1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * PF_INET protocol family socket handler. 7 * 8 * Authors: Ross Biro 9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 10 * Florian La Roche, <flla@stud.uni-sb.de> 11 * Alan Cox, <A.Cox@swansea.ac.uk> 12 * 13 * Changes (see also sock.c) 14 * 15 * piggy, 16 * Karl Knutson : Socket protocol table 17 * A.N.Kuznetsov : Socket death error in accept(). 18 * John Richardson : Fix non blocking error in connect() 19 * so sockets that fail to connect 20 * don't return -EINPROGRESS. 21 * Alan Cox : Asynchronous I/O support 22 * Alan Cox : Keep correct socket pointer on sock 23 * structures 24 * when accept() ed 25 * Alan Cox : Semantics of SO_LINGER aren't state 26 * moved to close when you look carefully. 27 * With this fixed and the accept bug fixed 28 * some RPC stuff seems happier. 29 * Niibe Yutaka : 4.4BSD style write async I/O 30 * Alan Cox, 31 * Tony Gale : Fixed reuse semantics. 32 * Alan Cox : bind() shouldn't abort existing but dead 33 * sockets. Stops FTP netin:.. I hope. 34 * Alan Cox : bind() works correctly for RAW sockets. 35 * Note that FreeBSD at least was broken 36 * in this respect so be careful with 37 * compatibility tests... 38 * Alan Cox : routing cache support 39 * Alan Cox : memzero the socket structure for 40 * compactness. 41 * Matt Day : nonblock connect error handler 42 * Alan Cox : Allow large numbers of pending sockets 43 * (eg for big web sites), but only if 44 * specifically application requested. 45 * Alan Cox : New buffering throughout IP. Used 46 * dumbly. 47 * Alan Cox : New buffering now used smartly. 48 * Alan Cox : BSD rather than common sense 49 * interpretation of listen. 50 * Germano Caronni : Assorted small races. 51 * Alan Cox : sendmsg/recvmsg basic support. 52 * Alan Cox : Only sendmsg/recvmsg now supported. 53 * Alan Cox : Locked down bind (see security list). 54 * Alan Cox : Loosened bind a little. 55 * Mike McLagan : ADD/DEL DLCI Ioctls 56 * Willy Konynenberg : Transparent proxying support. 57 * David S. Miller : New socket lookup architecture. 58 * Some other random speedups. 59 * Cyrus Durgin : Cleaned up file for kmod hacks. 60 * Andi Kleen : Fix inet_stream_connect TCP race. 61 * 62 * This program is free software; you can redistribute it and/or 63 * modify it under the terms of the GNU General Public License 64 * as published by the Free Software Foundation; either version 65 * 2 of the License, or (at your option) any later version. 66 */ 67 68 #define pr_fmt(fmt) "IPv4: " fmt 69 70 #include <linux/err.h> 71 #include <linux/errno.h> 72 #include <linux/types.h> 73 #include <linux/socket.h> 74 #include <linux/in.h> 75 #include <linux/kernel.h> 76 #include <linux/kmod.h> 77 #include <linux/sched.h> 78 #include <linux/timer.h> 79 #include <linux/string.h> 80 #include <linux/sockios.h> 81 #include <linux/net.h> 82 #include <linux/capability.h> 83 #include <linux/fcntl.h> 84 #include <linux/mm.h> 85 #include <linux/interrupt.h> 86 #include <linux/stat.h> 87 #include <linux/init.h> 88 #include <linux/poll.h> 89 #include <linux/netfilter_ipv4.h> 90 #include <linux/random.h> 91 #include <linux/slab.h> 92 93 #include <linux/uaccess.h> 94 95 #include <linux/inet.h> 96 #include <linux/igmp.h> 97 #include <linux/inetdevice.h> 98 #include <linux/netdevice.h> 99 #include <net/checksum.h> 100 #include <net/ip.h> 101 #include <net/protocol.h> 102 #include <net/arp.h> 103 #include <net/route.h> 104 #include <net/ip_fib.h> 105 #include <net/inet_connection_sock.h> 106 #include <net/tcp.h> 107 #include <net/udp.h> 108 #include <net/udplite.h> 109 #include <net/ping.h> 110 #include <linux/skbuff.h> 111 #include <net/sock.h> 112 #include <net/raw.h> 113 #include <net/icmp.h> 114 #include <net/inet_common.h> 115 #include <net/ip_tunnels.h> 116 #include <net/xfrm.h> 117 #include <net/net_namespace.h> 118 #include <net/secure_seq.h> 119 #ifdef CONFIG_IP_MROUTE 120 #include <linux/mroute.h> 121 #endif 122 #include <net/l3mdev.h> 123 124 #include <trace/events/sock.h> 125 126 /* The inetsw table contains everything that inet_create needs to 127 * build a new socket. 128 */ 129 static struct list_head inetsw[SOCK_MAX]; 130 static DEFINE_SPINLOCK(inetsw_lock); 131 132 /* New destruction routine */ 133 134 void inet_sock_destruct(struct sock *sk) 135 { 136 struct inet_sock *inet = inet_sk(sk); 137 138 __skb_queue_purge(&sk->sk_receive_queue); 139 __skb_queue_purge(&sk->sk_error_queue); 140 141 sk_mem_reclaim(sk); 142 143 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) { 144 pr_err("Attempt to release TCP socket in state %d %p\n", 145 sk->sk_state, sk); 146 return; 147 } 148 if (!sock_flag(sk, SOCK_DEAD)) { 149 pr_err("Attempt to release alive inet socket %p\n", sk); 150 return; 151 } 152 153 WARN_ON(atomic_read(&sk->sk_rmem_alloc)); 154 WARN_ON(refcount_read(&sk->sk_wmem_alloc)); 155 WARN_ON(sk->sk_wmem_queued); 156 WARN_ON(sk->sk_forward_alloc); 157 158 kfree(rcu_dereference_protected(inet->inet_opt, 1)); 159 dst_release(rcu_dereference_check(sk->sk_dst_cache, 1)); 160 dst_release(sk->sk_rx_dst); 161 sk_refcnt_debug_dec(sk); 162 } 163 EXPORT_SYMBOL(inet_sock_destruct); 164 165 /* 166 * The routines beyond this point handle the behaviour of an AF_INET 167 * socket object. Mostly it punts to the subprotocols of IP to do 168 * the work. 169 */ 170 171 /* 172 * Automatically bind an unbound socket. 173 */ 174 175 static int inet_autobind(struct sock *sk) 176 { 177 struct inet_sock *inet; 178 /* We may need to bind the socket. */ 179 lock_sock(sk); 180 inet = inet_sk(sk); 181 if (!inet->inet_num) { 182 if (sk->sk_prot->get_port(sk, 0)) { 183 release_sock(sk); 184 return -EAGAIN; 185 } 186 inet->inet_sport = htons(inet->inet_num); 187 } 188 release_sock(sk); 189 return 0; 190 } 191 192 /* 193 * Move a socket into listening state. 194 */ 195 int inet_listen(struct socket *sock, int backlog) 196 { 197 struct sock *sk = sock->sk; 198 unsigned char old_state; 199 int err, tcp_fastopen; 200 201 lock_sock(sk); 202 203 err = -EINVAL; 204 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM) 205 goto out; 206 207 old_state = sk->sk_state; 208 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN))) 209 goto out; 210 211 /* Really, if the socket is already in listen state 212 * we can only allow the backlog to be adjusted. 213 */ 214 if (old_state != TCP_LISTEN) { 215 /* Enable TFO w/o requiring TCP_FASTOPEN socket option. 216 * Note that only TCP sockets (SOCK_STREAM) will reach here. 217 * Also fastopen backlog may already been set via the option 218 * because the socket was in TCP_LISTEN state previously but 219 * was shutdown() rather than close(). 220 */ 221 tcp_fastopen = sock_net(sk)->ipv4.sysctl_tcp_fastopen; 222 if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) && 223 (tcp_fastopen & TFO_SERVER_ENABLE) && 224 !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) { 225 fastopen_queue_tune(sk, backlog); 226 tcp_fastopen_init_key_once(sock_net(sk)); 227 } 228 229 err = inet_csk_listen_start(sk, backlog); 230 if (err) 231 goto out; 232 } 233 sk->sk_max_ack_backlog = backlog; 234 err = 0; 235 236 out: 237 release_sock(sk); 238 return err; 239 } 240 EXPORT_SYMBOL(inet_listen); 241 242 /* 243 * Create an inet socket. 244 */ 245 246 static int inet_create(struct net *net, struct socket *sock, int protocol, 247 int kern) 248 { 249 struct sock *sk; 250 struct inet_protosw *answer; 251 struct inet_sock *inet; 252 struct proto *answer_prot; 253 unsigned char answer_flags; 254 int try_loading_module = 0; 255 int err; 256 257 if (protocol < 0 || protocol >= IPPROTO_MAX) 258 return -EINVAL; 259 260 sock->state = SS_UNCONNECTED; 261 262 /* Look for the requested type/protocol pair. */ 263 lookup_protocol: 264 err = -ESOCKTNOSUPPORT; 265 rcu_read_lock(); 266 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) { 267 268 err = 0; 269 /* Check the non-wild match. */ 270 if (protocol == answer->protocol) { 271 if (protocol != IPPROTO_IP) 272 break; 273 } else { 274 /* Check for the two wild cases. */ 275 if (IPPROTO_IP == protocol) { 276 protocol = answer->protocol; 277 break; 278 } 279 if (IPPROTO_IP == answer->protocol) 280 break; 281 } 282 err = -EPROTONOSUPPORT; 283 } 284 285 if (unlikely(err)) { 286 if (try_loading_module < 2) { 287 rcu_read_unlock(); 288 /* 289 * Be more specific, e.g. net-pf-2-proto-132-type-1 290 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM) 291 */ 292 if (++try_loading_module == 1) 293 request_module("net-pf-%d-proto-%d-type-%d", 294 PF_INET, protocol, sock->type); 295 /* 296 * Fall back to generic, e.g. net-pf-2-proto-132 297 * (net-pf-PF_INET-proto-IPPROTO_SCTP) 298 */ 299 else 300 request_module("net-pf-%d-proto-%d", 301 PF_INET, protocol); 302 goto lookup_protocol; 303 } else 304 goto out_rcu_unlock; 305 } 306 307 err = -EPERM; 308 if (sock->type == SOCK_RAW && !kern && 309 !ns_capable(net->user_ns, CAP_NET_RAW)) 310 goto out_rcu_unlock; 311 312 sock->ops = answer->ops; 313 answer_prot = answer->prot; 314 answer_flags = answer->flags; 315 rcu_read_unlock(); 316 317 WARN_ON(!answer_prot->slab); 318 319 err = -ENOBUFS; 320 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern); 321 if (!sk) 322 goto out; 323 324 err = 0; 325 if (INET_PROTOSW_REUSE & answer_flags) 326 sk->sk_reuse = SK_CAN_REUSE; 327 328 inet = inet_sk(sk); 329 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0; 330 331 inet->nodefrag = 0; 332 333 if (SOCK_RAW == sock->type) { 334 inet->inet_num = protocol; 335 if (IPPROTO_RAW == protocol) 336 inet->hdrincl = 1; 337 } 338 339 if (net->ipv4.sysctl_ip_no_pmtu_disc) 340 inet->pmtudisc = IP_PMTUDISC_DONT; 341 else 342 inet->pmtudisc = IP_PMTUDISC_WANT; 343 344 inet->inet_id = 0; 345 346 sock_init_data(sock, sk); 347 348 sk->sk_destruct = inet_sock_destruct; 349 sk->sk_protocol = protocol; 350 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv; 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 sk_refcnt_debug_inc(sk); 361 362 if (inet->inet_num) { 363 /* It assumes that any protocol which allows 364 * the user to assign a number at socket 365 * creation time automatically 366 * shares. 367 */ 368 inet->inet_sport = htons(inet->inet_num); 369 /* Add to protocol hash chains. */ 370 err = sk->sk_prot->hash(sk); 371 if (err) { 372 sk_common_release(sk); 373 goto out; 374 } 375 } 376 377 if (sk->sk_prot->init) { 378 err = sk->sk_prot->init(sk); 379 if (err) { 380 sk_common_release(sk); 381 goto out; 382 } 383 } 384 385 if (!kern) { 386 err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk); 387 if (err) { 388 sk_common_release(sk); 389 goto out; 390 } 391 } 392 out: 393 return err; 394 out_rcu_unlock: 395 rcu_read_unlock(); 396 goto out; 397 } 398 399 400 /* 401 * The peer socket should always be NULL (or else). When we call this 402 * function we are destroying the object and from then on nobody 403 * should refer to it. 404 */ 405 int inet_release(struct socket *sock) 406 { 407 struct sock *sk = sock->sk; 408 409 if (sk) { 410 long timeout; 411 412 /* Applications forget to leave groups before exiting */ 413 ip_mc_drop_socket(sk); 414 415 /* If linger is set, we don't return until the close 416 * is complete. Otherwise we return immediately. The 417 * actually closing is done the same either way. 418 * 419 * If the close is due to the process exiting, we never 420 * linger.. 421 */ 422 timeout = 0; 423 if (sock_flag(sk, SOCK_LINGER) && 424 !(current->flags & PF_EXITING)) 425 timeout = sk->sk_lingertime; 426 sock->sk = NULL; 427 sk->sk_prot->close(sk, timeout); 428 } 429 return 0; 430 } 431 EXPORT_SYMBOL(inet_release); 432 433 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) 434 { 435 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr; 436 struct sock *sk = sock->sk; 437 struct inet_sock *inet = inet_sk(sk); 438 struct net *net = sock_net(sk); 439 unsigned short snum; 440 int chk_addr_ret; 441 u32 tb_id = RT_TABLE_LOCAL; 442 int err; 443 444 /* If the socket has its own bind function then use it. (RAW) */ 445 if (sk->sk_prot->bind) { 446 err = sk->sk_prot->bind(sk, uaddr, addr_len); 447 goto out; 448 } 449 err = -EINVAL; 450 if (addr_len < sizeof(struct sockaddr_in)) 451 goto out; 452 453 if (addr->sin_family != AF_INET) { 454 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET) 455 * only if s_addr is INADDR_ANY. 456 */ 457 err = -EAFNOSUPPORT; 458 if (addr->sin_family != AF_UNSPEC || 459 addr->sin_addr.s_addr != htonl(INADDR_ANY)) 460 goto out; 461 } 462 463 tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id; 464 chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id); 465 466 /* Not specified by any standard per-se, however it breaks too 467 * many applications when removed. It is unfortunate since 468 * allowing applications to make a non-local bind solves 469 * several problems with systems using dynamic addressing. 470 * (ie. your servers still start up even if your ISDN link 471 * is temporarily down) 472 */ 473 err = -EADDRNOTAVAIL; 474 if (!net->ipv4.sysctl_ip_nonlocal_bind && 475 !(inet->freebind || inet->transparent) && 476 addr->sin_addr.s_addr != htonl(INADDR_ANY) && 477 chk_addr_ret != RTN_LOCAL && 478 chk_addr_ret != RTN_MULTICAST && 479 chk_addr_ret != RTN_BROADCAST) 480 goto out; 481 482 snum = ntohs(addr->sin_port); 483 err = -EACCES; 484 if (snum && snum < inet_prot_sock(net) && 485 !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE)) 486 goto out; 487 488 /* We keep a pair of addresses. rcv_saddr is the one 489 * used by hash lookups, and saddr is used for transmit. 490 * 491 * In the BSD API these are the same except where it 492 * would be illegal to use them (multicast/broadcast) in 493 * which case the sending device address is used. 494 */ 495 lock_sock(sk); 496 497 /* Check these errors (active socket, double bind). */ 498 err = -EINVAL; 499 if (sk->sk_state != TCP_CLOSE || inet->inet_num) 500 goto out_release_sock; 501 502 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr; 503 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST) 504 inet->inet_saddr = 0; /* Use device */ 505 506 /* Make sure we are allowed to bind here. */ 507 if ((snum || !inet->bind_address_no_port) && 508 sk->sk_prot->get_port(sk, snum)) { 509 inet->inet_saddr = inet->inet_rcv_saddr = 0; 510 err = -EADDRINUSE; 511 goto out_release_sock; 512 } 513 514 if (inet->inet_rcv_saddr) 515 sk->sk_userlocks |= SOCK_BINDADDR_LOCK; 516 if (snum) 517 sk->sk_userlocks |= SOCK_BINDPORT_LOCK; 518 inet->inet_sport = htons(inet->inet_num); 519 inet->inet_daddr = 0; 520 inet->inet_dport = 0; 521 sk_dst_reset(sk); 522 err = 0; 523 out_release_sock: 524 release_sock(sk); 525 out: 526 return err; 527 } 528 EXPORT_SYMBOL(inet_bind); 529 530 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr, 531 int addr_len, int flags) 532 { 533 struct sock *sk = sock->sk; 534 535 if (addr_len < sizeof(uaddr->sa_family)) 536 return -EINVAL; 537 if (uaddr->sa_family == AF_UNSPEC) 538 return sk->sk_prot->disconnect(sk, flags); 539 540 if (!inet_sk(sk)->inet_num && inet_autobind(sk)) 541 return -EAGAIN; 542 return sk->sk_prot->connect(sk, uaddr, addr_len); 543 } 544 EXPORT_SYMBOL(inet_dgram_connect); 545 546 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias) 547 { 548 DEFINE_WAIT_FUNC(wait, woken_wake_function); 549 550 add_wait_queue(sk_sleep(sk), &wait); 551 sk->sk_write_pending += writebias; 552 553 /* Basic assumption: if someone sets sk->sk_err, he _must_ 554 * change state of the socket from TCP_SYN_*. 555 * Connect() does not allow to get error notifications 556 * without closing the socket. 557 */ 558 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 559 release_sock(sk); 560 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo); 561 lock_sock(sk); 562 if (signal_pending(current) || !timeo) 563 break; 564 } 565 remove_wait_queue(sk_sleep(sk), &wait); 566 sk->sk_write_pending -= writebias; 567 return timeo; 568 } 569 570 /* 571 * Connect to a remote host. There is regrettably still a little 572 * TCP 'magic' in here. 573 */ 574 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr, 575 int addr_len, int flags, int is_sendmsg) 576 { 577 struct sock *sk = sock->sk; 578 int err; 579 long timeo; 580 581 /* 582 * uaddr can be NULL and addr_len can be 0 if: 583 * sk is a TCP fastopen active socket and 584 * TCP_FASTOPEN_CONNECT sockopt is set and 585 * we already have a valid cookie for this socket. 586 * In this case, user can call write() after connect(). 587 * write() will invoke tcp_sendmsg_fastopen() which calls 588 * __inet_stream_connect(). 589 */ 590 if (uaddr) { 591 if (addr_len < sizeof(uaddr->sa_family)) 592 return -EINVAL; 593 594 if (uaddr->sa_family == AF_UNSPEC) { 595 err = sk->sk_prot->disconnect(sk, flags); 596 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED; 597 goto out; 598 } 599 } 600 601 switch (sock->state) { 602 default: 603 err = -EINVAL; 604 goto out; 605 case SS_CONNECTED: 606 err = -EISCONN; 607 goto out; 608 case SS_CONNECTING: 609 if (inet_sk(sk)->defer_connect) 610 err = is_sendmsg ? -EINPROGRESS : -EISCONN; 611 else 612 err = -EALREADY; 613 /* Fall out of switch with err, set for this state */ 614 break; 615 case SS_UNCONNECTED: 616 err = -EISCONN; 617 if (sk->sk_state != TCP_CLOSE) 618 goto out; 619 620 err = sk->sk_prot->connect(sk, uaddr, addr_len); 621 if (err < 0) 622 goto out; 623 624 sock->state = SS_CONNECTING; 625 626 if (!err && inet_sk(sk)->defer_connect) 627 goto out; 628 629 /* Just entered SS_CONNECTING state; the only 630 * difference is that return value in non-blocking 631 * case is EINPROGRESS, rather than EALREADY. 632 */ 633 err = -EINPROGRESS; 634 break; 635 } 636 637 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK); 638 639 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 640 int writebias = (sk->sk_protocol == IPPROTO_TCP) && 641 tcp_sk(sk)->fastopen_req && 642 tcp_sk(sk)->fastopen_req->data ? 1 : 0; 643 644 /* Error code is set above */ 645 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias)) 646 goto out; 647 648 err = sock_intr_errno(timeo); 649 if (signal_pending(current)) 650 goto out; 651 } 652 653 /* Connection was closed by RST, timeout, ICMP error 654 * or another process disconnected us. 655 */ 656 if (sk->sk_state == TCP_CLOSE) 657 goto sock_error; 658 659 /* sk->sk_err may be not zero now, if RECVERR was ordered by user 660 * and error was received after socket entered established state. 661 * Hence, it is handled normally after connect() return successfully. 662 */ 663 664 sock->state = SS_CONNECTED; 665 err = 0; 666 out: 667 return err; 668 669 sock_error: 670 err = sock_error(sk) ? : -ECONNABORTED; 671 sock->state = SS_UNCONNECTED; 672 if (sk->sk_prot->disconnect(sk, flags)) 673 sock->state = SS_DISCONNECTING; 674 goto out; 675 } 676 EXPORT_SYMBOL(__inet_stream_connect); 677 678 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr, 679 int addr_len, int flags) 680 { 681 int err; 682 683 lock_sock(sock->sk); 684 err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0); 685 release_sock(sock->sk); 686 return err; 687 } 688 EXPORT_SYMBOL(inet_stream_connect); 689 690 /* 691 * Accept a pending connection. The TCP layer now gives BSD semantics. 692 */ 693 694 int inet_accept(struct socket *sock, struct socket *newsock, int flags, 695 bool kern) 696 { 697 struct sock *sk1 = sock->sk; 698 int err = -EINVAL; 699 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err, kern); 700 701 if (!sk2) 702 goto do_err; 703 704 lock_sock(sk2); 705 706 sock_rps_record_flow(sk2); 707 WARN_ON(!((1 << sk2->sk_state) & 708 (TCPF_ESTABLISHED | TCPF_SYN_RECV | 709 TCPF_CLOSE_WAIT | TCPF_CLOSE))); 710 711 sock_graft(sk2, newsock); 712 713 newsock->state = SS_CONNECTED; 714 err = 0; 715 release_sock(sk2); 716 do_err: 717 return err; 718 } 719 EXPORT_SYMBOL(inet_accept); 720 721 722 /* 723 * This does both peername and sockname. 724 */ 725 int inet_getname(struct socket *sock, struct sockaddr *uaddr, 726 int peer) 727 { 728 struct sock *sk = sock->sk; 729 struct inet_sock *inet = inet_sk(sk); 730 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr); 731 732 sin->sin_family = AF_INET; 733 if (peer) { 734 if (!inet->inet_dport || 735 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) && 736 peer == 1)) 737 return -ENOTCONN; 738 sin->sin_port = inet->inet_dport; 739 sin->sin_addr.s_addr = inet->inet_daddr; 740 } else { 741 __be32 addr = inet->inet_rcv_saddr; 742 if (!addr) 743 addr = inet->inet_saddr; 744 sin->sin_port = inet->inet_sport; 745 sin->sin_addr.s_addr = addr; 746 } 747 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 748 return sizeof(*sin); 749 } 750 EXPORT_SYMBOL(inet_getname); 751 752 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size) 753 { 754 struct sock *sk = sock->sk; 755 756 sock_rps_record_flow(sk); 757 758 /* We may need to bind the socket. */ 759 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind && 760 inet_autobind(sk)) 761 return -EAGAIN; 762 763 return sk->sk_prot->sendmsg(sk, msg, size); 764 } 765 EXPORT_SYMBOL(inet_sendmsg); 766 767 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, 768 size_t size, int flags) 769 { 770 struct sock *sk = sock->sk; 771 772 sock_rps_record_flow(sk); 773 774 /* We may need to bind the socket. */ 775 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind && 776 inet_autobind(sk)) 777 return -EAGAIN; 778 779 if (sk->sk_prot->sendpage) 780 return sk->sk_prot->sendpage(sk, page, offset, size, flags); 781 return sock_no_sendpage(sock, page, offset, size, flags); 782 } 783 EXPORT_SYMBOL(inet_sendpage); 784 785 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 786 int flags) 787 { 788 struct sock *sk = sock->sk; 789 int addr_len = 0; 790 int err; 791 792 if (likely(!(flags & MSG_ERRQUEUE))) 793 sock_rps_record_flow(sk); 794 795 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT, 796 flags & ~MSG_DONTWAIT, &addr_len); 797 if (err >= 0) 798 msg->msg_namelen = addr_len; 799 return err; 800 } 801 EXPORT_SYMBOL(inet_recvmsg); 802 803 int inet_shutdown(struct socket *sock, int how) 804 { 805 struct sock *sk = sock->sk; 806 int err = 0; 807 808 /* This should really check to make sure 809 * the socket is a TCP socket. (WHY AC...) 810 */ 811 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and 812 1->2 bit 2 snds. 813 2->3 */ 814 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */ 815 return -EINVAL; 816 817 lock_sock(sk); 818 if (sock->state == SS_CONNECTING) { 819 if ((1 << sk->sk_state) & 820 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE)) 821 sock->state = SS_DISCONNECTING; 822 else 823 sock->state = SS_CONNECTED; 824 } 825 826 switch (sk->sk_state) { 827 case TCP_CLOSE: 828 err = -ENOTCONN; 829 /* Hack to wake up other listeners, who can poll for 830 EPOLLHUP, even on eg. unconnected UDP sockets -- RR */ 831 /* fall through */ 832 default: 833 sk->sk_shutdown |= how; 834 if (sk->sk_prot->shutdown) 835 sk->sk_prot->shutdown(sk, how); 836 break; 837 838 /* Remaining two branches are temporary solution for missing 839 * close() in multithreaded environment. It is _not_ a good idea, 840 * but we have no choice until close() is repaired at VFS level. 841 */ 842 case TCP_LISTEN: 843 if (!(how & RCV_SHUTDOWN)) 844 break; 845 /* fall through */ 846 case TCP_SYN_SENT: 847 err = sk->sk_prot->disconnect(sk, O_NONBLOCK); 848 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED; 849 break; 850 } 851 852 /* Wake up anyone sleeping in poll. */ 853 sk->sk_state_change(sk); 854 release_sock(sk); 855 return err; 856 } 857 EXPORT_SYMBOL(inet_shutdown); 858 859 /* 860 * ioctl() calls you can issue on an INET socket. Most of these are 861 * device configuration and stuff and very rarely used. Some ioctls 862 * pass on to the socket itself. 863 * 864 * NOTE: I like the idea of a module for the config stuff. ie ifconfig 865 * loads the devconfigure module does its configuring and unloads it. 866 * There's a good 20K of config code hanging around the kernel. 867 */ 868 869 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 870 { 871 struct sock *sk = sock->sk; 872 int err = 0; 873 struct net *net = sock_net(sk); 874 void __user *p = (void __user *)arg; 875 struct ifreq ifr; 876 struct rtentry rt; 877 878 switch (cmd) { 879 case SIOCGSTAMP: 880 err = sock_get_timestamp(sk, (struct timeval __user *)arg); 881 break; 882 case SIOCGSTAMPNS: 883 err = sock_get_timestampns(sk, (struct timespec __user *)arg); 884 break; 885 case SIOCADDRT: 886 case SIOCDELRT: 887 if (copy_from_user(&rt, p, sizeof(struct rtentry))) 888 return -EFAULT; 889 err = ip_rt_ioctl(net, cmd, &rt); 890 break; 891 case SIOCRTMSG: 892 err = -EINVAL; 893 break; 894 case SIOCDARP: 895 case SIOCGARP: 896 case SIOCSARP: 897 err = arp_ioctl(net, cmd, (void __user *)arg); 898 break; 899 case SIOCGIFADDR: 900 case SIOCGIFBRDADDR: 901 case SIOCGIFNETMASK: 902 case SIOCGIFDSTADDR: 903 case SIOCGIFPFLAGS: 904 if (copy_from_user(&ifr, p, sizeof(struct ifreq))) 905 return -EFAULT; 906 err = devinet_ioctl(net, cmd, &ifr); 907 if (!err && copy_to_user(p, &ifr, sizeof(struct ifreq))) 908 err = -EFAULT; 909 break; 910 911 case SIOCSIFADDR: 912 case SIOCSIFBRDADDR: 913 case SIOCSIFNETMASK: 914 case SIOCSIFDSTADDR: 915 case SIOCSIFPFLAGS: 916 case SIOCSIFFLAGS: 917 if (copy_from_user(&ifr, p, sizeof(struct ifreq))) 918 return -EFAULT; 919 err = devinet_ioctl(net, cmd, &ifr); 920 break; 921 default: 922 if (sk->sk_prot->ioctl) 923 err = sk->sk_prot->ioctl(sk, cmd, arg); 924 else 925 err = -ENOIOCTLCMD; 926 break; 927 } 928 return err; 929 } 930 EXPORT_SYMBOL(inet_ioctl); 931 932 #ifdef CONFIG_COMPAT 933 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 934 { 935 struct sock *sk = sock->sk; 936 int err = -ENOIOCTLCMD; 937 938 if (sk->sk_prot->compat_ioctl) 939 err = sk->sk_prot->compat_ioctl(sk, cmd, arg); 940 941 return err; 942 } 943 #endif 944 945 const struct proto_ops inet_stream_ops = { 946 .family = PF_INET, 947 .owner = THIS_MODULE, 948 .release = inet_release, 949 .bind = inet_bind, 950 .connect = inet_stream_connect, 951 .socketpair = sock_no_socketpair, 952 .accept = inet_accept, 953 .getname = inet_getname, 954 .poll = tcp_poll, 955 .ioctl = inet_ioctl, 956 .listen = inet_listen, 957 .shutdown = inet_shutdown, 958 .setsockopt = sock_common_setsockopt, 959 .getsockopt = sock_common_getsockopt, 960 .sendmsg = inet_sendmsg, 961 .recvmsg = inet_recvmsg, 962 .mmap = sock_no_mmap, 963 .sendpage = inet_sendpage, 964 .splice_read = tcp_splice_read, 965 .read_sock = tcp_read_sock, 966 .sendmsg_locked = tcp_sendmsg_locked, 967 .sendpage_locked = tcp_sendpage_locked, 968 .peek_len = tcp_peek_len, 969 #ifdef CONFIG_COMPAT 970 .compat_setsockopt = compat_sock_common_setsockopt, 971 .compat_getsockopt = compat_sock_common_getsockopt, 972 .compat_ioctl = inet_compat_ioctl, 973 #endif 974 }; 975 EXPORT_SYMBOL(inet_stream_ops); 976 977 const struct proto_ops inet_dgram_ops = { 978 .family = PF_INET, 979 .owner = THIS_MODULE, 980 .release = inet_release, 981 .bind = inet_bind, 982 .connect = inet_dgram_connect, 983 .socketpair = sock_no_socketpair, 984 .accept = sock_no_accept, 985 .getname = inet_getname, 986 .poll = udp_poll, 987 .ioctl = inet_ioctl, 988 .listen = sock_no_listen, 989 .shutdown = inet_shutdown, 990 .setsockopt = sock_common_setsockopt, 991 .getsockopt = sock_common_getsockopt, 992 .sendmsg = inet_sendmsg, 993 .recvmsg = inet_recvmsg, 994 .mmap = sock_no_mmap, 995 .sendpage = inet_sendpage, 996 .set_peek_off = sk_set_peek_off, 997 #ifdef CONFIG_COMPAT 998 .compat_setsockopt = compat_sock_common_setsockopt, 999 .compat_getsockopt = compat_sock_common_getsockopt, 1000 .compat_ioctl = inet_compat_ioctl, 1001 #endif 1002 }; 1003 EXPORT_SYMBOL(inet_dgram_ops); 1004 1005 /* 1006 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without 1007 * udp_poll 1008 */ 1009 static const struct proto_ops inet_sockraw_ops = { 1010 .family = PF_INET, 1011 .owner = THIS_MODULE, 1012 .release = inet_release, 1013 .bind = inet_bind, 1014 .connect = inet_dgram_connect, 1015 .socketpair = sock_no_socketpair, 1016 .accept = sock_no_accept, 1017 .getname = inet_getname, 1018 .poll = datagram_poll, 1019 .ioctl = inet_ioctl, 1020 .listen = sock_no_listen, 1021 .shutdown = inet_shutdown, 1022 .setsockopt = sock_common_setsockopt, 1023 .getsockopt = sock_common_getsockopt, 1024 .sendmsg = inet_sendmsg, 1025 .recvmsg = inet_recvmsg, 1026 .mmap = sock_no_mmap, 1027 .sendpage = inet_sendpage, 1028 #ifdef CONFIG_COMPAT 1029 .compat_setsockopt = compat_sock_common_setsockopt, 1030 .compat_getsockopt = compat_sock_common_getsockopt, 1031 .compat_ioctl = inet_compat_ioctl, 1032 #endif 1033 }; 1034 1035 static const struct net_proto_family inet_family_ops = { 1036 .family = PF_INET, 1037 .create = inet_create, 1038 .owner = THIS_MODULE, 1039 }; 1040 1041 /* Upon startup we insert all the elements in inetsw_array[] into 1042 * the linked list inetsw. 1043 */ 1044 static struct inet_protosw inetsw_array[] = 1045 { 1046 { 1047 .type = SOCK_STREAM, 1048 .protocol = IPPROTO_TCP, 1049 .prot = &tcp_prot, 1050 .ops = &inet_stream_ops, 1051 .flags = INET_PROTOSW_PERMANENT | 1052 INET_PROTOSW_ICSK, 1053 }, 1054 1055 { 1056 .type = SOCK_DGRAM, 1057 .protocol = IPPROTO_UDP, 1058 .prot = &udp_prot, 1059 .ops = &inet_dgram_ops, 1060 .flags = INET_PROTOSW_PERMANENT, 1061 }, 1062 1063 { 1064 .type = SOCK_DGRAM, 1065 .protocol = IPPROTO_ICMP, 1066 .prot = &ping_prot, 1067 .ops = &inet_sockraw_ops, 1068 .flags = INET_PROTOSW_REUSE, 1069 }, 1070 1071 { 1072 .type = SOCK_RAW, 1073 .protocol = IPPROTO_IP, /* wild card */ 1074 .prot = &raw_prot, 1075 .ops = &inet_sockraw_ops, 1076 .flags = INET_PROTOSW_REUSE, 1077 } 1078 }; 1079 1080 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array) 1081 1082 void inet_register_protosw(struct inet_protosw *p) 1083 { 1084 struct list_head *lh; 1085 struct inet_protosw *answer; 1086 int protocol = p->protocol; 1087 struct list_head *last_perm; 1088 1089 spin_lock_bh(&inetsw_lock); 1090 1091 if (p->type >= SOCK_MAX) 1092 goto out_illegal; 1093 1094 /* If we are trying to override a permanent protocol, bail. */ 1095 last_perm = &inetsw[p->type]; 1096 list_for_each(lh, &inetsw[p->type]) { 1097 answer = list_entry(lh, struct inet_protosw, list); 1098 /* Check only the non-wild match. */ 1099 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0) 1100 break; 1101 if (protocol == answer->protocol) 1102 goto out_permanent; 1103 last_perm = lh; 1104 } 1105 1106 /* Add the new entry after the last permanent entry if any, so that 1107 * the new entry does not override a permanent entry when matched with 1108 * a wild-card protocol. But it is allowed to override any existing 1109 * non-permanent entry. This means that when we remove this entry, the 1110 * system automatically returns to the old behavior. 1111 */ 1112 list_add_rcu(&p->list, last_perm); 1113 out: 1114 spin_unlock_bh(&inetsw_lock); 1115 1116 return; 1117 1118 out_permanent: 1119 pr_err("Attempt to override permanent protocol %d\n", protocol); 1120 goto out; 1121 1122 out_illegal: 1123 pr_err("Ignoring attempt to register invalid socket type %d\n", 1124 p->type); 1125 goto out; 1126 } 1127 EXPORT_SYMBOL(inet_register_protosw); 1128 1129 void inet_unregister_protosw(struct inet_protosw *p) 1130 { 1131 if (INET_PROTOSW_PERMANENT & p->flags) { 1132 pr_err("Attempt to unregister permanent protocol %d\n", 1133 p->protocol); 1134 } else { 1135 spin_lock_bh(&inetsw_lock); 1136 list_del_rcu(&p->list); 1137 spin_unlock_bh(&inetsw_lock); 1138 1139 synchronize_net(); 1140 } 1141 } 1142 EXPORT_SYMBOL(inet_unregister_protosw); 1143 1144 static int inet_sk_reselect_saddr(struct sock *sk) 1145 { 1146 struct inet_sock *inet = inet_sk(sk); 1147 __be32 old_saddr = inet->inet_saddr; 1148 __be32 daddr = inet->inet_daddr; 1149 struct flowi4 *fl4; 1150 struct rtable *rt; 1151 __be32 new_saddr; 1152 struct ip_options_rcu *inet_opt; 1153 1154 inet_opt = rcu_dereference_protected(inet->inet_opt, 1155 lockdep_sock_is_held(sk)); 1156 if (inet_opt && inet_opt->opt.srr) 1157 daddr = inet_opt->opt.faddr; 1158 1159 /* Query new route. */ 1160 fl4 = &inet->cork.fl.u.ip4; 1161 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk), 1162 sk->sk_bound_dev_if, sk->sk_protocol, 1163 inet->inet_sport, inet->inet_dport, sk); 1164 if (IS_ERR(rt)) 1165 return PTR_ERR(rt); 1166 1167 sk_setup_caps(sk, &rt->dst); 1168 1169 new_saddr = fl4->saddr; 1170 1171 if (new_saddr == old_saddr) 1172 return 0; 1173 1174 if (sock_net(sk)->ipv4.sysctl_ip_dynaddr > 1) { 1175 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n", 1176 __func__, &old_saddr, &new_saddr); 1177 } 1178 1179 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr; 1180 1181 /* 1182 * XXX The only one ugly spot where we need to 1183 * XXX really change the sockets identity after 1184 * XXX it has entered the hashes. -DaveM 1185 * 1186 * Besides that, it does not check for connection 1187 * uniqueness. Wait for troubles. 1188 */ 1189 return __sk_prot_rehash(sk); 1190 } 1191 1192 int inet_sk_rebuild_header(struct sock *sk) 1193 { 1194 struct inet_sock *inet = inet_sk(sk); 1195 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0); 1196 __be32 daddr; 1197 struct ip_options_rcu *inet_opt; 1198 struct flowi4 *fl4; 1199 int err; 1200 1201 /* Route is OK, nothing to do. */ 1202 if (rt) 1203 return 0; 1204 1205 /* Reroute. */ 1206 rcu_read_lock(); 1207 inet_opt = rcu_dereference(inet->inet_opt); 1208 daddr = inet->inet_daddr; 1209 if (inet_opt && inet_opt->opt.srr) 1210 daddr = inet_opt->opt.faddr; 1211 rcu_read_unlock(); 1212 fl4 = &inet->cork.fl.u.ip4; 1213 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr, 1214 inet->inet_dport, inet->inet_sport, 1215 sk->sk_protocol, RT_CONN_FLAGS(sk), 1216 sk->sk_bound_dev_if); 1217 if (!IS_ERR(rt)) { 1218 err = 0; 1219 sk_setup_caps(sk, &rt->dst); 1220 } else { 1221 err = PTR_ERR(rt); 1222 1223 /* Routing failed... */ 1224 sk->sk_route_caps = 0; 1225 /* 1226 * Other protocols have to map its equivalent state to TCP_SYN_SENT. 1227 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme 1228 */ 1229 if (!sock_net(sk)->ipv4.sysctl_ip_dynaddr || 1230 sk->sk_state != TCP_SYN_SENT || 1231 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) || 1232 (err = inet_sk_reselect_saddr(sk)) != 0) 1233 sk->sk_err_soft = -err; 1234 } 1235 1236 return err; 1237 } 1238 EXPORT_SYMBOL(inet_sk_rebuild_header); 1239 1240 void inet_sk_set_state(struct sock *sk, int state) 1241 { 1242 trace_inet_sock_set_state(sk, sk->sk_state, state); 1243 sk->sk_state = state; 1244 } 1245 EXPORT_SYMBOL(inet_sk_set_state); 1246 1247 void inet_sk_state_store(struct sock *sk, int newstate) 1248 { 1249 trace_inet_sock_set_state(sk, sk->sk_state, newstate); 1250 smp_store_release(&sk->sk_state, newstate); 1251 } 1252 1253 struct sk_buff *inet_gso_segment(struct sk_buff *skb, 1254 netdev_features_t features) 1255 { 1256 bool udpfrag = false, fixedid = false, gso_partial, encap; 1257 struct sk_buff *segs = ERR_PTR(-EINVAL); 1258 const struct net_offload *ops; 1259 unsigned int offset = 0; 1260 struct iphdr *iph; 1261 int proto, tot_len; 1262 int nhoff; 1263 int ihl; 1264 int id; 1265 1266 skb_reset_network_header(skb); 1267 nhoff = skb_network_header(skb) - skb_mac_header(skb); 1268 if (unlikely(!pskb_may_pull(skb, sizeof(*iph)))) 1269 goto out; 1270 1271 iph = ip_hdr(skb); 1272 ihl = iph->ihl * 4; 1273 if (ihl < sizeof(*iph)) 1274 goto out; 1275 1276 id = ntohs(iph->id); 1277 proto = iph->protocol; 1278 1279 /* Warning: after this point, iph might be no longer valid */ 1280 if (unlikely(!pskb_may_pull(skb, ihl))) 1281 goto out; 1282 __skb_pull(skb, ihl); 1283 1284 encap = SKB_GSO_CB(skb)->encap_level > 0; 1285 if (encap) 1286 features &= skb->dev->hw_enc_features; 1287 SKB_GSO_CB(skb)->encap_level += ihl; 1288 1289 skb_reset_transport_header(skb); 1290 1291 segs = ERR_PTR(-EPROTONOSUPPORT); 1292 1293 if (!skb->encapsulation || encap) { 1294 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP); 1295 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID); 1296 1297 /* fixed ID is invalid if DF bit is not set */ 1298 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF))) 1299 goto out; 1300 } 1301 1302 ops = rcu_dereference(inet_offloads[proto]); 1303 if (likely(ops && ops->callbacks.gso_segment)) 1304 segs = ops->callbacks.gso_segment(skb, features); 1305 1306 if (IS_ERR_OR_NULL(segs)) 1307 goto out; 1308 1309 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL); 1310 1311 skb = segs; 1312 do { 1313 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff); 1314 if (udpfrag) { 1315 iph->frag_off = htons(offset >> 3); 1316 if (skb->next) 1317 iph->frag_off |= htons(IP_MF); 1318 offset += skb->len - nhoff - ihl; 1319 tot_len = skb->len - nhoff; 1320 } else if (skb_is_gso(skb)) { 1321 if (!fixedid) { 1322 iph->id = htons(id); 1323 id += skb_shinfo(skb)->gso_segs; 1324 } 1325 1326 if (gso_partial) 1327 tot_len = skb_shinfo(skb)->gso_size + 1328 SKB_GSO_CB(skb)->data_offset + 1329 skb->head - (unsigned char *)iph; 1330 else 1331 tot_len = skb->len - nhoff; 1332 } else { 1333 if (!fixedid) 1334 iph->id = htons(id++); 1335 tot_len = skb->len - nhoff; 1336 } 1337 iph->tot_len = htons(tot_len); 1338 ip_send_check(iph); 1339 if (encap) 1340 skb_reset_inner_headers(skb); 1341 skb->network_header = (u8 *)iph - skb->head; 1342 } while ((skb = skb->next)); 1343 1344 out: 1345 return segs; 1346 } 1347 EXPORT_SYMBOL(inet_gso_segment); 1348 1349 struct sk_buff **inet_gro_receive(struct sk_buff **head, struct sk_buff *skb) 1350 { 1351 const struct net_offload *ops; 1352 struct sk_buff **pp = NULL; 1353 struct sk_buff *p; 1354 const struct iphdr *iph; 1355 unsigned int hlen; 1356 unsigned int off; 1357 unsigned int id; 1358 int flush = 1; 1359 int proto; 1360 1361 off = skb_gro_offset(skb); 1362 hlen = off + sizeof(*iph); 1363 iph = skb_gro_header_fast(skb, off); 1364 if (skb_gro_header_hard(skb, hlen)) { 1365 iph = skb_gro_header_slow(skb, hlen, off); 1366 if (unlikely(!iph)) 1367 goto out; 1368 } 1369 1370 proto = iph->protocol; 1371 1372 rcu_read_lock(); 1373 ops = rcu_dereference(inet_offloads[proto]); 1374 if (!ops || !ops->callbacks.gro_receive) 1375 goto out_unlock; 1376 1377 if (*(u8 *)iph != 0x45) 1378 goto out_unlock; 1379 1380 if (ip_is_fragment(iph)) 1381 goto out_unlock; 1382 1383 if (unlikely(ip_fast_csum((u8 *)iph, 5))) 1384 goto out_unlock; 1385 1386 id = ntohl(*(__be32 *)&iph->id); 1387 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF)); 1388 id >>= 16; 1389 1390 for (p = *head; p; p = p->next) { 1391 struct iphdr *iph2; 1392 u16 flush_id; 1393 1394 if (!NAPI_GRO_CB(p)->same_flow) 1395 continue; 1396 1397 iph2 = (struct iphdr *)(p->data + off); 1398 /* The above works because, with the exception of the top 1399 * (inner most) layer, we only aggregate pkts with the same 1400 * hdr length so all the hdrs we'll need to verify will start 1401 * at the same offset. 1402 */ 1403 if ((iph->protocol ^ iph2->protocol) | 1404 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) | 1405 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) { 1406 NAPI_GRO_CB(p)->same_flow = 0; 1407 continue; 1408 } 1409 1410 /* All fields must match except length and checksum. */ 1411 NAPI_GRO_CB(p)->flush |= 1412 (iph->ttl ^ iph2->ttl) | 1413 (iph->tos ^ iph2->tos) | 1414 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF)); 1415 1416 NAPI_GRO_CB(p)->flush |= flush; 1417 1418 /* We need to store of the IP ID check to be included later 1419 * when we can verify that this packet does in fact belong 1420 * to a given flow. 1421 */ 1422 flush_id = (u16)(id - ntohs(iph2->id)); 1423 1424 /* This bit of code makes it much easier for us to identify 1425 * the cases where we are doing atomic vs non-atomic IP ID 1426 * checks. Specifically an atomic check can return IP ID 1427 * values 0 - 0xFFFF, while a non-atomic check can only 1428 * return 0 or 0xFFFF. 1429 */ 1430 if (!NAPI_GRO_CB(p)->is_atomic || 1431 !(iph->frag_off & htons(IP_DF))) { 1432 flush_id ^= NAPI_GRO_CB(p)->count; 1433 flush_id = flush_id ? 0xFFFF : 0; 1434 } 1435 1436 /* If the previous IP ID value was based on an atomic 1437 * datagram we can overwrite the value and ignore it. 1438 */ 1439 if (NAPI_GRO_CB(skb)->is_atomic) 1440 NAPI_GRO_CB(p)->flush_id = flush_id; 1441 else 1442 NAPI_GRO_CB(p)->flush_id |= flush_id; 1443 } 1444 1445 NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF)); 1446 NAPI_GRO_CB(skb)->flush |= flush; 1447 skb_set_network_header(skb, off); 1448 /* The above will be needed by the transport layer if there is one 1449 * immediately following this IP hdr. 1450 */ 1451 1452 /* Note : No need to call skb_gro_postpull_rcsum() here, 1453 * as we already checked checksum over ipv4 header was 0 1454 */ 1455 skb_gro_pull(skb, sizeof(*iph)); 1456 skb_set_transport_header(skb, skb_gro_offset(skb)); 1457 1458 pp = call_gro_receive(ops->callbacks.gro_receive, head, skb); 1459 1460 out_unlock: 1461 rcu_read_unlock(); 1462 1463 out: 1464 skb_gro_flush_final(skb, pp, flush); 1465 1466 return pp; 1467 } 1468 EXPORT_SYMBOL(inet_gro_receive); 1469 1470 static struct sk_buff **ipip_gro_receive(struct sk_buff **head, 1471 struct sk_buff *skb) 1472 { 1473 if (NAPI_GRO_CB(skb)->encap_mark) { 1474 NAPI_GRO_CB(skb)->flush = 1; 1475 return NULL; 1476 } 1477 1478 NAPI_GRO_CB(skb)->encap_mark = 1; 1479 1480 return inet_gro_receive(head, skb); 1481 } 1482 1483 #define SECONDS_PER_DAY 86400 1484 1485 /* inet_current_timestamp - Return IP network timestamp 1486 * 1487 * Return milliseconds since midnight in network byte order. 1488 */ 1489 __be32 inet_current_timestamp(void) 1490 { 1491 u32 secs; 1492 u32 msecs; 1493 struct timespec64 ts; 1494 1495 ktime_get_real_ts64(&ts); 1496 1497 /* Get secs since midnight. */ 1498 (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs); 1499 /* Convert to msecs. */ 1500 msecs = secs * MSEC_PER_SEC; 1501 /* Convert nsec to msec. */ 1502 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC; 1503 1504 /* Convert to network byte order. */ 1505 return htonl(msecs); 1506 } 1507 EXPORT_SYMBOL(inet_current_timestamp); 1508 1509 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len) 1510 { 1511 if (sk->sk_family == AF_INET) 1512 return ip_recv_error(sk, msg, len, addr_len); 1513 #if IS_ENABLED(CONFIG_IPV6) 1514 if (sk->sk_family == AF_INET6) 1515 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len); 1516 #endif 1517 return -EINVAL; 1518 } 1519 1520 int inet_gro_complete(struct sk_buff *skb, int nhoff) 1521 { 1522 __be16 newlen = htons(skb->len - nhoff); 1523 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff); 1524 const struct net_offload *ops; 1525 int proto = iph->protocol; 1526 int err = -ENOSYS; 1527 1528 if (skb->encapsulation) { 1529 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP)); 1530 skb_set_inner_network_header(skb, nhoff); 1531 } 1532 1533 csum_replace2(&iph->check, iph->tot_len, newlen); 1534 iph->tot_len = newlen; 1535 1536 rcu_read_lock(); 1537 ops = rcu_dereference(inet_offloads[proto]); 1538 if (WARN_ON(!ops || !ops->callbacks.gro_complete)) 1539 goto out_unlock; 1540 1541 /* Only need to add sizeof(*iph) to get to the next hdr below 1542 * because any hdr with option will have been flushed in 1543 * inet_gro_receive(). 1544 */ 1545 err = ops->callbacks.gro_complete(skb, nhoff + sizeof(*iph)); 1546 1547 out_unlock: 1548 rcu_read_unlock(); 1549 1550 return err; 1551 } 1552 EXPORT_SYMBOL(inet_gro_complete); 1553 1554 static int ipip_gro_complete(struct sk_buff *skb, int nhoff) 1555 { 1556 skb->encapsulation = 1; 1557 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4; 1558 return inet_gro_complete(skb, nhoff); 1559 } 1560 1561 int inet_ctl_sock_create(struct sock **sk, unsigned short family, 1562 unsigned short type, unsigned char protocol, 1563 struct net *net) 1564 { 1565 struct socket *sock; 1566 int rc = sock_create_kern(net, family, type, protocol, &sock); 1567 1568 if (rc == 0) { 1569 *sk = sock->sk; 1570 (*sk)->sk_allocation = GFP_ATOMIC; 1571 /* 1572 * Unhash it so that IP input processing does not even see it, 1573 * we do not wish this socket to see incoming packets. 1574 */ 1575 (*sk)->sk_prot->unhash(*sk); 1576 } 1577 return rc; 1578 } 1579 EXPORT_SYMBOL_GPL(inet_ctl_sock_create); 1580 1581 u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt) 1582 { 1583 return *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt); 1584 } 1585 EXPORT_SYMBOL_GPL(snmp_get_cpu_field); 1586 1587 unsigned long snmp_fold_field(void __percpu *mib, int offt) 1588 { 1589 unsigned long res = 0; 1590 int i; 1591 1592 for_each_possible_cpu(i) 1593 res += snmp_get_cpu_field(mib, i, offt); 1594 return res; 1595 } 1596 EXPORT_SYMBOL_GPL(snmp_fold_field); 1597 1598 #if BITS_PER_LONG==32 1599 1600 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt, 1601 size_t syncp_offset) 1602 { 1603 void *bhptr; 1604 struct u64_stats_sync *syncp; 1605 u64 v; 1606 unsigned int start; 1607 1608 bhptr = per_cpu_ptr(mib, cpu); 1609 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset); 1610 do { 1611 start = u64_stats_fetch_begin_irq(syncp); 1612 v = *(((u64 *)bhptr) + offt); 1613 } while (u64_stats_fetch_retry_irq(syncp, start)); 1614 1615 return v; 1616 } 1617 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64); 1618 1619 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset) 1620 { 1621 u64 res = 0; 1622 int cpu; 1623 1624 for_each_possible_cpu(cpu) { 1625 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset); 1626 } 1627 return res; 1628 } 1629 EXPORT_SYMBOL_GPL(snmp_fold_field64); 1630 #endif 1631 1632 #ifdef CONFIG_IP_MULTICAST 1633 static const struct net_protocol igmp_protocol = { 1634 .handler = igmp_rcv, 1635 .netns_ok = 1, 1636 }; 1637 #endif 1638 1639 /* thinking of making this const? Don't. 1640 * early_demux can change based on sysctl. 1641 */ 1642 static struct net_protocol tcp_protocol = { 1643 .early_demux = tcp_v4_early_demux, 1644 .early_demux_handler = tcp_v4_early_demux, 1645 .handler = tcp_v4_rcv, 1646 .err_handler = tcp_v4_err, 1647 .no_policy = 1, 1648 .netns_ok = 1, 1649 .icmp_strict_tag_validation = 1, 1650 }; 1651 1652 /* thinking of making this const? Don't. 1653 * early_demux can change based on sysctl. 1654 */ 1655 static struct net_protocol udp_protocol = { 1656 .early_demux = udp_v4_early_demux, 1657 .early_demux_handler = udp_v4_early_demux, 1658 .handler = udp_rcv, 1659 .err_handler = udp_err, 1660 .no_policy = 1, 1661 .netns_ok = 1, 1662 }; 1663 1664 static const struct net_protocol icmp_protocol = { 1665 .handler = icmp_rcv, 1666 .err_handler = icmp_err, 1667 .no_policy = 1, 1668 .netns_ok = 1, 1669 }; 1670 1671 static __net_init int ipv4_mib_init_net(struct net *net) 1672 { 1673 int i; 1674 1675 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib); 1676 if (!net->mib.tcp_statistics) 1677 goto err_tcp_mib; 1678 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib); 1679 if (!net->mib.ip_statistics) 1680 goto err_ip_mib; 1681 1682 for_each_possible_cpu(i) { 1683 struct ipstats_mib *af_inet_stats; 1684 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i); 1685 u64_stats_init(&af_inet_stats->syncp); 1686 } 1687 1688 net->mib.net_statistics = alloc_percpu(struct linux_mib); 1689 if (!net->mib.net_statistics) 1690 goto err_net_mib; 1691 net->mib.udp_statistics = alloc_percpu(struct udp_mib); 1692 if (!net->mib.udp_statistics) 1693 goto err_udp_mib; 1694 net->mib.udplite_statistics = alloc_percpu(struct udp_mib); 1695 if (!net->mib.udplite_statistics) 1696 goto err_udplite_mib; 1697 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib); 1698 if (!net->mib.icmp_statistics) 1699 goto err_icmp_mib; 1700 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib), 1701 GFP_KERNEL); 1702 if (!net->mib.icmpmsg_statistics) 1703 goto err_icmpmsg_mib; 1704 1705 tcp_mib_init(net); 1706 return 0; 1707 1708 err_icmpmsg_mib: 1709 free_percpu(net->mib.icmp_statistics); 1710 err_icmp_mib: 1711 free_percpu(net->mib.udplite_statistics); 1712 err_udplite_mib: 1713 free_percpu(net->mib.udp_statistics); 1714 err_udp_mib: 1715 free_percpu(net->mib.net_statistics); 1716 err_net_mib: 1717 free_percpu(net->mib.ip_statistics); 1718 err_ip_mib: 1719 free_percpu(net->mib.tcp_statistics); 1720 err_tcp_mib: 1721 return -ENOMEM; 1722 } 1723 1724 static __net_exit void ipv4_mib_exit_net(struct net *net) 1725 { 1726 kfree(net->mib.icmpmsg_statistics); 1727 free_percpu(net->mib.icmp_statistics); 1728 free_percpu(net->mib.udplite_statistics); 1729 free_percpu(net->mib.udp_statistics); 1730 free_percpu(net->mib.net_statistics); 1731 free_percpu(net->mib.ip_statistics); 1732 free_percpu(net->mib.tcp_statistics); 1733 } 1734 1735 static __net_initdata struct pernet_operations ipv4_mib_ops = { 1736 .init = ipv4_mib_init_net, 1737 .exit = ipv4_mib_exit_net, 1738 .async = true, 1739 }; 1740 1741 static int __init init_ipv4_mibs(void) 1742 { 1743 return register_pernet_subsys(&ipv4_mib_ops); 1744 } 1745 1746 static __net_init int inet_init_net(struct net *net) 1747 { 1748 /* 1749 * Set defaults for local port range 1750 */ 1751 seqlock_init(&net->ipv4.ip_local_ports.lock); 1752 net->ipv4.ip_local_ports.range[0] = 32768; 1753 net->ipv4.ip_local_ports.range[1] = 60999; 1754 1755 seqlock_init(&net->ipv4.ping_group_range.lock); 1756 /* 1757 * Sane defaults - nobody may create ping sockets. 1758 * Boot scripts should set this to distro-specific group. 1759 */ 1760 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1); 1761 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0); 1762 1763 /* Default values for sysctl-controlled parameters. 1764 * We set them here, in case sysctl is not compiled. 1765 */ 1766 net->ipv4.sysctl_ip_default_ttl = IPDEFTTL; 1767 net->ipv4.sysctl_ip_dynaddr = 0; 1768 net->ipv4.sysctl_ip_early_demux = 1; 1769 net->ipv4.sysctl_udp_early_demux = 1; 1770 net->ipv4.sysctl_tcp_early_demux = 1; 1771 #ifdef CONFIG_SYSCTL 1772 net->ipv4.sysctl_ip_prot_sock = PROT_SOCK; 1773 #endif 1774 1775 /* Some igmp sysctl, whose values are always used */ 1776 net->ipv4.sysctl_igmp_max_memberships = 20; 1777 net->ipv4.sysctl_igmp_max_msf = 10; 1778 /* IGMP reports for link-local multicast groups are enabled by default */ 1779 net->ipv4.sysctl_igmp_llm_reports = 1; 1780 net->ipv4.sysctl_igmp_qrv = 2; 1781 1782 return 0; 1783 } 1784 1785 static __net_exit void inet_exit_net(struct net *net) 1786 { 1787 } 1788 1789 static __net_initdata struct pernet_operations af_inet_ops = { 1790 .init = inet_init_net, 1791 .exit = inet_exit_net, 1792 .async = true, 1793 }; 1794 1795 static int __init init_inet_pernet_ops(void) 1796 { 1797 return register_pernet_subsys(&af_inet_ops); 1798 } 1799 1800 static int ipv4_proc_init(void); 1801 1802 /* 1803 * IP protocol layer initialiser 1804 */ 1805 1806 static struct packet_offload ip_packet_offload __read_mostly = { 1807 .type = cpu_to_be16(ETH_P_IP), 1808 .callbacks = { 1809 .gso_segment = inet_gso_segment, 1810 .gro_receive = inet_gro_receive, 1811 .gro_complete = inet_gro_complete, 1812 }, 1813 }; 1814 1815 static const struct net_offload ipip_offload = { 1816 .callbacks = { 1817 .gso_segment = inet_gso_segment, 1818 .gro_receive = ipip_gro_receive, 1819 .gro_complete = ipip_gro_complete, 1820 }, 1821 }; 1822 1823 static int __init ipip_offload_init(void) 1824 { 1825 return inet_add_offload(&ipip_offload, IPPROTO_IPIP); 1826 } 1827 1828 static int __init ipv4_offload_init(void) 1829 { 1830 /* 1831 * Add offloads 1832 */ 1833 if (udpv4_offload_init() < 0) 1834 pr_crit("%s: Cannot add UDP protocol offload\n", __func__); 1835 if (tcpv4_offload_init() < 0) 1836 pr_crit("%s: Cannot add TCP protocol offload\n", __func__); 1837 if (ipip_offload_init() < 0) 1838 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__); 1839 1840 dev_add_offload(&ip_packet_offload); 1841 return 0; 1842 } 1843 1844 fs_initcall(ipv4_offload_init); 1845 1846 static struct packet_type ip_packet_type __read_mostly = { 1847 .type = cpu_to_be16(ETH_P_IP), 1848 .func = ip_rcv, 1849 }; 1850 1851 static int __init inet_init(void) 1852 { 1853 struct inet_protosw *q; 1854 struct list_head *r; 1855 int rc = -EINVAL; 1856 1857 sock_skb_cb_check_size(sizeof(struct inet_skb_parm)); 1858 1859 rc = proto_register(&tcp_prot, 1); 1860 if (rc) 1861 goto out; 1862 1863 rc = proto_register(&udp_prot, 1); 1864 if (rc) 1865 goto out_unregister_tcp_proto; 1866 1867 rc = proto_register(&raw_prot, 1); 1868 if (rc) 1869 goto out_unregister_udp_proto; 1870 1871 rc = proto_register(&ping_prot, 1); 1872 if (rc) 1873 goto out_unregister_raw_proto; 1874 1875 /* 1876 * Tell SOCKET that we are alive... 1877 */ 1878 1879 (void)sock_register(&inet_family_ops); 1880 1881 #ifdef CONFIG_SYSCTL 1882 ip_static_sysctl_init(); 1883 #endif 1884 1885 /* 1886 * Add all the base protocols. 1887 */ 1888 1889 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0) 1890 pr_crit("%s: Cannot add ICMP protocol\n", __func__); 1891 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0) 1892 pr_crit("%s: Cannot add UDP protocol\n", __func__); 1893 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0) 1894 pr_crit("%s: Cannot add TCP protocol\n", __func__); 1895 #ifdef CONFIG_IP_MULTICAST 1896 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0) 1897 pr_crit("%s: Cannot add IGMP protocol\n", __func__); 1898 #endif 1899 1900 /* Register the socket-side information for inet_create. */ 1901 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r) 1902 INIT_LIST_HEAD(r); 1903 1904 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q) 1905 inet_register_protosw(q); 1906 1907 /* 1908 * Set the ARP module up 1909 */ 1910 1911 arp_init(); 1912 1913 /* 1914 * Set the IP module up 1915 */ 1916 1917 ip_init(); 1918 1919 /* Setup TCP slab cache for open requests. */ 1920 tcp_init(); 1921 1922 /* Setup UDP memory threshold */ 1923 udp_init(); 1924 1925 /* Add UDP-Lite (RFC 3828) */ 1926 udplite4_register(); 1927 1928 ping_init(); 1929 1930 /* 1931 * Set the ICMP layer up 1932 */ 1933 1934 if (icmp_init() < 0) 1935 panic("Failed to create the ICMP control socket.\n"); 1936 1937 /* 1938 * Initialise the multicast router 1939 */ 1940 #if defined(CONFIG_IP_MROUTE) 1941 if (ip_mr_init()) 1942 pr_crit("%s: Cannot init ipv4 mroute\n", __func__); 1943 #endif 1944 1945 if (init_inet_pernet_ops()) 1946 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__); 1947 /* 1948 * Initialise per-cpu ipv4 mibs 1949 */ 1950 1951 if (init_ipv4_mibs()) 1952 pr_crit("%s: Cannot init ipv4 mibs\n", __func__); 1953 1954 ipv4_proc_init(); 1955 1956 ipfrag_init(); 1957 1958 dev_add_pack(&ip_packet_type); 1959 1960 ip_tunnel_core_init(); 1961 1962 rc = 0; 1963 out: 1964 return rc; 1965 out_unregister_raw_proto: 1966 proto_unregister(&raw_prot); 1967 out_unregister_udp_proto: 1968 proto_unregister(&udp_prot); 1969 out_unregister_tcp_proto: 1970 proto_unregister(&tcp_prot); 1971 goto out; 1972 } 1973 1974 fs_initcall(inet_init); 1975 1976 /* ------------------------------------------------------------------------ */ 1977 1978 #ifdef CONFIG_PROC_FS 1979 static int __init ipv4_proc_init(void) 1980 { 1981 int rc = 0; 1982 1983 if (raw_proc_init()) 1984 goto out_raw; 1985 if (tcp4_proc_init()) 1986 goto out_tcp; 1987 if (udp4_proc_init()) 1988 goto out_udp; 1989 if (ping_proc_init()) 1990 goto out_ping; 1991 if (ip_misc_proc_init()) 1992 goto out_misc; 1993 out: 1994 return rc; 1995 out_misc: 1996 ping_proc_exit(); 1997 out_ping: 1998 udp4_proc_exit(); 1999 out_udp: 2000 tcp4_proc_exit(); 2001 out_tcp: 2002 raw_proc_exit(); 2003 out_raw: 2004 rc = -ENOMEM; 2005 goto out; 2006 } 2007 2008 #else /* CONFIG_PROC_FS */ 2009 static int __init ipv4_proc_init(void) 2010 { 2011 return 0; 2012 } 2013 #endif /* CONFIG_PROC_FS */ 2014