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 #include <linux/err.h> 69 #include <linux/errno.h> 70 #include <linux/types.h> 71 #include <linux/socket.h> 72 #include <linux/in.h> 73 #include <linux/kernel.h> 74 #include <linux/module.h> 75 #include <linux/sched.h> 76 #include <linux/timer.h> 77 #include <linux/string.h> 78 #include <linux/sockios.h> 79 #include <linux/net.h> 80 #include <linux/capability.h> 81 #include <linux/fcntl.h> 82 #include <linux/mm.h> 83 #include <linux/interrupt.h> 84 #include <linux/stat.h> 85 #include <linux/init.h> 86 #include <linux/poll.h> 87 #include <linux/netfilter_ipv4.h> 88 #include <linux/random.h> 89 #include <linux/slab.h> 90 91 #include <asm/uaccess.h> 92 #include <asm/system.h> 93 94 #include <linux/inet.h> 95 #include <linux/igmp.h> 96 #include <linux/inetdevice.h> 97 #include <linux/netdevice.h> 98 #include <net/checksum.h> 99 #include <net/ip.h> 100 #include <net/protocol.h> 101 #include <net/arp.h> 102 #include <net/route.h> 103 #include <net/ip_fib.h> 104 #include <net/inet_connection_sock.h> 105 #include <net/tcp.h> 106 #include <net/udp.h> 107 #include <net/udplite.h> 108 #include <net/ping.h> 109 #include <linux/skbuff.h> 110 #include <net/sock.h> 111 #include <net/raw.h> 112 #include <net/icmp.h> 113 #include <net/ipip.h> 114 #include <net/inet_common.h> 115 #include <net/xfrm.h> 116 #include <net/net_namespace.h> 117 #ifdef CONFIG_IP_MROUTE 118 #include <linux/mroute.h> 119 #endif 120 121 122 /* The inetsw table contains everything that inet_create needs to 123 * build a new socket. 124 */ 125 static struct list_head inetsw[SOCK_MAX]; 126 static DEFINE_SPINLOCK(inetsw_lock); 127 128 struct ipv4_config ipv4_config; 129 EXPORT_SYMBOL(ipv4_config); 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 sk_refcnt_debug_dec(sk); 160 } 161 EXPORT_SYMBOL(inet_sock_destruct); 162 163 /* 164 * The routines beyond this point handle the behaviour of an AF_INET 165 * socket object. Mostly it punts to the subprotocols of IP to do 166 * the work. 167 */ 168 169 /* 170 * Automatically bind an unbound socket. 171 */ 172 173 static int inet_autobind(struct sock *sk) 174 { 175 struct inet_sock *inet; 176 /* We may need to bind the socket. */ 177 lock_sock(sk); 178 inet = inet_sk(sk); 179 if (!inet->inet_num) { 180 if (sk->sk_prot->get_port(sk, 0)) { 181 release_sock(sk); 182 return -EAGAIN; 183 } 184 inet->inet_sport = htons(inet->inet_num); 185 } 186 release_sock(sk); 187 return 0; 188 } 189 190 /* 191 * Move a socket into listening state. 192 */ 193 int inet_listen(struct socket *sock, int backlog) 194 { 195 struct sock *sk = sock->sk; 196 unsigned char old_state; 197 int err; 198 199 lock_sock(sk); 200 201 err = -EINVAL; 202 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM) 203 goto out; 204 205 old_state = sk->sk_state; 206 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN))) 207 goto out; 208 209 /* Really, if the socket is already in listen state 210 * we can only allow the backlog to be adjusted. 211 */ 212 if (old_state != TCP_LISTEN) { 213 err = inet_csk_listen_start(sk, backlog); 214 if (err) 215 goto out; 216 } 217 sk->sk_max_ack_backlog = backlog; 218 err = 0; 219 220 out: 221 release_sock(sk); 222 return err; 223 } 224 EXPORT_SYMBOL(inet_listen); 225 226 u32 inet_ehash_secret __read_mostly; 227 EXPORT_SYMBOL(inet_ehash_secret); 228 229 /* 230 * inet_ehash_secret must be set exactly once 231 */ 232 void build_ehash_secret(void) 233 { 234 u32 rnd; 235 236 do { 237 get_random_bytes(&rnd, sizeof(rnd)); 238 } while (rnd == 0); 239 240 cmpxchg(&inet_ehash_secret, 0, rnd); 241 } 242 EXPORT_SYMBOL(build_ehash_secret); 243 244 static inline int inet_netns_ok(struct net *net, int protocol) 245 { 246 int hash; 247 const struct net_protocol *ipprot; 248 249 if (net_eq(net, &init_net)) 250 return 1; 251 252 hash = protocol & (MAX_INET_PROTOS - 1); 253 ipprot = rcu_dereference(inet_protos[hash]); 254 255 if (ipprot == NULL) 256 /* raw IP is OK */ 257 return 1; 258 return ipprot->netns_ok; 259 } 260 261 /* 262 * Create an inet socket. 263 */ 264 265 static int inet_create(struct net *net, struct socket *sock, int protocol, 266 int kern) 267 { 268 struct sock *sk; 269 struct inet_protosw *answer; 270 struct inet_sock *inet; 271 struct proto *answer_prot; 272 unsigned char answer_flags; 273 char answer_no_check; 274 int try_loading_module = 0; 275 int err; 276 277 if (unlikely(!inet_ehash_secret)) 278 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM) 279 build_ehash_secret(); 280 281 sock->state = SS_UNCONNECTED; 282 283 /* Look for the requested type/protocol pair. */ 284 lookup_protocol: 285 err = -ESOCKTNOSUPPORT; 286 rcu_read_lock(); 287 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) { 288 289 err = 0; 290 /* Check the non-wild match. */ 291 if (protocol == answer->protocol) { 292 if (protocol != IPPROTO_IP) 293 break; 294 } else { 295 /* Check for the two wild cases. */ 296 if (IPPROTO_IP == protocol) { 297 protocol = answer->protocol; 298 break; 299 } 300 if (IPPROTO_IP == answer->protocol) 301 break; 302 } 303 err = -EPROTONOSUPPORT; 304 } 305 306 if (unlikely(err)) { 307 if (try_loading_module < 2) { 308 rcu_read_unlock(); 309 /* 310 * Be more specific, e.g. net-pf-2-proto-132-type-1 311 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM) 312 */ 313 if (++try_loading_module == 1) 314 request_module("net-pf-%d-proto-%d-type-%d", 315 PF_INET, protocol, sock->type); 316 /* 317 * Fall back to generic, e.g. net-pf-2-proto-132 318 * (net-pf-PF_INET-proto-IPPROTO_SCTP) 319 */ 320 else 321 request_module("net-pf-%d-proto-%d", 322 PF_INET, protocol); 323 goto lookup_protocol; 324 } else 325 goto out_rcu_unlock; 326 } 327 328 err = -EPERM; 329 if (sock->type == SOCK_RAW && !kern && !capable(CAP_NET_RAW)) 330 goto out_rcu_unlock; 331 332 err = -EAFNOSUPPORT; 333 if (!inet_netns_ok(net, protocol)) 334 goto out_rcu_unlock; 335 336 sock->ops = answer->ops; 337 answer_prot = answer->prot; 338 answer_no_check = answer->no_check; 339 answer_flags = answer->flags; 340 rcu_read_unlock(); 341 342 WARN_ON(answer_prot->slab == NULL); 343 344 err = -ENOBUFS; 345 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot); 346 if (sk == NULL) 347 goto out; 348 349 err = 0; 350 sk->sk_no_check = answer_no_check; 351 if (INET_PROTOSW_REUSE & answer_flags) 352 sk->sk_reuse = 1; 353 354 inet = inet_sk(sk); 355 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0; 356 357 inet->nodefrag = 0; 358 359 if (SOCK_RAW == sock->type) { 360 inet->inet_num = protocol; 361 if (IPPROTO_RAW == protocol) 362 inet->hdrincl = 1; 363 } 364 365 if (ipv4_config.no_pmtu_disc) 366 inet->pmtudisc = IP_PMTUDISC_DONT; 367 else 368 inet->pmtudisc = IP_PMTUDISC_WANT; 369 370 inet->inet_id = 0; 371 372 sock_init_data(sock, sk); 373 374 sk->sk_destruct = inet_sock_destruct; 375 sk->sk_protocol = protocol; 376 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv; 377 378 inet->uc_ttl = -1; 379 inet->mc_loop = 1; 380 inet->mc_ttl = 1; 381 inet->mc_all = 1; 382 inet->mc_index = 0; 383 inet->mc_list = NULL; 384 385 sk_refcnt_debug_inc(sk); 386 387 if (inet->inet_num) { 388 /* It assumes that any protocol which allows 389 * the user to assign a number at socket 390 * creation time automatically 391 * shares. 392 */ 393 inet->inet_sport = htons(inet->inet_num); 394 /* Add to protocol hash chains. */ 395 sk->sk_prot->hash(sk); 396 } 397 398 if (sk->sk_prot->init) { 399 err = sk->sk_prot->init(sk); 400 if (err) 401 sk_common_release(sk); 402 } 403 out: 404 return err; 405 out_rcu_unlock: 406 rcu_read_unlock(); 407 goto out; 408 } 409 410 411 /* 412 * The peer socket should always be NULL (or else). When we call this 413 * function we are destroying the object and from then on nobody 414 * should refer to it. 415 */ 416 int inet_release(struct socket *sock) 417 { 418 struct sock *sk = sock->sk; 419 420 if (sk) { 421 long timeout; 422 423 sock_rps_reset_flow(sk); 424 425 /* Applications forget to leave groups before exiting */ 426 ip_mc_drop_socket(sk); 427 428 /* If linger is set, we don't return until the close 429 * is complete. Otherwise we return immediately. The 430 * actually closing is done the same either way. 431 * 432 * If the close is due to the process exiting, we never 433 * linger.. 434 */ 435 timeout = 0; 436 if (sock_flag(sk, SOCK_LINGER) && 437 !(current->flags & PF_EXITING)) 438 timeout = sk->sk_lingertime; 439 sock->sk = NULL; 440 sk->sk_prot->close(sk, timeout); 441 } 442 return 0; 443 } 444 EXPORT_SYMBOL(inet_release); 445 446 /* It is off by default, see below. */ 447 int sysctl_ip_nonlocal_bind __read_mostly; 448 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind); 449 450 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) 451 { 452 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr; 453 struct sock *sk = sock->sk; 454 struct inet_sock *inet = inet_sk(sk); 455 unsigned short snum; 456 int chk_addr_ret; 457 int err; 458 459 /* If the socket has its own bind function then use it. (RAW) */ 460 if (sk->sk_prot->bind) { 461 err = sk->sk_prot->bind(sk, uaddr, addr_len); 462 goto out; 463 } 464 err = -EINVAL; 465 if (addr_len < sizeof(struct sockaddr_in)) 466 goto out; 467 468 if (addr->sin_family != AF_INET) 469 goto out; 470 471 chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr); 472 473 /* Not specified by any standard per-se, however it breaks too 474 * many applications when removed. It is unfortunate since 475 * allowing applications to make a non-local bind solves 476 * several problems with systems using dynamic addressing. 477 * (ie. your servers still start up even if your ISDN link 478 * is temporarily down) 479 */ 480 err = -EADDRNOTAVAIL; 481 if (!sysctl_ip_nonlocal_bind && 482 !(inet->freebind || inet->transparent) && 483 addr->sin_addr.s_addr != htonl(INADDR_ANY) && 484 chk_addr_ret != RTN_LOCAL && 485 chk_addr_ret != RTN_MULTICAST && 486 chk_addr_ret != RTN_BROADCAST) 487 goto out; 488 489 snum = ntohs(addr->sin_port); 490 err = -EACCES; 491 if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE)) 492 goto out; 493 494 /* We keep a pair of addresses. rcv_saddr is the one 495 * used by hash lookups, and saddr is used for transmit. 496 * 497 * In the BSD API these are the same except where it 498 * would be illegal to use them (multicast/broadcast) in 499 * which case the sending device address is used. 500 */ 501 lock_sock(sk); 502 503 /* Check these errors (active socket, double bind). */ 504 err = -EINVAL; 505 if (sk->sk_state != TCP_CLOSE || inet->inet_num) 506 goto out_release_sock; 507 508 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr; 509 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST) 510 inet->inet_saddr = 0; /* Use device */ 511 512 /* Make sure we are allowed to bind here. */ 513 if (sk->sk_prot->get_port(sk, snum)) { 514 inet->inet_saddr = inet->inet_rcv_saddr = 0; 515 err = -EADDRINUSE; 516 goto out_release_sock; 517 } 518 519 if (inet->inet_rcv_saddr) 520 sk->sk_userlocks |= SOCK_BINDADDR_LOCK; 521 if (snum) 522 sk->sk_userlocks |= SOCK_BINDPORT_LOCK; 523 inet->inet_sport = htons(inet->inet_num); 524 inet->inet_daddr = 0; 525 inet->inet_dport = 0; 526 sk_dst_reset(sk); 527 err = 0; 528 out_release_sock: 529 release_sock(sk); 530 out: 531 return err; 532 } 533 EXPORT_SYMBOL(inet_bind); 534 535 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr, 536 int addr_len, int flags) 537 { 538 struct sock *sk = sock->sk; 539 540 if (addr_len < sizeof(uaddr->sa_family)) 541 return -EINVAL; 542 if (uaddr->sa_family == AF_UNSPEC) 543 return sk->sk_prot->disconnect(sk, flags); 544 545 if (!inet_sk(sk)->inet_num && inet_autobind(sk)) 546 return -EAGAIN; 547 return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len); 548 } 549 EXPORT_SYMBOL(inet_dgram_connect); 550 551 static long inet_wait_for_connect(struct sock *sk, long timeo) 552 { 553 DEFINE_WAIT(wait); 554 555 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 556 557 /* Basic assumption: if someone sets sk->sk_err, he _must_ 558 * change state of the socket from TCP_SYN_*. 559 * Connect() does not allow to get error notifications 560 * without closing the socket. 561 */ 562 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 563 release_sock(sk); 564 timeo = schedule_timeout(timeo); 565 lock_sock(sk); 566 if (signal_pending(current) || !timeo) 567 break; 568 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 569 } 570 finish_wait(sk_sleep(sk), &wait); 571 return timeo; 572 } 573 574 /* 575 * Connect to a remote host. There is regrettably still a little 576 * TCP 'magic' in here. 577 */ 578 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr, 579 int addr_len, int flags) 580 { 581 struct sock *sk = sock->sk; 582 int err; 583 long timeo; 584 585 if (addr_len < sizeof(uaddr->sa_family)) 586 return -EINVAL; 587 588 lock_sock(sk); 589 590 if (uaddr->sa_family == AF_UNSPEC) { 591 err = sk->sk_prot->disconnect(sk, flags); 592 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED; 593 goto out; 594 } 595 596 switch (sock->state) { 597 default: 598 err = -EINVAL; 599 goto out; 600 case SS_CONNECTED: 601 err = -EISCONN; 602 goto out; 603 case SS_CONNECTING: 604 err = -EALREADY; 605 /* Fall out of switch with err, set for this state */ 606 break; 607 case SS_UNCONNECTED: 608 err = -EISCONN; 609 if (sk->sk_state != TCP_CLOSE) 610 goto out; 611 612 err = sk->sk_prot->connect(sk, uaddr, addr_len); 613 if (err < 0) 614 goto out; 615 616 sock->state = SS_CONNECTING; 617 618 /* Just entered SS_CONNECTING state; the only 619 * difference is that return value in non-blocking 620 * case is EINPROGRESS, rather than EALREADY. 621 */ 622 err = -EINPROGRESS; 623 break; 624 } 625 626 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK); 627 628 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 629 /* Error code is set above */ 630 if (!timeo || !inet_wait_for_connect(sk, timeo)) 631 goto out; 632 633 err = sock_intr_errno(timeo); 634 if (signal_pending(current)) 635 goto out; 636 } 637 638 /* Connection was closed by RST, timeout, ICMP error 639 * or another process disconnected us. 640 */ 641 if (sk->sk_state == TCP_CLOSE) 642 goto sock_error; 643 644 /* sk->sk_err may be not zero now, if RECVERR was ordered by user 645 * and error was received after socket entered established state. 646 * Hence, it is handled normally after connect() return successfully. 647 */ 648 649 sock->state = SS_CONNECTED; 650 err = 0; 651 out: 652 release_sock(sk); 653 return err; 654 655 sock_error: 656 err = sock_error(sk) ? : -ECONNABORTED; 657 sock->state = SS_UNCONNECTED; 658 if (sk->sk_prot->disconnect(sk, flags)) 659 sock->state = SS_DISCONNECTING; 660 goto out; 661 } 662 EXPORT_SYMBOL(inet_stream_connect); 663 664 /* 665 * Accept a pending connection. The TCP layer now gives BSD semantics. 666 */ 667 668 int inet_accept(struct socket *sock, struct socket *newsock, int flags) 669 { 670 struct sock *sk1 = sock->sk; 671 int err = -EINVAL; 672 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err); 673 674 if (!sk2) 675 goto do_err; 676 677 lock_sock(sk2); 678 679 sock_rps_record_flow(sk2); 680 WARN_ON(!((1 << sk2->sk_state) & 681 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE))); 682 683 sock_graft(sk2, newsock); 684 685 newsock->state = SS_CONNECTED; 686 err = 0; 687 release_sock(sk2); 688 do_err: 689 return err; 690 } 691 EXPORT_SYMBOL(inet_accept); 692 693 694 /* 695 * This does both peername and sockname. 696 */ 697 int inet_getname(struct socket *sock, struct sockaddr *uaddr, 698 int *uaddr_len, int peer) 699 { 700 struct sock *sk = sock->sk; 701 struct inet_sock *inet = inet_sk(sk); 702 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr); 703 704 sin->sin_family = AF_INET; 705 if (peer) { 706 if (!inet->inet_dport || 707 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) && 708 peer == 1)) 709 return -ENOTCONN; 710 sin->sin_port = inet->inet_dport; 711 sin->sin_addr.s_addr = inet->inet_daddr; 712 } else { 713 __be32 addr = inet->inet_rcv_saddr; 714 if (!addr) 715 addr = inet->inet_saddr; 716 sin->sin_port = inet->inet_sport; 717 sin->sin_addr.s_addr = addr; 718 } 719 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 720 *uaddr_len = sizeof(*sin); 721 return 0; 722 } 723 EXPORT_SYMBOL(inet_getname); 724 725 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg, 726 size_t size) 727 { 728 struct sock *sk = sock->sk; 729 730 sock_rps_record_flow(sk); 731 732 /* We may need to bind the socket. */ 733 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind && 734 inet_autobind(sk)) 735 return -EAGAIN; 736 737 return sk->sk_prot->sendmsg(iocb, sk, msg, size); 738 } 739 EXPORT_SYMBOL(inet_sendmsg); 740 741 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, 742 size_t size, int flags) 743 { 744 struct sock *sk = sock->sk; 745 746 sock_rps_record_flow(sk); 747 748 /* We may need to bind the socket. */ 749 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind && 750 inet_autobind(sk)) 751 return -EAGAIN; 752 753 if (sk->sk_prot->sendpage) 754 return sk->sk_prot->sendpage(sk, page, offset, size, flags); 755 return sock_no_sendpage(sock, page, offset, size, flags); 756 } 757 EXPORT_SYMBOL(inet_sendpage); 758 759 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg, 760 size_t size, int flags) 761 { 762 struct sock *sk = sock->sk; 763 int addr_len = 0; 764 int err; 765 766 sock_rps_record_flow(sk); 767 768 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT, 769 flags & ~MSG_DONTWAIT, &addr_len); 770 if (err >= 0) 771 msg->msg_namelen = addr_len; 772 return err; 773 } 774 EXPORT_SYMBOL(inet_recvmsg); 775 776 int inet_shutdown(struct socket *sock, int how) 777 { 778 struct sock *sk = sock->sk; 779 int err = 0; 780 781 /* This should really check to make sure 782 * the socket is a TCP socket. (WHY AC...) 783 */ 784 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and 785 1->2 bit 2 snds. 786 2->3 */ 787 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */ 788 return -EINVAL; 789 790 lock_sock(sk); 791 if (sock->state == SS_CONNECTING) { 792 if ((1 << sk->sk_state) & 793 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE)) 794 sock->state = SS_DISCONNECTING; 795 else 796 sock->state = SS_CONNECTED; 797 } 798 799 switch (sk->sk_state) { 800 case TCP_CLOSE: 801 err = -ENOTCONN; 802 /* Hack to wake up other listeners, who can poll for 803 POLLHUP, even on eg. unconnected UDP sockets -- RR */ 804 default: 805 sk->sk_shutdown |= how; 806 if (sk->sk_prot->shutdown) 807 sk->sk_prot->shutdown(sk, how); 808 break; 809 810 /* Remaining two branches are temporary solution for missing 811 * close() in multithreaded environment. It is _not_ a good idea, 812 * but we have no choice until close() is repaired at VFS level. 813 */ 814 case TCP_LISTEN: 815 if (!(how & RCV_SHUTDOWN)) 816 break; 817 /* Fall through */ 818 case TCP_SYN_SENT: 819 err = sk->sk_prot->disconnect(sk, O_NONBLOCK); 820 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED; 821 break; 822 } 823 824 /* Wake up anyone sleeping in poll. */ 825 sk->sk_state_change(sk); 826 release_sock(sk); 827 return err; 828 } 829 EXPORT_SYMBOL(inet_shutdown); 830 831 /* 832 * ioctl() calls you can issue on an INET socket. Most of these are 833 * device configuration and stuff and very rarely used. Some ioctls 834 * pass on to the socket itself. 835 * 836 * NOTE: I like the idea of a module for the config stuff. ie ifconfig 837 * loads the devconfigure module does its configuring and unloads it. 838 * There's a good 20K of config code hanging around the kernel. 839 */ 840 841 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 842 { 843 struct sock *sk = sock->sk; 844 int err = 0; 845 struct net *net = sock_net(sk); 846 847 switch (cmd) { 848 case SIOCGSTAMP: 849 err = sock_get_timestamp(sk, (struct timeval __user *)arg); 850 break; 851 case SIOCGSTAMPNS: 852 err = sock_get_timestampns(sk, (struct timespec __user *)arg); 853 break; 854 case SIOCADDRT: 855 case SIOCDELRT: 856 case SIOCRTMSG: 857 err = ip_rt_ioctl(net, cmd, (void __user *)arg); 858 break; 859 case SIOCDARP: 860 case SIOCGARP: 861 case SIOCSARP: 862 err = arp_ioctl(net, cmd, (void __user *)arg); 863 break; 864 case SIOCGIFADDR: 865 case SIOCSIFADDR: 866 case SIOCGIFBRDADDR: 867 case SIOCSIFBRDADDR: 868 case SIOCGIFNETMASK: 869 case SIOCSIFNETMASK: 870 case SIOCGIFDSTADDR: 871 case SIOCSIFDSTADDR: 872 case SIOCSIFPFLAGS: 873 case SIOCGIFPFLAGS: 874 case SIOCSIFFLAGS: 875 err = devinet_ioctl(net, cmd, (void __user *)arg); 876 break; 877 default: 878 if (sk->sk_prot->ioctl) 879 err = sk->sk_prot->ioctl(sk, cmd, arg); 880 else 881 err = -ENOIOCTLCMD; 882 break; 883 } 884 return err; 885 } 886 EXPORT_SYMBOL(inet_ioctl); 887 888 #ifdef CONFIG_COMPAT 889 int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 890 { 891 struct sock *sk = sock->sk; 892 int err = -ENOIOCTLCMD; 893 894 if (sk->sk_prot->compat_ioctl) 895 err = sk->sk_prot->compat_ioctl(sk, cmd, arg); 896 897 return err; 898 } 899 #endif 900 901 const struct proto_ops inet_stream_ops = { 902 .family = PF_INET, 903 .owner = THIS_MODULE, 904 .release = inet_release, 905 .bind = inet_bind, 906 .connect = inet_stream_connect, 907 .socketpair = sock_no_socketpair, 908 .accept = inet_accept, 909 .getname = inet_getname, 910 .poll = tcp_poll, 911 .ioctl = inet_ioctl, 912 .listen = inet_listen, 913 .shutdown = inet_shutdown, 914 .setsockopt = sock_common_setsockopt, 915 .getsockopt = sock_common_getsockopt, 916 .sendmsg = inet_sendmsg, 917 .recvmsg = inet_recvmsg, 918 .mmap = sock_no_mmap, 919 .sendpage = inet_sendpage, 920 .splice_read = tcp_splice_read, 921 #ifdef CONFIG_COMPAT 922 .compat_setsockopt = compat_sock_common_setsockopt, 923 .compat_getsockopt = compat_sock_common_getsockopt, 924 .compat_ioctl = inet_compat_ioctl, 925 #endif 926 }; 927 EXPORT_SYMBOL(inet_stream_ops); 928 929 const struct proto_ops inet_dgram_ops = { 930 .family = PF_INET, 931 .owner = THIS_MODULE, 932 .release = inet_release, 933 .bind = inet_bind, 934 .connect = inet_dgram_connect, 935 .socketpair = sock_no_socketpair, 936 .accept = sock_no_accept, 937 .getname = inet_getname, 938 .poll = udp_poll, 939 .ioctl = inet_ioctl, 940 .listen = sock_no_listen, 941 .shutdown = inet_shutdown, 942 .setsockopt = sock_common_setsockopt, 943 .getsockopt = sock_common_getsockopt, 944 .sendmsg = inet_sendmsg, 945 .recvmsg = inet_recvmsg, 946 .mmap = sock_no_mmap, 947 .sendpage = inet_sendpage, 948 #ifdef CONFIG_COMPAT 949 .compat_setsockopt = compat_sock_common_setsockopt, 950 .compat_getsockopt = compat_sock_common_getsockopt, 951 .compat_ioctl = inet_compat_ioctl, 952 #endif 953 }; 954 EXPORT_SYMBOL(inet_dgram_ops); 955 956 /* 957 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without 958 * udp_poll 959 */ 960 static const struct proto_ops inet_sockraw_ops = { 961 .family = PF_INET, 962 .owner = THIS_MODULE, 963 .release = inet_release, 964 .bind = inet_bind, 965 .connect = inet_dgram_connect, 966 .socketpair = sock_no_socketpair, 967 .accept = sock_no_accept, 968 .getname = inet_getname, 969 .poll = datagram_poll, 970 .ioctl = inet_ioctl, 971 .listen = sock_no_listen, 972 .shutdown = inet_shutdown, 973 .setsockopt = sock_common_setsockopt, 974 .getsockopt = sock_common_getsockopt, 975 .sendmsg = inet_sendmsg, 976 .recvmsg = inet_recvmsg, 977 .mmap = sock_no_mmap, 978 .sendpage = inet_sendpage, 979 #ifdef CONFIG_COMPAT 980 .compat_setsockopt = compat_sock_common_setsockopt, 981 .compat_getsockopt = compat_sock_common_getsockopt, 982 .compat_ioctl = inet_compat_ioctl, 983 #endif 984 }; 985 986 static const struct net_proto_family inet_family_ops = { 987 .family = PF_INET, 988 .create = inet_create, 989 .owner = THIS_MODULE, 990 }; 991 992 /* Upon startup we insert all the elements in inetsw_array[] into 993 * the linked list inetsw. 994 */ 995 static struct inet_protosw inetsw_array[] = 996 { 997 { 998 .type = SOCK_STREAM, 999 .protocol = IPPROTO_TCP, 1000 .prot = &tcp_prot, 1001 .ops = &inet_stream_ops, 1002 .no_check = 0, 1003 .flags = INET_PROTOSW_PERMANENT | 1004 INET_PROTOSW_ICSK, 1005 }, 1006 1007 { 1008 .type = SOCK_DGRAM, 1009 .protocol = IPPROTO_UDP, 1010 .prot = &udp_prot, 1011 .ops = &inet_dgram_ops, 1012 .no_check = UDP_CSUM_DEFAULT, 1013 .flags = INET_PROTOSW_PERMANENT, 1014 }, 1015 1016 { 1017 .type = SOCK_DGRAM, 1018 .protocol = IPPROTO_ICMP, 1019 .prot = &ping_prot, 1020 .ops = &inet_dgram_ops, 1021 .no_check = UDP_CSUM_DEFAULT, 1022 .flags = INET_PROTOSW_REUSE, 1023 }, 1024 1025 { 1026 .type = SOCK_RAW, 1027 .protocol = IPPROTO_IP, /* wild card */ 1028 .prot = &raw_prot, 1029 .ops = &inet_sockraw_ops, 1030 .no_check = UDP_CSUM_DEFAULT, 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 answer = NULL; 1051 last_perm = &inetsw[p->type]; 1052 list_for_each(lh, &inetsw[p->type]) { 1053 answer = list_entry(lh, struct inet_protosw, list); 1054 1055 /* Check only the non-wild match. */ 1056 if (INET_PROTOSW_PERMANENT & answer->flags) { 1057 if (protocol == answer->protocol) 1058 break; 1059 last_perm = lh; 1060 } 1061 1062 answer = NULL; 1063 } 1064 if (answer) 1065 goto out_permanent; 1066 1067 /* Add the new entry after the last permanent entry if any, so that 1068 * the new entry does not override a permanent entry when matched with 1069 * a wild-card protocol. But it is allowed to override any existing 1070 * non-permanent entry. This means that when we remove this entry, the 1071 * system automatically returns to the old behavior. 1072 */ 1073 list_add_rcu(&p->list, last_perm); 1074 out: 1075 spin_unlock_bh(&inetsw_lock); 1076 1077 return; 1078 1079 out_permanent: 1080 printk(KERN_ERR "Attempt to override permanent protocol %d.\n", 1081 protocol); 1082 goto out; 1083 1084 out_illegal: 1085 printk(KERN_ERR 1086 "Ignoring attempt to register invalid socket type %d.\n", 1087 p->type); 1088 goto out; 1089 } 1090 EXPORT_SYMBOL(inet_register_protosw); 1091 1092 void inet_unregister_protosw(struct inet_protosw *p) 1093 { 1094 if (INET_PROTOSW_PERMANENT & p->flags) { 1095 printk(KERN_ERR 1096 "Attempt to unregister permanent protocol %d.\n", 1097 p->protocol); 1098 } else { 1099 spin_lock_bh(&inetsw_lock); 1100 list_del_rcu(&p->list); 1101 spin_unlock_bh(&inetsw_lock); 1102 1103 synchronize_net(); 1104 } 1105 } 1106 EXPORT_SYMBOL(inet_unregister_protosw); 1107 1108 /* 1109 * Shall we try to damage output packets if routing dev changes? 1110 */ 1111 1112 int sysctl_ip_dynaddr __read_mostly; 1113 1114 static int inet_sk_reselect_saddr(struct sock *sk) 1115 { 1116 struct inet_sock *inet = inet_sk(sk); 1117 __be32 old_saddr = inet->inet_saddr; 1118 __be32 daddr = inet->inet_daddr; 1119 struct flowi4 *fl4; 1120 struct rtable *rt; 1121 __be32 new_saddr; 1122 struct ip_options_rcu *inet_opt; 1123 1124 inet_opt = rcu_dereference_protected(inet->inet_opt, 1125 sock_owned_by_user(sk)); 1126 if (inet_opt && inet_opt->opt.srr) 1127 daddr = inet_opt->opt.faddr; 1128 1129 /* Query new route. */ 1130 fl4 = &inet->cork.fl.u.ip4; 1131 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk), 1132 sk->sk_bound_dev_if, sk->sk_protocol, 1133 inet->inet_sport, inet->inet_dport, sk, false); 1134 if (IS_ERR(rt)) 1135 return PTR_ERR(rt); 1136 1137 sk_setup_caps(sk, &rt->dst); 1138 1139 new_saddr = fl4->saddr; 1140 1141 if (new_saddr == old_saddr) 1142 return 0; 1143 1144 if (sysctl_ip_dynaddr > 1) { 1145 printk(KERN_INFO "%s(): shifting inet->saddr from %pI4 to %pI4\n", 1146 __func__, &old_saddr, &new_saddr); 1147 } 1148 1149 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr; 1150 1151 /* 1152 * XXX The only one ugly spot where we need to 1153 * XXX really change the sockets identity after 1154 * XXX it has entered the hashes. -DaveM 1155 * 1156 * Besides that, it does not check for connection 1157 * uniqueness. Wait for troubles. 1158 */ 1159 __sk_prot_rehash(sk); 1160 return 0; 1161 } 1162 1163 int inet_sk_rebuild_header(struct sock *sk) 1164 { 1165 struct inet_sock *inet = inet_sk(sk); 1166 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0); 1167 __be32 daddr; 1168 struct ip_options_rcu *inet_opt; 1169 struct flowi4 *fl4; 1170 int err; 1171 1172 /* Route is OK, nothing to do. */ 1173 if (rt) 1174 return 0; 1175 1176 /* Reroute. */ 1177 rcu_read_lock(); 1178 inet_opt = rcu_dereference(inet->inet_opt); 1179 daddr = inet->inet_daddr; 1180 if (inet_opt && inet_opt->opt.srr) 1181 daddr = inet_opt->opt.faddr; 1182 rcu_read_unlock(); 1183 fl4 = &inet->cork.fl.u.ip4; 1184 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr, 1185 inet->inet_dport, inet->inet_sport, 1186 sk->sk_protocol, RT_CONN_FLAGS(sk), 1187 sk->sk_bound_dev_if); 1188 if (!IS_ERR(rt)) { 1189 err = 0; 1190 sk_setup_caps(sk, &rt->dst); 1191 } else { 1192 err = PTR_ERR(rt); 1193 1194 /* Routing failed... */ 1195 sk->sk_route_caps = 0; 1196 /* 1197 * Other protocols have to map its equivalent state to TCP_SYN_SENT. 1198 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme 1199 */ 1200 if (!sysctl_ip_dynaddr || 1201 sk->sk_state != TCP_SYN_SENT || 1202 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) || 1203 (err = inet_sk_reselect_saddr(sk)) != 0) 1204 sk->sk_err_soft = -err; 1205 } 1206 1207 return err; 1208 } 1209 EXPORT_SYMBOL(inet_sk_rebuild_header); 1210 1211 static int inet_gso_send_check(struct sk_buff *skb) 1212 { 1213 const struct iphdr *iph; 1214 const struct net_protocol *ops; 1215 int proto; 1216 int ihl; 1217 int err = -EINVAL; 1218 1219 if (unlikely(!pskb_may_pull(skb, sizeof(*iph)))) 1220 goto out; 1221 1222 iph = ip_hdr(skb); 1223 ihl = iph->ihl * 4; 1224 if (ihl < sizeof(*iph)) 1225 goto out; 1226 1227 if (unlikely(!pskb_may_pull(skb, ihl))) 1228 goto out; 1229 1230 __skb_pull(skb, ihl); 1231 skb_reset_transport_header(skb); 1232 iph = ip_hdr(skb); 1233 proto = iph->protocol & (MAX_INET_PROTOS - 1); 1234 err = -EPROTONOSUPPORT; 1235 1236 rcu_read_lock(); 1237 ops = rcu_dereference(inet_protos[proto]); 1238 if (likely(ops && ops->gso_send_check)) 1239 err = ops->gso_send_check(skb); 1240 rcu_read_unlock(); 1241 1242 out: 1243 return err; 1244 } 1245 1246 static struct sk_buff *inet_gso_segment(struct sk_buff *skb, u32 features) 1247 { 1248 struct sk_buff *segs = ERR_PTR(-EINVAL); 1249 struct iphdr *iph; 1250 const struct net_protocol *ops; 1251 int proto; 1252 int ihl; 1253 int id; 1254 unsigned int offset = 0; 1255 1256 if (!(features & NETIF_F_V4_CSUM)) 1257 features &= ~NETIF_F_SG; 1258 1259 if (unlikely(skb_shinfo(skb)->gso_type & 1260 ~(SKB_GSO_TCPV4 | 1261 SKB_GSO_UDP | 1262 SKB_GSO_DODGY | 1263 SKB_GSO_TCP_ECN | 1264 0))) 1265 goto out; 1266 1267 if (unlikely(!pskb_may_pull(skb, sizeof(*iph)))) 1268 goto out; 1269 1270 iph = ip_hdr(skb); 1271 ihl = iph->ihl * 4; 1272 if (ihl < sizeof(*iph)) 1273 goto out; 1274 1275 if (unlikely(!pskb_may_pull(skb, ihl))) 1276 goto out; 1277 1278 __skb_pull(skb, ihl); 1279 skb_reset_transport_header(skb); 1280 iph = ip_hdr(skb); 1281 id = ntohs(iph->id); 1282 proto = iph->protocol & (MAX_INET_PROTOS - 1); 1283 segs = ERR_PTR(-EPROTONOSUPPORT); 1284 1285 rcu_read_lock(); 1286 ops = rcu_dereference(inet_protos[proto]); 1287 if (likely(ops && ops->gso_segment)) 1288 segs = ops->gso_segment(skb, features); 1289 rcu_read_unlock(); 1290 1291 if (!segs || IS_ERR(segs)) 1292 goto out; 1293 1294 skb = segs; 1295 do { 1296 iph = ip_hdr(skb); 1297 if (proto == IPPROTO_UDP) { 1298 iph->id = htons(id); 1299 iph->frag_off = htons(offset >> 3); 1300 if (skb->next != NULL) 1301 iph->frag_off |= htons(IP_MF); 1302 offset += (skb->len - skb->mac_len - iph->ihl * 4); 1303 } else 1304 iph->id = htons(id++); 1305 iph->tot_len = htons(skb->len - skb->mac_len); 1306 iph->check = 0; 1307 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl); 1308 } while ((skb = skb->next)); 1309 1310 out: 1311 return segs; 1312 } 1313 1314 static struct sk_buff **inet_gro_receive(struct sk_buff **head, 1315 struct sk_buff *skb) 1316 { 1317 const struct net_protocol *ops; 1318 struct sk_buff **pp = NULL; 1319 struct sk_buff *p; 1320 const struct iphdr *iph; 1321 unsigned int hlen; 1322 unsigned int off; 1323 unsigned int id; 1324 int flush = 1; 1325 int proto; 1326 1327 off = skb_gro_offset(skb); 1328 hlen = off + sizeof(*iph); 1329 iph = skb_gro_header_fast(skb, off); 1330 if (skb_gro_header_hard(skb, hlen)) { 1331 iph = skb_gro_header_slow(skb, hlen, off); 1332 if (unlikely(!iph)) 1333 goto out; 1334 } 1335 1336 proto = iph->protocol & (MAX_INET_PROTOS - 1); 1337 1338 rcu_read_lock(); 1339 ops = rcu_dereference(inet_protos[proto]); 1340 if (!ops || !ops->gro_receive) 1341 goto out_unlock; 1342 1343 if (*(u8 *)iph != 0x45) 1344 goto out_unlock; 1345 1346 if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl))) 1347 goto out_unlock; 1348 1349 id = ntohl(*(__be32 *)&iph->id); 1350 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF)); 1351 id >>= 16; 1352 1353 for (p = *head; p; p = p->next) { 1354 struct iphdr *iph2; 1355 1356 if (!NAPI_GRO_CB(p)->same_flow) 1357 continue; 1358 1359 iph2 = ip_hdr(p); 1360 1361 if ((iph->protocol ^ iph2->protocol) | 1362 (iph->tos ^ iph2->tos) | 1363 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) | 1364 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) { 1365 NAPI_GRO_CB(p)->same_flow = 0; 1366 continue; 1367 } 1368 1369 /* All fields must match except length and checksum. */ 1370 NAPI_GRO_CB(p)->flush |= 1371 (iph->ttl ^ iph2->ttl) | 1372 ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id); 1373 1374 NAPI_GRO_CB(p)->flush |= flush; 1375 } 1376 1377 NAPI_GRO_CB(skb)->flush |= flush; 1378 skb_gro_pull(skb, sizeof(*iph)); 1379 skb_set_transport_header(skb, skb_gro_offset(skb)); 1380 1381 pp = ops->gro_receive(head, skb); 1382 1383 out_unlock: 1384 rcu_read_unlock(); 1385 1386 out: 1387 NAPI_GRO_CB(skb)->flush |= flush; 1388 1389 return pp; 1390 } 1391 1392 static int inet_gro_complete(struct sk_buff *skb) 1393 { 1394 const struct net_protocol *ops; 1395 struct iphdr *iph = ip_hdr(skb); 1396 int proto = iph->protocol & (MAX_INET_PROTOS - 1); 1397 int err = -ENOSYS; 1398 __be16 newlen = htons(skb->len - skb_network_offset(skb)); 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_protos[proto]); 1405 if (WARN_ON(!ops || !ops->gro_complete)) 1406 goto out_unlock; 1407 1408 err = ops->gro_complete(skb); 1409 1410 out_unlock: 1411 rcu_read_unlock(); 1412 1413 return err; 1414 } 1415 1416 int inet_ctl_sock_create(struct sock **sk, unsigned short family, 1417 unsigned short type, unsigned char protocol, 1418 struct net *net) 1419 { 1420 struct socket *sock; 1421 int rc = sock_create_kern(family, type, protocol, &sock); 1422 1423 if (rc == 0) { 1424 *sk = sock->sk; 1425 (*sk)->sk_allocation = GFP_ATOMIC; 1426 /* 1427 * Unhash it so that IP input processing does not even see it, 1428 * we do not wish this socket to see incoming packets. 1429 */ 1430 (*sk)->sk_prot->unhash(*sk); 1431 1432 sk_change_net(*sk, net); 1433 } 1434 return rc; 1435 } 1436 EXPORT_SYMBOL_GPL(inet_ctl_sock_create); 1437 1438 unsigned long snmp_fold_field(void __percpu *mib[], int offt) 1439 { 1440 unsigned long res = 0; 1441 int i; 1442 1443 for_each_possible_cpu(i) { 1444 res += *(((unsigned long *) per_cpu_ptr(mib[0], i)) + offt); 1445 res += *(((unsigned long *) per_cpu_ptr(mib[1], i)) + offt); 1446 } 1447 return res; 1448 } 1449 EXPORT_SYMBOL_GPL(snmp_fold_field); 1450 1451 #if BITS_PER_LONG==32 1452 1453 u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset) 1454 { 1455 u64 res = 0; 1456 int cpu; 1457 1458 for_each_possible_cpu(cpu) { 1459 void *bhptr, *userptr; 1460 struct u64_stats_sync *syncp; 1461 u64 v_bh, v_user; 1462 unsigned int start; 1463 1464 /* first mib used by softirq context, we must use _bh() accessors */ 1465 bhptr = per_cpu_ptr(SNMP_STAT_BHPTR(mib), cpu); 1466 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset); 1467 do { 1468 start = u64_stats_fetch_begin_bh(syncp); 1469 v_bh = *(((u64 *) bhptr) + offt); 1470 } while (u64_stats_fetch_retry_bh(syncp, start)); 1471 1472 /* second mib used in USER context */ 1473 userptr = per_cpu_ptr(SNMP_STAT_USRPTR(mib), cpu); 1474 syncp = (struct u64_stats_sync *)(userptr + syncp_offset); 1475 do { 1476 start = u64_stats_fetch_begin(syncp); 1477 v_user = *(((u64 *) userptr) + offt); 1478 } while (u64_stats_fetch_retry(syncp, start)); 1479 1480 res += v_bh + v_user; 1481 } 1482 return res; 1483 } 1484 EXPORT_SYMBOL_GPL(snmp_fold_field64); 1485 #endif 1486 1487 int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align) 1488 { 1489 BUG_ON(ptr == NULL); 1490 ptr[0] = __alloc_percpu(mibsize, align); 1491 if (!ptr[0]) 1492 goto err0; 1493 ptr[1] = __alloc_percpu(mibsize, align); 1494 if (!ptr[1]) 1495 goto err1; 1496 return 0; 1497 err1: 1498 free_percpu(ptr[0]); 1499 ptr[0] = NULL; 1500 err0: 1501 return -ENOMEM; 1502 } 1503 EXPORT_SYMBOL_GPL(snmp_mib_init); 1504 1505 void snmp_mib_free(void __percpu *ptr[2]) 1506 { 1507 BUG_ON(ptr == NULL); 1508 free_percpu(ptr[0]); 1509 free_percpu(ptr[1]); 1510 ptr[0] = ptr[1] = NULL; 1511 } 1512 EXPORT_SYMBOL_GPL(snmp_mib_free); 1513 1514 #ifdef CONFIG_IP_MULTICAST 1515 static const struct net_protocol igmp_protocol = { 1516 .handler = igmp_rcv, 1517 .netns_ok = 1, 1518 }; 1519 #endif 1520 1521 static const struct net_protocol tcp_protocol = { 1522 .handler = tcp_v4_rcv, 1523 .err_handler = tcp_v4_err, 1524 .gso_send_check = tcp_v4_gso_send_check, 1525 .gso_segment = tcp_tso_segment, 1526 .gro_receive = tcp4_gro_receive, 1527 .gro_complete = tcp4_gro_complete, 1528 .no_policy = 1, 1529 .netns_ok = 1, 1530 }; 1531 1532 static const struct net_protocol udp_protocol = { 1533 .handler = udp_rcv, 1534 .err_handler = udp_err, 1535 .gso_send_check = udp4_ufo_send_check, 1536 .gso_segment = udp4_ufo_fragment, 1537 .no_policy = 1, 1538 .netns_ok = 1, 1539 }; 1540 1541 static const struct net_protocol icmp_protocol = { 1542 .handler = icmp_rcv, 1543 .err_handler = ping_err, 1544 .no_policy = 1, 1545 .netns_ok = 1, 1546 }; 1547 1548 static __net_init int ipv4_mib_init_net(struct net *net) 1549 { 1550 if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics, 1551 sizeof(struct tcp_mib), 1552 __alignof__(struct tcp_mib)) < 0) 1553 goto err_tcp_mib; 1554 if (snmp_mib_init((void __percpu **)net->mib.ip_statistics, 1555 sizeof(struct ipstats_mib), 1556 __alignof__(struct ipstats_mib)) < 0) 1557 goto err_ip_mib; 1558 if (snmp_mib_init((void __percpu **)net->mib.net_statistics, 1559 sizeof(struct linux_mib), 1560 __alignof__(struct linux_mib)) < 0) 1561 goto err_net_mib; 1562 if (snmp_mib_init((void __percpu **)net->mib.udp_statistics, 1563 sizeof(struct udp_mib), 1564 __alignof__(struct udp_mib)) < 0) 1565 goto err_udp_mib; 1566 if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics, 1567 sizeof(struct udp_mib), 1568 __alignof__(struct udp_mib)) < 0) 1569 goto err_udplite_mib; 1570 if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics, 1571 sizeof(struct icmp_mib), 1572 __alignof__(struct icmp_mib)) < 0) 1573 goto err_icmp_mib; 1574 if (snmp_mib_init((void __percpu **)net->mib.icmpmsg_statistics, 1575 sizeof(struct icmpmsg_mib), 1576 __alignof__(struct icmpmsg_mib)) < 0) 1577 goto err_icmpmsg_mib; 1578 1579 tcp_mib_init(net); 1580 return 0; 1581 1582 err_icmpmsg_mib: 1583 snmp_mib_free((void __percpu **)net->mib.icmp_statistics); 1584 err_icmp_mib: 1585 snmp_mib_free((void __percpu **)net->mib.udplite_statistics); 1586 err_udplite_mib: 1587 snmp_mib_free((void __percpu **)net->mib.udp_statistics); 1588 err_udp_mib: 1589 snmp_mib_free((void __percpu **)net->mib.net_statistics); 1590 err_net_mib: 1591 snmp_mib_free((void __percpu **)net->mib.ip_statistics); 1592 err_ip_mib: 1593 snmp_mib_free((void __percpu **)net->mib.tcp_statistics); 1594 err_tcp_mib: 1595 return -ENOMEM; 1596 } 1597 1598 static __net_exit void ipv4_mib_exit_net(struct net *net) 1599 { 1600 snmp_mib_free((void __percpu **)net->mib.icmpmsg_statistics); 1601 snmp_mib_free((void __percpu **)net->mib.icmp_statistics); 1602 snmp_mib_free((void __percpu **)net->mib.udplite_statistics); 1603 snmp_mib_free((void __percpu **)net->mib.udp_statistics); 1604 snmp_mib_free((void __percpu **)net->mib.net_statistics); 1605 snmp_mib_free((void __percpu **)net->mib.ip_statistics); 1606 snmp_mib_free((void __percpu **)net->mib.tcp_statistics); 1607 } 1608 1609 static __net_initdata struct pernet_operations ipv4_mib_ops = { 1610 .init = ipv4_mib_init_net, 1611 .exit = ipv4_mib_exit_net, 1612 }; 1613 1614 static int __init init_ipv4_mibs(void) 1615 { 1616 return register_pernet_subsys(&ipv4_mib_ops); 1617 } 1618 1619 static int ipv4_proc_init(void); 1620 1621 /* 1622 * IP protocol layer initialiser 1623 */ 1624 1625 static struct packet_type ip_packet_type __read_mostly = { 1626 .type = cpu_to_be16(ETH_P_IP), 1627 .func = ip_rcv, 1628 .gso_send_check = inet_gso_send_check, 1629 .gso_segment = inet_gso_segment, 1630 .gro_receive = inet_gro_receive, 1631 .gro_complete = inet_gro_complete, 1632 }; 1633 1634 static int __init inet_init(void) 1635 { 1636 struct sk_buff *dummy_skb; 1637 struct inet_protosw *q; 1638 struct list_head *r; 1639 int rc = -EINVAL; 1640 1641 BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb)); 1642 1643 sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL); 1644 if (!sysctl_local_reserved_ports) 1645 goto out; 1646 1647 rc = proto_register(&tcp_prot, 1); 1648 if (rc) 1649 goto out_free_reserved_ports; 1650 1651 rc = proto_register(&udp_prot, 1); 1652 if (rc) 1653 goto out_unregister_tcp_proto; 1654 1655 rc = proto_register(&raw_prot, 1); 1656 if (rc) 1657 goto out_unregister_udp_proto; 1658 1659 rc = proto_register(&ping_prot, 1); 1660 if (rc) 1661 goto out_unregister_raw_proto; 1662 1663 /* 1664 * Tell SOCKET that we are alive... 1665 */ 1666 1667 (void)sock_register(&inet_family_ops); 1668 1669 #ifdef CONFIG_SYSCTL 1670 ip_static_sysctl_init(); 1671 #endif 1672 1673 /* 1674 * Add all the base protocols. 1675 */ 1676 1677 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0) 1678 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n"); 1679 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0) 1680 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n"); 1681 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0) 1682 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n"); 1683 #ifdef CONFIG_IP_MULTICAST 1684 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0) 1685 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n"); 1686 #endif 1687 1688 /* Register the socket-side information for inet_create. */ 1689 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r) 1690 INIT_LIST_HEAD(r); 1691 1692 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q) 1693 inet_register_protosw(q); 1694 1695 /* 1696 * Set the ARP module up 1697 */ 1698 1699 arp_init(); 1700 1701 /* 1702 * Set the IP module up 1703 */ 1704 1705 ip_init(); 1706 1707 tcp_v4_init(); 1708 1709 /* Setup TCP slab cache for open requests. */ 1710 tcp_init(); 1711 1712 /* Setup UDP memory threshold */ 1713 udp_init(); 1714 1715 /* Add UDP-Lite (RFC 3828) */ 1716 udplite4_register(); 1717 1718 ping_init(); 1719 1720 /* 1721 * Set the ICMP layer up 1722 */ 1723 1724 if (icmp_init() < 0) 1725 panic("Failed to create the ICMP control socket.\n"); 1726 1727 /* 1728 * Initialise the multicast router 1729 */ 1730 #if defined(CONFIG_IP_MROUTE) 1731 if (ip_mr_init()) 1732 printk(KERN_CRIT "inet_init: Cannot init ipv4 mroute\n"); 1733 #endif 1734 /* 1735 * Initialise per-cpu ipv4 mibs 1736 */ 1737 1738 if (init_ipv4_mibs()) 1739 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n"); 1740 1741 ipv4_proc_init(); 1742 1743 ipfrag_init(); 1744 1745 dev_add_pack(&ip_packet_type); 1746 1747 rc = 0; 1748 out: 1749 return rc; 1750 out_unregister_raw_proto: 1751 proto_unregister(&raw_prot); 1752 out_unregister_udp_proto: 1753 proto_unregister(&udp_prot); 1754 out_unregister_tcp_proto: 1755 proto_unregister(&tcp_prot); 1756 out_free_reserved_ports: 1757 kfree(sysctl_local_reserved_ports); 1758 goto out; 1759 } 1760 1761 fs_initcall(inet_init); 1762 1763 /* ------------------------------------------------------------------------ */ 1764 1765 #ifdef CONFIG_PROC_FS 1766 static int __init ipv4_proc_init(void) 1767 { 1768 int rc = 0; 1769 1770 if (raw_proc_init()) 1771 goto out_raw; 1772 if (tcp4_proc_init()) 1773 goto out_tcp; 1774 if (udp4_proc_init()) 1775 goto out_udp; 1776 if (ping_proc_init()) 1777 goto out_ping; 1778 if (ip_misc_proc_init()) 1779 goto out_misc; 1780 out: 1781 return rc; 1782 out_misc: 1783 ping_proc_exit(); 1784 out_ping: 1785 udp4_proc_exit(); 1786 out_udp: 1787 tcp4_proc_exit(); 1788 out_tcp: 1789 raw_proc_exit(); 1790 out_raw: 1791 rc = -ENOMEM; 1792 goto out; 1793 } 1794 1795 #else /* CONFIG_PROC_FS */ 1796 static int __init ipv4_proc_init(void) 1797 { 1798 return 0; 1799 } 1800 #endif /* CONFIG_PROC_FS */ 1801 1802 MODULE_ALIAS_NETPROTO(PF_INET); 1803 1804