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/module.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 <asm/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(atomic_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 /* Check special setups for testing purpose to enable TFO w/o 215 * requiring TCP_FASTOPEN sockopt. 216 * Note that only TCP sockets (SOCK_STREAM) will reach here. 217 * Also fastopenq may already been allocated because this 218 * socket was in TCP_LISTEN state previously but was 219 * shutdown() (rather than close()). 220 */ 221 if ((sysctl_tcp_fastopen & TFO_SERVER_ENABLE) != 0 && 222 !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) { 223 if ((sysctl_tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) != 0) 224 fastopen_queue_tune(sk, backlog); 225 else if ((sysctl_tcp_fastopen & 226 TFO_SERVER_WO_SOCKOPT2) != 0) 227 fastopen_queue_tune(sk, 228 ((uint)sysctl_tcp_fastopen) >> 16); 229 230 tcp_fastopen_init_key_once(true); 231 } 232 err = inet_csk_listen_start(sk, backlog); 233 if (err) 234 goto out; 235 } 236 sk->sk_max_ack_backlog = backlog; 237 err = 0; 238 239 out: 240 release_sock(sk); 241 return err; 242 } 243 EXPORT_SYMBOL(inet_listen); 244 245 /* 246 * Create an inet socket. 247 */ 248 249 static int inet_create(struct net *net, struct socket *sock, int protocol, 250 int kern) 251 { 252 struct sock *sk; 253 struct inet_protosw *answer; 254 struct inet_sock *inet; 255 struct proto *answer_prot; 256 unsigned char answer_flags; 257 int try_loading_module = 0; 258 int err; 259 260 if (protocol < 0 || protocol >= IPPROTO_MAX) 261 return -EINVAL; 262 263 sock->state = SS_UNCONNECTED; 264 265 /* Look for the requested type/protocol pair. */ 266 lookup_protocol: 267 err = -ESOCKTNOSUPPORT; 268 rcu_read_lock(); 269 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) { 270 271 err = 0; 272 /* Check the non-wild match. */ 273 if (protocol == answer->protocol) { 274 if (protocol != IPPROTO_IP) 275 break; 276 } else { 277 /* Check for the two wild cases. */ 278 if (IPPROTO_IP == protocol) { 279 protocol = answer->protocol; 280 break; 281 } 282 if (IPPROTO_IP == answer->protocol) 283 break; 284 } 285 err = -EPROTONOSUPPORT; 286 } 287 288 if (unlikely(err)) { 289 if (try_loading_module < 2) { 290 rcu_read_unlock(); 291 /* 292 * Be more specific, e.g. net-pf-2-proto-132-type-1 293 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM) 294 */ 295 if (++try_loading_module == 1) 296 request_module("net-pf-%d-proto-%d-type-%d", 297 PF_INET, protocol, sock->type); 298 /* 299 * Fall back to generic, e.g. net-pf-2-proto-132 300 * (net-pf-PF_INET-proto-IPPROTO_SCTP) 301 */ 302 else 303 request_module("net-pf-%d-proto-%d", 304 PF_INET, protocol); 305 goto lookup_protocol; 306 } else 307 goto out_rcu_unlock; 308 } 309 310 err = -EPERM; 311 if (sock->type == SOCK_RAW && !kern && 312 !ns_capable(net->user_ns, CAP_NET_RAW)) 313 goto out_rcu_unlock; 314 315 sock->ops = answer->ops; 316 answer_prot = answer->prot; 317 answer_flags = answer->flags; 318 rcu_read_unlock(); 319 320 WARN_ON(!answer_prot->slab); 321 322 err = -ENOBUFS; 323 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern); 324 if (!sk) 325 goto out; 326 327 err = 0; 328 if (INET_PROTOSW_REUSE & answer_flags) 329 sk->sk_reuse = SK_CAN_REUSE; 330 331 inet = inet_sk(sk); 332 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0; 333 334 inet->nodefrag = 0; 335 336 if (SOCK_RAW == sock->type) { 337 inet->inet_num = protocol; 338 if (IPPROTO_RAW == protocol) 339 inet->hdrincl = 1; 340 } 341 342 if (net->ipv4.sysctl_ip_no_pmtu_disc) 343 inet->pmtudisc = IP_PMTUDISC_DONT; 344 else 345 inet->pmtudisc = IP_PMTUDISC_WANT; 346 347 inet->inet_id = 0; 348 349 sock_init_data(sock, sk); 350 351 sk->sk_destruct = inet_sock_destruct; 352 sk->sk_protocol = protocol; 353 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv; 354 355 inet->uc_ttl = -1; 356 inet->mc_loop = 1; 357 inet->mc_ttl = 1; 358 inet->mc_all = 1; 359 inet->mc_index = 0; 360 inet->mc_list = NULL; 361 inet->rcv_tos = 0; 362 363 sk_refcnt_debug_inc(sk); 364 365 if (inet->inet_num) { 366 /* It assumes that any protocol which allows 367 * the user to assign a number at socket 368 * creation time automatically 369 * shares. 370 */ 371 inet->inet_sport = htons(inet->inet_num); 372 /* Add to protocol hash chains. */ 373 err = sk->sk_prot->hash(sk); 374 if (err) { 375 sk_common_release(sk); 376 goto out; 377 } 378 } 379 380 if (sk->sk_prot->init) { 381 err = sk->sk_prot->init(sk); 382 if (err) 383 sk_common_release(sk); 384 } 385 out: 386 return err; 387 out_rcu_unlock: 388 rcu_read_unlock(); 389 goto out; 390 } 391 392 393 /* 394 * The peer socket should always be NULL (or else). When we call this 395 * function we are destroying the object and from then on nobody 396 * should refer to it. 397 */ 398 int inet_release(struct socket *sock) 399 { 400 struct sock *sk = sock->sk; 401 402 if (sk) { 403 long timeout; 404 405 /* Applications forget to leave groups before exiting */ 406 ip_mc_drop_socket(sk); 407 408 /* If linger is set, we don't return until the close 409 * is complete. Otherwise we return immediately. The 410 * actually closing is done the same either way. 411 * 412 * If the close is due to the process exiting, we never 413 * linger.. 414 */ 415 timeout = 0; 416 if (sock_flag(sk, SOCK_LINGER) && 417 !(current->flags & PF_EXITING)) 418 timeout = sk->sk_lingertime; 419 sock->sk = NULL; 420 sk->sk_prot->close(sk, timeout); 421 } 422 return 0; 423 } 424 EXPORT_SYMBOL(inet_release); 425 426 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) 427 { 428 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr; 429 struct sock *sk = sock->sk; 430 struct inet_sock *inet = inet_sk(sk); 431 struct net *net = sock_net(sk); 432 unsigned short snum; 433 int chk_addr_ret; 434 u32 tb_id = RT_TABLE_LOCAL; 435 int err; 436 437 /* If the socket has its own bind function then use it. (RAW) */ 438 if (sk->sk_prot->bind) { 439 err = sk->sk_prot->bind(sk, uaddr, addr_len); 440 goto out; 441 } 442 err = -EINVAL; 443 if (addr_len < sizeof(struct sockaddr_in)) 444 goto out; 445 446 if (addr->sin_family != AF_INET) { 447 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET) 448 * only if s_addr is INADDR_ANY. 449 */ 450 err = -EAFNOSUPPORT; 451 if (addr->sin_family != AF_UNSPEC || 452 addr->sin_addr.s_addr != htonl(INADDR_ANY)) 453 goto out; 454 } 455 456 tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id; 457 chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id); 458 459 /* Not specified by any standard per-se, however it breaks too 460 * many applications when removed. It is unfortunate since 461 * allowing applications to make a non-local bind solves 462 * several problems with systems using dynamic addressing. 463 * (ie. your servers still start up even if your ISDN link 464 * is temporarily down) 465 */ 466 err = -EADDRNOTAVAIL; 467 if (!net->ipv4.sysctl_ip_nonlocal_bind && 468 !(inet->freebind || inet->transparent) && 469 addr->sin_addr.s_addr != htonl(INADDR_ANY) && 470 chk_addr_ret != RTN_LOCAL && 471 chk_addr_ret != RTN_MULTICAST && 472 chk_addr_ret != RTN_BROADCAST) 473 goto out; 474 475 snum = ntohs(addr->sin_port); 476 err = -EACCES; 477 if (snum && snum < PROT_SOCK && 478 !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE)) 479 goto out; 480 481 /* We keep a pair of addresses. rcv_saddr is the one 482 * used by hash lookups, and saddr is used for transmit. 483 * 484 * In the BSD API these are the same except where it 485 * would be illegal to use them (multicast/broadcast) in 486 * which case the sending device address is used. 487 */ 488 lock_sock(sk); 489 490 /* Check these errors (active socket, double bind). */ 491 err = -EINVAL; 492 if (sk->sk_state != TCP_CLOSE || inet->inet_num) 493 goto out_release_sock; 494 495 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr; 496 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST) 497 inet->inet_saddr = 0; /* Use device */ 498 499 /* Make sure we are allowed to bind here. */ 500 if ((snum || !inet->bind_address_no_port) && 501 sk->sk_prot->get_port(sk, snum)) { 502 inet->inet_saddr = inet->inet_rcv_saddr = 0; 503 err = -EADDRINUSE; 504 goto out_release_sock; 505 } 506 507 if (inet->inet_rcv_saddr) 508 sk->sk_userlocks |= SOCK_BINDADDR_LOCK; 509 if (snum) 510 sk->sk_userlocks |= SOCK_BINDPORT_LOCK; 511 inet->inet_sport = htons(inet->inet_num); 512 inet->inet_daddr = 0; 513 inet->inet_dport = 0; 514 sk_dst_reset(sk); 515 err = 0; 516 out_release_sock: 517 release_sock(sk); 518 out: 519 return err; 520 } 521 EXPORT_SYMBOL(inet_bind); 522 523 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr, 524 int addr_len, int flags) 525 { 526 struct sock *sk = sock->sk; 527 528 if (addr_len < sizeof(uaddr->sa_family)) 529 return -EINVAL; 530 if (uaddr->sa_family == AF_UNSPEC) 531 return sk->sk_prot->disconnect(sk, flags); 532 533 if (!inet_sk(sk)->inet_num && inet_autobind(sk)) 534 return -EAGAIN; 535 return sk->sk_prot->connect(sk, uaddr, addr_len); 536 } 537 EXPORT_SYMBOL(inet_dgram_connect); 538 539 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias) 540 { 541 DEFINE_WAIT(wait); 542 543 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 544 sk->sk_write_pending += writebias; 545 546 /* Basic assumption: if someone sets sk->sk_err, he _must_ 547 * change state of the socket from TCP_SYN_*. 548 * Connect() does not allow to get error notifications 549 * without closing the socket. 550 */ 551 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 552 release_sock(sk); 553 timeo = schedule_timeout(timeo); 554 lock_sock(sk); 555 if (signal_pending(current) || !timeo) 556 break; 557 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 558 } 559 finish_wait(sk_sleep(sk), &wait); 560 sk->sk_write_pending -= writebias; 561 return timeo; 562 } 563 564 /* 565 * Connect to a remote host. There is regrettably still a little 566 * TCP 'magic' in here. 567 */ 568 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr, 569 int addr_len, int flags) 570 { 571 struct sock *sk = sock->sk; 572 int err; 573 long timeo; 574 575 if (addr_len < sizeof(uaddr->sa_family)) 576 return -EINVAL; 577 578 if (uaddr->sa_family == AF_UNSPEC) { 579 err = sk->sk_prot->disconnect(sk, flags); 580 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED; 581 goto out; 582 } 583 584 switch (sock->state) { 585 default: 586 err = -EINVAL; 587 goto out; 588 case SS_CONNECTED: 589 err = -EISCONN; 590 goto out; 591 case SS_CONNECTING: 592 err = -EALREADY; 593 /* Fall out of switch with err, set for this state */ 594 break; 595 case SS_UNCONNECTED: 596 err = -EISCONN; 597 if (sk->sk_state != TCP_CLOSE) 598 goto out; 599 600 err = sk->sk_prot->connect(sk, uaddr, addr_len); 601 if (err < 0) 602 goto out; 603 604 sock->state = SS_CONNECTING; 605 606 /* Just entered SS_CONNECTING state; the only 607 * difference is that return value in non-blocking 608 * case is EINPROGRESS, rather than EALREADY. 609 */ 610 err = -EINPROGRESS; 611 break; 612 } 613 614 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK); 615 616 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 617 int writebias = (sk->sk_protocol == IPPROTO_TCP) && 618 tcp_sk(sk)->fastopen_req && 619 tcp_sk(sk)->fastopen_req->data ? 1 : 0; 620 621 /* Error code is set above */ 622 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias)) 623 goto out; 624 625 err = sock_intr_errno(timeo); 626 if (signal_pending(current)) 627 goto out; 628 } 629 630 /* Connection was closed by RST, timeout, ICMP error 631 * or another process disconnected us. 632 */ 633 if (sk->sk_state == TCP_CLOSE) 634 goto sock_error; 635 636 /* sk->sk_err may be not zero now, if RECVERR was ordered by user 637 * and error was received after socket entered established state. 638 * Hence, it is handled normally after connect() return successfully. 639 */ 640 641 sock->state = SS_CONNECTED; 642 err = 0; 643 out: 644 return err; 645 646 sock_error: 647 err = sock_error(sk) ? : -ECONNABORTED; 648 sock->state = SS_UNCONNECTED; 649 if (sk->sk_prot->disconnect(sk, flags)) 650 sock->state = SS_DISCONNECTING; 651 goto out; 652 } 653 EXPORT_SYMBOL(__inet_stream_connect); 654 655 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr, 656 int addr_len, int flags) 657 { 658 int err; 659 660 lock_sock(sock->sk); 661 err = __inet_stream_connect(sock, uaddr, addr_len, flags); 662 release_sock(sock->sk); 663 return err; 664 } 665 EXPORT_SYMBOL(inet_stream_connect); 666 667 /* 668 * Accept a pending connection. The TCP layer now gives BSD semantics. 669 */ 670 671 int inet_accept(struct socket *sock, struct socket *newsock, int flags) 672 { 673 struct sock *sk1 = sock->sk; 674 int err = -EINVAL; 675 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err); 676 677 if (!sk2) 678 goto do_err; 679 680 lock_sock(sk2); 681 682 sock_rps_record_flow(sk2); 683 WARN_ON(!((1 << sk2->sk_state) & 684 (TCPF_ESTABLISHED | TCPF_SYN_RECV | 685 TCPF_CLOSE_WAIT | TCPF_CLOSE))); 686 687 sock_graft(sk2, newsock); 688 689 newsock->state = SS_CONNECTED; 690 err = 0; 691 release_sock(sk2); 692 do_err: 693 return err; 694 } 695 EXPORT_SYMBOL(inet_accept); 696 697 698 /* 699 * This does both peername and sockname. 700 */ 701 int inet_getname(struct socket *sock, struct sockaddr *uaddr, 702 int *uaddr_len, int peer) 703 { 704 struct sock *sk = sock->sk; 705 struct inet_sock *inet = inet_sk(sk); 706 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr); 707 708 sin->sin_family = AF_INET; 709 if (peer) { 710 if (!inet->inet_dport || 711 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) && 712 peer == 1)) 713 return -ENOTCONN; 714 sin->sin_port = inet->inet_dport; 715 sin->sin_addr.s_addr = inet->inet_daddr; 716 } else { 717 __be32 addr = inet->inet_rcv_saddr; 718 if (!addr) 719 addr = inet->inet_saddr; 720 sin->sin_port = inet->inet_sport; 721 sin->sin_addr.s_addr = addr; 722 } 723 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 724 *uaddr_len = sizeof(*sin); 725 return 0; 726 } 727 EXPORT_SYMBOL(inet_getname); 728 729 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size) 730 { 731 struct sock *sk = sock->sk; 732 733 sock_rps_record_flow(sk); 734 735 /* We may need to bind the socket. */ 736 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind && 737 inet_autobind(sk)) 738 return -EAGAIN; 739 740 return sk->sk_prot->sendmsg(sk, msg, size); 741 } 742 EXPORT_SYMBOL(inet_sendmsg); 743 744 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, 745 size_t size, int flags) 746 { 747 struct sock *sk = sock->sk; 748 749 sock_rps_record_flow(sk); 750 751 /* We may need to bind the socket. */ 752 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind && 753 inet_autobind(sk)) 754 return -EAGAIN; 755 756 if (sk->sk_prot->sendpage) 757 return sk->sk_prot->sendpage(sk, page, offset, size, flags); 758 return sock_no_sendpage(sock, page, offset, size, flags); 759 } 760 EXPORT_SYMBOL(inet_sendpage); 761 762 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 763 int flags) 764 { 765 struct sock *sk = sock->sk; 766 int addr_len = 0; 767 int err; 768 769 sock_rps_record_flow(sk); 770 771 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT, 772 flags & ~MSG_DONTWAIT, &addr_len); 773 if (err >= 0) 774 msg->msg_namelen = addr_len; 775 return err; 776 } 777 EXPORT_SYMBOL(inet_recvmsg); 778 779 int inet_shutdown(struct socket *sock, int how) 780 { 781 struct sock *sk = sock->sk; 782 int err = 0; 783 784 /* This should really check to make sure 785 * the socket is a TCP socket. (WHY AC...) 786 */ 787 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and 788 1->2 bit 2 snds. 789 2->3 */ 790 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */ 791 return -EINVAL; 792 793 lock_sock(sk); 794 if (sock->state == SS_CONNECTING) { 795 if ((1 << sk->sk_state) & 796 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE)) 797 sock->state = SS_DISCONNECTING; 798 else 799 sock->state = SS_CONNECTED; 800 } 801 802 switch (sk->sk_state) { 803 case TCP_CLOSE: 804 err = -ENOTCONN; 805 /* Hack to wake up other listeners, who can poll for 806 POLLHUP, even on eg. unconnected UDP sockets -- RR */ 807 default: 808 sk->sk_shutdown |= how; 809 if (sk->sk_prot->shutdown) 810 sk->sk_prot->shutdown(sk, how); 811 break; 812 813 /* Remaining two branches are temporary solution for missing 814 * close() in multithreaded environment. It is _not_ a good idea, 815 * but we have no choice until close() is repaired at VFS level. 816 */ 817 case TCP_LISTEN: 818 if (!(how & RCV_SHUTDOWN)) 819 break; 820 /* Fall through */ 821 case TCP_SYN_SENT: 822 err = sk->sk_prot->disconnect(sk, O_NONBLOCK); 823 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED; 824 break; 825 } 826 827 /* Wake up anyone sleeping in poll. */ 828 sk->sk_state_change(sk); 829 release_sock(sk); 830 return err; 831 } 832 EXPORT_SYMBOL(inet_shutdown); 833 834 /* 835 * ioctl() calls you can issue on an INET socket. Most of these are 836 * device configuration and stuff and very rarely used. Some ioctls 837 * pass on to the socket itself. 838 * 839 * NOTE: I like the idea of a module for the config stuff. ie ifconfig 840 * loads the devconfigure module does its configuring and unloads it. 841 * There's a good 20K of config code hanging around the kernel. 842 */ 843 844 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 845 { 846 struct sock *sk = sock->sk; 847 int err = 0; 848 struct net *net = sock_net(sk); 849 850 switch (cmd) { 851 case SIOCGSTAMP: 852 err = sock_get_timestamp(sk, (struct timeval __user *)arg); 853 break; 854 case SIOCGSTAMPNS: 855 err = sock_get_timestampns(sk, (struct timespec __user *)arg); 856 break; 857 case SIOCADDRT: 858 case SIOCDELRT: 859 case SIOCRTMSG: 860 err = ip_rt_ioctl(net, cmd, (void __user *)arg); 861 break; 862 case SIOCDARP: 863 case SIOCGARP: 864 case SIOCSARP: 865 err = arp_ioctl(net, cmd, (void __user *)arg); 866 break; 867 case SIOCGIFADDR: 868 case SIOCSIFADDR: 869 case SIOCGIFBRDADDR: 870 case SIOCSIFBRDADDR: 871 case SIOCGIFNETMASK: 872 case SIOCSIFNETMASK: 873 case SIOCGIFDSTADDR: 874 case SIOCSIFDSTADDR: 875 case SIOCSIFPFLAGS: 876 case SIOCGIFPFLAGS: 877 case SIOCSIFFLAGS: 878 err = devinet_ioctl(net, cmd, (void __user *)arg); 879 break; 880 default: 881 if (sk->sk_prot->ioctl) 882 err = sk->sk_prot->ioctl(sk, cmd, arg); 883 else 884 err = -ENOIOCTLCMD; 885 break; 886 } 887 return err; 888 } 889 EXPORT_SYMBOL(inet_ioctl); 890 891 #ifdef CONFIG_COMPAT 892 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 893 { 894 struct sock *sk = sock->sk; 895 int err = -ENOIOCTLCMD; 896 897 if (sk->sk_prot->compat_ioctl) 898 err = sk->sk_prot->compat_ioctl(sk, cmd, arg); 899 900 return err; 901 } 902 #endif 903 904 const struct proto_ops inet_stream_ops = { 905 .family = PF_INET, 906 .owner = THIS_MODULE, 907 .release = inet_release, 908 .bind = inet_bind, 909 .connect = inet_stream_connect, 910 .socketpair = sock_no_socketpair, 911 .accept = inet_accept, 912 .getname = inet_getname, 913 .poll = tcp_poll, 914 .ioctl = inet_ioctl, 915 .listen = inet_listen, 916 .shutdown = inet_shutdown, 917 .setsockopt = sock_common_setsockopt, 918 .getsockopt = sock_common_getsockopt, 919 .sendmsg = inet_sendmsg, 920 .recvmsg = inet_recvmsg, 921 .mmap = sock_no_mmap, 922 .sendpage = inet_sendpage, 923 .splice_read = tcp_splice_read, 924 #ifdef CONFIG_COMPAT 925 .compat_setsockopt = compat_sock_common_setsockopt, 926 .compat_getsockopt = compat_sock_common_getsockopt, 927 .compat_ioctl = inet_compat_ioctl, 928 #endif 929 }; 930 EXPORT_SYMBOL(inet_stream_ops); 931 932 const struct proto_ops inet_dgram_ops = { 933 .family = PF_INET, 934 .owner = THIS_MODULE, 935 .release = inet_release, 936 .bind = inet_bind, 937 .connect = inet_dgram_connect, 938 .socketpair = sock_no_socketpair, 939 .accept = sock_no_accept, 940 .getname = inet_getname, 941 .poll = udp_poll, 942 .ioctl = inet_ioctl, 943 .listen = sock_no_listen, 944 .shutdown = inet_shutdown, 945 .setsockopt = sock_common_setsockopt, 946 .getsockopt = sock_common_getsockopt, 947 .sendmsg = inet_sendmsg, 948 .recvmsg = inet_recvmsg, 949 .mmap = sock_no_mmap, 950 .sendpage = inet_sendpage, 951 .set_peek_off = sk_set_peek_off, 952 #ifdef CONFIG_COMPAT 953 .compat_setsockopt = compat_sock_common_setsockopt, 954 .compat_getsockopt = compat_sock_common_getsockopt, 955 .compat_ioctl = inet_compat_ioctl, 956 #endif 957 }; 958 EXPORT_SYMBOL(inet_dgram_ops); 959 960 /* 961 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without 962 * udp_poll 963 */ 964 static const struct proto_ops inet_sockraw_ops = { 965 .family = PF_INET, 966 .owner = THIS_MODULE, 967 .release = inet_release, 968 .bind = inet_bind, 969 .connect = inet_dgram_connect, 970 .socketpair = sock_no_socketpair, 971 .accept = sock_no_accept, 972 .getname = inet_getname, 973 .poll = datagram_poll, 974 .ioctl = inet_ioctl, 975 .listen = sock_no_listen, 976 .shutdown = inet_shutdown, 977 .setsockopt = sock_common_setsockopt, 978 .getsockopt = sock_common_getsockopt, 979 .sendmsg = inet_sendmsg, 980 .recvmsg = inet_recvmsg, 981 .mmap = sock_no_mmap, 982 .sendpage = inet_sendpage, 983 #ifdef CONFIG_COMPAT 984 .compat_setsockopt = compat_sock_common_setsockopt, 985 .compat_getsockopt = compat_sock_common_getsockopt, 986 .compat_ioctl = inet_compat_ioctl, 987 #endif 988 }; 989 990 static const struct net_proto_family inet_family_ops = { 991 .family = PF_INET, 992 .create = inet_create, 993 .owner = THIS_MODULE, 994 }; 995 996 /* Upon startup we insert all the elements in inetsw_array[] into 997 * the linked list inetsw. 998 */ 999 static struct inet_protosw inetsw_array[] = 1000 { 1001 { 1002 .type = SOCK_STREAM, 1003 .protocol = IPPROTO_TCP, 1004 .prot = &tcp_prot, 1005 .ops = &inet_stream_ops, 1006 .flags = INET_PROTOSW_PERMANENT | 1007 INET_PROTOSW_ICSK, 1008 }, 1009 1010 { 1011 .type = SOCK_DGRAM, 1012 .protocol = IPPROTO_UDP, 1013 .prot = &udp_prot, 1014 .ops = &inet_dgram_ops, 1015 .flags = INET_PROTOSW_PERMANENT, 1016 }, 1017 1018 { 1019 .type = SOCK_DGRAM, 1020 .protocol = IPPROTO_ICMP, 1021 .prot = &ping_prot, 1022 .ops = &inet_dgram_ops, 1023 .flags = INET_PROTOSW_REUSE, 1024 }, 1025 1026 { 1027 .type = SOCK_RAW, 1028 .protocol = IPPROTO_IP, /* wild card */ 1029 .prot = &raw_prot, 1030 .ops = &inet_sockraw_ops, 1031 .flags = INET_PROTOSW_REUSE, 1032 } 1033 }; 1034 1035 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array) 1036 1037 void inet_register_protosw(struct inet_protosw *p) 1038 { 1039 struct list_head *lh; 1040 struct inet_protosw *answer; 1041 int protocol = p->protocol; 1042 struct list_head *last_perm; 1043 1044 spin_lock_bh(&inetsw_lock); 1045 1046 if (p->type >= SOCK_MAX) 1047 goto out_illegal; 1048 1049 /* If we are trying to override a permanent protocol, bail. */ 1050 last_perm = &inetsw[p->type]; 1051 list_for_each(lh, &inetsw[p->type]) { 1052 answer = list_entry(lh, struct inet_protosw, list); 1053 /* Check only the non-wild match. */ 1054 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0) 1055 break; 1056 if (protocol == answer->protocol) 1057 goto out_permanent; 1058 last_perm = lh; 1059 } 1060 1061 /* Add the new entry after the last permanent entry if any, so that 1062 * the new entry does not override a permanent entry when matched with 1063 * a wild-card protocol. But it is allowed to override any existing 1064 * non-permanent entry. This means that when we remove this entry, the 1065 * system automatically returns to the old behavior. 1066 */ 1067 list_add_rcu(&p->list, last_perm); 1068 out: 1069 spin_unlock_bh(&inetsw_lock); 1070 1071 return; 1072 1073 out_permanent: 1074 pr_err("Attempt to override permanent protocol %d\n", protocol); 1075 goto out; 1076 1077 out_illegal: 1078 pr_err("Ignoring attempt to register invalid socket type %d\n", 1079 p->type); 1080 goto out; 1081 } 1082 EXPORT_SYMBOL(inet_register_protosw); 1083 1084 void inet_unregister_protosw(struct inet_protosw *p) 1085 { 1086 if (INET_PROTOSW_PERMANENT & p->flags) { 1087 pr_err("Attempt to unregister permanent protocol %d\n", 1088 p->protocol); 1089 } else { 1090 spin_lock_bh(&inetsw_lock); 1091 list_del_rcu(&p->list); 1092 spin_unlock_bh(&inetsw_lock); 1093 1094 synchronize_net(); 1095 } 1096 } 1097 EXPORT_SYMBOL(inet_unregister_protosw); 1098 1099 static int inet_sk_reselect_saddr(struct sock *sk) 1100 { 1101 struct inet_sock *inet = inet_sk(sk); 1102 __be32 old_saddr = inet->inet_saddr; 1103 __be32 daddr = inet->inet_daddr; 1104 struct flowi4 *fl4; 1105 struct rtable *rt; 1106 __be32 new_saddr; 1107 struct ip_options_rcu *inet_opt; 1108 1109 inet_opt = rcu_dereference_protected(inet->inet_opt, 1110 lockdep_sock_is_held(sk)); 1111 if (inet_opt && inet_opt->opt.srr) 1112 daddr = inet_opt->opt.faddr; 1113 1114 /* Query new route. */ 1115 fl4 = &inet->cork.fl.u.ip4; 1116 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk), 1117 sk->sk_bound_dev_if, sk->sk_protocol, 1118 inet->inet_sport, inet->inet_dport, sk); 1119 if (IS_ERR(rt)) 1120 return PTR_ERR(rt); 1121 1122 sk_setup_caps(sk, &rt->dst); 1123 1124 new_saddr = fl4->saddr; 1125 1126 if (new_saddr == old_saddr) 1127 return 0; 1128 1129 if (sock_net(sk)->ipv4.sysctl_ip_dynaddr > 1) { 1130 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n", 1131 __func__, &old_saddr, &new_saddr); 1132 } 1133 1134 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr; 1135 1136 /* 1137 * XXX The only one ugly spot where we need to 1138 * XXX really change the sockets identity after 1139 * XXX it has entered the hashes. -DaveM 1140 * 1141 * Besides that, it does not check for connection 1142 * uniqueness. Wait for troubles. 1143 */ 1144 return __sk_prot_rehash(sk); 1145 } 1146 1147 int inet_sk_rebuild_header(struct sock *sk) 1148 { 1149 struct inet_sock *inet = inet_sk(sk); 1150 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0); 1151 __be32 daddr; 1152 struct ip_options_rcu *inet_opt; 1153 struct flowi4 *fl4; 1154 int err; 1155 1156 /* Route is OK, nothing to do. */ 1157 if (rt) 1158 return 0; 1159 1160 /* Reroute. */ 1161 rcu_read_lock(); 1162 inet_opt = rcu_dereference(inet->inet_opt); 1163 daddr = inet->inet_daddr; 1164 if (inet_opt && inet_opt->opt.srr) 1165 daddr = inet_opt->opt.faddr; 1166 rcu_read_unlock(); 1167 fl4 = &inet->cork.fl.u.ip4; 1168 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr, 1169 inet->inet_dport, inet->inet_sport, 1170 sk->sk_protocol, RT_CONN_FLAGS(sk), 1171 sk->sk_bound_dev_if); 1172 if (!IS_ERR(rt)) { 1173 err = 0; 1174 sk_setup_caps(sk, &rt->dst); 1175 } else { 1176 err = PTR_ERR(rt); 1177 1178 /* Routing failed... */ 1179 sk->sk_route_caps = 0; 1180 /* 1181 * Other protocols have to map its equivalent state to TCP_SYN_SENT. 1182 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme 1183 */ 1184 if (!sock_net(sk)->ipv4.sysctl_ip_dynaddr || 1185 sk->sk_state != TCP_SYN_SENT || 1186 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) || 1187 (err = inet_sk_reselect_saddr(sk)) != 0) 1188 sk->sk_err_soft = -err; 1189 } 1190 1191 return err; 1192 } 1193 EXPORT_SYMBOL(inet_sk_rebuild_header); 1194 1195 struct sk_buff *inet_gso_segment(struct sk_buff *skb, 1196 netdev_features_t features) 1197 { 1198 bool udpfrag = false, fixedid = false, encap; 1199 struct sk_buff *segs = ERR_PTR(-EINVAL); 1200 const struct net_offload *ops; 1201 unsigned int offset = 0; 1202 struct iphdr *iph; 1203 int proto, tot_len; 1204 int nhoff; 1205 int ihl; 1206 int id; 1207 1208 skb_reset_network_header(skb); 1209 nhoff = skb_network_header(skb) - skb_mac_header(skb); 1210 if (unlikely(!pskb_may_pull(skb, sizeof(*iph)))) 1211 goto out; 1212 1213 iph = ip_hdr(skb); 1214 ihl = iph->ihl * 4; 1215 if (ihl < sizeof(*iph)) 1216 goto out; 1217 1218 id = ntohs(iph->id); 1219 proto = iph->protocol; 1220 1221 /* Warning: after this point, iph might be no longer valid */ 1222 if (unlikely(!pskb_may_pull(skb, ihl))) 1223 goto out; 1224 __skb_pull(skb, ihl); 1225 1226 encap = SKB_GSO_CB(skb)->encap_level > 0; 1227 if (encap) 1228 features &= skb->dev->hw_enc_features; 1229 SKB_GSO_CB(skb)->encap_level += ihl; 1230 1231 skb_reset_transport_header(skb); 1232 1233 segs = ERR_PTR(-EPROTONOSUPPORT); 1234 1235 if (!skb->encapsulation || encap) { 1236 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP); 1237 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID); 1238 1239 /* fixed ID is invalid if DF bit is not set */ 1240 if (fixedid && !(iph->frag_off & htons(IP_DF))) 1241 goto out; 1242 } 1243 1244 ops = rcu_dereference(inet_offloads[proto]); 1245 if (likely(ops && ops->callbacks.gso_segment)) 1246 segs = ops->callbacks.gso_segment(skb, features); 1247 1248 if (IS_ERR_OR_NULL(segs)) 1249 goto out; 1250 1251 skb = segs; 1252 do { 1253 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff); 1254 if (udpfrag) { 1255 iph->frag_off = htons(offset >> 3); 1256 if (skb->next) 1257 iph->frag_off |= htons(IP_MF); 1258 offset += skb->len - nhoff - ihl; 1259 tot_len = skb->len - nhoff; 1260 } else if (skb_is_gso(skb)) { 1261 if (!fixedid) { 1262 iph->id = htons(id); 1263 id += skb_shinfo(skb)->gso_segs; 1264 } 1265 tot_len = skb_shinfo(skb)->gso_size + 1266 SKB_GSO_CB(skb)->data_offset + 1267 skb->head - (unsigned char *)iph; 1268 } else { 1269 if (!fixedid) 1270 iph->id = htons(id++); 1271 tot_len = skb->len - nhoff; 1272 } 1273 iph->tot_len = htons(tot_len); 1274 ip_send_check(iph); 1275 if (encap) 1276 skb_reset_inner_headers(skb); 1277 skb->network_header = (u8 *)iph - skb->head; 1278 } while ((skb = skb->next)); 1279 1280 out: 1281 return segs; 1282 } 1283 EXPORT_SYMBOL(inet_gso_segment); 1284 1285 struct sk_buff **inet_gro_receive(struct sk_buff **head, struct sk_buff *skb) 1286 { 1287 const struct net_offload *ops; 1288 struct sk_buff **pp = NULL; 1289 struct sk_buff *p; 1290 const struct iphdr *iph; 1291 unsigned int hlen; 1292 unsigned int off; 1293 unsigned int id; 1294 int flush = 1; 1295 int proto; 1296 1297 off = skb_gro_offset(skb); 1298 hlen = off + sizeof(*iph); 1299 iph = skb_gro_header_fast(skb, off); 1300 if (skb_gro_header_hard(skb, hlen)) { 1301 iph = skb_gro_header_slow(skb, hlen, off); 1302 if (unlikely(!iph)) 1303 goto out; 1304 } 1305 1306 proto = iph->protocol; 1307 1308 rcu_read_lock(); 1309 ops = rcu_dereference(inet_offloads[proto]); 1310 if (!ops || !ops->callbacks.gro_receive) 1311 goto out_unlock; 1312 1313 if (*(u8 *)iph != 0x45) 1314 goto out_unlock; 1315 1316 if (unlikely(ip_fast_csum((u8 *)iph, 5))) 1317 goto out_unlock; 1318 1319 id = ntohl(*(__be32 *)&iph->id); 1320 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF)); 1321 id >>= 16; 1322 1323 for (p = *head; p; p = p->next) { 1324 struct iphdr *iph2; 1325 u16 flush_id; 1326 1327 if (!NAPI_GRO_CB(p)->same_flow) 1328 continue; 1329 1330 iph2 = (struct iphdr *)(p->data + off); 1331 /* The above works because, with the exception of the top 1332 * (inner most) layer, we only aggregate pkts with the same 1333 * hdr length so all the hdrs we'll need to verify will start 1334 * at the same offset. 1335 */ 1336 if ((iph->protocol ^ iph2->protocol) | 1337 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) | 1338 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) { 1339 NAPI_GRO_CB(p)->same_flow = 0; 1340 continue; 1341 } 1342 1343 /* All fields must match except length and checksum. */ 1344 NAPI_GRO_CB(p)->flush |= 1345 (iph->ttl ^ iph2->ttl) | 1346 (iph->tos ^ iph2->tos) | 1347 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF)); 1348 1349 NAPI_GRO_CB(p)->flush |= flush; 1350 1351 /* We need to store of the IP ID check to be included later 1352 * when we can verify that this packet does in fact belong 1353 * to a given flow. 1354 */ 1355 flush_id = (u16)(id - ntohs(iph2->id)); 1356 1357 /* This bit of code makes it much easier for us to identify 1358 * the cases where we are doing atomic vs non-atomic IP ID 1359 * checks. Specifically an atomic check can return IP ID 1360 * values 0 - 0xFFFF, while a non-atomic check can only 1361 * return 0 or 0xFFFF. 1362 */ 1363 if (!NAPI_GRO_CB(p)->is_atomic || 1364 !(iph->frag_off & htons(IP_DF))) { 1365 flush_id ^= NAPI_GRO_CB(p)->count; 1366 flush_id = flush_id ? 0xFFFF : 0; 1367 } 1368 1369 /* If the previous IP ID value was based on an atomic 1370 * datagram we can overwrite the value and ignore it. 1371 */ 1372 if (NAPI_GRO_CB(skb)->is_atomic) 1373 NAPI_GRO_CB(p)->flush_id = flush_id; 1374 else 1375 NAPI_GRO_CB(p)->flush_id |= flush_id; 1376 } 1377 1378 NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF)); 1379 NAPI_GRO_CB(skb)->flush |= flush; 1380 skb_set_network_header(skb, off); 1381 /* The above will be needed by the transport layer if there is one 1382 * immediately following this IP hdr. 1383 */ 1384 1385 /* Note : No need to call skb_gro_postpull_rcsum() here, 1386 * as we already checked checksum over ipv4 header was 0 1387 */ 1388 skb_gro_pull(skb, sizeof(*iph)); 1389 skb_set_transport_header(skb, skb_gro_offset(skb)); 1390 1391 pp = ops->callbacks.gro_receive(head, skb); 1392 1393 out_unlock: 1394 rcu_read_unlock(); 1395 1396 out: 1397 NAPI_GRO_CB(skb)->flush |= flush; 1398 1399 return pp; 1400 } 1401 EXPORT_SYMBOL(inet_gro_receive); 1402 1403 static struct sk_buff **ipip_gro_receive(struct sk_buff **head, 1404 struct sk_buff *skb) 1405 { 1406 if (NAPI_GRO_CB(skb)->encap_mark) { 1407 NAPI_GRO_CB(skb)->flush = 1; 1408 return NULL; 1409 } 1410 1411 NAPI_GRO_CB(skb)->encap_mark = 1; 1412 1413 return inet_gro_receive(head, skb); 1414 } 1415 1416 #define SECONDS_PER_DAY 86400 1417 1418 /* inet_current_timestamp - Return IP network timestamp 1419 * 1420 * Return milliseconds since midnight in network byte order. 1421 */ 1422 __be32 inet_current_timestamp(void) 1423 { 1424 u32 secs; 1425 u32 msecs; 1426 struct timespec64 ts; 1427 1428 ktime_get_real_ts64(&ts); 1429 1430 /* Get secs since midnight. */ 1431 (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs); 1432 /* Convert to msecs. */ 1433 msecs = secs * MSEC_PER_SEC; 1434 /* Convert nsec to msec. */ 1435 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC; 1436 1437 /* Convert to network byte order. */ 1438 return htonl(msecs); 1439 } 1440 EXPORT_SYMBOL(inet_current_timestamp); 1441 1442 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len) 1443 { 1444 if (sk->sk_family == AF_INET) 1445 return ip_recv_error(sk, msg, len, addr_len); 1446 #if IS_ENABLED(CONFIG_IPV6) 1447 if (sk->sk_family == AF_INET6) 1448 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len); 1449 #endif 1450 return -EINVAL; 1451 } 1452 1453 int inet_gro_complete(struct sk_buff *skb, int nhoff) 1454 { 1455 __be16 newlen = htons(skb->len - nhoff); 1456 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff); 1457 const struct net_offload *ops; 1458 int proto = iph->protocol; 1459 int err = -ENOSYS; 1460 1461 if (skb->encapsulation) 1462 skb_set_inner_network_header(skb, nhoff); 1463 1464 csum_replace2(&iph->check, iph->tot_len, newlen); 1465 iph->tot_len = newlen; 1466 1467 rcu_read_lock(); 1468 ops = rcu_dereference(inet_offloads[proto]); 1469 if (WARN_ON(!ops || !ops->callbacks.gro_complete)) 1470 goto out_unlock; 1471 1472 /* Only need to add sizeof(*iph) to get to the next hdr below 1473 * because any hdr with option will have been flushed in 1474 * inet_gro_receive(). 1475 */ 1476 err = ops->callbacks.gro_complete(skb, nhoff + sizeof(*iph)); 1477 1478 out_unlock: 1479 rcu_read_unlock(); 1480 1481 return err; 1482 } 1483 EXPORT_SYMBOL(inet_gro_complete); 1484 1485 static int ipip_gro_complete(struct sk_buff *skb, int nhoff) 1486 { 1487 skb->encapsulation = 1; 1488 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4; 1489 return inet_gro_complete(skb, nhoff); 1490 } 1491 1492 int inet_ctl_sock_create(struct sock **sk, unsigned short family, 1493 unsigned short type, unsigned char protocol, 1494 struct net *net) 1495 { 1496 struct socket *sock; 1497 int rc = sock_create_kern(net, family, type, protocol, &sock); 1498 1499 if (rc == 0) { 1500 *sk = sock->sk; 1501 (*sk)->sk_allocation = GFP_ATOMIC; 1502 /* 1503 * Unhash it so that IP input processing does not even see it, 1504 * we do not wish this socket to see incoming packets. 1505 */ 1506 (*sk)->sk_prot->unhash(*sk); 1507 } 1508 return rc; 1509 } 1510 EXPORT_SYMBOL_GPL(inet_ctl_sock_create); 1511 1512 u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt) 1513 { 1514 return *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt); 1515 } 1516 EXPORT_SYMBOL_GPL(snmp_get_cpu_field); 1517 1518 unsigned long snmp_fold_field(void __percpu *mib, int offt) 1519 { 1520 unsigned long res = 0; 1521 int i; 1522 1523 for_each_possible_cpu(i) 1524 res += snmp_get_cpu_field(mib, i, offt); 1525 return res; 1526 } 1527 EXPORT_SYMBOL_GPL(snmp_fold_field); 1528 1529 #if BITS_PER_LONG==32 1530 1531 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt, 1532 size_t syncp_offset) 1533 { 1534 void *bhptr; 1535 struct u64_stats_sync *syncp; 1536 u64 v; 1537 unsigned int start; 1538 1539 bhptr = per_cpu_ptr(mib, cpu); 1540 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset); 1541 do { 1542 start = u64_stats_fetch_begin_irq(syncp); 1543 v = *(((u64 *)bhptr) + offt); 1544 } while (u64_stats_fetch_retry_irq(syncp, start)); 1545 1546 return v; 1547 } 1548 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64); 1549 1550 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset) 1551 { 1552 u64 res = 0; 1553 int cpu; 1554 1555 for_each_possible_cpu(cpu) { 1556 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset); 1557 } 1558 return res; 1559 } 1560 EXPORT_SYMBOL_GPL(snmp_fold_field64); 1561 #endif 1562 1563 #ifdef CONFIG_IP_MULTICAST 1564 static const struct net_protocol igmp_protocol = { 1565 .handler = igmp_rcv, 1566 .netns_ok = 1, 1567 }; 1568 #endif 1569 1570 static const struct net_protocol tcp_protocol = { 1571 .early_demux = tcp_v4_early_demux, 1572 .handler = tcp_v4_rcv, 1573 .err_handler = tcp_v4_err, 1574 .no_policy = 1, 1575 .netns_ok = 1, 1576 .icmp_strict_tag_validation = 1, 1577 }; 1578 1579 static const struct net_protocol udp_protocol = { 1580 .early_demux = udp_v4_early_demux, 1581 .handler = udp_rcv, 1582 .err_handler = udp_err, 1583 .no_policy = 1, 1584 .netns_ok = 1, 1585 }; 1586 1587 static const struct net_protocol icmp_protocol = { 1588 .handler = icmp_rcv, 1589 .err_handler = icmp_err, 1590 .no_policy = 1, 1591 .netns_ok = 1, 1592 }; 1593 1594 static __net_init int ipv4_mib_init_net(struct net *net) 1595 { 1596 int i; 1597 1598 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib); 1599 if (!net->mib.tcp_statistics) 1600 goto err_tcp_mib; 1601 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib); 1602 if (!net->mib.ip_statistics) 1603 goto err_ip_mib; 1604 1605 for_each_possible_cpu(i) { 1606 struct ipstats_mib *af_inet_stats; 1607 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i); 1608 u64_stats_init(&af_inet_stats->syncp); 1609 } 1610 1611 net->mib.net_statistics = alloc_percpu(struct linux_mib); 1612 if (!net->mib.net_statistics) 1613 goto err_net_mib; 1614 net->mib.udp_statistics = alloc_percpu(struct udp_mib); 1615 if (!net->mib.udp_statistics) 1616 goto err_udp_mib; 1617 net->mib.udplite_statistics = alloc_percpu(struct udp_mib); 1618 if (!net->mib.udplite_statistics) 1619 goto err_udplite_mib; 1620 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib); 1621 if (!net->mib.icmp_statistics) 1622 goto err_icmp_mib; 1623 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib), 1624 GFP_KERNEL); 1625 if (!net->mib.icmpmsg_statistics) 1626 goto err_icmpmsg_mib; 1627 1628 tcp_mib_init(net); 1629 return 0; 1630 1631 err_icmpmsg_mib: 1632 free_percpu(net->mib.icmp_statistics); 1633 err_icmp_mib: 1634 free_percpu(net->mib.udplite_statistics); 1635 err_udplite_mib: 1636 free_percpu(net->mib.udp_statistics); 1637 err_udp_mib: 1638 free_percpu(net->mib.net_statistics); 1639 err_net_mib: 1640 free_percpu(net->mib.ip_statistics); 1641 err_ip_mib: 1642 free_percpu(net->mib.tcp_statistics); 1643 err_tcp_mib: 1644 return -ENOMEM; 1645 } 1646 1647 static __net_exit void ipv4_mib_exit_net(struct net *net) 1648 { 1649 kfree(net->mib.icmpmsg_statistics); 1650 free_percpu(net->mib.icmp_statistics); 1651 free_percpu(net->mib.udplite_statistics); 1652 free_percpu(net->mib.udp_statistics); 1653 free_percpu(net->mib.net_statistics); 1654 free_percpu(net->mib.ip_statistics); 1655 free_percpu(net->mib.tcp_statistics); 1656 } 1657 1658 static __net_initdata struct pernet_operations ipv4_mib_ops = { 1659 .init = ipv4_mib_init_net, 1660 .exit = ipv4_mib_exit_net, 1661 }; 1662 1663 static int __init init_ipv4_mibs(void) 1664 { 1665 return register_pernet_subsys(&ipv4_mib_ops); 1666 } 1667 1668 static __net_init int inet_init_net(struct net *net) 1669 { 1670 /* 1671 * Set defaults for local port range 1672 */ 1673 seqlock_init(&net->ipv4.ip_local_ports.lock); 1674 net->ipv4.ip_local_ports.range[0] = 32768; 1675 net->ipv4.ip_local_ports.range[1] = 60999; 1676 1677 seqlock_init(&net->ipv4.ping_group_range.lock); 1678 /* 1679 * Sane defaults - nobody may create ping sockets. 1680 * Boot scripts should set this to distro-specific group. 1681 */ 1682 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1); 1683 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0); 1684 1685 /* Default values for sysctl-controlled parameters. 1686 * We set them here, in case sysctl is not compiled. 1687 */ 1688 net->ipv4.sysctl_ip_default_ttl = IPDEFTTL; 1689 net->ipv4.sysctl_ip_dynaddr = 0; 1690 net->ipv4.sysctl_ip_early_demux = 1; 1691 1692 return 0; 1693 } 1694 1695 static __net_exit void inet_exit_net(struct net *net) 1696 { 1697 } 1698 1699 static __net_initdata struct pernet_operations af_inet_ops = { 1700 .init = inet_init_net, 1701 .exit = inet_exit_net, 1702 }; 1703 1704 static int __init init_inet_pernet_ops(void) 1705 { 1706 return register_pernet_subsys(&af_inet_ops); 1707 } 1708 1709 static int ipv4_proc_init(void); 1710 1711 /* 1712 * IP protocol layer initialiser 1713 */ 1714 1715 static struct packet_offload ip_packet_offload __read_mostly = { 1716 .type = cpu_to_be16(ETH_P_IP), 1717 .callbacks = { 1718 .gso_segment = inet_gso_segment, 1719 .gro_receive = inet_gro_receive, 1720 .gro_complete = inet_gro_complete, 1721 }, 1722 }; 1723 1724 static const struct net_offload ipip_offload = { 1725 .callbacks = { 1726 .gso_segment = inet_gso_segment, 1727 .gro_receive = ipip_gro_receive, 1728 .gro_complete = ipip_gro_complete, 1729 }, 1730 }; 1731 1732 static int __init ipv4_offload_init(void) 1733 { 1734 /* 1735 * Add offloads 1736 */ 1737 if (udpv4_offload_init() < 0) 1738 pr_crit("%s: Cannot add UDP protocol offload\n", __func__); 1739 if (tcpv4_offload_init() < 0) 1740 pr_crit("%s: Cannot add TCP protocol offload\n", __func__); 1741 1742 dev_add_offload(&ip_packet_offload); 1743 inet_add_offload(&ipip_offload, IPPROTO_IPIP); 1744 return 0; 1745 } 1746 1747 fs_initcall(ipv4_offload_init); 1748 1749 static struct packet_type ip_packet_type __read_mostly = { 1750 .type = cpu_to_be16(ETH_P_IP), 1751 .func = ip_rcv, 1752 }; 1753 1754 static int __init inet_init(void) 1755 { 1756 struct inet_protosw *q; 1757 struct list_head *r; 1758 int rc = -EINVAL; 1759 1760 sock_skb_cb_check_size(sizeof(struct inet_skb_parm)); 1761 1762 rc = proto_register(&tcp_prot, 1); 1763 if (rc) 1764 goto out; 1765 1766 rc = proto_register(&udp_prot, 1); 1767 if (rc) 1768 goto out_unregister_tcp_proto; 1769 1770 rc = proto_register(&raw_prot, 1); 1771 if (rc) 1772 goto out_unregister_udp_proto; 1773 1774 rc = proto_register(&ping_prot, 1); 1775 if (rc) 1776 goto out_unregister_raw_proto; 1777 1778 /* 1779 * Tell SOCKET that we are alive... 1780 */ 1781 1782 (void)sock_register(&inet_family_ops); 1783 1784 #ifdef CONFIG_SYSCTL 1785 ip_static_sysctl_init(); 1786 #endif 1787 1788 /* 1789 * Add all the base protocols. 1790 */ 1791 1792 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0) 1793 pr_crit("%s: Cannot add ICMP protocol\n", __func__); 1794 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0) 1795 pr_crit("%s: Cannot add UDP protocol\n", __func__); 1796 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0) 1797 pr_crit("%s: Cannot add TCP protocol\n", __func__); 1798 #ifdef CONFIG_IP_MULTICAST 1799 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0) 1800 pr_crit("%s: Cannot add IGMP protocol\n", __func__); 1801 #endif 1802 1803 /* Register the socket-side information for inet_create. */ 1804 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r) 1805 INIT_LIST_HEAD(r); 1806 1807 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q) 1808 inet_register_protosw(q); 1809 1810 /* 1811 * Set the ARP module up 1812 */ 1813 1814 arp_init(); 1815 1816 /* 1817 * Set the IP module up 1818 */ 1819 1820 ip_init(); 1821 1822 tcp_v4_init(); 1823 1824 /* Setup TCP slab cache for open requests. */ 1825 tcp_init(); 1826 1827 /* Setup UDP memory threshold */ 1828 udp_init(); 1829 1830 /* Add UDP-Lite (RFC 3828) */ 1831 udplite4_register(); 1832 1833 ping_init(); 1834 1835 /* 1836 * Set the ICMP layer up 1837 */ 1838 1839 if (icmp_init() < 0) 1840 panic("Failed to create the ICMP control socket.\n"); 1841 1842 /* 1843 * Initialise the multicast router 1844 */ 1845 #if defined(CONFIG_IP_MROUTE) 1846 if (ip_mr_init()) 1847 pr_crit("%s: Cannot init ipv4 mroute\n", __func__); 1848 #endif 1849 1850 if (init_inet_pernet_ops()) 1851 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__); 1852 /* 1853 * Initialise per-cpu ipv4 mibs 1854 */ 1855 1856 if (init_ipv4_mibs()) 1857 pr_crit("%s: Cannot init ipv4 mibs\n", __func__); 1858 1859 ipv4_proc_init(); 1860 1861 ipfrag_init(); 1862 1863 dev_add_pack(&ip_packet_type); 1864 1865 ip_tunnel_core_init(); 1866 1867 rc = 0; 1868 out: 1869 return rc; 1870 out_unregister_raw_proto: 1871 proto_unregister(&raw_prot); 1872 out_unregister_udp_proto: 1873 proto_unregister(&udp_prot); 1874 out_unregister_tcp_proto: 1875 proto_unregister(&tcp_prot); 1876 goto out; 1877 } 1878 1879 fs_initcall(inet_init); 1880 1881 /* ------------------------------------------------------------------------ */ 1882 1883 #ifdef CONFIG_PROC_FS 1884 static int __init ipv4_proc_init(void) 1885 { 1886 int rc = 0; 1887 1888 if (raw_proc_init()) 1889 goto out_raw; 1890 if (tcp4_proc_init()) 1891 goto out_tcp; 1892 if (udp4_proc_init()) 1893 goto out_udp; 1894 if (ping_proc_init()) 1895 goto out_ping; 1896 if (ip_misc_proc_init()) 1897 goto out_misc; 1898 out: 1899 return rc; 1900 out_misc: 1901 ping_proc_exit(); 1902 out_ping: 1903 udp4_proc_exit(); 1904 out_udp: 1905 tcp4_proc_exit(); 1906 out_tcp: 1907 raw_proc_exit(); 1908 out_raw: 1909 rc = -ENOMEM; 1910 goto out; 1911 } 1912 1913 #else /* CONFIG_PROC_FS */ 1914 static int __init ipv4_proc_init(void) 1915 { 1916 return 0; 1917 } 1918 #endif /* CONFIG_PROC_FS */ 1919 1920 MODULE_ALIAS_NETPROTO(PF_INET); 1921 1922