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