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> 123 124 125 /* The inetsw table contains everything that inet_create needs to 126 * build a new socket. 127 */ 128 static struct list_head inetsw[SOCK_MAX]; 129 static DEFINE_SPINLOCK(inetsw_lock); 130 131 /* New destruction routine */ 132 133 void inet_sock_destruct(struct sock *sk) 134 { 135 struct inet_sock *inet = inet_sk(sk); 136 137 __skb_queue_purge(&sk->sk_receive_queue); 138 __skb_queue_purge(&sk->sk_error_queue); 139 140 sk_mem_reclaim(sk); 141 142 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) { 143 pr_err("Attempt to release TCP socket in state %d %p\n", 144 sk->sk_state, sk); 145 return; 146 } 147 if (!sock_flag(sk, SOCK_DEAD)) { 148 pr_err("Attempt to release alive inet socket %p\n", sk); 149 return; 150 } 151 152 WARN_ON(atomic_read(&sk->sk_rmem_alloc)); 153 WARN_ON(refcount_read(&sk->sk_wmem_alloc)); 154 WARN_ON(sk->sk_wmem_queued); 155 WARN_ON(sk->sk_forward_alloc); 156 157 kfree(rcu_dereference_protected(inet->inet_opt, 1)); 158 dst_release(rcu_dereference_check(sk->sk_dst_cache, 1)); 159 dst_release(sk->sk_rx_dst); 160 sk_refcnt_debug_dec(sk); 161 } 162 EXPORT_SYMBOL(inet_sock_destruct); 163 164 /* 165 * The routines beyond this point handle the behaviour of an AF_INET 166 * socket object. Mostly it punts to the subprotocols of IP to do 167 * the work. 168 */ 169 170 /* 171 * Automatically bind an unbound socket. 172 */ 173 174 static int inet_autobind(struct sock *sk) 175 { 176 struct inet_sock *inet; 177 /* We may need to bind the socket. */ 178 lock_sock(sk); 179 inet = inet_sk(sk); 180 if (!inet->inet_num) { 181 if (sk->sk_prot->get_port(sk, 0)) { 182 release_sock(sk); 183 return -EAGAIN; 184 } 185 inet->inet_sport = htons(inet->inet_num); 186 } 187 release_sock(sk); 188 return 0; 189 } 190 191 /* 192 * Move a socket into listening state. 193 */ 194 int inet_listen(struct socket *sock, int backlog) 195 { 196 struct sock *sk = sock->sk; 197 unsigned char old_state; 198 int err, tcp_fastopen; 199 200 lock_sock(sk); 201 202 err = -EINVAL; 203 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM) 204 goto out; 205 206 old_state = sk->sk_state; 207 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN))) 208 goto out; 209 210 /* Really, if the socket is already in listen state 211 * we can only allow the backlog to be adjusted. 212 */ 213 if (old_state != TCP_LISTEN) { 214 /* Enable TFO w/o requiring TCP_FASTOPEN socket option. 215 * Note that only TCP sockets (SOCK_STREAM) will reach here. 216 * Also fastopen backlog may already been set via the option 217 * because the socket was in TCP_LISTEN state previously but 218 * was shutdown() rather than close(). 219 */ 220 tcp_fastopen = sock_net(sk)->ipv4.sysctl_tcp_fastopen; 221 if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) && 222 (tcp_fastopen & TFO_SERVER_ENABLE) && 223 !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) { 224 fastopen_queue_tune(sk, backlog); 225 tcp_fastopen_init_key_once(sock_net(sk)); 226 } 227 228 err = inet_csk_listen_start(sk, backlog); 229 if (err) 230 goto out; 231 } 232 sk->sk_max_ack_backlog = backlog; 233 err = 0; 234 235 out: 236 release_sock(sk); 237 return err; 238 } 239 EXPORT_SYMBOL(inet_listen); 240 241 /* 242 * Create an inet socket. 243 */ 244 245 static int inet_create(struct net *net, struct socket *sock, int protocol, 246 int kern) 247 { 248 struct sock *sk; 249 struct inet_protosw *answer; 250 struct inet_sock *inet; 251 struct proto *answer_prot; 252 unsigned char answer_flags; 253 int try_loading_module = 0; 254 int err; 255 256 if (protocol < 0 || protocol >= IPPROTO_MAX) 257 return -EINVAL; 258 259 sock->state = SS_UNCONNECTED; 260 261 /* Look for the requested type/protocol pair. */ 262 lookup_protocol: 263 err = -ESOCKTNOSUPPORT; 264 rcu_read_lock(); 265 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) { 266 267 err = 0; 268 /* Check the non-wild match. */ 269 if (protocol == answer->protocol) { 270 if (protocol != IPPROTO_IP) 271 break; 272 } else { 273 /* Check for the two wild cases. */ 274 if (IPPROTO_IP == protocol) { 275 protocol = answer->protocol; 276 break; 277 } 278 if (IPPROTO_IP == answer->protocol) 279 break; 280 } 281 err = -EPROTONOSUPPORT; 282 } 283 284 if (unlikely(err)) { 285 if (try_loading_module < 2) { 286 rcu_read_unlock(); 287 /* 288 * Be more specific, e.g. net-pf-2-proto-132-type-1 289 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM) 290 */ 291 if (++try_loading_module == 1) 292 request_module("net-pf-%d-proto-%d-type-%d", 293 PF_INET, protocol, sock->type); 294 /* 295 * Fall back to generic, e.g. net-pf-2-proto-132 296 * (net-pf-PF_INET-proto-IPPROTO_SCTP) 297 */ 298 else 299 request_module("net-pf-%d-proto-%d", 300 PF_INET, protocol); 301 goto lookup_protocol; 302 } else 303 goto out_rcu_unlock; 304 } 305 306 err = -EPERM; 307 if (sock->type == SOCK_RAW && !kern && 308 !ns_capable(net->user_ns, CAP_NET_RAW)) 309 goto out_rcu_unlock; 310 311 sock->ops = answer->ops; 312 answer_prot = answer->prot; 313 answer_flags = answer->flags; 314 rcu_read_unlock(); 315 316 WARN_ON(!answer_prot->slab); 317 318 err = -ENOBUFS; 319 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern); 320 if (!sk) 321 goto out; 322 323 err = 0; 324 if (INET_PROTOSW_REUSE & answer_flags) 325 sk->sk_reuse = SK_CAN_REUSE; 326 327 inet = inet_sk(sk); 328 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0; 329 330 inet->nodefrag = 0; 331 332 if (SOCK_RAW == sock->type) { 333 inet->inet_num = protocol; 334 if (IPPROTO_RAW == protocol) 335 inet->hdrincl = 1; 336 } 337 338 if (net->ipv4.sysctl_ip_no_pmtu_disc) 339 inet->pmtudisc = IP_PMTUDISC_DONT; 340 else 341 inet->pmtudisc = IP_PMTUDISC_WANT; 342 343 inet->inet_id = 0; 344 345 sock_init_data(sock, sk); 346 347 sk->sk_destruct = inet_sock_destruct; 348 sk->sk_protocol = protocol; 349 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv; 350 351 inet->uc_ttl = -1; 352 inet->mc_loop = 1; 353 inet->mc_ttl = 1; 354 inet->mc_all = 1; 355 inet->mc_index = 0; 356 inet->mc_list = NULL; 357 inet->rcv_tos = 0; 358 359 sk_refcnt_debug_inc(sk); 360 361 if (inet->inet_num) { 362 /* It assumes that any protocol which allows 363 * the user to assign a number at socket 364 * creation time automatically 365 * shares. 366 */ 367 inet->inet_sport = htons(inet->inet_num); 368 /* Add to protocol hash chains. */ 369 err = sk->sk_prot->hash(sk); 370 if (err) { 371 sk_common_release(sk); 372 goto out; 373 } 374 } 375 376 if (sk->sk_prot->init) { 377 err = sk->sk_prot->init(sk); 378 if (err) { 379 sk_common_release(sk); 380 goto out; 381 } 382 } 383 384 if (!kern) { 385 err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk); 386 if (err) { 387 sk_common_release(sk); 388 goto out; 389 } 390 } 391 out: 392 return err; 393 out_rcu_unlock: 394 rcu_read_unlock(); 395 goto out; 396 } 397 398 399 /* 400 * The peer socket should always be NULL (or else). When we call this 401 * function we are destroying the object and from then on nobody 402 * should refer to it. 403 */ 404 int inet_release(struct socket *sock) 405 { 406 struct sock *sk = sock->sk; 407 408 if (sk) { 409 long timeout; 410 411 /* Applications forget to leave groups before exiting */ 412 ip_mc_drop_socket(sk); 413 414 /* If linger is set, we don't return until the close 415 * is complete. Otherwise we return immediately. The 416 * actually closing is done the same either way. 417 * 418 * If the close is due to the process exiting, we never 419 * linger.. 420 */ 421 timeout = 0; 422 if (sock_flag(sk, SOCK_LINGER) && 423 !(current->flags & PF_EXITING)) 424 timeout = sk->sk_lingertime; 425 sock->sk = NULL; 426 sk->sk_prot->close(sk, timeout); 427 } 428 return 0; 429 } 430 EXPORT_SYMBOL(inet_release); 431 432 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) 433 { 434 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr; 435 struct sock *sk = sock->sk; 436 struct inet_sock *inet = inet_sk(sk); 437 struct net *net = sock_net(sk); 438 unsigned short snum; 439 int chk_addr_ret; 440 u32 tb_id = RT_TABLE_LOCAL; 441 int err; 442 443 /* If the socket has its own bind function then use it. (RAW) */ 444 if (sk->sk_prot->bind) { 445 err = sk->sk_prot->bind(sk, uaddr, addr_len); 446 goto out; 447 } 448 err = -EINVAL; 449 if (addr_len < sizeof(struct sockaddr_in)) 450 goto out; 451 452 if (addr->sin_family != AF_INET) { 453 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET) 454 * only if s_addr is INADDR_ANY. 455 */ 456 err = -EAFNOSUPPORT; 457 if (addr->sin_family != AF_UNSPEC || 458 addr->sin_addr.s_addr != htonl(INADDR_ANY)) 459 goto out; 460 } 461 462 tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id; 463 chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id); 464 465 /* Not specified by any standard per-se, however it breaks too 466 * many applications when removed. It is unfortunate since 467 * allowing applications to make a non-local bind solves 468 * several problems with systems using dynamic addressing. 469 * (ie. your servers still start up even if your ISDN link 470 * is temporarily down) 471 */ 472 err = -EADDRNOTAVAIL; 473 if (!net->ipv4.sysctl_ip_nonlocal_bind && 474 !(inet->freebind || inet->transparent) && 475 addr->sin_addr.s_addr != htonl(INADDR_ANY) && 476 chk_addr_ret != RTN_LOCAL && 477 chk_addr_ret != RTN_MULTICAST && 478 chk_addr_ret != RTN_BROADCAST) 479 goto out; 480 481 snum = ntohs(addr->sin_port); 482 err = -EACCES; 483 if (snum && snum < inet_prot_sock(net) && 484 !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE)) 485 goto out; 486 487 /* We keep a pair of addresses. rcv_saddr is the one 488 * used by hash lookups, and saddr is used for transmit. 489 * 490 * In the BSD API these are the same except where it 491 * would be illegal to use them (multicast/broadcast) in 492 * which case the sending device address is used. 493 */ 494 lock_sock(sk); 495 496 /* Check these errors (active socket, double bind). */ 497 err = -EINVAL; 498 if (sk->sk_state != TCP_CLOSE || inet->inet_num) 499 goto out_release_sock; 500 501 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr; 502 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST) 503 inet->inet_saddr = 0; /* Use device */ 504 505 /* Make sure we are allowed to bind here. */ 506 if ((snum || !inet->bind_address_no_port) && 507 sk->sk_prot->get_port(sk, snum)) { 508 inet->inet_saddr = inet->inet_rcv_saddr = 0; 509 err = -EADDRINUSE; 510 goto out_release_sock; 511 } 512 513 if (inet->inet_rcv_saddr) 514 sk->sk_userlocks |= SOCK_BINDADDR_LOCK; 515 if (snum) 516 sk->sk_userlocks |= SOCK_BINDPORT_LOCK; 517 inet->inet_sport = htons(inet->inet_num); 518 inet->inet_daddr = 0; 519 inet->inet_dport = 0; 520 sk_dst_reset(sk); 521 err = 0; 522 out_release_sock: 523 release_sock(sk); 524 out: 525 return err; 526 } 527 EXPORT_SYMBOL(inet_bind); 528 529 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr, 530 int addr_len, int flags) 531 { 532 struct sock *sk = sock->sk; 533 534 if (addr_len < sizeof(uaddr->sa_family)) 535 return -EINVAL; 536 if (uaddr->sa_family == AF_UNSPEC) 537 return sk->sk_prot->disconnect(sk, flags); 538 539 if (!inet_sk(sk)->inet_num && inet_autobind(sk)) 540 return -EAGAIN; 541 return sk->sk_prot->connect(sk, uaddr, addr_len); 542 } 543 EXPORT_SYMBOL(inet_dgram_connect); 544 545 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias) 546 { 547 DEFINE_WAIT_FUNC(wait, woken_wake_function); 548 549 add_wait_queue(sk_sleep(sk), &wait); 550 sk->sk_write_pending += writebias; 551 552 /* Basic assumption: if someone sets sk->sk_err, he _must_ 553 * change state of the socket from TCP_SYN_*. 554 * Connect() does not allow to get error notifications 555 * without closing the socket. 556 */ 557 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 558 release_sock(sk); 559 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo); 560 lock_sock(sk); 561 if (signal_pending(current) || !timeo) 562 break; 563 } 564 remove_wait_queue(sk_sleep(sk), &wait); 565 sk->sk_write_pending -= writebias; 566 return timeo; 567 } 568 569 /* 570 * Connect to a remote host. There is regrettably still a little 571 * TCP 'magic' in here. 572 */ 573 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr, 574 int addr_len, int flags, int is_sendmsg) 575 { 576 struct sock *sk = sock->sk; 577 int err; 578 long timeo; 579 580 /* 581 * uaddr can be NULL and addr_len can be 0 if: 582 * sk is a TCP fastopen active socket and 583 * TCP_FASTOPEN_CONNECT sockopt is set and 584 * we already have a valid cookie for this socket. 585 * In this case, user can call write() after connect(). 586 * write() will invoke tcp_sendmsg_fastopen() which calls 587 * __inet_stream_connect(). 588 */ 589 if (uaddr) { 590 if (addr_len < sizeof(uaddr->sa_family)) 591 return -EINVAL; 592 593 if (uaddr->sa_family == AF_UNSPEC) { 594 err = sk->sk_prot->disconnect(sk, flags); 595 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED; 596 goto out; 597 } 598 } 599 600 switch (sock->state) { 601 default: 602 err = -EINVAL; 603 goto out; 604 case SS_CONNECTED: 605 err = -EISCONN; 606 goto out; 607 case SS_CONNECTING: 608 if (inet_sk(sk)->defer_connect) 609 err = is_sendmsg ? -EINPROGRESS : -EISCONN; 610 else 611 err = -EALREADY; 612 /* Fall out of switch with err, set for this state */ 613 break; 614 case SS_UNCONNECTED: 615 err = -EISCONN; 616 if (sk->sk_state != TCP_CLOSE) 617 goto out; 618 619 err = sk->sk_prot->connect(sk, uaddr, addr_len); 620 if (err < 0) 621 goto out; 622 623 sock->state = SS_CONNECTING; 624 625 if (!err && inet_sk(sk)->defer_connect) 626 goto out; 627 628 /* Just entered SS_CONNECTING state; the only 629 * difference is that return value in non-blocking 630 * case is EINPROGRESS, rather than EALREADY. 631 */ 632 err = -EINPROGRESS; 633 break; 634 } 635 636 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK); 637 638 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 639 int writebias = (sk->sk_protocol == IPPROTO_TCP) && 640 tcp_sk(sk)->fastopen_req && 641 tcp_sk(sk)->fastopen_req->data ? 1 : 0; 642 643 /* Error code is set above */ 644 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias)) 645 goto out; 646 647 err = sock_intr_errno(timeo); 648 if (signal_pending(current)) 649 goto out; 650 } 651 652 /* Connection was closed by RST, timeout, ICMP error 653 * or another process disconnected us. 654 */ 655 if (sk->sk_state == TCP_CLOSE) 656 goto sock_error; 657 658 /* sk->sk_err may be not zero now, if RECVERR was ordered by user 659 * and error was received after socket entered established state. 660 * Hence, it is handled normally after connect() return successfully. 661 */ 662 663 sock->state = SS_CONNECTED; 664 err = 0; 665 out: 666 return err; 667 668 sock_error: 669 err = sock_error(sk) ? : -ECONNABORTED; 670 sock->state = SS_UNCONNECTED; 671 if (sk->sk_prot->disconnect(sk, flags)) 672 sock->state = SS_DISCONNECTING; 673 goto out; 674 } 675 EXPORT_SYMBOL(__inet_stream_connect); 676 677 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr, 678 int addr_len, int flags) 679 { 680 int err; 681 682 lock_sock(sock->sk); 683 err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0); 684 release_sock(sock->sk); 685 return err; 686 } 687 EXPORT_SYMBOL(inet_stream_connect); 688 689 /* 690 * Accept a pending connection. The TCP layer now gives BSD semantics. 691 */ 692 693 int inet_accept(struct socket *sock, struct socket *newsock, int flags, 694 bool kern) 695 { 696 struct sock *sk1 = sock->sk; 697 int err = -EINVAL; 698 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err, kern); 699 700 if (!sk2) 701 goto do_err; 702 703 lock_sock(sk2); 704 705 sock_rps_record_flow(sk2); 706 WARN_ON(!((1 << sk2->sk_state) & 707 (TCPF_ESTABLISHED | TCPF_SYN_RECV | 708 TCPF_CLOSE_WAIT | TCPF_CLOSE))); 709 710 sock_graft(sk2, newsock); 711 712 newsock->state = SS_CONNECTED; 713 err = 0; 714 release_sock(sk2); 715 do_err: 716 return err; 717 } 718 EXPORT_SYMBOL(inet_accept); 719 720 721 /* 722 * This does both peername and sockname. 723 */ 724 int inet_getname(struct socket *sock, struct sockaddr *uaddr, 725 int *uaddr_len, int peer) 726 { 727 struct sock *sk = sock->sk; 728 struct inet_sock *inet = inet_sk(sk); 729 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr); 730 731 sin->sin_family = AF_INET; 732 if (peer) { 733 if (!inet->inet_dport || 734 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) && 735 peer == 1)) 736 return -ENOTCONN; 737 sin->sin_port = inet->inet_dport; 738 sin->sin_addr.s_addr = inet->inet_daddr; 739 } else { 740 __be32 addr = inet->inet_rcv_saddr; 741 if (!addr) 742 addr = inet->inet_saddr; 743 sin->sin_port = inet->inet_sport; 744 sin->sin_addr.s_addr = addr; 745 } 746 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 747 *uaddr_len = sizeof(*sin); 748 return 0; 749 } 750 EXPORT_SYMBOL(inet_getname); 751 752 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size) 753 { 754 struct sock *sk = sock->sk; 755 756 sock_rps_record_flow(sk); 757 758 /* We may need to bind the socket. */ 759 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind && 760 inet_autobind(sk)) 761 return -EAGAIN; 762 763 return sk->sk_prot->sendmsg(sk, msg, size); 764 } 765 EXPORT_SYMBOL(inet_sendmsg); 766 767 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, 768 size_t size, int flags) 769 { 770 struct sock *sk = sock->sk; 771 772 sock_rps_record_flow(sk); 773 774 /* We may need to bind the socket. */ 775 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind && 776 inet_autobind(sk)) 777 return -EAGAIN; 778 779 if (sk->sk_prot->sendpage) 780 return sk->sk_prot->sendpage(sk, page, offset, size, flags); 781 return sock_no_sendpage(sock, page, offset, size, flags); 782 } 783 EXPORT_SYMBOL(inet_sendpage); 784 785 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 786 int flags) 787 { 788 struct sock *sk = sock->sk; 789 int addr_len = 0; 790 int err; 791 792 sock_rps_record_flow(sk); 793 794 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT, 795 flags & ~MSG_DONTWAIT, &addr_len); 796 if (err >= 0) 797 msg->msg_namelen = addr_len; 798 return err; 799 } 800 EXPORT_SYMBOL(inet_recvmsg); 801 802 int inet_shutdown(struct socket *sock, int how) 803 { 804 struct sock *sk = sock->sk; 805 int err = 0; 806 807 /* This should really check to make sure 808 * the socket is a TCP socket. (WHY AC...) 809 */ 810 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and 811 1->2 bit 2 snds. 812 2->3 */ 813 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */ 814 return -EINVAL; 815 816 lock_sock(sk); 817 if (sock->state == SS_CONNECTING) { 818 if ((1 << sk->sk_state) & 819 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE)) 820 sock->state = SS_DISCONNECTING; 821 else 822 sock->state = SS_CONNECTED; 823 } 824 825 switch (sk->sk_state) { 826 case TCP_CLOSE: 827 err = -ENOTCONN; 828 /* Hack to wake up other listeners, who can poll for 829 POLLHUP, even on eg. unconnected UDP sockets -- RR */ 830 default: 831 sk->sk_shutdown |= how; 832 if (sk->sk_prot->shutdown) 833 sk->sk_prot->shutdown(sk, how); 834 break; 835 836 /* Remaining two branches are temporary solution for missing 837 * close() in multithreaded environment. It is _not_ a good idea, 838 * but we have no choice until close() is repaired at VFS level. 839 */ 840 case TCP_LISTEN: 841 if (!(how & RCV_SHUTDOWN)) 842 break; 843 /* Fall through */ 844 case TCP_SYN_SENT: 845 err = sk->sk_prot->disconnect(sk, O_NONBLOCK); 846 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED; 847 break; 848 } 849 850 /* Wake up anyone sleeping in poll. */ 851 sk->sk_state_change(sk); 852 release_sock(sk); 853 return err; 854 } 855 EXPORT_SYMBOL(inet_shutdown); 856 857 /* 858 * ioctl() calls you can issue on an INET socket. Most of these are 859 * device configuration and stuff and very rarely used. Some ioctls 860 * pass on to the socket itself. 861 * 862 * NOTE: I like the idea of a module for the config stuff. ie ifconfig 863 * loads the devconfigure module does its configuring and unloads it. 864 * There's a good 20K of config code hanging around the kernel. 865 */ 866 867 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 868 { 869 struct sock *sk = sock->sk; 870 int err = 0; 871 struct net *net = sock_net(sk); 872 873 switch (cmd) { 874 case SIOCGSTAMP: 875 err = sock_get_timestamp(sk, (struct timeval __user *)arg); 876 break; 877 case SIOCGSTAMPNS: 878 err = sock_get_timestampns(sk, (struct timespec __user *)arg); 879 break; 880 case SIOCADDRT: 881 case SIOCDELRT: 882 case SIOCRTMSG: 883 err = ip_rt_ioctl(net, cmd, (void __user *)arg); 884 break; 885 case SIOCDARP: 886 case SIOCGARP: 887 case SIOCSARP: 888 err = arp_ioctl(net, cmd, (void __user *)arg); 889 break; 890 case SIOCGIFADDR: 891 case SIOCSIFADDR: 892 case SIOCGIFBRDADDR: 893 case SIOCSIFBRDADDR: 894 case SIOCGIFNETMASK: 895 case SIOCSIFNETMASK: 896 case SIOCGIFDSTADDR: 897 case SIOCSIFDSTADDR: 898 case SIOCSIFPFLAGS: 899 case SIOCGIFPFLAGS: 900 case SIOCSIFFLAGS: 901 err = devinet_ioctl(net, cmd, (void __user *)arg); 902 break; 903 default: 904 if (sk->sk_prot->ioctl) 905 err = sk->sk_prot->ioctl(sk, cmd, arg); 906 else 907 err = -ENOIOCTLCMD; 908 break; 909 } 910 return err; 911 } 912 EXPORT_SYMBOL(inet_ioctl); 913 914 #ifdef CONFIG_COMPAT 915 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 916 { 917 struct sock *sk = sock->sk; 918 int err = -ENOIOCTLCMD; 919 920 if (sk->sk_prot->compat_ioctl) 921 err = sk->sk_prot->compat_ioctl(sk, cmd, arg); 922 923 return err; 924 } 925 #endif 926 927 const struct proto_ops inet_stream_ops = { 928 .family = PF_INET, 929 .owner = THIS_MODULE, 930 .release = inet_release, 931 .bind = inet_bind, 932 .connect = inet_stream_connect, 933 .socketpair = sock_no_socketpair, 934 .accept = inet_accept, 935 .getname = inet_getname, 936 .poll = tcp_poll, 937 .ioctl = inet_ioctl, 938 .listen = inet_listen, 939 .shutdown = inet_shutdown, 940 .setsockopt = sock_common_setsockopt, 941 .getsockopt = sock_common_getsockopt, 942 .sendmsg = inet_sendmsg, 943 .recvmsg = inet_recvmsg, 944 .mmap = sock_no_mmap, 945 .sendpage = inet_sendpage, 946 .splice_read = tcp_splice_read, 947 .read_sock = tcp_read_sock, 948 .sendmsg_locked = tcp_sendmsg_locked, 949 .sendpage_locked = tcp_sendpage_locked, 950 .peek_len = tcp_peek_len, 951 #ifdef CONFIG_COMPAT 952 .compat_setsockopt = compat_sock_common_setsockopt, 953 .compat_getsockopt = compat_sock_common_getsockopt, 954 .compat_ioctl = inet_compat_ioctl, 955 #endif 956 }; 957 EXPORT_SYMBOL(inet_stream_ops); 958 959 const struct proto_ops inet_dgram_ops = { 960 .family = PF_INET, 961 .owner = THIS_MODULE, 962 .release = inet_release, 963 .bind = inet_bind, 964 .connect = inet_dgram_connect, 965 .socketpair = sock_no_socketpair, 966 .accept = sock_no_accept, 967 .getname = inet_getname, 968 .poll = udp_poll, 969 .ioctl = inet_ioctl, 970 .listen = sock_no_listen, 971 .shutdown = inet_shutdown, 972 .setsockopt = sock_common_setsockopt, 973 .getsockopt = sock_common_getsockopt, 974 .sendmsg = inet_sendmsg, 975 .recvmsg = inet_recvmsg, 976 .mmap = sock_no_mmap, 977 .sendpage = inet_sendpage, 978 .set_peek_off = sk_set_peek_off, 979 #ifdef CONFIG_COMPAT 980 .compat_setsockopt = compat_sock_common_setsockopt, 981 .compat_getsockopt = compat_sock_common_getsockopt, 982 .compat_ioctl = inet_compat_ioctl, 983 #endif 984 }; 985 EXPORT_SYMBOL(inet_dgram_ops); 986 987 /* 988 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without 989 * udp_poll 990 */ 991 static const struct proto_ops inet_sockraw_ops = { 992 .family = PF_INET, 993 .owner = THIS_MODULE, 994 .release = inet_release, 995 .bind = inet_bind, 996 .connect = inet_dgram_connect, 997 .socketpair = sock_no_socketpair, 998 .accept = sock_no_accept, 999 .getname = inet_getname, 1000 .poll = datagram_poll, 1001 .ioctl = inet_ioctl, 1002 .listen = sock_no_listen, 1003 .shutdown = inet_shutdown, 1004 .setsockopt = sock_common_setsockopt, 1005 .getsockopt = sock_common_getsockopt, 1006 .sendmsg = inet_sendmsg, 1007 .recvmsg = inet_recvmsg, 1008 .mmap = sock_no_mmap, 1009 .sendpage = inet_sendpage, 1010 #ifdef CONFIG_COMPAT 1011 .compat_setsockopt = compat_sock_common_setsockopt, 1012 .compat_getsockopt = compat_sock_common_getsockopt, 1013 .compat_ioctl = inet_compat_ioctl, 1014 #endif 1015 }; 1016 1017 static const struct net_proto_family inet_family_ops = { 1018 .family = PF_INET, 1019 .create = inet_create, 1020 .owner = THIS_MODULE, 1021 }; 1022 1023 /* Upon startup we insert all the elements in inetsw_array[] into 1024 * the linked list inetsw. 1025 */ 1026 static struct inet_protosw inetsw_array[] = 1027 { 1028 { 1029 .type = SOCK_STREAM, 1030 .protocol = IPPROTO_TCP, 1031 .prot = &tcp_prot, 1032 .ops = &inet_stream_ops, 1033 .flags = INET_PROTOSW_PERMANENT | 1034 INET_PROTOSW_ICSK, 1035 }, 1036 1037 { 1038 .type = SOCK_DGRAM, 1039 .protocol = IPPROTO_UDP, 1040 .prot = &udp_prot, 1041 .ops = &inet_dgram_ops, 1042 .flags = INET_PROTOSW_PERMANENT, 1043 }, 1044 1045 { 1046 .type = SOCK_DGRAM, 1047 .protocol = IPPROTO_ICMP, 1048 .prot = &ping_prot, 1049 .ops = &inet_sockraw_ops, 1050 .flags = INET_PROTOSW_REUSE, 1051 }, 1052 1053 { 1054 .type = SOCK_RAW, 1055 .protocol = IPPROTO_IP, /* wild card */ 1056 .prot = &raw_prot, 1057 .ops = &inet_sockraw_ops, 1058 .flags = INET_PROTOSW_REUSE, 1059 } 1060 }; 1061 1062 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array) 1063 1064 void inet_register_protosw(struct inet_protosw *p) 1065 { 1066 struct list_head *lh; 1067 struct inet_protosw *answer; 1068 int protocol = p->protocol; 1069 struct list_head *last_perm; 1070 1071 spin_lock_bh(&inetsw_lock); 1072 1073 if (p->type >= SOCK_MAX) 1074 goto out_illegal; 1075 1076 /* If we are trying to override a permanent protocol, bail. */ 1077 last_perm = &inetsw[p->type]; 1078 list_for_each(lh, &inetsw[p->type]) { 1079 answer = list_entry(lh, struct inet_protosw, list); 1080 /* Check only the non-wild match. */ 1081 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0) 1082 break; 1083 if (protocol == answer->protocol) 1084 goto out_permanent; 1085 last_perm = lh; 1086 } 1087 1088 /* Add the new entry after the last permanent entry if any, so that 1089 * the new entry does not override a permanent entry when matched with 1090 * a wild-card protocol. But it is allowed to override any existing 1091 * non-permanent entry. This means that when we remove this entry, the 1092 * system automatically returns to the old behavior. 1093 */ 1094 list_add_rcu(&p->list, last_perm); 1095 out: 1096 spin_unlock_bh(&inetsw_lock); 1097 1098 return; 1099 1100 out_permanent: 1101 pr_err("Attempt to override permanent protocol %d\n", protocol); 1102 goto out; 1103 1104 out_illegal: 1105 pr_err("Ignoring attempt to register invalid socket type %d\n", 1106 p->type); 1107 goto out; 1108 } 1109 EXPORT_SYMBOL(inet_register_protosw); 1110 1111 void inet_unregister_protosw(struct inet_protosw *p) 1112 { 1113 if (INET_PROTOSW_PERMANENT & p->flags) { 1114 pr_err("Attempt to unregister permanent protocol %d\n", 1115 p->protocol); 1116 } else { 1117 spin_lock_bh(&inetsw_lock); 1118 list_del_rcu(&p->list); 1119 spin_unlock_bh(&inetsw_lock); 1120 1121 synchronize_net(); 1122 } 1123 } 1124 EXPORT_SYMBOL(inet_unregister_protosw); 1125 1126 static int inet_sk_reselect_saddr(struct sock *sk) 1127 { 1128 struct inet_sock *inet = inet_sk(sk); 1129 __be32 old_saddr = inet->inet_saddr; 1130 __be32 daddr = inet->inet_daddr; 1131 struct flowi4 *fl4; 1132 struct rtable *rt; 1133 __be32 new_saddr; 1134 struct ip_options_rcu *inet_opt; 1135 1136 inet_opt = rcu_dereference_protected(inet->inet_opt, 1137 lockdep_sock_is_held(sk)); 1138 if (inet_opt && inet_opt->opt.srr) 1139 daddr = inet_opt->opt.faddr; 1140 1141 /* Query new route. */ 1142 fl4 = &inet->cork.fl.u.ip4; 1143 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk), 1144 sk->sk_bound_dev_if, sk->sk_protocol, 1145 inet->inet_sport, inet->inet_dport, sk); 1146 if (IS_ERR(rt)) 1147 return PTR_ERR(rt); 1148 1149 sk_setup_caps(sk, &rt->dst); 1150 1151 new_saddr = fl4->saddr; 1152 1153 if (new_saddr == old_saddr) 1154 return 0; 1155 1156 if (sock_net(sk)->ipv4.sysctl_ip_dynaddr > 1) { 1157 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n", 1158 __func__, &old_saddr, &new_saddr); 1159 } 1160 1161 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr; 1162 1163 /* 1164 * XXX The only one ugly spot where we need to 1165 * XXX really change the sockets identity after 1166 * XXX it has entered the hashes. -DaveM 1167 * 1168 * Besides that, it does not check for connection 1169 * uniqueness. Wait for troubles. 1170 */ 1171 return __sk_prot_rehash(sk); 1172 } 1173 1174 int inet_sk_rebuild_header(struct sock *sk) 1175 { 1176 struct inet_sock *inet = inet_sk(sk); 1177 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0); 1178 __be32 daddr; 1179 struct ip_options_rcu *inet_opt; 1180 struct flowi4 *fl4; 1181 int err; 1182 1183 /* Route is OK, nothing to do. */ 1184 if (rt) 1185 return 0; 1186 1187 /* Reroute. */ 1188 rcu_read_lock(); 1189 inet_opt = rcu_dereference(inet->inet_opt); 1190 daddr = inet->inet_daddr; 1191 if (inet_opt && inet_opt->opt.srr) 1192 daddr = inet_opt->opt.faddr; 1193 rcu_read_unlock(); 1194 fl4 = &inet->cork.fl.u.ip4; 1195 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr, 1196 inet->inet_dport, inet->inet_sport, 1197 sk->sk_protocol, RT_CONN_FLAGS(sk), 1198 sk->sk_bound_dev_if); 1199 if (!IS_ERR(rt)) { 1200 err = 0; 1201 sk_setup_caps(sk, &rt->dst); 1202 } else { 1203 err = PTR_ERR(rt); 1204 1205 /* Routing failed... */ 1206 sk->sk_route_caps = 0; 1207 /* 1208 * Other protocols have to map its equivalent state to TCP_SYN_SENT. 1209 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme 1210 */ 1211 if (!sock_net(sk)->ipv4.sysctl_ip_dynaddr || 1212 sk->sk_state != TCP_SYN_SENT || 1213 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) || 1214 (err = inet_sk_reselect_saddr(sk)) != 0) 1215 sk->sk_err_soft = -err; 1216 } 1217 1218 return err; 1219 } 1220 EXPORT_SYMBOL(inet_sk_rebuild_header); 1221 1222 struct sk_buff *inet_gso_segment(struct sk_buff *skb, 1223 netdev_features_t features) 1224 { 1225 bool fixedid = false, gso_partial, encap; 1226 struct sk_buff *segs = ERR_PTR(-EINVAL); 1227 const struct net_offload *ops; 1228 struct iphdr *iph; 1229 int proto, tot_len; 1230 int nhoff; 1231 int ihl; 1232 int id; 1233 1234 skb_reset_network_header(skb); 1235 nhoff = skb_network_header(skb) - skb_mac_header(skb); 1236 if (unlikely(!pskb_may_pull(skb, sizeof(*iph)))) 1237 goto out; 1238 1239 iph = ip_hdr(skb); 1240 ihl = iph->ihl * 4; 1241 if (ihl < sizeof(*iph)) 1242 goto out; 1243 1244 id = ntohs(iph->id); 1245 proto = iph->protocol; 1246 1247 /* Warning: after this point, iph might be no longer valid */ 1248 if (unlikely(!pskb_may_pull(skb, ihl))) 1249 goto out; 1250 __skb_pull(skb, ihl); 1251 1252 encap = SKB_GSO_CB(skb)->encap_level > 0; 1253 if (encap) 1254 features &= skb->dev->hw_enc_features; 1255 SKB_GSO_CB(skb)->encap_level += ihl; 1256 1257 skb_reset_transport_header(skb); 1258 1259 segs = ERR_PTR(-EPROTONOSUPPORT); 1260 1261 if (!skb->encapsulation || encap) { 1262 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID); 1263 1264 /* fixed ID is invalid if DF bit is not set */ 1265 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF))) 1266 goto out; 1267 } 1268 1269 ops = rcu_dereference(inet_offloads[proto]); 1270 if (likely(ops && ops->callbacks.gso_segment)) 1271 segs = ops->callbacks.gso_segment(skb, features); 1272 1273 if (IS_ERR_OR_NULL(segs)) 1274 goto out; 1275 1276 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL); 1277 1278 skb = segs; 1279 do { 1280 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff); 1281 if (skb_is_gso(skb)) { 1282 if (!fixedid) { 1283 iph->id = htons(id); 1284 id += skb_shinfo(skb)->gso_segs; 1285 } 1286 1287 if (gso_partial) 1288 tot_len = skb_shinfo(skb)->gso_size + 1289 SKB_GSO_CB(skb)->data_offset + 1290 skb->head - (unsigned char *)iph; 1291 else 1292 tot_len = skb->len - nhoff; 1293 } else { 1294 if (!fixedid) 1295 iph->id = htons(id++); 1296 tot_len = skb->len - nhoff; 1297 } 1298 iph->tot_len = htons(tot_len); 1299 ip_send_check(iph); 1300 if (encap) 1301 skb_reset_inner_headers(skb); 1302 skb->network_header = (u8 *)iph - skb->head; 1303 } while ((skb = skb->next)); 1304 1305 out: 1306 return segs; 1307 } 1308 EXPORT_SYMBOL(inet_gso_segment); 1309 1310 struct sk_buff **inet_gro_receive(struct sk_buff **head, struct sk_buff *skb) 1311 { 1312 const struct net_offload *ops; 1313 struct sk_buff **pp = NULL; 1314 struct sk_buff *p; 1315 const struct iphdr *iph; 1316 unsigned int hlen; 1317 unsigned int off; 1318 unsigned int id; 1319 int flush = 1; 1320 int proto; 1321 1322 off = skb_gro_offset(skb); 1323 hlen = off + sizeof(*iph); 1324 iph = skb_gro_header_fast(skb, off); 1325 if (skb_gro_header_hard(skb, hlen)) { 1326 iph = skb_gro_header_slow(skb, hlen, off); 1327 if (unlikely(!iph)) 1328 goto out; 1329 } 1330 1331 proto = iph->protocol; 1332 1333 rcu_read_lock(); 1334 ops = rcu_dereference(inet_offloads[proto]); 1335 if (!ops || !ops->callbacks.gro_receive) 1336 goto out_unlock; 1337 1338 if (*(u8 *)iph != 0x45) 1339 goto out_unlock; 1340 1341 if (ip_is_fragment(iph)) 1342 goto out_unlock; 1343 1344 if (unlikely(ip_fast_csum((u8 *)iph, 5))) 1345 goto out_unlock; 1346 1347 id = ntohl(*(__be32 *)&iph->id); 1348 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF)); 1349 id >>= 16; 1350 1351 for (p = *head; p; p = p->next) { 1352 struct iphdr *iph2; 1353 u16 flush_id; 1354 1355 if (!NAPI_GRO_CB(p)->same_flow) 1356 continue; 1357 1358 iph2 = (struct iphdr *)(p->data + off); 1359 /* The above works because, with the exception of the top 1360 * (inner most) layer, we only aggregate pkts with the same 1361 * hdr length so all the hdrs we'll need to verify will start 1362 * at the same offset. 1363 */ 1364 if ((iph->protocol ^ iph2->protocol) | 1365 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) | 1366 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) { 1367 NAPI_GRO_CB(p)->same_flow = 0; 1368 continue; 1369 } 1370 1371 /* All fields must match except length and checksum. */ 1372 NAPI_GRO_CB(p)->flush |= 1373 (iph->ttl ^ iph2->ttl) | 1374 (iph->tos ^ iph2->tos) | 1375 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF)); 1376 1377 NAPI_GRO_CB(p)->flush |= flush; 1378 1379 /* We need to store of the IP ID check to be included later 1380 * when we can verify that this packet does in fact belong 1381 * to a given flow. 1382 */ 1383 flush_id = (u16)(id - ntohs(iph2->id)); 1384 1385 /* This bit of code makes it much easier for us to identify 1386 * the cases where we are doing atomic vs non-atomic IP ID 1387 * checks. Specifically an atomic check can return IP ID 1388 * values 0 - 0xFFFF, while a non-atomic check can only 1389 * return 0 or 0xFFFF. 1390 */ 1391 if (!NAPI_GRO_CB(p)->is_atomic || 1392 !(iph->frag_off & htons(IP_DF))) { 1393 flush_id ^= NAPI_GRO_CB(p)->count; 1394 flush_id = flush_id ? 0xFFFF : 0; 1395 } 1396 1397 /* If the previous IP ID value was based on an atomic 1398 * datagram we can overwrite the value and ignore it. 1399 */ 1400 if (NAPI_GRO_CB(skb)->is_atomic) 1401 NAPI_GRO_CB(p)->flush_id = flush_id; 1402 else 1403 NAPI_GRO_CB(p)->flush_id |= flush_id; 1404 } 1405 1406 NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF)); 1407 NAPI_GRO_CB(skb)->flush |= flush; 1408 skb_set_network_header(skb, off); 1409 /* The above will be needed by the transport layer if there is one 1410 * immediately following this IP hdr. 1411 */ 1412 1413 /* Note : No need to call skb_gro_postpull_rcsum() here, 1414 * as we already checked checksum over ipv4 header was 0 1415 */ 1416 skb_gro_pull(skb, sizeof(*iph)); 1417 skb_set_transport_header(skb, skb_gro_offset(skb)); 1418 1419 pp = call_gro_receive(ops->callbacks.gro_receive, head, skb); 1420 1421 out_unlock: 1422 rcu_read_unlock(); 1423 1424 out: 1425 skb_gro_flush_final(skb, pp, flush); 1426 1427 return pp; 1428 } 1429 EXPORT_SYMBOL(inet_gro_receive); 1430 1431 static struct sk_buff **ipip_gro_receive(struct sk_buff **head, 1432 struct sk_buff *skb) 1433 { 1434 if (NAPI_GRO_CB(skb)->encap_mark) { 1435 NAPI_GRO_CB(skb)->flush = 1; 1436 return NULL; 1437 } 1438 1439 NAPI_GRO_CB(skb)->encap_mark = 1; 1440 1441 return inet_gro_receive(head, skb); 1442 } 1443 1444 #define SECONDS_PER_DAY 86400 1445 1446 /* inet_current_timestamp - Return IP network timestamp 1447 * 1448 * Return milliseconds since midnight in network byte order. 1449 */ 1450 __be32 inet_current_timestamp(void) 1451 { 1452 u32 secs; 1453 u32 msecs; 1454 struct timespec64 ts; 1455 1456 ktime_get_real_ts64(&ts); 1457 1458 /* Get secs since midnight. */ 1459 (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs); 1460 /* Convert to msecs. */ 1461 msecs = secs * MSEC_PER_SEC; 1462 /* Convert nsec to msec. */ 1463 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC; 1464 1465 /* Convert to network byte order. */ 1466 return htonl(msecs); 1467 } 1468 EXPORT_SYMBOL(inet_current_timestamp); 1469 1470 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len) 1471 { 1472 if (sk->sk_family == AF_INET) 1473 return ip_recv_error(sk, msg, len, addr_len); 1474 #if IS_ENABLED(CONFIG_IPV6) 1475 if (sk->sk_family == AF_INET6) 1476 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len); 1477 #endif 1478 return -EINVAL; 1479 } 1480 1481 int inet_gro_complete(struct sk_buff *skb, int nhoff) 1482 { 1483 __be16 newlen = htons(skb->len - nhoff); 1484 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff); 1485 const struct net_offload *ops; 1486 int proto = iph->protocol; 1487 int err = -ENOSYS; 1488 1489 if (skb->encapsulation) { 1490 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP)); 1491 skb_set_inner_network_header(skb, nhoff); 1492 } 1493 1494 csum_replace2(&iph->check, iph->tot_len, newlen); 1495 iph->tot_len = newlen; 1496 1497 rcu_read_lock(); 1498 ops = rcu_dereference(inet_offloads[proto]); 1499 if (WARN_ON(!ops || !ops->callbacks.gro_complete)) 1500 goto out_unlock; 1501 1502 /* Only need to add sizeof(*iph) to get to the next hdr below 1503 * because any hdr with option will have been flushed in 1504 * inet_gro_receive(). 1505 */ 1506 err = ops->callbacks.gro_complete(skb, nhoff + sizeof(*iph)); 1507 1508 out_unlock: 1509 rcu_read_unlock(); 1510 1511 return err; 1512 } 1513 EXPORT_SYMBOL(inet_gro_complete); 1514 1515 static int ipip_gro_complete(struct sk_buff *skb, int nhoff) 1516 { 1517 skb->encapsulation = 1; 1518 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4; 1519 return inet_gro_complete(skb, nhoff); 1520 } 1521 1522 int inet_ctl_sock_create(struct sock **sk, unsigned short family, 1523 unsigned short type, unsigned char protocol, 1524 struct net *net) 1525 { 1526 struct socket *sock; 1527 int rc = sock_create_kern(net, family, type, protocol, &sock); 1528 1529 if (rc == 0) { 1530 *sk = sock->sk; 1531 (*sk)->sk_allocation = GFP_ATOMIC; 1532 /* 1533 * Unhash it so that IP input processing does not even see it, 1534 * we do not wish this socket to see incoming packets. 1535 */ 1536 (*sk)->sk_prot->unhash(*sk); 1537 } 1538 return rc; 1539 } 1540 EXPORT_SYMBOL_GPL(inet_ctl_sock_create); 1541 1542 u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt) 1543 { 1544 return *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt); 1545 } 1546 EXPORT_SYMBOL_GPL(snmp_get_cpu_field); 1547 1548 unsigned long snmp_fold_field(void __percpu *mib, int offt) 1549 { 1550 unsigned long res = 0; 1551 int i; 1552 1553 for_each_possible_cpu(i) 1554 res += snmp_get_cpu_field(mib, i, offt); 1555 return res; 1556 } 1557 EXPORT_SYMBOL_GPL(snmp_fold_field); 1558 1559 #if BITS_PER_LONG==32 1560 1561 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt, 1562 size_t syncp_offset) 1563 { 1564 void *bhptr; 1565 struct u64_stats_sync *syncp; 1566 u64 v; 1567 unsigned int start; 1568 1569 bhptr = per_cpu_ptr(mib, cpu); 1570 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset); 1571 do { 1572 start = u64_stats_fetch_begin_irq(syncp); 1573 v = *(((u64 *)bhptr) + offt); 1574 } while (u64_stats_fetch_retry_irq(syncp, start)); 1575 1576 return v; 1577 } 1578 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64); 1579 1580 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset) 1581 { 1582 u64 res = 0; 1583 int cpu; 1584 1585 for_each_possible_cpu(cpu) { 1586 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset); 1587 } 1588 return res; 1589 } 1590 EXPORT_SYMBOL_GPL(snmp_fold_field64); 1591 #endif 1592 1593 #ifdef CONFIG_IP_MULTICAST 1594 static const struct net_protocol igmp_protocol = { 1595 .handler = igmp_rcv, 1596 .netns_ok = 1, 1597 }; 1598 #endif 1599 1600 /* thinking of making this const? Don't. 1601 * early_demux can change based on sysctl. 1602 */ 1603 static struct net_protocol tcp_protocol = { 1604 .early_demux = tcp_v4_early_demux, 1605 .early_demux_handler = tcp_v4_early_demux, 1606 .handler = tcp_v4_rcv, 1607 .err_handler = tcp_v4_err, 1608 .no_policy = 1, 1609 .netns_ok = 1, 1610 .icmp_strict_tag_validation = 1, 1611 }; 1612 1613 /* thinking of making this const? Don't. 1614 * early_demux can change based on sysctl. 1615 */ 1616 static struct net_protocol udp_protocol = { 1617 .early_demux = udp_v4_early_demux, 1618 .early_demux_handler = udp_v4_early_demux, 1619 .handler = udp_rcv, 1620 .err_handler = udp_err, 1621 .no_policy = 1, 1622 .netns_ok = 1, 1623 }; 1624 1625 static const struct net_protocol icmp_protocol = { 1626 .handler = icmp_rcv, 1627 .err_handler = icmp_err, 1628 .no_policy = 1, 1629 .netns_ok = 1, 1630 }; 1631 1632 static __net_init int ipv4_mib_init_net(struct net *net) 1633 { 1634 int i; 1635 1636 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib); 1637 if (!net->mib.tcp_statistics) 1638 goto err_tcp_mib; 1639 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib); 1640 if (!net->mib.ip_statistics) 1641 goto err_ip_mib; 1642 1643 for_each_possible_cpu(i) { 1644 struct ipstats_mib *af_inet_stats; 1645 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i); 1646 u64_stats_init(&af_inet_stats->syncp); 1647 } 1648 1649 net->mib.net_statistics = alloc_percpu(struct linux_mib); 1650 if (!net->mib.net_statistics) 1651 goto err_net_mib; 1652 net->mib.udp_statistics = alloc_percpu(struct udp_mib); 1653 if (!net->mib.udp_statistics) 1654 goto err_udp_mib; 1655 net->mib.udplite_statistics = alloc_percpu(struct udp_mib); 1656 if (!net->mib.udplite_statistics) 1657 goto err_udplite_mib; 1658 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib); 1659 if (!net->mib.icmp_statistics) 1660 goto err_icmp_mib; 1661 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib), 1662 GFP_KERNEL); 1663 if (!net->mib.icmpmsg_statistics) 1664 goto err_icmpmsg_mib; 1665 1666 tcp_mib_init(net); 1667 return 0; 1668 1669 err_icmpmsg_mib: 1670 free_percpu(net->mib.icmp_statistics); 1671 err_icmp_mib: 1672 free_percpu(net->mib.udplite_statistics); 1673 err_udplite_mib: 1674 free_percpu(net->mib.udp_statistics); 1675 err_udp_mib: 1676 free_percpu(net->mib.net_statistics); 1677 err_net_mib: 1678 free_percpu(net->mib.ip_statistics); 1679 err_ip_mib: 1680 free_percpu(net->mib.tcp_statistics); 1681 err_tcp_mib: 1682 return -ENOMEM; 1683 } 1684 1685 static __net_exit void ipv4_mib_exit_net(struct net *net) 1686 { 1687 kfree(net->mib.icmpmsg_statistics); 1688 free_percpu(net->mib.icmp_statistics); 1689 free_percpu(net->mib.udplite_statistics); 1690 free_percpu(net->mib.udp_statistics); 1691 free_percpu(net->mib.net_statistics); 1692 free_percpu(net->mib.ip_statistics); 1693 free_percpu(net->mib.tcp_statistics); 1694 } 1695 1696 static __net_initdata struct pernet_operations ipv4_mib_ops = { 1697 .init = ipv4_mib_init_net, 1698 .exit = ipv4_mib_exit_net, 1699 }; 1700 1701 static int __init init_ipv4_mibs(void) 1702 { 1703 return register_pernet_subsys(&ipv4_mib_ops); 1704 } 1705 1706 static __net_init int inet_init_net(struct net *net) 1707 { 1708 /* 1709 * Set defaults for local port range 1710 */ 1711 seqlock_init(&net->ipv4.ip_local_ports.lock); 1712 net->ipv4.ip_local_ports.range[0] = 32768; 1713 net->ipv4.ip_local_ports.range[1] = 60999; 1714 1715 seqlock_init(&net->ipv4.ping_group_range.lock); 1716 /* 1717 * Sane defaults - nobody may create ping sockets. 1718 * Boot scripts should set this to distro-specific group. 1719 */ 1720 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1); 1721 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0); 1722 1723 /* Default values for sysctl-controlled parameters. 1724 * We set them here, in case sysctl is not compiled. 1725 */ 1726 net->ipv4.sysctl_ip_default_ttl = IPDEFTTL; 1727 net->ipv4.sysctl_ip_dynaddr = 0; 1728 net->ipv4.sysctl_ip_early_demux = 1; 1729 net->ipv4.sysctl_udp_early_demux = 1; 1730 net->ipv4.sysctl_tcp_early_demux = 1; 1731 #ifdef CONFIG_SYSCTL 1732 net->ipv4.sysctl_ip_prot_sock = PROT_SOCK; 1733 #endif 1734 1735 /* Some igmp sysctl, whose values are always used */ 1736 net->ipv4.sysctl_igmp_max_memberships = 20; 1737 net->ipv4.sysctl_igmp_max_msf = 10; 1738 /* IGMP reports for link-local multicast groups are enabled by default */ 1739 net->ipv4.sysctl_igmp_llm_reports = 1; 1740 net->ipv4.sysctl_igmp_qrv = 2; 1741 1742 return 0; 1743 } 1744 1745 static __net_exit void inet_exit_net(struct net *net) 1746 { 1747 } 1748 1749 static __net_initdata struct pernet_operations af_inet_ops = { 1750 .init = inet_init_net, 1751 .exit = inet_exit_net, 1752 }; 1753 1754 static int __init init_inet_pernet_ops(void) 1755 { 1756 return register_pernet_subsys(&af_inet_ops); 1757 } 1758 1759 static int ipv4_proc_init(void); 1760 1761 /* 1762 * IP protocol layer initialiser 1763 */ 1764 1765 static struct packet_offload ip_packet_offload __read_mostly = { 1766 .type = cpu_to_be16(ETH_P_IP), 1767 .callbacks = { 1768 .gso_segment = inet_gso_segment, 1769 .gro_receive = inet_gro_receive, 1770 .gro_complete = inet_gro_complete, 1771 }, 1772 }; 1773 1774 static const struct net_offload ipip_offload = { 1775 .callbacks = { 1776 .gso_segment = inet_gso_segment, 1777 .gro_receive = ipip_gro_receive, 1778 .gro_complete = ipip_gro_complete, 1779 }, 1780 }; 1781 1782 static int __init ipip_offload_init(void) 1783 { 1784 return inet_add_offload(&ipip_offload, IPPROTO_IPIP); 1785 } 1786 1787 static int __init ipv4_offload_init(void) 1788 { 1789 /* 1790 * Add offloads 1791 */ 1792 if (udpv4_offload_init() < 0) 1793 pr_crit("%s: Cannot add UDP protocol offload\n", __func__); 1794 if (tcpv4_offload_init() < 0) 1795 pr_crit("%s: Cannot add TCP protocol offload\n", __func__); 1796 if (ipip_offload_init() < 0) 1797 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__); 1798 1799 dev_add_offload(&ip_packet_offload); 1800 return 0; 1801 } 1802 1803 fs_initcall(ipv4_offload_init); 1804 1805 static struct packet_type ip_packet_type __read_mostly = { 1806 .type = cpu_to_be16(ETH_P_IP), 1807 .func = ip_rcv, 1808 }; 1809 1810 static int __init inet_init(void) 1811 { 1812 struct inet_protosw *q; 1813 struct list_head *r; 1814 int rc = -EINVAL; 1815 1816 sock_skb_cb_check_size(sizeof(struct inet_skb_parm)); 1817 1818 rc = proto_register(&tcp_prot, 1); 1819 if (rc) 1820 goto out; 1821 1822 rc = proto_register(&udp_prot, 1); 1823 if (rc) 1824 goto out_unregister_tcp_proto; 1825 1826 rc = proto_register(&raw_prot, 1); 1827 if (rc) 1828 goto out_unregister_udp_proto; 1829 1830 rc = proto_register(&ping_prot, 1); 1831 if (rc) 1832 goto out_unregister_raw_proto; 1833 1834 /* 1835 * Tell SOCKET that we are alive... 1836 */ 1837 1838 (void)sock_register(&inet_family_ops); 1839 1840 #ifdef CONFIG_SYSCTL 1841 ip_static_sysctl_init(); 1842 #endif 1843 1844 /* 1845 * Add all the base protocols. 1846 */ 1847 1848 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0) 1849 pr_crit("%s: Cannot add ICMP protocol\n", __func__); 1850 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0) 1851 pr_crit("%s: Cannot add UDP protocol\n", __func__); 1852 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0) 1853 pr_crit("%s: Cannot add TCP protocol\n", __func__); 1854 #ifdef CONFIG_IP_MULTICAST 1855 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0) 1856 pr_crit("%s: Cannot add IGMP protocol\n", __func__); 1857 #endif 1858 1859 /* Register the socket-side information for inet_create. */ 1860 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r) 1861 INIT_LIST_HEAD(r); 1862 1863 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q) 1864 inet_register_protosw(q); 1865 1866 /* 1867 * Set the ARP module up 1868 */ 1869 1870 arp_init(); 1871 1872 /* 1873 * Set the IP module up 1874 */ 1875 1876 ip_init(); 1877 1878 /* Setup TCP slab cache for open requests. */ 1879 tcp_init(); 1880 1881 /* Setup UDP memory threshold */ 1882 udp_init(); 1883 1884 /* Add UDP-Lite (RFC 3828) */ 1885 udplite4_register(); 1886 1887 ping_init(); 1888 1889 /* 1890 * Set the ICMP layer up 1891 */ 1892 1893 if (icmp_init() < 0) 1894 panic("Failed to create the ICMP control socket.\n"); 1895 1896 /* 1897 * Initialise the multicast router 1898 */ 1899 #if defined(CONFIG_IP_MROUTE) 1900 if (ip_mr_init()) 1901 pr_crit("%s: Cannot init ipv4 mroute\n", __func__); 1902 #endif 1903 1904 if (init_inet_pernet_ops()) 1905 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__); 1906 /* 1907 * Initialise per-cpu ipv4 mibs 1908 */ 1909 1910 if (init_ipv4_mibs()) 1911 pr_crit("%s: Cannot init ipv4 mibs\n", __func__); 1912 1913 ipv4_proc_init(); 1914 1915 ipfrag_init(); 1916 1917 dev_add_pack(&ip_packet_type); 1918 1919 ip_tunnel_core_init(); 1920 1921 rc = 0; 1922 out: 1923 return rc; 1924 out_unregister_raw_proto: 1925 proto_unregister(&raw_prot); 1926 out_unregister_udp_proto: 1927 proto_unregister(&udp_prot); 1928 out_unregister_tcp_proto: 1929 proto_unregister(&tcp_prot); 1930 goto out; 1931 } 1932 1933 fs_initcall(inet_init); 1934 1935 /* ------------------------------------------------------------------------ */ 1936 1937 #ifdef CONFIG_PROC_FS 1938 static int __init ipv4_proc_init(void) 1939 { 1940 int rc = 0; 1941 1942 if (raw_proc_init()) 1943 goto out_raw; 1944 if (tcp4_proc_init()) 1945 goto out_tcp; 1946 if (udp4_proc_init()) 1947 goto out_udp; 1948 if (ping_proc_init()) 1949 goto out_ping; 1950 if (ip_misc_proc_init()) 1951 goto out_misc; 1952 out: 1953 return rc; 1954 out_misc: 1955 ping_proc_exit(); 1956 out_ping: 1957 udp4_proc_exit(); 1958 out_udp: 1959 tcp4_proc_exit(); 1960 out_tcp: 1961 raw_proc_exit(); 1962 out_raw: 1963 rc = -ENOMEM; 1964 goto out; 1965 } 1966 1967 #else /* CONFIG_PROC_FS */ 1968 static int __init ipv4_proc_init(void) 1969 { 1970 return 0; 1971 } 1972 #endif /* CONFIG_PROC_FS */ 1973