1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * PF_INET protocol family socket handler. 8 * 9 * Authors: Ross Biro 10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 11 * Florian La Roche, <flla@stud.uni-sb.de> 12 * Alan Cox, <A.Cox@swansea.ac.uk> 13 * 14 * Changes (see also sock.c) 15 * 16 * piggy, 17 * Karl Knutson : Socket protocol table 18 * A.N.Kuznetsov : Socket death error in accept(). 19 * John Richardson : Fix non blocking error in connect() 20 * so sockets that fail to connect 21 * don't return -EINPROGRESS. 22 * Alan Cox : Asynchronous I/O support 23 * Alan Cox : Keep correct socket pointer on sock 24 * structures 25 * when accept() ed 26 * Alan Cox : Semantics of SO_LINGER aren't state 27 * moved to close when you look carefully. 28 * With this fixed and the accept bug fixed 29 * some RPC stuff seems happier. 30 * Niibe Yutaka : 4.4BSD style write async I/O 31 * Alan Cox, 32 * Tony Gale : Fixed reuse semantics. 33 * Alan Cox : bind() shouldn't abort existing but dead 34 * sockets. Stops FTP netin:.. I hope. 35 * Alan Cox : bind() works correctly for RAW sockets. 36 * Note that FreeBSD at least was broken 37 * in this respect so be careful with 38 * compatibility tests... 39 * Alan Cox : routing cache support 40 * Alan Cox : memzero the socket structure for 41 * compactness. 42 * Matt Day : nonblock connect error handler 43 * Alan Cox : Allow large numbers of pending sockets 44 * (eg for big web sites), but only if 45 * specifically application requested. 46 * Alan Cox : New buffering throughout IP. Used 47 * dumbly. 48 * Alan Cox : New buffering now used smartly. 49 * Alan Cox : BSD rather than common sense 50 * interpretation of listen. 51 * Germano Caronni : Assorted small races. 52 * Alan Cox : sendmsg/recvmsg basic support. 53 * Alan Cox : Only sendmsg/recvmsg now supported. 54 * Alan Cox : Locked down bind (see security list). 55 * Alan Cox : Loosened bind a little. 56 * Mike McLagan : ADD/DEL DLCI Ioctls 57 * Willy Konynenberg : Transparent proxying support. 58 * David S. Miller : New socket lookup architecture. 59 * Some other random speedups. 60 * Cyrus Durgin : Cleaned up file for kmod hacks. 61 * Andi Kleen : Fix inet_stream_connect TCP race. 62 */ 63 64 #define pr_fmt(fmt) "IPv4: " fmt 65 66 #include <linux/err.h> 67 #include <linux/errno.h> 68 #include <linux/types.h> 69 #include <linux/socket.h> 70 #include <linux/in.h> 71 #include <linux/kernel.h> 72 #include <linux/kmod.h> 73 #include <linux/sched.h> 74 #include <linux/timer.h> 75 #include <linux/string.h> 76 #include <linux/sockios.h> 77 #include <linux/net.h> 78 #include <linux/capability.h> 79 #include <linux/fcntl.h> 80 #include <linux/mm.h> 81 #include <linux/interrupt.h> 82 #include <linux/stat.h> 83 #include <linux/init.h> 84 #include <linux/poll.h> 85 #include <linux/netfilter_ipv4.h> 86 #include <linux/random.h> 87 #include <linux/slab.h> 88 89 #include <linux/uaccess.h> 90 91 #include <linux/inet.h> 92 #include <linux/igmp.h> 93 #include <linux/inetdevice.h> 94 #include <linux/netdevice.h> 95 #include <net/checksum.h> 96 #include <net/ip.h> 97 #include <net/protocol.h> 98 #include <net/arp.h> 99 #include <net/route.h> 100 #include <net/ip_fib.h> 101 #include <net/inet_connection_sock.h> 102 #include <net/gro.h> 103 #include <net/gso.h> 104 #include <net/tcp.h> 105 #include <net/udp.h> 106 #include <net/udplite.h> 107 #include <net/ping.h> 108 #include <linux/skbuff.h> 109 #include <net/sock.h> 110 #include <net/raw.h> 111 #include <net/icmp.h> 112 #include <net/inet_common.h> 113 #include <net/ip_tunnels.h> 114 #include <net/xfrm.h> 115 #include <net/net_namespace.h> 116 #include <net/secure_seq.h> 117 #ifdef CONFIG_IP_MROUTE 118 #include <linux/mroute.h> 119 #endif 120 #include <net/l3mdev.h> 121 #include <net/compat.h> 122 123 #include <trace/events/sock.h> 124 125 /* The inetsw table contains everything that inet_create needs to 126 * build a new socket. 127 */ 128 static struct list_head inetsw[SOCK_MAX]; 129 static DEFINE_SPINLOCK(inetsw_lock); 130 131 /* New destruction routine */ 132 133 void inet_sock_destruct(struct sock *sk) 134 { 135 struct inet_sock *inet = inet_sk(sk); 136 137 __skb_queue_purge(&sk->sk_receive_queue); 138 __skb_queue_purge(&sk->sk_error_queue); 139 140 sk_mem_reclaim_final(sk); 141 142 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) { 143 pr_err("Attempt to release TCP socket in state %d %p\n", 144 sk->sk_state, sk); 145 return; 146 } 147 if (!sock_flag(sk, SOCK_DEAD)) { 148 pr_err("Attempt to release alive inet socket %p\n", sk); 149 return; 150 } 151 152 WARN_ON_ONCE(atomic_read(&sk->sk_rmem_alloc)); 153 WARN_ON_ONCE(refcount_read(&sk->sk_wmem_alloc)); 154 WARN_ON_ONCE(sk->sk_wmem_queued); 155 WARN_ON_ONCE(sk_forward_alloc_get(sk)); 156 157 kfree(rcu_dereference_protected(inet->inet_opt, 1)); 158 dst_release(rcu_dereference_protected(sk->sk_dst_cache, 1)); 159 dst_release(rcu_dereference_protected(sk->sk_rx_dst, 1)); 160 } 161 EXPORT_SYMBOL(inet_sock_destruct); 162 163 /* 164 * The routines beyond this point handle the behaviour of an AF_INET 165 * socket object. Mostly it punts to the subprotocols of IP to do 166 * the work. 167 */ 168 169 /* 170 * Automatically bind an unbound socket. 171 */ 172 173 static int inet_autobind(struct sock *sk) 174 { 175 struct inet_sock *inet; 176 /* We may need to bind the socket. */ 177 lock_sock(sk); 178 inet = inet_sk(sk); 179 if (!inet->inet_num) { 180 if (sk->sk_prot->get_port(sk, 0)) { 181 release_sock(sk); 182 return -EAGAIN; 183 } 184 inet->inet_sport = htons(inet->inet_num); 185 } 186 release_sock(sk); 187 return 0; 188 } 189 190 /* 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 = READ_ONCE(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 (READ_ONCE(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 sk->sk_txrehash = READ_ONCE(net->core.sysctl_txrehash); 351 352 inet->uc_ttl = -1; 353 inet->mc_loop = 1; 354 inet->mc_ttl = 1; 355 inet->mc_all = 1; 356 inet->mc_index = 0; 357 inet->mc_list = NULL; 358 inet->rcv_tos = 0; 359 360 if (inet->inet_num) { 361 /* It assumes that any protocol which allows 362 * the user to assign a number at socket 363 * creation time automatically 364 * shares. 365 */ 366 inet->inet_sport = htons(inet->inet_num); 367 /* Add to protocol hash chains. */ 368 err = sk->sk_prot->hash(sk); 369 if (err) { 370 sk_common_release(sk); 371 goto out; 372 } 373 } 374 375 if (sk->sk_prot->init) { 376 err = sk->sk_prot->init(sk); 377 if (err) { 378 sk_common_release(sk); 379 goto out; 380 } 381 } 382 383 if (!kern) { 384 err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk); 385 if (err) { 386 sk_common_release(sk); 387 goto out; 388 } 389 } 390 out: 391 return err; 392 out_rcu_unlock: 393 rcu_read_unlock(); 394 goto out; 395 } 396 397 398 /* 399 * The peer socket should always be NULL (or else). When we call this 400 * function we are destroying the object and from then on nobody 401 * should refer to it. 402 */ 403 int inet_release(struct socket *sock) 404 { 405 struct sock *sk = sock->sk; 406 407 if (sk) { 408 long timeout; 409 410 if (!sk->sk_kern_sock) 411 BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk); 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 u32 flags = BIND_WITH_LOCK; 438 int err; 439 440 /* If the socket has its own bind function then use it. (RAW) */ 441 if (sk->sk_prot->bind) { 442 return sk->sk_prot->bind(sk, uaddr, addr_len); 443 } 444 if (addr_len < sizeof(struct sockaddr_in)) 445 return -EINVAL; 446 447 /* BPF prog is run before any checks are done so that if the prog 448 * changes context in a wrong way it will be caught. 449 */ 450 err = BPF_CGROUP_RUN_PROG_INET_BIND_LOCK(sk, uaddr, 451 CGROUP_INET4_BIND, &flags); 452 if (err) 453 return err; 454 455 return __inet_bind(sk, uaddr, addr_len, flags); 456 } 457 EXPORT_SYMBOL(inet_bind); 458 459 int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len, 460 u32 flags) 461 { 462 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr; 463 struct inet_sock *inet = inet_sk(sk); 464 struct net *net = sock_net(sk); 465 unsigned short snum; 466 int chk_addr_ret; 467 u32 tb_id = RT_TABLE_LOCAL; 468 int err; 469 470 if (addr->sin_family != AF_INET) { 471 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET) 472 * only if s_addr is INADDR_ANY. 473 */ 474 err = -EAFNOSUPPORT; 475 if (addr->sin_family != AF_UNSPEC || 476 addr->sin_addr.s_addr != htonl(INADDR_ANY)) 477 goto out; 478 } 479 480 tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id; 481 chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id); 482 483 /* Not specified by any standard per-se, however it breaks too 484 * many applications when removed. It is unfortunate since 485 * allowing applications to make a non-local bind solves 486 * several problems with systems using dynamic addressing. 487 * (ie. your servers still start up even if your ISDN link 488 * is temporarily down) 489 */ 490 err = -EADDRNOTAVAIL; 491 if (!inet_addr_valid_or_nonlocal(net, inet, addr->sin_addr.s_addr, 492 chk_addr_ret)) 493 goto out; 494 495 snum = ntohs(addr->sin_port); 496 err = -EACCES; 497 if (!(flags & BIND_NO_CAP_NET_BIND_SERVICE) && 498 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 err = sk->sk_prot->get_port(sk, snum); 525 if (err) { 526 inet->inet_saddr = inet->inet_rcv_saddr = 0; 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 if (sk->sk_prot->put_port) 534 sk->sk_prot->put_port(sk); 535 goto out_release_sock; 536 } 537 } 538 } 539 540 if (inet->inet_rcv_saddr) 541 sk->sk_userlocks |= SOCK_BINDADDR_LOCK; 542 if (snum) 543 sk->sk_userlocks |= SOCK_BINDPORT_LOCK; 544 inet->inet_sport = htons(inet->inet_num); 545 inet->inet_daddr = 0; 546 inet->inet_dport = 0; 547 sk_dst_reset(sk); 548 err = 0; 549 out_release_sock: 550 if (flags & BIND_WITH_LOCK) 551 release_sock(sk); 552 out: 553 return err; 554 } 555 556 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr, 557 int addr_len, int flags) 558 { 559 struct sock *sk = sock->sk; 560 const struct proto *prot; 561 int err; 562 563 if (addr_len < sizeof(uaddr->sa_family)) 564 return -EINVAL; 565 566 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 567 prot = READ_ONCE(sk->sk_prot); 568 569 if (uaddr->sa_family == AF_UNSPEC) 570 return prot->disconnect(sk, flags); 571 572 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) { 573 err = prot->pre_connect(sk, uaddr, addr_len); 574 if (err) 575 return err; 576 } 577 578 if (data_race(!inet_sk(sk)->inet_num) && inet_autobind(sk)) 579 return -EAGAIN; 580 return prot->connect(sk, uaddr, addr_len); 581 } 582 EXPORT_SYMBOL(inet_dgram_connect); 583 584 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias) 585 { 586 DEFINE_WAIT_FUNC(wait, woken_wake_function); 587 588 add_wait_queue(sk_sleep(sk), &wait); 589 sk->sk_write_pending += writebias; 590 sk->sk_wait_pending++; 591 592 /* Basic assumption: if someone sets sk->sk_err, he _must_ 593 * change state of the socket from TCP_SYN_*. 594 * Connect() does not allow to get error notifications 595 * without closing the socket. 596 */ 597 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 598 release_sock(sk); 599 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo); 600 lock_sock(sk); 601 if (signal_pending(current) || !timeo) 602 break; 603 } 604 remove_wait_queue(sk_sleep(sk), &wait); 605 sk->sk_write_pending -= writebias; 606 sk->sk_wait_pending--; 607 return timeo; 608 } 609 610 /* 611 * Connect to a remote host. There is regrettably still a little 612 * TCP 'magic' in here. 613 */ 614 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr, 615 int addr_len, int flags, int is_sendmsg) 616 { 617 struct sock *sk = sock->sk; 618 int err; 619 long timeo; 620 621 /* 622 * uaddr can be NULL and addr_len can be 0 if: 623 * sk is a TCP fastopen active socket and 624 * TCP_FASTOPEN_CONNECT sockopt is set and 625 * we already have a valid cookie for this socket. 626 * In this case, user can call write() after connect(). 627 * write() will invoke tcp_sendmsg_fastopen() which calls 628 * __inet_stream_connect(). 629 */ 630 if (uaddr) { 631 if (addr_len < sizeof(uaddr->sa_family)) 632 return -EINVAL; 633 634 if (uaddr->sa_family == AF_UNSPEC) { 635 err = sk->sk_prot->disconnect(sk, flags); 636 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED; 637 goto out; 638 } 639 } 640 641 switch (sock->state) { 642 default: 643 err = -EINVAL; 644 goto out; 645 case SS_CONNECTED: 646 err = -EISCONN; 647 goto out; 648 case SS_CONNECTING: 649 if (inet_sk(sk)->defer_connect) 650 err = is_sendmsg ? -EINPROGRESS : -EISCONN; 651 else 652 err = -EALREADY; 653 /* Fall out of switch with err, set for this state */ 654 break; 655 case SS_UNCONNECTED: 656 err = -EISCONN; 657 if (sk->sk_state != TCP_CLOSE) 658 goto out; 659 660 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) { 661 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len); 662 if (err) 663 goto out; 664 } 665 666 err = sk->sk_prot->connect(sk, uaddr, addr_len); 667 if (err < 0) 668 goto out; 669 670 sock->state = SS_CONNECTING; 671 672 if (!err && inet_sk(sk)->defer_connect) 673 goto out; 674 675 /* Just entered SS_CONNECTING state; the only 676 * difference is that return value in non-blocking 677 * case is EINPROGRESS, rather than EALREADY. 678 */ 679 err = -EINPROGRESS; 680 break; 681 } 682 683 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK); 684 685 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 686 int writebias = (sk->sk_protocol == IPPROTO_TCP) && 687 tcp_sk(sk)->fastopen_req && 688 tcp_sk(sk)->fastopen_req->data ? 1 : 0; 689 690 /* Error code is set above */ 691 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias)) 692 goto out; 693 694 err = sock_intr_errno(timeo); 695 if (signal_pending(current)) 696 goto out; 697 } 698 699 /* Connection was closed by RST, timeout, ICMP error 700 * or another process disconnected us. 701 */ 702 if (sk->sk_state == TCP_CLOSE) 703 goto sock_error; 704 705 /* sk->sk_err may be not zero now, if RECVERR was ordered by user 706 * and error was received after socket entered established state. 707 * Hence, it is handled normally after connect() return successfully. 708 */ 709 710 sock->state = SS_CONNECTED; 711 err = 0; 712 out: 713 return err; 714 715 sock_error: 716 err = sock_error(sk) ? : -ECONNABORTED; 717 sock->state = SS_UNCONNECTED; 718 if (sk->sk_prot->disconnect(sk, flags)) 719 sock->state = SS_DISCONNECTING; 720 goto out; 721 } 722 EXPORT_SYMBOL(__inet_stream_connect); 723 724 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr, 725 int addr_len, int flags) 726 { 727 int err; 728 729 lock_sock(sock->sk); 730 err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0); 731 release_sock(sock->sk); 732 return err; 733 } 734 EXPORT_SYMBOL(inet_stream_connect); 735 736 void __inet_accept(struct socket *sock, struct socket *newsock, struct sock *newsk) 737 { 738 sock_rps_record_flow(newsk); 739 WARN_ON(!((1 << newsk->sk_state) & 740 (TCPF_ESTABLISHED | TCPF_SYN_RECV | 741 TCPF_CLOSE_WAIT | TCPF_CLOSE))); 742 743 if (test_bit(SOCK_SUPPORT_ZC, &sock->flags)) 744 set_bit(SOCK_SUPPORT_ZC, &newsock->flags); 745 sock_graft(newsk, newsock); 746 747 newsock->state = SS_CONNECTED; 748 } 749 750 /* 751 * Accept a pending connection. The TCP layer now gives BSD semantics. 752 */ 753 754 int inet_accept(struct socket *sock, struct socket *newsock, int flags, 755 bool kern) 756 { 757 struct sock *sk1 = sock->sk, *sk2; 758 int err = -EINVAL; 759 760 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 761 sk2 = READ_ONCE(sk1->sk_prot)->accept(sk1, flags, &err, kern); 762 if (!sk2) 763 return err; 764 765 lock_sock(sk2); 766 __inet_accept(sock, newsock, sk2); 767 release_sock(sk2); 768 return 0; 769 } 770 EXPORT_SYMBOL(inet_accept); 771 772 /* 773 * This does both peername and sockname. 774 */ 775 int inet_getname(struct socket *sock, struct sockaddr *uaddr, 776 int peer) 777 { 778 struct sock *sk = sock->sk; 779 struct inet_sock *inet = inet_sk(sk); 780 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr); 781 782 sin->sin_family = AF_INET; 783 lock_sock(sk); 784 if (peer) { 785 if (!inet->inet_dport || 786 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) && 787 peer == 1)) { 788 release_sock(sk); 789 return -ENOTCONN; 790 } 791 sin->sin_port = inet->inet_dport; 792 sin->sin_addr.s_addr = inet->inet_daddr; 793 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin, 794 CGROUP_INET4_GETPEERNAME); 795 } else { 796 __be32 addr = inet->inet_rcv_saddr; 797 if (!addr) 798 addr = inet->inet_saddr; 799 sin->sin_port = inet->inet_sport; 800 sin->sin_addr.s_addr = addr; 801 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin, 802 CGROUP_INET4_GETSOCKNAME); 803 } 804 release_sock(sk); 805 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 806 return sizeof(*sin); 807 } 808 EXPORT_SYMBOL(inet_getname); 809 810 int inet_send_prepare(struct sock *sk) 811 { 812 sock_rps_record_flow(sk); 813 814 /* We may need to bind the socket. */ 815 if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind && 816 inet_autobind(sk)) 817 return -EAGAIN; 818 819 return 0; 820 } 821 EXPORT_SYMBOL_GPL(inet_send_prepare); 822 823 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size) 824 { 825 struct sock *sk = sock->sk; 826 827 if (unlikely(inet_send_prepare(sk))) 828 return -EAGAIN; 829 830 return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg, 831 sk, msg, size); 832 } 833 EXPORT_SYMBOL(inet_sendmsg); 834 835 void inet_splice_eof(struct socket *sock) 836 { 837 const struct proto *prot; 838 struct sock *sk = sock->sk; 839 840 if (unlikely(inet_send_prepare(sk))) 841 return; 842 843 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 844 prot = READ_ONCE(sk->sk_prot); 845 if (prot->splice_eof) 846 prot->splice_eof(sock); 847 } 848 EXPORT_SYMBOL_GPL(inet_splice_eof); 849 850 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, 851 size_t size, int flags) 852 { 853 struct sock *sk = sock->sk; 854 const struct proto *prot; 855 856 if (unlikely(inet_send_prepare(sk))) 857 return -EAGAIN; 858 859 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 860 prot = READ_ONCE(sk->sk_prot); 861 if (prot->sendpage) 862 return prot->sendpage(sk, page, offset, size, flags); 863 return sock_no_sendpage(sock, page, offset, size, flags); 864 } 865 EXPORT_SYMBOL(inet_sendpage); 866 867 INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *, 868 size_t, int, int *)); 869 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 870 int flags) 871 { 872 struct sock *sk = sock->sk; 873 int addr_len = 0; 874 int err; 875 876 if (likely(!(flags & MSG_ERRQUEUE))) 877 sock_rps_record_flow(sk); 878 879 err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg, 880 sk, msg, size, flags, &addr_len); 881 if (err >= 0) 882 msg->msg_namelen = addr_len; 883 return err; 884 } 885 EXPORT_SYMBOL(inet_recvmsg); 886 887 int inet_shutdown(struct socket *sock, int how) 888 { 889 struct sock *sk = sock->sk; 890 int err = 0; 891 892 /* This should really check to make sure 893 * the socket is a TCP socket. (WHY AC...) 894 */ 895 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and 896 1->2 bit 2 snds. 897 2->3 */ 898 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */ 899 return -EINVAL; 900 901 lock_sock(sk); 902 if (sock->state == SS_CONNECTING) { 903 if ((1 << sk->sk_state) & 904 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE)) 905 sock->state = SS_DISCONNECTING; 906 else 907 sock->state = SS_CONNECTED; 908 } 909 910 switch (sk->sk_state) { 911 case TCP_CLOSE: 912 err = -ENOTCONN; 913 /* Hack to wake up other listeners, who can poll for 914 EPOLLHUP, even on eg. unconnected UDP sockets -- RR */ 915 fallthrough; 916 default: 917 WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | how); 918 if (sk->sk_prot->shutdown) 919 sk->sk_prot->shutdown(sk, how); 920 break; 921 922 /* Remaining two branches are temporary solution for missing 923 * close() in multithreaded environment. It is _not_ a good idea, 924 * but we have no choice until close() is repaired at VFS level. 925 */ 926 case TCP_LISTEN: 927 if (!(how & RCV_SHUTDOWN)) 928 break; 929 fallthrough; 930 case TCP_SYN_SENT: 931 err = sk->sk_prot->disconnect(sk, O_NONBLOCK); 932 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED; 933 break; 934 } 935 936 /* Wake up anyone sleeping in poll. */ 937 sk->sk_state_change(sk); 938 release_sock(sk); 939 return err; 940 } 941 EXPORT_SYMBOL(inet_shutdown); 942 943 /* 944 * ioctl() calls you can issue on an INET socket. Most of these are 945 * device configuration and stuff and very rarely used. Some ioctls 946 * pass on to the socket itself. 947 * 948 * NOTE: I like the idea of a module for the config stuff. ie ifconfig 949 * loads the devconfigure module does its configuring and unloads it. 950 * There's a good 20K of config code hanging around the kernel. 951 */ 952 953 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 954 { 955 struct sock *sk = sock->sk; 956 int err = 0; 957 struct net *net = sock_net(sk); 958 void __user *p = (void __user *)arg; 959 struct ifreq ifr; 960 struct rtentry rt; 961 962 switch (cmd) { 963 case SIOCADDRT: 964 case SIOCDELRT: 965 if (copy_from_user(&rt, p, sizeof(struct rtentry))) 966 return -EFAULT; 967 err = ip_rt_ioctl(net, cmd, &rt); 968 break; 969 case SIOCRTMSG: 970 err = -EINVAL; 971 break; 972 case SIOCDARP: 973 case SIOCGARP: 974 case SIOCSARP: 975 err = arp_ioctl(net, cmd, (void __user *)arg); 976 break; 977 case SIOCGIFADDR: 978 case SIOCGIFBRDADDR: 979 case SIOCGIFNETMASK: 980 case SIOCGIFDSTADDR: 981 case SIOCGIFPFLAGS: 982 if (get_user_ifreq(&ifr, NULL, p)) 983 return -EFAULT; 984 err = devinet_ioctl(net, cmd, &ifr); 985 if (!err && put_user_ifreq(&ifr, p)) 986 err = -EFAULT; 987 break; 988 989 case SIOCSIFADDR: 990 case SIOCSIFBRDADDR: 991 case SIOCSIFNETMASK: 992 case SIOCSIFDSTADDR: 993 case SIOCSIFPFLAGS: 994 case SIOCSIFFLAGS: 995 if (get_user_ifreq(&ifr, NULL, p)) 996 return -EFAULT; 997 err = devinet_ioctl(net, cmd, &ifr); 998 break; 999 default: 1000 if (sk->sk_prot->ioctl) 1001 err = sk->sk_prot->ioctl(sk, cmd, arg); 1002 else 1003 err = -ENOIOCTLCMD; 1004 break; 1005 } 1006 return err; 1007 } 1008 EXPORT_SYMBOL(inet_ioctl); 1009 1010 #ifdef CONFIG_COMPAT 1011 static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd, 1012 struct compat_rtentry __user *ur) 1013 { 1014 compat_uptr_t rtdev; 1015 struct rtentry rt; 1016 1017 if (copy_from_user(&rt.rt_dst, &ur->rt_dst, 1018 3 * sizeof(struct sockaddr)) || 1019 get_user(rt.rt_flags, &ur->rt_flags) || 1020 get_user(rt.rt_metric, &ur->rt_metric) || 1021 get_user(rt.rt_mtu, &ur->rt_mtu) || 1022 get_user(rt.rt_window, &ur->rt_window) || 1023 get_user(rt.rt_irtt, &ur->rt_irtt) || 1024 get_user(rtdev, &ur->rt_dev)) 1025 return -EFAULT; 1026 1027 rt.rt_dev = compat_ptr(rtdev); 1028 return ip_rt_ioctl(sock_net(sk), cmd, &rt); 1029 } 1030 1031 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 1032 { 1033 void __user *argp = compat_ptr(arg); 1034 struct sock *sk = sock->sk; 1035 1036 switch (cmd) { 1037 case SIOCADDRT: 1038 case SIOCDELRT: 1039 return inet_compat_routing_ioctl(sk, cmd, argp); 1040 default: 1041 if (!sk->sk_prot->compat_ioctl) 1042 return -ENOIOCTLCMD; 1043 return sk->sk_prot->compat_ioctl(sk, cmd, arg); 1044 } 1045 } 1046 #endif /* CONFIG_COMPAT */ 1047 1048 const struct proto_ops inet_stream_ops = { 1049 .family = PF_INET, 1050 .owner = THIS_MODULE, 1051 .release = inet_release, 1052 .bind = inet_bind, 1053 .connect = inet_stream_connect, 1054 .socketpair = sock_no_socketpair, 1055 .accept = inet_accept, 1056 .getname = inet_getname, 1057 .poll = tcp_poll, 1058 .ioctl = inet_ioctl, 1059 .gettstamp = sock_gettstamp, 1060 .listen = inet_listen, 1061 .shutdown = inet_shutdown, 1062 .setsockopt = sock_common_setsockopt, 1063 .getsockopt = sock_common_getsockopt, 1064 .sendmsg = inet_sendmsg, 1065 .recvmsg = inet_recvmsg, 1066 #ifdef CONFIG_MMU 1067 .mmap = tcp_mmap, 1068 #endif 1069 .splice_eof = inet_splice_eof, 1070 .sendpage = inet_sendpage, 1071 .splice_read = tcp_splice_read, 1072 .read_sock = tcp_read_sock, 1073 .read_skb = tcp_read_skb, 1074 .sendmsg_locked = tcp_sendmsg_locked, 1075 .sendpage_locked = tcp_sendpage_locked, 1076 .peek_len = tcp_peek_len, 1077 #ifdef CONFIG_COMPAT 1078 .compat_ioctl = inet_compat_ioctl, 1079 #endif 1080 .set_rcvlowat = tcp_set_rcvlowat, 1081 }; 1082 EXPORT_SYMBOL(inet_stream_ops); 1083 1084 const struct proto_ops inet_dgram_ops = { 1085 .family = PF_INET, 1086 .owner = THIS_MODULE, 1087 .release = inet_release, 1088 .bind = inet_bind, 1089 .connect = inet_dgram_connect, 1090 .socketpair = sock_no_socketpair, 1091 .accept = sock_no_accept, 1092 .getname = inet_getname, 1093 .poll = udp_poll, 1094 .ioctl = inet_ioctl, 1095 .gettstamp = sock_gettstamp, 1096 .listen = sock_no_listen, 1097 .shutdown = inet_shutdown, 1098 .setsockopt = sock_common_setsockopt, 1099 .getsockopt = sock_common_getsockopt, 1100 .sendmsg = inet_sendmsg, 1101 .read_skb = udp_read_skb, 1102 .recvmsg = inet_recvmsg, 1103 .mmap = sock_no_mmap, 1104 .splice_eof = inet_splice_eof, 1105 .sendpage = inet_sendpage, 1106 .set_peek_off = sk_set_peek_off, 1107 #ifdef CONFIG_COMPAT 1108 .compat_ioctl = inet_compat_ioctl, 1109 #endif 1110 }; 1111 EXPORT_SYMBOL(inet_dgram_ops); 1112 1113 /* 1114 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without 1115 * udp_poll 1116 */ 1117 static const struct proto_ops inet_sockraw_ops = { 1118 .family = PF_INET, 1119 .owner = THIS_MODULE, 1120 .release = inet_release, 1121 .bind = inet_bind, 1122 .connect = inet_dgram_connect, 1123 .socketpair = sock_no_socketpair, 1124 .accept = sock_no_accept, 1125 .getname = inet_getname, 1126 .poll = datagram_poll, 1127 .ioctl = inet_ioctl, 1128 .gettstamp = sock_gettstamp, 1129 .listen = sock_no_listen, 1130 .shutdown = inet_shutdown, 1131 .setsockopt = sock_common_setsockopt, 1132 .getsockopt = sock_common_getsockopt, 1133 .sendmsg = inet_sendmsg, 1134 .recvmsg = inet_recvmsg, 1135 .mmap = sock_no_mmap, 1136 .splice_eof = inet_splice_eof, 1137 .sendpage = inet_sendpage, 1138 #ifdef CONFIG_COMPAT 1139 .compat_ioctl = inet_compat_ioctl, 1140 #endif 1141 }; 1142 1143 static const struct net_proto_family inet_family_ops = { 1144 .family = PF_INET, 1145 .create = inet_create, 1146 .owner = THIS_MODULE, 1147 }; 1148 1149 /* Upon startup we insert all the elements in inetsw_array[] into 1150 * the linked list inetsw. 1151 */ 1152 static struct inet_protosw inetsw_array[] = 1153 { 1154 { 1155 .type = SOCK_STREAM, 1156 .protocol = IPPROTO_TCP, 1157 .prot = &tcp_prot, 1158 .ops = &inet_stream_ops, 1159 .flags = INET_PROTOSW_PERMANENT | 1160 INET_PROTOSW_ICSK, 1161 }, 1162 1163 { 1164 .type = SOCK_DGRAM, 1165 .protocol = IPPROTO_UDP, 1166 .prot = &udp_prot, 1167 .ops = &inet_dgram_ops, 1168 .flags = INET_PROTOSW_PERMANENT, 1169 }, 1170 1171 { 1172 .type = SOCK_DGRAM, 1173 .protocol = IPPROTO_ICMP, 1174 .prot = &ping_prot, 1175 .ops = &inet_sockraw_ops, 1176 .flags = INET_PROTOSW_REUSE, 1177 }, 1178 1179 { 1180 .type = SOCK_RAW, 1181 .protocol = IPPROTO_IP, /* wild card */ 1182 .prot = &raw_prot, 1183 .ops = &inet_sockraw_ops, 1184 .flags = INET_PROTOSW_REUSE, 1185 } 1186 }; 1187 1188 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array) 1189 1190 void inet_register_protosw(struct inet_protosw *p) 1191 { 1192 struct list_head *lh; 1193 struct inet_protosw *answer; 1194 int protocol = p->protocol; 1195 struct list_head *last_perm; 1196 1197 spin_lock_bh(&inetsw_lock); 1198 1199 if (p->type >= SOCK_MAX) 1200 goto out_illegal; 1201 1202 /* If we are trying to override a permanent protocol, bail. */ 1203 last_perm = &inetsw[p->type]; 1204 list_for_each(lh, &inetsw[p->type]) { 1205 answer = list_entry(lh, struct inet_protosw, list); 1206 /* Check only the non-wild match. */ 1207 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0) 1208 break; 1209 if (protocol == answer->protocol) 1210 goto out_permanent; 1211 last_perm = lh; 1212 } 1213 1214 /* Add the new entry after the last permanent entry if any, so that 1215 * the new entry does not override a permanent entry when matched with 1216 * a wild-card protocol. But it is allowed to override any existing 1217 * non-permanent entry. This means that when we remove this entry, the 1218 * system automatically returns to the old behavior. 1219 */ 1220 list_add_rcu(&p->list, last_perm); 1221 out: 1222 spin_unlock_bh(&inetsw_lock); 1223 1224 return; 1225 1226 out_permanent: 1227 pr_err("Attempt to override permanent protocol %d\n", protocol); 1228 goto out; 1229 1230 out_illegal: 1231 pr_err("Ignoring attempt to register invalid socket type %d\n", 1232 p->type); 1233 goto out; 1234 } 1235 EXPORT_SYMBOL(inet_register_protosw); 1236 1237 void inet_unregister_protosw(struct inet_protosw *p) 1238 { 1239 if (INET_PROTOSW_PERMANENT & p->flags) { 1240 pr_err("Attempt to unregister permanent protocol %d\n", 1241 p->protocol); 1242 } else { 1243 spin_lock_bh(&inetsw_lock); 1244 list_del_rcu(&p->list); 1245 spin_unlock_bh(&inetsw_lock); 1246 1247 synchronize_net(); 1248 } 1249 } 1250 EXPORT_SYMBOL(inet_unregister_protosw); 1251 1252 static int inet_sk_reselect_saddr(struct sock *sk) 1253 { 1254 struct inet_sock *inet = inet_sk(sk); 1255 __be32 old_saddr = inet->inet_saddr; 1256 __be32 daddr = inet->inet_daddr; 1257 struct flowi4 *fl4; 1258 struct rtable *rt; 1259 __be32 new_saddr; 1260 struct ip_options_rcu *inet_opt; 1261 int err; 1262 1263 inet_opt = rcu_dereference_protected(inet->inet_opt, 1264 lockdep_sock_is_held(sk)); 1265 if (inet_opt && inet_opt->opt.srr) 1266 daddr = inet_opt->opt.faddr; 1267 1268 /* Query new route. */ 1269 fl4 = &inet->cork.fl.u.ip4; 1270 rt = ip_route_connect(fl4, daddr, 0, sk->sk_bound_dev_if, 1271 sk->sk_protocol, inet->inet_sport, 1272 inet->inet_dport, sk); 1273 if (IS_ERR(rt)) 1274 return PTR_ERR(rt); 1275 1276 new_saddr = fl4->saddr; 1277 1278 if (new_saddr == old_saddr) { 1279 sk_setup_caps(sk, &rt->dst); 1280 return 0; 1281 } 1282 1283 err = inet_bhash2_update_saddr(sk, &new_saddr, AF_INET); 1284 if (err) { 1285 ip_rt_put(rt); 1286 return err; 1287 } 1288 1289 sk_setup_caps(sk, &rt->dst); 1290 1291 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) > 1) { 1292 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n", 1293 __func__, &old_saddr, &new_saddr); 1294 } 1295 1296 /* 1297 * XXX The only one ugly spot where we need to 1298 * XXX really change the sockets identity after 1299 * XXX it has entered the hashes. -DaveM 1300 * 1301 * Besides that, it does not check for connection 1302 * uniqueness. Wait for troubles. 1303 */ 1304 return __sk_prot_rehash(sk); 1305 } 1306 1307 int inet_sk_rebuild_header(struct sock *sk) 1308 { 1309 struct inet_sock *inet = inet_sk(sk); 1310 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0); 1311 __be32 daddr; 1312 struct ip_options_rcu *inet_opt; 1313 struct flowi4 *fl4; 1314 int err; 1315 1316 /* Route is OK, nothing to do. */ 1317 if (rt) 1318 return 0; 1319 1320 /* Reroute. */ 1321 rcu_read_lock(); 1322 inet_opt = rcu_dereference(inet->inet_opt); 1323 daddr = inet->inet_daddr; 1324 if (inet_opt && inet_opt->opt.srr) 1325 daddr = inet_opt->opt.faddr; 1326 rcu_read_unlock(); 1327 fl4 = &inet->cork.fl.u.ip4; 1328 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr, 1329 inet->inet_dport, inet->inet_sport, 1330 sk->sk_protocol, RT_CONN_FLAGS(sk), 1331 sk->sk_bound_dev_if); 1332 if (!IS_ERR(rt)) { 1333 err = 0; 1334 sk_setup_caps(sk, &rt->dst); 1335 } else { 1336 err = PTR_ERR(rt); 1337 1338 /* Routing failed... */ 1339 sk->sk_route_caps = 0; 1340 /* 1341 * Other protocols have to map its equivalent state to TCP_SYN_SENT. 1342 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme 1343 */ 1344 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) || 1345 sk->sk_state != TCP_SYN_SENT || 1346 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) || 1347 (err = inet_sk_reselect_saddr(sk)) != 0) 1348 WRITE_ONCE(sk->sk_err_soft, -err); 1349 } 1350 1351 return err; 1352 } 1353 EXPORT_SYMBOL(inet_sk_rebuild_header); 1354 1355 void inet_sk_set_state(struct sock *sk, int state) 1356 { 1357 trace_inet_sock_set_state(sk, sk->sk_state, state); 1358 sk->sk_state = state; 1359 } 1360 EXPORT_SYMBOL(inet_sk_set_state); 1361 1362 void inet_sk_state_store(struct sock *sk, int newstate) 1363 { 1364 trace_inet_sock_set_state(sk, sk->sk_state, newstate); 1365 smp_store_release(&sk->sk_state, newstate); 1366 } 1367 1368 struct sk_buff *inet_gso_segment(struct sk_buff *skb, 1369 netdev_features_t features) 1370 { 1371 bool udpfrag = false, fixedid = false, gso_partial, encap; 1372 struct sk_buff *segs = ERR_PTR(-EINVAL); 1373 const struct net_offload *ops; 1374 unsigned int offset = 0; 1375 struct iphdr *iph; 1376 int proto, tot_len; 1377 int nhoff; 1378 int ihl; 1379 int id; 1380 1381 skb_reset_network_header(skb); 1382 nhoff = skb_network_header(skb) - skb_mac_header(skb); 1383 if (unlikely(!pskb_may_pull(skb, sizeof(*iph)))) 1384 goto out; 1385 1386 iph = ip_hdr(skb); 1387 ihl = iph->ihl * 4; 1388 if (ihl < sizeof(*iph)) 1389 goto out; 1390 1391 id = ntohs(iph->id); 1392 proto = iph->protocol; 1393 1394 /* Warning: after this point, iph might be no longer valid */ 1395 if (unlikely(!pskb_may_pull(skb, ihl))) 1396 goto out; 1397 __skb_pull(skb, ihl); 1398 1399 encap = SKB_GSO_CB(skb)->encap_level > 0; 1400 if (encap) 1401 features &= skb->dev->hw_enc_features; 1402 SKB_GSO_CB(skb)->encap_level += ihl; 1403 1404 skb_reset_transport_header(skb); 1405 1406 segs = ERR_PTR(-EPROTONOSUPPORT); 1407 1408 if (!skb->encapsulation || encap) { 1409 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP); 1410 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID); 1411 1412 /* fixed ID is invalid if DF bit is not set */ 1413 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF))) 1414 goto out; 1415 } 1416 1417 ops = rcu_dereference(inet_offloads[proto]); 1418 if (likely(ops && ops->callbacks.gso_segment)) { 1419 segs = ops->callbacks.gso_segment(skb, features); 1420 if (!segs) 1421 skb->network_header = skb_mac_header(skb) + nhoff - skb->head; 1422 } 1423 1424 if (IS_ERR_OR_NULL(segs)) 1425 goto out; 1426 1427 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL); 1428 1429 skb = segs; 1430 do { 1431 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff); 1432 if (udpfrag) { 1433 iph->frag_off = htons(offset >> 3); 1434 if (skb->next) 1435 iph->frag_off |= htons(IP_MF); 1436 offset += skb->len - nhoff - ihl; 1437 tot_len = skb->len - nhoff; 1438 } else if (skb_is_gso(skb)) { 1439 if (!fixedid) { 1440 iph->id = htons(id); 1441 id += skb_shinfo(skb)->gso_segs; 1442 } 1443 1444 if (gso_partial) 1445 tot_len = skb_shinfo(skb)->gso_size + 1446 SKB_GSO_CB(skb)->data_offset + 1447 skb->head - (unsigned char *)iph; 1448 else 1449 tot_len = skb->len - nhoff; 1450 } else { 1451 if (!fixedid) 1452 iph->id = htons(id++); 1453 tot_len = skb->len - nhoff; 1454 } 1455 iph->tot_len = htons(tot_len); 1456 ip_send_check(iph); 1457 if (encap) 1458 skb_reset_inner_headers(skb); 1459 skb->network_header = (u8 *)iph - skb->head; 1460 skb_reset_mac_len(skb); 1461 } while ((skb = skb->next)); 1462 1463 out: 1464 return segs; 1465 } 1466 1467 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb, 1468 netdev_features_t features) 1469 { 1470 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4)) 1471 return ERR_PTR(-EINVAL); 1472 1473 return inet_gso_segment(skb, features); 1474 } 1475 1476 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb) 1477 { 1478 const struct net_offload *ops; 1479 struct sk_buff *pp = NULL; 1480 const struct iphdr *iph; 1481 struct sk_buff *p; 1482 unsigned int hlen; 1483 unsigned int off; 1484 unsigned int id; 1485 int flush = 1; 1486 int proto; 1487 1488 off = skb_gro_offset(skb); 1489 hlen = off + sizeof(*iph); 1490 iph = skb_gro_header(skb, hlen, off); 1491 if (unlikely(!iph)) 1492 goto out; 1493 1494 proto = iph->protocol; 1495 1496 ops = rcu_dereference(inet_offloads[proto]); 1497 if (!ops || !ops->callbacks.gro_receive) 1498 goto out; 1499 1500 if (*(u8 *)iph != 0x45) 1501 goto out; 1502 1503 if (ip_is_fragment(iph)) 1504 goto out; 1505 1506 if (unlikely(ip_fast_csum((u8 *)iph, 5))) 1507 goto out; 1508 1509 NAPI_GRO_CB(skb)->proto = proto; 1510 id = ntohl(*(__be32 *)&iph->id); 1511 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF)); 1512 id >>= 16; 1513 1514 list_for_each_entry(p, head, list) { 1515 struct iphdr *iph2; 1516 u16 flush_id; 1517 1518 if (!NAPI_GRO_CB(p)->same_flow) 1519 continue; 1520 1521 iph2 = (struct iphdr *)(p->data + off); 1522 /* The above works because, with the exception of the top 1523 * (inner most) layer, we only aggregate pkts with the same 1524 * hdr length so all the hdrs we'll need to verify will start 1525 * at the same offset. 1526 */ 1527 if ((iph->protocol ^ iph2->protocol) | 1528 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) | 1529 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) { 1530 NAPI_GRO_CB(p)->same_flow = 0; 1531 continue; 1532 } 1533 1534 /* All fields must match except length and checksum. */ 1535 NAPI_GRO_CB(p)->flush |= 1536 (iph->ttl ^ iph2->ttl) | 1537 (iph->tos ^ iph2->tos) | 1538 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF)); 1539 1540 NAPI_GRO_CB(p)->flush |= flush; 1541 1542 /* We need to store of the IP ID check to be included later 1543 * when we can verify that this packet does in fact belong 1544 * to a given flow. 1545 */ 1546 flush_id = (u16)(id - ntohs(iph2->id)); 1547 1548 /* This bit of code makes it much easier for us to identify 1549 * the cases where we are doing atomic vs non-atomic IP ID 1550 * checks. Specifically an atomic check can return IP ID 1551 * values 0 - 0xFFFF, while a non-atomic check can only 1552 * return 0 or 0xFFFF. 1553 */ 1554 if (!NAPI_GRO_CB(p)->is_atomic || 1555 !(iph->frag_off & htons(IP_DF))) { 1556 flush_id ^= NAPI_GRO_CB(p)->count; 1557 flush_id = flush_id ? 0xFFFF : 0; 1558 } 1559 1560 /* If the previous IP ID value was based on an atomic 1561 * datagram we can overwrite the value and ignore it. 1562 */ 1563 if (NAPI_GRO_CB(skb)->is_atomic) 1564 NAPI_GRO_CB(p)->flush_id = flush_id; 1565 else 1566 NAPI_GRO_CB(p)->flush_id |= flush_id; 1567 } 1568 1569 NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF)); 1570 NAPI_GRO_CB(skb)->flush |= flush; 1571 skb_set_network_header(skb, off); 1572 /* The above will be needed by the transport layer if there is one 1573 * immediately following this IP hdr. 1574 */ 1575 1576 /* Note : No need to call skb_gro_postpull_rcsum() here, 1577 * as we already checked checksum over ipv4 header was 0 1578 */ 1579 skb_gro_pull(skb, sizeof(*iph)); 1580 skb_set_transport_header(skb, skb_gro_offset(skb)); 1581 1582 pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive, 1583 ops->callbacks.gro_receive, head, skb); 1584 1585 out: 1586 skb_gro_flush_final(skb, pp, flush); 1587 1588 return pp; 1589 } 1590 1591 static struct sk_buff *ipip_gro_receive(struct list_head *head, 1592 struct sk_buff *skb) 1593 { 1594 if (NAPI_GRO_CB(skb)->encap_mark) { 1595 NAPI_GRO_CB(skb)->flush = 1; 1596 return NULL; 1597 } 1598 1599 NAPI_GRO_CB(skb)->encap_mark = 1; 1600 1601 return inet_gro_receive(head, skb); 1602 } 1603 1604 #define SECONDS_PER_DAY 86400 1605 1606 /* inet_current_timestamp - Return IP network timestamp 1607 * 1608 * Return milliseconds since midnight in network byte order. 1609 */ 1610 __be32 inet_current_timestamp(void) 1611 { 1612 u32 secs; 1613 u32 msecs; 1614 struct timespec64 ts; 1615 1616 ktime_get_real_ts64(&ts); 1617 1618 /* Get secs since midnight. */ 1619 (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs); 1620 /* Convert to msecs. */ 1621 msecs = secs * MSEC_PER_SEC; 1622 /* Convert nsec to msec. */ 1623 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC; 1624 1625 /* Convert to network byte order. */ 1626 return htonl(msecs); 1627 } 1628 EXPORT_SYMBOL(inet_current_timestamp); 1629 1630 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len) 1631 { 1632 if (sk->sk_family == AF_INET) 1633 return ip_recv_error(sk, msg, len, addr_len); 1634 #if IS_ENABLED(CONFIG_IPV6) 1635 if (sk->sk_family == AF_INET6) 1636 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len); 1637 #endif 1638 return -EINVAL; 1639 } 1640 1641 int inet_gro_complete(struct sk_buff *skb, int nhoff) 1642 { 1643 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff); 1644 const struct net_offload *ops; 1645 __be16 totlen = iph->tot_len; 1646 int proto = iph->protocol; 1647 int err = -ENOSYS; 1648 1649 if (skb->encapsulation) { 1650 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP)); 1651 skb_set_inner_network_header(skb, nhoff); 1652 } 1653 1654 iph_set_totlen(iph, skb->len - nhoff); 1655 csum_replace2(&iph->check, totlen, iph->tot_len); 1656 1657 ops = rcu_dereference(inet_offloads[proto]); 1658 if (WARN_ON(!ops || !ops->callbacks.gro_complete)) 1659 goto out; 1660 1661 /* Only need to add sizeof(*iph) to get to the next hdr below 1662 * because any hdr with option will have been flushed in 1663 * inet_gro_receive(). 1664 */ 1665 err = INDIRECT_CALL_2(ops->callbacks.gro_complete, 1666 tcp4_gro_complete, udp4_gro_complete, 1667 skb, nhoff + sizeof(*iph)); 1668 1669 out: 1670 return err; 1671 } 1672 1673 static int ipip_gro_complete(struct sk_buff *skb, int nhoff) 1674 { 1675 skb->encapsulation = 1; 1676 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4; 1677 return inet_gro_complete(skb, nhoff); 1678 } 1679 1680 int inet_ctl_sock_create(struct sock **sk, unsigned short family, 1681 unsigned short type, unsigned char protocol, 1682 struct net *net) 1683 { 1684 struct socket *sock; 1685 int rc = sock_create_kern(net, family, type, protocol, &sock); 1686 1687 if (rc == 0) { 1688 *sk = sock->sk; 1689 (*sk)->sk_allocation = GFP_ATOMIC; 1690 (*sk)->sk_use_task_frag = false; 1691 /* 1692 * Unhash it so that IP input processing does not even see it, 1693 * we do not wish this socket to see incoming packets. 1694 */ 1695 (*sk)->sk_prot->unhash(*sk); 1696 } 1697 return rc; 1698 } 1699 EXPORT_SYMBOL_GPL(inet_ctl_sock_create); 1700 1701 unsigned long snmp_fold_field(void __percpu *mib, int offt) 1702 { 1703 unsigned long res = 0; 1704 int i; 1705 1706 for_each_possible_cpu(i) 1707 res += snmp_get_cpu_field(mib, i, offt); 1708 return res; 1709 } 1710 EXPORT_SYMBOL_GPL(snmp_fold_field); 1711 1712 #if BITS_PER_LONG==32 1713 1714 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt, 1715 size_t syncp_offset) 1716 { 1717 void *bhptr; 1718 struct u64_stats_sync *syncp; 1719 u64 v; 1720 unsigned int start; 1721 1722 bhptr = per_cpu_ptr(mib, cpu); 1723 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset); 1724 do { 1725 start = u64_stats_fetch_begin(syncp); 1726 v = *(((u64 *)bhptr) + offt); 1727 } while (u64_stats_fetch_retry(syncp, start)); 1728 1729 return v; 1730 } 1731 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64); 1732 1733 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset) 1734 { 1735 u64 res = 0; 1736 int cpu; 1737 1738 for_each_possible_cpu(cpu) { 1739 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset); 1740 } 1741 return res; 1742 } 1743 EXPORT_SYMBOL_GPL(snmp_fold_field64); 1744 #endif 1745 1746 #ifdef CONFIG_IP_MULTICAST 1747 static const struct net_protocol igmp_protocol = { 1748 .handler = igmp_rcv, 1749 }; 1750 #endif 1751 1752 static const struct net_protocol tcp_protocol = { 1753 .handler = tcp_v4_rcv, 1754 .err_handler = tcp_v4_err, 1755 .no_policy = 1, 1756 .icmp_strict_tag_validation = 1, 1757 }; 1758 1759 static const struct net_protocol udp_protocol = { 1760 .handler = udp_rcv, 1761 .err_handler = udp_err, 1762 .no_policy = 1, 1763 }; 1764 1765 static const struct net_protocol icmp_protocol = { 1766 .handler = icmp_rcv, 1767 .err_handler = icmp_err, 1768 .no_policy = 1, 1769 }; 1770 1771 static __net_init int ipv4_mib_init_net(struct net *net) 1772 { 1773 int i; 1774 1775 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib); 1776 if (!net->mib.tcp_statistics) 1777 goto err_tcp_mib; 1778 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib); 1779 if (!net->mib.ip_statistics) 1780 goto err_ip_mib; 1781 1782 for_each_possible_cpu(i) { 1783 struct ipstats_mib *af_inet_stats; 1784 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i); 1785 u64_stats_init(&af_inet_stats->syncp); 1786 } 1787 1788 net->mib.net_statistics = alloc_percpu(struct linux_mib); 1789 if (!net->mib.net_statistics) 1790 goto err_net_mib; 1791 net->mib.udp_statistics = alloc_percpu(struct udp_mib); 1792 if (!net->mib.udp_statistics) 1793 goto err_udp_mib; 1794 net->mib.udplite_statistics = alloc_percpu(struct udp_mib); 1795 if (!net->mib.udplite_statistics) 1796 goto err_udplite_mib; 1797 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib); 1798 if (!net->mib.icmp_statistics) 1799 goto err_icmp_mib; 1800 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib), 1801 GFP_KERNEL); 1802 if (!net->mib.icmpmsg_statistics) 1803 goto err_icmpmsg_mib; 1804 1805 tcp_mib_init(net); 1806 return 0; 1807 1808 err_icmpmsg_mib: 1809 free_percpu(net->mib.icmp_statistics); 1810 err_icmp_mib: 1811 free_percpu(net->mib.udplite_statistics); 1812 err_udplite_mib: 1813 free_percpu(net->mib.udp_statistics); 1814 err_udp_mib: 1815 free_percpu(net->mib.net_statistics); 1816 err_net_mib: 1817 free_percpu(net->mib.ip_statistics); 1818 err_ip_mib: 1819 free_percpu(net->mib.tcp_statistics); 1820 err_tcp_mib: 1821 return -ENOMEM; 1822 } 1823 1824 static __net_exit void ipv4_mib_exit_net(struct net *net) 1825 { 1826 kfree(net->mib.icmpmsg_statistics); 1827 free_percpu(net->mib.icmp_statistics); 1828 free_percpu(net->mib.udplite_statistics); 1829 free_percpu(net->mib.udp_statistics); 1830 free_percpu(net->mib.net_statistics); 1831 free_percpu(net->mib.ip_statistics); 1832 free_percpu(net->mib.tcp_statistics); 1833 #ifdef CONFIG_MPTCP 1834 /* allocated on demand, see mptcp_init_sock() */ 1835 free_percpu(net->mib.mptcp_statistics); 1836 #endif 1837 } 1838 1839 static __net_initdata struct pernet_operations ipv4_mib_ops = { 1840 .init = ipv4_mib_init_net, 1841 .exit = ipv4_mib_exit_net, 1842 }; 1843 1844 static int __init init_ipv4_mibs(void) 1845 { 1846 return register_pernet_subsys(&ipv4_mib_ops); 1847 } 1848 1849 static __net_init int inet_init_net(struct net *net) 1850 { 1851 /* 1852 * Set defaults for local port range 1853 */ 1854 seqlock_init(&net->ipv4.ip_local_ports.lock); 1855 net->ipv4.ip_local_ports.range[0] = 32768; 1856 net->ipv4.ip_local_ports.range[1] = 60999; 1857 1858 seqlock_init(&net->ipv4.ping_group_range.lock); 1859 /* 1860 * Sane defaults - nobody may create ping sockets. 1861 * Boot scripts should set this to distro-specific group. 1862 */ 1863 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1); 1864 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0); 1865 1866 /* Default values for sysctl-controlled parameters. 1867 * We set them here, in case sysctl is not compiled. 1868 */ 1869 net->ipv4.sysctl_ip_default_ttl = IPDEFTTL; 1870 net->ipv4.sysctl_ip_fwd_update_priority = 1; 1871 net->ipv4.sysctl_ip_dynaddr = 0; 1872 net->ipv4.sysctl_ip_early_demux = 1; 1873 net->ipv4.sysctl_udp_early_demux = 1; 1874 net->ipv4.sysctl_tcp_early_demux = 1; 1875 net->ipv4.sysctl_nexthop_compat_mode = 1; 1876 #ifdef CONFIG_SYSCTL 1877 net->ipv4.sysctl_ip_prot_sock = PROT_SOCK; 1878 #endif 1879 1880 /* Some igmp sysctl, whose values are always used */ 1881 net->ipv4.sysctl_igmp_max_memberships = 20; 1882 net->ipv4.sysctl_igmp_max_msf = 10; 1883 /* IGMP reports for link-local multicast groups are enabled by default */ 1884 net->ipv4.sysctl_igmp_llm_reports = 1; 1885 net->ipv4.sysctl_igmp_qrv = 2; 1886 1887 net->ipv4.sysctl_fib_notify_on_flag_change = 0; 1888 1889 return 0; 1890 } 1891 1892 static __net_initdata struct pernet_operations af_inet_ops = { 1893 .init = inet_init_net, 1894 }; 1895 1896 static int __init init_inet_pernet_ops(void) 1897 { 1898 return register_pernet_subsys(&af_inet_ops); 1899 } 1900 1901 static int ipv4_proc_init(void); 1902 1903 /* 1904 * IP protocol layer initialiser 1905 */ 1906 1907 static struct packet_offload ip_packet_offload __read_mostly = { 1908 .type = cpu_to_be16(ETH_P_IP), 1909 .callbacks = { 1910 .gso_segment = inet_gso_segment, 1911 .gro_receive = inet_gro_receive, 1912 .gro_complete = inet_gro_complete, 1913 }, 1914 }; 1915 1916 static const struct net_offload ipip_offload = { 1917 .callbacks = { 1918 .gso_segment = ipip_gso_segment, 1919 .gro_receive = ipip_gro_receive, 1920 .gro_complete = ipip_gro_complete, 1921 }, 1922 }; 1923 1924 static int __init ipip_offload_init(void) 1925 { 1926 return inet_add_offload(&ipip_offload, IPPROTO_IPIP); 1927 } 1928 1929 static int __init ipv4_offload_init(void) 1930 { 1931 /* 1932 * Add offloads 1933 */ 1934 if (udpv4_offload_init() < 0) 1935 pr_crit("%s: Cannot add UDP protocol offload\n", __func__); 1936 if (tcpv4_offload_init() < 0) 1937 pr_crit("%s: Cannot add TCP protocol offload\n", __func__); 1938 if (ipip_offload_init() < 0) 1939 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__); 1940 1941 dev_add_offload(&ip_packet_offload); 1942 return 0; 1943 } 1944 1945 fs_initcall(ipv4_offload_init); 1946 1947 static struct packet_type ip_packet_type __read_mostly = { 1948 .type = cpu_to_be16(ETH_P_IP), 1949 .func = ip_rcv, 1950 .list_func = ip_list_rcv, 1951 }; 1952 1953 static int __init inet_init(void) 1954 { 1955 struct inet_protosw *q; 1956 struct list_head *r; 1957 int rc; 1958 1959 sock_skb_cb_check_size(sizeof(struct inet_skb_parm)); 1960 1961 raw_hashinfo_init(&raw_v4_hashinfo); 1962 1963 rc = proto_register(&tcp_prot, 1); 1964 if (rc) 1965 goto out; 1966 1967 rc = proto_register(&udp_prot, 1); 1968 if (rc) 1969 goto out_unregister_tcp_proto; 1970 1971 rc = proto_register(&raw_prot, 1); 1972 if (rc) 1973 goto out_unregister_udp_proto; 1974 1975 rc = proto_register(&ping_prot, 1); 1976 if (rc) 1977 goto out_unregister_raw_proto; 1978 1979 /* 1980 * Tell SOCKET that we are alive... 1981 */ 1982 1983 (void)sock_register(&inet_family_ops); 1984 1985 #ifdef CONFIG_SYSCTL 1986 ip_static_sysctl_init(); 1987 #endif 1988 1989 /* 1990 * Add all the base protocols. 1991 */ 1992 1993 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0) 1994 pr_crit("%s: Cannot add ICMP protocol\n", __func__); 1995 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0) 1996 pr_crit("%s: Cannot add UDP protocol\n", __func__); 1997 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0) 1998 pr_crit("%s: Cannot add TCP protocol\n", __func__); 1999 #ifdef CONFIG_IP_MULTICAST 2000 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0) 2001 pr_crit("%s: Cannot add IGMP protocol\n", __func__); 2002 #endif 2003 2004 /* Register the socket-side information for inet_create. */ 2005 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r) 2006 INIT_LIST_HEAD(r); 2007 2008 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q) 2009 inet_register_protosw(q); 2010 2011 /* 2012 * Set the ARP module up 2013 */ 2014 2015 arp_init(); 2016 2017 /* 2018 * Set the IP module up 2019 */ 2020 2021 ip_init(); 2022 2023 /* Initialise per-cpu ipv4 mibs */ 2024 if (init_ipv4_mibs()) 2025 panic("%s: Cannot init ipv4 mibs\n", __func__); 2026 2027 /* Setup TCP slab cache for open requests. */ 2028 tcp_init(); 2029 2030 /* Setup UDP memory threshold */ 2031 udp_init(); 2032 2033 /* Add UDP-Lite (RFC 3828) */ 2034 udplite4_register(); 2035 2036 raw_init(); 2037 2038 ping_init(); 2039 2040 /* 2041 * Set the ICMP layer up 2042 */ 2043 2044 if (icmp_init() < 0) 2045 panic("Failed to create the ICMP control socket.\n"); 2046 2047 /* 2048 * Initialise the multicast router 2049 */ 2050 #if defined(CONFIG_IP_MROUTE) 2051 if (ip_mr_init()) 2052 pr_crit("%s: Cannot init ipv4 mroute\n", __func__); 2053 #endif 2054 2055 if (init_inet_pernet_ops()) 2056 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__); 2057 2058 ipv4_proc_init(); 2059 2060 ipfrag_init(); 2061 2062 dev_add_pack(&ip_packet_type); 2063 2064 ip_tunnel_core_init(); 2065 2066 rc = 0; 2067 out: 2068 return rc; 2069 out_unregister_raw_proto: 2070 proto_unregister(&raw_prot); 2071 out_unregister_udp_proto: 2072 proto_unregister(&udp_prot); 2073 out_unregister_tcp_proto: 2074 proto_unregister(&tcp_prot); 2075 goto out; 2076 } 2077 2078 fs_initcall(inet_init); 2079 2080 /* ------------------------------------------------------------------------ */ 2081 2082 #ifdef CONFIG_PROC_FS 2083 static int __init ipv4_proc_init(void) 2084 { 2085 int rc = 0; 2086 2087 if (raw_proc_init()) 2088 goto out_raw; 2089 if (tcp4_proc_init()) 2090 goto out_tcp; 2091 if (udp4_proc_init()) 2092 goto out_udp; 2093 if (ping_proc_init()) 2094 goto out_ping; 2095 if (ip_misc_proc_init()) 2096 goto out_misc; 2097 out: 2098 return rc; 2099 out_misc: 2100 ping_proc_exit(); 2101 out_ping: 2102 udp4_proc_exit(); 2103 out_udp: 2104 tcp4_proc_exit(); 2105 out_tcp: 2106 raw_proc_exit(); 2107 out_raw: 2108 rc = -ENOMEM; 2109 goto out; 2110 } 2111 2112 #else /* CONFIG_PROC_FS */ 2113 static int __init ipv4_proc_init(void) 2114 { 2115 return 0; 2116 } 2117 #endif /* CONFIG_PROC_FS */ 2118