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
  90 #include <asm/uaccess.h>
  91 #include <asm/system.h>
  92
  93 #include <linux/inet.h>
  94 #include <linux/igmp.h>
  95 #include <linux/inetdevice.h>
  96 #include <linux/netdevice.h>
  97 #include <net/checksum.h>
  98 #include <net/ip.h>
  99 #include <net/protocol.h>
 100 #include <net/arp.h>
 101 #include <net/route.h>
 102 #include <net/ip_fib.h>
 103 #include <net/inet_connection_sock.h>
 104 #include <net/tcp.h>
 105 #include <net/udp.h>
 106 #include <net/udplite.h>
 107 #include <linux/skbuff.h>
 108 #include <net/sock.h>
 109 #include <net/raw.h>
 110 #include <net/icmp.h>
 111 #include <net/ipip.h>
 112 #include <net/inet_common.h>
 113 #include <net/xfrm.h>
 114 #include <net/net_namespace.h>
 115 #ifdef CONFIG_IP_MROUTE
 116 #include <linux/mroute.h>
 117 #endif
 118
 119
 120 /* The inetsw table contains everything that inet_create needs to
 121  * build a new socket.
 122  */
 123 static struct list_head inetsw[SOCK_MAX];
 124 static DEFINE_SPINLOCK(inetsw_lock);
 125
 126 struct ipv4_config ipv4_config;
 127 EXPORT_SYMBOL(ipv4_config);
 128
 129 /* New destruction routine */
 130
 131 void inet_sock_destruct(struct sock *sk)
 132 {
 133         struct inet_sock *inet = inet_sk(sk);
 134
 135         __skb_queue_purge(&sk->sk_receive_queue);
 136         __skb_queue_purge(&sk->sk_error_queue);
 137
 138         sk_mem_reclaim(sk);
 139
 140         if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
 141                 pr_err("Attempt to release TCP socket in state %d %p\n",
 142                        sk->sk_state, sk);
 143                 return;
 144         }
 145         if (!sock_flag(sk, SOCK_DEAD)) {
 146                 pr_err("Attempt to release alive inet socket %p\n", sk);
 147                 return;
 148         }
 149
 150         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
 151         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
 152         WARN_ON(sk->sk_wmem_queued);
 153         WARN_ON(sk->sk_forward_alloc);
 154
 155         kfree(inet->opt);
 156         dst_release(sk->sk_dst_cache);
 157         sk_refcnt_debug_dec(sk);
 158 }
 159 EXPORT_SYMBOL(inet_sock_destruct);
 160
 161 /*
 162  *      The routines beyond this point handle the behaviour of an AF_INET
 163  *      socket object. Mostly it punts to the subprotocols of IP to do
 164  *      the work.
 165  */
 166
 167 /*
 168  *      Automatically bind an unbound socket.
 169  */
 170
 171 static int inet_autobind(struct sock *sk)
 172 {
 173         struct inet_sock *inet;
 174         /* We may need to bind the socket. */
 175         lock_sock(sk);
 176         inet = inet_sk(sk);
 177         if (!inet->inet_num) {
 178                 if (sk->sk_prot->get_port(sk, 0)) {
 179                         release_sock(sk);
 180                         return -EAGAIN;
 181                 }
 182                 inet->inet_sport = htons(inet->inet_num);
 183         }
 184         release_sock(sk);
 185         return 0;
 186 }
 187
 188 /*
 189  *      Move a socket into listening state.
 190  */
 191 int inet_listen(struct socket *sock, int backlog)
 192 {
 193         struct sock *sk = sock->sk;
 194         unsigned char old_state;
 195         int err;
 196
 197         lock_sock(sk);
 198
 199         err = -EINVAL;
 200         if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
 201                 goto out;
 202
 203         old_state = sk->sk_state;
 204         if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
 205                 goto out;
 206
 207         /* Really, if the socket is already in listen state
 208          * we can only allow the backlog to be adjusted.
 209          */
 210         if (old_state != TCP_LISTEN) {
 211                 err = inet_csk_listen_start(sk, backlog);
 212                 if (err)
 213                         goto out;
 214         }
 215         sk->sk_max_ack_backlog = backlog;
 216         err = 0;
 217
 218 out:
 219         release_sock(sk);
 220         return err;
 221 }
 222 EXPORT_SYMBOL(inet_listen);
 223
 224 u32 inet_ehash_secret __read_mostly;
 225 EXPORT_SYMBOL(inet_ehash_secret);
 226
 227 /*
 228  * inet_ehash_secret must be set exactly once
 229  * Instead of using a dedicated spinlock, we (ab)use inetsw_lock
 230  */
 231 void build_ehash_secret(void)
 232 {
 233         u32 rnd;
 234         do {
 235                 get_random_bytes(&rnd, sizeof(rnd));
 236         } while (rnd == 0);
 237         spin_lock_bh(&inetsw_lock);
 238         if (!inet_ehash_secret)
 239                 inet_ehash_secret = rnd;
 240         spin_unlock_bh(&inetsw_lock);
 241 }
 242 EXPORT_SYMBOL(build_ehash_secret);
 243
 244 static inline int inet_netns_ok(struct net *net, int protocol)
 245 {
 246         int hash;
 247         const struct net_protocol *ipprot;
 248
 249         if (net_eq(net, &init_net))
 250                 return 1;
 251
 252         hash = protocol & (MAX_INET_PROTOS - 1);
 253         ipprot = rcu_dereference(inet_protos[hash]);
 254
 255         if (ipprot == NULL)
 256                 /* raw IP is OK */
 257                 return 1;
 258         return ipprot->netns_ok;
 259 }
 260
 261 /*
 262  *      Create an inet socket.
 263  */
 264
 265 static int inet_create(struct net *net, struct socket *sock, int protocol,
 266                        int kern)
 267 {
 268         struct sock *sk;
 269         struct inet_protosw *answer;
 270         struct inet_sock *inet;
 271         struct proto *answer_prot;
 272         unsigned char answer_flags;
 273         char answer_no_check;
 274         int try_loading_module = 0;
 275         int err;
 276
 277         if (unlikely(!inet_ehash_secret))
 278                 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
 279                         build_ehash_secret();
 280
 281         sock->state = SS_UNCONNECTED;
 282
 283         /* Look for the requested type/protocol pair. */
 284 lookup_protocol:
 285         err = -ESOCKTNOSUPPORT;
 286         rcu_read_lock();
 287         list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
 288
 289                 err = 0;
 290                 /* Check the non-wild match. */
 291                 if (protocol == answer->protocol) {
 292                         if (protocol != IPPROTO_IP)
 293                                 break;
 294                 } else {
 295                         /* Check for the two wild cases. */
 296                         if (IPPROTO_IP == protocol) {
 297                                 protocol = answer->protocol;
 298                                 break;
 299                         }
 300                         if (IPPROTO_IP == answer->protocol)
 301                                 break;
 302                 }
 303                 err = -EPROTONOSUPPORT;
 304         }
 305
 306         if (unlikely(err)) {
 307                 if (try_loading_module < 2) {
 308                         rcu_read_unlock();
 309                         /*
 310                          * Be more specific, e.g. net-pf-2-proto-132-type-1
 311                          * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
 312                          */
 313                         if (++try_loading_module == 1)
 314                                 request_module("net-pf-%d-proto-%d-type-%d",
 315                                                PF_INET, protocol, sock->type);
 316                         /*
 317                          * Fall back to generic, e.g. net-pf-2-proto-132
 318                          * (net-pf-PF_INET-proto-IPPROTO_SCTP)
 319                          */
 320                         else
 321                                 request_module("net-pf-%d-proto-%d",
 322                                                PF_INET, protocol);
 323                         goto lookup_protocol;
 324                 } else
 325                         goto out_rcu_unlock;
 326         }
 327
 328         err = -EPERM;
 329         if (sock->type == SOCK_RAW && !kern && !capable(CAP_NET_RAW))
 330                 goto out_rcu_unlock;
 331
 332         err = -EAFNOSUPPORT;
 333         if (!inet_netns_ok(net, protocol))
 334                 goto out_rcu_unlock;
 335
 336         sock->ops = answer->ops;
 337         answer_prot = answer->prot;
 338         answer_no_check = answer->no_check;
 339         answer_flags = answer->flags;
 340         rcu_read_unlock();
 341
 342         WARN_ON(answer_prot->slab == NULL);
 343
 344         err = -ENOBUFS;
 345         sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
 346         if (sk == NULL)
 347                 goto out;
 348
 349         err = 0;
 350         sk->sk_no_check = answer_no_check;
 351         if (INET_PROTOSW_REUSE & answer_flags)
 352                 sk->sk_reuse = 1;
 353
 354         inet = inet_sk(sk);
 355         inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
 356
 357         if (SOCK_RAW == sock->type) {
 358                 inet->inet_num = protocol;
 359                 if (IPPROTO_RAW == protocol)
 360                         inet->hdrincl = 1;
 361         }
 362
 363         if (ipv4_config.no_pmtu_disc)
 364                 inet->pmtudisc = IP_PMTUDISC_DONT;
 365         else
 366                 inet->pmtudisc = IP_PMTUDISC_WANT;
 367
 368         inet->inet_id = 0;
 369
 370         sock_init_data(sock, sk);
 371
 372         sk->sk_destruct    = inet_sock_destruct;
 373         sk->sk_protocol    = protocol;
 374         sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
 375
 376         inet->uc_ttl    = -1;
 377         inet->mc_loop   = 1;
 378         inet->mc_ttl    = 1;
 379         inet->mc_all    = 1;
 380         inet->mc_index  = 0;
 381         inet->mc_list   = NULL;
 382
 383         sk_refcnt_debug_inc(sk);
 384
 385         if (inet->inet_num) {
 386                 /* It assumes that any protocol which allows
 387                  * the user to assign a number at socket
 388                  * creation time automatically
 389                  * shares.
 390                  */
 391                 inet->inet_sport = htons(inet->inet_num);
 392                 /* Add to protocol hash chains. */
 393                 sk->sk_prot->hash(sk);
 394         }
 395
 396         if (sk->sk_prot->init) {
 397                 err = sk->sk_prot->init(sk);
 398                 if (err)
 399                         sk_common_release(sk);
 400         }
 401 out:
 402         return err;
 403 out_rcu_unlock:
 404         rcu_read_unlock();
 405         goto out;
 406 }
 407
 408
 409 /*
 410  *      The peer socket should always be NULL (or else). When we call this
 411  *      function we are destroying the object and from then on nobody
 412  *      should refer to it.
 413  */
 414 int inet_release(struct socket *sock)
 415 {
 416         struct sock *sk = sock->sk;
 417
 418         if (sk) {
 419                 long timeout;
 420
 421                 /* Applications forget to leave groups before exiting */
 422                 ip_mc_drop_socket(sk);
 423
 424                 /* If linger is set, we don't return until the close
 425                  * is complete.  Otherwise we return immediately. The
 426                  * actually closing is done the same either way.
 427                  *
 428                  * If the close is due to the process exiting, we never
 429                  * linger..
 430                  */
 431                 timeout = 0;
 432                 if (sock_flag(sk, SOCK_LINGER) &&
 433                     !(current->flags & PF_EXITING))
 434                         timeout = sk->sk_lingertime;
 435                 sock->sk = NULL;
 436                 sk->sk_prot->close(sk, timeout);
 437         }
 438         return 0;
 439 }
 440 EXPORT_SYMBOL(inet_release);
 441
 442 /* It is off by default, see below. */
 443 int sysctl_ip_nonlocal_bind __read_mostly;
 444 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
 445
 446 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 447 {
 448         struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
 449         struct sock *sk = sock->sk;
 450         struct inet_sock *inet = inet_sk(sk);
 451         unsigned short snum;
 452         int chk_addr_ret;
 453         int err;
 454
 455         /* If the socket has its own bind function then use it. (RAW) */
 456         if (sk->sk_prot->bind) {
 457                 err = sk->sk_prot->bind(sk, uaddr, addr_len);
 458                 goto out;
 459         }
 460         err = -EINVAL;
 461         if (addr_len < sizeof(struct sockaddr_in))
 462                 goto out;
 463
 464         chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr);
 465
 466         /* Not specified by any standard per-se, however it breaks too
 467          * many applications when removed.  It is unfortunate since
 468          * allowing applications to make a non-local bind solves
 469          * several problems with systems using dynamic addressing.
 470          * (ie. your servers still start up even if your ISDN link
 471          *  is temporarily down)
 472          */
 473         err = -EADDRNOTAVAIL;
 474         if (!sysctl_ip_nonlocal_bind &&
 475             !(inet->freebind || inet->transparent) &&
 476             addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
 477             chk_addr_ret != RTN_LOCAL &&
 478             chk_addr_ret != RTN_MULTICAST &&
 479             chk_addr_ret != RTN_BROADCAST)
 480                 goto out;
 481
 482         snum = ntohs(addr->sin_port);
 483         err = -EACCES;
 484         if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
 485                 goto out;
 486
 487         /*      We keep a pair of addresses. rcv_saddr is the one
 488          *      used by hash lookups, and saddr is used for transmit.
 489          *
 490          *      In the BSD API these are the same except where it
 491          *      would be illegal to use them (multicast/broadcast) in
 492          *      which case the sending device address is used.
 493          */
 494         lock_sock(sk);
 495
 496         /* Check these errors (active socket, double bind). */
 497         err = -EINVAL;
 498         if (sk->sk_state != TCP_CLOSE || inet->inet_num)
 499                 goto out_release_sock;
 500
 501         inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
 502         if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
 503                 inet->inet_saddr = 0;  /* Use device */
 504
 505         /* Make sure we are allowed to bind here. */
 506         if (sk->sk_prot->get_port(sk, snum)) {
 507                 inet->inet_saddr = inet->inet_rcv_saddr = 0;
 508                 err = -EADDRINUSE;
 509                 goto out_release_sock;
 510         }
 511
 512         if (inet->inet_rcv_saddr)
 513                 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
 514         if (snum)
 515                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
 516         inet->inet_sport = htons(inet->inet_num);
 517         inet->inet_daddr = 0;
 518         inet->inet_dport = 0;
 519         sk_dst_reset(sk);
 520         err = 0;
 521 out_release_sock:
 522         release_sock(sk);
 523 out:
 524         return err;
 525 }
 526 EXPORT_SYMBOL(inet_bind);
 527
 528 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
 529                        int addr_len, int flags)
 530 {
 531         struct sock *sk = sock->sk;
 532
 533         if (uaddr->sa_family == AF_UNSPEC)
 534                 return sk->sk_prot->disconnect(sk, flags);
 535
 536         if (!inet_sk(sk)->inet_num && inet_autobind(sk))
 537                 return -EAGAIN;
 538         return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
 539 }
 540 EXPORT_SYMBOL(inet_dgram_connect);
 541
 542 static long inet_wait_for_connect(struct sock *sk, long timeo)
 543 {
 544         DEFINE_WAIT(wait);
 545
 546         prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 547
 548         /* Basic assumption: if someone sets sk->sk_err, he _must_
 549          * change state of the socket from TCP_SYN_*.
 550          * Connect() does not allow to get error notifications
 551          * without closing the socket.
 552          */
 553         while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 554                 release_sock(sk);
 555                 timeo = schedule_timeout(timeo);
 556                 lock_sock(sk);
 557                 if (signal_pending(current) || !timeo)
 558                         break;
 559                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 560         }
 561         finish_wait(sk->sk_sleep, &wait);
 562         return timeo;
 563 }
 564
 565 /*
 566  *      Connect to a remote host. There is regrettably still a little
 567  *      TCP 'magic' in here.
 568  */
 569 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 570                         int addr_len, int flags)
 571 {
 572         struct sock *sk = sock->sk;
 573         int err;
 574         long timeo;
 575
 576         lock_sock(sk);
 577
 578         if (uaddr->sa_family == AF_UNSPEC) {
 579                 err = sk->sk_prot->disconnect(sk, flags);
 580                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 581                 goto out;
 582         }
 583
 584         switch (sock->state) {
 585         default:
 586                 err = -EINVAL;
 587                 goto out;
 588         case SS_CONNECTED:
 589                 err = -EISCONN;
 590                 goto out;
 591         case SS_CONNECTING:
 592                 err = -EALREADY;
 593                 /* Fall out of switch with err, set for this state */
 594                 break;
 595         case SS_UNCONNECTED:
 596                 err = -EISCONN;
 597                 if (sk->sk_state != TCP_CLOSE)
 598                         goto out;
 599
 600                 err = sk->sk_prot->connect(sk, uaddr, addr_len);
 601                 if (err < 0)
 602                         goto out;
 603
 604                 sock->state = SS_CONNECTING;
 605
 606                 /* Just entered SS_CONNECTING state; the only
 607                  * difference is that return value in non-blocking
 608                  * case is EINPROGRESS, rather than EALREADY.
 609                  */
 610                 err = -EINPROGRESS;
 611                 break;
 612         }
 613
 614         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
 615
 616         if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 617                 /* Error code is set above */
 618                 if (!timeo || !inet_wait_for_connect(sk, timeo))
 619                         goto out;
 620
 621                 err = sock_intr_errno(timeo);
 622                 if (signal_pending(current))
 623                         goto out;
 624         }
 625
 626         /* Connection was closed by RST, timeout, ICMP error
 627          * or another process disconnected us.
 628          */
 629         if (sk->sk_state == TCP_CLOSE)
 630                 goto sock_error;
 631
 632         /* sk->sk_err may be not zero now, if RECVERR was ordered by user
 633          * and error was received after socket entered established state.
 634          * Hence, it is handled normally after connect() return successfully.
 635          */
 636
 637         sock->state = SS_CONNECTED;
 638         err = 0;
 639 out:
 640         release_sock(sk);
 641         return err;
 642
 643 sock_error:
 644         err = sock_error(sk) ? : -ECONNABORTED;
 645         sock->state = SS_UNCONNECTED;
 646         if (sk->sk_prot->disconnect(sk, flags))
 647                 sock->state = SS_DISCONNECTING;
 648         goto out;
 649 }
 650 EXPORT_SYMBOL(inet_stream_connect);
 651
 652 /*
 653  *      Accept a pending connection. The TCP layer now gives BSD semantics.
 654  */
 655
 656 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
 657 {
 658         struct sock *sk1 = sock->sk;
 659         int err = -EINVAL;
 660         struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
 661
 662         if (!sk2)
 663                 goto do_err;
 664
 665         lock_sock(sk2);
 666
 667         WARN_ON(!((1 << sk2->sk_state) &
 668                   (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE)));
 669
 670         sock_graft(sk2, newsock);
 671
 672         newsock->state = SS_CONNECTED;
 673         err = 0;
 674         release_sock(sk2);
 675 do_err:
 676         return err;
 677 }
 678 EXPORT_SYMBOL(inet_accept);
 679
 680
 681 /*
 682  *      This does both peername and sockname.
 683  */
 684 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
 685                         int *uaddr_len, int peer)
 686 {
 687         struct sock *sk         = sock->sk;
 688         struct inet_sock *inet  = inet_sk(sk);
 689         DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
 690
 691         sin->sin_family = AF_INET;
 692         if (peer) {
 693                 if (!inet->inet_dport ||
 694                     (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
 695                      peer == 1))
 696                         return -ENOTCONN;
 697                 sin->sin_port = inet->inet_dport;
 698                 sin->sin_addr.s_addr = inet->inet_daddr;
 699         } else {
 700                 __be32 addr = inet->inet_rcv_saddr;
 701                 if (!addr)
 702                         addr = inet->inet_saddr;
 703                 sin->sin_port = inet->inet_sport;
 704                 sin->sin_addr.s_addr = addr;
 705         }
 706         memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 707         *uaddr_len = sizeof(*sin);
 708         return 0;
 709 }
 710 EXPORT_SYMBOL(inet_getname);
 711
 712 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
 713                  size_t size)
 714 {
 715         struct sock *sk = sock->sk;
 716
 717         /* We may need to bind the socket. */
 718         if (!inet_sk(sk)->inet_num && inet_autobind(sk))
 719                 return -EAGAIN;
 720
 721         return sk->sk_prot->sendmsg(iocb, sk, msg, size);
 722 }
 723 EXPORT_SYMBOL(inet_sendmsg);
 724
 725
 726 static ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
 727                              size_t size, int flags)
 728 {
 729         struct sock *sk = sock->sk;
 730
 731         /* We may need to bind the socket. */
 732         if (!inet_sk(sk)->inet_num && inet_autobind(sk))
 733                 return -EAGAIN;
 734
 735         if (sk->sk_prot->sendpage)
 736                 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
 737         return sock_no_sendpage(sock, page, offset, size, flags);
 738 }
 739
 740
 741 int inet_shutdown(struct socket *sock, int how)
 742 {
 743         struct sock *sk = sock->sk;
 744         int err = 0;
 745
 746         /* This should really check to make sure
 747          * the socket is a TCP socket. (WHY AC...)
 748          */
 749         how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
 750                        1->2 bit 2 snds.
 751                        2->3 */
 752         if ((how & ~SHUTDOWN_MASK) || !how)     /* MAXINT->0 */
 753                 return -EINVAL;
 754
 755         lock_sock(sk);
 756         if (sock->state == SS_CONNECTING) {
 757                 if ((1 << sk->sk_state) &
 758                     (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
 759                         sock->state = SS_DISCONNECTING;
 760                 else
 761                         sock->state = SS_CONNECTED;
 762         }
 763
 764         switch (sk->sk_state) {
 765         case TCP_CLOSE:
 766                 err = -ENOTCONN;
 767                 /* Hack to wake up other listeners, who can poll for
 768                    POLLHUP, even on eg. unconnected UDP sockets -- RR */
 769         default:
 770                 sk->sk_shutdown |= how;
 771                 if (sk->sk_prot->shutdown)
 772                         sk->sk_prot->shutdown(sk, how);
 773                 break;
 774
 775         /* Remaining two branches are temporary solution for missing
 776          * close() in multithreaded environment. It is _not_ a good idea,
 777          * but we have no choice until close() is repaired at VFS level.
 778          */
 779         case TCP_LISTEN:
 780                 if (!(how & RCV_SHUTDOWN))
 781                         break;
 782                 /* Fall through */
 783         case TCP_SYN_SENT:
 784                 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
 785                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 786                 break;
 787         }
 788
 789         /* Wake up anyone sleeping in poll. */
 790         sk->sk_state_change(sk);
 791         release_sock(sk);
 792         return err;
 793 }
 794 EXPORT_SYMBOL(inet_shutdown);
 795
 796 /*
 797  *      ioctl() calls you can issue on an INET socket. Most of these are
 798  *      device configuration and stuff and very rarely used. Some ioctls
 799  *      pass on to the socket itself.
 800  *
 801  *      NOTE: I like the idea of a module for the config stuff. ie ifconfig
 802  *      loads the devconfigure module does its configuring and unloads it.
 803  *      There's a good 20K of config code hanging around the kernel.
 804  */
 805
 806 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 807 {
 808         struct sock *sk = sock->sk;
 809         int err = 0;
 810         struct net *net = sock_net(sk);
 811
 812         switch (cmd) {
 813         case SIOCGSTAMP:
 814                 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
 815                 break;
 816         case SIOCGSTAMPNS:
 817                 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
 818                 break;
 819         case SIOCADDRT:
 820         case SIOCDELRT:
 821         case SIOCRTMSG:
 822                 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
 823                 break;
 824         case SIOCDARP:
 825         case SIOCGARP:
 826         case SIOCSARP:
 827                 err = arp_ioctl(net, cmd, (void __user *)arg);
 828                 break;
 829         case SIOCGIFADDR:
 830         case SIOCSIFADDR:
 831         case SIOCGIFBRDADDR:
 832         case SIOCSIFBRDADDR:
 833         case SIOCGIFNETMASK:
 834         case SIOCSIFNETMASK:
 835         case SIOCGIFDSTADDR:
 836         case SIOCSIFDSTADDR:
 837         case SIOCSIFPFLAGS:
 838         case SIOCGIFPFLAGS:
 839         case SIOCSIFFLAGS:
 840                 err = devinet_ioctl(net, cmd, (void __user *)arg);
 841                 break;
 842         default:
 843                 if (sk->sk_prot->ioctl)
 844                         err = sk->sk_prot->ioctl(sk, cmd, arg);
 845                 else
 846                         err = -ENOIOCTLCMD;
 847                 break;
 848         }
 849         return err;
 850 }
 851 EXPORT_SYMBOL(inet_ioctl);
 852
 853 const struct proto_ops inet_stream_ops = {
 854         .family            = PF_INET,
 855         .owner             = THIS_MODULE,
 856         .release           = inet_release,
 857         .bind              = inet_bind,
 858         .connect           = inet_stream_connect,
 859         .socketpair        = sock_no_socketpair,
 860         .accept            = inet_accept,
 861         .getname           = inet_getname,
 862         .poll              = tcp_poll,
 863         .ioctl             = inet_ioctl,
 864         .listen            = inet_listen,
 865         .shutdown          = inet_shutdown,
 866         .setsockopt        = sock_common_setsockopt,
 867         .getsockopt        = sock_common_getsockopt,
 868         .sendmsg           = tcp_sendmsg,
 869         .recvmsg           = sock_common_recvmsg,
 870         .mmap              = sock_no_mmap,
 871         .sendpage          = tcp_sendpage,
 872         .splice_read       = tcp_splice_read,
 873 #ifdef CONFIG_COMPAT
 874         .compat_setsockopt = compat_sock_common_setsockopt,
 875         .compat_getsockopt = compat_sock_common_getsockopt,
 876 #endif
 877 };
 878 EXPORT_SYMBOL(inet_stream_ops);
 879
 880 const struct proto_ops inet_dgram_ops = {
 881         .family            = PF_INET,
 882         .owner             = THIS_MODULE,
 883         .release           = inet_release,
 884         .bind              = inet_bind,
 885         .connect           = inet_dgram_connect,
 886         .socketpair        = sock_no_socketpair,
 887         .accept            = sock_no_accept,
 888         .getname           = inet_getname,
 889         .poll              = udp_poll,
 890         .ioctl             = inet_ioctl,
 891         .listen            = sock_no_listen,
 892         .shutdown          = inet_shutdown,
 893         .setsockopt        = sock_common_setsockopt,
 894         .getsockopt        = sock_common_getsockopt,
 895         .sendmsg           = inet_sendmsg,
 896         .recvmsg           = sock_common_recvmsg,
 897         .mmap              = sock_no_mmap,
 898         .sendpage          = inet_sendpage,
 899 #ifdef CONFIG_COMPAT
 900         .compat_setsockopt = compat_sock_common_setsockopt,
 901         .compat_getsockopt = compat_sock_common_getsockopt,
 902 #endif
 903 };
 904 EXPORT_SYMBOL(inet_dgram_ops);
 905
 906 /*
 907  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
 908  * udp_poll
 909  */
 910 static const struct proto_ops inet_sockraw_ops = {
 911         .family            = PF_INET,
 912         .owner             = THIS_MODULE,
 913         .release           = inet_release,
 914         .bind              = inet_bind,
 915         .connect           = inet_dgram_connect,
 916         .socketpair        = sock_no_socketpair,
 917         .accept            = sock_no_accept,
 918         .getname           = inet_getname,
 919         .poll              = datagram_poll,
 920         .ioctl             = inet_ioctl,
 921         .listen            = sock_no_listen,
 922         .shutdown          = inet_shutdown,
 923         .setsockopt        = sock_common_setsockopt,
 924         .getsockopt        = sock_common_getsockopt,
 925         .sendmsg           = inet_sendmsg,
 926         .recvmsg           = sock_common_recvmsg,
 927         .mmap              = sock_no_mmap,
 928         .sendpage          = inet_sendpage,
 929 #ifdef CONFIG_COMPAT
 930         .compat_setsockopt = compat_sock_common_setsockopt,
 931         .compat_getsockopt = compat_sock_common_getsockopt,
 932 #endif
 933 };
 934
 935 static const struct net_proto_family inet_family_ops = {
 936         .family = PF_INET,
 937         .create = inet_create,
 938         .owner  = THIS_MODULE,
 939 };
 940
 941 /* Upon startup we insert all the elements in inetsw_array[] into
 942  * the linked list inetsw.
 943  */
 944 static struct inet_protosw inetsw_array[] =
 945 {
 946         {
 947                 .type =       SOCK_STREAM,
 948                 .protocol =   IPPROTO_TCP,
 949                 .prot =       &tcp_prot,
 950                 .ops =        &inet_stream_ops,
 951                 .no_check =   0,
 952                 .flags =      INET_PROTOSW_PERMANENT |
 953                               INET_PROTOSW_ICSK,
 954         },
 955
 956         {
 957                 .type =       SOCK_DGRAM,
 958                 .protocol =   IPPROTO_UDP,
 959                 .prot =       &udp_prot,
 960                 .ops =        &inet_dgram_ops,
 961                 .no_check =   UDP_CSUM_DEFAULT,
 962                 .flags =      INET_PROTOSW_PERMANENT,
 963        },
 964
 965
 966        {
 967                .type =       SOCK_RAW,
 968                .protocol =   IPPROTO_IP,        /* wild card */
 969                .prot =       &raw_prot,
 970                .ops =        &inet_sockraw_ops,
 971                .no_check =   UDP_CSUM_DEFAULT,
 972                .flags =      INET_PROTOSW_REUSE,
 973        }
 974 };
 975
 976 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
 977
 978 void inet_register_protosw(struct inet_protosw *p)
 979 {
 980         struct list_head *lh;
 981         struct inet_protosw *answer;
 982         int protocol = p->protocol;
 983         struct list_head *last_perm;
 984
 985         spin_lock_bh(&inetsw_lock);
 986
 987         if (p->type >= SOCK_MAX)
 988                 goto out_illegal;
 989
 990         /* If we are trying to override a permanent protocol, bail. */
 991         answer = NULL;
 992         last_perm = &inetsw[p->type];
 993         list_for_each(lh, &inetsw[p->type]) {
 994                 answer = list_entry(lh, struct inet_protosw, list);
 995
 996                 /* Check only the non-wild match. */
 997                 if (INET_PROTOSW_PERMANENT & answer->flags) {
 998                         if (protocol == answer->protocol)
 999                                 break;
1000                         last_perm = lh;
1001                 }
1002
1003                 answer = NULL;
1004         }
1005         if (answer)
1006                 goto out_permanent;
1007
1008         /* Add the new entry after the last permanent entry if any, so that
1009          * the new entry does not override a permanent entry when matched with
1010          * a wild-card protocol. But it is allowed to override any existing
1011          * non-permanent entry.  This means that when we remove this entry, the
1012          * system automatically returns to the old behavior.
1013          */
1014         list_add_rcu(&p->list, last_perm);
1015 out:
1016         spin_unlock_bh(&inetsw_lock);
1017
1018         return;
1019
1020 out_permanent:
1021         printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
1022                protocol);
1023         goto out;
1024
1025 out_illegal:
1026         printk(KERN_ERR
1027                "Ignoring attempt to register invalid socket type %d.\n",
1028                p->type);
1029         goto out;
1030 }
1031 EXPORT_SYMBOL(inet_register_protosw);
1032
1033 void inet_unregister_protosw(struct inet_protosw *p)
1034 {
1035         if (INET_PROTOSW_PERMANENT & p->flags) {
1036                 printk(KERN_ERR
1037                        "Attempt to unregister permanent protocol %d.\n",
1038                        p->protocol);
1039         } else {
1040                 spin_lock_bh(&inetsw_lock);
1041                 list_del_rcu(&p->list);
1042                 spin_unlock_bh(&inetsw_lock);
1043
1044                 synchronize_net();
1045         }
1046 }
1047 EXPORT_SYMBOL(inet_unregister_protosw);
1048
1049 /*
1050  *      Shall we try to damage output packets if routing dev changes?
1051  */
1052
1053 int sysctl_ip_dynaddr __read_mostly;
1054
1055 static int inet_sk_reselect_saddr(struct sock *sk)
1056 {
1057         struct inet_sock *inet = inet_sk(sk);
1058         int err;
1059         struct rtable *rt;
1060         __be32 old_saddr = inet->inet_saddr;
1061         __be32 new_saddr;
1062         __be32 daddr = inet->inet_daddr;
1063
1064         if (inet->opt && inet->opt->srr)
1065                 daddr = inet->opt->faddr;
1066
1067         /* Query new route. */
1068         err = ip_route_connect(&rt, daddr, 0,
1069                                RT_CONN_FLAGS(sk),
1070                                sk->sk_bound_dev_if,
1071                                sk->sk_protocol,
1072                                inet->inet_sport, inet->inet_dport, sk, 0);
1073         if (err)
1074                 return err;
1075
1076         sk_setup_caps(sk, &rt->u.dst);
1077
1078         new_saddr = rt->rt_src;
1079
1080         if (new_saddr == old_saddr)
1081                 return 0;
1082
1083         if (sysctl_ip_dynaddr > 1) {
1084                 printk(KERN_INFO "%s(): shifting inet->saddr from %pI4 to %pI4\n",
1085                        __func__, &old_saddr, &new_saddr);
1086         }
1087
1088         inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1089
1090         /*
1091          * XXX The only one ugly spot where we need to
1092          * XXX really change the sockets identity after
1093          * XXX it has entered the hashes. -DaveM
1094          *
1095          * Besides that, it does not check for connection
1096          * uniqueness. Wait for troubles.
1097          */
1098         __sk_prot_rehash(sk);
1099         return 0;
1100 }
1101
1102 int inet_sk_rebuild_header(struct sock *sk)
1103 {
1104         struct inet_sock *inet = inet_sk(sk);
1105         struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1106         __be32 daddr;
1107         int err;
1108
1109         /* Route is OK, nothing to do. */
1110         if (rt)
1111                 return 0;
1112
1113         /* Reroute. */
1114         daddr = inet->inet_daddr;
1115         if (inet->opt && inet->opt->srr)
1116                 daddr = inet->opt->faddr;
1117 {
1118         struct flowi fl = {
1119                 .oif = sk->sk_bound_dev_if,
1120                 .mark = sk->sk_mark,
1121                 .nl_u = {
1122                         .ip4_u = {
1123                                 .daddr  = daddr,
1124                                 .saddr  = inet->inet_saddr,
1125                                 .tos    = RT_CONN_FLAGS(sk),
1126                         },
1127                 },
1128                 .proto = sk->sk_protocol,
1129                 .flags = inet_sk_flowi_flags(sk),
1130                 .uli_u = {
1131                         .ports = {
1132                                 .sport = inet->inet_sport,
1133                                 .dport = inet->inet_dport,
1134                         },
1135                 },
1136         };
1137
1138         security_sk_classify_flow(sk, &fl);
1139         err = ip_route_output_flow(sock_net(sk), &rt, &fl, sk, 0);
1140 }
1141         if (!err)
1142                 sk_setup_caps(sk, &rt->u.dst);
1143         else {
1144                 /* Routing failed... */
1145                 sk->sk_route_caps = 0;
1146                 /*
1147                  * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1148                  * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1149                  */
1150                 if (!sysctl_ip_dynaddr ||
1151                     sk->sk_state != TCP_SYN_SENT ||
1152                     (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1153                     (err = inet_sk_reselect_saddr(sk)) != 0)
1154                         sk->sk_err_soft = -err;
1155         }
1156
1157         return err;
1158 }
1159 EXPORT_SYMBOL(inet_sk_rebuild_header);
1160
1161 static int inet_gso_send_check(struct sk_buff *skb)
1162 {
1163         struct iphdr *iph;
1164         const struct net_protocol *ops;
1165         int proto;
1166         int ihl;
1167         int err = -EINVAL;
1168
1169         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1170                 goto out;
1171
1172         iph = ip_hdr(skb);
1173         ihl = iph->ihl * 4;
1174         if (ihl < sizeof(*iph))
1175                 goto out;
1176
1177         if (unlikely(!pskb_may_pull(skb, ihl)))
1178                 goto out;
1179
1180         __skb_pull(skb, ihl);
1181         skb_reset_transport_header(skb);
1182         iph = ip_hdr(skb);
1183         proto = iph->protocol & (MAX_INET_PROTOS - 1);
1184         err = -EPROTONOSUPPORT;
1185
1186         rcu_read_lock();
1187         ops = rcu_dereference(inet_protos[proto]);
1188         if (likely(ops && ops->gso_send_check))
1189                 err = ops->gso_send_check(skb);
1190         rcu_read_unlock();
1191
1192 out:
1193         return err;
1194 }
1195
1196 static struct sk_buff *inet_gso_segment(struct sk_buff *skb, int features)
1197 {
1198         struct sk_buff *segs = ERR_PTR(-EINVAL);
1199         struct iphdr *iph;
1200         const struct net_protocol *ops;
1201         int proto;
1202         int ihl;
1203         int id;
1204         unsigned int offset = 0;
1205
1206         if (!(features & NETIF_F_V4_CSUM))
1207                 features &= ~NETIF_F_SG;
1208
1209         if (unlikely(skb_shinfo(skb)->gso_type &
1210                      ~(SKB_GSO_TCPV4 |
1211                        SKB_GSO_UDP |
1212                        SKB_GSO_DODGY |
1213                        SKB_GSO_TCP_ECN |
1214                        0)))
1215                 goto out;
1216
1217         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1218                 goto out;
1219
1220         iph = ip_hdr(skb);
1221         ihl = iph->ihl * 4;
1222         if (ihl < sizeof(*iph))
1223                 goto out;
1224
1225         if (unlikely(!pskb_may_pull(skb, ihl)))
1226                 goto out;
1227
1228         __skb_pull(skb, ihl);
1229         skb_reset_transport_header(skb);
1230         iph = ip_hdr(skb);
1231         id = ntohs(iph->id);
1232         proto = iph->protocol & (MAX_INET_PROTOS - 1);
1233         segs = ERR_PTR(-EPROTONOSUPPORT);
1234
1235         rcu_read_lock();
1236         ops = rcu_dereference(inet_protos[proto]);
1237         if (likely(ops && ops->gso_segment))
1238                 segs = ops->gso_segment(skb, features);
1239         rcu_read_unlock();
1240
1241         if (!segs || IS_ERR(segs))
1242                 goto out;
1243
1244         skb = segs;
1245         do {
1246                 iph = ip_hdr(skb);
1247                 if (proto == IPPROTO_UDP) {
1248                         iph->id = htons(id);
1249                         iph->frag_off = htons(offset >> 3);
1250                         if (skb->next != NULL)
1251                                 iph->frag_off |= htons(IP_MF);
1252                         offset += (skb->len - skb->mac_len - iph->ihl * 4);
1253                 } else
1254                         iph->id = htons(id++);
1255                 iph->tot_len = htons(skb->len - skb->mac_len);
1256                 iph->check = 0;
1257                 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1258         } while ((skb = skb->next));
1259
1260 out:
1261         return segs;
1262 }
1263
1264 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1265                                          struct sk_buff *skb)
1266 {
1267         const struct net_protocol *ops;
1268         struct sk_buff **pp = NULL;
1269         struct sk_buff *p;
1270         struct iphdr *iph;
1271         unsigned int hlen;
1272         unsigned int off;
1273         unsigned int id;
1274         int flush = 1;
1275         int proto;
1276
1277         off = skb_gro_offset(skb);
1278         hlen = off + sizeof(*iph);
1279         iph = skb_gro_header_fast(skb, off);
1280         if (skb_gro_header_hard(skb, hlen)) {
1281                 iph = skb_gro_header_slow(skb, hlen, off);
1282                 if (unlikely(!iph))
1283                         goto out;
1284         }
1285
1286         proto = iph->protocol & (MAX_INET_PROTOS - 1);
1287
1288         rcu_read_lock();
1289         ops = rcu_dereference(inet_protos[proto]);
1290         if (!ops || !ops->gro_receive)
1291                 goto out_unlock;
1292
1293         if (*(u8 *)iph != 0x45)
1294                 goto out_unlock;
1295
1296         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1297                 goto out_unlock;
1298
1299         id = ntohl(*(u32 *)&iph->id);
1300         flush = (u16)((ntohl(*(u32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF));
1301         id >>= 16;
1302
1303         for (p = *head; p; p = p->next) {
1304                 struct iphdr *iph2;
1305
1306                 if (!NAPI_GRO_CB(p)->same_flow)
1307                         continue;
1308
1309                 iph2 = ip_hdr(p);
1310
1311                 if ((iph->protocol ^ iph2->protocol) |
1312                     (iph->tos ^ iph2->tos) |
1313                     (iph->saddr ^ iph2->saddr) |
1314                     (iph->daddr ^ iph2->daddr)) {
1315                         NAPI_GRO_CB(p)->same_flow = 0;
1316                         continue;
1317                 }
1318
1319                 /* All fields must match except length and checksum. */
1320                 NAPI_GRO_CB(p)->flush |=
1321                         (iph->ttl ^ iph2->ttl) |
1322                         ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1323
1324                 NAPI_GRO_CB(p)->flush |= flush;
1325         }
1326
1327         NAPI_GRO_CB(skb)->flush |= flush;
1328         skb_gro_pull(skb, sizeof(*iph));
1329         skb_set_transport_header(skb, skb_gro_offset(skb));
1330
1331         pp = ops->gro_receive(head, skb);
1332
1333 out_unlock:
1334         rcu_read_unlock();
1335
1336 out:
1337         NAPI_GRO_CB(skb)->flush |= flush;
1338
1339         return pp;
1340 }
1341
1342 static int inet_gro_complete(struct sk_buff *skb)
1343 {
1344         const struct net_protocol *ops;
1345         struct iphdr *iph = ip_hdr(skb);
1346         int proto = iph->protocol & (MAX_INET_PROTOS - 1);
1347         int err = -ENOSYS;
1348         __be16 newlen = htons(skb->len - skb_network_offset(skb));
1349
1350         csum_replace2(&iph->check, iph->tot_len, newlen);
1351         iph->tot_len = newlen;
1352
1353         rcu_read_lock();
1354         ops = rcu_dereference(inet_protos[proto]);
1355         if (WARN_ON(!ops || !ops->gro_complete))
1356                 goto out_unlock;
1357
1358         err = ops->gro_complete(skb);
1359
1360 out_unlock:
1361         rcu_read_unlock();
1362
1363         return err;
1364 }
1365
1366 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1367                          unsigned short type, unsigned char protocol,
1368                          struct net *net)
1369 {
1370         struct socket *sock;
1371         int rc = sock_create_kern(family, type, protocol, &sock);
1372
1373         if (rc == 0) {
1374                 *sk = sock->sk;
1375                 (*sk)->sk_allocation = GFP_ATOMIC;
1376                 /*
1377                  * Unhash it so that IP input processing does not even see it,
1378                  * we do not wish this socket to see incoming packets.
1379                  */
1380                 (*sk)->sk_prot->unhash(*sk);
1381
1382                 sk_change_net(*sk, net);
1383         }
1384         return rc;
1385 }
1386 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1387
1388 unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1389 {
1390         unsigned long res = 0;
1391         int i;
1392
1393         for_each_possible_cpu(i) {
1394                 res += *(((unsigned long *) per_cpu_ptr(mib[0], i)) + offt);
1395                 res += *(((unsigned long *) per_cpu_ptr(mib[1], i)) + offt);
1396         }
1397         return res;
1398 }
1399 EXPORT_SYMBOL_GPL(snmp_fold_field);
1400
1401 int snmp_mib_init(void __percpu *ptr[2], size_t mibsize)
1402 {
1403         BUG_ON(ptr == NULL);
1404         ptr[0] = __alloc_percpu(mibsize, __alignof__(unsigned long long));
1405         if (!ptr[0])
1406                 goto err0;
1407         ptr[1] = __alloc_percpu(mibsize, __alignof__(unsigned long long));
1408         if (!ptr[1])
1409                 goto err1;
1410         return 0;
1411 err1:
1412         free_percpu(ptr[0]);
1413         ptr[0] = NULL;
1414 err0:
1415         return -ENOMEM;
1416 }
1417 EXPORT_SYMBOL_GPL(snmp_mib_init);
1418
1419 void snmp_mib_free(void __percpu *ptr[2])
1420 {
1421         BUG_ON(ptr == NULL);
1422         free_percpu(ptr[0]);
1423         free_percpu(ptr[1]);
1424         ptr[0] = ptr[1] = NULL;
1425 }
1426 EXPORT_SYMBOL_GPL(snmp_mib_free);
1427
1428 #ifdef CONFIG_IP_MULTICAST
1429 static const struct net_protocol igmp_protocol = {
1430         .handler =      igmp_rcv,
1431         .netns_ok =     1,
1432 };
1433 #endif
1434
1435 static const struct net_protocol tcp_protocol = {
1436         .handler =      tcp_v4_rcv,
1437         .err_handler =  tcp_v4_err,
1438         .gso_send_check = tcp_v4_gso_send_check,
1439         .gso_segment =  tcp_tso_segment,
1440         .gro_receive =  tcp4_gro_receive,
1441         .gro_complete = tcp4_gro_complete,
1442         .no_policy =    1,
1443         .netns_ok =     1,
1444 };
1445
1446 static const struct net_protocol udp_protocol = {
1447         .handler =      udp_rcv,
1448         .err_handler =  udp_err,
1449         .gso_send_check = udp4_ufo_send_check,
1450         .gso_segment = udp4_ufo_fragment,
1451         .no_policy =    1,
1452         .netns_ok =     1,
1453 };
1454
1455 static const struct net_protocol icmp_protocol = {
1456         .handler =      icmp_rcv,
1457         .no_policy =    1,
1458         .netns_ok =     1,
1459 };
1460
1461 static __net_init int ipv4_mib_init_net(struct net *net)
1462 {
1463         if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1464                           sizeof(struct tcp_mib)) < 0)
1465                 goto err_tcp_mib;
1466         if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1467                           sizeof(struct ipstats_mib)) < 0)
1468                 goto err_ip_mib;
1469         if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1470                           sizeof(struct linux_mib)) < 0)
1471                 goto err_net_mib;
1472         if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1473                           sizeof(struct udp_mib)) < 0)
1474                 goto err_udp_mib;
1475         if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1476                           sizeof(struct udp_mib)) < 0)
1477                 goto err_udplite_mib;
1478         if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1479                           sizeof(struct icmp_mib)) < 0)
1480                 goto err_icmp_mib;
1481         if (snmp_mib_init((void __percpu **)net->mib.icmpmsg_statistics,
1482                           sizeof(struct icmpmsg_mib)) < 0)
1483                 goto err_icmpmsg_mib;
1484
1485         tcp_mib_init(net);
1486         return 0;
1487
1488 err_icmpmsg_mib:
1489         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1490 err_icmp_mib:
1491         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1492 err_udplite_mib:
1493         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1494 err_udp_mib:
1495         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1496 err_net_mib:
1497         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1498 err_ip_mib:
1499         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1500 err_tcp_mib:
1501         return -ENOMEM;
1502 }
1503
1504 static __net_exit void ipv4_mib_exit_net(struct net *net)
1505 {
1506         snmp_mib_free((void __percpu **)net->mib.icmpmsg_statistics);
1507         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1508         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1509         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1510         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1511         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1512         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1513 }
1514
1515 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1516         .init = ipv4_mib_init_net,
1517         .exit = ipv4_mib_exit_net,
1518 };
1519
1520 static int __init init_ipv4_mibs(void)
1521 {
1522         return register_pernet_subsys(&ipv4_mib_ops);
1523 }
1524
1525 static int ipv4_proc_init(void);
1526
1527 /*
1528  *      IP protocol layer initialiser
1529  */
1530
1531 static struct packet_type ip_packet_type __read_mostly = {
1532         .type = cpu_to_be16(ETH_P_IP),
1533         .func = ip_rcv,
1534         .gso_send_check = inet_gso_send_check,
1535         .gso_segment = inet_gso_segment,
1536         .gro_receive = inet_gro_receive,
1537         .gro_complete = inet_gro_complete,
1538 };
1539
1540 static int __init inet_init(void)
1541 {
1542         struct sk_buff *dummy_skb;
1543         struct inet_protosw *q;
1544         struct list_head *r;
1545         int rc = -EINVAL;
1546
1547         BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1548
1549         rc = proto_register(&tcp_prot, 1);
1550         if (rc)
1551                 goto out;
1552
1553         rc = proto_register(&udp_prot, 1);
1554         if (rc)
1555                 goto out_unregister_tcp_proto;
1556
1557         rc = proto_register(&raw_prot, 1);
1558         if (rc)
1559                 goto out_unregister_udp_proto;
1560
1561         /*
1562          *      Tell SOCKET that we are alive...
1563          */
1564
1565         (void)sock_register(&inet_family_ops);
1566
1567 #ifdef CONFIG_SYSCTL
1568         ip_static_sysctl_init();
1569 #endif
1570
1571         /*
1572          *      Add all the base protocols.
1573          */
1574
1575         if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1576                 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1577         if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1578                 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1579         if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1580                 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1581 #ifdef CONFIG_IP_MULTICAST
1582         if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1583                 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1584 #endif
1585
1586         /* Register the socket-side information for inet_create. */
1587         for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1588                 INIT_LIST_HEAD(r);
1589
1590         for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1591                 inet_register_protosw(q);
1592
1593         /*
1594          *      Set the ARP module up
1595          */
1596
1597         arp_init();
1598
1599         /*
1600          *      Set the IP module up
1601          */
1602
1603         ip_init();
1604
1605         tcp_v4_init();
1606
1607         /* Setup TCP slab cache for open requests. */
1608         tcp_init();
1609
1610         /* Setup UDP memory threshold */
1611         udp_init();
1612
1613         /* Add UDP-Lite (RFC 3828) */
1614         udplite4_register();
1615
1616         /*
1617          *      Set the ICMP layer up
1618          */
1619
1620         if (icmp_init() < 0)
1621                 panic("Failed to create the ICMP control socket.\n");
1622
1623         /*
1624          *      Initialise the multicast router
1625          */
1626 #if defined(CONFIG_IP_MROUTE)
1627         if (ip_mr_init())
1628                 printk(KERN_CRIT "inet_init: Cannot init ipv4 mroute\n");
1629 #endif
1630         /*
1631          *      Initialise per-cpu ipv4 mibs
1632          */
1633
1634         if (init_ipv4_mibs())
1635                 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n");
1636
1637         ipv4_proc_init();
1638
1639         ipfrag_init();
1640
1641         dev_add_pack(&ip_packet_type);
1642
1643         rc = 0;
1644 out:
1645         return rc;
1646 out_unregister_udp_proto:
1647         proto_unregister(&udp_prot);
1648 out_unregister_tcp_proto:
1649         proto_unregister(&tcp_prot);
1650         goto out;
1651 }
1652
1653 fs_initcall(inet_init);
1654
1655 /* ------------------------------------------------------------------------ */
1656
1657 #ifdef CONFIG_PROC_FS
1658 static int __init ipv4_proc_init(void)
1659 {
1660         int rc = 0;
1661
1662         if (raw_proc_init())
1663                 goto out_raw;
1664         if (tcp4_proc_init())
1665                 goto out_tcp;
1666         if (udp4_proc_init())
1667                 goto out_udp;
1668         if (ip_misc_proc_init())
1669                 goto out_misc;
1670 out:
1671         return rc;
1672 out_misc:
1673         udp4_proc_exit();
1674 out_udp:
1675         tcp4_proc_exit();
1676 out_tcp:
1677         raw_proc_exit();
1678 out_raw:
1679         rc = -ENOMEM;
1680         goto out;
1681 }
1682
1683 #else /* CONFIG_PROC_FS */
1684 static int __init ipv4_proc_init(void)
1685 {
1686         return 0;
1687 }
1688 #endif /* CONFIG_PROC_FS */
1689
1690 MODULE_ALIAS_NETPROTO(PF_INET);
1691