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 119 120 121 /* The inetsw table contains everything that inet_create needs to 122 * build a new socket. 123 */ 124 static struct list_head inetsw[SOCK_MAX]; 125 static DEFINE_SPINLOCK(inetsw_lock); 126 127 struct ipv4_config ipv4_config; 128 EXPORT_SYMBOL(ipv4_config); 129 130 /* New destruction routine */ 131 132 void inet_sock_destruct(struct sock *sk) 133 { 134 struct inet_sock *inet = inet_sk(sk); 135 136 __skb_queue_purge(&sk->sk_receive_queue); 137 __skb_queue_purge(&sk->sk_error_queue); 138 139 sk_mem_reclaim(sk); 140 141 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) { 142 pr_err("Attempt to release TCP socket in state %d %p\n", 143 sk->sk_state, sk); 144 return; 145 } 146 if (!sock_flag(sk, SOCK_DEAD)) { 147 pr_err("Attempt to release alive inet socket %p\n", sk); 148 return; 149 } 150 151 WARN_ON(atomic_read(&sk->sk_rmem_alloc)); 152 WARN_ON(atomic_read(&sk->sk_wmem_alloc)); 153 WARN_ON(sk->sk_wmem_queued); 154 WARN_ON(sk->sk_forward_alloc); 155 156 kfree(inet->opt); 157 dst_release(rcu_dereference_check(sk->sk_dst_cache, 1)); 158 sk_refcnt_debug_dec(sk); 159 } 160 EXPORT_SYMBOL(inet_sock_destruct); 161 162 /* 163 * The routines beyond this point handle the behaviour of an AF_INET 164 * socket object. Mostly it punts to the subprotocols of IP to do 165 * the work. 166 */ 167 168 /* 169 * Automatically bind an unbound socket. 170 */ 171 172 static int inet_autobind(struct sock *sk) 173 { 174 struct inet_sock *inet; 175 /* We may need to bind the socket. */ 176 lock_sock(sk); 177 inet = inet_sk(sk); 178 if (!inet->inet_num) { 179 if (sk->sk_prot->get_port(sk, 0)) { 180 release_sock(sk); 181 return -EAGAIN; 182 } 183 inet->inet_sport = htons(inet->inet_num); 184 } 185 release_sock(sk); 186 return 0; 187 } 188 189 /* 190 * Move a socket into listening state. 191 */ 192 int inet_listen(struct socket *sock, int backlog) 193 { 194 struct sock *sk = sock->sk; 195 unsigned char old_state; 196 int err; 197 198 lock_sock(sk); 199 200 err = -EINVAL; 201 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM) 202 goto out; 203 204 old_state = sk->sk_state; 205 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN))) 206 goto out; 207 208 /* Really, if the socket is already in listen state 209 * we can only allow the backlog to be adjusted. 210 */ 211 if (old_state != TCP_LISTEN) { 212 err = inet_csk_listen_start(sk, backlog); 213 if (err) 214 goto out; 215 } 216 sk->sk_max_ack_backlog = backlog; 217 err = 0; 218 219 out: 220 release_sock(sk); 221 return err; 222 } 223 EXPORT_SYMBOL(inet_listen); 224 225 u32 inet_ehash_secret __read_mostly; 226 EXPORT_SYMBOL(inet_ehash_secret); 227 228 /* 229 * inet_ehash_secret must be set exactly once 230 * Instead of using a dedicated spinlock, we (ab)use inetsw_lock 231 */ 232 void build_ehash_secret(void) 233 { 234 u32 rnd; 235 do { 236 get_random_bytes(&rnd, sizeof(rnd)); 237 } while (rnd == 0); 238 spin_lock_bh(&inetsw_lock); 239 if (!inet_ehash_secret) 240 inet_ehash_secret = rnd; 241 spin_unlock_bh(&inetsw_lock); 242 } 243 EXPORT_SYMBOL(build_ehash_secret); 244 245 static inline int inet_netns_ok(struct net *net, int protocol) 246 { 247 int hash; 248 const struct net_protocol *ipprot; 249 250 if (net_eq(net, &init_net)) 251 return 1; 252 253 hash = protocol & (MAX_INET_PROTOS - 1); 254 ipprot = rcu_dereference(inet_protos[hash]); 255 256 if (ipprot == NULL) 257 /* raw IP is OK */ 258 return 1; 259 return ipprot->netns_ok; 260 } 261 262 /* 263 * Create an inet socket. 264 */ 265 266 static int inet_create(struct net *net, struct socket *sock, int protocol, 267 int kern) 268 { 269 struct sock *sk; 270 struct inet_protosw *answer; 271 struct inet_sock *inet; 272 struct proto *answer_prot; 273 unsigned char answer_flags; 274 char answer_no_check; 275 int try_loading_module = 0; 276 int err; 277 278 if (unlikely(!inet_ehash_secret)) 279 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM) 280 build_ehash_secret(); 281 282 sock->state = SS_UNCONNECTED; 283 284 /* Look for the requested type/protocol pair. */ 285 lookup_protocol: 286 err = -ESOCKTNOSUPPORT; 287 rcu_read_lock(); 288 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) { 289 290 err = 0; 291 /* Check the non-wild match. */ 292 if (protocol == answer->protocol) { 293 if (protocol != IPPROTO_IP) 294 break; 295 } else { 296 /* Check for the two wild cases. */ 297 if (IPPROTO_IP == protocol) { 298 protocol = answer->protocol; 299 break; 300 } 301 if (IPPROTO_IP == answer->protocol) 302 break; 303 } 304 err = -EPROTONOSUPPORT; 305 } 306 307 if (unlikely(err)) { 308 if (try_loading_module < 2) { 309 rcu_read_unlock(); 310 /* 311 * Be more specific, e.g. net-pf-2-proto-132-type-1 312 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM) 313 */ 314 if (++try_loading_module == 1) 315 request_module("net-pf-%d-proto-%d-type-%d", 316 PF_INET, protocol, sock->type); 317 /* 318 * Fall back to generic, e.g. net-pf-2-proto-132 319 * (net-pf-PF_INET-proto-IPPROTO_SCTP) 320 */ 321 else 322 request_module("net-pf-%d-proto-%d", 323 PF_INET, protocol); 324 goto lookup_protocol; 325 } else 326 goto out_rcu_unlock; 327 } 328 329 err = -EPERM; 330 if (sock->type == SOCK_RAW && !kern && !capable(CAP_NET_RAW)) 331 goto out_rcu_unlock; 332 333 err = -EAFNOSUPPORT; 334 if (!inet_netns_ok(net, protocol)) 335 goto out_rcu_unlock; 336 337 sock->ops = answer->ops; 338 answer_prot = answer->prot; 339 answer_no_check = answer->no_check; 340 answer_flags = answer->flags; 341 rcu_read_unlock(); 342 343 WARN_ON(answer_prot->slab == NULL); 344 345 err = -ENOBUFS; 346 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot); 347 if (sk == NULL) 348 goto out; 349 350 err = 0; 351 sk->sk_no_check = answer_no_check; 352 if (INET_PROTOSW_REUSE & answer_flags) 353 sk->sk_reuse = 1; 354 355 inet = inet_sk(sk); 356 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0; 357 358 inet->nodefrag = 0; 359 360 if (SOCK_RAW == sock->type) { 361 inet->inet_num = protocol; 362 if (IPPROTO_RAW == protocol) 363 inet->hdrincl = 1; 364 } 365 366 if (ipv4_config.no_pmtu_disc) 367 inet->pmtudisc = IP_PMTUDISC_DONT; 368 else 369 inet->pmtudisc = IP_PMTUDISC_WANT; 370 371 inet->inet_id = 0; 372 373 sock_init_data(sock, sk); 374 375 sk->sk_destruct = inet_sock_destruct; 376 sk->sk_protocol = protocol; 377 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv; 378 379 inet->uc_ttl = -1; 380 inet->mc_loop = 1; 381 inet->mc_ttl = 1; 382 inet->mc_all = 1; 383 inet->mc_index = 0; 384 inet->mc_list = NULL; 385 386 sk_refcnt_debug_inc(sk); 387 388 if (inet->inet_num) { 389 /* It assumes that any protocol which allows 390 * the user to assign a number at socket 391 * creation time automatically 392 * shares. 393 */ 394 inet->inet_sport = htons(inet->inet_num); 395 /* Add to protocol hash chains. */ 396 sk->sk_prot->hash(sk); 397 } 398 399 if (sk->sk_prot->init) { 400 err = sk->sk_prot->init(sk); 401 if (err) 402 sk_common_release(sk); 403 } 404 out: 405 return err; 406 out_rcu_unlock: 407 rcu_read_unlock(); 408 goto out; 409 } 410 411 412 /* 413 * The peer socket should always be NULL (or else). When we call this 414 * function we are destroying the object and from then on nobody 415 * should refer to it. 416 */ 417 int inet_release(struct socket *sock) 418 { 419 struct sock *sk = sock->sk; 420 421 if (sk) { 422 long timeout; 423 424 sock_rps_reset_flow(sk); 425 426 /* Applications forget to leave groups before exiting */ 427 ip_mc_drop_socket(sk); 428 429 /* If linger is set, we don't return until the close 430 * is complete. Otherwise we return immediately. The 431 * actually closing is done the same either way. 432 * 433 * If the close is due to the process exiting, we never 434 * linger.. 435 */ 436 timeout = 0; 437 if (sock_flag(sk, SOCK_LINGER) && 438 !(current->flags & PF_EXITING)) 439 timeout = sk->sk_lingertime; 440 sock->sk = NULL; 441 sk->sk_prot->close(sk, timeout); 442 } 443 return 0; 444 } 445 EXPORT_SYMBOL(inet_release); 446 447 /* It is off by default, see below. */ 448 int sysctl_ip_nonlocal_bind __read_mostly; 449 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind); 450 451 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) 452 { 453 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr; 454 struct sock *sk = sock->sk; 455 struct inet_sock *inet = inet_sk(sk); 456 unsigned short snum; 457 int chk_addr_ret; 458 int err; 459 460 /* If the socket has its own bind function then use it. (RAW) */ 461 if (sk->sk_prot->bind) { 462 err = sk->sk_prot->bind(sk, uaddr, addr_len); 463 goto out; 464 } 465 err = -EINVAL; 466 if (addr_len < sizeof(struct sockaddr_in)) 467 goto out; 468 469 chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr); 470 471 /* Not specified by any standard per-se, however it breaks too 472 * many applications when removed. It is unfortunate since 473 * allowing applications to make a non-local bind solves 474 * several problems with systems using dynamic addressing. 475 * (ie. your servers still start up even if your ISDN link 476 * is temporarily down) 477 */ 478 err = -EADDRNOTAVAIL; 479 if (!sysctl_ip_nonlocal_bind && 480 !(inet->freebind || inet->transparent) && 481 addr->sin_addr.s_addr != htonl(INADDR_ANY) && 482 chk_addr_ret != RTN_LOCAL && 483 chk_addr_ret != RTN_MULTICAST && 484 chk_addr_ret != RTN_BROADCAST) 485 goto out; 486 487 snum = ntohs(addr->sin_port); 488 err = -EACCES; 489 if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE)) 490 goto out; 491 492 /* We keep a pair of addresses. rcv_saddr is the one 493 * used by hash lookups, and saddr is used for transmit. 494 * 495 * In the BSD API these are the same except where it 496 * would be illegal to use them (multicast/broadcast) in 497 * which case the sending device address is used. 498 */ 499 lock_sock(sk); 500 501 /* Check these errors (active socket, double bind). */ 502 err = -EINVAL; 503 if (sk->sk_state != TCP_CLOSE || inet->inet_num) 504 goto out_release_sock; 505 506 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr; 507 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST) 508 inet->inet_saddr = 0; /* Use device */ 509 510 /* Make sure we are allowed to bind here. */ 511 if (sk->sk_prot->get_port(sk, snum)) { 512 inet->inet_saddr = inet->inet_rcv_saddr = 0; 513 err = -EADDRINUSE; 514 goto out_release_sock; 515 } 516 517 if (inet->inet_rcv_saddr) 518 sk->sk_userlocks |= SOCK_BINDADDR_LOCK; 519 if (snum) 520 sk->sk_userlocks |= SOCK_BINDPORT_LOCK; 521 inet->inet_sport = htons(inet->inet_num); 522 inet->inet_daddr = 0; 523 inet->inet_dport = 0; 524 sk_dst_reset(sk); 525 err = 0; 526 out_release_sock: 527 release_sock(sk); 528 out: 529 return err; 530 } 531 EXPORT_SYMBOL(inet_bind); 532 533 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr, 534 int addr_len, int flags) 535 { 536 struct sock *sk = sock->sk; 537 538 if (addr_len < sizeof(uaddr->sa_family)) 539 return -EINVAL; 540 if (uaddr->sa_family == AF_UNSPEC) 541 return sk->sk_prot->disconnect(sk, flags); 542 543 if (!inet_sk(sk)->inet_num && inet_autobind(sk)) 544 return -EAGAIN; 545 return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len); 546 } 547 EXPORT_SYMBOL(inet_dgram_connect); 548 549 static long inet_wait_for_connect(struct sock *sk, long timeo) 550 { 551 DEFINE_WAIT(wait); 552 553 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 554 555 /* Basic assumption: if someone sets sk->sk_err, he _must_ 556 * change state of the socket from TCP_SYN_*. 557 * Connect() does not allow to get error notifications 558 * without closing the socket. 559 */ 560 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 561 release_sock(sk); 562 timeo = schedule_timeout(timeo); 563 lock_sock(sk); 564 if (signal_pending(current) || !timeo) 565 break; 566 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 567 } 568 finish_wait(sk_sleep(sk), &wait); 569 return timeo; 570 } 571 572 /* 573 * Connect to a remote host. There is regrettably still a little 574 * TCP 'magic' in here. 575 */ 576 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr, 577 int addr_len, int flags) 578 { 579 struct sock *sk = sock->sk; 580 int err; 581 long timeo; 582 583 if (addr_len < sizeof(uaddr->sa_family)) 584 return -EINVAL; 585 586 lock_sock(sk); 587 588 if (uaddr->sa_family == AF_UNSPEC) { 589 err = sk->sk_prot->disconnect(sk, flags); 590 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED; 591 goto out; 592 } 593 594 switch (sock->state) { 595 default: 596 err = -EINVAL; 597 goto out; 598 case SS_CONNECTED: 599 err = -EISCONN; 600 goto out; 601 case SS_CONNECTING: 602 err = -EALREADY; 603 /* Fall out of switch with err, set for this state */ 604 break; 605 case SS_UNCONNECTED: 606 err = -EISCONN; 607 if (sk->sk_state != TCP_CLOSE) 608 goto out; 609 610 err = sk->sk_prot->connect(sk, uaddr, addr_len); 611 if (err < 0) 612 goto out; 613 614 sock->state = SS_CONNECTING; 615 616 /* Just entered SS_CONNECTING state; the only 617 * difference is that return value in non-blocking 618 * case is EINPROGRESS, rather than EALREADY. 619 */ 620 err = -EINPROGRESS; 621 break; 622 } 623 624 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK); 625 626 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 627 /* Error code is set above */ 628 if (!timeo || !inet_wait_for_connect(sk, timeo)) 629 goto out; 630 631 err = sock_intr_errno(timeo); 632 if (signal_pending(current)) 633 goto out; 634 } 635 636 /* Connection was closed by RST, timeout, ICMP error 637 * or another process disconnected us. 638 */ 639 if (sk->sk_state == TCP_CLOSE) 640 goto sock_error; 641 642 /* sk->sk_err may be not zero now, if RECVERR was ordered by user 643 * and error was received after socket entered established state. 644 * Hence, it is handled normally after connect() return successfully. 645 */ 646 647 sock->state = SS_CONNECTED; 648 err = 0; 649 out: 650 release_sock(sk); 651 return err; 652 653 sock_error: 654 err = sock_error(sk) ? : -ECONNABORTED; 655 sock->state = SS_UNCONNECTED; 656 if (sk->sk_prot->disconnect(sk, flags)) 657 sock->state = SS_DISCONNECTING; 658 goto out; 659 } 660 EXPORT_SYMBOL(inet_stream_connect); 661 662 /* 663 * Accept a pending connection. The TCP layer now gives BSD semantics. 664 */ 665 666 int inet_accept(struct socket *sock, struct socket *newsock, int flags) 667 { 668 struct sock *sk1 = sock->sk; 669 int err = -EINVAL; 670 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err); 671 672 if (!sk2) 673 goto do_err; 674 675 lock_sock(sk2); 676 677 WARN_ON(!((1 << sk2->sk_state) & 678 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE))); 679 680 sock_graft(sk2, newsock); 681 682 newsock->state = SS_CONNECTED; 683 err = 0; 684 release_sock(sk2); 685 do_err: 686 return err; 687 } 688 EXPORT_SYMBOL(inet_accept); 689 690 691 /* 692 * This does both peername and sockname. 693 */ 694 int inet_getname(struct socket *sock, struct sockaddr *uaddr, 695 int *uaddr_len, int peer) 696 { 697 struct sock *sk = sock->sk; 698 struct inet_sock *inet = inet_sk(sk); 699 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr); 700 701 sin->sin_family = AF_INET; 702 if (peer) { 703 if (!inet->inet_dport || 704 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) && 705 peer == 1)) 706 return -ENOTCONN; 707 sin->sin_port = inet->inet_dport; 708 sin->sin_addr.s_addr = inet->inet_daddr; 709 } else { 710 __be32 addr = inet->inet_rcv_saddr; 711 if (!addr) 712 addr = inet->inet_saddr; 713 sin->sin_port = inet->inet_sport; 714 sin->sin_addr.s_addr = addr; 715 } 716 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 717 *uaddr_len = sizeof(*sin); 718 return 0; 719 } 720 EXPORT_SYMBOL(inet_getname); 721 722 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg, 723 size_t size) 724 { 725 struct sock *sk = sock->sk; 726 727 sock_rps_record_flow(sk); 728 729 /* We may need to bind the socket. */ 730 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind && 731 inet_autobind(sk)) 732 return -EAGAIN; 733 734 return sk->sk_prot->sendmsg(iocb, sk, msg, size); 735 } 736 EXPORT_SYMBOL(inet_sendmsg); 737 738 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, 739 size_t size, int flags) 740 { 741 struct sock *sk = sock->sk; 742 743 sock_rps_record_flow(sk); 744 745 /* We may need to bind the socket. */ 746 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind && 747 inet_autobind(sk)) 748 return -EAGAIN; 749 750 if (sk->sk_prot->sendpage) 751 return sk->sk_prot->sendpage(sk, page, offset, size, flags); 752 return sock_no_sendpage(sock, page, offset, size, flags); 753 } 754 EXPORT_SYMBOL(inet_sendpage); 755 756 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg, 757 size_t size, int flags) 758 { 759 struct sock *sk = sock->sk; 760 int addr_len = 0; 761 int err; 762 763 sock_rps_record_flow(sk); 764 765 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT, 766 flags & ~MSG_DONTWAIT, &addr_len); 767 if (err >= 0) 768 msg->msg_namelen = addr_len; 769 return err; 770 } 771 EXPORT_SYMBOL(inet_recvmsg); 772 773 int inet_shutdown(struct socket *sock, int how) 774 { 775 struct sock *sk = sock->sk; 776 int err = 0; 777 778 /* This should really check to make sure 779 * the socket is a TCP socket. (WHY AC...) 780 */ 781 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and 782 1->2 bit 2 snds. 783 2->3 */ 784 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */ 785 return -EINVAL; 786 787 lock_sock(sk); 788 if (sock->state == SS_CONNECTING) { 789 if ((1 << sk->sk_state) & 790 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE)) 791 sock->state = SS_DISCONNECTING; 792 else 793 sock->state = SS_CONNECTED; 794 } 795 796 switch (sk->sk_state) { 797 case TCP_CLOSE: 798 err = -ENOTCONN; 799 /* Hack to wake up other listeners, who can poll for 800 POLLHUP, even on eg. unconnected UDP sockets -- RR */ 801 default: 802 sk->sk_shutdown |= how; 803 if (sk->sk_prot->shutdown) 804 sk->sk_prot->shutdown(sk, how); 805 break; 806 807 /* Remaining two branches are temporary solution for missing 808 * close() in multithreaded environment. It is _not_ a good idea, 809 * but we have no choice until close() is repaired at VFS level. 810 */ 811 case TCP_LISTEN: 812 if (!(how & RCV_SHUTDOWN)) 813 break; 814 /* Fall through */ 815 case TCP_SYN_SENT: 816 err = sk->sk_prot->disconnect(sk, O_NONBLOCK); 817 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED; 818 break; 819 } 820 821 /* Wake up anyone sleeping in poll. */ 822 sk->sk_state_change(sk); 823 release_sock(sk); 824 return err; 825 } 826 EXPORT_SYMBOL(inet_shutdown); 827 828 /* 829 * ioctl() calls you can issue on an INET socket. Most of these are 830 * device configuration and stuff and very rarely used. Some ioctls 831 * pass on to the socket itself. 832 * 833 * NOTE: I like the idea of a module for the config stuff. ie ifconfig 834 * loads the devconfigure module does its configuring and unloads it. 835 * There's a good 20K of config code hanging around the kernel. 836 */ 837 838 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 839 { 840 struct sock *sk = sock->sk; 841 int err = 0; 842 struct net *net = sock_net(sk); 843 844 switch (cmd) { 845 case SIOCGSTAMP: 846 err = sock_get_timestamp(sk, (struct timeval __user *)arg); 847 break; 848 case SIOCGSTAMPNS: 849 err = sock_get_timestampns(sk, (struct timespec __user *)arg); 850 break; 851 case SIOCADDRT: 852 case SIOCDELRT: 853 case SIOCRTMSG: 854 err = ip_rt_ioctl(net, cmd, (void __user *)arg); 855 break; 856 case SIOCDARP: 857 case SIOCGARP: 858 case SIOCSARP: 859 err = arp_ioctl(net, cmd, (void __user *)arg); 860 break; 861 case SIOCGIFADDR: 862 case SIOCSIFADDR: 863 case SIOCGIFBRDADDR: 864 case SIOCSIFBRDADDR: 865 case SIOCGIFNETMASK: 866 case SIOCSIFNETMASK: 867 case SIOCGIFDSTADDR: 868 case SIOCSIFDSTADDR: 869 case SIOCSIFPFLAGS: 870 case SIOCGIFPFLAGS: 871 case SIOCSIFFLAGS: 872 err = devinet_ioctl(net, cmd, (void __user *)arg); 873 break; 874 default: 875 if (sk->sk_prot->ioctl) 876 err = sk->sk_prot->ioctl(sk, cmd, arg); 877 else 878 err = -ENOIOCTLCMD; 879 break; 880 } 881 return err; 882 } 883 EXPORT_SYMBOL(inet_ioctl); 884 885 const struct proto_ops inet_stream_ops = { 886 .family = PF_INET, 887 .owner = THIS_MODULE, 888 .release = inet_release, 889 .bind = inet_bind, 890 .connect = inet_stream_connect, 891 .socketpair = sock_no_socketpair, 892 .accept = inet_accept, 893 .getname = inet_getname, 894 .poll = tcp_poll, 895 .ioctl = inet_ioctl, 896 .listen = inet_listen, 897 .shutdown = inet_shutdown, 898 .setsockopt = sock_common_setsockopt, 899 .getsockopt = sock_common_getsockopt, 900 .sendmsg = inet_sendmsg, 901 .recvmsg = inet_recvmsg, 902 .mmap = sock_no_mmap, 903 .sendpage = inet_sendpage, 904 .splice_read = tcp_splice_read, 905 #ifdef CONFIG_COMPAT 906 .compat_setsockopt = compat_sock_common_setsockopt, 907 .compat_getsockopt = compat_sock_common_getsockopt, 908 #endif 909 }; 910 EXPORT_SYMBOL(inet_stream_ops); 911 912 const struct proto_ops inet_dgram_ops = { 913 .family = PF_INET, 914 .owner = THIS_MODULE, 915 .release = inet_release, 916 .bind = inet_bind, 917 .connect = inet_dgram_connect, 918 .socketpair = sock_no_socketpair, 919 .accept = sock_no_accept, 920 .getname = inet_getname, 921 .poll = udp_poll, 922 .ioctl = inet_ioctl, 923 .listen = sock_no_listen, 924 .shutdown = inet_shutdown, 925 .setsockopt = sock_common_setsockopt, 926 .getsockopt = sock_common_getsockopt, 927 .sendmsg = inet_sendmsg, 928 .recvmsg = inet_recvmsg, 929 .mmap = sock_no_mmap, 930 .sendpage = inet_sendpage, 931 #ifdef CONFIG_COMPAT 932 .compat_setsockopt = compat_sock_common_setsockopt, 933 .compat_getsockopt = compat_sock_common_getsockopt, 934 #endif 935 }; 936 EXPORT_SYMBOL(inet_dgram_ops); 937 938 /* 939 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without 940 * udp_poll 941 */ 942 static const struct proto_ops inet_sockraw_ops = { 943 .family = PF_INET, 944 .owner = THIS_MODULE, 945 .release = inet_release, 946 .bind = inet_bind, 947 .connect = inet_dgram_connect, 948 .socketpair = sock_no_socketpair, 949 .accept = sock_no_accept, 950 .getname = inet_getname, 951 .poll = datagram_poll, 952 .ioctl = inet_ioctl, 953 .listen = sock_no_listen, 954 .shutdown = inet_shutdown, 955 .setsockopt = sock_common_setsockopt, 956 .getsockopt = sock_common_getsockopt, 957 .sendmsg = inet_sendmsg, 958 .recvmsg = inet_recvmsg, 959 .mmap = sock_no_mmap, 960 .sendpage = inet_sendpage, 961 #ifdef CONFIG_COMPAT 962 .compat_setsockopt = compat_sock_common_setsockopt, 963 .compat_getsockopt = compat_sock_common_getsockopt, 964 #endif 965 }; 966 967 static const struct net_proto_family inet_family_ops = { 968 .family = PF_INET, 969 .create = inet_create, 970 .owner = THIS_MODULE, 971 }; 972 973 /* Upon startup we insert all the elements in inetsw_array[] into 974 * the linked list inetsw. 975 */ 976 static struct inet_protosw inetsw_array[] = 977 { 978 { 979 .type = SOCK_STREAM, 980 .protocol = IPPROTO_TCP, 981 .prot = &tcp_prot, 982 .ops = &inet_stream_ops, 983 .no_check = 0, 984 .flags = INET_PROTOSW_PERMANENT | 985 INET_PROTOSW_ICSK, 986 }, 987 988 { 989 .type = SOCK_DGRAM, 990 .protocol = IPPROTO_UDP, 991 .prot = &udp_prot, 992 .ops = &inet_dgram_ops, 993 .no_check = UDP_CSUM_DEFAULT, 994 .flags = INET_PROTOSW_PERMANENT, 995 }, 996 997 998 { 999 .type = SOCK_RAW, 1000 .protocol = IPPROTO_IP, /* wild card */ 1001 .prot = &raw_prot, 1002 .ops = &inet_sockraw_ops, 1003 .no_check = UDP_CSUM_DEFAULT, 1004 .flags = INET_PROTOSW_REUSE, 1005 } 1006 }; 1007 1008 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array) 1009 1010 void inet_register_protosw(struct inet_protosw *p) 1011 { 1012 struct list_head *lh; 1013 struct inet_protosw *answer; 1014 int protocol = p->protocol; 1015 struct list_head *last_perm; 1016 1017 spin_lock_bh(&inetsw_lock); 1018 1019 if (p->type >= SOCK_MAX) 1020 goto out_illegal; 1021 1022 /* If we are trying to override a permanent protocol, bail. */ 1023 answer = NULL; 1024 last_perm = &inetsw[p->type]; 1025 list_for_each(lh, &inetsw[p->type]) { 1026 answer = list_entry(lh, struct inet_protosw, list); 1027 1028 /* Check only the non-wild match. */ 1029 if (INET_PROTOSW_PERMANENT & answer->flags) { 1030 if (protocol == answer->protocol) 1031 break; 1032 last_perm = lh; 1033 } 1034 1035 answer = NULL; 1036 } 1037 if (answer) 1038 goto out_permanent; 1039 1040 /* Add the new entry after the last permanent entry if any, so that 1041 * the new entry does not override a permanent entry when matched with 1042 * a wild-card protocol. But it is allowed to override any existing 1043 * non-permanent entry. This means that when we remove this entry, the 1044 * system automatically returns to the old behavior. 1045 */ 1046 list_add_rcu(&p->list, last_perm); 1047 out: 1048 spin_unlock_bh(&inetsw_lock); 1049 1050 return; 1051 1052 out_permanent: 1053 printk(KERN_ERR "Attempt to override permanent protocol %d.\n", 1054 protocol); 1055 goto out; 1056 1057 out_illegal: 1058 printk(KERN_ERR 1059 "Ignoring attempt to register invalid socket type %d.\n", 1060 p->type); 1061 goto out; 1062 } 1063 EXPORT_SYMBOL(inet_register_protosw); 1064 1065 void inet_unregister_protosw(struct inet_protosw *p) 1066 { 1067 if (INET_PROTOSW_PERMANENT & p->flags) { 1068 printk(KERN_ERR 1069 "Attempt to unregister permanent protocol %d.\n", 1070 p->protocol); 1071 } else { 1072 spin_lock_bh(&inetsw_lock); 1073 list_del_rcu(&p->list); 1074 spin_unlock_bh(&inetsw_lock); 1075 1076 synchronize_net(); 1077 } 1078 } 1079 EXPORT_SYMBOL(inet_unregister_protosw); 1080 1081 /* 1082 * Shall we try to damage output packets if routing dev changes? 1083 */ 1084 1085 int sysctl_ip_dynaddr __read_mostly; 1086 1087 static int inet_sk_reselect_saddr(struct sock *sk) 1088 { 1089 struct inet_sock *inet = inet_sk(sk); 1090 int err; 1091 struct rtable *rt; 1092 __be32 old_saddr = inet->inet_saddr; 1093 __be32 new_saddr; 1094 __be32 daddr = inet->inet_daddr; 1095 1096 if (inet->opt && inet->opt->srr) 1097 daddr = inet->opt->faddr; 1098 1099 /* Query new route. */ 1100 err = ip_route_connect(&rt, daddr, 0, 1101 RT_CONN_FLAGS(sk), 1102 sk->sk_bound_dev_if, 1103 sk->sk_protocol, 1104 inet->inet_sport, inet->inet_dport, sk, 0); 1105 if (err) 1106 return err; 1107 1108 sk_setup_caps(sk, &rt->dst); 1109 1110 new_saddr = rt->rt_src; 1111 1112 if (new_saddr == old_saddr) 1113 return 0; 1114 1115 if (sysctl_ip_dynaddr > 1) { 1116 printk(KERN_INFO "%s(): shifting inet->saddr from %pI4 to %pI4\n", 1117 __func__, &old_saddr, &new_saddr); 1118 } 1119 1120 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr; 1121 1122 /* 1123 * XXX The only one ugly spot where we need to 1124 * XXX really change the sockets identity after 1125 * XXX it has entered the hashes. -DaveM 1126 * 1127 * Besides that, it does not check for connection 1128 * uniqueness. Wait for troubles. 1129 */ 1130 __sk_prot_rehash(sk); 1131 return 0; 1132 } 1133 1134 int inet_sk_rebuild_header(struct sock *sk) 1135 { 1136 struct inet_sock *inet = inet_sk(sk); 1137 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0); 1138 __be32 daddr; 1139 int err; 1140 1141 /* Route is OK, nothing to do. */ 1142 if (rt) 1143 return 0; 1144 1145 /* Reroute. */ 1146 daddr = inet->inet_daddr; 1147 if (inet->opt && inet->opt->srr) 1148 daddr = inet->opt->faddr; 1149 { 1150 struct flowi fl = { 1151 .oif = sk->sk_bound_dev_if, 1152 .mark = sk->sk_mark, 1153 .nl_u = { 1154 .ip4_u = { 1155 .daddr = daddr, 1156 .saddr = inet->inet_saddr, 1157 .tos = RT_CONN_FLAGS(sk), 1158 }, 1159 }, 1160 .proto = sk->sk_protocol, 1161 .flags = inet_sk_flowi_flags(sk), 1162 .uli_u = { 1163 .ports = { 1164 .sport = inet->inet_sport, 1165 .dport = inet->inet_dport, 1166 }, 1167 }, 1168 }; 1169 1170 security_sk_classify_flow(sk, &fl); 1171 err = ip_route_output_flow(sock_net(sk), &rt, &fl, sk, 0); 1172 } 1173 if (!err) 1174 sk_setup_caps(sk, &rt->dst); 1175 else { 1176 /* Routing failed... */ 1177 sk->sk_route_caps = 0; 1178 /* 1179 * Other protocols have to map its equivalent state to TCP_SYN_SENT. 1180 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme 1181 */ 1182 if (!sysctl_ip_dynaddr || 1183 sk->sk_state != TCP_SYN_SENT || 1184 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) || 1185 (err = inet_sk_reselect_saddr(sk)) != 0) 1186 sk->sk_err_soft = -err; 1187 } 1188 1189 return err; 1190 } 1191 EXPORT_SYMBOL(inet_sk_rebuild_header); 1192 1193 static int inet_gso_send_check(struct sk_buff *skb) 1194 { 1195 struct iphdr *iph; 1196 const struct net_protocol *ops; 1197 int proto; 1198 int ihl; 1199 int err = -EINVAL; 1200 1201 if (unlikely(!pskb_may_pull(skb, sizeof(*iph)))) 1202 goto out; 1203 1204 iph = ip_hdr(skb); 1205 ihl = iph->ihl * 4; 1206 if (ihl < sizeof(*iph)) 1207 goto out; 1208 1209 if (unlikely(!pskb_may_pull(skb, ihl))) 1210 goto out; 1211 1212 __skb_pull(skb, ihl); 1213 skb_reset_transport_header(skb); 1214 iph = ip_hdr(skb); 1215 proto = iph->protocol & (MAX_INET_PROTOS - 1); 1216 err = -EPROTONOSUPPORT; 1217 1218 rcu_read_lock(); 1219 ops = rcu_dereference(inet_protos[proto]); 1220 if (likely(ops && ops->gso_send_check)) 1221 err = ops->gso_send_check(skb); 1222 rcu_read_unlock(); 1223 1224 out: 1225 return err; 1226 } 1227 1228 static struct sk_buff *inet_gso_segment(struct sk_buff *skb, int features) 1229 { 1230 struct sk_buff *segs = ERR_PTR(-EINVAL); 1231 struct iphdr *iph; 1232 const struct net_protocol *ops; 1233 int proto; 1234 int ihl; 1235 int id; 1236 unsigned int offset = 0; 1237 1238 if (!(features & NETIF_F_V4_CSUM)) 1239 features &= ~NETIF_F_SG; 1240 1241 if (unlikely(skb_shinfo(skb)->gso_type & 1242 ~(SKB_GSO_TCPV4 | 1243 SKB_GSO_UDP | 1244 SKB_GSO_DODGY | 1245 SKB_GSO_TCP_ECN | 1246 0))) 1247 goto out; 1248 1249 if (unlikely(!pskb_may_pull(skb, sizeof(*iph)))) 1250 goto out; 1251 1252 iph = ip_hdr(skb); 1253 ihl = iph->ihl * 4; 1254 if (ihl < sizeof(*iph)) 1255 goto out; 1256 1257 if (unlikely(!pskb_may_pull(skb, ihl))) 1258 goto out; 1259 1260 __skb_pull(skb, ihl); 1261 skb_reset_transport_header(skb); 1262 iph = ip_hdr(skb); 1263 id = ntohs(iph->id); 1264 proto = iph->protocol & (MAX_INET_PROTOS - 1); 1265 segs = ERR_PTR(-EPROTONOSUPPORT); 1266 1267 rcu_read_lock(); 1268 ops = rcu_dereference(inet_protos[proto]); 1269 if (likely(ops && ops->gso_segment)) 1270 segs = ops->gso_segment(skb, features); 1271 rcu_read_unlock(); 1272 1273 if (!segs || IS_ERR(segs)) 1274 goto out; 1275 1276 skb = segs; 1277 do { 1278 iph = ip_hdr(skb); 1279 if (proto == IPPROTO_UDP) { 1280 iph->id = htons(id); 1281 iph->frag_off = htons(offset >> 3); 1282 if (skb->next != NULL) 1283 iph->frag_off |= htons(IP_MF); 1284 offset += (skb->len - skb->mac_len - iph->ihl * 4); 1285 } else 1286 iph->id = htons(id++); 1287 iph->tot_len = htons(skb->len - skb->mac_len); 1288 iph->check = 0; 1289 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl); 1290 } while ((skb = skb->next)); 1291 1292 out: 1293 return segs; 1294 } 1295 1296 static struct sk_buff **inet_gro_receive(struct sk_buff **head, 1297 struct sk_buff *skb) 1298 { 1299 const struct net_protocol *ops; 1300 struct sk_buff **pp = NULL; 1301 struct sk_buff *p; 1302 struct iphdr *iph; 1303 unsigned int hlen; 1304 unsigned int off; 1305 unsigned int id; 1306 int flush = 1; 1307 int proto; 1308 1309 off = skb_gro_offset(skb); 1310 hlen = off + sizeof(*iph); 1311 iph = skb_gro_header_fast(skb, off); 1312 if (skb_gro_header_hard(skb, hlen)) { 1313 iph = skb_gro_header_slow(skb, hlen, off); 1314 if (unlikely(!iph)) 1315 goto out; 1316 } 1317 1318 proto = iph->protocol & (MAX_INET_PROTOS - 1); 1319 1320 rcu_read_lock(); 1321 ops = rcu_dereference(inet_protos[proto]); 1322 if (!ops || !ops->gro_receive) 1323 goto out_unlock; 1324 1325 if (*(u8 *)iph != 0x45) 1326 goto out_unlock; 1327 1328 if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl))) 1329 goto out_unlock; 1330 1331 id = ntohl(*(__be32 *)&iph->id); 1332 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF)); 1333 id >>= 16; 1334 1335 for (p = *head; p; p = p->next) { 1336 struct iphdr *iph2; 1337 1338 if (!NAPI_GRO_CB(p)->same_flow) 1339 continue; 1340 1341 iph2 = ip_hdr(p); 1342 1343 if ((iph->protocol ^ iph2->protocol) | 1344 (iph->tos ^ iph2->tos) | 1345 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) | 1346 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) { 1347 NAPI_GRO_CB(p)->same_flow = 0; 1348 continue; 1349 } 1350 1351 /* All fields must match except length and checksum. */ 1352 NAPI_GRO_CB(p)->flush |= 1353 (iph->ttl ^ iph2->ttl) | 1354 ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id); 1355 1356 NAPI_GRO_CB(p)->flush |= flush; 1357 } 1358 1359 NAPI_GRO_CB(skb)->flush |= flush; 1360 skb_gro_pull(skb, sizeof(*iph)); 1361 skb_set_transport_header(skb, skb_gro_offset(skb)); 1362 1363 pp = ops->gro_receive(head, skb); 1364 1365 out_unlock: 1366 rcu_read_unlock(); 1367 1368 out: 1369 NAPI_GRO_CB(skb)->flush |= flush; 1370 1371 return pp; 1372 } 1373 1374 static int inet_gro_complete(struct sk_buff *skb) 1375 { 1376 const struct net_protocol *ops; 1377 struct iphdr *iph = ip_hdr(skb); 1378 int proto = iph->protocol & (MAX_INET_PROTOS - 1); 1379 int err = -ENOSYS; 1380 __be16 newlen = htons(skb->len - skb_network_offset(skb)); 1381 1382 csum_replace2(&iph->check, iph->tot_len, newlen); 1383 iph->tot_len = newlen; 1384 1385 rcu_read_lock(); 1386 ops = rcu_dereference(inet_protos[proto]); 1387 if (WARN_ON(!ops || !ops->gro_complete)) 1388 goto out_unlock; 1389 1390 err = ops->gro_complete(skb); 1391 1392 out_unlock: 1393 rcu_read_unlock(); 1394 1395 return err; 1396 } 1397 1398 int inet_ctl_sock_create(struct sock **sk, unsigned short family, 1399 unsigned short type, unsigned char protocol, 1400 struct net *net) 1401 { 1402 struct socket *sock; 1403 int rc = sock_create_kern(family, type, protocol, &sock); 1404 1405 if (rc == 0) { 1406 *sk = sock->sk; 1407 (*sk)->sk_allocation = GFP_ATOMIC; 1408 /* 1409 * Unhash it so that IP input processing does not even see it, 1410 * we do not wish this socket to see incoming packets. 1411 */ 1412 (*sk)->sk_prot->unhash(*sk); 1413 1414 sk_change_net(*sk, net); 1415 } 1416 return rc; 1417 } 1418 EXPORT_SYMBOL_GPL(inet_ctl_sock_create); 1419 1420 unsigned long snmp_fold_field(void __percpu *mib[], int offt) 1421 { 1422 unsigned long res = 0; 1423 int i; 1424 1425 for_each_possible_cpu(i) { 1426 res += *(((unsigned long *) per_cpu_ptr(mib[0], i)) + offt); 1427 res += *(((unsigned long *) per_cpu_ptr(mib[1], i)) + offt); 1428 } 1429 return res; 1430 } 1431 EXPORT_SYMBOL_GPL(snmp_fold_field); 1432 1433 #if BITS_PER_LONG==32 1434 1435 u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset) 1436 { 1437 u64 res = 0; 1438 int cpu; 1439 1440 for_each_possible_cpu(cpu) { 1441 void *bhptr, *userptr; 1442 struct u64_stats_sync *syncp; 1443 u64 v_bh, v_user; 1444 unsigned int start; 1445 1446 /* first mib used by softirq context, we must use _bh() accessors */ 1447 bhptr = per_cpu_ptr(SNMP_STAT_BHPTR(mib), cpu); 1448 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset); 1449 do { 1450 start = u64_stats_fetch_begin_bh(syncp); 1451 v_bh = *(((u64 *) bhptr) + offt); 1452 } while (u64_stats_fetch_retry_bh(syncp, start)); 1453 1454 /* second mib used in USER context */ 1455 userptr = per_cpu_ptr(SNMP_STAT_USRPTR(mib), cpu); 1456 syncp = (struct u64_stats_sync *)(userptr + syncp_offset); 1457 do { 1458 start = u64_stats_fetch_begin(syncp); 1459 v_user = *(((u64 *) userptr) + offt); 1460 } while (u64_stats_fetch_retry(syncp, start)); 1461 1462 res += v_bh + v_user; 1463 } 1464 return res; 1465 } 1466 EXPORT_SYMBOL_GPL(snmp_fold_field64); 1467 #endif 1468 1469 int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align) 1470 { 1471 BUG_ON(ptr == NULL); 1472 ptr[0] = __alloc_percpu(mibsize, align); 1473 if (!ptr[0]) 1474 goto err0; 1475 ptr[1] = __alloc_percpu(mibsize, align); 1476 if (!ptr[1]) 1477 goto err1; 1478 return 0; 1479 err1: 1480 free_percpu(ptr[0]); 1481 ptr[0] = NULL; 1482 err0: 1483 return -ENOMEM; 1484 } 1485 EXPORT_SYMBOL_GPL(snmp_mib_init); 1486 1487 void snmp_mib_free(void __percpu *ptr[2]) 1488 { 1489 BUG_ON(ptr == NULL); 1490 free_percpu(ptr[0]); 1491 free_percpu(ptr[1]); 1492 ptr[0] = ptr[1] = NULL; 1493 } 1494 EXPORT_SYMBOL_GPL(snmp_mib_free); 1495 1496 #ifdef CONFIG_IP_MULTICAST 1497 static const struct net_protocol igmp_protocol = { 1498 .handler = igmp_rcv, 1499 .netns_ok = 1, 1500 }; 1501 #endif 1502 1503 static const struct net_protocol tcp_protocol = { 1504 .handler = tcp_v4_rcv, 1505 .err_handler = tcp_v4_err, 1506 .gso_send_check = tcp_v4_gso_send_check, 1507 .gso_segment = tcp_tso_segment, 1508 .gro_receive = tcp4_gro_receive, 1509 .gro_complete = tcp4_gro_complete, 1510 .no_policy = 1, 1511 .netns_ok = 1, 1512 }; 1513 1514 static const struct net_protocol udp_protocol = { 1515 .handler = udp_rcv, 1516 .err_handler = udp_err, 1517 .gso_send_check = udp4_ufo_send_check, 1518 .gso_segment = udp4_ufo_fragment, 1519 .no_policy = 1, 1520 .netns_ok = 1, 1521 }; 1522 1523 static const struct net_protocol icmp_protocol = { 1524 .handler = icmp_rcv, 1525 .no_policy = 1, 1526 .netns_ok = 1, 1527 }; 1528 1529 static __net_init int ipv4_mib_init_net(struct net *net) 1530 { 1531 if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics, 1532 sizeof(struct tcp_mib), 1533 __alignof__(struct tcp_mib)) < 0) 1534 goto err_tcp_mib; 1535 if (snmp_mib_init((void __percpu **)net->mib.ip_statistics, 1536 sizeof(struct ipstats_mib), 1537 __alignof__(struct ipstats_mib)) < 0) 1538 goto err_ip_mib; 1539 if (snmp_mib_init((void __percpu **)net->mib.net_statistics, 1540 sizeof(struct linux_mib), 1541 __alignof__(struct linux_mib)) < 0) 1542 goto err_net_mib; 1543 if (snmp_mib_init((void __percpu **)net->mib.udp_statistics, 1544 sizeof(struct udp_mib), 1545 __alignof__(struct udp_mib)) < 0) 1546 goto err_udp_mib; 1547 if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics, 1548 sizeof(struct udp_mib), 1549 __alignof__(struct udp_mib)) < 0) 1550 goto err_udplite_mib; 1551 if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics, 1552 sizeof(struct icmp_mib), 1553 __alignof__(struct icmp_mib)) < 0) 1554 goto err_icmp_mib; 1555 if (snmp_mib_init((void __percpu **)net->mib.icmpmsg_statistics, 1556 sizeof(struct icmpmsg_mib), 1557 __alignof__(struct icmpmsg_mib)) < 0) 1558 goto err_icmpmsg_mib; 1559 1560 tcp_mib_init(net); 1561 return 0; 1562 1563 err_icmpmsg_mib: 1564 snmp_mib_free((void __percpu **)net->mib.icmp_statistics); 1565 err_icmp_mib: 1566 snmp_mib_free((void __percpu **)net->mib.udplite_statistics); 1567 err_udplite_mib: 1568 snmp_mib_free((void __percpu **)net->mib.udp_statistics); 1569 err_udp_mib: 1570 snmp_mib_free((void __percpu **)net->mib.net_statistics); 1571 err_net_mib: 1572 snmp_mib_free((void __percpu **)net->mib.ip_statistics); 1573 err_ip_mib: 1574 snmp_mib_free((void __percpu **)net->mib.tcp_statistics); 1575 err_tcp_mib: 1576 return -ENOMEM; 1577 } 1578 1579 static __net_exit void ipv4_mib_exit_net(struct net *net) 1580 { 1581 snmp_mib_free((void __percpu **)net->mib.icmpmsg_statistics); 1582 snmp_mib_free((void __percpu **)net->mib.icmp_statistics); 1583 snmp_mib_free((void __percpu **)net->mib.udplite_statistics); 1584 snmp_mib_free((void __percpu **)net->mib.udp_statistics); 1585 snmp_mib_free((void __percpu **)net->mib.net_statistics); 1586 snmp_mib_free((void __percpu **)net->mib.ip_statistics); 1587 snmp_mib_free((void __percpu **)net->mib.tcp_statistics); 1588 } 1589 1590 static __net_initdata struct pernet_operations ipv4_mib_ops = { 1591 .init = ipv4_mib_init_net, 1592 .exit = ipv4_mib_exit_net, 1593 }; 1594 1595 static int __init init_ipv4_mibs(void) 1596 { 1597 return register_pernet_subsys(&ipv4_mib_ops); 1598 } 1599 1600 static int ipv4_proc_init(void); 1601 1602 /* 1603 * IP protocol layer initialiser 1604 */ 1605 1606 static struct packet_type ip_packet_type __read_mostly = { 1607 .type = cpu_to_be16(ETH_P_IP), 1608 .func = ip_rcv, 1609 .gso_send_check = inet_gso_send_check, 1610 .gso_segment = inet_gso_segment, 1611 .gro_receive = inet_gro_receive, 1612 .gro_complete = inet_gro_complete, 1613 }; 1614 1615 static int __init inet_init(void) 1616 { 1617 struct sk_buff *dummy_skb; 1618 struct inet_protosw *q; 1619 struct list_head *r; 1620 int rc = -EINVAL; 1621 1622 BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb)); 1623 1624 sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL); 1625 if (!sysctl_local_reserved_ports) 1626 goto out; 1627 1628 rc = proto_register(&tcp_prot, 1); 1629 if (rc) 1630 goto out_free_reserved_ports; 1631 1632 rc = proto_register(&udp_prot, 1); 1633 if (rc) 1634 goto out_unregister_tcp_proto; 1635 1636 rc = proto_register(&raw_prot, 1); 1637 if (rc) 1638 goto out_unregister_udp_proto; 1639 1640 /* 1641 * Tell SOCKET that we are alive... 1642 */ 1643 1644 (void)sock_register(&inet_family_ops); 1645 1646 #ifdef CONFIG_SYSCTL 1647 ip_static_sysctl_init(); 1648 #endif 1649 1650 /* 1651 * Add all the base protocols. 1652 */ 1653 1654 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0) 1655 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n"); 1656 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0) 1657 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n"); 1658 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0) 1659 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n"); 1660 #ifdef CONFIG_IP_MULTICAST 1661 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0) 1662 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n"); 1663 #endif 1664 1665 /* Register the socket-side information for inet_create. */ 1666 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r) 1667 INIT_LIST_HEAD(r); 1668 1669 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q) 1670 inet_register_protosw(q); 1671 1672 /* 1673 * Set the ARP module up 1674 */ 1675 1676 arp_init(); 1677 1678 /* 1679 * Set the IP module up 1680 */ 1681 1682 ip_init(); 1683 1684 tcp_v4_init(); 1685 1686 /* Setup TCP slab cache for open requests. */ 1687 tcp_init(); 1688 1689 /* Setup UDP memory threshold */ 1690 udp_init(); 1691 1692 /* Add UDP-Lite (RFC 3828) */ 1693 udplite4_register(); 1694 1695 /* 1696 * Set the ICMP layer up 1697 */ 1698 1699 if (icmp_init() < 0) 1700 panic("Failed to create the ICMP control socket.\n"); 1701 1702 /* 1703 * Initialise the multicast router 1704 */ 1705 #if defined(CONFIG_IP_MROUTE) 1706 if (ip_mr_init()) 1707 printk(KERN_CRIT "inet_init: Cannot init ipv4 mroute\n"); 1708 #endif 1709 /* 1710 * Initialise per-cpu ipv4 mibs 1711 */ 1712 1713 if (init_ipv4_mibs()) 1714 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n"); 1715 1716 ipv4_proc_init(); 1717 1718 ipfrag_init(); 1719 1720 dev_add_pack(&ip_packet_type); 1721 1722 rc = 0; 1723 out: 1724 return rc; 1725 out_unregister_udp_proto: 1726 proto_unregister(&udp_prot); 1727 out_unregister_tcp_proto: 1728 proto_unregister(&tcp_prot); 1729 out_free_reserved_ports: 1730 kfree(sysctl_local_reserved_ports); 1731 goto out; 1732 } 1733 1734 fs_initcall(inet_init); 1735 1736 /* ------------------------------------------------------------------------ */ 1737 1738 #ifdef CONFIG_PROC_FS 1739 static int __init ipv4_proc_init(void) 1740 { 1741 int rc = 0; 1742 1743 if (raw_proc_init()) 1744 goto out_raw; 1745 if (tcp4_proc_init()) 1746 goto out_tcp; 1747 if (udp4_proc_init()) 1748 goto out_udp; 1749 if (ip_misc_proc_init()) 1750 goto out_misc; 1751 out: 1752 return rc; 1753 out_misc: 1754 udp4_proc_exit(); 1755 out_udp: 1756 tcp4_proc_exit(); 1757 out_tcp: 1758 raw_proc_exit(); 1759 out_raw: 1760 rc = -ENOMEM; 1761 goto out; 1762 } 1763 1764 #else /* CONFIG_PROC_FS */ 1765 static int __init ipv4_proc_init(void) 1766 { 1767 return 0; 1768 } 1769 #endif /* CONFIG_PROC_FS */ 1770 1771 MODULE_ALIAS_NETPROTO(PF_INET); 1772 1773