net/ipv4/af_inet.c

   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 && inet_autobind(sk))
 731                 return -EAGAIN;
 732
 733         return sk->sk_prot->sendmsg(iocb, sk, msg, size);
 734 }
 735 EXPORT_SYMBOL(inet_sendmsg);
 736
 737 static ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
 738                              size_t size, int flags)
 739 {
 740         struct sock *sk = sock->sk;
 741
 742         sock_rps_record_flow(sk);
 743
 744         /* We may need to bind the socket. */
 745         if (!inet_sk(sk)->inet_num && inet_autobind(sk))
 746                 return -EAGAIN;
 747
 748         if (sk->sk_prot->sendpage)
 749                 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
 750         return sock_no_sendpage(sock, page, offset, size, flags);
 751 }
 752
 753 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
 754                  size_t size, int flags)
 755 {
 756         struct sock *sk = sock->sk;
 757         int addr_len = 0;
 758         int err;
 759
 760         sock_rps_record_flow(sk);
 761
 762         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
 763                                    flags & ~MSG_DONTWAIT, &addr_len);
 764         if (err >= 0)
 765                 msg->msg_namelen = addr_len;
 766         return err;
 767 }
 768 EXPORT_SYMBOL(inet_recvmsg);
 769
 770 int inet_shutdown(struct socket *sock, int how)
 771 {
 772         struct sock *sk = sock->sk;
 773         int err = 0;
 774
 775         /* This should really check to make sure
 776          * the socket is a TCP socket. (WHY AC...)
 777          */
 778         how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
 779                        1->2 bit 2 snds.
 780                        2->3 */
 781         if ((how & ~SHUTDOWN_MASK) || !how)     /* MAXINT->0 */
 782                 return -EINVAL;
 783
 784         lock_sock(sk);
 785         if (sock->state == SS_CONNECTING) {
 786                 if ((1 << sk->sk_state) &
 787                     (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
 788                         sock->state = SS_DISCONNECTING;
 789                 else
 790                         sock->state = SS_CONNECTED;
 791         }
 792
 793         switch (sk->sk_state) {
 794         case TCP_CLOSE:
 795                 err = -ENOTCONN;
 796                 /* Hack to wake up other listeners, who can poll for
 797                    POLLHUP, even on eg. unconnected UDP sockets -- RR */
 798         default:
 799                 sk->sk_shutdown |= how;
 800                 if (sk->sk_prot->shutdown)
 801                         sk->sk_prot->shutdown(sk, how);
 802                 break;
 803
 804         /* Remaining two branches are temporary solution for missing
 805          * close() in multithreaded environment. It is _not_ a good idea,
 806          * but we have no choice until close() is repaired at VFS level.
 807          */
 808         case TCP_LISTEN:
 809                 if (!(how & RCV_SHUTDOWN))
 810                         break;
 811                 /* Fall through */
 812         case TCP_SYN_SENT:
 813                 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
 814                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 815                 break;
 816         }
 817
 818         /* Wake up anyone sleeping in poll. */
 819         sk->sk_state_change(sk);
 820         release_sock(sk);
 821         return err;
 822 }
 823 EXPORT_SYMBOL(inet_shutdown);
 824
 825 /*
 826  *      ioctl() calls you can issue on an INET socket. Most of these are
 827  *      device configuration and stuff and very rarely used. Some ioctls
 828  *      pass on to the socket itself.
 829  *
 830  *      NOTE: I like the idea of a module for the config stuff. ie ifconfig
 831  *      loads the devconfigure module does its configuring and unloads it.
 832  *      There's a good 20K of config code hanging around the kernel.
 833  */
 834
 835 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 836 {
 837         struct sock *sk = sock->sk;
 838         int err = 0;
 839         struct net *net = sock_net(sk);
 840
 841         switch (cmd) {
 842         case SIOCGSTAMP:
 843                 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
 844                 break;
 845         case SIOCGSTAMPNS:
 846                 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
 847                 break;
 848         case SIOCADDRT:
 849         case SIOCDELRT:
 850         case SIOCRTMSG:
 851                 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
 852                 break;
 853         case SIOCDARP:
 854         case SIOCGARP:
 855         case SIOCSARP:
 856                 err = arp_ioctl(net, cmd, (void __user *)arg);
 857                 break;
 858         case SIOCGIFADDR:
 859         case SIOCSIFADDR:
 860         case SIOCGIFBRDADDR:
 861         case SIOCSIFBRDADDR:
 862         case SIOCGIFNETMASK:
 863         case SIOCSIFNETMASK:
 864         case SIOCGIFDSTADDR:
 865         case SIOCSIFDSTADDR:
 866         case SIOCSIFPFLAGS:
 867         case SIOCGIFPFLAGS:
 868         case SIOCSIFFLAGS:
 869                 err = devinet_ioctl(net, cmd, (void __user *)arg);
 870                 break;
 871         default:
 872                 if (sk->sk_prot->ioctl)
 873                         err = sk->sk_prot->ioctl(sk, cmd, arg);
 874                 else
 875                         err = -ENOIOCTLCMD;
 876                 break;
 877         }
 878         return err;
 879 }
 880 EXPORT_SYMBOL(inet_ioctl);
 881
 882 const struct proto_ops inet_stream_ops = {
 883         .family            = PF_INET,
 884         .owner             = THIS_MODULE,
 885         .release           = inet_release,
 886         .bind              = inet_bind,
 887         .connect           = inet_stream_connect,
 888         .socketpair        = sock_no_socketpair,
 889         .accept            = inet_accept,
 890         .getname           = inet_getname,
 891         .poll              = tcp_poll,
 892         .ioctl             = inet_ioctl,
 893         .listen            = inet_listen,
 894         .shutdown          = inet_shutdown,
 895         .setsockopt        = sock_common_setsockopt,
 896         .getsockopt        = sock_common_getsockopt,
 897         .sendmsg           = tcp_sendmsg,
 898         .recvmsg           = inet_recvmsg,
 899         .mmap              = sock_no_mmap,
 900         .sendpage          = tcp_sendpage,
 901         .splice_read       = tcp_splice_read,
 902 #ifdef CONFIG_COMPAT
 903         .compat_setsockopt = compat_sock_common_setsockopt,
 904         .compat_getsockopt = compat_sock_common_getsockopt,
 905 #endif
 906 };
 907 EXPORT_SYMBOL(inet_stream_ops);
 908
 909 const struct proto_ops inet_dgram_ops = {
 910         .family            = PF_INET,
 911         .owner             = THIS_MODULE,
 912         .release           = inet_release,
 913         .bind              = inet_bind,
 914         .connect           = inet_dgram_connect,
 915         .socketpair        = sock_no_socketpair,
 916         .accept            = sock_no_accept,
 917         .getname           = inet_getname,
 918         .poll              = udp_poll,
 919         .ioctl             = inet_ioctl,
 920         .listen            = sock_no_listen,
 921         .shutdown          = inet_shutdown,
 922         .setsockopt        = sock_common_setsockopt,
 923         .getsockopt        = sock_common_getsockopt,
 924         .sendmsg           = inet_sendmsg,
 925         .recvmsg           = inet_recvmsg,
 926         .mmap              = sock_no_mmap,
 927         .sendpage          = inet_sendpage,
 928 #ifdef CONFIG_COMPAT
 929         .compat_setsockopt = compat_sock_common_setsockopt,
 930         .compat_getsockopt = compat_sock_common_getsockopt,
 931 #endif
 932 };
 933 EXPORT_SYMBOL(inet_dgram_ops);
 934
 935 /*
 936  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
 937  * udp_poll
 938  */
 939 static const struct proto_ops inet_sockraw_ops = {
 940         .family            = PF_INET,
 941         .owner             = THIS_MODULE,
 942         .release           = inet_release,
 943         .bind              = inet_bind,
 944         .connect           = inet_dgram_connect,
 945         .socketpair        = sock_no_socketpair,
 946         .accept            = sock_no_accept,
 947         .getname           = inet_getname,
 948         .poll              = datagram_poll,
 949         .ioctl             = inet_ioctl,
 950         .listen            = sock_no_listen,
 951         .shutdown          = inet_shutdown,
 952         .setsockopt        = sock_common_setsockopt,
 953         .getsockopt        = sock_common_getsockopt,
 954         .sendmsg           = inet_sendmsg,
 955         .recvmsg           = inet_recvmsg,
 956         .mmap              = sock_no_mmap,
 957         .sendpage          = inet_sendpage,
 958 #ifdef CONFIG_COMPAT
 959         .compat_setsockopt = compat_sock_common_setsockopt,
 960         .compat_getsockopt = compat_sock_common_getsockopt,
 961 #endif
 962 };
 963
 964 static const struct net_proto_family inet_family_ops = {
 965         .family = PF_INET,
 966         .create = inet_create,
 967         .owner  = THIS_MODULE,
 968 };
 969
 970 /* Upon startup we insert all the elements in inetsw_array[] into
 971  * the linked list inetsw.
 972  */
 973 static struct inet_protosw inetsw_array[] =
 974 {
 975         {
 976                 .type =       SOCK_STREAM,
 977                 .protocol =   IPPROTO_TCP,
 978                 .prot =       &tcp_prot,
 979                 .ops =        &inet_stream_ops,
 980                 .no_check =   0,
 981                 .flags =      INET_PROTOSW_PERMANENT |
 982                               INET_PROTOSW_ICSK,
 983         },
 984
 985         {
 986                 .type =       SOCK_DGRAM,
 987                 .protocol =   IPPROTO_UDP,
 988                 .prot =       &udp_prot,
 989                 .ops =        &inet_dgram_ops,
 990                 .no_check =   UDP_CSUM_DEFAULT,
 991                 .flags =      INET_PROTOSW_PERMANENT,
 992        },
 993
 994
 995        {
 996                .type =       SOCK_RAW,
 997                .protocol =   IPPROTO_IP,        /* wild card */
 998                .prot =       &raw_prot,
 999                .ops =        &inet_sockraw_ops,
1000                .no_check =   UDP_CSUM_DEFAULT,
1001                .flags =      INET_PROTOSW_REUSE,
1002        }
1003 };
1004
1005 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1006
1007 void inet_register_protosw(struct inet_protosw *p)
1008 {
1009         struct list_head *lh;
1010         struct inet_protosw *answer;
1011         int protocol = p->protocol;
1012         struct list_head *last_perm;
1013
1014         spin_lock_bh(&inetsw_lock);
1015
1016         if (p->type >= SOCK_MAX)
1017                 goto out_illegal;
1018
1019         /* If we are trying to override a permanent protocol, bail. */
1020         answer = NULL;
1021         last_perm = &inetsw[p->type];
1022         list_for_each(lh, &inetsw[p->type]) {
1023                 answer = list_entry(lh, struct inet_protosw, list);
1024
1025                 /* Check only the non-wild match. */
1026                 if (INET_PROTOSW_PERMANENT & answer->flags) {
1027                         if (protocol == answer->protocol)
1028                                 break;
1029                         last_perm = lh;
1030                 }
1031
1032                 answer = NULL;
1033         }
1034         if (answer)
1035                 goto out_permanent;
1036
1037         /* Add the new entry after the last permanent entry if any, so that
1038          * the new entry does not override a permanent entry when matched with
1039          * a wild-card protocol. But it is allowed to override any existing
1040          * non-permanent entry.  This means that when we remove this entry, the
1041          * system automatically returns to the old behavior.
1042          */
1043         list_add_rcu(&p->list, last_perm);
1044 out:
1045         spin_unlock_bh(&inetsw_lock);
1046
1047         return;
1048
1049 out_permanent:
1050         printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
1051                protocol);
1052         goto out;
1053
1054 out_illegal:
1055         printk(KERN_ERR
1056                "Ignoring attempt to register invalid socket type %d.\n",
1057                p->type);
1058         goto out;
1059 }
1060 EXPORT_SYMBOL(inet_register_protosw);
1061
1062 void inet_unregister_protosw(struct inet_protosw *p)
1063 {
1064         if (INET_PROTOSW_PERMANENT & p->flags) {
1065                 printk(KERN_ERR
1066                        "Attempt to unregister permanent protocol %d.\n",
1067                        p->protocol);
1068         } else {
1069                 spin_lock_bh(&inetsw_lock);
1070                 list_del_rcu(&p->list);
1071                 spin_unlock_bh(&inetsw_lock);
1072
1073                 synchronize_net();
1074         }
1075 }
1076 EXPORT_SYMBOL(inet_unregister_protosw);
1077
1078 /*
1079  *      Shall we try to damage output packets if routing dev changes?
1080  */
1081
1082 int sysctl_ip_dynaddr __read_mostly;
1083
1084 static int inet_sk_reselect_saddr(struct sock *sk)
1085 {
1086         struct inet_sock *inet = inet_sk(sk);
1087         int err;
1088         struct rtable *rt;
1089         __be32 old_saddr = inet->inet_saddr;
1090         __be32 new_saddr;
1091         __be32 daddr = inet->inet_daddr;
1092
1093         if (inet->opt && inet->opt->srr)
1094                 daddr = inet->opt->faddr;
1095
1096         /* Query new route. */
1097         err = ip_route_connect(&rt, daddr, 0,
1098                                RT_CONN_FLAGS(sk),
1099                                sk->sk_bound_dev_if,
1100                                sk->sk_protocol,
1101                                inet->inet_sport, inet->inet_dport, sk, 0);
1102         if (err)
1103                 return err;
1104
1105         sk_setup_caps(sk, &rt->dst);
1106
1107         new_saddr = rt->rt_src;
1108
1109         if (new_saddr == old_saddr)
1110                 return 0;
1111
1112         if (sysctl_ip_dynaddr > 1) {
1113                 printk(KERN_INFO "%s(): shifting inet->saddr from %pI4 to %pI4\n",
1114                        __func__, &old_saddr, &new_saddr);
1115         }
1116
1117         inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1118
1119         /*
1120          * XXX The only one ugly spot where we need to
1121          * XXX really change the sockets identity after
1122          * XXX it has entered the hashes. -DaveM
1123          *
1124          * Besides that, it does not check for connection
1125          * uniqueness. Wait for troubles.
1126          */
1127         __sk_prot_rehash(sk);
1128         return 0;
1129 }
1130
1131 int inet_sk_rebuild_header(struct sock *sk)
1132 {
1133         struct inet_sock *inet = inet_sk(sk);
1134         struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1135         __be32 daddr;
1136         int err;
1137
1138         /* Route is OK, nothing to do. */
1139         if (rt)
1140                 return 0;
1141
1142         /* Reroute. */
1143         daddr = inet->inet_daddr;
1144         if (inet->opt && inet->opt->srr)
1145                 daddr = inet->opt->faddr;
1146 {
1147         struct flowi fl = {
1148                 .oif = sk->sk_bound_dev_if,
1149                 .mark = sk->sk_mark,
1150                 .nl_u = {
1151                         .ip4_u = {
1152                                 .daddr  = daddr,
1153                                 .saddr  = inet->inet_saddr,
1154                                 .tos    = RT_CONN_FLAGS(sk),
1155                         },
1156                 },
1157                 .proto = sk->sk_protocol,
1158                 .flags = inet_sk_flowi_flags(sk),
1159                 .uli_u = {
1160                         .ports = {
1161                                 .sport = inet->inet_sport,
1162                                 .dport = inet->inet_dport,
1163                         },
1164                 },
1165         };
1166
1167         security_sk_classify_flow(sk, &fl);
1168         err = ip_route_output_flow(sock_net(sk), &rt, &fl, sk, 0);
1169 }
1170         if (!err)
1171                 sk_setup_caps(sk, &rt->dst);
1172         else {
1173                 /* Routing failed... */
1174                 sk->sk_route_caps = 0;
1175                 /*
1176                  * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1177                  * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1178                  */
1179                 if (!sysctl_ip_dynaddr ||
1180                     sk->sk_state != TCP_SYN_SENT ||
1181                     (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1182                     (err = inet_sk_reselect_saddr(sk)) != 0)
1183                         sk->sk_err_soft = -err;
1184         }
1185
1186         return err;
1187 }
1188 EXPORT_SYMBOL(inet_sk_rebuild_header);
1189
1190 static int inet_gso_send_check(struct sk_buff *skb)
1191 {
1192         struct iphdr *iph;
1193         const struct net_protocol *ops;
1194         int proto;
1195         int ihl;
1196         int err = -EINVAL;
1197
1198         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1199                 goto out;
1200
1201         iph = ip_hdr(skb);
1202         ihl = iph->ihl * 4;
1203         if (ihl < sizeof(*iph))
1204                 goto out;
1205
1206         if (unlikely(!pskb_may_pull(skb, ihl)))
1207                 goto out;
1208
1209         __skb_pull(skb, ihl);
1210         skb_reset_transport_header(skb);
1211         iph = ip_hdr(skb);
1212         proto = iph->protocol & (MAX_INET_PROTOS - 1);
1213         err = -EPROTONOSUPPORT;
1214
1215         rcu_read_lock();
1216         ops = rcu_dereference(inet_protos[proto]);
1217         if (likely(ops && ops->gso_send_check))
1218                 err = ops->gso_send_check(skb);
1219         rcu_read_unlock();
1220
1221 out:
1222         return err;
1223 }
1224
1225 static struct sk_buff *inet_gso_segment(struct sk_buff *skb, int features)
1226 {
1227         struct sk_buff *segs = ERR_PTR(-EINVAL);
1228         struct iphdr *iph;
1229         const struct net_protocol *ops;
1230         int proto;
1231         int ihl;
1232         int id;
1233         unsigned int offset = 0;
1234
1235         if (!(features & NETIF_F_V4_CSUM))
1236                 features &= ~NETIF_F_SG;
1237
1238         if (unlikely(skb_shinfo(skb)->gso_type &
1239                      ~(SKB_GSO_TCPV4 |
1240                        SKB_GSO_UDP |
1241                        SKB_GSO_DODGY |
1242                        SKB_GSO_TCP_ECN |
1243                        0)))
1244                 goto out;
1245
1246         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1247                 goto out;
1248
1249         iph = ip_hdr(skb);
1250         ihl = iph->ihl * 4;
1251         if (ihl < sizeof(*iph))
1252                 goto out;
1253
1254         if (unlikely(!pskb_may_pull(skb, ihl)))
1255                 goto out;
1256
1257         __skb_pull(skb, ihl);
1258         skb_reset_transport_header(skb);
1259         iph = ip_hdr(skb);
1260         id = ntohs(iph->id);
1261         proto = iph->protocol & (MAX_INET_PROTOS - 1);
1262         segs = ERR_PTR(-EPROTONOSUPPORT);
1263
1264         rcu_read_lock();
1265         ops = rcu_dereference(inet_protos[proto]);
1266         if (likely(ops && ops->gso_segment))
1267                 segs = ops->gso_segment(skb, features);
1268         rcu_read_unlock();
1269
1270         if (!segs || IS_ERR(segs))
1271                 goto out;
1272
1273         skb = segs;
1274         do {
1275                 iph = ip_hdr(skb);
1276                 if (proto == IPPROTO_UDP) {
1277                         iph->id = htons(id);
1278                         iph->frag_off = htons(offset >> 3);
1279                         if (skb->next != NULL)
1280                                 iph->frag_off |= htons(IP_MF);
1281                         offset += (skb->len - skb->mac_len - iph->ihl * 4);
1282                 } else
1283                         iph->id = htons(id++);
1284                 iph->tot_len = htons(skb->len - skb->mac_len);
1285                 iph->check = 0;
1286                 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1287         } while ((skb = skb->next));
1288
1289 out:
1290         return segs;
1291 }
1292
1293 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1294                                          struct sk_buff *skb)
1295 {
1296         const struct net_protocol *ops;
1297         struct sk_buff **pp = NULL;
1298         struct sk_buff *p;
1299         struct iphdr *iph;
1300         unsigned int hlen;
1301         unsigned int off;
1302         unsigned int id;
1303         int flush = 1;
1304         int proto;
1305
1306         off = skb_gro_offset(skb);
1307         hlen = off + sizeof(*iph);
1308         iph = skb_gro_header_fast(skb, off);
1309         if (skb_gro_header_hard(skb, hlen)) {
1310                 iph = skb_gro_header_slow(skb, hlen, off);
1311                 if (unlikely(!iph))
1312                         goto out;
1313         }
1314
1315         proto = iph->protocol & (MAX_INET_PROTOS - 1);
1316
1317         rcu_read_lock();
1318         ops = rcu_dereference(inet_protos[proto]);
1319         if (!ops || !ops->gro_receive)
1320                 goto out_unlock;
1321
1322         if (*(u8 *)iph != 0x45)
1323                 goto out_unlock;
1324
1325         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1326                 goto out_unlock;
1327
1328         id = ntohl(*(__be32 *)&iph->id);
1329         flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF));
1330         id >>= 16;
1331
1332         for (p = *head; p; p = p->next) {
1333                 struct iphdr *iph2;
1334
1335                 if (!NAPI_GRO_CB(p)->same_flow)
1336                         continue;
1337
1338                 iph2 = ip_hdr(p);
1339
1340                 if ((iph->protocol ^ iph2->protocol) |
1341                     (iph->tos ^ iph2->tos) |
1342                     ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1343                     ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1344                         NAPI_GRO_CB(p)->same_flow = 0;
1345                         continue;
1346                 }
1347
1348                 /* All fields must match except length and checksum. */
1349                 NAPI_GRO_CB(p)->flush |=
1350                         (iph->ttl ^ iph2->ttl) |
1351                         ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1352
1353                 NAPI_GRO_CB(p)->flush |= flush;
1354         }
1355
1356         NAPI_GRO_CB(skb)->flush |= flush;
1357         skb_gro_pull(skb, sizeof(*iph));
1358         skb_set_transport_header(skb, skb_gro_offset(skb));
1359
1360         pp = ops->gro_receive(head, skb);
1361
1362 out_unlock:
1363         rcu_read_unlock();
1364
1365 out:
1366         NAPI_GRO_CB(skb)->flush |= flush;
1367
1368         return pp;
1369 }
1370
1371 static int inet_gro_complete(struct sk_buff *skb)
1372 {
1373         const struct net_protocol *ops;
1374         struct iphdr *iph = ip_hdr(skb);
1375         int proto = iph->protocol & (MAX_INET_PROTOS - 1);
1376         int err = -ENOSYS;
1377         __be16 newlen = htons(skb->len - skb_network_offset(skb));
1378
1379         csum_replace2(&iph->check, iph->tot_len, newlen);
1380         iph->tot_len = newlen;
1381
1382         rcu_read_lock();
1383         ops = rcu_dereference(inet_protos[proto]);
1384         if (WARN_ON(!ops || !ops->gro_complete))
1385                 goto out_unlock;
1386
1387         err = ops->gro_complete(skb);
1388
1389 out_unlock:
1390         rcu_read_unlock();
1391
1392         return err;
1393 }
1394
1395 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1396                          unsigned short type, unsigned char protocol,
1397                          struct net *net)
1398 {
1399         struct socket *sock;
1400         int rc = sock_create_kern(family, type, protocol, &sock);
1401
1402         if (rc == 0) {
1403                 *sk = sock->sk;
1404                 (*sk)->sk_allocation = GFP_ATOMIC;
1405                 /*
1406                  * Unhash it so that IP input processing does not even see it,
1407                  * we do not wish this socket to see incoming packets.
1408                  */
1409                 (*sk)->sk_prot->unhash(*sk);
1410
1411                 sk_change_net(*sk, net);
1412         }
1413         return rc;
1414 }
1415 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1416
1417 unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1418 {
1419         unsigned long res = 0;
1420         int i;
1421
1422         for_each_possible_cpu(i) {
1423                 res += *(((unsigned long *) per_cpu_ptr(mib[0], i)) + offt);
1424                 res += *(((unsigned long *) per_cpu_ptr(mib[1], i)) + offt);
1425         }
1426         return res;
1427 }
1428 EXPORT_SYMBOL_GPL(snmp_fold_field);
1429
1430 #if BITS_PER_LONG==32
1431
1432 u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset)
1433 {
1434         u64 res = 0;
1435         int cpu;
1436
1437         for_each_possible_cpu(cpu) {
1438                 void *bhptr, *userptr;
1439                 struct u64_stats_sync *syncp;
1440                 u64 v_bh, v_user;
1441                 unsigned int start;
1442
1443                 /* first mib used by softirq context, we must use _bh() accessors */
1444                 bhptr = per_cpu_ptr(SNMP_STAT_BHPTR(mib), cpu);
1445                 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1446                 do {
1447                         start = u64_stats_fetch_begin_bh(syncp);
1448                         v_bh = *(((u64 *) bhptr) + offt);
1449                 } while (u64_stats_fetch_retry_bh(syncp, start));
1450
1451                 /* second mib used in USER context */
1452                 userptr = per_cpu_ptr(SNMP_STAT_USRPTR(mib), cpu);
1453                 syncp = (struct u64_stats_sync *)(userptr + syncp_offset);
1454                 do {
1455                         start = u64_stats_fetch_begin(syncp);
1456                         v_user = *(((u64 *) userptr) + offt);
1457                 } while (u64_stats_fetch_retry(syncp, start));
1458
1459                 res += v_bh + v_user;
1460         }
1461         return res;
1462 }
1463 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1464 #endif
1465
1466 int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align)
1467 {
1468         BUG_ON(ptr == NULL);
1469         ptr[0] = __alloc_percpu(mibsize, align);
1470         if (!ptr[0])
1471                 goto err0;
1472         ptr[1] = __alloc_percpu(mibsize, align);
1473         if (!ptr[1])
1474                 goto err1;
1475         return 0;
1476 err1:
1477         free_percpu(ptr[0]);
1478         ptr[0] = NULL;
1479 err0:
1480         return -ENOMEM;
1481 }
1482 EXPORT_SYMBOL_GPL(snmp_mib_init);
1483
1484 void snmp_mib_free(void __percpu *ptr[2])
1485 {
1486         BUG_ON(ptr == NULL);
1487         free_percpu(ptr[0]);
1488         free_percpu(ptr[1]);
1489         ptr[0] = ptr[1] = NULL;
1490 }
1491 EXPORT_SYMBOL_GPL(snmp_mib_free);
1492
1493 #ifdef CONFIG_IP_MULTICAST
1494 static const struct net_protocol igmp_protocol = {
1495         .handler =      igmp_rcv,
1496         .netns_ok =     1,
1497 };
1498 #endif
1499
1500 static const struct net_protocol tcp_protocol = {
1501         .handler =      tcp_v4_rcv,
1502         .err_handler =  tcp_v4_err,
1503         .gso_send_check = tcp_v4_gso_send_check,
1504         .gso_segment =  tcp_tso_segment,
1505         .gro_receive =  tcp4_gro_receive,
1506         .gro_complete = tcp4_gro_complete,
1507         .no_policy =    1,
1508         .netns_ok =     1,
1509 };
1510
1511 static const struct net_protocol udp_protocol = {
1512         .handler =      udp_rcv,
1513         .err_handler =  udp_err,
1514         .gso_send_check = udp4_ufo_send_check,
1515         .gso_segment = udp4_ufo_fragment,
1516         .no_policy =    1,
1517         .netns_ok =     1,
1518 };
1519
1520 static const struct net_protocol icmp_protocol = {
1521         .handler =      icmp_rcv,
1522         .no_policy =    1,
1523         .netns_ok =     1,
1524 };
1525
1526 static __net_init int ipv4_mib_init_net(struct net *net)
1527 {
1528         if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1529                           sizeof(struct tcp_mib),
1530                           __alignof__(struct tcp_mib)) < 0)
1531                 goto err_tcp_mib;
1532         if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1533                           sizeof(struct ipstats_mib),
1534                           __alignof__(struct ipstats_mib)) < 0)
1535                 goto err_ip_mib;
1536         if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1537                           sizeof(struct linux_mib),
1538                           __alignof__(struct linux_mib)) < 0)
1539                 goto err_net_mib;
1540         if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1541                           sizeof(struct udp_mib),
1542                           __alignof__(struct udp_mib)) < 0)
1543                 goto err_udp_mib;
1544         if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1545                           sizeof(struct udp_mib),
1546                           __alignof__(struct udp_mib)) < 0)
1547                 goto err_udplite_mib;
1548         if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1549                           sizeof(struct icmp_mib),
1550                           __alignof__(struct icmp_mib)) < 0)
1551                 goto err_icmp_mib;
1552         if (snmp_mib_init((void __percpu **)net->mib.icmpmsg_statistics,
1553                           sizeof(struct icmpmsg_mib),
1554                           __alignof__(struct icmpmsg_mib)) < 0)
1555                 goto err_icmpmsg_mib;
1556
1557         tcp_mib_init(net);
1558         return 0;
1559
1560 err_icmpmsg_mib:
1561         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1562 err_icmp_mib:
1563         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1564 err_udplite_mib:
1565         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1566 err_udp_mib:
1567         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1568 err_net_mib:
1569         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1570 err_ip_mib:
1571         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1572 err_tcp_mib:
1573         return -ENOMEM;
1574 }
1575
1576 static __net_exit void ipv4_mib_exit_net(struct net *net)
1577 {
1578         snmp_mib_free((void __percpu **)net->mib.icmpmsg_statistics);
1579         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1580         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1581         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1582         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1583         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1584         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1585 }
1586
1587 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1588         .init = ipv4_mib_init_net,
1589         .exit = ipv4_mib_exit_net,
1590 };
1591
1592 static int __init init_ipv4_mibs(void)
1593 {
1594         return register_pernet_subsys(&ipv4_mib_ops);
1595 }
1596
1597 static int ipv4_proc_init(void);
1598
1599 /*
1600  *      IP protocol layer initialiser
1601  */
1602
1603 static struct packet_type ip_packet_type __read_mostly = {
1604         .type = cpu_to_be16(ETH_P_IP),
1605         .func = ip_rcv,
1606         .gso_send_check = inet_gso_send_check,
1607         .gso_segment = inet_gso_segment,
1608         .gro_receive = inet_gro_receive,
1609         .gro_complete = inet_gro_complete,
1610 };
1611
1612 static int __init inet_init(void)
1613 {
1614         struct sk_buff *dummy_skb;
1615         struct inet_protosw *q;
1616         struct list_head *r;
1617         int rc = -EINVAL;
1618
1619         BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1620
1621         sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL);
1622         if (!sysctl_local_reserved_ports)
1623                 goto out;
1624
1625         rc = proto_register(&tcp_prot, 1);
1626         if (rc)
1627                 goto out_free_reserved_ports;
1628
1629         rc = proto_register(&udp_prot, 1);
1630         if (rc)
1631                 goto out_unregister_tcp_proto;
1632
1633         rc = proto_register(&raw_prot, 1);
1634         if (rc)
1635                 goto out_unregister_udp_proto;
1636
1637         /*
1638          *      Tell SOCKET that we are alive...
1639          */
1640
1641         (void)sock_register(&inet_family_ops);
1642
1643 #ifdef CONFIG_SYSCTL
1644         ip_static_sysctl_init();
1645 #endif
1646
1647         /*
1648          *      Add all the base protocols.
1649          */
1650
1651         if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1652                 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1653         if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1654                 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1655         if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1656                 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1657 #ifdef CONFIG_IP_MULTICAST
1658         if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1659                 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1660 #endif
1661
1662         /* Register the socket-side information for inet_create. */
1663         for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1664                 INIT_LIST_HEAD(r);
1665
1666         for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1667                 inet_register_protosw(q);
1668
1669         /*
1670          *      Set the ARP module up
1671          */
1672
1673         arp_init();
1674
1675         /*
1676          *      Set the IP module up
1677          */
1678
1679         ip_init();
1680
1681         tcp_v4_init();
1682
1683         /* Setup TCP slab cache for open requests. */
1684         tcp_init();
1685
1686         /* Setup UDP memory threshold */
1687         udp_init();
1688
1689         /* Add UDP-Lite (RFC 3828) */
1690         udplite4_register();
1691
1692         /*
1693          *      Set the ICMP layer up
1694          */
1695
1696         if (icmp_init() < 0)
1697                 panic("Failed to create the ICMP control socket.\n");
1698
1699         /*
1700          *      Initialise the multicast router
1701          */
1702 #if defined(CONFIG_IP_MROUTE)
1703         if (ip_mr_init())
1704                 printk(KERN_CRIT "inet_init: Cannot init ipv4 mroute\n");
1705 #endif
1706         /*
1707          *      Initialise per-cpu ipv4 mibs
1708          */
1709
1710         if (init_ipv4_mibs())
1711                 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n");
1712
1713         ipv4_proc_init();
1714
1715         ipfrag_init();
1716
1717         dev_add_pack(&ip_packet_type);
1718
1719         rc = 0;
1720 out:
1721         return rc;
1722 out_unregister_udp_proto:
1723         proto_unregister(&udp_prot);
1724 out_unregister_tcp_proto:
1725         proto_unregister(&tcp_prot);
1726 out_free_reserved_ports:
1727         kfree(sysctl_local_reserved_ports);
1728         goto out;
1729 }
1730
1731 fs_initcall(inet_init);
1732
1733 /* ------------------------------------------------------------------------ */
1734
1735 #ifdef CONFIG_PROC_FS
1736 static int __init ipv4_proc_init(void)
1737 {
1738         int rc = 0;
1739
1740         if (raw_proc_init())
1741                 goto out_raw;
1742         if (tcp4_proc_init())
1743                 goto out_tcp;
1744         if (udp4_proc_init())
1745                 goto out_udp;
1746         if (ip_misc_proc_init())
1747                 goto out_misc;
1748 out:
1749         return rc;
1750 out_misc:
1751         udp4_proc_exit();
1752 out_udp:
1753         tcp4_proc_exit();
1754 out_tcp:
1755         raw_proc_exit();
1756 out_raw:
1757         rc = -ENOMEM;
1758         goto out;
1759 }
1760
1761 #else /* CONFIG_PROC_FS */
1762 static int __init ipv4_proc_init(void)
1763 {
1764         return 0;
1765 }
1766 #endif /* CONFIG_PROC_FS */
1767
1768 MODULE_ALIAS_NETPROTO(PF_INET);
1769