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 <linux/skbuff.h> 109#include <net/sock.h> 110#include <net/raw.h> 111#include <net/icmp.h> 112#include <net/ipip.h> 113#include <net/inet_common.h> 114#include <net/xfrm.h> 115#include <net/net_namespace.h> 116#ifdef CONFIG_IP_MROUTE 117#include <linux/mroute.h> 118#endif 119 120#include <typedefs.h> 121#include <bcmdefs.h> 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 129struct ipv4_config ipv4_config; 130EXPORT_SYMBOL(ipv4_config); 131 132/* New destruction routine */ 133 134void inet_sock_destruct(struct sock *sk) 135{ 136 struct inet_sock *inet = inet_sk(sk); 137 138 __skb_queue_purge(&sk->sk_receive_queue); 139 __skb_queue_purge(&sk->sk_error_queue); 140 141 sk_mem_reclaim(sk); 142 143 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) { 144 pr_err("Attempt to release TCP socket in state %d %p\n", 145 sk->sk_state, sk); 146 return; 147 } 148 if (!sock_flag(sk, SOCK_DEAD)) { 149 pr_err("Attempt to release alive inet socket %p\n", sk); 150 return; 151 } 152 153 WARN_ON(atomic_read(&sk->sk_rmem_alloc)); 154 WARN_ON(atomic_read(&sk->sk_wmem_alloc)); 155 WARN_ON(sk->sk_wmem_queued); 156 WARN_ON(sk->sk_forward_alloc); 157 158 kfree(inet->opt); 159 dst_release(rcu_dereference_check(sk->sk_dst_cache, 1)); 160 sk_refcnt_debug_dec(sk); 161} 162EXPORT_SYMBOL(inet_sock_destruct); 163 164/* 165 * The routines beyond this point handle the behaviour of an AF_INET 166 * socket object. Mostly it punts to the subprotocols of IP to do 167 * the work. 168 */ 169 170/* 171 * Automatically bind an unbound socket. 172 */ 173 174static int inet_autobind(struct sock *sk) 175{ 176 struct inet_sock *inet; 177 /* We may need to bind the socket. */ 178 lock_sock(sk); 179 inet = inet_sk(sk); 180 if (!inet->inet_num) { 181 if (sk->sk_prot->get_port(sk, 0)) { 182 release_sock(sk); 183 return -EAGAIN; 184 } 185 inet->inet_sport = htons(inet->inet_num); 186 } 187 release_sock(sk); 188 return 0; 189} 190 191/* 192 * Move a socket into listening state. 193 */ 194int inet_listen(struct socket *sock, int backlog) 195{ 196 struct sock *sk = sock->sk; 197 unsigned char old_state; 198 int err; 199 200 lock_sock(sk); 201 202 err = -EINVAL; 203 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM) 204 goto out; 205 206 old_state = sk->sk_state; 207 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN))) 208 goto out; 209 210 /* Really, if the socket is already in listen state 211 * we can only allow the backlog to be adjusted. 212 */ 213 if (old_state != TCP_LISTEN) { 214 err = inet_csk_listen_start(sk, backlog); 215 if (err) 216 goto out; 217 } 218 sk->sk_max_ack_backlog = backlog; 219 err = 0; 220 221out: 222 release_sock(sk); 223 return err; 224} 225EXPORT_SYMBOL(inet_listen); 226 227u32 inet_ehash_secret __read_mostly; 228EXPORT_SYMBOL(inet_ehash_secret); 229 230/* 231 * inet_ehash_secret must be set exactly once 232 * Instead of using a dedicated spinlock, we (ab)use inetsw_lock 233 */ 234void build_ehash_secret(void) 235{ 236 u32 rnd; 237 do { 238 get_random_bytes(&rnd, sizeof(rnd)); 239 } while (rnd == 0); 240 spin_lock_bh(&inetsw_lock); 241 if (!inet_ehash_secret) 242 inet_ehash_secret = rnd; 243 spin_unlock_bh(&inetsw_lock); 244} 245EXPORT_SYMBOL(build_ehash_secret); 246 247static inline int inet_netns_ok(struct net *net, int protocol) 248{ 249 int hash; 250 const struct net_protocol *ipprot; 251 252 if (net_eq(net, &init_net)) 253 return 1; 254 255 hash = protocol & (MAX_INET_PROTOS - 1); 256 ipprot = rcu_dereference(inet_protos[hash]); 257 258 if (ipprot == NULL) 259 /* raw IP is OK */ 260 return 1; 261 return ipprot->netns_ok; 262} 263 264/* 265 * Create an inet socket. 266 */ 267 268static int inet_create(struct net *net, struct socket *sock, int protocol, 269 int kern) 270{ 271 struct sock *sk; 272 struct inet_protosw *answer; 273 struct inet_sock *inet; 274 struct proto *answer_prot; 275 unsigned char answer_flags; 276 char answer_no_check; 277 int try_loading_module = 0; 278 int err; 279 280 if (unlikely(!inet_ehash_secret)) 281 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM) 282 build_ehash_secret(); 283 284 sock->state = SS_UNCONNECTED; 285 286 /* Look for the requested type/protocol pair. */ 287lookup_protocol: 288 err = -ESOCKTNOSUPPORT; 289 rcu_read_lock(); 290 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) { 291 292 err = 0; 293 /* Check the non-wild match. */ 294 if (protocol == answer->protocol) { 295 if (protocol != IPPROTO_IP) 296 break; 297 } else { 298 /* Check for the two wild cases. */ 299 if (IPPROTO_IP == protocol) { 300 protocol = answer->protocol; 301 break; 302 } 303 if (IPPROTO_IP == answer->protocol) 304 break; 305 } 306 err = -EPROTONOSUPPORT; 307 } 308 309 if (unlikely(err)) { 310 if (try_loading_module < 2) { 311 rcu_read_unlock(); 312 /* 313 * Be more specific, e.g. net-pf-2-proto-132-type-1 314 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM) 315 */ 316 if (++try_loading_module == 1) 317 request_module("net-pf-%d-proto-%d-type-%d", 318 PF_INET, protocol, sock->type); 319 /* 320 * Fall back to generic, e.g. net-pf-2-proto-132 321 * (net-pf-PF_INET-proto-IPPROTO_SCTP) 322 */ 323 else 324 request_module("net-pf-%d-proto-%d", 325 PF_INET, protocol); 326 goto lookup_protocol; 327 } else 328 goto out_rcu_unlock; 329 } 330 331 err = -EPERM; 332 if (sock->type == SOCK_RAW && !kern && !capable(CAP_NET_RAW)) 333 goto out_rcu_unlock; 334 335 err = -EAFNOSUPPORT; 336 if (!inet_netns_ok(net, protocol)) 337 goto out_rcu_unlock; 338 339 sock->ops = answer->ops; 340 answer_prot = answer->prot; 341 answer_no_check = answer->no_check; 342 answer_flags = answer->flags; 343 rcu_read_unlock(); 344 345 WARN_ON(answer_prot->slab == NULL); 346 347 err = -ENOBUFS; 348 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot); 349 if (sk == NULL) 350 goto out; 351 352 err = 0; 353 sk->sk_no_check = answer_no_check; 354 if (INET_PROTOSW_REUSE & answer_flags) 355 sk->sk_reuse = 1; 356 357 inet = inet_sk(sk); 358 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0; 359 360 inet->nodefrag = 0; 361 362 if (SOCK_RAW == sock->type) { 363 inet->inet_num = protocol; 364 if (IPPROTO_RAW == protocol) 365 inet->hdrincl = 1; 366 } 367 368 if (ipv4_config.no_pmtu_disc) 369 inet->pmtudisc = IP_PMTUDISC_DONT; 370 else 371 inet->pmtudisc = IP_PMTUDISC_WANT; 372 373 inet->inet_id = 0; 374 375 sock_init_data(sock, sk); 376 377 sk->sk_destruct = inet_sock_destruct; 378 sk->sk_protocol = protocol; 379 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv; 380 381 inet->uc_ttl = -1; 382 inet->mc_loop = 1; 383 inet->mc_ttl = 1; 384 inet->mc_all = 1; 385 inet->mc_index = 0; 386 inet->mc_list = NULL; 387 388 sk_refcnt_debug_inc(sk); 389 390 if (inet->inet_num) { 391 /* It assumes that any protocol which allows 392 * the user to assign a number at socket 393 * creation time automatically 394 * shares. 395 */ 396 inet->inet_sport = htons(inet->inet_num); 397 /* Add to protocol hash chains. */ 398 sk->sk_prot->hash(sk); 399 } 400 401 if (sk->sk_prot->init) { 402 err = sk->sk_prot->init(sk); 403 if (err) 404 sk_common_release(sk); 405 } 406out: 407 return err; 408out_rcu_unlock: 409 rcu_read_unlock(); 410 goto out; 411} 412 413 414/* 415 * The peer socket should always be NULL (or else). When we call this 416 * function we are destroying the object and from then on nobody 417 * should refer to it. 418 */ 419int inet_release(struct socket *sock) 420{ 421 struct sock *sk = sock->sk; 422 423 if (sk) { 424 long timeout; 425 426 sock_rps_reset_flow(sk); 427 428 /* Applications forget to leave groups before exiting */ 429 ip_mc_drop_socket(sk); 430 431 /* If linger is set, we don't return until the close 432 * is complete. Otherwise we return immediately. The 433 * actually closing is done the same either way. 434 * 435 * If the close is due to the process exiting, we never 436 * linger.. 437 */ 438 timeout = 0; 439 if (sock_flag(sk, SOCK_LINGER) && 440 !(current->flags & PF_EXITING)) 441 timeout = sk->sk_lingertime; 442 sock->sk = NULL; 443 sk->sk_prot->close(sk, timeout); 444 } 445 return 0; 446} 447EXPORT_SYMBOL(inet_release); 448 449/* It is off by default, see below. */ 450int sysctl_ip_nonlocal_bind __read_mostly; 451EXPORT_SYMBOL(sysctl_ip_nonlocal_bind); 452 453int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) 454{ 455 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr; 456 struct sock *sk = sock->sk; 457 struct inet_sock *inet = inet_sk(sk); 458 unsigned short snum; 459 int chk_addr_ret; 460 int err; 461 462 /* If the socket has its own bind function then use it. (RAW) */ 463 if (sk->sk_prot->bind) { 464 err = sk->sk_prot->bind(sk, uaddr, addr_len); 465 goto out; 466 } 467 err = -EINVAL; 468 if (addr_len < sizeof(struct sockaddr_in)) 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; 528out_release_sock: 529 release_sock(sk); 530out: 531 return err; 532} 533EXPORT_SYMBOL(inet_bind); 534 535int 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} 549EXPORT_SYMBOL(inet_dgram_connect); 550 551static 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 */ 578int 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; 651out: 652 release_sock(sk); 653 return err; 654 655sock_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} 662EXPORT_SYMBOL(inet_stream_connect); 663 664/* 665 * Accept a pending connection. The TCP layer now gives BSD semantics. 666 */ 667 668int 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 WARN_ON(!((1 << sk2->sk_state) & 680 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE))); 681 682 sock_graft(sk2, newsock); 683 684 newsock->state = SS_CONNECTED; 685 err = 0; 686 release_sock(sk2); 687do_err: 688 return err; 689} 690EXPORT_SYMBOL(inet_accept); 691 692 693/* 694 * This does both peername and sockname. 695 */ 696int inet_getname(struct socket *sock, struct sockaddr *uaddr, 697 int *uaddr_len, int peer) 698{ 699 struct sock *sk = sock->sk; 700 struct inet_sock *inet = inet_sk(sk); 701 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr); 702 703 sin->sin_family = AF_INET; 704 if (peer) { 705 if (!inet->inet_dport || 706 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) && 707 peer == 1)) 708 return -ENOTCONN; 709 sin->sin_port = inet->inet_dport; 710 sin->sin_addr.s_addr = inet->inet_daddr; 711 } else { 712 __be32 addr = inet->inet_rcv_saddr; 713 if (!addr) 714 addr = inet->inet_saddr; 715 sin->sin_port = inet->inet_sport; 716 sin->sin_addr.s_addr = addr; 717 } 718 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 719 *uaddr_len = sizeof(*sin); 720 return 0; 721} 722EXPORT_SYMBOL(inet_getname); 723 724int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg, 725 size_t size) 726{ 727 struct sock *sk = sock->sk; 728 729 sock_rps_record_flow(sk); 730 731 /* We may need to bind the socket. */ 732 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind && 733 inet_autobind(sk)) 734 return -EAGAIN; 735 736 return sk->sk_prot->sendmsg(iocb, sk, msg, size); 737} 738EXPORT_SYMBOL(inet_sendmsg); 739 740ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, 741 size_t size, int flags) 742{ 743 struct sock *sk = sock->sk; 744 745 sock_rps_record_flow(sk); 746 747 /* We may need to bind the socket. */ 748 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind && 749 inet_autobind(sk)) 750 return -EAGAIN; 751 752 if (sk->sk_prot->sendpage) 753 return sk->sk_prot->sendpage(sk, page, offset, size, flags); 754 return sock_no_sendpage(sock, page, offset, size, flags); 755} 756EXPORT_SYMBOL(inet_sendpage); 757 758int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg, 759 size_t size, int flags) 760{ 761 struct sock *sk = sock->sk; 762 int addr_len = 0; 763 int err; 764 765 sock_rps_record_flow(sk); 766 767 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT, 768 flags & ~MSG_DONTWAIT, &addr_len); 769 if (err >= 0) 770 msg->msg_namelen = addr_len; 771 return err; 772} 773EXPORT_SYMBOL(inet_recvmsg); 774 775int inet_shutdown(struct socket *sock, int how) 776{ 777 struct sock *sk = sock->sk; 778 int err = 0; 779 780 /* This should really check to make sure 781 * the socket is a TCP socket. (WHY AC...) 782 */ 783 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and 784 1->2 bit 2 snds. 785 2->3 */ 786 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */ 787 return -EINVAL; 788 789 lock_sock(sk); 790 if (sock->state == SS_CONNECTING) { 791 if ((1 << sk->sk_state) & 792 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE)) 793 sock->state = SS_DISCONNECTING; 794 else 795 sock->state = SS_CONNECTED; 796 } 797 798 switch (sk->sk_state) { 799 case TCP_CLOSE: 800 err = -ENOTCONN; 801 /* Hack to wake up other listeners, who can poll for 802 POLLHUP, even on eg. unconnected UDP sockets -- RR */ 803 default: 804 sk->sk_shutdown |= how; 805 if (sk->sk_prot->shutdown) 806 sk->sk_prot->shutdown(sk, how); 807 break; 808 809 /* Remaining two branches are temporary solution for missing 810 * close() in multithreaded environment. It is _not_ a good idea, 811 * but we have no choice until close() is repaired at VFS level. 812 */ 813 case TCP_LISTEN: 814 if (!(how & RCV_SHUTDOWN)) 815 break; 816 /* Fall through */ 817 case TCP_SYN_SENT: 818 err = sk->sk_prot->disconnect(sk, O_NONBLOCK); 819 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED; 820 break; 821 } 822 823 /* Wake up anyone sleeping in poll. */ 824 sk->sk_state_change(sk); 825 release_sock(sk); 826 return err; 827} 828EXPORT_SYMBOL(inet_shutdown); 829 830/* 831 * ioctl() calls you can issue on an INET socket. Most of these are 832 * device configuration and stuff and very rarely used. Some ioctls 833 * pass on to the socket itself. 834 * 835 * NOTE: I like the idea of a module for the config stuff. ie ifconfig 836 * loads the devconfigure module does its configuring and unloads it. 837 * There's a good 20K of config code hanging around the kernel. 838 */ 839 840int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 841{ 842 struct sock *sk = sock->sk; 843 int err = 0; 844 struct net *net = sock_net(sk); 845 846 switch (cmd) { 847 case SIOCGSTAMP: 848 err = sock_get_timestamp(sk, (struct timeval __user *)arg); 849 break; 850 case SIOCGSTAMPNS: 851 err = sock_get_timestampns(sk, (struct timespec __user *)arg); 852 break; 853 case SIOCADDRT: 854 case SIOCDELRT: 855 case SIOCRTMSG: 856 err = ip_rt_ioctl(net, cmd, (void __user *)arg); 857 break; 858 case SIOCDARP: 859 case SIOCGARP: 860 case SIOCSARP: 861 case SIOCPIDARP: //Foxconn tab tsng add, 2013/05/23 ,ioctl pid to arp 862 err = arp_ioctl(net, cmd, (void __user *)arg); 863 break; 864/*foxconn add start,edward zhang, 2012/11/16 @arp protection*/ 865#ifdef ARP_PROTECTION 866 case SIOCREJARP: 867 return arp_ioctl(net, cmd, (void __user *)arg); 868#endif 869/*foxconn add end,edward zhang, 2012/11/16 @arp protection*/ 870 case SIOCGIFADDR: 871 case SIOCSIFADDR: 872 case SIOCGIFBRDADDR: 873 case SIOCSIFBRDADDR: 874 case SIOCGIFNETMASK: 875 case SIOCSIFNETMASK: 876 case SIOCGIFDSTADDR: 877 case SIOCSIFDSTADDR: 878 case SIOCSIFPFLAGS: 879 case SIOCGIFPFLAGS: 880 case SIOCSIFFLAGS: 881 err = devinet_ioctl(net, cmd, (void __user *)arg); 882 break; 883 default: 884 if (sk->sk_prot->ioctl) 885 err = sk->sk_prot->ioctl(sk, cmd, arg); 886 else 887 err = -ENOIOCTLCMD; 888 break; 889 } 890 return err; 891} 892EXPORT_SYMBOL(inet_ioctl); 893 894const struct proto_ops inet_stream_ops = { 895 .family = PF_INET, 896 .owner = THIS_MODULE, 897 .release = inet_release, 898 .bind = inet_bind, 899 .connect = inet_stream_connect, 900 .socketpair = sock_no_socketpair, 901 .accept = inet_accept, 902 .getname = inet_getname, 903 .poll = tcp_poll, 904 .ioctl = inet_ioctl, 905 .listen = inet_listen, 906 .shutdown = inet_shutdown, 907 .setsockopt = sock_common_setsockopt, 908 .getsockopt = sock_common_getsockopt, 909 .sendmsg = inet_sendmsg, 910 .recvmsg = inet_recvmsg, 911 .mmap = sock_no_mmap, 912 .sendpage = inet_sendpage, 913 .splice_read = tcp_splice_read, 914#ifdef CONFIG_COMPAT 915 .compat_setsockopt = compat_sock_common_setsockopt, 916 .compat_getsockopt = compat_sock_common_getsockopt, 917#endif 918}; 919EXPORT_SYMBOL(inet_stream_ops); 920 921const struct proto_ops inet_dgram_ops = { 922 .family = PF_INET, 923 .owner = THIS_MODULE, 924 .release = inet_release, 925 .bind = inet_bind, 926 .connect = inet_dgram_connect, 927 .socketpair = sock_no_socketpair, 928 .accept = sock_no_accept, 929 .getname = inet_getname, 930 .poll = udp_poll, 931 .ioctl = inet_ioctl, 932 .listen = sock_no_listen, 933 .shutdown = inet_shutdown, 934 .setsockopt = sock_common_setsockopt, 935 .getsockopt = sock_common_getsockopt, 936 .sendmsg = inet_sendmsg, 937 .recvmsg = inet_recvmsg, 938 .mmap = sock_no_mmap, 939 .sendpage = inet_sendpage, 940#ifdef CONFIG_COMPAT 941 .compat_setsockopt = compat_sock_common_setsockopt, 942 .compat_getsockopt = compat_sock_common_getsockopt, 943#endif 944}; 945EXPORT_SYMBOL(inet_dgram_ops); 946 947/* 948 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without 949 * udp_poll 950 */ 951static const struct proto_ops inet_sockraw_ops = { 952 .family = PF_INET, 953 .owner = THIS_MODULE, 954 .release = inet_release, 955 .bind = inet_bind, 956 .connect = inet_dgram_connect, 957 .socketpair = sock_no_socketpair, 958 .accept = sock_no_accept, 959 .getname = inet_getname, 960 .poll = datagram_poll, 961 .ioctl = inet_ioctl, 962 .listen = sock_no_listen, 963 .shutdown = inet_shutdown, 964 .setsockopt = sock_common_setsockopt, 965 .getsockopt = sock_common_getsockopt, 966 .sendmsg = inet_sendmsg, 967 .recvmsg = inet_recvmsg, 968 .mmap = sock_no_mmap, 969 .sendpage = inet_sendpage, 970#ifdef CONFIG_COMPAT 971 .compat_setsockopt = compat_sock_common_setsockopt, 972 .compat_getsockopt = compat_sock_common_getsockopt, 973#endif 974}; 975 976static const struct net_proto_family inet_family_ops = { 977 .family = PF_INET, 978 .create = inet_create, 979 .owner = THIS_MODULE, 980}; 981 982/* Upon startup we insert all the elements in inetsw_array[] into 983 * the linked list inetsw. 984 */ 985static struct inet_protosw inetsw_array[] = 986{ 987 { 988 .type = SOCK_STREAM, 989 .protocol = IPPROTO_TCP, 990 .prot = &tcp_prot, 991 .ops = &inet_stream_ops, 992 .no_check = 0, 993 .flags = INET_PROTOSW_PERMANENT | 994 INET_PROTOSW_ICSK, 995 }, 996 997 { 998 .type = SOCK_DGRAM, 999 .protocol = IPPROTO_UDP, 1000 .prot = &udp_prot, 1001 .ops = &inet_dgram_ops, 1002 .no_check = UDP_CSUM_DEFAULT, 1003 .flags = INET_PROTOSW_PERMANENT, 1004 }, 1005 1006 1007 { 1008 .type = SOCK_RAW, 1009 .protocol = IPPROTO_IP, /* wild card */ 1010 .prot = &raw_prot, 1011 .ops = &inet_sockraw_ops, 1012 .no_check = UDP_CSUM_DEFAULT, 1013 .flags = INET_PROTOSW_REUSE, 1014 } 1015}; 1016 1017#define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array) 1018 1019void inet_register_protosw(struct inet_protosw *p) 1020{ 1021 struct list_head *lh; 1022 struct inet_protosw *answer; 1023 int protocol = p->protocol; 1024 struct list_head *last_perm; 1025 1026 spin_lock_bh(&inetsw_lock); 1027 1028 if (p->type >= SOCK_MAX) 1029 goto out_illegal; 1030 1031 /* If we are trying to override a permanent protocol, bail. */ 1032 answer = NULL; 1033 last_perm = &inetsw[p->type]; 1034 list_for_each(lh, &inetsw[p->type]) { 1035 answer = list_entry(lh, struct inet_protosw, list); 1036 1037 /* Check only the non-wild match. */ 1038 if (INET_PROTOSW_PERMANENT & answer->flags) { 1039 if (protocol == answer->protocol) 1040 break; 1041 last_perm = lh; 1042 } 1043 1044 answer = NULL; 1045 } 1046 if (answer) 1047 goto out_permanent; 1048 1049 /* Add the new entry after the last permanent entry if any, so that 1050 * the new entry does not override a permanent entry when matched with 1051 * a wild-card protocol. But it is allowed to override any existing 1052 * non-permanent entry. This means that when we remove this entry, the 1053 * system automatically returns to the old behavior. 1054 */ 1055 list_add_rcu(&p->list, last_perm); 1056out: 1057 spin_unlock_bh(&inetsw_lock); 1058 1059 return; 1060 1061out_permanent: 1062 printk(KERN_ERR "Attempt to override permanent protocol %d.\n", 1063 protocol); 1064 goto out; 1065 1066out_illegal: 1067 printk(KERN_ERR 1068 "Ignoring attempt to register invalid socket type %d.\n", 1069 p->type); 1070 goto out; 1071} 1072EXPORT_SYMBOL(inet_register_protosw); 1073 1074void inet_unregister_protosw(struct inet_protosw *p) 1075{ 1076 if (INET_PROTOSW_PERMANENT & p->flags) { 1077 printk(KERN_ERR 1078 "Attempt to unregister permanent protocol %d.\n", 1079 p->protocol); 1080 } else { 1081 spin_lock_bh(&inetsw_lock); 1082 list_del_rcu(&p->list); 1083 spin_unlock_bh(&inetsw_lock); 1084 1085 synchronize_net(); 1086 } 1087} 1088EXPORT_SYMBOL(inet_unregister_protosw); 1089 1090/* 1091 * Shall we try to damage output packets if routing dev changes? 1092 */ 1093 1094int sysctl_ip_dynaddr __read_mostly; 1095 1096static int inet_sk_reselect_saddr(struct sock *sk) 1097{ 1098 struct inet_sock *inet = inet_sk(sk); 1099 int err; 1100 struct rtable *rt; 1101 __be32 old_saddr = inet->inet_saddr; 1102 __be32 new_saddr; 1103 __be32 daddr = inet->inet_daddr; 1104 1105 if (inet->opt && inet->opt->srr) 1106 daddr = inet->opt->faddr; 1107 1108 /* Query new route. */ 1109 err = ip_route_connect(&rt, daddr, 0, 1110 RT_CONN_FLAGS(sk), 1111 sk->sk_bound_dev_if, 1112 sk->sk_protocol, 1113 inet->inet_sport, inet->inet_dport, sk, 0); 1114 if (err) 1115 return err; 1116 1117 sk_setup_caps(sk, &rt->dst); 1118 1119 new_saddr = rt->rt_src; 1120 1121 if (new_saddr == old_saddr) 1122 return 0; 1123 1124 if (sysctl_ip_dynaddr > 1) { 1125 printk(KERN_INFO "%s(): shifting inet->saddr from %pI4 to %pI4\n", 1126 __func__, &old_saddr, &new_saddr); 1127 } 1128 1129 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr; 1130 1131 __sk_prot_rehash(sk); 1132 return 0; 1133} 1134 1135int inet_sk_rebuild_header(struct sock *sk) 1136{ 1137 struct inet_sock *inet = inet_sk(sk); 1138 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0); 1139 __be32 daddr; 1140 int err; 1141 1142 /* Route is OK, nothing to do. */ 1143 if (rt) 1144 return 0; 1145 1146 /* Reroute. */ 1147 daddr = inet->inet_daddr; 1148 if (inet->opt && inet->opt->srr) 1149 daddr = inet->opt->faddr; 1150{ 1151 struct flowi fl = { 1152 .oif = sk->sk_bound_dev_if, 1153 .mark = sk->sk_mark, 1154 .nl_u = { 1155 .ip4_u = { 1156 .daddr = daddr, 1157 .saddr = inet->inet_saddr, 1158 .tos = RT_CONN_FLAGS(sk), 1159 }, 1160 }, 1161 .proto = sk->sk_protocol, 1162 .flags = inet_sk_flowi_flags(sk), 1163 .uli_u = { 1164 .ports = { 1165 .sport = inet->inet_sport, 1166 .dport = inet->inet_dport, 1167 }, 1168 }, 1169 }; 1170 1171 security_sk_classify_flow(sk, &fl); 1172 err = ip_route_output_flow(sock_net(sk), &rt, &fl, sk, 0); 1173} 1174 if (!err) 1175 sk_setup_caps(sk, &rt->dst); 1176 else { 1177 /* Routing failed... */ 1178 sk->sk_route_caps = 0; 1179 /* 1180 * Other protocols have to map its equivalent state to TCP_SYN_SENT. 1181 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme 1182 */ 1183 if (!sysctl_ip_dynaddr || 1184 sk->sk_state != TCP_SYN_SENT || 1185 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) || 1186 (err = inet_sk_reselect_saddr(sk)) != 0) 1187 sk->sk_err_soft = -err; 1188 } 1189 1190 return err; 1191} 1192EXPORT_SYMBOL(inet_sk_rebuild_header); 1193 1194static int inet_gso_send_check(struct sk_buff *skb) 1195{ 1196 struct iphdr *iph; 1197 const struct net_protocol *ops; 1198 int proto; 1199 int ihl; 1200 int err = -EINVAL; 1201 1202 if (unlikely(!pskb_may_pull(skb, sizeof(*iph)))) 1203 goto out; 1204 1205 iph = ip_hdr(skb); 1206 ihl = iph->ihl * 4; 1207 if (ihl < sizeof(*iph)) 1208 goto out; 1209 1210 if (unlikely(!pskb_may_pull(skb, ihl))) 1211 goto out; 1212 1213 __skb_pull(skb, ihl); 1214 skb_reset_transport_header(skb); 1215 iph = ip_hdr(skb); 1216 proto = iph->protocol & (MAX_INET_PROTOS - 1); 1217 err = -EPROTONOSUPPORT; 1218 1219 rcu_read_lock(); 1220 ops = rcu_dereference(inet_protos[proto]); 1221 if (likely(ops && ops->gso_send_check)) 1222 err = ops->gso_send_check(skb); 1223 rcu_read_unlock(); 1224 1225out: 1226 return err; 1227} 1228 1229static struct sk_buff * BCMFASTPATH_HOST inet_gso_segment(struct sk_buff *skb, int features) 1230{ 1231 struct sk_buff *segs = ERR_PTR(-EINVAL); 1232 struct iphdr *iph; 1233 const struct net_protocol *ops; 1234 int proto; 1235 int ihl; 1236 int id; 1237 unsigned int offset = 0; 1238 1239 if (!(features & NETIF_F_V4_CSUM)) 1240 features &= ~NETIF_F_SG; 1241 1242 if (unlikely(skb_shinfo(skb)->gso_type & 1243 ~(SKB_GSO_TCPV4 | 1244 SKB_GSO_UDP | 1245 SKB_GSO_DODGY | 1246 SKB_GSO_TCP_ECN | 1247 0))) 1248 goto out; 1249 1250 if (unlikely(!pskb_may_pull(skb, sizeof(*iph)))) 1251 goto out; 1252 1253 iph = ip_hdr(skb); 1254 ihl = iph->ihl * 4; 1255 if (ihl < sizeof(*iph)) 1256 goto out; 1257 1258 if (unlikely(!pskb_may_pull(skb, ihl))) 1259 goto out; 1260 1261 __skb_pull(skb, ihl); 1262 skb_reset_transport_header(skb); 1263 iph = ip_hdr(skb); 1264 id = ntohs(iph->id); 1265 proto = iph->protocol & (MAX_INET_PROTOS - 1); 1266 segs = ERR_PTR(-EPROTONOSUPPORT); 1267 1268 rcu_read_lock(); 1269 ops = rcu_dereference(inet_protos[proto]); 1270 if (likely(ops && ops->gso_segment)) 1271 segs = ops->gso_segment(skb, features); 1272 rcu_read_unlock(); 1273 1274 if (!segs || IS_ERR(segs)) 1275 goto out; 1276 1277 skb = segs; 1278 do { 1279 iph = ip_hdr(skb); 1280 if (proto == IPPROTO_UDP) { 1281 iph->id = htons(id); 1282 iph->frag_off = htons(offset >> 3); 1283 if (skb->next != NULL) 1284 iph->frag_off |= htons(IP_MF); 1285 offset += (skb->len - skb->mac_len - iph->ihl * 4); 1286 } else 1287 iph->id = htons(id++); 1288 iph->tot_len = htons(skb->len - skb->mac_len); 1289 iph->check = 0; 1290 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl); 1291 } while ((skb = skb->next)); 1292 1293out: 1294 return segs; 1295} 1296 1297static struct sk_buff ** BCMFASTPATH_HOST inet_gro_receive(struct sk_buff **head, 1298 struct sk_buff *skb) 1299{ 1300 const struct net_protocol *ops; 1301 struct sk_buff **pp = NULL; 1302 struct sk_buff *p; 1303 struct iphdr *iph; 1304 unsigned int hlen; 1305 unsigned int off; 1306 unsigned int id; 1307 int flush = 1; 1308 int proto; 1309 1310 off = skb_gro_offset(skb); 1311 hlen = off + sizeof(*iph); 1312 iph = skb_gro_header_fast(skb, off); 1313 if (skb_gro_header_hard(skb, hlen)) { 1314 iph = skb_gro_header_slow(skb, hlen, off); 1315 if (unlikely(!iph)) 1316 goto out; 1317 } 1318 1319 proto = iph->protocol & (MAX_INET_PROTOS - 1); 1320 1321 rcu_read_lock(); 1322 ops = rcu_dereference(inet_protos[proto]); 1323 if (!ops || !ops->gro_receive) 1324 goto out_unlock; 1325 1326 if (*(u8 *)iph != 0x45) 1327 goto out_unlock; 1328 1329 if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl))) 1330 goto out_unlock; 1331 1332 id = ntohl(*(__be32 *)&iph->id); 1333 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF)); 1334 id >>= 16; 1335 1336 for (p = *head; p; p = p->next) { 1337 struct iphdr *iph2; 1338 1339 if (!NAPI_GRO_CB(p)->same_flow) 1340 continue; 1341 1342 iph2 = ip_hdr(p); 1343 1344 if ((iph->protocol ^ iph2->protocol) | 1345 (iph->tos ^ iph2->tos) | 1346 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) | 1347 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) { 1348 NAPI_GRO_CB(p)->same_flow = 0; 1349 continue; 1350 } 1351 1352 /* All fields must match except length and checksum. */ 1353 NAPI_GRO_CB(p)->flush |= 1354 (iph->ttl ^ iph2->ttl) | 1355 ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id); 1356 1357 NAPI_GRO_CB(p)->flush |= flush; 1358 } 1359 1360 NAPI_GRO_CB(skb)->flush |= flush; 1361 skb_gro_pull(skb, sizeof(*iph)); 1362 skb_set_transport_header(skb, skb_gro_offset(skb)); 1363 1364 pp = ops->gro_receive(head, skb); 1365 1366out_unlock: 1367 rcu_read_unlock(); 1368 1369out: 1370 NAPI_GRO_CB(skb)->flush |= flush; 1371 1372 return pp; 1373} 1374 1375static int BCMFASTPATH_HOST inet_gro_complete(struct sk_buff *skb) 1376{ 1377 const struct net_protocol *ops; 1378 struct iphdr *iph = ip_hdr(skb); 1379 int proto = iph->protocol & (MAX_INET_PROTOS - 1); 1380 int err = -ENOSYS; 1381 __be16 newlen = htons(skb->len - skb_network_offset(skb)); 1382 1383 csum_replace2(&iph->check, iph->tot_len, newlen); 1384 iph->tot_len = newlen; 1385 1386 rcu_read_lock(); 1387 ops = rcu_dereference(inet_protos[proto]); 1388 if (WARN_ON(!ops || !ops->gro_complete)) 1389 goto out_unlock; 1390 1391 err = ops->gro_complete(skb); 1392 1393out_unlock: 1394 rcu_read_unlock(); 1395 1396 return err; 1397} 1398 1399int inet_ctl_sock_create(struct sock **sk, unsigned short family, 1400 unsigned short type, unsigned char protocol, 1401 struct net *net) 1402{ 1403 struct socket *sock; 1404 int rc = sock_create_kern(family, type, protocol, &sock); 1405 1406 if (rc == 0) { 1407 *sk = sock->sk; 1408 (*sk)->sk_allocation = GFP_ATOMIC; 1409 /* 1410 * Unhash it so that IP input processing does not even see it, 1411 * we do not wish this socket to see incoming packets. 1412 */ 1413 (*sk)->sk_prot->unhash(*sk); 1414 1415 sk_change_net(*sk, net); 1416 } 1417 return rc; 1418} 1419EXPORT_SYMBOL_GPL(inet_ctl_sock_create); 1420 1421unsigned long snmp_fold_field(void __percpu *mib[], int offt) 1422{ 1423 unsigned long res = 0; 1424 int i; 1425 1426 for_each_possible_cpu(i) { 1427 res += *(((unsigned long *) per_cpu_ptr(mib[0], i)) + offt); 1428 res += *(((unsigned long *) per_cpu_ptr(mib[1], i)) + offt); 1429 } 1430 return res; 1431} 1432EXPORT_SYMBOL_GPL(snmp_fold_field); 1433 1434#if BITS_PER_LONG==32 1435 1436u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset) 1437{ 1438 u64 res = 0; 1439 int cpu; 1440 1441 for_each_possible_cpu(cpu) { 1442 void *bhptr, *userptr; 1443 struct u64_stats_sync *syncp; 1444 u64 v_bh, v_user; 1445 unsigned int start; 1446 1447 /* first mib used by softirq context, we must use _bh() accessors */ 1448 bhptr = per_cpu_ptr(SNMP_STAT_BHPTR(mib), cpu); 1449 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset); 1450 do { 1451 start = u64_stats_fetch_begin_bh(syncp); 1452 v_bh = *(((u64 *) bhptr) + offt); 1453 } while (u64_stats_fetch_retry_bh(syncp, start)); 1454 1455 /* second mib used in USER context */ 1456 userptr = per_cpu_ptr(SNMP_STAT_USRPTR(mib), cpu); 1457 syncp = (struct u64_stats_sync *)(userptr + syncp_offset); 1458 do { 1459 start = u64_stats_fetch_begin(syncp); 1460 v_user = *(((u64 *) userptr) + offt); 1461 } while (u64_stats_fetch_retry(syncp, start)); 1462 1463 res += v_bh + v_user; 1464 } 1465 return res; 1466} 1467EXPORT_SYMBOL_GPL(snmp_fold_field64); 1468#endif 1469 1470int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align) 1471{ 1472 BUG_ON(ptr == NULL); 1473 ptr[0] = __alloc_percpu(mibsize, align); 1474 if (!ptr[0]) 1475 goto err0; 1476 ptr[1] = __alloc_percpu(mibsize, align); 1477 if (!ptr[1]) 1478 goto err1; 1479 return 0; 1480err1: 1481 free_percpu(ptr[0]); 1482 ptr[0] = NULL; 1483err0: 1484 return -ENOMEM; 1485} 1486EXPORT_SYMBOL_GPL(snmp_mib_init); 1487 1488void snmp_mib_free(void __percpu *ptr[2]) 1489{ 1490 BUG_ON(ptr == NULL); 1491 free_percpu(ptr[0]); 1492 free_percpu(ptr[1]); 1493 ptr[0] = ptr[1] = NULL; 1494} 1495EXPORT_SYMBOL_GPL(snmp_mib_free); 1496 1497#ifdef CONFIG_IP_MULTICAST 1498static const struct net_protocol igmp_protocol = { 1499 .handler = igmp_rcv, 1500 .netns_ok = 1, 1501}; 1502#endif 1503 1504static const struct net_protocol tcp_protocol = { 1505 .handler = tcp_v4_rcv, 1506 .err_handler = tcp_v4_err, 1507 .gso_send_check = tcp_v4_gso_send_check, 1508 .gso_segment = tcp_tso_segment, 1509 .gro_receive = tcp4_gro_receive, 1510 .gro_complete = tcp4_gro_complete, 1511 .no_policy = 1, 1512 .netns_ok = 1, 1513}; 1514 1515static const struct net_protocol udp_protocol = { 1516 .handler = udp_rcv, 1517 .err_handler = udp_err, 1518 .gso_send_check = udp4_ufo_send_check, 1519 .gso_segment = udp4_ufo_fragment, 1520 .no_policy = 1, 1521 .netns_ok = 1, 1522}; 1523 1524static const struct net_protocol icmp_protocol = { 1525 .handler = icmp_rcv, 1526 .no_policy = 1, 1527 .netns_ok = 1, 1528}; 1529 1530static __net_init int ipv4_mib_init_net(struct net *net) 1531{ 1532 if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics, 1533 sizeof(struct tcp_mib), 1534 __alignof__(struct tcp_mib)) < 0) 1535 goto err_tcp_mib; 1536 if (snmp_mib_init((void __percpu **)net->mib.ip_statistics, 1537 sizeof(struct ipstats_mib), 1538 __alignof__(struct ipstats_mib)) < 0) 1539 goto err_ip_mib; 1540 if (snmp_mib_init((void __percpu **)net->mib.net_statistics, 1541 sizeof(struct linux_mib), 1542 __alignof__(struct linux_mib)) < 0) 1543 goto err_net_mib; 1544 if (snmp_mib_init((void __percpu **)net->mib.udp_statistics, 1545 sizeof(struct udp_mib), 1546 __alignof__(struct udp_mib)) < 0) 1547 goto err_udp_mib; 1548 if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics, 1549 sizeof(struct udp_mib), 1550 __alignof__(struct udp_mib)) < 0) 1551 goto err_udplite_mib; 1552 if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics, 1553 sizeof(struct icmp_mib), 1554 __alignof__(struct icmp_mib)) < 0) 1555 goto err_icmp_mib; 1556 if (snmp_mib_init((void __percpu **)net->mib.icmpmsg_statistics, 1557 sizeof(struct icmpmsg_mib), 1558 __alignof__(struct icmpmsg_mib)) < 0) 1559 goto err_icmpmsg_mib; 1560 1561 tcp_mib_init(net); 1562 return 0; 1563 1564err_icmpmsg_mib: 1565 snmp_mib_free((void __percpu **)net->mib.icmp_statistics); 1566err_icmp_mib: 1567 snmp_mib_free((void __percpu **)net->mib.udplite_statistics); 1568err_udplite_mib: 1569 snmp_mib_free((void __percpu **)net->mib.udp_statistics); 1570err_udp_mib: 1571 snmp_mib_free((void __percpu **)net->mib.net_statistics); 1572err_net_mib: 1573 snmp_mib_free((void __percpu **)net->mib.ip_statistics); 1574err_ip_mib: 1575 snmp_mib_free((void __percpu **)net->mib.tcp_statistics); 1576err_tcp_mib: 1577 return -ENOMEM; 1578} 1579 1580static __net_exit void ipv4_mib_exit_net(struct net *net) 1581{ 1582 snmp_mib_free((void __percpu **)net->mib.icmpmsg_statistics); 1583 snmp_mib_free((void __percpu **)net->mib.icmp_statistics); 1584 snmp_mib_free((void __percpu **)net->mib.udplite_statistics); 1585 snmp_mib_free((void __percpu **)net->mib.udp_statistics); 1586 snmp_mib_free((void __percpu **)net->mib.net_statistics); 1587 snmp_mib_free((void __percpu **)net->mib.ip_statistics); 1588 snmp_mib_free((void __percpu **)net->mib.tcp_statistics); 1589} 1590 1591static __net_initdata struct pernet_operations ipv4_mib_ops = { 1592 .init = ipv4_mib_init_net, 1593 .exit = ipv4_mib_exit_net, 1594}; 1595 1596static int __init init_ipv4_mibs(void) 1597{ 1598 return register_pernet_subsys(&ipv4_mib_ops); 1599} 1600 1601static int ipv4_proc_init(void); 1602 1603/* 1604 * IP protocol layer initialiser 1605 */ 1606 1607static struct packet_type ip_packet_type __read_mostly = { 1608 .type = cpu_to_be16(ETH_P_IP), 1609 .func = ip_rcv, 1610 .gso_send_check = inet_gso_send_check, 1611 .gso_segment = inet_gso_segment, 1612 .gro_receive = inet_gro_receive, 1613 .gro_complete = inet_gro_complete, 1614}; 1615 1616static int __init inet_init(void) 1617{ 1618 struct sk_buff *dummy_skb; 1619 struct inet_protosw *q; 1620 struct list_head *r; 1621 int rc = -EINVAL; 1622 1623 BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb)); 1624 1625 sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL); 1626 if (!sysctl_local_reserved_ports) 1627 goto out; 1628 1629 rc = proto_register(&tcp_prot, 1); 1630 if (rc) 1631 goto out_free_reserved_ports; 1632 1633 rc = proto_register(&udp_prot, 1); 1634 if (rc) 1635 goto out_unregister_tcp_proto; 1636 1637 rc = proto_register(&raw_prot, 1); 1638 if (rc) 1639 goto out_unregister_udp_proto; 1640 1641 /* 1642 * Tell SOCKET that we are alive... 1643 */ 1644 1645 (void)sock_register(&inet_family_ops); 1646 1647#ifdef CONFIG_SYSCTL 1648 ip_static_sysctl_init(); 1649#endif 1650 1651 /* 1652 * Add all the base protocols. 1653 */ 1654 1655 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0) 1656 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n"); 1657 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0) 1658 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n"); 1659 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0) 1660 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n"); 1661#ifdef CONFIG_IP_MULTICAST 1662 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0) 1663 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n"); 1664#endif 1665 1666 /* Register the socket-side information for inet_create. */ 1667 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r) 1668 INIT_LIST_HEAD(r); 1669 1670 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q) 1671 inet_register_protosw(q); 1672 1673 /* 1674 * Set the ARP module up 1675 */ 1676 1677 arp_init(); 1678 1679 /* 1680 * Set the IP module up 1681 */ 1682 1683 ip_init(); 1684 1685 tcp_v4_init(); 1686 1687 /* Setup TCP slab cache for open requests. */ 1688 tcp_init(); 1689 1690 /* Setup UDP memory threshold */ 1691 udp_init(); 1692 1693 /* Add UDP-Lite (RFC 3828) */ 1694 udplite4_register(); 1695 1696 /* 1697 * Set the ICMP layer up 1698 */ 1699 1700 if (icmp_init() < 0) 1701 panic("Failed to create the ICMP control socket.\n"); 1702 1703 /* 1704 * Initialise the multicast router 1705 */ 1706#if defined(CONFIG_IP_MROUTE) 1707 if (ip_mr_init()) 1708 printk(KERN_CRIT "inet_init: Cannot init ipv4 mroute\n"); 1709#endif 1710 /* 1711 * Initialise per-cpu ipv4 mibs 1712 */ 1713 1714 if (init_ipv4_mibs()) 1715 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n"); 1716 1717 ipv4_proc_init(); 1718 1719 ipfrag_init(); 1720 1721 dev_add_pack(&ip_packet_type); 1722 1723 rc = 0; 1724out: 1725 return rc; 1726out_unregister_udp_proto: 1727 proto_unregister(&udp_prot); 1728out_unregister_tcp_proto: 1729 proto_unregister(&tcp_prot); 1730out_free_reserved_ports: 1731 kfree(sysctl_local_reserved_ports); 1732 goto out; 1733} 1734 1735fs_initcall(inet_init); 1736 1737/* ------------------------------------------------------------------------ */ 1738 1739#ifdef CONFIG_PROC_FS 1740static int __init ipv4_proc_init(void) 1741{ 1742 int rc = 0; 1743 1744 if (raw_proc_init()) 1745 goto out_raw; 1746 if (tcp4_proc_init()) 1747 goto out_tcp; 1748 if (udp4_proc_init()) 1749 goto out_udp; 1750 if (ip_misc_proc_init()) 1751 goto out_misc; 1752out: 1753 return rc; 1754out_misc: 1755 udp4_proc_exit(); 1756out_udp: 1757 tcp4_proc_exit(); 1758out_tcp: 1759 raw_proc_exit(); 1760out_raw: 1761 rc = -ENOMEM; 1762 goto out; 1763} 1764 1765#else /* CONFIG_PROC_FS */ 1766static int __init ipv4_proc_init(void) 1767{ 1768 return 0; 1769} 1770#endif /* CONFIG_PROC_FS */ 1771 1772MODULE_ALIAS_NETPROTO(PF_INET); 1773