1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * PF_INET protocol family socket handler. 8 * 9 * Authors: Ross Biro 10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 11 * Florian La Roche, <flla@stud.uni-sb.de> 12 * Alan Cox, <A.Cox@swansea.ac.uk> 13 * 14 * Changes (see also sock.c) 15 * 16 * piggy, 17 * Karl Knutson : Socket protocol table 18 * A.N.Kuznetsov : Socket death error in accept(). 19 * John Richardson : Fix non blocking error in connect() 20 * so sockets that fail to connect 21 * don't return -EINPROGRESS. 22 * Alan Cox : Asynchronous I/O support 23 * Alan Cox : Keep correct socket pointer on sock 24 * structures 25 * when accept() ed 26 * Alan Cox : Semantics of SO_LINGER aren't state 27 * moved to close when you look carefully. 28 * With this fixed and the accept bug fixed 29 * some RPC stuff seems happier. 30 * Niibe Yutaka : 4.4BSD style write async I/O 31 * Alan Cox, 32 * Tony Gale : Fixed reuse semantics. 33 * Alan Cox : bind() shouldn't abort existing but dead 34 * sockets. Stops FTP netin:.. I hope. 35 * Alan Cox : bind() works correctly for RAW sockets. 36 * Note that FreeBSD at least was broken 37 * in this respect so be careful with 38 * compatibility tests... 39 * Alan Cox : routing cache support 40 * Alan Cox : memzero the socket structure for 41 * compactness. 42 * Matt Day : nonblock connect error handler 43 * Alan Cox : Allow large numbers of pending sockets 44 * (eg for big web sites), but only if 45 * specifically application requested. 46 * Alan Cox : New buffering throughout IP. Used 47 * dumbly. 48 * Alan Cox : New buffering now used smartly. 49 * Alan Cox : BSD rather than common sense 50 * interpretation of listen. 51 * Germano Caronni : Assorted small races. 52 * Alan Cox : sendmsg/recvmsg basic support. 53 * Alan Cox : Only sendmsg/recvmsg now supported. 54 * Alan Cox : Locked down bind (see security list). 55 * Alan Cox : Loosened bind a little. 56 * Mike McLagan : ADD/DEL DLCI Ioctls 57 * Willy Konynenberg : Transparent proxying support. 58 * David S. Miller : New socket lookup architecture. 59 * Some other random speedups. 60 * Cyrus Durgin : Cleaned up file for kmod hacks. 61 * Andi Kleen : Fix inet_stream_connect TCP race. 62 */ 63 64 #define pr_fmt(fmt) "IPv4: " fmt 65 66 #include <linux/err.h> 67 #include <linux/errno.h> 68 #include <linux/types.h> 69 #include <linux/socket.h> 70 #include <linux/in.h> 71 #include <linux/kernel.h> 72 #include <linux/kmod.h> 73 #include <linux/sched.h> 74 #include <linux/timer.h> 75 #include <linux/string.h> 76 #include <linux/sockios.h> 77 #include <linux/net.h> 78 #include <linux/capability.h> 79 #include <linux/fcntl.h> 80 #include <linux/mm.h> 81 #include <linux/interrupt.h> 82 #include <linux/stat.h> 83 #include <linux/init.h> 84 #include <linux/poll.h> 85 #include <linux/netfilter_ipv4.h> 86 #include <linux/random.h> 87 #include <linux/slab.h> 88 89 #include <linux/uaccess.h> 90 91 #include <linux/inet.h> 92 #include <linux/igmp.h> 93 #include <linux/inetdevice.h> 94 #include <linux/netdevice.h> 95 #include <net/checksum.h> 96 #include <net/ip.h> 97 #include <net/protocol.h> 98 #include <net/arp.h> 99 #include <net/route.h> 100 #include <net/ip_fib.h> 101 #include <net/inet_connection_sock.h> 102 #include <net/gro.h> 103 #include <net/gso.h> 104 #include <net/tcp.h> 105 #include <net/udp.h> 106 #include <net/udplite.h> 107 #include <net/ping.h> 108 #include <linux/skbuff.h> 109 #include <net/sock.h> 110 #include <net/raw.h> 111 #include <net/icmp.h> 112 #include <net/inet_common.h> 113 #include <net/ip_tunnels.h> 114 #include <net/xfrm.h> 115 #include <net/net_namespace.h> 116 #include <net/secure_seq.h> 117 #ifdef CONFIG_IP_MROUTE 118 #include <linux/mroute.h> 119 #endif 120 #include <net/l3mdev.h> 121 #include <net/compat.h> 122 123 #include <trace/events/sock.h> 124 125 /* The inetsw table contains everything that inet_create needs to 126 * build a new socket. 127 */ 128 static struct list_head inetsw[SOCK_MAX]; 129 static DEFINE_SPINLOCK(inetsw_lock); 130 131 /* New destruction routine */ 132 133 void inet_sock_destruct(struct sock *sk) 134 { 135 struct inet_sock *inet = inet_sk(sk); 136 137 __skb_queue_purge(&sk->sk_receive_queue); 138 __skb_queue_purge(&sk->sk_error_queue); 139 140 sk_mem_reclaim_final(sk); 141 142 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) { 143 pr_err("Attempt to release TCP socket in state %d %p\n", 144 sk->sk_state, sk); 145 return; 146 } 147 if (!sock_flag(sk, SOCK_DEAD)) { 148 pr_err("Attempt to release alive inet socket %p\n", sk); 149 return; 150 } 151 152 WARN_ON_ONCE(atomic_read(&sk->sk_rmem_alloc)); 153 WARN_ON_ONCE(refcount_read(&sk->sk_wmem_alloc)); 154 WARN_ON_ONCE(sk->sk_wmem_queued); 155 WARN_ON_ONCE(sk_forward_alloc_get(sk)); 156 157 kfree(rcu_dereference_protected(inet->inet_opt, 1)); 158 dst_release(rcu_dereference_protected(sk->sk_dst_cache, 1)); 159 dst_release(rcu_dereference_protected(sk->sk_rx_dst, 1)); 160 } 161 EXPORT_SYMBOL(inet_sock_destruct); 162 163 /* 164 * The routines beyond this point handle the behaviour of an AF_INET 165 * socket object. Mostly it punts to the subprotocols of IP to do 166 * the work. 167 */ 168 169 /* 170 * Automatically bind an unbound socket. 171 */ 172 173 static int inet_autobind(struct sock *sk) 174 { 175 struct inet_sock *inet; 176 /* We may need to bind the socket. */ 177 lock_sock(sk); 178 inet = inet_sk(sk); 179 if (!inet->inet_num) { 180 if (sk->sk_prot->get_port(sk, 0)) { 181 release_sock(sk); 182 return -EAGAIN; 183 } 184 inet->inet_sport = htons(inet->inet_num); 185 } 186 release_sock(sk); 187 return 0; 188 } 189 190 int __inet_listen_sk(struct sock *sk, int backlog) 191 { 192 unsigned char old_state = sk->sk_state; 193 int err, tcp_fastopen; 194 195 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN))) 196 return -EINVAL; 197 198 WRITE_ONCE(sk->sk_max_ack_backlog, backlog); 199 /* Really, if the socket is already in listen state 200 * we can only allow the backlog to be adjusted. 201 */ 202 if (old_state != TCP_LISTEN) { 203 /* Enable TFO w/o requiring TCP_FASTOPEN socket option. 204 * Note that only TCP sockets (SOCK_STREAM) will reach here. 205 * Also fastopen backlog may already been set via the option 206 * because the socket was in TCP_LISTEN state previously but 207 * was shutdown() rather than close(). 208 */ 209 tcp_fastopen = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fastopen); 210 if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) && 211 (tcp_fastopen & TFO_SERVER_ENABLE) && 212 !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) { 213 fastopen_queue_tune(sk, backlog); 214 tcp_fastopen_init_key_once(sock_net(sk)); 215 } 216 217 err = inet_csk_listen_start(sk); 218 if (err) 219 return err; 220 221 tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL); 222 } 223 return 0; 224 } 225 226 /* 227 * Move a socket into listening state. 228 */ 229 int inet_listen(struct socket *sock, int backlog) 230 { 231 struct sock *sk = sock->sk; 232 int err = -EINVAL; 233 234 lock_sock(sk); 235 236 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM) 237 goto out; 238 239 err = __inet_listen_sk(sk, backlog); 240 241 out: 242 release_sock(sk); 243 return err; 244 } 245 EXPORT_SYMBOL(inet_listen); 246 247 /* 248 * Create an inet socket. 249 */ 250 251 static int inet_create(struct net *net, struct socket *sock, int protocol, 252 int kern) 253 { 254 struct sock *sk; 255 struct inet_protosw *answer; 256 struct inet_sock *inet; 257 struct proto *answer_prot; 258 unsigned char answer_flags; 259 int try_loading_module = 0; 260 int err; 261 262 if (protocol < 0 || protocol >= IPPROTO_MAX) 263 return -EINVAL; 264 265 sock->state = SS_UNCONNECTED; 266 267 /* Look for the requested type/protocol pair. */ 268 lookup_protocol: 269 err = -ESOCKTNOSUPPORT; 270 rcu_read_lock(); 271 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) { 272 273 err = 0; 274 /* Check the non-wild match. */ 275 if (protocol == answer->protocol) { 276 if (protocol != IPPROTO_IP) 277 break; 278 } else { 279 /* Check for the two wild cases. */ 280 if (IPPROTO_IP == protocol) { 281 protocol = answer->protocol; 282 break; 283 } 284 if (IPPROTO_IP == answer->protocol) 285 break; 286 } 287 err = -EPROTONOSUPPORT; 288 } 289 290 if (unlikely(err)) { 291 if (try_loading_module < 2) { 292 rcu_read_unlock(); 293 /* 294 * Be more specific, e.g. net-pf-2-proto-132-type-1 295 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM) 296 */ 297 if (++try_loading_module == 1) 298 request_module("net-pf-%d-proto-%d-type-%d", 299 PF_INET, protocol, sock->type); 300 /* 301 * Fall back to generic, e.g. net-pf-2-proto-132 302 * (net-pf-PF_INET-proto-IPPROTO_SCTP) 303 */ 304 else 305 request_module("net-pf-%d-proto-%d", 306 PF_INET, protocol); 307 goto lookup_protocol; 308 } else 309 goto out_rcu_unlock; 310 } 311 312 err = -EPERM; 313 if (sock->type == SOCK_RAW && !kern && 314 !ns_capable(net->user_ns, CAP_NET_RAW)) 315 goto out_rcu_unlock; 316 317 sock->ops = answer->ops; 318 answer_prot = answer->prot; 319 answer_flags = answer->flags; 320 rcu_read_unlock(); 321 322 WARN_ON(!answer_prot->slab); 323 324 err = -ENOMEM; 325 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern); 326 if (!sk) 327 goto out; 328 329 err = 0; 330 if (INET_PROTOSW_REUSE & answer_flags) 331 sk->sk_reuse = SK_CAN_REUSE; 332 333 inet = inet_sk(sk); 334 inet_assign_bit(IS_ICSK, sk, INET_PROTOSW_ICSK & answer_flags); 335 336 inet_clear_bit(NODEFRAG, sk); 337 338 if (SOCK_RAW == sock->type) { 339 inet->inet_num = protocol; 340 if (IPPROTO_RAW == protocol) 341 inet_set_bit(HDRINCL, sk); 342 } 343 344 if (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc)) 345 inet->pmtudisc = IP_PMTUDISC_DONT; 346 else 347 inet->pmtudisc = IP_PMTUDISC_WANT; 348 349 atomic_set(&inet->inet_id, 0); 350 351 sock_init_data(sock, sk); 352 353 sk->sk_destruct = inet_sock_destruct; 354 sk->sk_protocol = protocol; 355 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv; 356 sk->sk_txrehash = READ_ONCE(net->core.sysctl_txrehash); 357 358 inet->uc_ttl = -1; 359 inet_set_bit(MC_LOOP, sk); 360 inet->mc_ttl = 1; 361 inet_set_bit(MC_ALL, sk); 362 inet->mc_index = 0; 363 inet->mc_list = NULL; 364 inet->rcv_tos = 0; 365 366 if (inet->inet_num) { 367 /* It assumes that any protocol which allows 368 * the user to assign a number at socket 369 * creation time automatically 370 * shares. 371 */ 372 inet->inet_sport = htons(inet->inet_num); 373 /* Add to protocol hash chains. */ 374 err = sk->sk_prot->hash(sk); 375 if (err) { 376 sk_common_release(sk); 377 goto out; 378 } 379 } 380 381 if (sk->sk_prot->init) { 382 err = sk->sk_prot->init(sk); 383 if (err) { 384 sk_common_release(sk); 385 goto out; 386 } 387 } 388 389 if (!kern) { 390 err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk); 391 if (err) { 392 sk_common_release(sk); 393 goto out; 394 } 395 } 396 out: 397 return err; 398 out_rcu_unlock: 399 rcu_read_unlock(); 400 goto out; 401 } 402 403 404 /* 405 * The peer socket should always be NULL (or else). When we call this 406 * function we are destroying the object and from then on nobody 407 * should refer to it. 408 */ 409 int inet_release(struct socket *sock) 410 { 411 struct sock *sk = sock->sk; 412 413 if (sk) { 414 long timeout; 415 416 if (!sk->sk_kern_sock) 417 BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk); 418 419 /* Applications forget to leave groups before exiting */ 420 ip_mc_drop_socket(sk); 421 422 /* If linger is set, we don't return until the close 423 * is complete. Otherwise we return immediately. The 424 * actually closing is done the same either way. 425 * 426 * If the close is due to the process exiting, we never 427 * linger.. 428 */ 429 timeout = 0; 430 if (sock_flag(sk, SOCK_LINGER) && 431 !(current->flags & PF_EXITING)) 432 timeout = sk->sk_lingertime; 433 sk->sk_prot->close(sk, timeout); 434 sock->sk = NULL; 435 } 436 return 0; 437 } 438 EXPORT_SYMBOL(inet_release); 439 440 int inet_bind_sk(struct sock *sk, struct sockaddr *uaddr, int addr_len) 441 { 442 u32 flags = BIND_WITH_LOCK; 443 int err; 444 445 /* If the socket has its own bind function then use it. (RAW) */ 446 if (sk->sk_prot->bind) { 447 return sk->sk_prot->bind(sk, uaddr, addr_len); 448 } 449 if (addr_len < sizeof(struct sockaddr_in)) 450 return -EINVAL; 451 452 /* BPF prog is run before any checks are done so that if the prog 453 * changes context in a wrong way it will be caught. 454 */ 455 err = BPF_CGROUP_RUN_PROG_INET_BIND_LOCK(sk, uaddr, &addr_len, 456 CGROUP_INET4_BIND, &flags); 457 if (err) 458 return err; 459 460 return __inet_bind(sk, uaddr, addr_len, flags); 461 } 462 463 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) 464 { 465 return inet_bind_sk(sock->sk, uaddr, addr_len); 466 } 467 EXPORT_SYMBOL(inet_bind); 468 469 int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len, 470 u32 flags) 471 { 472 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr; 473 struct inet_sock *inet = inet_sk(sk); 474 struct net *net = sock_net(sk); 475 unsigned short snum; 476 int chk_addr_ret; 477 u32 tb_id = RT_TABLE_LOCAL; 478 int err; 479 480 if (addr->sin_family != AF_INET) { 481 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET) 482 * only if s_addr is INADDR_ANY. 483 */ 484 err = -EAFNOSUPPORT; 485 if (addr->sin_family != AF_UNSPEC || 486 addr->sin_addr.s_addr != htonl(INADDR_ANY)) 487 goto out; 488 } 489 490 tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id; 491 chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id); 492 493 /* Not specified by any standard per-se, however it breaks too 494 * many applications when removed. It is unfortunate since 495 * allowing applications to make a non-local bind solves 496 * several problems with systems using dynamic addressing. 497 * (ie. your servers still start up even if your ISDN link 498 * is temporarily down) 499 */ 500 err = -EADDRNOTAVAIL; 501 if (!inet_addr_valid_or_nonlocal(net, inet, addr->sin_addr.s_addr, 502 chk_addr_ret)) 503 goto out; 504 505 snum = ntohs(addr->sin_port); 506 err = -EACCES; 507 if (!(flags & BIND_NO_CAP_NET_BIND_SERVICE) && 508 snum && inet_port_requires_bind_service(net, snum) && 509 !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE)) 510 goto out; 511 512 /* We keep a pair of addresses. rcv_saddr is the one 513 * used by hash lookups, and saddr is used for transmit. 514 * 515 * In the BSD API these are the same except where it 516 * would be illegal to use them (multicast/broadcast) in 517 * which case the sending device address is used. 518 */ 519 if (flags & BIND_WITH_LOCK) 520 lock_sock(sk); 521 522 /* Check these errors (active socket, double bind). */ 523 err = -EINVAL; 524 if (sk->sk_state != TCP_CLOSE || inet->inet_num) 525 goto out_release_sock; 526 527 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr; 528 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST) 529 inet->inet_saddr = 0; /* Use device */ 530 531 /* Make sure we are allowed to bind here. */ 532 if (snum || !(inet_test_bit(BIND_ADDRESS_NO_PORT, sk) || 533 (flags & BIND_FORCE_ADDRESS_NO_PORT))) { 534 err = sk->sk_prot->get_port(sk, snum); 535 if (err) { 536 inet->inet_saddr = inet->inet_rcv_saddr = 0; 537 goto out_release_sock; 538 } 539 if (!(flags & BIND_FROM_BPF)) { 540 err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk); 541 if (err) { 542 inet->inet_saddr = inet->inet_rcv_saddr = 0; 543 if (sk->sk_prot->put_port) 544 sk->sk_prot->put_port(sk); 545 goto out_release_sock; 546 } 547 } 548 } 549 550 if (inet->inet_rcv_saddr) 551 sk->sk_userlocks |= SOCK_BINDADDR_LOCK; 552 if (snum) 553 sk->sk_userlocks |= SOCK_BINDPORT_LOCK; 554 inet->inet_sport = htons(inet->inet_num); 555 inet->inet_daddr = 0; 556 inet->inet_dport = 0; 557 sk_dst_reset(sk); 558 err = 0; 559 out_release_sock: 560 if (flags & BIND_WITH_LOCK) 561 release_sock(sk); 562 out: 563 return err; 564 } 565 566 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr, 567 int addr_len, int flags) 568 { 569 struct sock *sk = sock->sk; 570 const struct proto *prot; 571 int err; 572 573 if (addr_len < sizeof(uaddr->sa_family)) 574 return -EINVAL; 575 576 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 577 prot = READ_ONCE(sk->sk_prot); 578 579 if (uaddr->sa_family == AF_UNSPEC) 580 return prot->disconnect(sk, flags); 581 582 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) { 583 err = prot->pre_connect(sk, uaddr, addr_len); 584 if (err) 585 return err; 586 } 587 588 if (data_race(!inet_sk(sk)->inet_num) && inet_autobind(sk)) 589 return -EAGAIN; 590 return prot->connect(sk, uaddr, addr_len); 591 } 592 EXPORT_SYMBOL(inet_dgram_connect); 593 594 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias) 595 { 596 DEFINE_WAIT_FUNC(wait, woken_wake_function); 597 598 add_wait_queue(sk_sleep(sk), &wait); 599 sk->sk_write_pending += writebias; 600 sk->sk_wait_pending++; 601 602 /* Basic assumption: if someone sets sk->sk_err, he _must_ 603 * change state of the socket from TCP_SYN_*. 604 * Connect() does not allow to get error notifications 605 * without closing the socket. 606 */ 607 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 608 release_sock(sk); 609 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo); 610 lock_sock(sk); 611 if (signal_pending(current) || !timeo) 612 break; 613 } 614 remove_wait_queue(sk_sleep(sk), &wait); 615 sk->sk_write_pending -= writebias; 616 sk->sk_wait_pending--; 617 return timeo; 618 } 619 620 /* 621 * Connect to a remote host. There is regrettably still a little 622 * TCP 'magic' in here. 623 */ 624 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr, 625 int addr_len, int flags, int is_sendmsg) 626 { 627 struct sock *sk = sock->sk; 628 int err; 629 long timeo; 630 631 /* 632 * uaddr can be NULL and addr_len can be 0 if: 633 * sk is a TCP fastopen active socket and 634 * TCP_FASTOPEN_CONNECT sockopt is set and 635 * we already have a valid cookie for this socket. 636 * In this case, user can call write() after connect(). 637 * write() will invoke tcp_sendmsg_fastopen() which calls 638 * __inet_stream_connect(). 639 */ 640 if (uaddr) { 641 if (addr_len < sizeof(uaddr->sa_family)) 642 return -EINVAL; 643 644 if (uaddr->sa_family == AF_UNSPEC) { 645 err = sk->sk_prot->disconnect(sk, flags); 646 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED; 647 goto out; 648 } 649 } 650 651 switch (sock->state) { 652 default: 653 err = -EINVAL; 654 goto out; 655 case SS_CONNECTED: 656 err = -EISCONN; 657 goto out; 658 case SS_CONNECTING: 659 if (inet_test_bit(DEFER_CONNECT, sk)) 660 err = is_sendmsg ? -EINPROGRESS : -EISCONN; 661 else 662 err = -EALREADY; 663 /* Fall out of switch with err, set for this state */ 664 break; 665 case SS_UNCONNECTED: 666 err = -EISCONN; 667 if (sk->sk_state != TCP_CLOSE) 668 goto out; 669 670 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) { 671 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len); 672 if (err) 673 goto out; 674 } 675 676 err = sk->sk_prot->connect(sk, uaddr, addr_len); 677 if (err < 0) 678 goto out; 679 680 sock->state = SS_CONNECTING; 681 682 if (!err && inet_test_bit(DEFER_CONNECT, sk)) 683 goto out; 684 685 /* Just entered SS_CONNECTING state; the only 686 * difference is that return value in non-blocking 687 * case is EINPROGRESS, rather than EALREADY. 688 */ 689 err = -EINPROGRESS; 690 break; 691 } 692 693 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK); 694 695 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 696 int writebias = (sk->sk_protocol == IPPROTO_TCP) && 697 tcp_sk(sk)->fastopen_req && 698 tcp_sk(sk)->fastopen_req->data ? 1 : 0; 699 700 /* Error code is set above */ 701 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias)) 702 goto out; 703 704 err = sock_intr_errno(timeo); 705 if (signal_pending(current)) 706 goto out; 707 } 708 709 /* Connection was closed by RST, timeout, ICMP error 710 * or another process disconnected us. 711 */ 712 if (sk->sk_state == TCP_CLOSE) 713 goto sock_error; 714 715 /* sk->sk_err may be not zero now, if RECVERR was ordered by user 716 * and error was received after socket entered established state. 717 * Hence, it is handled normally after connect() return successfully. 718 */ 719 720 sock->state = SS_CONNECTED; 721 err = 0; 722 out: 723 return err; 724 725 sock_error: 726 err = sock_error(sk) ? : -ECONNABORTED; 727 sock->state = SS_UNCONNECTED; 728 if (sk->sk_prot->disconnect(sk, flags)) 729 sock->state = SS_DISCONNECTING; 730 goto out; 731 } 732 EXPORT_SYMBOL(__inet_stream_connect); 733 734 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr, 735 int addr_len, int flags) 736 { 737 int err; 738 739 lock_sock(sock->sk); 740 err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0); 741 release_sock(sock->sk); 742 return err; 743 } 744 EXPORT_SYMBOL(inet_stream_connect); 745 746 void __inet_accept(struct socket *sock, struct socket *newsock, struct sock *newsk) 747 { 748 sock_rps_record_flow(newsk); 749 WARN_ON(!((1 << newsk->sk_state) & 750 (TCPF_ESTABLISHED | TCPF_SYN_RECV | 751 TCPF_CLOSE_WAIT | TCPF_CLOSE))); 752 753 if (test_bit(SOCK_SUPPORT_ZC, &sock->flags)) 754 set_bit(SOCK_SUPPORT_ZC, &newsock->flags); 755 sock_graft(newsk, newsock); 756 757 newsock->state = SS_CONNECTED; 758 } 759 760 /* 761 * Accept a pending connection. The TCP layer now gives BSD semantics. 762 */ 763 764 int inet_accept(struct socket *sock, struct socket *newsock, int flags, 765 bool kern) 766 { 767 struct sock *sk1 = sock->sk, *sk2; 768 int err = -EINVAL; 769 770 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 771 sk2 = READ_ONCE(sk1->sk_prot)->accept(sk1, flags, &err, kern); 772 if (!sk2) 773 return err; 774 775 lock_sock(sk2); 776 __inet_accept(sock, newsock, sk2); 777 release_sock(sk2); 778 return 0; 779 } 780 EXPORT_SYMBOL(inet_accept); 781 782 /* 783 * This does both peername and sockname. 784 */ 785 int inet_getname(struct socket *sock, struct sockaddr *uaddr, 786 int peer) 787 { 788 struct sock *sk = sock->sk; 789 struct inet_sock *inet = inet_sk(sk); 790 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr); 791 int sin_addr_len = sizeof(*sin); 792 793 sin->sin_family = AF_INET; 794 lock_sock(sk); 795 if (peer) { 796 if (!inet->inet_dport || 797 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) && 798 peer == 1)) { 799 release_sock(sk); 800 return -ENOTCONN; 801 } 802 sin->sin_port = inet->inet_dport; 803 sin->sin_addr.s_addr = inet->inet_daddr; 804 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin, &sin_addr_len, 805 CGROUP_INET4_GETPEERNAME); 806 } else { 807 __be32 addr = inet->inet_rcv_saddr; 808 if (!addr) 809 addr = inet->inet_saddr; 810 sin->sin_port = inet->inet_sport; 811 sin->sin_addr.s_addr = addr; 812 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin, &sin_addr_len, 813 CGROUP_INET4_GETSOCKNAME); 814 } 815 release_sock(sk); 816 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 817 return sin_addr_len; 818 } 819 EXPORT_SYMBOL(inet_getname); 820 821 int inet_send_prepare(struct sock *sk) 822 { 823 sock_rps_record_flow(sk); 824 825 /* We may need to bind the socket. */ 826 if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind && 827 inet_autobind(sk)) 828 return -EAGAIN; 829 830 return 0; 831 } 832 EXPORT_SYMBOL_GPL(inet_send_prepare); 833 834 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size) 835 { 836 struct sock *sk = sock->sk; 837 838 if (unlikely(inet_send_prepare(sk))) 839 return -EAGAIN; 840 841 return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg, 842 sk, msg, size); 843 } 844 EXPORT_SYMBOL(inet_sendmsg); 845 846 void inet_splice_eof(struct socket *sock) 847 { 848 const struct proto *prot; 849 struct sock *sk = sock->sk; 850 851 if (unlikely(inet_send_prepare(sk))) 852 return; 853 854 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 855 prot = READ_ONCE(sk->sk_prot); 856 if (prot->splice_eof) 857 prot->splice_eof(sock); 858 } 859 EXPORT_SYMBOL_GPL(inet_splice_eof); 860 861 INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *, 862 size_t, int, int *)); 863 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 864 int flags) 865 { 866 struct sock *sk = sock->sk; 867 int addr_len = 0; 868 int err; 869 870 if (likely(!(flags & MSG_ERRQUEUE))) 871 sock_rps_record_flow(sk); 872 873 err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg, 874 sk, msg, size, flags, &addr_len); 875 if (err >= 0) 876 msg->msg_namelen = addr_len; 877 return err; 878 } 879 EXPORT_SYMBOL(inet_recvmsg); 880 881 int inet_shutdown(struct socket *sock, int how) 882 { 883 struct sock *sk = sock->sk; 884 int err = 0; 885 886 /* This should really check to make sure 887 * the socket is a TCP socket. (WHY AC...) 888 */ 889 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and 890 1->2 bit 2 snds. 891 2->3 */ 892 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */ 893 return -EINVAL; 894 895 lock_sock(sk); 896 if (sock->state == SS_CONNECTING) { 897 if ((1 << sk->sk_state) & 898 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE)) 899 sock->state = SS_DISCONNECTING; 900 else 901 sock->state = SS_CONNECTED; 902 } 903 904 switch (sk->sk_state) { 905 case TCP_CLOSE: 906 err = -ENOTCONN; 907 /* Hack to wake up other listeners, who can poll for 908 EPOLLHUP, even on eg. unconnected UDP sockets -- RR */ 909 fallthrough; 910 default: 911 WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | how); 912 if (sk->sk_prot->shutdown) 913 sk->sk_prot->shutdown(sk, how); 914 break; 915 916 /* Remaining two branches are temporary solution for missing 917 * close() in multithreaded environment. It is _not_ a good idea, 918 * but we have no choice until close() is repaired at VFS level. 919 */ 920 case TCP_LISTEN: 921 if (!(how & RCV_SHUTDOWN)) 922 break; 923 fallthrough; 924 case TCP_SYN_SENT: 925 err = sk->sk_prot->disconnect(sk, O_NONBLOCK); 926 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED; 927 break; 928 } 929 930 /* Wake up anyone sleeping in poll. */ 931 sk->sk_state_change(sk); 932 release_sock(sk); 933 return err; 934 } 935 EXPORT_SYMBOL(inet_shutdown); 936 937 /* 938 * ioctl() calls you can issue on an INET socket. Most of these are 939 * device configuration and stuff and very rarely used. Some ioctls 940 * pass on to the socket itself. 941 * 942 * NOTE: I like the idea of a module for the config stuff. ie ifconfig 943 * loads the devconfigure module does its configuring and unloads it. 944 * There's a good 20K of config code hanging around the kernel. 945 */ 946 947 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 948 { 949 struct sock *sk = sock->sk; 950 int err = 0; 951 struct net *net = sock_net(sk); 952 void __user *p = (void __user *)arg; 953 struct ifreq ifr; 954 struct rtentry rt; 955 956 switch (cmd) { 957 case SIOCADDRT: 958 case SIOCDELRT: 959 if (copy_from_user(&rt, p, sizeof(struct rtentry))) 960 return -EFAULT; 961 err = ip_rt_ioctl(net, cmd, &rt); 962 break; 963 case SIOCRTMSG: 964 err = -EINVAL; 965 break; 966 case SIOCDARP: 967 case SIOCGARP: 968 case SIOCSARP: 969 err = arp_ioctl(net, cmd, (void __user *)arg); 970 break; 971 case SIOCGIFADDR: 972 case SIOCGIFBRDADDR: 973 case SIOCGIFNETMASK: 974 case SIOCGIFDSTADDR: 975 case SIOCGIFPFLAGS: 976 if (get_user_ifreq(&ifr, NULL, p)) 977 return -EFAULT; 978 err = devinet_ioctl(net, cmd, &ifr); 979 if (!err && put_user_ifreq(&ifr, p)) 980 err = -EFAULT; 981 break; 982 983 case SIOCSIFADDR: 984 case SIOCSIFBRDADDR: 985 case SIOCSIFNETMASK: 986 case SIOCSIFDSTADDR: 987 case SIOCSIFPFLAGS: 988 case SIOCSIFFLAGS: 989 if (get_user_ifreq(&ifr, NULL, p)) 990 return -EFAULT; 991 err = devinet_ioctl(net, cmd, &ifr); 992 break; 993 default: 994 if (sk->sk_prot->ioctl) 995 err = sk_ioctl(sk, cmd, (void __user *)arg); 996 else 997 err = -ENOIOCTLCMD; 998 break; 999 } 1000 return err; 1001 } 1002 EXPORT_SYMBOL(inet_ioctl); 1003 1004 #ifdef CONFIG_COMPAT 1005 static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd, 1006 struct compat_rtentry __user *ur) 1007 { 1008 compat_uptr_t rtdev; 1009 struct rtentry rt; 1010 1011 if (copy_from_user(&rt.rt_dst, &ur->rt_dst, 1012 3 * sizeof(struct sockaddr)) || 1013 get_user(rt.rt_flags, &ur->rt_flags) || 1014 get_user(rt.rt_metric, &ur->rt_metric) || 1015 get_user(rt.rt_mtu, &ur->rt_mtu) || 1016 get_user(rt.rt_window, &ur->rt_window) || 1017 get_user(rt.rt_irtt, &ur->rt_irtt) || 1018 get_user(rtdev, &ur->rt_dev)) 1019 return -EFAULT; 1020 1021 rt.rt_dev = compat_ptr(rtdev); 1022 return ip_rt_ioctl(sock_net(sk), cmd, &rt); 1023 } 1024 1025 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 1026 { 1027 void __user *argp = compat_ptr(arg); 1028 struct sock *sk = sock->sk; 1029 1030 switch (cmd) { 1031 case SIOCADDRT: 1032 case SIOCDELRT: 1033 return inet_compat_routing_ioctl(sk, cmd, argp); 1034 default: 1035 if (!sk->sk_prot->compat_ioctl) 1036 return -ENOIOCTLCMD; 1037 return sk->sk_prot->compat_ioctl(sk, cmd, arg); 1038 } 1039 } 1040 #endif /* CONFIG_COMPAT */ 1041 1042 const struct proto_ops inet_stream_ops = { 1043 .family = PF_INET, 1044 .owner = THIS_MODULE, 1045 .release = inet_release, 1046 .bind = inet_bind, 1047 .connect = inet_stream_connect, 1048 .socketpair = sock_no_socketpair, 1049 .accept = inet_accept, 1050 .getname = inet_getname, 1051 .poll = tcp_poll, 1052 .ioctl = inet_ioctl, 1053 .gettstamp = sock_gettstamp, 1054 .listen = inet_listen, 1055 .shutdown = inet_shutdown, 1056 .setsockopt = sock_common_setsockopt, 1057 .getsockopt = sock_common_getsockopt, 1058 .sendmsg = inet_sendmsg, 1059 .recvmsg = inet_recvmsg, 1060 #ifdef CONFIG_MMU 1061 .mmap = tcp_mmap, 1062 #endif 1063 .splice_eof = inet_splice_eof, 1064 .splice_read = tcp_splice_read, 1065 .read_sock = tcp_read_sock, 1066 .read_skb = tcp_read_skb, 1067 .sendmsg_locked = tcp_sendmsg_locked, 1068 .peek_len = tcp_peek_len, 1069 #ifdef CONFIG_COMPAT 1070 .compat_ioctl = inet_compat_ioctl, 1071 #endif 1072 .set_rcvlowat = tcp_set_rcvlowat, 1073 }; 1074 EXPORT_SYMBOL(inet_stream_ops); 1075 1076 const struct proto_ops inet_dgram_ops = { 1077 .family = PF_INET, 1078 .owner = THIS_MODULE, 1079 .release = inet_release, 1080 .bind = inet_bind, 1081 .connect = inet_dgram_connect, 1082 .socketpair = sock_no_socketpair, 1083 .accept = sock_no_accept, 1084 .getname = inet_getname, 1085 .poll = udp_poll, 1086 .ioctl = inet_ioctl, 1087 .gettstamp = sock_gettstamp, 1088 .listen = sock_no_listen, 1089 .shutdown = inet_shutdown, 1090 .setsockopt = sock_common_setsockopt, 1091 .getsockopt = sock_common_getsockopt, 1092 .sendmsg = inet_sendmsg, 1093 .read_skb = udp_read_skb, 1094 .recvmsg = inet_recvmsg, 1095 .mmap = sock_no_mmap, 1096 .splice_eof = inet_splice_eof, 1097 .set_peek_off = sk_set_peek_off, 1098 #ifdef CONFIG_COMPAT 1099 .compat_ioctl = inet_compat_ioctl, 1100 #endif 1101 }; 1102 EXPORT_SYMBOL(inet_dgram_ops); 1103 1104 /* 1105 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without 1106 * udp_poll 1107 */ 1108 static const struct proto_ops inet_sockraw_ops = { 1109 .family = PF_INET, 1110 .owner = THIS_MODULE, 1111 .release = inet_release, 1112 .bind = inet_bind, 1113 .connect = inet_dgram_connect, 1114 .socketpair = sock_no_socketpair, 1115 .accept = sock_no_accept, 1116 .getname = inet_getname, 1117 .poll = datagram_poll, 1118 .ioctl = inet_ioctl, 1119 .gettstamp = sock_gettstamp, 1120 .listen = sock_no_listen, 1121 .shutdown = inet_shutdown, 1122 .setsockopt = sock_common_setsockopt, 1123 .getsockopt = sock_common_getsockopt, 1124 .sendmsg = inet_sendmsg, 1125 .recvmsg = inet_recvmsg, 1126 .mmap = sock_no_mmap, 1127 .splice_eof = inet_splice_eof, 1128 #ifdef CONFIG_COMPAT 1129 .compat_ioctl = inet_compat_ioctl, 1130 #endif 1131 }; 1132 1133 static const struct net_proto_family inet_family_ops = { 1134 .family = PF_INET, 1135 .create = inet_create, 1136 .owner = THIS_MODULE, 1137 }; 1138 1139 /* Upon startup we insert all the elements in inetsw_array[] into 1140 * the linked list inetsw. 1141 */ 1142 static struct inet_protosw inetsw_array[] = 1143 { 1144 { 1145 .type = SOCK_STREAM, 1146 .protocol = IPPROTO_TCP, 1147 .prot = &tcp_prot, 1148 .ops = &inet_stream_ops, 1149 .flags = INET_PROTOSW_PERMANENT | 1150 INET_PROTOSW_ICSK, 1151 }, 1152 1153 { 1154 .type = SOCK_DGRAM, 1155 .protocol = IPPROTO_UDP, 1156 .prot = &udp_prot, 1157 .ops = &inet_dgram_ops, 1158 .flags = INET_PROTOSW_PERMANENT, 1159 }, 1160 1161 { 1162 .type = SOCK_DGRAM, 1163 .protocol = IPPROTO_ICMP, 1164 .prot = &ping_prot, 1165 .ops = &inet_sockraw_ops, 1166 .flags = INET_PROTOSW_REUSE, 1167 }, 1168 1169 { 1170 .type = SOCK_RAW, 1171 .protocol = IPPROTO_IP, /* wild card */ 1172 .prot = &raw_prot, 1173 .ops = &inet_sockraw_ops, 1174 .flags = INET_PROTOSW_REUSE, 1175 } 1176 }; 1177 1178 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array) 1179 1180 void inet_register_protosw(struct inet_protosw *p) 1181 { 1182 struct list_head *lh; 1183 struct inet_protosw *answer; 1184 int protocol = p->protocol; 1185 struct list_head *last_perm; 1186 1187 spin_lock_bh(&inetsw_lock); 1188 1189 if (p->type >= SOCK_MAX) 1190 goto out_illegal; 1191 1192 /* If we are trying to override a permanent protocol, bail. */ 1193 last_perm = &inetsw[p->type]; 1194 list_for_each(lh, &inetsw[p->type]) { 1195 answer = list_entry(lh, struct inet_protosw, list); 1196 /* Check only the non-wild match. */ 1197 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0) 1198 break; 1199 if (protocol == answer->protocol) 1200 goto out_permanent; 1201 last_perm = lh; 1202 } 1203 1204 /* Add the new entry after the last permanent entry if any, so that 1205 * the new entry does not override a permanent entry when matched with 1206 * a wild-card protocol. But it is allowed to override any existing 1207 * non-permanent entry. This means that when we remove this entry, the 1208 * system automatically returns to the old behavior. 1209 */ 1210 list_add_rcu(&p->list, last_perm); 1211 out: 1212 spin_unlock_bh(&inetsw_lock); 1213 1214 return; 1215 1216 out_permanent: 1217 pr_err("Attempt to override permanent protocol %d\n", protocol); 1218 goto out; 1219 1220 out_illegal: 1221 pr_err("Ignoring attempt to register invalid socket type %d\n", 1222 p->type); 1223 goto out; 1224 } 1225 EXPORT_SYMBOL(inet_register_protosw); 1226 1227 void inet_unregister_protosw(struct inet_protosw *p) 1228 { 1229 if (INET_PROTOSW_PERMANENT & p->flags) { 1230 pr_err("Attempt to unregister permanent protocol %d\n", 1231 p->protocol); 1232 } else { 1233 spin_lock_bh(&inetsw_lock); 1234 list_del_rcu(&p->list); 1235 spin_unlock_bh(&inetsw_lock); 1236 1237 synchronize_net(); 1238 } 1239 } 1240 EXPORT_SYMBOL(inet_unregister_protosw); 1241 1242 static int inet_sk_reselect_saddr(struct sock *sk) 1243 { 1244 struct inet_sock *inet = inet_sk(sk); 1245 __be32 old_saddr = inet->inet_saddr; 1246 __be32 daddr = inet->inet_daddr; 1247 struct flowi4 *fl4; 1248 struct rtable *rt; 1249 __be32 new_saddr; 1250 struct ip_options_rcu *inet_opt; 1251 int err; 1252 1253 inet_opt = rcu_dereference_protected(inet->inet_opt, 1254 lockdep_sock_is_held(sk)); 1255 if (inet_opt && inet_opt->opt.srr) 1256 daddr = inet_opt->opt.faddr; 1257 1258 /* Query new route. */ 1259 fl4 = &inet->cork.fl.u.ip4; 1260 rt = ip_route_connect(fl4, daddr, 0, sk->sk_bound_dev_if, 1261 sk->sk_protocol, inet->inet_sport, 1262 inet->inet_dport, sk); 1263 if (IS_ERR(rt)) 1264 return PTR_ERR(rt); 1265 1266 new_saddr = fl4->saddr; 1267 1268 if (new_saddr == old_saddr) { 1269 sk_setup_caps(sk, &rt->dst); 1270 return 0; 1271 } 1272 1273 err = inet_bhash2_update_saddr(sk, &new_saddr, AF_INET); 1274 if (err) { 1275 ip_rt_put(rt); 1276 return err; 1277 } 1278 1279 sk_setup_caps(sk, &rt->dst); 1280 1281 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) > 1) { 1282 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n", 1283 __func__, &old_saddr, &new_saddr); 1284 } 1285 1286 /* 1287 * XXX The only one ugly spot where we need to 1288 * XXX really change the sockets identity after 1289 * XXX it has entered the hashes. -DaveM 1290 * 1291 * Besides that, it does not check for connection 1292 * uniqueness. Wait for troubles. 1293 */ 1294 return __sk_prot_rehash(sk); 1295 } 1296 1297 int inet_sk_rebuild_header(struct sock *sk) 1298 { 1299 struct inet_sock *inet = inet_sk(sk); 1300 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0); 1301 __be32 daddr; 1302 struct ip_options_rcu *inet_opt; 1303 struct flowi4 *fl4; 1304 int err; 1305 1306 /* Route is OK, nothing to do. */ 1307 if (rt) 1308 return 0; 1309 1310 /* Reroute. */ 1311 rcu_read_lock(); 1312 inet_opt = rcu_dereference(inet->inet_opt); 1313 daddr = inet->inet_daddr; 1314 if (inet_opt && inet_opt->opt.srr) 1315 daddr = inet_opt->opt.faddr; 1316 rcu_read_unlock(); 1317 fl4 = &inet->cork.fl.u.ip4; 1318 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr, 1319 inet->inet_dport, inet->inet_sport, 1320 sk->sk_protocol, RT_CONN_FLAGS(sk), 1321 sk->sk_bound_dev_if); 1322 if (!IS_ERR(rt)) { 1323 err = 0; 1324 sk_setup_caps(sk, &rt->dst); 1325 } else { 1326 err = PTR_ERR(rt); 1327 1328 /* Routing failed... */ 1329 sk->sk_route_caps = 0; 1330 /* 1331 * Other protocols have to map its equivalent state to TCP_SYN_SENT. 1332 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme 1333 */ 1334 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) || 1335 sk->sk_state != TCP_SYN_SENT || 1336 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) || 1337 (err = inet_sk_reselect_saddr(sk)) != 0) 1338 WRITE_ONCE(sk->sk_err_soft, -err); 1339 } 1340 1341 return err; 1342 } 1343 EXPORT_SYMBOL(inet_sk_rebuild_header); 1344 1345 void inet_sk_set_state(struct sock *sk, int state) 1346 { 1347 trace_inet_sock_set_state(sk, sk->sk_state, state); 1348 sk->sk_state = state; 1349 } 1350 EXPORT_SYMBOL(inet_sk_set_state); 1351 1352 void inet_sk_state_store(struct sock *sk, int newstate) 1353 { 1354 trace_inet_sock_set_state(sk, sk->sk_state, newstate); 1355 smp_store_release(&sk->sk_state, newstate); 1356 } 1357 1358 struct sk_buff *inet_gso_segment(struct sk_buff *skb, 1359 netdev_features_t features) 1360 { 1361 bool udpfrag = false, fixedid = false, gso_partial, encap; 1362 struct sk_buff *segs = ERR_PTR(-EINVAL); 1363 const struct net_offload *ops; 1364 unsigned int offset = 0; 1365 struct iphdr *iph; 1366 int proto, tot_len; 1367 int nhoff; 1368 int ihl; 1369 int id; 1370 1371 skb_reset_network_header(skb); 1372 nhoff = skb_network_header(skb) - skb_mac_header(skb); 1373 if (unlikely(!pskb_may_pull(skb, sizeof(*iph)))) 1374 goto out; 1375 1376 iph = ip_hdr(skb); 1377 ihl = iph->ihl * 4; 1378 if (ihl < sizeof(*iph)) 1379 goto out; 1380 1381 id = ntohs(iph->id); 1382 proto = iph->protocol; 1383 1384 /* Warning: after this point, iph might be no longer valid */ 1385 if (unlikely(!pskb_may_pull(skb, ihl))) 1386 goto out; 1387 __skb_pull(skb, ihl); 1388 1389 encap = SKB_GSO_CB(skb)->encap_level > 0; 1390 if (encap) 1391 features &= skb->dev->hw_enc_features; 1392 SKB_GSO_CB(skb)->encap_level += ihl; 1393 1394 skb_reset_transport_header(skb); 1395 1396 segs = ERR_PTR(-EPROTONOSUPPORT); 1397 1398 if (!skb->encapsulation || encap) { 1399 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP); 1400 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID); 1401 1402 /* fixed ID is invalid if DF bit is not set */ 1403 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF))) 1404 goto out; 1405 } 1406 1407 ops = rcu_dereference(inet_offloads[proto]); 1408 if (likely(ops && ops->callbacks.gso_segment)) { 1409 segs = ops->callbacks.gso_segment(skb, features); 1410 if (!segs) 1411 skb->network_header = skb_mac_header(skb) + nhoff - skb->head; 1412 } 1413 1414 if (IS_ERR_OR_NULL(segs)) 1415 goto out; 1416 1417 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL); 1418 1419 skb = segs; 1420 do { 1421 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff); 1422 if (udpfrag) { 1423 iph->frag_off = htons(offset >> 3); 1424 if (skb->next) 1425 iph->frag_off |= htons(IP_MF); 1426 offset += skb->len - nhoff - ihl; 1427 tot_len = skb->len - nhoff; 1428 } else if (skb_is_gso(skb)) { 1429 if (!fixedid) { 1430 iph->id = htons(id); 1431 id += skb_shinfo(skb)->gso_segs; 1432 } 1433 1434 if (gso_partial) 1435 tot_len = skb_shinfo(skb)->gso_size + 1436 SKB_GSO_CB(skb)->data_offset + 1437 skb->head - (unsigned char *)iph; 1438 else 1439 tot_len = skb->len - nhoff; 1440 } else { 1441 if (!fixedid) 1442 iph->id = htons(id++); 1443 tot_len = skb->len - nhoff; 1444 } 1445 iph->tot_len = htons(tot_len); 1446 ip_send_check(iph); 1447 if (encap) 1448 skb_reset_inner_headers(skb); 1449 skb->network_header = (u8 *)iph - skb->head; 1450 skb_reset_mac_len(skb); 1451 } while ((skb = skb->next)); 1452 1453 out: 1454 return segs; 1455 } 1456 1457 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb, 1458 netdev_features_t features) 1459 { 1460 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4)) 1461 return ERR_PTR(-EINVAL); 1462 1463 return inet_gso_segment(skb, features); 1464 } 1465 1466 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb) 1467 { 1468 const struct net_offload *ops; 1469 struct sk_buff *pp = NULL; 1470 const struct iphdr *iph; 1471 struct sk_buff *p; 1472 unsigned int hlen; 1473 unsigned int off; 1474 unsigned int id; 1475 int flush = 1; 1476 int proto; 1477 1478 off = skb_gro_offset(skb); 1479 hlen = off + sizeof(*iph); 1480 iph = skb_gro_header(skb, hlen, off); 1481 if (unlikely(!iph)) 1482 goto out; 1483 1484 proto = iph->protocol; 1485 1486 ops = rcu_dereference(inet_offloads[proto]); 1487 if (!ops || !ops->callbacks.gro_receive) 1488 goto out; 1489 1490 if (*(u8 *)iph != 0x45) 1491 goto out; 1492 1493 if (ip_is_fragment(iph)) 1494 goto out; 1495 1496 if (unlikely(ip_fast_csum((u8 *)iph, 5))) 1497 goto out; 1498 1499 NAPI_GRO_CB(skb)->proto = proto; 1500 id = ntohl(*(__be32 *)&iph->id); 1501 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF)); 1502 id >>= 16; 1503 1504 list_for_each_entry(p, head, list) { 1505 struct iphdr *iph2; 1506 u16 flush_id; 1507 1508 if (!NAPI_GRO_CB(p)->same_flow) 1509 continue; 1510 1511 iph2 = (struct iphdr *)(p->data + off); 1512 /* The above works because, with the exception of the top 1513 * (inner most) layer, we only aggregate pkts with the same 1514 * hdr length so all the hdrs we'll need to verify will start 1515 * at the same offset. 1516 */ 1517 if ((iph->protocol ^ iph2->protocol) | 1518 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) | 1519 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) { 1520 NAPI_GRO_CB(p)->same_flow = 0; 1521 continue; 1522 } 1523 1524 /* All fields must match except length and checksum. */ 1525 NAPI_GRO_CB(p)->flush |= 1526 (iph->ttl ^ iph2->ttl) | 1527 (iph->tos ^ iph2->tos) | 1528 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF)); 1529 1530 NAPI_GRO_CB(p)->flush |= flush; 1531 1532 /* We need to store of the IP ID check to be included later 1533 * when we can verify that this packet does in fact belong 1534 * to a given flow. 1535 */ 1536 flush_id = (u16)(id - ntohs(iph2->id)); 1537 1538 /* This bit of code makes it much easier for us to identify 1539 * the cases where we are doing atomic vs non-atomic IP ID 1540 * checks. Specifically an atomic check can return IP ID 1541 * values 0 - 0xFFFF, while a non-atomic check can only 1542 * return 0 or 0xFFFF. 1543 */ 1544 if (!NAPI_GRO_CB(p)->is_atomic || 1545 !(iph->frag_off & htons(IP_DF))) { 1546 flush_id ^= NAPI_GRO_CB(p)->count; 1547 flush_id = flush_id ? 0xFFFF : 0; 1548 } 1549 1550 /* If the previous IP ID value was based on an atomic 1551 * datagram we can overwrite the value and ignore it. 1552 */ 1553 if (NAPI_GRO_CB(skb)->is_atomic) 1554 NAPI_GRO_CB(p)->flush_id = flush_id; 1555 else 1556 NAPI_GRO_CB(p)->flush_id |= flush_id; 1557 } 1558 1559 NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF)); 1560 NAPI_GRO_CB(skb)->flush |= flush; 1561 skb_set_network_header(skb, off); 1562 /* The above will be needed by the transport layer if there is one 1563 * immediately following this IP hdr. 1564 */ 1565 1566 /* Note : No need to call skb_gro_postpull_rcsum() here, 1567 * as we already checked checksum over ipv4 header was 0 1568 */ 1569 skb_gro_pull(skb, sizeof(*iph)); 1570 skb_set_transport_header(skb, skb_gro_offset(skb)); 1571 1572 pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive, 1573 ops->callbacks.gro_receive, head, skb); 1574 1575 out: 1576 skb_gro_flush_final(skb, pp, flush); 1577 1578 return pp; 1579 } 1580 1581 static struct sk_buff *ipip_gro_receive(struct list_head *head, 1582 struct sk_buff *skb) 1583 { 1584 if (NAPI_GRO_CB(skb)->encap_mark) { 1585 NAPI_GRO_CB(skb)->flush = 1; 1586 return NULL; 1587 } 1588 1589 NAPI_GRO_CB(skb)->encap_mark = 1; 1590 1591 return inet_gro_receive(head, skb); 1592 } 1593 1594 #define SECONDS_PER_DAY 86400 1595 1596 /* inet_current_timestamp - Return IP network timestamp 1597 * 1598 * Return milliseconds since midnight in network byte order. 1599 */ 1600 __be32 inet_current_timestamp(void) 1601 { 1602 u32 secs; 1603 u32 msecs; 1604 struct timespec64 ts; 1605 1606 ktime_get_real_ts64(&ts); 1607 1608 /* Get secs since midnight. */ 1609 (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs); 1610 /* Convert to msecs. */ 1611 msecs = secs * MSEC_PER_SEC; 1612 /* Convert nsec to msec. */ 1613 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC; 1614 1615 /* Convert to network byte order. */ 1616 return htonl(msecs); 1617 } 1618 EXPORT_SYMBOL(inet_current_timestamp); 1619 1620 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len) 1621 { 1622 if (sk->sk_family == AF_INET) 1623 return ip_recv_error(sk, msg, len, addr_len); 1624 #if IS_ENABLED(CONFIG_IPV6) 1625 if (sk->sk_family == AF_INET6) 1626 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len); 1627 #endif 1628 return -EINVAL; 1629 } 1630 1631 int inet_gro_complete(struct sk_buff *skb, int nhoff) 1632 { 1633 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff); 1634 const struct net_offload *ops; 1635 __be16 totlen = iph->tot_len; 1636 int proto = iph->protocol; 1637 int err = -ENOSYS; 1638 1639 if (skb->encapsulation) { 1640 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP)); 1641 skb_set_inner_network_header(skb, nhoff); 1642 } 1643 1644 iph_set_totlen(iph, skb->len - nhoff); 1645 csum_replace2(&iph->check, totlen, iph->tot_len); 1646 1647 ops = rcu_dereference(inet_offloads[proto]); 1648 if (WARN_ON(!ops || !ops->callbacks.gro_complete)) 1649 goto out; 1650 1651 /* Only need to add sizeof(*iph) to get to the next hdr below 1652 * because any hdr with option will have been flushed in 1653 * inet_gro_receive(). 1654 */ 1655 err = INDIRECT_CALL_2(ops->callbacks.gro_complete, 1656 tcp4_gro_complete, udp4_gro_complete, 1657 skb, nhoff + sizeof(*iph)); 1658 1659 out: 1660 return err; 1661 } 1662 1663 static int ipip_gro_complete(struct sk_buff *skb, int nhoff) 1664 { 1665 skb->encapsulation = 1; 1666 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4; 1667 return inet_gro_complete(skb, nhoff); 1668 } 1669 1670 int inet_ctl_sock_create(struct sock **sk, unsigned short family, 1671 unsigned short type, unsigned char protocol, 1672 struct net *net) 1673 { 1674 struct socket *sock; 1675 int rc = sock_create_kern(net, family, type, protocol, &sock); 1676 1677 if (rc == 0) { 1678 *sk = sock->sk; 1679 (*sk)->sk_allocation = GFP_ATOMIC; 1680 (*sk)->sk_use_task_frag = false; 1681 /* 1682 * Unhash it so that IP input processing does not even see it, 1683 * we do not wish this socket to see incoming packets. 1684 */ 1685 (*sk)->sk_prot->unhash(*sk); 1686 } 1687 return rc; 1688 } 1689 EXPORT_SYMBOL_GPL(inet_ctl_sock_create); 1690 1691 unsigned long snmp_fold_field(void __percpu *mib, int offt) 1692 { 1693 unsigned long res = 0; 1694 int i; 1695 1696 for_each_possible_cpu(i) 1697 res += snmp_get_cpu_field(mib, i, offt); 1698 return res; 1699 } 1700 EXPORT_SYMBOL_GPL(snmp_fold_field); 1701 1702 #if BITS_PER_LONG==32 1703 1704 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt, 1705 size_t syncp_offset) 1706 { 1707 void *bhptr; 1708 struct u64_stats_sync *syncp; 1709 u64 v; 1710 unsigned int start; 1711 1712 bhptr = per_cpu_ptr(mib, cpu); 1713 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset); 1714 do { 1715 start = u64_stats_fetch_begin(syncp); 1716 v = *(((u64 *)bhptr) + offt); 1717 } while (u64_stats_fetch_retry(syncp, start)); 1718 1719 return v; 1720 } 1721 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64); 1722 1723 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset) 1724 { 1725 u64 res = 0; 1726 int cpu; 1727 1728 for_each_possible_cpu(cpu) { 1729 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset); 1730 } 1731 return res; 1732 } 1733 EXPORT_SYMBOL_GPL(snmp_fold_field64); 1734 #endif 1735 1736 #ifdef CONFIG_IP_MULTICAST 1737 static const struct net_protocol igmp_protocol = { 1738 .handler = igmp_rcv, 1739 }; 1740 #endif 1741 1742 static const struct net_protocol tcp_protocol = { 1743 .handler = tcp_v4_rcv, 1744 .err_handler = tcp_v4_err, 1745 .no_policy = 1, 1746 .icmp_strict_tag_validation = 1, 1747 }; 1748 1749 static const struct net_protocol udp_protocol = { 1750 .handler = udp_rcv, 1751 .err_handler = udp_err, 1752 .no_policy = 1, 1753 }; 1754 1755 static const struct net_protocol icmp_protocol = { 1756 .handler = icmp_rcv, 1757 .err_handler = icmp_err, 1758 .no_policy = 1, 1759 }; 1760 1761 static __net_init int ipv4_mib_init_net(struct net *net) 1762 { 1763 int i; 1764 1765 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib); 1766 if (!net->mib.tcp_statistics) 1767 goto err_tcp_mib; 1768 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib); 1769 if (!net->mib.ip_statistics) 1770 goto err_ip_mib; 1771 1772 for_each_possible_cpu(i) { 1773 struct ipstats_mib *af_inet_stats; 1774 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i); 1775 u64_stats_init(&af_inet_stats->syncp); 1776 } 1777 1778 net->mib.net_statistics = alloc_percpu(struct linux_mib); 1779 if (!net->mib.net_statistics) 1780 goto err_net_mib; 1781 net->mib.udp_statistics = alloc_percpu(struct udp_mib); 1782 if (!net->mib.udp_statistics) 1783 goto err_udp_mib; 1784 net->mib.udplite_statistics = alloc_percpu(struct udp_mib); 1785 if (!net->mib.udplite_statistics) 1786 goto err_udplite_mib; 1787 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib); 1788 if (!net->mib.icmp_statistics) 1789 goto err_icmp_mib; 1790 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib), 1791 GFP_KERNEL); 1792 if (!net->mib.icmpmsg_statistics) 1793 goto err_icmpmsg_mib; 1794 1795 tcp_mib_init(net); 1796 return 0; 1797 1798 err_icmpmsg_mib: 1799 free_percpu(net->mib.icmp_statistics); 1800 err_icmp_mib: 1801 free_percpu(net->mib.udplite_statistics); 1802 err_udplite_mib: 1803 free_percpu(net->mib.udp_statistics); 1804 err_udp_mib: 1805 free_percpu(net->mib.net_statistics); 1806 err_net_mib: 1807 free_percpu(net->mib.ip_statistics); 1808 err_ip_mib: 1809 free_percpu(net->mib.tcp_statistics); 1810 err_tcp_mib: 1811 return -ENOMEM; 1812 } 1813 1814 static __net_exit void ipv4_mib_exit_net(struct net *net) 1815 { 1816 kfree(net->mib.icmpmsg_statistics); 1817 free_percpu(net->mib.icmp_statistics); 1818 free_percpu(net->mib.udplite_statistics); 1819 free_percpu(net->mib.udp_statistics); 1820 free_percpu(net->mib.net_statistics); 1821 free_percpu(net->mib.ip_statistics); 1822 free_percpu(net->mib.tcp_statistics); 1823 #ifdef CONFIG_MPTCP 1824 /* allocated on demand, see mptcp_init_sock() */ 1825 free_percpu(net->mib.mptcp_statistics); 1826 #endif 1827 } 1828 1829 static __net_initdata struct pernet_operations ipv4_mib_ops = { 1830 .init = ipv4_mib_init_net, 1831 .exit = ipv4_mib_exit_net, 1832 }; 1833 1834 static int __init init_ipv4_mibs(void) 1835 { 1836 return register_pernet_subsys(&ipv4_mib_ops); 1837 } 1838 1839 static __net_init int inet_init_net(struct net *net) 1840 { 1841 /* 1842 * Set defaults for local port range 1843 */ 1844 seqlock_init(&net->ipv4.ip_local_ports.lock); 1845 net->ipv4.ip_local_ports.range[0] = 32768; 1846 net->ipv4.ip_local_ports.range[1] = 60999; 1847 1848 seqlock_init(&net->ipv4.ping_group_range.lock); 1849 /* 1850 * Sane defaults - nobody may create ping sockets. 1851 * Boot scripts should set this to distro-specific group. 1852 */ 1853 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1); 1854 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0); 1855 1856 /* Default values for sysctl-controlled parameters. 1857 * We set them here, in case sysctl is not compiled. 1858 */ 1859 net->ipv4.sysctl_ip_default_ttl = IPDEFTTL; 1860 net->ipv4.sysctl_ip_fwd_update_priority = 1; 1861 net->ipv4.sysctl_ip_dynaddr = 0; 1862 net->ipv4.sysctl_ip_early_demux = 1; 1863 net->ipv4.sysctl_udp_early_demux = 1; 1864 net->ipv4.sysctl_tcp_early_demux = 1; 1865 net->ipv4.sysctl_nexthop_compat_mode = 1; 1866 #ifdef CONFIG_SYSCTL 1867 net->ipv4.sysctl_ip_prot_sock = PROT_SOCK; 1868 #endif 1869 1870 /* Some igmp sysctl, whose values are always used */ 1871 net->ipv4.sysctl_igmp_max_memberships = 20; 1872 net->ipv4.sysctl_igmp_max_msf = 10; 1873 /* IGMP reports for link-local multicast groups are enabled by default */ 1874 net->ipv4.sysctl_igmp_llm_reports = 1; 1875 net->ipv4.sysctl_igmp_qrv = 2; 1876 1877 net->ipv4.sysctl_fib_notify_on_flag_change = 0; 1878 1879 return 0; 1880 } 1881 1882 static __net_initdata struct pernet_operations af_inet_ops = { 1883 .init = inet_init_net, 1884 }; 1885 1886 static int __init init_inet_pernet_ops(void) 1887 { 1888 return register_pernet_subsys(&af_inet_ops); 1889 } 1890 1891 static int ipv4_proc_init(void); 1892 1893 /* 1894 * IP protocol layer initialiser 1895 */ 1896 1897 static struct packet_offload ip_packet_offload __read_mostly = { 1898 .type = cpu_to_be16(ETH_P_IP), 1899 .callbacks = { 1900 .gso_segment = inet_gso_segment, 1901 .gro_receive = inet_gro_receive, 1902 .gro_complete = inet_gro_complete, 1903 }, 1904 }; 1905 1906 static const struct net_offload ipip_offload = { 1907 .callbacks = { 1908 .gso_segment = ipip_gso_segment, 1909 .gro_receive = ipip_gro_receive, 1910 .gro_complete = ipip_gro_complete, 1911 }, 1912 }; 1913 1914 static int __init ipip_offload_init(void) 1915 { 1916 return inet_add_offload(&ipip_offload, IPPROTO_IPIP); 1917 } 1918 1919 static int __init ipv4_offload_init(void) 1920 { 1921 /* 1922 * Add offloads 1923 */ 1924 if (udpv4_offload_init() < 0) 1925 pr_crit("%s: Cannot add UDP protocol offload\n", __func__); 1926 if (tcpv4_offload_init() < 0) 1927 pr_crit("%s: Cannot add TCP protocol offload\n", __func__); 1928 if (ipip_offload_init() < 0) 1929 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__); 1930 1931 dev_add_offload(&ip_packet_offload); 1932 return 0; 1933 } 1934 1935 fs_initcall(ipv4_offload_init); 1936 1937 static struct packet_type ip_packet_type __read_mostly = { 1938 .type = cpu_to_be16(ETH_P_IP), 1939 .func = ip_rcv, 1940 .list_func = ip_list_rcv, 1941 }; 1942 1943 static int __init inet_init(void) 1944 { 1945 struct inet_protosw *q; 1946 struct list_head *r; 1947 int rc; 1948 1949 sock_skb_cb_check_size(sizeof(struct inet_skb_parm)); 1950 1951 raw_hashinfo_init(&raw_v4_hashinfo); 1952 1953 rc = proto_register(&tcp_prot, 1); 1954 if (rc) 1955 goto out; 1956 1957 rc = proto_register(&udp_prot, 1); 1958 if (rc) 1959 goto out_unregister_tcp_proto; 1960 1961 rc = proto_register(&raw_prot, 1); 1962 if (rc) 1963 goto out_unregister_udp_proto; 1964 1965 rc = proto_register(&ping_prot, 1); 1966 if (rc) 1967 goto out_unregister_raw_proto; 1968 1969 /* 1970 * Tell SOCKET that we are alive... 1971 */ 1972 1973 (void)sock_register(&inet_family_ops); 1974 1975 #ifdef CONFIG_SYSCTL 1976 ip_static_sysctl_init(); 1977 #endif 1978 1979 /* 1980 * Add all the base protocols. 1981 */ 1982 1983 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0) 1984 pr_crit("%s: Cannot add ICMP protocol\n", __func__); 1985 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0) 1986 pr_crit("%s: Cannot add UDP protocol\n", __func__); 1987 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0) 1988 pr_crit("%s: Cannot add TCP protocol\n", __func__); 1989 #ifdef CONFIG_IP_MULTICAST 1990 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0) 1991 pr_crit("%s: Cannot add IGMP protocol\n", __func__); 1992 #endif 1993 1994 /* Register the socket-side information for inet_create. */ 1995 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r) 1996 INIT_LIST_HEAD(r); 1997 1998 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q) 1999 inet_register_protosw(q); 2000 2001 /* 2002 * Set the ARP module up 2003 */ 2004 2005 arp_init(); 2006 2007 /* 2008 * Set the IP module up 2009 */ 2010 2011 ip_init(); 2012 2013 /* Initialise per-cpu ipv4 mibs */ 2014 if (init_ipv4_mibs()) 2015 panic("%s: Cannot init ipv4 mibs\n", __func__); 2016 2017 /* Setup TCP slab cache for open requests. */ 2018 tcp_init(); 2019 2020 /* Setup UDP memory threshold */ 2021 udp_init(); 2022 2023 /* Add UDP-Lite (RFC 3828) */ 2024 udplite4_register(); 2025 2026 raw_init(); 2027 2028 ping_init(); 2029 2030 /* 2031 * Set the ICMP layer up 2032 */ 2033 2034 if (icmp_init() < 0) 2035 panic("Failed to create the ICMP control socket.\n"); 2036 2037 /* 2038 * Initialise the multicast router 2039 */ 2040 #if defined(CONFIG_IP_MROUTE) 2041 if (ip_mr_init()) 2042 pr_crit("%s: Cannot init ipv4 mroute\n", __func__); 2043 #endif 2044 2045 if (init_inet_pernet_ops()) 2046 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__); 2047 2048 ipv4_proc_init(); 2049 2050 ipfrag_init(); 2051 2052 dev_add_pack(&ip_packet_type); 2053 2054 ip_tunnel_core_init(); 2055 2056 rc = 0; 2057 out: 2058 return rc; 2059 out_unregister_raw_proto: 2060 proto_unregister(&raw_prot); 2061 out_unregister_udp_proto: 2062 proto_unregister(&udp_prot); 2063 out_unregister_tcp_proto: 2064 proto_unregister(&tcp_prot); 2065 goto out; 2066 } 2067 2068 fs_initcall(inet_init); 2069 2070 /* ------------------------------------------------------------------------ */ 2071 2072 #ifdef CONFIG_PROC_FS 2073 static int __init ipv4_proc_init(void) 2074 { 2075 int rc = 0; 2076 2077 if (raw_proc_init()) 2078 goto out_raw; 2079 if (tcp4_proc_init()) 2080 goto out_tcp; 2081 if (udp4_proc_init()) 2082 goto out_udp; 2083 if (ping_proc_init()) 2084 goto out_ping; 2085 if (ip_misc_proc_init()) 2086 goto out_misc; 2087 out: 2088 return rc; 2089 out_misc: 2090 ping_proc_exit(); 2091 out_ping: 2092 udp4_proc_exit(); 2093 out_udp: 2094 tcp4_proc_exit(); 2095 out_tcp: 2096 raw_proc_exit(); 2097 out_raw: 2098 rc = -ENOMEM; 2099 goto out; 2100 } 2101 2102 #else /* CONFIG_PROC_FS */ 2103 static int __init ipv4_proc_init(void) 2104 { 2105 return 0; 2106 } 2107 #endif /* CONFIG_PROC_FS */ 2108