap) 1336 features &= skb->dev->hw_enc_features; 1337 SKB_GSO_CB(skb)->encap_level += ihl; 1338 1339 skb_reset_transport_header(skb); 1340 1341 segs = ERR_PTR(-EPROTONOSUPPORT); 1342 1343 if (!skb->encapsulation || encap) { 1344 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP); 1345 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID); 1346 1347 /* fixed ID is invalid if DF bit is not set */ 1348 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF))) 1349 goto out; 1350 } 1351 1352 ops = rcu_dereference(inet_offloads[proto]); 1353 if (likely(ops && ops->callbacks.gso_segment)) 1354 segs = ops->callbacks.gso_segment(skb, features); 1355 1356 if (IS_ERR_OR_NULL(segs)) 1357 goto out; 1358 1359 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL); 1360 1361 skb = segs; 1362 do { 1363 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff); 1364 if (udpfrag) { 1365 iph->frag_off = htons(offset >> 3); 1366 if (skb->next) 1367 iph->frag_off |= htons(IP_MF); 1368 offset += skb->len - nhoff - ihl; 1369 tot_len = skb->len - nhoff; 1370 } else if (skb_is_gso(skb)) { 1371 if (!fixedid) { 1372 iph->id = htons(id); 1373 id += skb_shinfo(skb)->gso_segs; 1374 } 1375 1376 if (gso_partial) 1377 tot_len = skb_shinfo(skb)->gso_size + 1378 SKB_GSO_CB(skb)->data_offset + 1379 skb->head - (unsigned char *)iph; 1380 else 1381 tot_len = skb->len - nhoff; 1382 } else { 1383 if (!fixedid) 1384 iph->id = htons(id++); 1385 tot_len = skb->len - nhoff; 1386 } 1387 iph->tot_len = htons(tot_len); 1388 ip_send_check(iph); 1389 if (encap) 1390 skb_reset_inner_headers(skb); 1391 skb->network_header = (u8 *)iph - skb->head; 1392 skb_reset_mac_len(skb); 1393 } while ((skb = skb->next)); 1394 1395 out: 1396 return segs; 1397 } 1398 EXPORT_SYMBOL(inet_gso_segment); 1399 1400 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb, 1401 netdev_features_t features) 1402 { 1403 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4)) 1404 return ERR_PTR(-EINVAL); 1405 1406 return inet_gso_segment(skb, features); 1407 } 1408 1409 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *, 1410 struct sk_buff *)); 1411 INDIRECT_CALLABLE_DECLARE(struct sk_buff *udp4_gro_receive(struct list_head *, 1412 struct sk_buff *)); 1413 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb) 1414 { 1415 const struct net_offload *ops; 1416 struct sk_buff *pp = NULL; 1417 const struct iphdr *iph; 1418 struct sk_buff *p; 1419 unsigned int hlen; 1420 unsigned int off; 1421 unsigned int id; 1422 int flush = 1; 1423 int proto; 1424 1425 off = skb_gro_offset(skb); 1426 hlen = off + sizeof(*iph); 1427 iph = skb_gro_header_fast(skb, off); 1428 if (skb_gro_header_hard(skb, hlen)) { 1429 iph = skb_gro_header_slow(skb, hlen, off); 1430 if (unlikely(!iph)) 1431 goto out; 1432 } 1433 1434 proto = iph->protocol; 1435 1436 rcu_read_lock(); 1437 ops = rcu_dereference(inet_offloads[proto]); 1438 if (!ops || !ops->callbacks.gro_receive) 1439 goto out_unlock; 1440 1441 if (*(u8 *)iph != 0x45) 1442 goto out_unlock; 1443 1444 if (ip_is_fragment(iph)) 1445 goto out_unlock; 1446 1447 if (unlikely(ip_fast_csum((u8 *)iph, 5))) 1448 goto out_unlock; 1449 1450 id = ntohl(*(__be32 *)&iph->id); 1451 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF)); 1452 id >>= 16; 1453 1454 list_for_each_entry(p, head, list) { 1455 struct iphdr *iph2; 1456 u16 flush_id; 1457 1458 if (!NAPI_GRO_CB(p)->same_flow) 1459 continue; 1460 1461 iph2 = (struct iphdr *)(p->data + off); 1462 /* The above works because, with the exception of the top 1463 * (inner most) layer, we only aggregate pkts with the same 1464 * hdr length so all the hdrs we'll need to verify will start 1465 * at the same offset. 1466 */ 1467 if ((iph->protocol ^ iph2->protocol) | 1468 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) | 1469 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) { 1470 NAPI_GRO_CB(p)->same_flow = 0; 1471 continue; 1472 } 1473 1474 /* All fields must match except length and checksum. */ 1475 NAPI_GRO_CB(p)->flush |= 1476 (iph->ttl ^ iph2->ttl) | 1477 (iph->tos ^ iph2->tos) | 1478 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF)); 1479 1480 NAPI_GRO_CB(p)->flush |= flush; 1481 1482 /* We need to store of the IP ID check to be included later 1483 * when we can verify that this packet does in fact belong 1484 * to a given flow. 1485 */ 1486 flush_id = (u16)(id - ntohs(iph2->id)); 1487 1488 /* This bit of code makes it much easier for us to identify 1489 * the cases where we are doing atomic vs non-atomic IP ID 1490 * checks. Specifically an atomic check can return IP ID 1491 * values 0 - 0xFFFF, while a non-atomic check can only 1492 * return 0 or 0xFFFF. 1493 */ 1494 if (!NAPI_GRO_CB(p)->is_atomic || 1495 !(iph->frag_off & htons(IP_DF))) { 1496 flush_id ^= NAPI_GRO_CB(p)->count; 1497 flush_id = flush_id ? 0xFFFF : 0; 1498 } 1499 1500 /* If the previous IP ID value was based on an atomic 1501 * datagram we can overwrite the value and ignore it. 1502 */ 1503 if (NAPI_GRO_CB(skb)->is_atomic) 1504 NAPI_GRO_CB(p)->flush_id = flush_id; 1505 else 1506 NAPI_GRO_CB(p)->flush_id |= flush_id; 1507 } 1508 1509 NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF)); 1510 NAPI_GRO_CB(skb)->flush |= flush; 1511 skb_set_network_header(skb, off); 1512 /* The above will be needed by the transport layer if there is one 1513 * immediately following this IP hdr. 1514 */ 1515 1516 /* Note : No need to call skb_gro_postpull_rcsum() here, 1517 * as we already checked checksum over ipv4 header was 0 1518 */ 1519 skb_gro_pull(skb, sizeof(*iph)); 1520 skb_set_transport_header(skb, skb_gro_offset(skb)); 1521 1522 pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive, 1523 ops->callbacks.gro_receive, head, skb); 1524 1525 out_unlock: 1526 rcu_read_unlock(); 1527 1528 out: 1529 skb_gro_flush_final(skb, pp, flush); 1530 1531 return pp; 1532 } 1533 EXPORT_SYMBOL(inet_gro_receive); 1534 1535 static struct sk_buff *ipip_gro_receive(struct list_head *head, 1536 struct sk_buff *skb) 1537 { 1538 if (NAPI_GRO_CB(skb)->encap_mark) { 1539 NAPI_GRO_CB(skb)->flush = 1; 1540 return NULL; 1541 } 1542 1543 NAPI_GRO_CB(skb)->encap_mark = 1; 1544 1545 return inet_gro_receive(head, skb); 1546 } 1547 1548 #define SECONDS_PER_DAY 86400 1549 1550 /* inet_current_timestamp - Return IP network timestamp 1551 * 1552 * Return milliseconds since midnight in network byte order. 1553 */ 1554 __be32 inet_current_timestamp(void) 1555 { 1556 u32 secs; 1557 u32 msecs; 1558 struct timespec64 ts; 1559 1560 ktime_get_real_ts64(&ts); 1561 1562 /* Get secs since midnight. */ 1563 (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs); 1564 /* Convert to msecs. */ 1565 msecs = secs * MSEC_PER_SEC; 1566 /* Convert nsec to msec. */ 1567 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC; 1568 1569 /* Convert to network byte order. */ 1570 return htonl(msecs); 1571 } 1572 EXPORT_SYMBOL(inet_current_timestamp); 1573 1574 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len) 1575 { 1576 if (sk->sk_family == AF_INET) 1577 return ip_recv_error(sk, msg, len, addr_len); 1578 #if IS_ENABLED(CONFIG_IPV6) 1579 if (sk->sk_family == AF_INET6) 1580 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len); 1581 #endif 1582 return -EINVAL; 1583 } 1584 1585 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *, int)); 1586 INDIRECT_CALLABLE_DECLARE(int udp4_gro_complete(struct sk_buff *, int)); 1587 int inet_gro_complete(struct sk_buff *skb, int nhoff) 1588 { 1589 __be16 newlen = htons(skb->len - nhoff); 1590 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff); 1591 const struct net_offload *ops; 1592 int proto = iph->protocol; 1593 int err = -ENOSYS; 1594 1595 if (skb->encapsulation) { 1596 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP)); 1597 skb_set_inner_network_header(skb, nhoff); 1598 } 1599 1600 csum_replace2(&iph->check, iph->tot_len, newlen); 1601 iph->tot_len = newlen; 1602 1603 rcu_read_lock(); 1604 ops = rcu_dereference(inet_offloads[proto]); 1605 if (WARN_ON(!ops || !ops->callbacks.gro_complete)) 1606 goto out_unlock; 1607 1608 /* Only need to add sizeof(*iph) to get to the next hdr below 1609 * because any hdr with option will have been flushed in 1610 * inet_gro_receive(). 1611 */ 1612 err = INDIRECT_CALL_2(ops->callbacks.gro_complete, 1613 tcp4_gro_complete, udp4_gro_complete, 1614 skb, nhoff + sizeof(*iph)); 1615 1616 out_unlock: 1617 rcu_read_unlock(); 1618 1619 return err; 1620 } 1621 EXPORT_SYMBOL(inet_gro_complete); 1622 1623 static int ipip_gro_complete(struct sk_buff *skb, int nhoff) 1624 { 1625 skb->encapsulation = 1; 1626 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4; 1627 return inet_gro_complete(skb, nhoff); 1628 } 1629 1630 int inet_ctl_sock_create(struct sock **sk, unsigned short family, 1631 unsigned short type, unsigned char protocol, 1632 struct net *net) 1633 { 1634 struct socket *sock; 1635 int rc = sock_create_kern(net, family, type, protocol, &sock); 1636 1637 if (rc == 0) { 1638 *sk = sock->sk; 1639 (*sk)->sk_allocation = GFP_ATOMIC; 1640 /* 1641 * Unhash it so that IP input processing does not even see it, 1642 * we do not wish this socket to see incoming packets. 1643 */ 1644 (*sk)->sk_prot->unhash(*sk); 1645 } 1646 return rc; 1647 } 1648 EXPORT_SYMBOL_GPL(inet_ctl_sock_create); 1649 1650 u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt) 1651 { 1652 return *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt); 1653 } 1654 EXPORT_SYMBOL_GPL(snmp_get_cpu_field); 1655 1656 unsigned long snmp_fold_field(void __percpu *mib, int offt) 1657 { 1658 unsigned long res = 0; 1659 int i; 1660 1661 for_each_possible_cpu(i) 1662 res += snmp_get_cpu_field(mib, i, offt); 1663 return res; 1664 } 1665 EXPORT_SYMBOL_GPL(snmp_fold_field); 1666 1667 #if BITS_PER_LONG==32 1668 1669 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt, 1670 size_t syncp_offset) 1671 { 1672 void *bhptr; 1673 struct u64_stats_sync *syncp; 1674 u64 v; 1675 unsigned int start; 1676 1677 bhptr = per_cpu_ptr(mib, cpu); 1678 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset); 1679 do { 1680 start = u64_stats_fetch_begin_irq(syncp); 1681 v = *(((u64 *)bhptr) + offt); 1682 } while (u64_stats_fetch_retry_irq(syncp, start)); 1683 1684 return v; 1685 } 1686 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64); 1687 1688 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset) 1689 { 1690 u64 res = 0; 1691 int cpu; 1692 1693 for_each_possible_cpu(cpu) { 1694 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset); 1695 } 1696 return res; 1697 } 1698 EXPORT_SYMBOL_GPL(snmp_fold_field64); 1699 #endif 1700 1701 #ifdef CONFIG_IP_MULTICAST 1702 static const struct net_protocol igmp_protocol = { 1703 .handler = igmp_rcv, 1704 .netns_ok = 1, 1705 }; 1706 #endif 1707 1708 /* thinking of making this const? Don't. 1709 * early_demux can change based on sysctl. 1710 */ 1711 static struct net_protocol tcp_protocol = { 1712 .early_demux = tcp_v4_early_demux, 1713 .early_demux_handler = tcp_v4_early_demux, 1714 .handler = tcp_v4_rcv, 1715 .err_handler = tcp_v4_err, 1716 .no_policy = 1, 1717 .netns_ok = 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 .netns_ok = 1, 1731 }; 1732 1733 static const struct net_protocol icmp_protocol = { 1734 .handler = icmp_rcv, 1735 .err_handler = icmp_err, 1736 .no_policy = 1, 1737 .netns_ok = 1, 1738 }; 1739 1740 static __net_init int ipv4_mib_init_net(struct net *net) 1741 { 1742 int i; 1743 1744 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib); 1745 if (!net->mib.tcp_statistics) 1746 goto err_tcp_mib; 1747 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib); 1748 if (!net->mib.ip_statistics) 1749 goto err_ip_mib; 1750 1751 for_each_possible_cpu(i) { 1752 struct ipstats_mib *af_inet_stats; 1753 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i); 1754 u64_stats_init(&af_inet_stats->syncp); 1755 } 1756 1757 net->mib.net_statistics = alloc_percpu(struct linux_mib); 1758 if (!net->mib.net_statistics) 1759 goto err_net_mib; 1760 net->mib.udp_statistics = alloc_percpu(struct udp_mib); 1761 if (!net->mib.udp_statistics) 1762 goto err_udp_mib; 1763 net->mib.udplite_statistics = alloc_percpu(struct udp_mib); 1764 if (!net->mib.udplite_statistics) 1765 goto err_udplite_mib; 1766 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib); 1767 if (!net->mib.icmp_statistics) 1768 goto err_icmp_mib; 1769 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib), 1770 GFP_KERNEL); 1771 if (!net->mib.icmpmsg_statistics) 1772 goto err_icmpmsg_mib; 1773 1774 tcp_mib_init(net); 1775 return 0; 1776 1777 err_icmpmsg_mib: 1778 free_percpu(net->mib.icmp_statistics); 1779 err_icmp_mib: 1780 free_percpu(net->mib.udplite_statistics); 1781 err_udplite_mib: 1782 free_percpu(net->mib.udp_statistics); 1783 err_udp_mib: 1784 free_percpu(net->mib.net_statistics); 1785 err_net_mib: 1786 free_percpu(net->mib.ip_statistics); 1787 err_ip_mib: 1788 free_percpu(net->mib.tcp_statistics); 1789 err_tcp_mib: 1790 return -ENOMEM; 1791 } 1792 1793 static __net_exit void ipv4_mib_exit_net(struct net *net) 1794 { 1795 kfree(net->mib.icmpmsg_statistics); 1796 free_percpu(net->mib.icmp_statistics); 1797 free_percpu(net->mib.udplite_statistics); 1798 free_percpu(net->mib.udp_statistics); 1799 free_percpu(net->mib.net_statistics); 1800 free_percpu(net->mib.ip_statistics); 1801 free_percpu(net->mib.tcp_statistics); 1802 #ifdef CONFIG_MPTCP 1803 /* allocated on demand, see mptcp_init_sock() */ 1804 free_percpu(net->mib.mptcp_statistics); 1805 #endif 1806 } 1807 1808 static __net_initdata struct pernet_operations ipv4_mib_ops = { 1809 .init = ipv4_mib_init_net, 1810 .exit = ipv4_mib_exit_net, 1811 }; 1812 1813 static int __init init_ipv4_mibs(void) 1814 { 1815 return register_pernet_subsys(&ipv4_mib_ops); 1816 } 1817 1818 static __net_init int inet_init_net(struct net *net) 1819 { 1820 /* 1821 * Set defaults for local port range 1822 */ 1823 seqlock_init(&net->ipv4.ip_local_ports.lock); 1824 net->ipv4.ip_local_ports.range[0] = 32768; 1825 net->ipv4.ip_local_ports.range[1] = 60999; 1826 1827 seqlock_init(&net->ipv4.ping_group_range.lock); 1828 /* 1829 * Sane defaults - nobody may create ping sockets. 1830 * Boot scripts should set this to distro-specific group. 1831 */ 1832 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1); 1833 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0); 1834 1835 /* Default values for sysctl-controlled parameters. 1836 * We set them here, in case sysctl is not compiled. 1837 */ 1838 net->ipv4.sysctl_ip_default_ttl = IPDEFTTL; 1839 net->ipv4.sysctl_ip_fwd_update_priority = 1; 1840 net->ipv4.sysctl_ip_dynaddr = 0; 1841 net->ipv4.sysctl_ip_early_demux = 1; 1842 net->ipv4.sysctl_udp_early_demux = 1; 1843 net->ipv4.sysctl_tcp_early_demux = 1; 1844 net->ipv4.sysctl_nexthop_compat_mode = 1; 1845 #ifdef CONFIG_SYSCTL 1846 net->ipv4.sysctl_ip_prot_sock = PROT_SOCK; 1847 #endif 1848 1849 /* Some igmp sysctl, whose values are always used */ 1850 net->ipv4.sysctl_igmp_max_memberships = 20; 1851 net->ipv4.sysctl_igmp_max_msf = 10; 1852 /* IGMP reports for link-local multicast groups are enabled by default */ 1853 net->ipv4.sysctl_igmp_llm_reports = 1; 1854 net->ipv4.sysctl_igmp_qrv = 2; 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