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 *uaddr_len, 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 *uaddr_len = sizeof(*sin); 749 return 0; 750 } 751 EXPORT_SYMBOL(inet_getname); 752 753 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size) 754 { 755 struct sock *sk = sock->sk; 756 757 sock_rps_record_flow(sk); 758 759 /* We may need to bind the socket. */ 760 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind && 761 inet_autobind(sk)) 762 return -EAGAIN; 763 764 return sk->sk_prot->sendmsg(sk, msg, size); 765 } 766 EXPORT_SYMBOL(inet_sendmsg); 767 768 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, 769 size_t size, int flags) 770 { 771 struct sock *sk = sock->sk; 772 773 sock_rps_record_flow(sk); 774 775 /* We may need to bind the socket. */ 776 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind && 777 inet_autobind(sk)) 778 return -EAGAIN; 779 780 if (sk->sk_prot->sendpage) 781 return sk->sk_prot->sendpage(sk, page, offset, size, flags); 782 return sock_no_sendpage(sock, page, offset, size, flags); 783 } 784 EXPORT_SYMBOL(inet_sendpage); 785 786 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 787 int flags) 788 { 789 struct sock *sk = sock->sk; 790 int addr_len = 0; 791 int err; 792 793 if (likely(!(flags & MSG_ERRQUEUE))) 794 sock_rps_record_flow(sk); 795 796 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT, 797 flags & ~MSG_DONTWAIT, &addr_len); 798 if (err >= 0) 799 msg->msg_namelen = addr_len; 800 return err; 801 } 802 EXPORT_SYMBOL(inet_recvmsg); 803 804 int inet_shutdown(struct socket *sock, int how) 805 { 806 struct sock *sk = sock->sk; 807 int err = 0; 808 809 /* This should really check to make sure 810 * the socket is a TCP socket. (WHY AC...) 811 */ 812 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and 813 1->2 bit 2 snds. 814 2->3 */ 815 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */ 816 return -EINVAL; 817 818 lock_sock(sk); 819 if (sock->state == SS_CONNECTING) { 820 if ((1 << sk->sk_state) & 821 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE)) 822 sock->state = SS_DISCONNECTING; 823 else 824 sock->state = SS_CONNECTED; 825 } 826 827 switch (sk->sk_state) { 828 case TCP_CLOSE: 829 err = -ENOTCONN; 830 /* Hack to wake up other listeners, who can poll for 831 POLLHUP, even on eg. unconnected UDP sockets -- RR */ 832 /* fall through */ 833 default: 834 sk->sk_shutdown |= how; 835 if (sk->sk_prot->shutdown) 836 sk->sk_prot->shutdown(sk, how); 837 break; 838 839 /* Remaining two branches are temporary solution for missing 840 * close() in multithreaded environment. It is _not_ a good idea, 841 * but we have no choice until close() is repaired at VFS level. 842 */ 843 case TCP_LISTEN: 844 if (!(how & RCV_SHUTDOWN)) 845 break; 846 /* fall through */ 847 case TCP_SYN_SENT: 848 err = sk->sk_prot->disconnect(sk, O_NONBLOCK); 849 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED; 850 break; 851 } 852 853 /* Wake up anyone sleeping in poll. */ 854 sk->sk_state_change(sk); 855 release_sock(sk); 856 return err; 857 } 858 EXPORT_SYMBOL(inet_shutdown); 859 860 /* 861 * ioctl() calls you can issue on an INET socket. Most of these are 862 * device configuration and stuff and very rarely used. Some ioctls 863 * pass on to the socket itself. 864 * 865 * NOTE: I like the idea of a module for the config stuff. ie ifconfig 866 * loads the devconfigure module does its configuring and unloads it. 867 * There's a good 20K of config code hanging around the kernel. 868 */ 869 870 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 871 { 872 struct sock *sk = sock->sk; 873 int err = 0; 874 struct net *net = sock_net(sk); 875 876 switch (cmd) { 877 case SIOCGSTAMP: 878 err = sock_get_timestamp(sk, (struct timeval __user *)arg); 879 break; 880 case SIOCGSTAMPNS: 881 err = sock_get_timestampns(sk, (struct timespec __user *)arg); 882 break; 883 case SIOCADDRT: 884 case SIOCDELRT: 885 case SIOCRTMSG: 886 err = ip_rt_ioctl(net, cmd, (void __user *)arg); 887 break; 888 case SIOCDARP: 889 case SIOCGARP: 890 case SIOCSARP: 891 err = arp_ioctl(net, cmd, (void __user *)arg); 892 break; 893 case SIOCGIFADDR: 894 case SIOCSIFADDR: 895 case SIOCGIFBRDADDR: 896 case SIOCSIFBRDADDR: 897 case SIOCGIFNETMASK: 898 case SIOCSIFNETMASK: 899 case SIOCGIFDSTADDR: 900 case SIOCSIFDSTADDR: 901 case SIOCSIFPFLAGS: 902 case SIOCGIFPFLAGS: 903 case SIOCSIFFLAGS: 904 err = devinet_ioctl(net, cmd, (void __user *)arg); 905 break; 906 default: 907 if (sk->sk_prot->ioctl) 908 err = sk->sk_prot->ioctl(sk, cmd, arg); 909 else 910 err = -ENOIOCTLCMD; 911 break; 912 } 913 return err; 914 } 915 EXPORT_SYMBOL(inet_ioctl); 916 917 #ifdef CONFIG_COMPAT 918 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 919 { 920 struct sock *sk = sock->sk; 921 int err = -ENOIOCTLCMD; 922 923 if (sk->sk_prot->compat_ioctl) 924 err = sk->sk_prot->compat_ioctl(sk, cmd, arg); 925 926 return err; 927 } 928 #endif 929 930 const struct proto_ops inet_stream_ops = { 931 .family = PF_INET, 932 .owner = THIS_MODULE, 933 .release = inet_release, 934 .bind = inet_bind, 935 .connect = inet_stream_connect, 936 .socketpair = sock_no_socketpair, 937 .accept = inet_accept, 938 .getname = inet_getname, 939 .poll = tcp_poll, 940 .ioctl = inet_ioctl, 941 .listen = inet_listen, 942 .shutdown = inet_shutdown, 943 .setsockopt = sock_common_setsockopt, 944 .getsockopt = sock_common_getsockopt, 945 .sendmsg = inet_sendmsg, 946 .recvmsg = inet_recvmsg, 947 .mmap = sock_no_mmap, 948 .sendpage = inet_sendpage, 949 .splice_read = tcp_splice_read, 950 .read_sock = tcp_read_sock, 951 .sendmsg_locked = tcp_sendmsg_locked, 952 .sendpage_locked = tcp_sendpage_locked, 953 .peek_len = tcp_peek_len, 954 #ifdef CONFIG_COMPAT 955 .compat_setsockopt = compat_sock_common_setsockopt, 956 .compat_getsockopt = compat_sock_common_getsockopt, 957 .compat_ioctl = inet_compat_ioctl, 958 #endif 959 }; 960 EXPORT_SYMBOL(inet_stream_ops); 961 962 const struct proto_ops inet_dgram_ops = { 963 .family = PF_INET, 964 .owner = THIS_MODULE, 965 .release = inet_release, 966 .bind = inet_bind, 967 .connect = inet_dgram_connect, 968 .socketpair = sock_no_socketpair, 969 .accept = sock_no_accept, 970 .getname = inet_getname, 971 .poll = udp_poll, 972 .ioctl = inet_ioctl, 973 .listen = sock_no_listen, 974 .shutdown = inet_shutdown, 975 .setsockopt = sock_common_setsockopt, 976 .getsockopt = sock_common_getsockopt, 977 .sendmsg = inet_sendmsg, 978 .recvmsg = inet_recvmsg, 979 .mmap = sock_no_mmap, 980 .sendpage = inet_sendpage, 981 .set_peek_off = sk_set_peek_off, 982 #ifdef CONFIG_COMPAT 983 .compat_setsockopt = compat_sock_common_setsockopt, 984 .compat_getsockopt = compat_sock_common_getsockopt, 985 .compat_ioctl = inet_compat_ioctl, 986 #endif 987 }; 988 EXPORT_SYMBOL(inet_dgram_ops); 989 990 /* 991 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without 992 * udp_poll 993 */ 994 static const struct proto_ops inet_sockraw_ops = { 995 .family = PF_INET, 996 .owner = THIS_MODULE, 997 .release = inet_release, 998 .bind = inet_bind, 999 .connect = inet_dgram_connect, 1000 .socketpair = sock_no_socketpair, 1001 .accept = sock_no_accept, 1002 .getname = inet_getname, 1003 .poll = datagram_poll, 1004 .ioctl = inet_ioctl, 1005 .listen = sock_no_listen, 1006 .shutdown = inet_shutdown, 1007 .setsockopt = sock_common_setsockopt, 1008 .getsockopt = sock_common_getsockopt, 1009 .sendmsg = inet_sendmsg, 1010 .recvmsg = inet_recvmsg, 1011 .mmap = sock_no_mmap, 1012 .sendpage = inet_sendpage, 1013 #ifdef CONFIG_COMPAT 1014 .compat_setsockopt = compat_sock_common_setsockopt, 1015 .compat_getsockopt = compat_sock_common_getsockopt, 1016 .compat_ioctl = inet_compat_ioctl, 1017 #endif 1018 }; 1019 1020 static const struct net_proto_family inet_family_ops = { 1021 .family = PF_INET, 1022 .create = inet_create, 1023 .owner = THIS_MODULE, 1024 }; 1025 1026 /* Upon startup we insert all the elements in inetsw_array[] into 1027 * the linked list inetsw. 1028 */ 1029 static struct inet_protosw inetsw_array[] = 1030 { 1031 { 1032 .type = SOCK_STREAM, 1033 .protocol = IPPROTO_TCP, 1034 .prot = &tcp_prot, 1035 .ops = &inet_stream_ops, 1036 .flags = INET_PROTOSW_PERMANENT | 1037 INET_PROTOSW_ICSK, 1038 }, 1039 1040 { 1041 .type = SOCK_DGRAM, 1042 .protocol = IPPROTO_UDP, 1043 .prot = &udp_prot, 1044 .ops = &inet_dgram_ops, 1045 .flags = INET_PROTOSW_PERMANENT, 1046 }, 1047 1048 { 1049 .type = SOCK_DGRAM, 1050 .protocol = IPPROTO_ICMP, 1051 .prot = &ping_prot, 1052 .ops = &inet_sockraw_ops, 1053 .flags = INET_PROTOSW_REUSE, 1054 }, 1055 1056 { 1057 .type = SOCK_RAW, 1058 .protocol = IPPROTO_IP, /* wild card */ 1059 .prot = &raw_prot, 1060 .ops = &inet_sockraw_ops, 1061 .flags = INET_PROTOSW_REUSE, 1062 } 1063 }; 1064 1065 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array) 1066 1067 void inet_register_protosw(struct inet_protosw *p) 1068 { 1069 struct list_head *lh; 1070 struct inet_protosw *answer; 1071 int protocol = p->protocol; 1072 struct list_head *last_perm; 1073 1074 spin_lock_bh(&inetsw_lock); 1075 1076 if (p->type >= SOCK_MAX) 1077 goto out_illegal; 1078 1079 /* If we are trying to override a permanent protocol, bail. */ 1080 last_perm = &inetsw[p->type]; 1081 list_for_each(lh, &inetsw[p->type]) { 1082 answer = list_entry(lh, struct inet_protosw, list); 1083 /* Check only the non-wild match. */ 1084 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0) 1085 break; 1086 if (protocol == answer->protocol) 1087 goto out_permanent; 1088 last_perm = lh; 1089 } 1090 1091 /* Add the new entry after the last permanent entry if any, so that 1092 * the new entry does not override a permanent entry when matched with 1093 * a wild-card protocol. But it is allowed to override any existing 1094 * non-permanent entry. This means that when we remove this entry, the 1095 * system automatically returns to the old behavior. 1096 */ 1097 list_add_rcu(&p->list, last_perm); 1098 out: 1099 spin_unlock_bh(&inetsw_lock); 1100 1101 return; 1102 1103 out_permanent: 1104 pr_err("Attempt to override permanent protocol %d\n", protocol); 1105 goto out; 1106 1107 out_illegal: 1108 pr_err("Ignoring attempt to register invalid socket type %d\n", 1109 p->type); 1110 goto out; 1111 } 1112 EXPORT_SYMBOL(inet_register_protosw); 1113 1114 void inet_unregister_protosw(struct inet_protosw *p) 1115 { 1116 if (INET_PROTOSW_PERMANENT & p->flags) { 1117 pr_err("Attempt to unregister permanent protocol %d\n", 1118 p->protocol); 1119 } else { 1120 spin_lock_bh(&inetsw_lock); 1121 list_del_rcu(&p->list); 1122 spin_unlock_bh(&inetsw_lock); 1123 1124 synchronize_net(); 1125 } 1126 } 1127 EXPORT_SYMBOL(inet_unregister_protosw); 1128 1129 static int inet_sk_reselect_saddr(struct sock *sk) 1130 { 1131 struct inet_sock *inet = inet_sk(sk); 1132 __be32 old_saddr = inet->inet_saddr; 1133 __be32 daddr = inet->inet_daddr; 1134 struct flowi4 *fl4; 1135 struct rtable *rt; 1136 __be32 new_saddr; 1137 struct ip_options_rcu *inet_opt; 1138 1139 inet_opt = rcu_dereference_protected(inet->inet_opt, 1140 lockdep_sock_is_held(sk)); 1141 if (inet_opt && inet_opt->opt.srr) 1142 daddr = inet_opt->opt.faddr; 1143 1144 /* Query new route. */ 1145 fl4 = &inet->cork.fl.u.ip4; 1146 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk), 1147 sk->sk_bound_dev_if, sk->sk_protocol, 1148 inet->inet_sport, inet->inet_dport, sk); 1149 if (IS_ERR(rt)) 1150 return PTR_ERR(rt); 1151 1152 sk_setup_caps(sk, &rt->dst); 1153 1154 new_saddr = fl4->saddr; 1155 1156 if (new_saddr == old_saddr) 1157 return 0; 1158 1159 if (sock_net(sk)->ipv4.sysctl_ip_dynaddr > 1) { 1160 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n", 1161 __func__, &old_saddr, &new_saddr); 1162 } 1163 1164 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr; 1165 1166 /* 1167 * XXX The only one ugly spot where we need to 1168 * XXX really change the sockets identity after 1169 * XXX it has entered the hashes. -DaveM 1170 * 1171 * Besides that, it does not check for connection 1172 * uniqueness. Wait for troubles. 1173 */ 1174 return __sk_prot_rehash(sk); 1175 } 1176 1177 int inet_sk_rebuild_header(struct sock *sk) 1178 { 1179 struct inet_sock *inet = inet_sk(sk); 1180 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0); 1181 __be32 daddr; 1182 struct ip_options_rcu *inet_opt; 1183 struct flowi4 *fl4; 1184 int err; 1185 1186 /* Route is OK, nothing to do. */ 1187 if (rt) 1188 return 0; 1189 1190 /* Reroute. */ 1191 rcu_read_lock(); 1192 inet_opt = rcu_dereference(inet->inet_opt); 1193 daddr = inet->inet_daddr; 1194 if (inet_opt && inet_opt->opt.srr) 1195 daddr = inet_opt->opt.faddr; 1196 rcu_read_unlock(); 1197 fl4 = &inet->cork.fl.u.ip4; 1198 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr, 1199 inet->inet_dport, inet->inet_sport, 1200 sk->sk_protocol, RT_CONN_FLAGS(sk), 1201 sk->sk_bound_dev_if); 1202 if (!IS_ERR(rt)) { 1203 err = 0; 1204 sk_setup_caps(sk, &rt->dst); 1205 } else { 1206 err = PTR_ERR(rt); 1207 1208 /* Routing failed... */ 1209 sk->sk_route_caps = 0; 1210 /* 1211 * Other protocols have to map its equivalent state to TCP_SYN_SENT. 1212 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme 1213 */ 1214 if (!sock_net(sk)->ipv4.sysctl_ip_dynaddr || 1215 sk->sk_state != TCP_SYN_SENT || 1216 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) || 1217 (err = inet_sk_reselect_saddr(sk)) != 0) 1218 sk->sk_err_soft = -err; 1219 } 1220 1221 return err; 1222 } 1223 EXPORT_SYMBOL(inet_sk_rebuild_header); 1224 1225 void inet_sk_set_state(struct sock *sk, int state) 1226 { 1227 trace_inet_sock_set_state(sk, sk->sk_state, state); 1228 sk->sk_state = state; 1229 } 1230 EXPORT_SYMBOL(inet_sk_set_state); 1231 1232 void inet_sk_state_store(struct sock *sk, int newstate) 1233 { 1234 trace_inet_sock_set_state(sk, sk->sk_state, newstate); 1235 smp_store_release(&sk->sk_state, newstate); 1236 } 1237 1238 struct sk_buff *inet_gso_segment(struct sk_buff *skb, 1239 netdev_features_t features) 1240 { 1241 bool udpfrag = false, fixedid = false, gso_partial, encap; 1242 struct sk_buff *segs = ERR_PTR(-EINVAL); 1243 const struct net_offload *ops; 1244 unsigned int offset = 0; 1245 struct iphdr *iph; 1246 int proto, tot_len; 1247 int nhoff; 1248 int ihl; 1249 int id; 1250 1251 skb_reset_network_header(skb); 1252 nhoff = skb_network_header(skb) - skb_mac_header(skb); 1253 if (unlikely(!pskb_may_pull(skb, sizeof(*iph)))) 1254 goto out; 1255 1256 iph = ip_hdr(skb); 1257 ihl = iph->ihl * 4; 1258 if (ihl < sizeof(*iph)) 1259 goto out; 1260 1261 id = ntohs(iph->id); 1262 proto = iph->protocol; 1263 1264 /* Warning: after this point, iph might be no longer valid */ 1265 if (unlikely(!pskb_may_pull(skb, ihl))) 1266 goto out; 1267 __skb_pull(skb, ihl); 1268 1269 encap = SKB_GSO_CB(skb)->encap_level > 0; 1270 if (encap) 1271 features &= skb->dev->hw_enc_features; 1272 SKB_GSO_CB(skb)->encap_level += ihl; 1273 1274 skb_reset_transport_header(skb); 1275 1276 segs = ERR_PTR(-EPROTONOSUPPORT); 1277 1278 if (!skb->encapsulation || encap) { 1279 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP); 1280 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID); 1281 1282 /* fixed ID is invalid if DF bit is not set */ 1283 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF))) 1284 goto out; 1285 } 1286 1287 ops = rcu_dereference(inet_offloads[proto]); 1288 if (likely(ops && ops->callbacks.gso_segment)) 1289 segs = ops->callbacks.gso_segment(skb, features); 1290 1291 if (IS_ERR_OR_NULL(segs)) 1292 goto out; 1293 1294 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL); 1295 1296 skb = segs; 1297 do { 1298 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff); 1299 if (udpfrag) { 1300 iph->frag_off = htons(offset >> 3); 1301 if (skb->next) 1302 iph->frag_off |= htons(IP_MF); 1303 offset += skb->len - nhoff - ihl; 1304 tot_len = skb->len - nhoff; 1305 } else if (skb_is_gso(skb)) { 1306 if (!fixedid) { 1307 iph->id = htons(id); 1308 id += skb_shinfo(skb)->gso_segs; 1309 } 1310 1311 if (gso_partial) 1312 tot_len = skb_shinfo(skb)->gso_size + 1313 SKB_GSO_CB(skb)->data_offset + 1314 skb->head - (unsigned char *)iph; 1315 else 1316 tot_len = skb->len - nhoff; 1317 } else { 1318 if (!fixedid) 1319 iph->id = htons(id++); 1320 tot_len = skb->len - nhoff; 1321 } 1322 iph->tot_len = htons(tot_len); 1323 ip_send_check(iph); 1324 if (encap) 1325 skb_reset_inner_headers(skb); 1326 skb->network_header = (u8 *)iph - skb->head; 1327 } while ((skb = skb->next)); 1328 1329 out: 1330 return segs; 1331 } 1332 EXPORT_SYMBOL(inet_gso_segment); 1333 1334 struct sk_buff **inet_gro_receive(struct sk_buff **head, struct sk_buff *skb) 1335 { 1336 const struct net_offload *ops; 1337 struct sk_buff **pp = NULL; 1338 struct sk_buff *p; 1339 const struct iphdr *iph; 1340 unsigned int hlen; 1341 unsigned int off; 1342 unsigned int id; 1343 int flush = 1; 1344 int proto; 1345 1346 off = skb_gro_offset(skb); 1347 hlen = off + sizeof(*iph); 1348 iph = skb_gro_header_fast(skb, off); 1349 if (skb_gro_header_hard(skb, hlen)) { 1350 iph = skb_gro_header_slow(skb, hlen, off); 1351 if (unlikely(!iph)) 1352 goto out; 1353 } 1354 1355 proto = iph->protocol; 1356 1357 rcu_read_lock(); 1358 ops = rcu_dereference(inet_offloads[proto]); 1359 if (!ops || !ops->callbacks.gro_receive) 1360 goto out_unlock; 1361 1362 if (*(u8 *)iph != 0x45) 1363 goto out_unlock; 1364 1365 if (ip_is_fragment(iph)) 1366 goto out_unlock; 1367 1368 if (unlikely(ip_fast_csum((u8 *)iph, 5))) 1369 goto out_unlock; 1370 1371 id = ntohl(*(__be32 *)&iph->id); 1372 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF)); 1373 id >>= 16; 1374 1375 for (p = *head; p; p = p->next) { 1376 struct iphdr *iph2; 1377 u16 flush_id; 1378 1379 if (!NAPI_GRO_CB(p)->same_flow) 1380 continue; 1381 1382 iph2 = (struct iphdr *)(p->data + off); 1383 /* The above works because, with the exception of the top 1384 * (inner most) layer, we only aggregate pkts with the same 1385 * hdr length so all the hdrs we'll need to verify will start 1386 * at the same offset. 1387 */ 1388 if ((iph->protocol ^ iph2->protocol) | 1389 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) | 1390 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) { 1391 NAPI_GRO_CB(p)->same_flow = 0; 1392 continue; 1393 } 1394 1395 /* All fields must match except length and checksum. */ 1396 NAPI_GRO_CB(p)->flush |= 1397 (iph->ttl ^ iph2->ttl) | 1398 (iph->tos ^ iph2->tos) | 1399 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF)); 1400 1401 NAPI_GRO_CB(p)->flush |= flush; 1402 1403 /* We need to store of the IP ID check to be included later 1404 * when we can verify that this packet does in fact belong 1405 * to a given flow. 1406 */ 1407 flush_id = (u16)(id - ntohs(iph2->id)); 1408 1409 /* This bit of code makes it much easier for us to identify 1410 * the cases where we are doing atomic vs non-atomic IP ID 1411 * checks. Specifically an atomic check can return IP ID 1412 * values 0 - 0xFFFF, while a non-atomic check can only 1413 * return 0 or 0xFFFF. 1414 */ 1415 if (!NAPI_GRO_CB(p)->is_atomic || 1416 !(iph->frag_off & htons(IP_DF))) { 1417 flush_id ^= NAPI_GRO_CB(p)->count; 1418 flush_id = flush_id ? 0xFFFF : 0; 1419 } 1420 1421 /* If the previous IP ID value was based on an atomic 1422 * datagram we can overwrite the value and ignore it. 1423 */ 1424 if (NAPI_GRO_CB(skb)->is_atomic) 1425 NAPI_GRO_CB(p)->flush_id = flush_id; 1426 else 1427 NAPI_GRO_CB(p)->flush_id |= flush_id; 1428 } 1429 1430 NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF)); 1431 NAPI_GRO_CB(skb)->flush |= flush; 1432 skb_set_network_header(skb, off); 1433 /* The above will be needed by the transport layer if there is one 1434 * immediately following this IP hdr. 1435 */ 1436 1437 /* Note : No need to call skb_gro_postpull_rcsum() here, 1438 * as we already checked checksum over ipv4 header was 0 1439 */ 1440 skb_gro_pull(skb, sizeof(*iph)); 1441 skb_set_transport_header(skb, skb_gro_offset(skb)); 1442 1443 pp = call_gro_receive(ops->callbacks.gro_receive, head, skb); 1444 1445 out_unlock: 1446 rcu_read_unlock(); 1447 1448 out: 1449 skb_gro_flush_final(skb, pp, flush); 1450 1451 return pp; 1452 } 1453 EXPORT_SYMBOL(inet_gro_receive); 1454 1455 static struct sk_buff **ipip_gro_receive(struct sk_buff **head, 1456 struct sk_buff *skb) 1457 { 1458 if (NAPI_GRO_CB(skb)->encap_mark) { 1459 NAPI_GRO_CB(skb)->flush = 1; 1460 return NULL; 1461 } 1462 1463 NAPI_GRO_CB(skb)->encap_mark = 1; 1464 1465 return inet_gro_receive(head, skb); 1466 } 1467 1468 #define SECONDS_PER_DAY 86400 1469 1470 /* inet_current_timestamp - Return IP network timestamp 1471 * 1472 * Return milliseconds since midnight in network byte order. 1473 */ 1474 __be32 inet_current_timestamp(void) 1475 { 1476 u32 secs; 1477 u32 msecs; 1478 struct timespec64 ts; 1479 1480 ktime_get_real_ts64(&ts); 1481 1482 /* Get secs since midnight. */ 1483 (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs); 1484 /* Convert to msecs. */ 1485 msecs = secs * MSEC_PER_SEC; 1486 /* Convert nsec to msec. */ 1487 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC; 1488 1489 /* Convert to network byte order. */ 1490 return htonl(msecs); 1491 } 1492 EXPORT_SYMBOL(inet_current_timestamp); 1493 1494 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len) 1495 { 1496 if (sk->sk_family == AF_INET) 1497 return ip_recv_error(sk, msg, len, addr_len); 1498 #if IS_ENABLED(CONFIG_IPV6) 1499 if (sk->sk_family == AF_INET6) 1500 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len); 1501 #endif 1502 return -EINVAL; 1503 } 1504 1505 int inet_gro_complete(struct sk_buff *skb, int nhoff) 1506 { 1507 __be16 newlen = htons(skb->len - nhoff); 1508 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff); 1509 const struct net_offload *ops; 1510 int proto = iph->protocol; 1511 int err = -ENOSYS; 1512 1513 if (skb->encapsulation) { 1514 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP)); 1515 skb_set_inner_network_header(skb, nhoff); 1516 } 1517 1518 csum_replace2(&iph->check, iph->tot_len, newlen); 1519 iph->tot_len = newlen; 1520 1521 rcu_read_lock(); 1522 ops = rcu_dereference(inet_offloads[proto]); 1523 if (WARN_ON(!ops || !ops->callbacks.gro_complete)) 1524 goto out_unlock; 1525 1526 /* Only need to add sizeof(*iph) to get to the next hdr below 1527 * because any hdr with option will have been flushed in 1528 * inet_gro_receive(). 1529 */ 1530 err = ops->callbacks.gro_complete(skb, nhoff + sizeof(*iph)); 1531 1532 out_unlock: 1533 rcu_read_unlock(); 1534 1535 return err; 1536 } 1537 EXPORT_SYMBOL(inet_gro_complete); 1538 1539 static int ipip_gro_complete(struct sk_buff *skb, int nhoff) 1540 { 1541 skb->encapsulation = 1; 1542 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4; 1543 return inet_gro_complete(skb, nhoff); 1544 } 1545 1546 int inet_ctl_sock_create(struct sock **sk, unsigned short family, 1547 unsigned short type, unsigned char protocol, 1548 struct net *net) 1549 { 1550 struct socket *sock; 1551 int rc = sock_create_kern(net, family, type, protocol, &sock); 1552 1553 if (rc == 0) { 1554 *sk = sock->sk; 1555 (*sk)->sk_allocation = GFP_ATOMIC; 1556 /* 1557 * Unhash it so that IP input processing does not even see it, 1558 * we do not wish this socket to see incoming packets. 1559 */ 1560 (*sk)->sk_prot->unhash(*sk); 1561 } 1562 return rc; 1563 } 1564 EXPORT_SYMBOL_GPL(inet_ctl_sock_create); 1565 1566 u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt) 1567 { 1568 return *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt); 1569 } 1570 EXPORT_SYMBOL_GPL(snmp_get_cpu_field); 1571 1572 unsigned long snmp_fold_field(void __percpu *mib, int offt) 1573 { 1574 unsigned long res = 0; 1575 int i; 1576 1577 for_each_possible_cpu(i) 1578 res += snmp_get_cpu_field(mib, i, offt); 1579 return res; 1580 } 1581 EXPORT_SYMBOL_GPL(snmp_fold_field); 1582 1583 #if BITS_PER_LONG==32 1584 1585 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt, 1586 size_t syncp_offset) 1587 { 1588 void *bhptr; 1589 struct u64_stats_sync *syncp; 1590 u64 v; 1591 unsigned int start; 1592 1593 bhptr = per_cpu_ptr(mib, cpu); 1594 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset); 1595 do { 1596 start = u64_stats_fetch_begin_irq(syncp); 1597 v = *(((u64 *)bhptr) + offt); 1598 } while (u64_stats_fetch_retry_irq(syncp, start)); 1599 1600 return v; 1601 } 1602 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64); 1603 1604 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset) 1605 { 1606 u64 res = 0; 1607 int cpu; 1608 1609 for_each_possible_cpu(cpu) { 1610 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset); 1611 } 1612 return res; 1613 } 1614 EXPORT_SYMBOL_GPL(snmp_fold_field64); 1615 #endif 1616 1617 #ifdef CONFIG_IP_MULTICAST 1618 static const struct net_protocol igmp_protocol = { 1619 .handler = igmp_rcv, 1620 .netns_ok = 1, 1621 }; 1622 #endif 1623 1624 /* thinking of making this const? Don't. 1625 * early_demux can change based on sysctl. 1626 */ 1627 static struct net_protocol tcp_protocol = { 1628 .early_demux = tcp_v4_early_demux, 1629 .early_demux_handler = tcp_v4_early_demux, 1630 .handler = tcp_v4_rcv, 1631 .err_handler = tcp_v4_err, 1632 .no_policy = 1, 1633 .netns_ok = 1, 1634 .icmp_strict_tag_validation = 1, 1635 }; 1636 1637 /* thinking of making this const? Don't. 1638 * early_demux can change based on sysctl. 1639 */ 1640 static struct net_protocol udp_protocol = { 1641 .early_demux = udp_v4_early_demux, 1642 .early_demux_handler = udp_v4_early_demux, 1643 .handler = udp_rcv, 1644 .err_handler = udp_err, 1645 .no_policy = 1, 1646 .netns_ok = 1, 1647 }; 1648 1649 static const struct net_protocol icmp_protocol = { 1650 .handler = icmp_rcv, 1651 .err_handler = icmp_err, 1652 .no_policy = 1, 1653 .netns_ok = 1, 1654 }; 1655 1656 static __net_init int ipv4_mib_init_net(struct net *net) 1657 { 1658 int i; 1659 1660 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib); 1661 if (!net->mib.tcp_statistics) 1662 goto err_tcp_mib; 1663 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib); 1664 if (!net->mib.ip_statistics) 1665 goto err_ip_mib; 1666 1667 for_each_possible_cpu(i) { 1668 struct ipstats_mib *af_inet_stats; 1669 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i); 1670 u64_stats_init(&af_inet_stats->syncp); 1671 } 1672 1673 net->mib.net_statistics = alloc_percpu(struct linux_mib); 1674 if (!net->mib.net_statistics) 1675 goto err_net_mib; 1676 net->mib.udp_statistics = alloc_percpu(struct udp_mib); 1677 if (!net->mib.udp_statistics) 1678 goto err_udp_mib; 1679 net->mib.udplite_statistics = alloc_percpu(struct udp_mib); 1680 if (!net->mib.udplite_statistics) 1681 goto err_udplite_mib; 1682 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib); 1683 if (!net->mib.icmp_statistics) 1684 goto err_icmp_mib; 1685 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib), 1686 GFP_KERNEL); 1687 if (!net->mib.icmpmsg_statistics) 1688 goto err_icmpmsg_mib; 1689 1690 tcp_mib_init(net); 1691 return 0; 1692 1693 err_icmpmsg_mib: 1694 free_percpu(net->mib.icmp_statistics); 1695 err_icmp_mib: 1696 free_percpu(net->mib.udplite_statistics); 1697 err_udplite_mib: 1698 free_percpu(net->mib.udp_statistics); 1699 err_udp_mib: 1700 free_percpu(net->mib.net_statistics); 1701 err_net_mib: 1702 free_percpu(net->mib.ip_statistics); 1703 err_ip_mib: 1704 free_percpu(net->mib.tcp_statistics); 1705 err_tcp_mib: 1706 return -ENOMEM; 1707 } 1708 1709 static __net_exit void ipv4_mib_exit_net(struct net *net) 1710 { 1711 kfree(net->mib.icmpmsg_statistics); 1712 free_percpu(net->mib.icmp_statistics); 1713 free_percpu(net->mib.udplite_statistics); 1714 free_percpu(net->mib.udp_statistics); 1715 free_percpu(net->mib.net_statistics); 1716 free_percpu(net->mib.ip_statistics); 1717 free_percpu(net->mib.tcp_statistics); 1718 } 1719 1720 static __net_initdata struct pernet_operations ipv4_mib_ops = { 1721 .init = ipv4_mib_init_net, 1722 .exit = ipv4_mib_exit_net, 1723 }; 1724 1725 static int __init init_ipv4_mibs(void) 1726 { 1727 return register_pernet_subsys(&ipv4_mib_ops); 1728 } 1729 1730 static __net_init int inet_init_net(struct net *net) 1731 { 1732 /* 1733 * Set defaults for local port range 1734 */ 1735 seqlock_init(&net->ipv4.ip_local_ports.lock); 1736 net->ipv4.ip_local_ports.range[0] = 32768; 1737 net->ipv4.ip_local_ports.range[1] = 60999; 1738 1739 seqlock_init(&net->ipv4.ping_group_range.lock); 1740 /* 1741 * Sane defaults - nobody may create ping sockets. 1742 * Boot scripts should set this to distro-specific group. 1743 */ 1744 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1); 1745 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0); 1746 1747 /* Default values for sysctl-controlled parameters. 1748 * We set them here, in case sysctl is not compiled. 1749 */ 1750 net->ipv4.sysctl_ip_default_ttl = IPDEFTTL; 1751 net->ipv4.sysctl_ip_dynaddr = 0; 1752 net->ipv4.sysctl_ip_early_demux = 1; 1753 net->ipv4.sysctl_udp_early_demux = 1; 1754 net->ipv4.sysctl_tcp_early_demux = 1; 1755 #ifdef CONFIG_SYSCTL 1756 net->ipv4.sysctl_ip_prot_sock = PROT_SOCK; 1757 #endif 1758 1759 /* Some igmp sysctl, whose values are always used */ 1760 net->ipv4.sysctl_igmp_max_memberships = 20; 1761 net->ipv4.sysctl_igmp_max_msf = 10; 1762 /* IGMP reports for link-local multicast groups are enabled by default */ 1763 net->ipv4.sysctl_igmp_llm_reports = 1; 1764 net->ipv4.sysctl_igmp_qrv = 2; 1765 1766 return 0; 1767 } 1768 1769 static __net_exit void inet_exit_net(struct net *net) 1770 { 1771 } 1772 1773 static __net_initdata struct pernet_operations af_inet_ops = { 1774 .init = inet_init_net, 1775 .exit = inet_exit_net, 1776 }; 1777 1778 static int __init init_inet_pernet_ops(void) 1779 { 1780 return register_pernet_subsys(&af_inet_ops); 1781 } 1782 1783 static int ipv4_proc_init(void); 1784 1785 /* 1786 * IP protocol layer initialiser 1787 */ 1788 1789 static struct packet_offload ip_packet_offload __read_mostly = { 1790 .type = cpu_to_be16(ETH_P_IP), 1791 .callbacks = { 1792 .gso_segment = inet_gso_segment, 1793 .gro_receive = inet_gro_receive, 1794 .gro_complete = inet_gro_complete, 1795 }, 1796 }; 1797 1798 static const struct net_offload ipip_offload = { 1799 .callbacks = { 1800 .gso_segment = inet_gso_segment, 1801 .gro_receive = ipip_gro_receive, 1802 .gro_complete = ipip_gro_complete, 1803 }, 1804 }; 1805 1806 static int __init ipip_offload_init(void) 1807 { 1808 return inet_add_offload(&ipip_offload, IPPROTO_IPIP); 1809 } 1810 1811 static int __init ipv4_offload_init(void) 1812 { 1813 /* 1814 * Add offloads 1815 */ 1816 if (udpv4_offload_init() < 0) 1817 pr_crit("%s: Cannot add UDP protocol offload\n", __func__); 1818 if (tcpv4_offload_init() < 0) 1819 pr_crit("%s: Cannot add TCP protocol offload\n", __func__); 1820 if (ipip_offload_init() < 0) 1821 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__); 1822 1823 dev_add_offload(&ip_packet_offload); 1824 return 0; 1825 } 1826 1827 fs_initcall(ipv4_offload_init); 1828 1829 static struct packet_type ip_packet_type __read_mostly = { 1830 .type = cpu_to_be16(ETH_P_IP), 1831 .func = ip_rcv, 1832 }; 1833 1834 static int __init inet_init(void) 1835 { 1836 struct inet_protosw *q; 1837 struct list_head *r; 1838 int rc = -EINVAL; 1839 1840 sock_skb_cb_check_size(sizeof(struct inet_skb_parm)); 1841 1842 rc = proto_register(&tcp_prot, 1); 1843 if (rc) 1844 goto out; 1845 1846 rc = proto_register(&udp_prot, 1); 1847 if (rc) 1848 goto out_unregister_tcp_proto; 1849 1850 rc = proto_register(&raw_prot, 1); 1851 if (rc) 1852 goto out_unregister_udp_proto; 1853 1854 rc = proto_register(&ping_prot, 1); 1855 if (rc) 1856 goto out_unregister_raw_proto; 1857 1858 /* 1859 * Tell SOCKET that we are alive... 1860 */ 1861 1862 (void)sock_register(&inet_family_ops); 1863 1864 #ifdef CONFIG_SYSCTL 1865 ip_static_sysctl_init(); 1866 #endif 1867 1868 /* 1869 * Add all the base protocols. 1870 */ 1871 1872 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0) 1873 pr_crit("%s: Cannot add ICMP protocol\n", __func__); 1874 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0) 1875 pr_crit("%s: Cannot add UDP protocol\n", __func__); 1876 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0) 1877 pr_crit("%s: Cannot add TCP protocol\n", __func__); 1878 #ifdef CONFIG_IP_MULTICAST 1879 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0) 1880 pr_crit("%s: Cannot add IGMP protocol\n", __func__); 1881 #endif 1882 1883 /* Register the socket-side information for inet_create. */ 1884 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r) 1885 INIT_LIST_HEAD(r); 1886 1887 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q) 1888 inet_register_protosw(q); 1889 1890 /* 1891 * Set the ARP module up 1892 */ 1893 1894 arp_init(); 1895 1896 /* 1897 * Set the IP module up 1898 */ 1899 1900 ip_init(); 1901 1902 /* Setup TCP slab cache for open requests. */ 1903 tcp_init(); 1904 1905 /* Setup UDP memory threshold */ 1906 udp_init(); 1907 1908 /* Add UDP-Lite (RFC 3828) */ 1909 udplite4_register(); 1910 1911 ping_init(); 1912 1913 /* 1914 * Set the ICMP layer up 1915 */ 1916 1917 if (icmp_init() < 0) 1918 panic("Failed to create the ICMP control socket.\n"); 1919 1920 /* 1921 * Initialise the multicast router 1922 */ 1923 #if defined(CONFIG_IP_MROUTE) 1924 if (ip_mr_init()) 1925 pr_crit("%s: Cannot init ipv4 mroute\n", __func__); 1926 #endif 1927 1928 if (init_inet_pernet_ops()) 1929 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__); 1930 /* 1931 * Initialise per-cpu ipv4 mibs 1932 */ 1933 1934 if (init_ipv4_mibs()) 1935 pr_crit("%s: Cannot init ipv4 mibs\n", __func__); 1936 1937 ipv4_proc_init(); 1938 1939 ipfrag_init(); 1940 1941 dev_add_pack(&ip_packet_type); 1942 1943 ip_tunnel_core_init(); 1944 1945 rc = 0; 1946 out: 1947 return rc; 1948 out_unregister_raw_proto: 1949 proto_unregister(&raw_prot); 1950 out_unregister_udp_proto: 1951 proto_unregister(&udp_prot); 1952 out_unregister_tcp_proto: 1953 proto_unregister(&tcp_prot); 1954 goto out; 1955 } 1956 1957 fs_initcall(inet_init); 1958 1959 /* ------------------------------------------------------------------------ */ 1960 1961 #ifdef CONFIG_PROC_FS 1962 static int __init ipv4_proc_init(void) 1963 { 1964 int rc = 0; 1965 1966 if (raw_proc_init()) 1967 goto out_raw; 1968 if (tcp4_proc_init()) 1969 goto out_tcp; 1970 if (udp4_proc_init()) 1971 goto out_udp; 1972 if (ping_proc_init()) 1973 goto out_ping; 1974 if (ip_misc_proc_init()) 1975 goto out_misc; 1976 out: 1977 return rc; 1978 out_misc: 1979 ping_proc_exit(); 1980 out_ping: 1981 udp4_proc_exit(); 1982 out_udp: 1983 tcp4_proc_exit(); 1984 out_tcp: 1985 raw_proc_exit(); 1986 out_raw: 1987 rc = -ENOMEM; 1988 goto out; 1989 } 1990 1991 #else /* CONFIG_PROC_FS */ 1992 static int __init ipv4_proc_init(void) 1993 { 1994 return 0; 1995 } 1996 #endif /* CONFIG_PROC_FS */ 1997