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 EPOLLHUP, 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 void __user *p = (void __user *)arg; 876 struct ifreq ifr; 877 struct rtentry rt; 878 879 switch (cmd) { 880 case SIOCGSTAMP: 881 err = sock_get_timestamp(sk, (struct timeval __user *)arg); 882 break; 883 case SIOCGSTAMPNS: 884 err = sock_get_timestampns(sk, (struct timespec __user *)arg); 885 break; 886 case SIOCADDRT: 887 case SIOCDELRT: 888 if (copy_from_user(&rt, p, sizeof(struct rtentry))) 889 return -EFAULT; 890 err = ip_rt_ioctl(net, cmd, &rt); 891 break; 892 case SIOCRTMSG: 893 err = -EINVAL; 894 break; 895 case SIOCDARP: 896 case SIOCGARP: 897 case SIOCSARP: 898 err = arp_ioctl(net, cmd, (void __user *)arg); 899 break; 900 case SIOCGIFADDR: 901 case SIOCGIFBRDADDR: 902 case SIOCGIFNETMASK: 903 case SIOCGIFDSTADDR: 904 case SIOCGIFPFLAGS: 905 if (copy_from_user(&ifr, p, sizeof(struct ifreq))) 906 return -EFAULT; 907 err = devinet_ioctl(net, cmd, &ifr); 908 if (!err && copy_to_user(p, &ifr, sizeof(struct ifreq))) 909 err = -EFAULT; 910 break; 911 912 case SIOCSIFADDR: 913 case SIOCSIFBRDADDR: 914 case SIOCSIFNETMASK: 915 case SIOCSIFDSTADDR: 916 case SIOCSIFPFLAGS: 917 case SIOCSIFFLAGS: 918 if (copy_from_user(&ifr, p, sizeof(struct ifreq))) 919 return -EFAULT; 920 err = devinet_ioctl(net, cmd, &ifr); 921 break; 922 default: 923 if (sk->sk_prot->ioctl) 924 err = sk->sk_prot->ioctl(sk, cmd, arg); 925 else 926 err = -ENOIOCTLCMD; 927 break; 928 } 929 return err; 930 } 931 EXPORT_SYMBOL(inet_ioctl); 932 933 #ifdef CONFIG_COMPAT 934 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 935 { 936 struct sock *sk = sock->sk; 937 int err = -ENOIOCTLCMD; 938 939 if (sk->sk_prot->compat_ioctl) 940 err = sk->sk_prot->compat_ioctl(sk, cmd, arg); 941 942 return err; 943 } 944 #endif 945 946 const struct proto_ops inet_stream_ops = { 947 .family = PF_INET, 948 .owner = THIS_MODULE, 949 .release = inet_release, 950 .bind = inet_bind, 951 .connect = inet_stream_connect, 952 .socketpair = sock_no_socketpair, 953 .accept = inet_accept, 954 .getname = inet_getname, 955 .poll = tcp_poll, 956 .ioctl = inet_ioctl, 957 .listen = inet_listen, 958 .shutdown = inet_shutdown, 959 .setsockopt = sock_common_setsockopt, 960 .getsockopt = sock_common_getsockopt, 961 .sendmsg = inet_sendmsg, 962 .recvmsg = inet_recvmsg, 963 .mmap = sock_no_mmap, 964 .sendpage = inet_sendpage, 965 .splice_read = tcp_splice_read, 966 .read_sock = tcp_read_sock, 967 .sendmsg_locked = tcp_sendmsg_locked, 968 .sendpage_locked = tcp_sendpage_locked, 969 .peek_len = tcp_peek_len, 970 #ifdef CONFIG_COMPAT 971 .compat_setsockopt = compat_sock_common_setsockopt, 972 .compat_getsockopt = compat_sock_common_getsockopt, 973 .compat_ioctl = inet_compat_ioctl, 974 #endif 975 }; 976 EXPORT_SYMBOL(inet_stream_ops); 977 978 const struct proto_ops inet_dgram_ops = { 979 .family = PF_INET, 980 .owner = THIS_MODULE, 981 .release = inet_release, 982 .bind = inet_bind, 983 .connect = inet_dgram_connect, 984 .socketpair = sock_no_socketpair, 985 .accept = sock_no_accept, 986 .getname = inet_getname, 987 .poll = udp_poll, 988 .ioctl = inet_ioctl, 989 .listen = sock_no_listen, 990 .shutdown = inet_shutdown, 991 .setsockopt = sock_common_setsockopt, 992 .getsockopt = sock_common_getsockopt, 993 .sendmsg = inet_sendmsg, 994 .recvmsg = inet_recvmsg, 995 .mmap = sock_no_mmap, 996 .sendpage = inet_sendpage, 997 .set_peek_off = sk_set_peek_off, 998 #ifdef CONFIG_COMPAT 999 .compat_setsockopt = compat_sock_common_setsockopt, 1000 .compat_getsockopt = compat_sock_common_getsockopt, 1001 .compat_ioctl = inet_compat_ioctl, 1002 #endif 1003 }; 1004 EXPORT_SYMBOL(inet_dgram_ops); 1005 1006 /* 1007 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without 1008 * udp_poll 1009 */ 1010 static const struct proto_ops inet_sockraw_ops = { 1011 .family = PF_INET, 1012 .owner = THIS_MODULE, 1013 .release = inet_release, 1014 .bind = inet_bind, 1015 .connect = inet_dgram_connect, 1016 .socketpair = sock_no_socketpair, 1017 .accept = sock_no_accept, 1018 .getname = inet_getname, 1019 .poll = datagram_poll, 1020 .ioctl = inet_ioctl, 1021 .listen = sock_no_listen, 1022 .shutdown = inet_shutdown, 1023 .setsockopt = sock_common_setsockopt, 1024 .getsockopt = sock_common_getsockopt, 1025 .sendmsg = inet_sendmsg, 1026 .recvmsg = inet_recvmsg, 1027 .mmap = sock_no_mmap, 1028 .sendpage = inet_sendpage, 1029 #ifdef CONFIG_COMPAT 1030 .compat_setsockopt = compat_sock_common_setsockopt, 1031 .compat_getsockopt = compat_sock_common_getsockopt, 1032 .compat_ioctl = inet_compat_ioctl, 1033 #endif 1034 }; 1035 1036 static const struct net_proto_family inet_family_ops = { 1037 .family = PF_INET, 1038 .create = inet_create, 1039 .owner = THIS_MODULE, 1040 }; 1041 1042 /* Upon startup we insert all the elements in inetsw_array[] into 1043 * the linked list inetsw. 1044 */ 1045 static struct inet_protosw inetsw_array[] = 1046 { 1047 { 1048 .type = SOCK_STREAM, 1049 .protocol = IPPROTO_TCP, 1050 .prot = &tcp_prot, 1051 .ops = &inet_stream_ops, 1052 .flags = INET_PROTOSW_PERMANENT | 1053 INET_PROTOSW_ICSK, 1054 }, 1055 1056 { 1057 .type = SOCK_DGRAM, 1058 .protocol = IPPROTO_UDP, 1059 .prot = &udp_prot, 1060 .ops = &inet_dgram_ops, 1061 .flags = INET_PROTOSW_PERMANENT, 1062 }, 1063 1064 { 1065 .type = SOCK_DGRAM, 1066 .protocol = IPPROTO_ICMP, 1067 .prot = &ping_prot, 1068 .ops = &inet_sockraw_ops, 1069 .flags = INET_PROTOSW_REUSE, 1070 }, 1071 1072 { 1073 .type = SOCK_RAW, 1074 .protocol = IPPROTO_IP, /* wild card */ 1075 .prot = &raw_prot, 1076 .ops = &inet_sockraw_ops, 1077 .flags = INET_PROTOSW_REUSE, 1078 } 1079 }; 1080 1081 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array) 1082 1083 void inet_register_protosw(struct inet_protosw *p) 1084 { 1085 struct list_head *lh; 1086 struct inet_protosw *answer; 1087 int protocol = p->protocol; 1088 struct list_head *last_perm; 1089 1090 spin_lock_bh(&inetsw_lock); 1091 1092 if (p->type >= SOCK_MAX) 1093 goto out_illegal; 1094 1095 /* If we are trying to override a permanent protocol, bail. */ 1096 last_perm = &inetsw[p->type]; 1097 list_for_each(lh, &inetsw[p->type]) { 1098 answer = list_entry(lh, struct inet_protosw, list); 1099 /* Check only the non-wild match. */ 1100 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0) 1101 break; 1102 if (protocol == answer->protocol) 1103 goto out_permanent; 1104 last_perm = lh; 1105 } 1106 1107 /* Add the new entry after the last permanent entry if any, so that 1108 * the new entry does not override a permanent entry when matched with 1109 * a wild-card protocol. But it is allowed to override any existing 1110 * non-permanent entry. This means that when we remove this entry, the 1111 * system automatically returns to the old behavior. 1112 */ 1113 list_add_rcu(&p->list, last_perm); 1114 out: 1115 spin_unlock_bh(&inetsw_lock); 1116 1117 return; 1118 1119 out_permanent: 1120 pr_err("Attempt to override permanent protocol %d\n", protocol); 1121 goto out; 1122 1123 out_illegal: 1124 pr_err("Ignoring attempt to register invalid socket type %d\n", 1125 p->type); 1126 goto out; 1127 } 1128 EXPORT_SYMBOL(inet_register_protosw); 1129 1130 void inet_unregister_protosw(struct inet_protosw *p) 1131 { 1132 if (INET_PROTOSW_PERMANENT & p->flags) { 1133 pr_err("Attempt to unregister permanent protocol %d\n", 1134 p->protocol); 1135 } else { 1136 spin_lock_bh(&inetsw_lock); 1137 list_del_rcu(&p->list); 1138 spin_unlock_bh(&inetsw_lock); 1139 1140 synchronize_net(); 1141 } 1142 } 1143 EXPORT_SYMBOL(inet_unregister_protosw); 1144 1145 static int inet_sk_reselect_saddr(struct sock *sk) 1146 { 1147 struct inet_sock *inet = inet_sk(sk); 1148 __be32 old_saddr = inet->inet_saddr; 1149 __be32 daddr = inet->inet_daddr; 1150 struct flowi4 *fl4; 1151 struct rtable *rt; 1152 __be32 new_saddr; 1153 struct ip_options_rcu *inet_opt; 1154 1155 inet_opt = rcu_dereference_protected(inet->inet_opt, 1156 lockdep_sock_is_held(sk)); 1157 if (inet_opt && inet_opt->opt.srr) 1158 daddr = inet_opt->opt.faddr; 1159 1160 /* Query new route. */ 1161 fl4 = &inet->cork.fl.u.ip4; 1162 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk), 1163 sk->sk_bound_dev_if, sk->sk_protocol, 1164 inet->inet_sport, inet->inet_dport, sk); 1165 if (IS_ERR(rt)) 1166 return PTR_ERR(rt); 1167 1168 sk_setup_caps(sk, &rt->dst); 1169 1170 new_saddr = fl4->saddr; 1171 1172 if (new_saddr == old_saddr) 1173 return 0; 1174 1175 if (sock_net(sk)->ipv4.sysctl_ip_dynaddr > 1) { 1176 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n", 1177 __func__, &old_saddr, &new_saddr); 1178 } 1179 1180 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr; 1181 1182 /* 1183 * XXX The only one ugly spot where we need to 1184 * XXX really change the sockets identity after 1185 * XXX it has entered the hashes. -DaveM 1186 * 1187 * Besides that, it does not check for connection 1188 * uniqueness. Wait for troubles. 1189 */ 1190 return __sk_prot_rehash(sk); 1191 } 1192 1193 int inet_sk_rebuild_header(struct sock *sk) 1194 { 1195 struct inet_sock *inet = inet_sk(sk); 1196 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0); 1197 __be32 daddr; 1198 struct ip_options_rcu *inet_opt; 1199 struct flowi4 *fl4; 1200 int err; 1201 1202 /* Route is OK, nothing to do. */ 1203 if (rt) 1204 return 0; 1205 1206 /* Reroute. */ 1207 rcu_read_lock(); 1208 inet_opt = rcu_dereference(inet->inet_opt); 1209 daddr = inet->inet_daddr; 1210 if (inet_opt && inet_opt->opt.srr) 1211 daddr = inet_opt->opt.faddr; 1212 rcu_read_unlock(); 1213 fl4 = &inet->cork.fl.u.ip4; 1214 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr, 1215 inet->inet_dport, inet->inet_sport, 1216 sk->sk_protocol, RT_CONN_FLAGS(sk), 1217 sk->sk_bound_dev_if); 1218 if (!IS_ERR(rt)) { 1219 err = 0; 1220 sk_setup_caps(sk, &rt->dst); 1221 } else { 1222 err = PTR_ERR(rt); 1223 1224 /* Routing failed... */ 1225 sk->sk_route_caps = 0; 1226 /* 1227 * Other protocols have to map its equivalent state to TCP_SYN_SENT. 1228 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme 1229 */ 1230 if (!sock_net(sk)->ipv4.sysctl_ip_dynaddr || 1231 sk->sk_state != TCP_SYN_SENT || 1232 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) || 1233 (err = inet_sk_reselect_saddr(sk)) != 0) 1234 sk->sk_err_soft = -err; 1235 } 1236 1237 return err; 1238 } 1239 EXPORT_SYMBOL(inet_sk_rebuild_header); 1240 1241 void inet_sk_set_state(struct sock *sk, int state) 1242 { 1243 trace_inet_sock_set_state(sk, sk->sk_state, state); 1244 sk->sk_state = state; 1245 } 1246 EXPORT_SYMBOL(inet_sk_set_state); 1247 1248 void inet_sk_state_store(struct sock *sk, int newstate) 1249 { 1250 trace_inet_sock_set_state(sk, sk->sk_state, newstate); 1251 smp_store_release(&sk->sk_state, newstate); 1252 } 1253 1254 struct sk_buff *inet_gso_segment(struct sk_buff *skb, 1255 netdev_features_t features) 1256 { 1257 bool udpfrag = false, fixedid = false, gso_partial, encap; 1258 struct sk_buff *segs = ERR_PTR(-EINVAL); 1259 const struct net_offload *ops; 1260 unsigned int offset = 0; 1261 struct iphdr *iph; 1262 int proto, tot_len; 1263 int nhoff; 1264 int ihl; 1265 int id; 1266 1267 skb_reset_network_header(skb); 1268 nhoff = skb_network_header(skb) - skb_mac_header(skb); 1269 if (unlikely(!pskb_may_pull(skb, sizeof(*iph)))) 1270 goto out; 1271 1272 iph = ip_hdr(skb); 1273 ihl = iph->ihl * 4; 1274 if (ihl < sizeof(*iph)) 1275 goto out; 1276 1277 id = ntohs(iph->id); 1278 proto = iph->protocol; 1279 1280 /* Warning: after this point, iph might be no longer valid */ 1281 if (unlikely(!pskb_may_pull(skb, ihl))) 1282 goto out; 1283 __skb_pull(skb, ihl); 1284 1285 encap = SKB_GSO_CB(skb)->encap_level > 0; 1286 if (encap) 1287 features &= skb->dev->hw_enc_features; 1288 SKB_GSO_CB(skb)->encap_level += ihl; 1289 1290 skb_reset_transport_header(skb); 1291 1292 segs = ERR_PTR(-EPROTONOSUPPORT); 1293 1294 if (!skb->encapsulation || encap) { 1295 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP); 1296 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID); 1297 1298 /* fixed ID is invalid if DF bit is not set */ 1299 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF))) 1300 goto out; 1301 } 1302 1303 ops = rcu_dereference(inet_offloads[proto]); 1304 if (likely(ops && ops->callbacks.gso_segment)) 1305 segs = ops->callbacks.gso_segment(skb, features); 1306 1307 if (IS_ERR_OR_NULL(segs)) 1308 goto out; 1309 1310 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL); 1311 1312 skb = segs; 1313 do { 1314 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff); 1315 if (udpfrag) { 1316 iph->frag_off = htons(offset >> 3); 1317 if (skb->next) 1318 iph->frag_off |= htons(IP_MF); 1319 offset += skb->len - nhoff - ihl; 1320 tot_len = skb->len - nhoff; 1321 } else if (skb_is_gso(skb)) { 1322 if (!fixedid) { 1323 iph->id = htons(id); 1324 id += skb_shinfo(skb)->gso_segs; 1325 } 1326 1327 if (gso_partial) 1328 tot_len = skb_shinfo(skb)->gso_size + 1329 SKB_GSO_CB(skb)->data_offset + 1330 skb->head - (unsigned char *)iph; 1331 else 1332 tot_len = skb->len - nhoff; 1333 } else { 1334 if (!fixedid) 1335 iph->id = htons(id++); 1336 tot_len = skb->len - nhoff; 1337 } 1338 iph->tot_len = htons(tot_len); 1339 ip_send_check(iph); 1340 if (encap) 1341 skb_reset_inner_headers(skb); 1342 skb->network_header = (u8 *)iph - skb->head; 1343 } while ((skb = skb->next)); 1344 1345 out: 1346 return segs; 1347 } 1348 EXPORT_SYMBOL(inet_gso_segment); 1349 1350 struct sk_buff **inet_gro_receive(struct sk_buff **head, struct sk_buff *skb) 1351 { 1352 const struct net_offload *ops; 1353 struct sk_buff **pp = NULL; 1354 struct sk_buff *p; 1355 const struct iphdr *iph; 1356 unsigned int hlen; 1357 unsigned int off; 1358 unsigned int id; 1359 int flush = 1; 1360 int proto; 1361 1362 off = skb_gro_offset(skb); 1363 hlen = off + sizeof(*iph); 1364 iph = skb_gro_header_fast(skb, off); 1365 if (skb_gro_header_hard(skb, hlen)) { 1366 iph = skb_gro_header_slow(skb, hlen, off); 1367 if (unlikely(!iph)) 1368 goto out; 1369 } 1370 1371 proto = iph->protocol; 1372 1373 rcu_read_lock(); 1374 ops = rcu_dereference(inet_offloads[proto]); 1375 if (!ops || !ops->callbacks.gro_receive) 1376 goto out_unlock; 1377 1378 if (*(u8 *)iph != 0x45) 1379 goto out_unlock; 1380 1381 if (ip_is_fragment(iph)) 1382 goto out_unlock; 1383 1384 if (unlikely(ip_fast_csum((u8 *)iph, 5))) 1385 goto out_unlock; 1386 1387 id = ntohl(*(__be32 *)&iph->id); 1388 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF)); 1389 id >>= 16; 1390 1391 for (p = *head; p; p = p->next) { 1392 struct iphdr *iph2; 1393 u16 flush_id; 1394 1395 if (!NAPI_GRO_CB(p)->same_flow) 1396 continue; 1397 1398 iph2 = (struct iphdr *)(p->data + off); 1399 /* The above works because, with the exception of the top 1400 * (inner most) layer, we only aggregate pkts with the same 1401 * hdr length so all the hdrs we'll need to verify will start 1402 * at the same offset. 1403 */ 1404 if ((iph->protocol ^ iph2->protocol) | 1405 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) | 1406 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) { 1407 NAPI_GRO_CB(p)->same_flow = 0; 1408 continue; 1409 } 1410 1411 /* All fields must match except length and checksum. */ 1412 NAPI_GRO_CB(p)->flush |= 1413 (iph->ttl ^ iph2->ttl) | 1414 (iph->tos ^ iph2->tos) | 1415 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF)); 1416 1417 NAPI_GRO_CB(p)->flush |= flush; 1418 1419 /* We need to store of the IP ID check to be included later 1420 * when we can verify that this packet does in fact belong 1421 * to a given flow. 1422 */ 1423 flush_id = (u16)(id - ntohs(iph2->id)); 1424 1425 /* This bit of code makes it much easier for us to identify 1426 * the cases where we are doing atomic vs non-atomic IP ID 1427 * checks. Specifically an atomic check can return IP ID 1428 * values 0 - 0xFFFF, while a non-atomic check can only 1429 * return 0 or 0xFFFF. 1430 */ 1431 if (!NAPI_GRO_CB(p)->is_atomic || 1432 !(iph->frag_off & htons(IP_DF))) { 1433 flush_id ^= NAPI_GRO_CB(p)->count; 1434 flush_id = flush_id ? 0xFFFF : 0; 1435 } 1436 1437 /* If the previous IP ID value was based on an atomic 1438 * datagram we can overwrite the value and ignore it. 1439 */ 1440 if (NAPI_GRO_CB(skb)->is_atomic) 1441 NAPI_GRO_CB(p)->flush_id = flush_id; 1442 else 1443 NAPI_GRO_CB(p)->flush_id |= flush_id; 1444 } 1445 1446 NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF)); 1447 NAPI_GRO_CB(skb)->flush |= flush; 1448 skb_set_network_header(skb, off); 1449 /* The above will be needed by the transport layer if there is one 1450 * immediately following this IP hdr. 1451 */ 1452 1453 /* Note : No need to call skb_gro_postpull_rcsum() here, 1454 * as we already checked checksum over ipv4 header was 0 1455 */ 1456 skb_gro_pull(skb, sizeof(*iph)); 1457 skb_set_transport_header(skb, skb_gro_offset(skb)); 1458 1459 pp = call_gro_receive(ops->callbacks.gro_receive, head, skb); 1460 1461 out_unlock: 1462 rcu_read_unlock(); 1463 1464 out: 1465 skb_gro_flush_final(skb, pp, flush); 1466 1467 return pp; 1468 } 1469 EXPORT_SYMBOL(inet_gro_receive); 1470 1471 static struct sk_buff **ipip_gro_receive(struct sk_buff **head, 1472 struct sk_buff *skb) 1473 { 1474 if (NAPI_GRO_CB(skb)->encap_mark) { 1475 NAPI_GRO_CB(skb)->flush = 1; 1476 return NULL; 1477 } 1478 1479 NAPI_GRO_CB(skb)->encap_mark = 1; 1480 1481 return inet_gro_receive(head, skb); 1482 } 1483 1484 #define SECONDS_PER_DAY 86400 1485 1486 /* inet_current_timestamp - Return IP network timestamp 1487 * 1488 * Return milliseconds since midnight in network byte order. 1489 */ 1490 __be32 inet_current_timestamp(void) 1491 { 1492 u32 secs; 1493 u32 msecs; 1494 struct timespec64 ts; 1495 1496 ktime_get_real_ts64(&ts); 1497 1498 /* Get secs since midnight. */ 1499 (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs); 1500 /* Convert to msecs. */ 1501 msecs = secs * MSEC_PER_SEC; 1502 /* Convert nsec to msec. */ 1503 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC; 1504 1505 /* Convert to network byte order. */ 1506 return htonl(msecs); 1507 } 1508 EXPORT_SYMBOL(inet_current_timestamp); 1509 1510 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len) 1511 { 1512 if (sk->sk_family == AF_INET) 1513 return ip_recv_error(sk, msg, len, addr_len); 1514 #if IS_ENABLED(CONFIG_IPV6) 1515 if (sk->sk_family == AF_INET6) 1516 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len); 1517 #endif 1518 return -EINVAL; 1519 } 1520 1521 int inet_gro_complete(struct sk_buff *skb, int nhoff) 1522 { 1523 __be16 newlen = htons(skb->len - nhoff); 1524 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff); 1525 const struct net_offload *ops; 1526 int proto = iph->protocol; 1527 int err = -ENOSYS; 1528 1529 if (skb->encapsulation) { 1530 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP)); 1531 skb_set_inner_network_header(skb, nhoff); 1532 } 1533 1534 csum_replace2(&iph->check, iph->tot_len, newlen); 1535 iph->tot_len = newlen; 1536 1537 rcu_read_lock(); 1538 ops = rcu_dereference(inet_offloads[proto]); 1539 if (WARN_ON(!ops || !ops->callbacks.gro_complete)) 1540 goto out_unlock; 1541 1542 /* Only need to add sizeof(*iph) to get to the next hdr below 1543 * because any hdr with option will have been flushed in 1544 * inet_gro_receive(). 1545 */ 1546 err = ops->callbacks.gro_complete(skb, nhoff + sizeof(*iph)); 1547 1548 out_unlock: 1549 rcu_read_unlock(); 1550 1551 return err; 1552 } 1553 EXPORT_SYMBOL(inet_gro_complete); 1554 1555 static int ipip_gro_complete(struct sk_buff *skb, int nhoff) 1556 { 1557 skb->encapsulation = 1; 1558 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4; 1559 return inet_gro_complete(skb, nhoff); 1560 } 1561 1562 int inet_ctl_sock_create(struct sock **sk, unsigned short family, 1563 unsigned short type, unsigned char protocol, 1564 struct net *net) 1565 { 1566 struct socket *sock; 1567 int rc = sock_create_kern(net, family, type, protocol, &sock); 1568 1569 if (rc == 0) { 1570 *sk = sock->sk; 1571 (*sk)->sk_allocation = GFP_ATOMIC; 1572 /* 1573 * Unhash it so that IP input processing does not even see it, 1574 * we do not wish this socket to see incoming packets. 1575 */ 1576 (*sk)->sk_prot->unhash(*sk); 1577 } 1578 return rc; 1579 } 1580 EXPORT_SYMBOL_GPL(inet_ctl_sock_create); 1581 1582 u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt) 1583 { 1584 return *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt); 1585 } 1586 EXPORT_SYMBOL_GPL(snmp_get_cpu_field); 1587 1588 unsigned long snmp_fold_field(void __percpu *mib, int offt) 1589 { 1590 unsigned long res = 0; 1591 int i; 1592 1593 for_each_possible_cpu(i) 1594 res += snmp_get_cpu_field(mib, i, offt); 1595 return res; 1596 } 1597 EXPORT_SYMBOL_GPL(snmp_fold_field); 1598 1599 #if BITS_PER_LONG==32 1600 1601 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt, 1602 size_t syncp_offset) 1603 { 1604 void *bhptr; 1605 struct u64_stats_sync *syncp; 1606 u64 v; 1607 unsigned int start; 1608 1609 bhptr = per_cpu_ptr(mib, cpu); 1610 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset); 1611 do { 1612 start = u64_stats_fetch_begin_irq(syncp); 1613 v = *(((u64 *)bhptr) + offt); 1614 } while (u64_stats_fetch_retry_irq(syncp, start)); 1615 1616 return v; 1617 } 1618 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64); 1619 1620 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset) 1621 { 1622 u64 res = 0; 1623 int cpu; 1624 1625 for_each_possible_cpu(cpu) { 1626 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset); 1627 } 1628 return res; 1629 } 1630 EXPORT_SYMBOL_GPL(snmp_fold_field64); 1631 #endif 1632 1633 #ifdef CONFIG_IP_MULTICAST 1634 static const struct net_protocol igmp_protocol = { 1635 .handler = igmp_rcv, 1636 .netns_ok = 1, 1637 }; 1638 #endif 1639 1640 /* thinking of making this const? Don't. 1641 * early_demux can change based on sysctl. 1642 */ 1643 static struct net_protocol tcp_protocol = { 1644 .early_demux = tcp_v4_early_demux, 1645 .early_demux_handler = tcp_v4_early_demux, 1646 .handler = tcp_v4_rcv, 1647 .err_handler = tcp_v4_err, 1648 .no_policy = 1, 1649 .netns_ok = 1, 1650 .icmp_strict_tag_validation = 1, 1651 }; 1652 1653 /* thinking of making this const? Don't. 1654 * early_demux can change based on sysctl. 1655 */ 1656 static struct net_protocol udp_protocol = { 1657 .early_demux = udp_v4_early_demux, 1658 .early_demux_handler = udp_v4_early_demux, 1659 .handler = udp_rcv, 1660 .err_handler = udp_err, 1661 .no_policy = 1, 1662 .netns_ok = 1, 1663 }; 1664 1665 static const struct net_protocol icmp_protocol = { 1666 .handler = icmp_rcv, 1667 .err_handler = icmp_err, 1668 .no_policy = 1, 1669 .netns_ok = 1, 1670 }; 1671 1672 static __net_init int ipv4_mib_init_net(struct net *net) 1673 { 1674 int i; 1675 1676 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib); 1677 if (!net->mib.tcp_statistics) 1678 goto err_tcp_mib; 1679 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib); 1680 if (!net->mib.ip_statistics) 1681 goto err_ip_mib; 1682 1683 for_each_possible_cpu(i) { 1684 struct ipstats_mib *af_inet_stats; 1685 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i); 1686 u64_stats_init(&af_inet_stats->syncp); 1687 } 1688 1689 net->mib.net_statistics = alloc_percpu(struct linux_mib); 1690 if (!net->mib.net_statistics) 1691 goto err_net_mib; 1692 net->mib.udp_statistics = alloc_percpu(struct udp_mib); 1693 if (!net->mib.udp_statistics) 1694 goto err_udp_mib; 1695 net->mib.udplite_statistics = alloc_percpu(struct udp_mib); 1696 if (!net->mib.udplite_statistics) 1697 goto err_udplite_mib; 1698 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib); 1699 if (!net->mib.icmp_statistics) 1700 goto err_icmp_mib; 1701 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib), 1702 GFP_KERNEL); 1703 if (!net->mib.icmpmsg_statistics) 1704 goto err_icmpmsg_mib; 1705 1706 tcp_mib_init(net); 1707 return 0; 1708 1709 err_icmpmsg_mib: 1710 free_percpu(net->mib.icmp_statistics); 1711 err_icmp_mib: 1712 free_percpu(net->mib.udplite_statistics); 1713 err_udplite_mib: 1714 free_percpu(net->mib.udp_statistics); 1715 err_udp_mib: 1716 free_percpu(net->mib.net_statistics); 1717 err_net_mib: 1718 free_percpu(net->mib.ip_statistics); 1719 err_ip_mib: 1720 free_percpu(net->mib.tcp_statistics); 1721 err_tcp_mib: 1722 return -ENOMEM; 1723 } 1724 1725 static __net_exit void ipv4_mib_exit_net(struct net *net) 1726 { 1727 kfree(net->mib.icmpmsg_statistics); 1728 free_percpu(net->mib.icmp_statistics); 1729 free_percpu(net->mib.udplite_statistics); 1730 free_percpu(net->mib.udp_statistics); 1731 free_percpu(net->mib.net_statistics); 1732 free_percpu(net->mib.ip_statistics); 1733 free_percpu(net->mib.tcp_statistics); 1734 } 1735 1736 static __net_initdata struct pernet_operations ipv4_mib_ops = { 1737 .init = ipv4_mib_init_net, 1738 .exit = ipv4_mib_exit_net, 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 }; 1793 1794 static int __init init_inet_pernet_ops(void) 1795 { 1796 return register_pernet_subsys(&af_inet_ops); 1797 } 1798 1799 static int ipv4_proc_init(void); 1800 1801 /* 1802 * IP protocol layer initialiser 1803 */ 1804 1805 static struct packet_offload ip_packet_offload __read_mostly = { 1806 .type = cpu_to_be16(ETH_P_IP), 1807 .callbacks = { 1808 .gso_segment = inet_gso_segment, 1809 .gro_receive = inet_gro_receive, 1810 .gro_complete = inet_gro_complete, 1811 }, 1812 }; 1813 1814 static const struct net_offload ipip_offload = { 1815 .callbacks = { 1816 .gso_segment = inet_gso_segment, 1817 .gro_receive = ipip_gro_receive, 1818 .gro_complete = ipip_gro_complete, 1819 }, 1820 }; 1821 1822 static int __init ipip_offload_init(void) 1823 { 1824 return inet_add_offload(&ipip_offload, IPPROTO_IPIP); 1825 } 1826 1827 static int __init ipv4_offload_init(void) 1828 { 1829 /* 1830 * Add offloads 1831 */ 1832 if (udpv4_offload_init() < 0) 1833 pr_crit("%s: Cannot add UDP protocol offload\n", __func__); 1834 if (tcpv4_offload_init() < 0) 1835 pr_crit("%s: Cannot add TCP protocol offload\n", __func__); 1836 if (ipip_offload_init() < 0) 1837 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__); 1838 1839 dev_add_offload(&ip_packet_offload); 1840 return 0; 1841 } 1842 1843 fs_initcall(ipv4_offload_init); 1844 1845 static struct packet_type ip_packet_type __read_mostly = { 1846 .type = cpu_to_be16(ETH_P_IP), 1847 .func = ip_rcv, 1848 }; 1849 1850 static int __init inet_init(void) 1851 { 1852 struct inet_protosw *q; 1853 struct list_head *r; 1854 int rc = -EINVAL; 1855 1856 sock_skb_cb_check_size(sizeof(struct inet_skb_parm)); 1857 1858 rc = proto_register(&tcp_prot, 1); 1859 if (rc) 1860 goto out; 1861 1862 rc = proto_register(&udp_prot, 1); 1863 if (rc) 1864 goto out_unregister_tcp_proto; 1865 1866 rc = proto_register(&raw_prot, 1); 1867 if (rc) 1868 goto out_unregister_udp_proto; 1869 1870 rc = proto_register(&ping_prot, 1); 1871 if (rc) 1872 goto out_unregister_raw_proto; 1873 1874 /* 1875 * Tell SOCKET that we are alive... 1876 */ 1877 1878 (void)sock_register(&inet_family_ops); 1879 1880 #ifdef CONFIG_SYSCTL 1881 ip_static_sysctl_init(); 1882 #endif 1883 1884 /* 1885 * Add all the base protocols. 1886 */ 1887 1888 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0) 1889 pr_crit("%s: Cannot add ICMP protocol\n", __func__); 1890 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0) 1891 pr_crit("%s: Cannot add UDP protocol\n", __func__); 1892 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0) 1893 pr_crit("%s: Cannot add TCP protocol\n", __func__); 1894 #ifdef CONFIG_IP_MULTICAST 1895 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0) 1896 pr_crit("%s: Cannot add IGMP protocol\n", __func__); 1897 #endif 1898 1899 /* Register the socket-side information for inet_create. */ 1900 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r) 1901 INIT_LIST_HEAD(r); 1902 1903 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q) 1904 inet_register_protosw(q); 1905 1906 /* 1907 * Set the ARP module up 1908 */ 1909 1910 arp_init(); 1911 1912 /* 1913 * Set the IP module up 1914 */ 1915 1916 ip_init(); 1917 1918 /* Setup TCP slab cache for open requests. */ 1919 tcp_init(); 1920 1921 /* Setup UDP memory threshold */ 1922 udp_init(); 1923 1924 /* Add UDP-Lite (RFC 3828) */ 1925 udplite4_register(); 1926 1927 ping_init(); 1928 1929 /* 1930 * Set the ICMP layer up 1931 */ 1932 1933 if (icmp_init() < 0) 1934 panic("Failed to create the ICMP control socket.\n"); 1935 1936 /* 1937 * Initialise the multicast router 1938 */ 1939 #if defined(CONFIG_IP_MROUTE) 1940 if (ip_mr_init()) 1941 pr_crit("%s: Cannot init ipv4 mroute\n", __func__); 1942 #endif 1943 1944 if (init_inet_pernet_ops()) 1945 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__); 1946 /* 1947 * Initialise per-cpu ipv4 mibs 1948 */ 1949 1950 if (init_ipv4_mibs()) 1951 pr_crit("%s: Cannot init ipv4 mibs\n", __func__); 1952 1953 ipv4_proc_init(); 1954 1955 ipfrag_init(); 1956 1957 dev_add_pack(&ip_packet_type); 1958 1959 ip_tunnel_core_init(); 1960 1961 rc = 0; 1962 out: 1963 return rc; 1964 out_unregister_raw_proto: 1965 proto_unregister(&raw_prot); 1966 out_unregister_udp_proto: 1967 proto_unregister(&udp_prot); 1968 out_unregister_tcp_proto: 1969 proto_unregister(&tcp_prot); 1970 goto out; 1971 } 1972 1973 fs_initcall(inet_init); 1974 1975 /* ------------------------------------------------------------------------ */ 1976 1977 #ifdef CONFIG_PROC_FS 1978 static int __init ipv4_proc_init(void) 1979 { 1980 int rc = 0; 1981 1982 if (raw_proc_init()) 1983 goto out_raw; 1984 if (tcp4_proc_init()) 1985 goto out_tcp; 1986 if (udp4_proc_init()) 1987 goto out_udp; 1988 if (ping_proc_init()) 1989 goto out_ping; 1990 if (ip_misc_proc_init()) 1991 goto out_misc; 1992 out: 1993 return rc; 1994 out_misc: 1995 ping_proc_exit(); 1996 out_ping: 1997 udp4_proc_exit(); 1998 out_udp: 1999 tcp4_proc_exit(); 2000 out_tcp: 2001 raw_proc_exit(); 2002 out_raw: 2003 rc = -ENOMEM; 2004 goto out; 2005 } 2006 2007 #else /* CONFIG_PROC_FS */ 2008 static int __init ipv4_proc_init(void) 2009 { 2010 return 0; 2011 } 2012 #endif /* CONFIG_PROC_FS */ 2013