af_inet.c revision 2c804d0f8fc7799981d9fdd8c88653541b28c1a7
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/xfrm.h> 116#include <net/net_namespace.h> 117#include <net/secure_seq.h> 118#ifdef CONFIG_IP_MROUTE 119#include <linux/mroute.h> 120#endif 121 122 123/* The inetsw table contains everything that inet_create needs to 124 * build a new socket. 125 */ 126static struct list_head inetsw[SOCK_MAX]; 127static DEFINE_SPINLOCK(inetsw_lock); 128 129/* New destruction routine */ 130 131void inet_sock_destruct(struct sock *sk) 132{ 133 struct inet_sock *inet = inet_sk(sk); 134 135 __skb_queue_purge(&sk->sk_receive_queue); 136 __skb_queue_purge(&sk->sk_error_queue); 137 138 sk_mem_reclaim(sk); 139 140 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) { 141 pr_err("Attempt to release TCP socket in state %d %p\n", 142 sk->sk_state, sk); 143 return; 144 } 145 if (!sock_flag(sk, SOCK_DEAD)) { 146 pr_err("Attempt to release alive inet socket %p\n", sk); 147 return; 148 } 149 150 WARN_ON(atomic_read(&sk->sk_rmem_alloc)); 151 WARN_ON(atomic_read(&sk->sk_wmem_alloc)); 152 WARN_ON(sk->sk_wmem_queued); 153 WARN_ON(sk->sk_forward_alloc); 154 155 kfree(rcu_dereference_protected(inet->inet_opt, 1)); 156 dst_release(rcu_dereference_check(sk->sk_dst_cache, 1)); 157 dst_release(sk->sk_rx_dst); 158 sk_refcnt_debug_dec(sk); 159} 160EXPORT_SYMBOL(inet_sock_destruct); 161 162/* 163 * The routines beyond this point handle the behaviour of an AF_INET 164 * socket object. Mostly it punts to the subprotocols of IP to do 165 * the work. 166 */ 167 168/* 169 * Automatically bind an unbound socket. 170 */ 171 172static int inet_autobind(struct sock *sk) 173{ 174 struct inet_sock *inet; 175 /* We may need to bind the socket. */ 176 lock_sock(sk); 177 inet = inet_sk(sk); 178 if (!inet->inet_num) { 179 if (sk->sk_prot->get_port(sk, 0)) { 180 release_sock(sk); 181 return -EAGAIN; 182 } 183 inet->inet_sport = htons(inet->inet_num); 184 } 185 release_sock(sk); 186 return 0; 187} 188 189/* 190 * Move a socket into listening state. 191 */ 192int inet_listen(struct socket *sock, int backlog) 193{ 194 struct sock *sk = sock->sk; 195 unsigned char old_state; 196 int err; 197 198 lock_sock(sk); 199 200 err = -EINVAL; 201 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM) 202 goto out; 203 204 old_state = sk->sk_state; 205 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN))) 206 goto out; 207 208 /* Really, if the socket is already in listen state 209 * we can only allow the backlog to be adjusted. 210 */ 211 if (old_state != TCP_LISTEN) { 212 /* Check special setups for testing purpose to enable TFO w/o 213 * requiring TCP_FASTOPEN sockopt. 214 * Note that only TCP sockets (SOCK_STREAM) will reach here. 215 * Also fastopenq may already been allocated because this 216 * socket was in TCP_LISTEN state previously but was 217 * shutdown() (rather than close()). 218 */ 219 if ((sysctl_tcp_fastopen & TFO_SERVER_ENABLE) != 0 && 220 inet_csk(sk)->icsk_accept_queue.fastopenq == NULL) { 221 if ((sysctl_tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) != 0) 222 err = fastopen_init_queue(sk, backlog); 223 else if ((sysctl_tcp_fastopen & 224 TFO_SERVER_WO_SOCKOPT2) != 0) 225 err = fastopen_init_queue(sk, 226 ((uint)sysctl_tcp_fastopen) >> 16); 227 else 228 err = 0; 229 if (err) 230 goto out; 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 239out: 240 release_sock(sk); 241 return err; 242} 243EXPORT_SYMBOL(inet_listen); 244 245/* 246 * Create an inet socket. 247 */ 248 249static 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 sock->state = SS_UNCONNECTED; 261 262 /* Look for the requested type/protocol pair. */ 263lookup_protocol: 264 err = -ESOCKTNOSUPPORT; 265 rcu_read_lock(); 266 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) { 267 268 err = 0; 269 /* Check the non-wild match. */ 270 if (protocol == answer->protocol) { 271 if (protocol != IPPROTO_IP) 272 break; 273 } else { 274 /* Check for the two wild cases. */ 275 if (IPPROTO_IP == protocol) { 276 protocol = answer->protocol; 277 break; 278 } 279 if (IPPROTO_IP == answer->protocol) 280 break; 281 } 282 err = -EPROTONOSUPPORT; 283 } 284 285 if (unlikely(err)) { 286 if (try_loading_module < 2) { 287 rcu_read_unlock(); 288 /* 289 * Be more specific, e.g. net-pf-2-proto-132-type-1 290 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM) 291 */ 292 if (++try_loading_module == 1) 293 request_module("net-pf-%d-proto-%d-type-%d", 294 PF_INET, protocol, sock->type); 295 /* 296 * Fall back to generic, e.g. net-pf-2-proto-132 297 * (net-pf-PF_INET-proto-IPPROTO_SCTP) 298 */ 299 else 300 request_module("net-pf-%d-proto-%d", 301 PF_INET, protocol); 302 goto lookup_protocol; 303 } else 304 goto out_rcu_unlock; 305 } 306 307 err = -EPERM; 308 if (sock->type == SOCK_RAW && !kern && 309 !ns_capable(net->user_ns, CAP_NET_RAW)) 310 goto out_rcu_unlock; 311 312 sock->ops = answer->ops; 313 answer_prot = answer->prot; 314 answer_flags = answer->flags; 315 rcu_read_unlock(); 316 317 WARN_ON(answer_prot->slab == NULL); 318 319 err = -ENOBUFS; 320 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot); 321 if (sk == NULL) 322 goto out; 323 324 err = 0; 325 if (INET_PROTOSW_REUSE & answer_flags) 326 sk->sk_reuse = SK_CAN_REUSE; 327 328 inet = inet_sk(sk); 329 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0; 330 331 inet->nodefrag = 0; 332 333 if (SOCK_RAW == sock->type) { 334 inet->inet_num = protocol; 335 if (IPPROTO_RAW == protocol) 336 inet->hdrincl = 1; 337 } 338 339 if (net->ipv4.sysctl_ip_no_pmtu_disc) 340 inet->pmtudisc = IP_PMTUDISC_DONT; 341 else 342 inet->pmtudisc = IP_PMTUDISC_WANT; 343 344 inet->inet_id = 0; 345 346 sock_init_data(sock, sk); 347 348 sk->sk_destruct = inet_sock_destruct; 349 sk->sk_protocol = protocol; 350 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv; 351 352 inet->uc_ttl = -1; 353 inet->mc_loop = 1; 354 inet->mc_ttl = 1; 355 inet->mc_all = 1; 356 inet->mc_index = 0; 357 inet->mc_list = NULL; 358 inet->rcv_tos = 0; 359 360 sk_refcnt_debug_inc(sk); 361 362 if (inet->inet_num) { 363 /* It assumes that any protocol which allows 364 * the user to assign a number at socket 365 * creation time automatically 366 * shares. 367 */ 368 inet->inet_sport = htons(inet->inet_num); 369 /* Add to protocol hash chains. */ 370 sk->sk_prot->hash(sk); 371 } 372 373 if (sk->sk_prot->init) { 374 err = sk->sk_prot->init(sk); 375 if (err) 376 sk_common_release(sk); 377 } 378out: 379 return err; 380out_rcu_unlock: 381 rcu_read_unlock(); 382 goto out; 383} 384 385 386/* 387 * The peer socket should always be NULL (or else). When we call this 388 * function we are destroying the object and from then on nobody 389 * should refer to it. 390 */ 391int inet_release(struct socket *sock) 392{ 393 struct sock *sk = sock->sk; 394 395 if (sk) { 396 long timeout; 397 398 sock_rps_reset_flow(sk); 399 400 /* Applications forget to leave groups before exiting */ 401 ip_mc_drop_socket(sk); 402 403 /* If linger is set, we don't return until the close 404 * is complete. Otherwise we return immediately. The 405 * actually closing is done the same either way. 406 * 407 * If the close is due to the process exiting, we never 408 * linger.. 409 */ 410 timeout = 0; 411 if (sock_flag(sk, SOCK_LINGER) && 412 !(current->flags & PF_EXITING)) 413 timeout = sk->sk_lingertime; 414 sock->sk = NULL; 415 sk->sk_prot->close(sk, timeout); 416 } 417 return 0; 418} 419EXPORT_SYMBOL(inet_release); 420 421int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) 422{ 423 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr; 424 struct sock *sk = sock->sk; 425 struct inet_sock *inet = inet_sk(sk); 426 struct net *net = sock_net(sk); 427 unsigned short snum; 428 int chk_addr_ret; 429 int err; 430 431 /* If the socket has its own bind function then use it. (RAW) */ 432 if (sk->sk_prot->bind) { 433 err = sk->sk_prot->bind(sk, uaddr, addr_len); 434 goto out; 435 } 436 err = -EINVAL; 437 if (addr_len < sizeof(struct sockaddr_in)) 438 goto out; 439 440 if (addr->sin_family != AF_INET) { 441 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET) 442 * only if s_addr is INADDR_ANY. 443 */ 444 err = -EAFNOSUPPORT; 445 if (addr->sin_family != AF_UNSPEC || 446 addr->sin_addr.s_addr != htonl(INADDR_ANY)) 447 goto out; 448 } 449 450 chk_addr_ret = inet_addr_type(net, addr->sin_addr.s_addr); 451 452 /* Not specified by any standard per-se, however it breaks too 453 * many applications when removed. It is unfortunate since 454 * allowing applications to make a non-local bind solves 455 * several problems with systems using dynamic addressing. 456 * (ie. your servers still start up even if your ISDN link 457 * is temporarily down) 458 */ 459 err = -EADDRNOTAVAIL; 460 if (!net->ipv4.sysctl_ip_nonlocal_bind && 461 !(inet->freebind || inet->transparent) && 462 addr->sin_addr.s_addr != htonl(INADDR_ANY) && 463 chk_addr_ret != RTN_LOCAL && 464 chk_addr_ret != RTN_MULTICAST && 465 chk_addr_ret != RTN_BROADCAST) 466 goto out; 467 468 snum = ntohs(addr->sin_port); 469 err = -EACCES; 470 if (snum && snum < PROT_SOCK && 471 !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE)) 472 goto out; 473 474 /* We keep a pair of addresses. rcv_saddr is the one 475 * used by hash lookups, and saddr is used for transmit. 476 * 477 * In the BSD API these are the same except where it 478 * would be illegal to use them (multicast/broadcast) in 479 * which case the sending device address is used. 480 */ 481 lock_sock(sk); 482 483 /* Check these errors (active socket, double bind). */ 484 err = -EINVAL; 485 if (sk->sk_state != TCP_CLOSE || inet->inet_num) 486 goto out_release_sock; 487 488 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr; 489 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST) 490 inet->inet_saddr = 0; /* Use device */ 491 492 /* Make sure we are allowed to bind here. */ 493 if (sk->sk_prot->get_port(sk, snum)) { 494 inet->inet_saddr = inet->inet_rcv_saddr = 0; 495 err = -EADDRINUSE; 496 goto out_release_sock; 497 } 498 499 if (inet->inet_rcv_saddr) 500 sk->sk_userlocks |= SOCK_BINDADDR_LOCK; 501 if (snum) 502 sk->sk_userlocks |= SOCK_BINDPORT_LOCK; 503 inet->inet_sport = htons(inet->inet_num); 504 inet->inet_daddr = 0; 505 inet->inet_dport = 0; 506 sk_dst_reset(sk); 507 err = 0; 508out_release_sock: 509 release_sock(sk); 510out: 511 return err; 512} 513EXPORT_SYMBOL(inet_bind); 514 515int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr, 516 int addr_len, int flags) 517{ 518 struct sock *sk = sock->sk; 519 520 if (addr_len < sizeof(uaddr->sa_family)) 521 return -EINVAL; 522 if (uaddr->sa_family == AF_UNSPEC) 523 return sk->sk_prot->disconnect(sk, flags); 524 525 if (!inet_sk(sk)->inet_num && inet_autobind(sk)) 526 return -EAGAIN; 527 return sk->sk_prot->connect(sk, uaddr, addr_len); 528} 529EXPORT_SYMBOL(inet_dgram_connect); 530 531static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias) 532{ 533 DEFINE_WAIT(wait); 534 535 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 536 sk->sk_write_pending += writebias; 537 538 /* Basic assumption: if someone sets sk->sk_err, he _must_ 539 * change state of the socket from TCP_SYN_*. 540 * Connect() does not allow to get error notifications 541 * without closing the socket. 542 */ 543 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 544 release_sock(sk); 545 timeo = schedule_timeout(timeo); 546 lock_sock(sk); 547 if (signal_pending(current) || !timeo) 548 break; 549 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 550 } 551 finish_wait(sk_sleep(sk), &wait); 552 sk->sk_write_pending -= writebias; 553 return timeo; 554} 555 556/* 557 * Connect to a remote host. There is regrettably still a little 558 * TCP 'magic' in here. 559 */ 560int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr, 561 int addr_len, int flags) 562{ 563 struct sock *sk = sock->sk; 564 int err; 565 long timeo; 566 567 if (addr_len < sizeof(uaddr->sa_family)) 568 return -EINVAL; 569 570 if (uaddr->sa_family == AF_UNSPEC) { 571 err = sk->sk_prot->disconnect(sk, flags); 572 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED; 573 goto out; 574 } 575 576 switch (sock->state) { 577 default: 578 err = -EINVAL; 579 goto out; 580 case SS_CONNECTED: 581 err = -EISCONN; 582 goto out; 583 case SS_CONNECTING: 584 err = -EALREADY; 585 /* Fall out of switch with err, set for this state */ 586 break; 587 case SS_UNCONNECTED: 588 err = -EISCONN; 589 if (sk->sk_state != TCP_CLOSE) 590 goto out; 591 592 err = sk->sk_prot->connect(sk, uaddr, addr_len); 593 if (err < 0) 594 goto out; 595 596 sock->state = SS_CONNECTING; 597 598 /* Just entered SS_CONNECTING state; the only 599 * difference is that return value in non-blocking 600 * case is EINPROGRESS, rather than EALREADY. 601 */ 602 err = -EINPROGRESS; 603 break; 604 } 605 606 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK); 607 608 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 609 int writebias = (sk->sk_protocol == IPPROTO_TCP) && 610 tcp_sk(sk)->fastopen_req && 611 tcp_sk(sk)->fastopen_req->data ? 1 : 0; 612 613 /* Error code is set above */ 614 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias)) 615 goto out; 616 617 err = sock_intr_errno(timeo); 618 if (signal_pending(current)) 619 goto out; 620 } 621 622 /* Connection was closed by RST, timeout, ICMP error 623 * or another process disconnected us. 624 */ 625 if (sk->sk_state == TCP_CLOSE) 626 goto sock_error; 627 628 /* sk->sk_err may be not zero now, if RECVERR was ordered by user 629 * and error was received after socket entered established state. 630 * Hence, it is handled normally after connect() return successfully. 631 */ 632 633 sock->state = SS_CONNECTED; 634 err = 0; 635out: 636 return err; 637 638sock_error: 639 err = sock_error(sk) ? : -ECONNABORTED; 640 sock->state = SS_UNCONNECTED; 641 if (sk->sk_prot->disconnect(sk, flags)) 642 sock->state = SS_DISCONNECTING; 643 goto out; 644} 645EXPORT_SYMBOL(__inet_stream_connect); 646 647int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr, 648 int addr_len, int flags) 649{ 650 int err; 651 652 lock_sock(sock->sk); 653 err = __inet_stream_connect(sock, uaddr, addr_len, flags); 654 release_sock(sock->sk); 655 return err; 656} 657EXPORT_SYMBOL(inet_stream_connect); 658 659/* 660 * Accept a pending connection. The TCP layer now gives BSD semantics. 661 */ 662 663int inet_accept(struct socket *sock, struct socket *newsock, int flags) 664{ 665 struct sock *sk1 = sock->sk; 666 int err = -EINVAL; 667 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err); 668 669 if (!sk2) 670 goto do_err; 671 672 lock_sock(sk2); 673 674 sock_rps_record_flow(sk2); 675 WARN_ON(!((1 << sk2->sk_state) & 676 (TCPF_ESTABLISHED | TCPF_SYN_RECV | 677 TCPF_CLOSE_WAIT | TCPF_CLOSE))); 678 679 sock_graft(sk2, newsock); 680 681 newsock->state = SS_CONNECTED; 682 err = 0; 683 release_sock(sk2); 684do_err: 685 return err; 686} 687EXPORT_SYMBOL(inet_accept); 688 689 690/* 691 * This does both peername and sockname. 692 */ 693int inet_getname(struct socket *sock, struct sockaddr *uaddr, 694 int *uaddr_len, int peer) 695{ 696 struct sock *sk = sock->sk; 697 struct inet_sock *inet = inet_sk(sk); 698 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr); 699 700 sin->sin_family = AF_INET; 701 if (peer) { 702 if (!inet->inet_dport || 703 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) && 704 peer == 1)) 705 return -ENOTCONN; 706 sin->sin_port = inet->inet_dport; 707 sin->sin_addr.s_addr = inet->inet_daddr; 708 } else { 709 __be32 addr = inet->inet_rcv_saddr; 710 if (!addr) 711 addr = inet->inet_saddr; 712 sin->sin_port = inet->inet_sport; 713 sin->sin_addr.s_addr = addr; 714 } 715 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 716 *uaddr_len = sizeof(*sin); 717 return 0; 718} 719EXPORT_SYMBOL(inet_getname); 720 721int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg, 722 size_t size) 723{ 724 struct sock *sk = sock->sk; 725 726 sock_rps_record_flow(sk); 727 728 /* We may need to bind the socket. */ 729 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind && 730 inet_autobind(sk)) 731 return -EAGAIN; 732 733 return sk->sk_prot->sendmsg(iocb, sk, msg, size); 734} 735EXPORT_SYMBOL(inet_sendmsg); 736 737ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, 738 size_t size, int flags) 739{ 740 struct sock *sk = sock->sk; 741 742 sock_rps_record_flow(sk); 743 744 /* We may need to bind the socket. */ 745 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind && 746 inet_autobind(sk)) 747 return -EAGAIN; 748 749 if (sk->sk_prot->sendpage) 750 return sk->sk_prot->sendpage(sk, page, offset, size, flags); 751 return sock_no_sendpage(sock, page, offset, size, flags); 752} 753EXPORT_SYMBOL(inet_sendpage); 754 755int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg, 756 size_t size, int flags) 757{ 758 struct sock *sk = sock->sk; 759 int addr_len = 0; 760 int err; 761 762 sock_rps_record_flow(sk); 763 764 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT, 765 flags & ~MSG_DONTWAIT, &addr_len); 766 if (err >= 0) 767 msg->msg_namelen = addr_len; 768 return err; 769} 770EXPORT_SYMBOL(inet_recvmsg); 771 772int inet_shutdown(struct socket *sock, int how) 773{ 774 struct sock *sk = sock->sk; 775 int err = 0; 776 777 /* This should really check to make sure 778 * the socket is a TCP socket. (WHY AC...) 779 */ 780 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and 781 1->2 bit 2 snds. 782 2->3 */ 783 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */ 784 return -EINVAL; 785 786 lock_sock(sk); 787 if (sock->state == SS_CONNECTING) { 788 if ((1 << sk->sk_state) & 789 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE)) 790 sock->state = SS_DISCONNECTING; 791 else 792 sock->state = SS_CONNECTED; 793 } 794 795 switch (sk->sk_state) { 796 case TCP_CLOSE: 797 err = -ENOTCONN; 798 /* Hack to wake up other listeners, who can poll for 799 POLLHUP, even on eg. unconnected UDP sockets -- RR */ 800 default: 801 sk->sk_shutdown |= how; 802 if (sk->sk_prot->shutdown) 803 sk->sk_prot->shutdown(sk, how); 804 break; 805 806 /* Remaining two branches are temporary solution for missing 807 * close() in multithreaded environment. It is _not_ a good idea, 808 * but we have no choice until close() is repaired at VFS level. 809 */ 810 case TCP_LISTEN: 811 if (!(how & RCV_SHUTDOWN)) 812 break; 813 /* Fall through */ 814 case TCP_SYN_SENT: 815 err = sk->sk_prot->disconnect(sk, O_NONBLOCK); 816 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED; 817 break; 818 } 819 820 /* Wake up anyone sleeping in poll. */ 821 sk->sk_state_change(sk); 822 release_sock(sk); 823 return err; 824} 825EXPORT_SYMBOL(inet_shutdown); 826 827/* 828 * ioctl() calls you can issue on an INET socket. Most of these are 829 * device configuration and stuff and very rarely used. Some ioctls 830 * pass on to the socket itself. 831 * 832 * NOTE: I like the idea of a module for the config stuff. ie ifconfig 833 * loads the devconfigure module does its configuring and unloads it. 834 * There's a good 20K of config code hanging around the kernel. 835 */ 836 837int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 838{ 839 struct sock *sk = sock->sk; 840 int err = 0; 841 struct net *net = sock_net(sk); 842 843 switch (cmd) { 844 case SIOCGSTAMP: 845 err = sock_get_timestamp(sk, (struct timeval __user *)arg); 846 break; 847 case SIOCGSTAMPNS: 848 err = sock_get_timestampns(sk, (struct timespec __user *)arg); 849 break; 850 case SIOCADDRT: 851 case SIOCDELRT: 852 case SIOCRTMSG: 853 err = ip_rt_ioctl(net, cmd, (void __user *)arg); 854 break; 855 case SIOCDARP: 856 case SIOCGARP: 857 case SIOCSARP: 858 err = arp_ioctl(net, cmd, (void __user *)arg); 859 break; 860 case SIOCGIFADDR: 861 case SIOCSIFADDR: 862 case SIOCGIFBRDADDR: 863 case SIOCSIFBRDADDR: 864 case SIOCGIFNETMASK: 865 case SIOCSIFNETMASK: 866 case SIOCGIFDSTADDR: 867 case SIOCSIFDSTADDR: 868 case SIOCSIFPFLAGS: 869 case SIOCGIFPFLAGS: 870 case SIOCSIFFLAGS: 871 err = devinet_ioctl(net, cmd, (void __user *)arg); 872 break; 873 default: 874 if (sk->sk_prot->ioctl) 875 err = sk->sk_prot->ioctl(sk, cmd, arg); 876 else 877 err = -ENOIOCTLCMD; 878 break; 879 } 880 return err; 881} 882EXPORT_SYMBOL(inet_ioctl); 883 884#ifdef CONFIG_COMPAT 885static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 886{ 887 struct sock *sk = sock->sk; 888 int err = -ENOIOCTLCMD; 889 890 if (sk->sk_prot->compat_ioctl) 891 err = sk->sk_prot->compat_ioctl(sk, cmd, arg); 892 893 return err; 894} 895#endif 896 897const struct proto_ops inet_stream_ops = { 898 .family = PF_INET, 899 .owner = THIS_MODULE, 900 .release = inet_release, 901 .bind = inet_bind, 902 .connect = inet_stream_connect, 903 .socketpair = sock_no_socketpair, 904 .accept = inet_accept, 905 .getname = inet_getname, 906 .poll = tcp_poll, 907 .ioctl = inet_ioctl, 908 .listen = inet_listen, 909 .shutdown = inet_shutdown, 910 .setsockopt = sock_common_setsockopt, 911 .getsockopt = sock_common_getsockopt, 912 .sendmsg = inet_sendmsg, 913 .recvmsg = inet_recvmsg, 914 .mmap = sock_no_mmap, 915 .sendpage = inet_sendpage, 916 .splice_read = tcp_splice_read, 917#ifdef CONFIG_COMPAT 918 .compat_setsockopt = compat_sock_common_setsockopt, 919 .compat_getsockopt = compat_sock_common_getsockopt, 920 .compat_ioctl = inet_compat_ioctl, 921#endif 922}; 923EXPORT_SYMBOL(inet_stream_ops); 924 925const struct proto_ops inet_dgram_ops = { 926 .family = PF_INET, 927 .owner = THIS_MODULE, 928 .release = inet_release, 929 .bind = inet_bind, 930 .connect = inet_dgram_connect, 931 .socketpair = sock_no_socketpair, 932 .accept = sock_no_accept, 933 .getname = inet_getname, 934 .poll = udp_poll, 935 .ioctl = inet_ioctl, 936 .listen = sock_no_listen, 937 .shutdown = inet_shutdown, 938 .setsockopt = sock_common_setsockopt, 939 .getsockopt = sock_common_getsockopt, 940 .sendmsg = inet_sendmsg, 941 .recvmsg = inet_recvmsg, 942 .mmap = sock_no_mmap, 943 .sendpage = inet_sendpage, 944#ifdef CONFIG_COMPAT 945 .compat_setsockopt = compat_sock_common_setsockopt, 946 .compat_getsockopt = compat_sock_common_getsockopt, 947 .compat_ioctl = inet_compat_ioctl, 948#endif 949}; 950EXPORT_SYMBOL(inet_dgram_ops); 951 952/* 953 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without 954 * udp_poll 955 */ 956static const struct proto_ops inet_sockraw_ops = { 957 .family = PF_INET, 958 .owner = THIS_MODULE, 959 .release = inet_release, 960 .bind = inet_bind, 961 .connect = inet_dgram_connect, 962 .socketpair = sock_no_socketpair, 963 .accept = sock_no_accept, 964 .getname = inet_getname, 965 .poll = datagram_poll, 966 .ioctl = inet_ioctl, 967 .listen = sock_no_listen, 968 .shutdown = inet_shutdown, 969 .setsockopt = sock_common_setsockopt, 970 .getsockopt = sock_common_getsockopt, 971 .sendmsg = inet_sendmsg, 972 .recvmsg = inet_recvmsg, 973 .mmap = sock_no_mmap, 974 .sendpage = inet_sendpage, 975#ifdef CONFIG_COMPAT 976 .compat_setsockopt = compat_sock_common_setsockopt, 977 .compat_getsockopt = compat_sock_common_getsockopt, 978 .compat_ioctl = inet_compat_ioctl, 979#endif 980}; 981 982static const struct net_proto_family inet_family_ops = { 983 .family = PF_INET, 984 .create = inet_create, 985 .owner = THIS_MODULE, 986}; 987 988/* Upon startup we insert all the elements in inetsw_array[] into 989 * the linked list inetsw. 990 */ 991static struct inet_protosw inetsw_array[] = 992{ 993 { 994 .type = SOCK_STREAM, 995 .protocol = IPPROTO_TCP, 996 .prot = &tcp_prot, 997 .ops = &inet_stream_ops, 998 .flags = INET_PROTOSW_PERMANENT | 999 INET_PROTOSW_ICSK, 1000 }, 1001 1002 { 1003 .type = SOCK_DGRAM, 1004 .protocol = IPPROTO_UDP, 1005 .prot = &udp_prot, 1006 .ops = &inet_dgram_ops, 1007 .flags = INET_PROTOSW_PERMANENT, 1008 }, 1009 1010 { 1011 .type = SOCK_DGRAM, 1012 .protocol = IPPROTO_ICMP, 1013 .prot = &ping_prot, 1014 .ops = &inet_dgram_ops, 1015 .flags = INET_PROTOSW_REUSE, 1016 }, 1017 1018 { 1019 .type = SOCK_RAW, 1020 .protocol = IPPROTO_IP, /* wild card */ 1021 .prot = &raw_prot, 1022 .ops = &inet_sockraw_ops, 1023 .flags = INET_PROTOSW_REUSE, 1024 } 1025}; 1026 1027#define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array) 1028 1029void inet_register_protosw(struct inet_protosw *p) 1030{ 1031 struct list_head *lh; 1032 struct inet_protosw *answer; 1033 int protocol = p->protocol; 1034 struct list_head *last_perm; 1035 1036 spin_lock_bh(&inetsw_lock); 1037 1038 if (p->type >= SOCK_MAX) 1039 goto out_illegal; 1040 1041 /* If we are trying to override a permanent protocol, bail. */ 1042 answer = NULL; 1043 last_perm = &inetsw[p->type]; 1044 list_for_each(lh, &inetsw[p->type]) { 1045 answer = list_entry(lh, struct inet_protosw, list); 1046 1047 /* Check only the non-wild match. */ 1048 if (INET_PROTOSW_PERMANENT & answer->flags) { 1049 if (protocol == answer->protocol) 1050 break; 1051 last_perm = lh; 1052 } 1053 1054 answer = NULL; 1055 } 1056 if (answer) 1057 goto out_permanent; 1058 1059 /* Add the new entry after the last permanent entry if any, so that 1060 * the new entry does not override a permanent entry when matched with 1061 * a wild-card protocol. But it is allowed to override any existing 1062 * non-permanent entry. This means that when we remove this entry, the 1063 * system automatically returns to the old behavior. 1064 */ 1065 list_add_rcu(&p->list, last_perm); 1066out: 1067 spin_unlock_bh(&inetsw_lock); 1068 1069 return; 1070 1071out_permanent: 1072 pr_err("Attempt to override permanent protocol %d\n", protocol); 1073 goto out; 1074 1075out_illegal: 1076 pr_err("Ignoring attempt to register invalid socket type %d\n", 1077 p->type); 1078 goto out; 1079} 1080EXPORT_SYMBOL(inet_register_protosw); 1081 1082void inet_unregister_protosw(struct inet_protosw *p) 1083{ 1084 if (INET_PROTOSW_PERMANENT & p->flags) { 1085 pr_err("Attempt to unregister permanent protocol %d\n", 1086 p->protocol); 1087 } else { 1088 spin_lock_bh(&inetsw_lock); 1089 list_del_rcu(&p->list); 1090 spin_unlock_bh(&inetsw_lock); 1091 1092 synchronize_net(); 1093 } 1094} 1095EXPORT_SYMBOL(inet_unregister_protosw); 1096 1097/* 1098 * Shall we try to damage output packets if routing dev changes? 1099 */ 1100 1101int sysctl_ip_dynaddr __read_mostly; 1102 1103static int inet_sk_reselect_saddr(struct sock *sk) 1104{ 1105 struct inet_sock *inet = inet_sk(sk); 1106 __be32 old_saddr = inet->inet_saddr; 1107 __be32 daddr = inet->inet_daddr; 1108 struct flowi4 *fl4; 1109 struct rtable *rt; 1110 __be32 new_saddr; 1111 struct ip_options_rcu *inet_opt; 1112 1113 inet_opt = rcu_dereference_protected(inet->inet_opt, 1114 sock_owned_by_user(sk)); 1115 if (inet_opt && inet_opt->opt.srr) 1116 daddr = inet_opt->opt.faddr; 1117 1118 /* Query new route. */ 1119 fl4 = &inet->cork.fl.u.ip4; 1120 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk), 1121 sk->sk_bound_dev_if, sk->sk_protocol, 1122 inet->inet_sport, inet->inet_dport, sk); 1123 if (IS_ERR(rt)) 1124 return PTR_ERR(rt); 1125 1126 sk_setup_caps(sk, &rt->dst); 1127 1128 new_saddr = fl4->saddr; 1129 1130 if (new_saddr == old_saddr) 1131 return 0; 1132 1133 if (sysctl_ip_dynaddr > 1) { 1134 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n", 1135 __func__, &old_saddr, &new_saddr); 1136 } 1137 1138 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr; 1139 1140 /* 1141 * XXX The only one ugly spot where we need to 1142 * XXX really change the sockets identity after 1143 * XXX it has entered the hashes. -DaveM 1144 * 1145 * Besides that, it does not check for connection 1146 * uniqueness. Wait for troubles. 1147 */ 1148 __sk_prot_rehash(sk); 1149 return 0; 1150} 1151 1152int inet_sk_rebuild_header(struct sock *sk) 1153{ 1154 struct inet_sock *inet = inet_sk(sk); 1155 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0); 1156 __be32 daddr; 1157 struct ip_options_rcu *inet_opt; 1158 struct flowi4 *fl4; 1159 int err; 1160 1161 /* Route is OK, nothing to do. */ 1162 if (rt) 1163 return 0; 1164 1165 /* Reroute. */ 1166 rcu_read_lock(); 1167 inet_opt = rcu_dereference(inet->inet_opt); 1168 daddr = inet->inet_daddr; 1169 if (inet_opt && inet_opt->opt.srr) 1170 daddr = inet_opt->opt.faddr; 1171 rcu_read_unlock(); 1172 fl4 = &inet->cork.fl.u.ip4; 1173 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr, 1174 inet->inet_dport, inet->inet_sport, 1175 sk->sk_protocol, RT_CONN_FLAGS(sk), 1176 sk->sk_bound_dev_if); 1177 if (!IS_ERR(rt)) { 1178 err = 0; 1179 sk_setup_caps(sk, &rt->dst); 1180 } else { 1181 err = PTR_ERR(rt); 1182 1183 /* Routing failed... */ 1184 sk->sk_route_caps = 0; 1185 /* 1186 * Other protocols have to map its equivalent state to TCP_SYN_SENT. 1187 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme 1188 */ 1189 if (!sysctl_ip_dynaddr || 1190 sk->sk_state != TCP_SYN_SENT || 1191 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) || 1192 (err = inet_sk_reselect_saddr(sk)) != 0) 1193 sk->sk_err_soft = -err; 1194 } 1195 1196 return err; 1197} 1198EXPORT_SYMBOL(inet_sk_rebuild_header); 1199 1200static struct sk_buff *inet_gso_segment(struct sk_buff *skb, 1201 netdev_features_t features) 1202{ 1203 struct sk_buff *segs = ERR_PTR(-EINVAL); 1204 const struct net_offload *ops; 1205 unsigned int offset = 0; 1206 bool udpfrag, encap; 1207 struct iphdr *iph; 1208 int proto; 1209 int nhoff; 1210 int ihl; 1211 int id; 1212 1213 if (unlikely(skb_shinfo(skb)->gso_type & 1214 ~(SKB_GSO_TCPV4 | 1215 SKB_GSO_UDP | 1216 SKB_GSO_DODGY | 1217 SKB_GSO_TCP_ECN | 1218 SKB_GSO_GRE | 1219 SKB_GSO_GRE_CSUM | 1220 SKB_GSO_IPIP | 1221 SKB_GSO_SIT | 1222 SKB_GSO_TCPV6 | 1223 SKB_GSO_UDP_TUNNEL | 1224 SKB_GSO_UDP_TUNNEL_CSUM | 1225 SKB_GSO_MPLS | 1226 0))) 1227 goto out; 1228 1229 skb_reset_network_header(skb); 1230 nhoff = skb_network_header(skb) - skb_mac_header(skb); 1231 if (unlikely(!pskb_may_pull(skb, sizeof(*iph)))) 1232 goto out; 1233 1234 iph = ip_hdr(skb); 1235 ihl = iph->ihl * 4; 1236 if (ihl < sizeof(*iph)) 1237 goto out; 1238 1239 id = ntohs(iph->id); 1240 proto = iph->protocol; 1241 1242 /* Warning: after this point, iph might be no longer valid */ 1243 if (unlikely(!pskb_may_pull(skb, ihl))) 1244 goto out; 1245 __skb_pull(skb, ihl); 1246 1247 encap = SKB_GSO_CB(skb)->encap_level > 0; 1248 if (encap) 1249 features = skb->dev->hw_enc_features & netif_skb_features(skb); 1250 SKB_GSO_CB(skb)->encap_level += ihl; 1251 1252 skb_reset_transport_header(skb); 1253 1254 segs = ERR_PTR(-EPROTONOSUPPORT); 1255 1256 if (skb->encapsulation && 1257 skb_shinfo(skb)->gso_type & (SKB_GSO_SIT|SKB_GSO_IPIP)) 1258 udpfrag = proto == IPPROTO_UDP && encap; 1259 else 1260 udpfrag = proto == IPPROTO_UDP && !skb->encapsulation; 1261 1262 ops = rcu_dereference(inet_offloads[proto]); 1263 if (likely(ops && ops->callbacks.gso_segment)) 1264 segs = ops->callbacks.gso_segment(skb, features); 1265 1266 if (IS_ERR_OR_NULL(segs)) 1267 goto out; 1268 1269 skb = segs; 1270 do { 1271 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff); 1272 if (udpfrag) { 1273 iph->id = htons(id); 1274 iph->frag_off = htons(offset >> 3); 1275 if (skb->next != NULL) 1276 iph->frag_off |= htons(IP_MF); 1277 offset += skb->len - nhoff - ihl; 1278 } else { 1279 iph->id = htons(id++); 1280 } 1281 iph->tot_len = htons(skb->len - nhoff); 1282 ip_send_check(iph); 1283 if (encap) 1284 skb_reset_inner_headers(skb); 1285 skb->network_header = (u8 *)iph - skb->head; 1286 } while ((skb = skb->next)); 1287 1288out: 1289 return segs; 1290} 1291 1292static struct sk_buff **inet_gro_receive(struct sk_buff **head, 1293 struct sk_buff *skb) 1294{ 1295 const struct net_offload *ops; 1296 struct sk_buff **pp = NULL; 1297 struct sk_buff *p; 1298 const struct iphdr *iph; 1299 unsigned int hlen; 1300 unsigned int off; 1301 unsigned int id; 1302 int flush = 1; 1303 int proto; 1304 1305 off = skb_gro_offset(skb); 1306 hlen = off + sizeof(*iph); 1307 iph = skb_gro_header_fast(skb, off); 1308 if (skb_gro_header_hard(skb, hlen)) { 1309 iph = skb_gro_header_slow(skb, hlen, off); 1310 if (unlikely(!iph)) 1311 goto out; 1312 } 1313 1314 proto = iph->protocol; 1315 1316 rcu_read_lock(); 1317 ops = rcu_dereference(inet_offloads[proto]); 1318 if (!ops || !ops->callbacks.gro_receive) 1319 goto out_unlock; 1320 1321 if (*(u8 *)iph != 0x45) 1322 goto out_unlock; 1323 1324 if (unlikely(ip_fast_csum((u8 *)iph, 5))) 1325 goto out_unlock; 1326 1327 id = ntohl(*(__be32 *)&iph->id); 1328 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF)); 1329 id >>= 16; 1330 1331 for (p = *head; p; p = p->next) { 1332 struct iphdr *iph2; 1333 1334 if (!NAPI_GRO_CB(p)->same_flow) 1335 continue; 1336 1337 iph2 = (struct iphdr *)(p->data + off); 1338 /* The above works because, with the exception of the top 1339 * (inner most) layer, we only aggregate pkts with the same 1340 * hdr length so all the hdrs we'll need to verify will start 1341 * at the same offset. 1342 */ 1343 if ((iph->protocol ^ iph2->protocol) | 1344 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) | 1345 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) { 1346 NAPI_GRO_CB(p)->same_flow = 0; 1347 continue; 1348 } 1349 1350 /* All fields must match except length and checksum. */ 1351 NAPI_GRO_CB(p)->flush |= 1352 (iph->ttl ^ iph2->ttl) | 1353 (iph->tos ^ iph2->tos) | 1354 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF)); 1355 1356 /* Save the IP ID check to be included later when we get to 1357 * the transport layer so only the inner most IP ID is checked. 1358 * This is because some GSO/TSO implementations do not 1359 * correctly increment the IP ID for the outer hdrs. 1360 */ 1361 NAPI_GRO_CB(p)->flush_id = 1362 ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id); 1363 NAPI_GRO_CB(p)->flush |= flush; 1364 } 1365 1366 NAPI_GRO_CB(skb)->flush |= flush; 1367 skb_set_network_header(skb, off); 1368 /* The above will be needed by the transport layer if there is one 1369 * immediately following this IP hdr. 1370 */ 1371 1372 /* Note : No need to call skb_gro_postpull_rcsum() here, 1373 * as we already checked checksum over ipv4 header was 0 1374 */ 1375 skb_gro_pull(skb, sizeof(*iph)); 1376 skb_set_transport_header(skb, skb_gro_offset(skb)); 1377 1378 pp = ops->callbacks.gro_receive(head, skb); 1379 1380out_unlock: 1381 rcu_read_unlock(); 1382 1383out: 1384 NAPI_GRO_CB(skb)->flush |= flush; 1385 1386 return pp; 1387} 1388 1389static int inet_gro_complete(struct sk_buff *skb, int nhoff) 1390{ 1391 __be16 newlen = htons(skb->len - nhoff); 1392 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff); 1393 const struct net_offload *ops; 1394 int proto = iph->protocol; 1395 int err = -ENOSYS; 1396 1397 if (skb->encapsulation) 1398 skb_set_inner_network_header(skb, nhoff); 1399 1400 csum_replace2(&iph->check, iph->tot_len, newlen); 1401 iph->tot_len = newlen; 1402 1403 rcu_read_lock(); 1404 ops = rcu_dereference(inet_offloads[proto]); 1405 if (WARN_ON(!ops || !ops->callbacks.gro_complete)) 1406 goto out_unlock; 1407 1408 /* Only need to add sizeof(*iph) to get to the next hdr below 1409 * because any hdr with option will have been flushed in 1410 * inet_gro_receive(). 1411 */ 1412 err = ops->callbacks.gro_complete(skb, nhoff + sizeof(*iph)); 1413 1414out_unlock: 1415 rcu_read_unlock(); 1416 1417 return err; 1418} 1419 1420int inet_ctl_sock_create(struct sock **sk, unsigned short family, 1421 unsigned short type, unsigned char protocol, 1422 struct net *net) 1423{ 1424 struct socket *sock; 1425 int rc = sock_create_kern(family, type, protocol, &sock); 1426 1427 if (rc == 0) { 1428 *sk = sock->sk; 1429 (*sk)->sk_allocation = GFP_ATOMIC; 1430 /* 1431 * Unhash it so that IP input processing does not even see it, 1432 * we do not wish this socket to see incoming packets. 1433 */ 1434 (*sk)->sk_prot->unhash(*sk); 1435 1436 sk_change_net(*sk, net); 1437 } 1438 return rc; 1439} 1440EXPORT_SYMBOL_GPL(inet_ctl_sock_create); 1441 1442unsigned long snmp_fold_field(void __percpu *mib, int offt) 1443{ 1444 unsigned long res = 0; 1445 int i; 1446 1447 for_each_possible_cpu(i) 1448 res += *(((unsigned long *) per_cpu_ptr(mib, i)) + offt); 1449 return res; 1450} 1451EXPORT_SYMBOL_GPL(snmp_fold_field); 1452 1453#if BITS_PER_LONG==32 1454 1455u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset) 1456{ 1457 u64 res = 0; 1458 int cpu; 1459 1460 for_each_possible_cpu(cpu) { 1461 void *bhptr; 1462 struct u64_stats_sync *syncp; 1463 u64 v; 1464 unsigned int start; 1465 1466 bhptr = per_cpu_ptr(mib, cpu); 1467 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset); 1468 do { 1469 start = u64_stats_fetch_begin_irq(syncp); 1470 v = *(((u64 *) bhptr) + offt); 1471 } while (u64_stats_fetch_retry_irq(syncp, start)); 1472 1473 res += v; 1474 } 1475 return res; 1476} 1477EXPORT_SYMBOL_GPL(snmp_fold_field64); 1478#endif 1479 1480#ifdef CONFIG_IP_MULTICAST 1481static const struct net_protocol igmp_protocol = { 1482 .handler = igmp_rcv, 1483 .netns_ok = 1, 1484}; 1485#endif 1486 1487static const struct net_protocol tcp_protocol = { 1488 .early_demux = tcp_v4_early_demux, 1489 .handler = tcp_v4_rcv, 1490 .err_handler = tcp_v4_err, 1491 .no_policy = 1, 1492 .netns_ok = 1, 1493 .icmp_strict_tag_validation = 1, 1494}; 1495 1496static const struct net_protocol udp_protocol = { 1497 .early_demux = udp_v4_early_demux, 1498 .handler = udp_rcv, 1499 .err_handler = udp_err, 1500 .no_policy = 1, 1501 .netns_ok = 1, 1502}; 1503 1504static const struct net_protocol icmp_protocol = { 1505 .handler = icmp_rcv, 1506 .err_handler = icmp_err, 1507 .no_policy = 1, 1508 .netns_ok = 1, 1509}; 1510 1511static __net_init int ipv4_mib_init_net(struct net *net) 1512{ 1513 int i; 1514 1515 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib); 1516 if (!net->mib.tcp_statistics) 1517 goto err_tcp_mib; 1518 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib); 1519 if (!net->mib.ip_statistics) 1520 goto err_ip_mib; 1521 1522 for_each_possible_cpu(i) { 1523 struct ipstats_mib *af_inet_stats; 1524 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i); 1525 u64_stats_init(&af_inet_stats->syncp); 1526 } 1527 1528 net->mib.net_statistics = alloc_percpu(struct linux_mib); 1529 if (!net->mib.net_statistics) 1530 goto err_net_mib; 1531 net->mib.udp_statistics = alloc_percpu(struct udp_mib); 1532 if (!net->mib.udp_statistics) 1533 goto err_udp_mib; 1534 net->mib.udplite_statistics = alloc_percpu(struct udp_mib); 1535 if (!net->mib.udplite_statistics) 1536 goto err_udplite_mib; 1537 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib); 1538 if (!net->mib.icmp_statistics) 1539 goto err_icmp_mib; 1540 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib), 1541 GFP_KERNEL); 1542 if (!net->mib.icmpmsg_statistics) 1543 goto err_icmpmsg_mib; 1544 1545 tcp_mib_init(net); 1546 return 0; 1547 1548err_icmpmsg_mib: 1549 free_percpu(net->mib.icmp_statistics); 1550err_icmp_mib: 1551 free_percpu(net->mib.udplite_statistics); 1552err_udplite_mib: 1553 free_percpu(net->mib.udp_statistics); 1554err_udp_mib: 1555 free_percpu(net->mib.net_statistics); 1556err_net_mib: 1557 free_percpu(net->mib.ip_statistics); 1558err_ip_mib: 1559 free_percpu(net->mib.tcp_statistics); 1560err_tcp_mib: 1561 return -ENOMEM; 1562} 1563 1564static __net_exit void ipv4_mib_exit_net(struct net *net) 1565{ 1566 kfree(net->mib.icmpmsg_statistics); 1567 free_percpu(net->mib.icmp_statistics); 1568 free_percpu(net->mib.udplite_statistics); 1569 free_percpu(net->mib.udp_statistics); 1570 free_percpu(net->mib.net_statistics); 1571 free_percpu(net->mib.ip_statistics); 1572 free_percpu(net->mib.tcp_statistics); 1573} 1574 1575static __net_initdata struct pernet_operations ipv4_mib_ops = { 1576 .init = ipv4_mib_init_net, 1577 .exit = ipv4_mib_exit_net, 1578}; 1579 1580static int __init init_ipv4_mibs(void) 1581{ 1582 return register_pernet_subsys(&ipv4_mib_ops); 1583} 1584 1585static __net_init int inet_init_net(struct net *net) 1586{ 1587 /* 1588 * Set defaults for local port range 1589 */ 1590 seqlock_init(&net->ipv4.ip_local_ports.lock); 1591 net->ipv4.ip_local_ports.range[0] = 32768; 1592 net->ipv4.ip_local_ports.range[1] = 61000; 1593 1594 seqlock_init(&net->ipv4.ping_group_range.lock); 1595 /* 1596 * Sane defaults - nobody may create ping sockets. 1597 * Boot scripts should set this to distro-specific group. 1598 */ 1599 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1); 1600 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0); 1601 return 0; 1602} 1603 1604static __net_exit void inet_exit_net(struct net *net) 1605{ 1606} 1607 1608static __net_initdata struct pernet_operations af_inet_ops = { 1609 .init = inet_init_net, 1610 .exit = inet_exit_net, 1611}; 1612 1613static int __init init_inet_pernet_ops(void) 1614{ 1615 return register_pernet_subsys(&af_inet_ops); 1616} 1617 1618static int ipv4_proc_init(void); 1619 1620/* 1621 * IP protocol layer initialiser 1622 */ 1623 1624static struct packet_offload ip_packet_offload __read_mostly = { 1625 .type = cpu_to_be16(ETH_P_IP), 1626 .callbacks = { 1627 .gso_segment = inet_gso_segment, 1628 .gro_receive = inet_gro_receive, 1629 .gro_complete = inet_gro_complete, 1630 }, 1631}; 1632 1633static const struct net_offload ipip_offload = { 1634 .callbacks = { 1635 .gso_segment = inet_gso_segment, 1636 .gro_receive = inet_gro_receive, 1637 .gro_complete = inet_gro_complete, 1638 }, 1639}; 1640 1641static int __init ipv4_offload_init(void) 1642{ 1643 /* 1644 * Add offloads 1645 */ 1646 if (udpv4_offload_init() < 0) 1647 pr_crit("%s: Cannot add UDP protocol offload\n", __func__); 1648 if (tcpv4_offload_init() < 0) 1649 pr_crit("%s: Cannot add TCP protocol offload\n", __func__); 1650 1651 dev_add_offload(&ip_packet_offload); 1652 inet_add_offload(&ipip_offload, IPPROTO_IPIP); 1653 return 0; 1654} 1655 1656fs_initcall(ipv4_offload_init); 1657 1658static struct packet_type ip_packet_type __read_mostly = { 1659 .type = cpu_to_be16(ETH_P_IP), 1660 .func = ip_rcv, 1661}; 1662 1663static int __init inet_init(void) 1664{ 1665 struct inet_protosw *q; 1666 struct list_head *r; 1667 int rc = -EINVAL; 1668 1669 BUILD_BUG_ON(sizeof(struct inet_skb_parm) > FIELD_SIZEOF(struct sk_buff, cb)); 1670 1671 rc = proto_register(&tcp_prot, 1); 1672 if (rc) 1673 goto out; 1674 1675 rc = proto_register(&udp_prot, 1); 1676 if (rc) 1677 goto out_unregister_tcp_proto; 1678 1679 rc = proto_register(&raw_prot, 1); 1680 if (rc) 1681 goto out_unregister_udp_proto; 1682 1683 rc = proto_register(&ping_prot, 1); 1684 if (rc) 1685 goto out_unregister_raw_proto; 1686 1687 /* 1688 * Tell SOCKET that we are alive... 1689 */ 1690 1691 (void)sock_register(&inet_family_ops); 1692 1693#ifdef CONFIG_SYSCTL 1694 ip_static_sysctl_init(); 1695#endif 1696 1697 /* 1698 * Add all the base protocols. 1699 */ 1700 1701 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0) 1702 pr_crit("%s: Cannot add ICMP protocol\n", __func__); 1703 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0) 1704 pr_crit("%s: Cannot add UDP protocol\n", __func__); 1705 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0) 1706 pr_crit("%s: Cannot add TCP protocol\n", __func__); 1707#ifdef CONFIG_IP_MULTICAST 1708 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0) 1709 pr_crit("%s: Cannot add IGMP protocol\n", __func__); 1710#endif 1711 1712 /* Register the socket-side information for inet_create. */ 1713 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r) 1714 INIT_LIST_HEAD(r); 1715 1716 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q) 1717 inet_register_protosw(q); 1718 1719 /* 1720 * Set the ARP module up 1721 */ 1722 1723 arp_init(); 1724 1725 /* 1726 * Set the IP module up 1727 */ 1728 1729 ip_init(); 1730 1731 tcp_v4_init(); 1732 1733 /* Setup TCP slab cache for open requests. */ 1734 tcp_init(); 1735 1736 /* Setup UDP memory threshold */ 1737 udp_init(); 1738 1739 /* Add UDP-Lite (RFC 3828) */ 1740 udplite4_register(); 1741 1742 ping_init(); 1743 1744 /* 1745 * Set the ICMP layer up 1746 */ 1747 1748 if (icmp_init() < 0) 1749 panic("Failed to create the ICMP control socket.\n"); 1750 1751 /* 1752 * Initialise the multicast router 1753 */ 1754#if defined(CONFIG_IP_MROUTE) 1755 if (ip_mr_init()) 1756 pr_crit("%s: Cannot init ipv4 mroute\n", __func__); 1757#endif 1758 1759 if (init_inet_pernet_ops()) 1760 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__); 1761 /* 1762 * Initialise per-cpu ipv4 mibs 1763 */ 1764 1765 if (init_ipv4_mibs()) 1766 pr_crit("%s: Cannot init ipv4 mibs\n", __func__); 1767 1768 ipv4_proc_init(); 1769 1770 ipfrag_init(); 1771 1772 dev_add_pack(&ip_packet_type); 1773 1774 rc = 0; 1775out: 1776 return rc; 1777out_unregister_raw_proto: 1778 proto_unregister(&raw_prot); 1779out_unregister_udp_proto: 1780 proto_unregister(&udp_prot); 1781out_unregister_tcp_proto: 1782 proto_unregister(&tcp_prot); 1783 goto out; 1784} 1785 1786fs_initcall(inet_init); 1787 1788/* ------------------------------------------------------------------------ */ 1789 1790#ifdef CONFIG_PROC_FS 1791static int __init ipv4_proc_init(void) 1792{ 1793 int rc = 0; 1794 1795 if (raw_proc_init()) 1796 goto out_raw; 1797 if (tcp4_proc_init()) 1798 goto out_tcp; 1799 if (udp4_proc_init()) 1800 goto out_udp; 1801 if (ping_proc_init()) 1802 goto out_ping; 1803 if (ip_misc_proc_init()) 1804 goto out_misc; 1805out: 1806 return rc; 1807out_misc: 1808 ping_proc_exit(); 1809out_ping: 1810 udp4_proc_exit(); 1811out_udp: 1812 tcp4_proc_exit(); 1813out_tcp: 1814 raw_proc_exit(); 1815out_raw: 1816 rc = -ENOMEM; 1817 goto out; 1818} 1819 1820#else /* CONFIG_PROC_FS */ 1821static int __init ipv4_proc_init(void) 1822{ 1823 return 0; 1824} 1825#endif /* CONFIG_PROC_FS */ 1826 1827MODULE_ALIAS_NETPROTO(PF_INET); 1828 1829