SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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->num) {
 178                 if (sk->sk_prot->get_port(sk, 0)) {
 179                         release_sock(sk);
 180                         return -EAGAIN;
 181                 }
 182                 inet->sport = htons(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 {
 267         struct sock *sk;
 268         struct inet_protosw *answer;
 269         struct inet_sock *inet;
 270         struct proto *answer_prot;
 271         unsigned char answer_flags;
 272         char answer_no_check;
 273         int try_loading_module = 0;
 274         int err;
 275
 276         if (unlikely(!inet_ehash_secret))
 277                 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
 278                         build_ehash_secret();
 279
 280         sock->state = SS_UNCONNECTED;
 281
 282         /* Look for the requested type/protocol pair. */
 283 lookup_protocol:
 284         err = -ESOCKTNOSUPPORT;
 285         rcu_read_lock();
 286         list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
 287
 288                 err = 0;
 289                 /* Check the non-wild match. */
 290                 if (protocol == answer->protocol) {
 291                         if (protocol != IPPROTO_IP)
 292                                 break;
 293                 } else {
 294                         /* Check for the two wild cases. */
 295                         if (IPPROTO_IP == protocol) {
 296                                 protocol = answer->protocol;
 297                                 break;
 298                         }
 299                         if (IPPROTO_IP == answer->protocol)
 300                                 break;
 301                 }
 302                 err = -EPROTONOSUPPORT;
 303         }
 304
 305         if (unlikely(err)) {
 306                 if (try_loading_module < 2) {
 307                         rcu_read_unlock();
 308                         /*
 309                          * Be more specific, e.g. net-pf-2-proto-132-type-1
 310                          * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
 311                          */
 312                         if (++try_loading_module == 1)
 313                                 request_module("net-pf-%d-proto-%d-type-%d",
 314                                                PF_INET, protocol, sock->type);
 315                         /*
 316                          * Fall back to generic, e.g. net-pf-2-proto-132
 317                          * (net-pf-PF_INET-proto-IPPROTO_SCTP)
 318                          */
 319                         else
 320                                 request_module("net-pf-%d-proto-%d",
 321                                                PF_INET, protocol);
 322                         goto lookup_protocol;
 323                 } else
 324                         goto out_rcu_unlock;
 325         }
 326
 327         err = -EPERM;
 328         if (answer->capability > 0 && !capable(answer->capability))
 329                 goto out_rcu_unlock;
 330
 331         err = -EAFNOSUPPORT;
 332         if (!inet_netns_ok(net, protocol))
 333                 goto out_rcu_unlock;
 334
 335         sock->ops = answer->ops;
 336         answer_prot = answer->prot;
 337         answer_no_check = answer->no_check;
 338         answer_flags = answer->flags;
 339         rcu_read_unlock();
 340
 341         WARN_ON(answer_prot->slab == NULL);
 342
 343         err = -ENOBUFS;
 344         sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
 345         if (sk == NULL)
 346                 goto out;
 347
 348         err = 0;
 349         sk->sk_no_check = answer_no_check;
 350         if (INET_PROTOSW_REUSE & answer_flags)
 351                 sk->sk_reuse = 1;
 352
 353         inet = inet_sk(sk);
 354         inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
 355
 356         if (SOCK_RAW == sock->type) {
 357                 inet->num = protocol;
 358                 if (IPPROTO_RAW == protocol)
 359                         inet->hdrincl = 1;
 360         }
 361
 362         if (ipv4_config.no_pmtu_disc)
 363                 inet->pmtudisc = IP_PMTUDISC_DONT;
 364         else
 365                 inet->pmtudisc = IP_PMTUDISC_WANT;
 366
 367         inet->id = 0;
 368
 369         sock_init_data(sock, sk);
 370
 371         sk->sk_destruct    = inet_sock_destruct;
 372         sk->sk_protocol    = protocol;
 373         sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
 374
 375         inet->uc_ttl    = -1;
 376         inet->mc_loop   = 1;
 377         inet->mc_ttl    = 1;
 378         inet->mc_all    = 1;
 379         inet->mc_index  = 0;
 380         inet->mc_list   = NULL;
 381
 382         sk_refcnt_debug_inc(sk);
 383
 384         if (inet->num) {
 385                 /* It assumes that any protocol which allows
 386                  * the user to assign a number at socket
 387                  * creation time automatically
 388                  * shares.
 389                  */
 390                 inet->sport = htons(inet->num);
 391                 /* Add to protocol hash chains. */
 392                 sk->sk_prot->hash(sk);
 393         }
 394
 395         if (sk->sk_prot->init) {
 396                 err = sk->sk_prot->init(sk);
 397                 if (err)
 398                         sk_common_release(sk);
 399         }
 400 out:
 401         return err;
 402 out_rcu_unlock:
 403         rcu_read_unlock();
 404         goto out;
 405 }
 406
 407
 408 /*
 409  *      The peer socket should always be NULL (or else). When we call this
 410  *      function we are destroying the object and from then on nobody
 411  *      should refer to it.
 412  */
 413 int inet_release(struct socket *sock)
 414 {
 415         struct sock *sk = sock->sk;
 416
 417         if (sk) {
 418                 long timeout;
 419
 420                 /* Applications forget to leave groups before exiting */
 421                 ip_mc_drop_socket(sk);
 422
 423                 /* If linger is set, we don't return until the close
 424                  * is complete.  Otherwise we return immediately. The
 425                  * actually closing is done the same either way.
 426                  *
 427                  * If the close is due to the process exiting, we never
 428                  * linger..
 429                  */
 430                 timeout = 0;
 431                 if (sock_flag(sk, SOCK_LINGER) &&
 432                     !(current->flags & PF_EXITING))
 433                         timeout = sk->sk_lingertime;
 434                 sock->sk = NULL;
 435                 sk->sk_prot->close(sk, timeout);
 436         }
 437         return 0;
 438 }
 439 EXPORT_SYMBOL(inet_release);
 440
 441 /* It is off by default, see below. */
 442 int sysctl_ip_nonlocal_bind __read_mostly;
 443 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
 444
 445 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 446 {
 447         struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
 448         struct sock *sk = sock->sk;
 449         struct inet_sock *inet = inet_sk(sk);
 450         unsigned short snum;
 451         int chk_addr_ret;
 452         int err;
 453
 454         /* If the socket has its own bind function then use it. (RAW) */
 455         if (sk->sk_prot->bind) {
 456                 err = sk->sk_prot->bind(sk, uaddr, addr_len);
 457                 goto out;
 458         }
 459         err = -EINVAL;
 460         if (addr_len < sizeof(struct sockaddr_in))
 461                 goto out;
 462
 463         chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr);
 464
 465         /* Not specified by any standard per-se, however it breaks too
 466          * many applications when removed.  It is unfortunate since
 467          * allowing applications to make a non-local bind solves
 468          * several problems with systems using dynamic addressing.
 469          * (ie. your servers still start up even if your ISDN link
 470          *  is temporarily down)
 471          */
 472         err = -EADDRNOTAVAIL;
 473         if (!sysctl_ip_nonlocal_bind &&
 474             !(inet->freebind || inet->transparent) &&
 475             addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
 476             chk_addr_ret != RTN_LOCAL &&
 477             chk_addr_ret != RTN_MULTICAST &&
 478             chk_addr_ret != RTN_BROADCAST)
 479                 goto out;
 480
 481         snum = ntohs(addr->sin_port);
 482         err = -EACCES;
 483         if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
 484                 goto out;
 485
 486         /*      We keep a pair of addresses. rcv_saddr is the one
 487          *      used by hash lookups, and saddr is used for transmit.
 488          *
 489          *      In the BSD API these are the same except where it
 490          *      would be illegal to use them (multicast/broadcast) in
 491          *      which case the sending device address is used.
 492          */
 493         lock_sock(sk);
 494
 495         /* Check these errors (active socket, double bind). */
 496         err = -EINVAL;
 497         if (sk->sk_state != TCP_CLOSE || inet->num)
 498                 goto out_release_sock;
 499
 500         inet->rcv_saddr = inet->saddr = addr->sin_addr.s_addr;
 501         if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
 502                 inet->saddr = 0;  /* Use device */
 503
 504         /* Make sure we are allowed to bind here. */
 505         if (sk->sk_prot->get_port(sk, snum)) {
 506                 inet->saddr = inet->rcv_saddr = 0;
 507                 err = -EADDRINUSE;
 508                 goto out_release_sock;
 509         }
 510
 511         if (inet->rcv_saddr)
 512                 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
 513         if (snum)
 514                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
 515         inet->sport = htons(inet->num);
 516         inet->daddr = 0;
 517         inet->dport = 0;
 518         sk_dst_reset(sk);
 519         err = 0;
 520 out_release_sock:
 521         release_sock(sk);
 522 out:
 523         return err;
 524 }
 525 EXPORT_SYMBOL(inet_bind);
 526
 527 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
 528                        int addr_len, int flags)
 529 {
 530         struct sock *sk = sock->sk;
 531
 532         if (uaddr->sa_family == AF_UNSPEC)
 533                 return sk->sk_prot->disconnect(sk, flags);
 534
 535         if (!inet_sk(sk)->num && inet_autobind(sk))
 536                 return -EAGAIN;
 537         return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
 538 }
 539 EXPORT_SYMBOL(inet_dgram_connect);
 540
 541 static long inet_wait_for_connect(struct sock *sk, long timeo)
 542 {
 543         DEFINE_WAIT(wait);
 544
 545         prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 546
 547         /* Basic assumption: if someone sets sk->sk_err, he _must_
 548          * change state of the socket from TCP_SYN_*.
 549          * Connect() does not allow to get error notifications
 550          * without closing the socket.
 551          */
 552         while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 553                 release_sock(sk);
 554                 timeo = schedule_timeout(timeo);
 555                 lock_sock(sk);
 556                 if (signal_pending(current) || !timeo)
 557                         break;
 558                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 559         }
 560         finish_wait(sk->sk_sleep, &wait);
 561         return timeo;
 562 }
 563
 564 /*
 565  *      Connect to a remote host. There is regrettably still a little
 566  *      TCP 'magic' in here.
 567  */
 568 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 569                         int addr_len, int flags)
 570 {
 571         struct sock *sk = sock->sk;
 572         int err;
 573         long timeo;
 574
 575         lock_sock(sk);
 576
 577         if (uaddr->sa_family == AF_UNSPEC) {
 578                 err = sk->sk_prot->disconnect(sk, flags);
 579                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 580                 goto out;
 581         }
 582
 583         switch (sock->state) {
 584         default:
 585                 err = -EINVAL;
 586                 goto out;
 587         case SS_CONNECTED:
 588                 err = -EISCONN;
 589                 goto out;
 590         case SS_CONNECTING:
 591                 err = -EALREADY;
 592                 /* Fall out of switch with err, set for this state */
 593                 break;
 594         case SS_UNCONNECTED:
 595                 err = -EISCONN;
 596                 if (sk->sk_state != TCP_CLOSE)
 597                         goto out;
 598
 599                 err = sk->sk_prot->connect(sk, uaddr, addr_len);
 600                 if (err < 0)
 601                         goto out;
 602
 603                 sock->state = SS_CONNECTING;
 604
 605                 /* Just entered SS_CONNECTING state; the only
 606                  * difference is that return value in non-blocking
 607                  * case is EINPROGRESS, rather than EALREADY.
 608                  */
 609                 err = -EINPROGRESS;
 610                 break;
 611         }
 612
 613         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
 614
 615         if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 616                 /* Error code is set above */
 617                 if (!timeo || !inet_wait_for_connect(sk, timeo))
 618                         goto out;
 619
 620                 err = sock_intr_errno(timeo);
 621                 if (signal_pending(current))
 622                         goto out;
 623         }
 624
 625         /* Connection was closed by RST, timeout, ICMP error
 626          * or another process disconnected us.
 627          */
 628         if (sk->sk_state == TCP_CLOSE)
 629                 goto sock_error;
 630
 631         /* sk->sk_err may be not zero now, if RECVERR was ordered by user
 632          * and error was received after socket entered established state.
 633          * Hence, it is handled normally after connect() return successfully.
 634          */
 635
 636         sock->state = SS_CONNECTED;
 637         err = 0;
 638 out:
 639         release_sock(sk);
 640         return err;
 641
 642 sock_error:
 643         err = sock_error(sk) ? : -ECONNABORTED;
 644         sock->state = SS_UNCONNECTED;
 645         if (sk->sk_prot->disconnect(sk, flags))
 646                 sock->state = SS_DISCONNECTING;
 647         goto out;
 648 }
 649 EXPORT_SYMBOL(inet_stream_connect);
 650
 651 /*
 652  *      Accept a pending connection. The TCP layer now gives BSD semantics.
 653  */
 654
 655 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
 656 {
 657         struct sock *sk1 = sock->sk;
 658         int err = -EINVAL;
 659         struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
 660
 661         if (!sk2)
 662                 goto do_err;
 663
 664         lock_sock(sk2);
 665
 666         WARN_ON(!((1 << sk2->sk_state) &
 667                   (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE)));
 668
 669         sock_graft(sk2, newsock);
 670
 671         newsock->state = SS_CONNECTED;
 672         err = 0;
 673         release_sock(sk2);
 674 do_err:
 675         return err;
 676 }
 677 EXPORT_SYMBOL(inet_accept);
 678
 679
 680 /*
 681  *      This does both peername and sockname.
 682  */
 683 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
 684                         int *uaddr_len, int peer)
 685 {
 686         struct sock *sk         = sock->sk;
 687         struct inet_sock *inet  = inet_sk(sk);
 688         struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
 689
 690         sin->sin_family = AF_INET;
 691         if (peer) {
 692                 if (!inet->dport ||
 693                     (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
 694                      peer == 1))
 695                         return -ENOTCONN;
 696                 sin->sin_port = inet->dport;
 697                 sin->sin_addr.s_addr = inet->daddr;
 698         } else {
 699                 __be32 addr = inet->rcv_saddr;
 700                 if (!addr)
 701                         addr = inet->saddr;
 702                 sin->sin_port = inet->sport;
 703                 sin->sin_addr.s_addr = addr;
 704         }
 705         memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 706         *uaddr_len = sizeof(*sin);
 707         return 0;
 708 }
 709 EXPORT_SYMBOL(inet_getname);
 710
 711 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
 712                  size_t size)
 713 {
 714         struct sock *sk = sock->sk;
 715
 716         /* We may need to bind the socket. */
 717         if (!inet_sk(sk)->num && inet_autobind(sk))
 718                 return -EAGAIN;
 719
 720         return sk->sk_prot->sendmsg(iocb, sk, msg, size);
 721 }
 722 EXPORT_SYMBOL(inet_sendmsg);
 723
 724
 725 static ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
 726                              size_t size, int flags)
 727 {
 728         struct sock *sk = sock->sk;
 729
 730         /* We may need to bind the socket. */
 731         if (!inet_sk(sk)->num && inet_autobind(sk))
 732                 return -EAGAIN;
 733
 734         if (sk->sk_prot->sendpage)
 735                 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
 736         return sock_no_sendpage(sock, page, offset, size, flags);
 737 }
 738
 739
 740 int inet_shutdown(struct socket *sock, int how)
 741 {
 742         struct sock *sk = sock->sk;
 743         int err = 0;
 744
 745         /* This should really check to make sure
 746          * the socket is a TCP socket. (WHY AC...)
 747          */
 748         how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
 749                        1->2 bit 2 snds.
 750                        2->3 */
 751         if ((how & ~SHUTDOWN_MASK) || !how)     /* MAXINT->0 */
 752                 return -EINVAL;
 753
 754         lock_sock(sk);
 755         if (sock->state == SS_CONNECTING) {
 756                 if ((1 << sk->sk_state) &
 757                     (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
 758                         sock->state = SS_DISCONNECTING;
 759                 else
 760                         sock->state = SS_CONNECTED;
 761         }
 762
 763         switch (sk->sk_state) {
 764         case TCP_CLOSE:
 765                 err = -ENOTCONN;
 766                 /* Hack to wake up other listeners, who can poll for
 767                    POLLHUP, even on eg. unconnected UDP sockets -- RR */
 768         default:
 769                 sk->sk_shutdown |= how;
 770                 if (sk->sk_prot->shutdown)
 771                         sk->sk_prot->shutdown(sk, how);
 772                 break;
 773
 774         /* Remaining two branches are temporary solution for missing
 775          * close() in multithreaded environment. It is _not_ a good idea,
 776          * but we have no choice until close() is repaired at VFS level.
 777          */
 778         case TCP_LISTEN:
 779                 if (!(how & RCV_SHUTDOWN))
 780                         break;
 781                 /* Fall through */
 782         case TCP_SYN_SENT:
 783                 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
 784                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 785                 break;
 786         }
 787
 788         /* Wake up anyone sleeping in poll. */
 789         sk->sk_state_change(sk);
 790         release_sock(sk);
 791         return err;
 792 }
 793 EXPORT_SYMBOL(inet_shutdown);
 794
 795 /*
 796  *      ioctl() calls you can issue on an INET socket. Most of these are
 797  *      device configuration and stuff and very rarely used. Some ioctls
 798  *      pass on to the socket itself.
 799  *
 800  *      NOTE: I like the idea of a module for the config stuff. ie ifconfig
 801  *      loads the devconfigure module does its configuring and unloads it.
 802  *      There's a good 20K of config code hanging around the kernel.
 803  */
 804
 805 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 806 {
 807         struct sock *sk = sock->sk;
 808         int err = 0;
 809         struct net *net = sock_net(sk);
 810
 811         switch (cmd) {
 812         case SIOCGSTAMP:
 813                 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
 814                 break;
 815         case SIOCGSTAMPNS:
 816                 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
 817                 break;
 818         case SIOCADDRT:
 819         case SIOCDELRT:
 820         case SIOCRTMSG:
 821                 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
 822                 break;
 823         case SIOCDARP:
 824         case SIOCGARP:
 825         case SIOCSARP:
 826                 err = arp_ioctl(net, cmd, (void __user *)arg);
 827                 break;
 828         case SIOCGIFADDR:
 829         case SIOCSIFADDR:
 830         case SIOCGIFBRDADDR:
 831         case SIOCSIFBRDADDR:
 832         case SIOCGIFNETMASK:
 833         case SIOCSIFNETMASK:
 834         case SIOCGIFDSTADDR:
 835         case SIOCSIFDSTADDR:
 836         case SIOCSIFPFLAGS:
 837         case SIOCGIFPFLAGS:
 838         case SIOCSIFFLAGS:
 839                 err = devinet_ioctl(net, cmd, (void __user *)arg);
 840                 break;
 841         default:
 842                 if (sk->sk_prot->ioctl)
 843                         err = sk->sk_prot->ioctl(sk, cmd, arg);
 844                 else
 845                         err = -ENOIOCTLCMD;
 846                 break;
 847         }
 848         return err;
 849 }
 850 EXPORT_SYMBOL(inet_ioctl);
 851
 852 const struct proto_ops inet_stream_ops = {
 853         .family            = PF_INET,
 854         .owner             = THIS_MODULE,
 855         .release           = inet_release,
 856         .bind              = inet_bind,
 857         .connect           = inet_stream_connect,
 858         .socketpair        = sock_no_socketpair,
 859         .accept            = inet_accept,
 860         .getname           = inet_getname,
 861         .poll              = tcp_poll,
 862         .ioctl             = inet_ioctl,
 863         .listen            = inet_listen,
 864         .shutdown          = inet_shutdown,
 865         .setsockopt        = sock_common_setsockopt,
 866         .getsockopt        = sock_common_getsockopt,
 867         .sendmsg           = tcp_sendmsg,
 868         .recvmsg           = sock_common_recvmsg,
 869         .mmap              = sock_no_mmap,
 870         .sendpage          = tcp_sendpage,
 871         .splice_read       = tcp_splice_read,
 872 #ifdef CONFIG_COMPAT
 873         .compat_setsockopt = compat_sock_common_setsockopt,
 874         .compat_getsockopt = compat_sock_common_getsockopt,
 875 #endif
 876 };
 877 EXPORT_SYMBOL(inet_stream_ops);
 878
 879 const struct proto_ops inet_dgram_ops = {
 880         .family            = PF_INET,
 881         .owner             = THIS_MODULE,
 882         .release           = inet_release,
 883         .bind              = inet_bind,
 884         .connect           = inet_dgram_connect,
 885         .socketpair        = sock_no_socketpair,
 886         .accept            = sock_no_accept,
 887         .getname           = inet_getname,
 888         .poll              = udp_poll,
 889         .ioctl             = inet_ioctl,
 890         .listen            = sock_no_listen,
 891         .shutdown          = inet_shutdown,
 892         .setsockopt        = sock_common_setsockopt,
 893         .getsockopt        = sock_common_getsockopt,
 894         .sendmsg           = inet_sendmsg,
 895         .recvmsg           = sock_common_recvmsg,
 896         .mmap              = sock_no_mmap,
 897         .sendpage          = inet_sendpage,
 898 #ifdef CONFIG_COMPAT
 899         .compat_setsockopt = compat_sock_common_setsockopt,
 900         .compat_getsockopt = compat_sock_common_getsockopt,
 901 #endif
 902 };
 903 EXPORT_SYMBOL(inet_dgram_ops);
 904
 905 /*
 906  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
 907  * udp_poll
 908  */
 909 static const struct proto_ops inet_sockraw_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              = datagram_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           = sock_common_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
 934 static struct net_proto_family inet_family_ops = {
 935         .family = PF_INET,
 936         .create = inet_create,
 937         .owner  = THIS_MODULE,
 938 };
 939
 940 /* Upon startup we insert all the elements in inetsw_array[] into
 941  * the linked list inetsw.
 942  */
 943 static struct inet_protosw inetsw_array[] =
 944 {
 945         {
 946                 .type =       SOCK_STREAM,
 947                 .protocol =   IPPROTO_TCP,
 948                 .prot =       &tcp_prot,
 949                 .ops =        &inet_stream_ops,
 950                 .capability = -1,
 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                 .capability = -1,
 962                 .no_check =   UDP_CSUM_DEFAULT,
 963                 .flags =      INET_PROTOSW_PERMANENT,
 964        },
 965
 966
 967        {
 968                .type =       SOCK_RAW,
 969                .protocol =   IPPROTO_IP,        /* wild card */
 970                .prot =       &raw_prot,
 971                .ops =        &inet_sockraw_ops,
 972                .capability = CAP_NET_RAW,
 973                .no_check =   UDP_CSUM_DEFAULT,
 974                .flags =      INET_PROTOSW_REUSE,
 975        }
 976 };
 977
 978 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
 979
 980 void inet_register_protosw(struct inet_protosw *p)
 981 {
 982         struct list_head *lh;
 983         struct inet_protosw *answer;
 984         int protocol = p->protocol;
 985         struct list_head *last_perm;
 986
 987         spin_lock_bh(&inetsw_lock);
 988
 989         if (p->type >= SOCK_MAX)
 990                 goto out_illegal;
 991
 992         /* If we are trying to override a permanent protocol, bail. */
 993         answer = NULL;
 994         last_perm = &inetsw[p->type];
 995         list_for_each(lh, &inetsw[p->type]) {
 996                 answer = list_entry(lh, struct inet_protosw, list);
 997
 998                 /* Check only the non-wild match. */
 999                 if (INET_PROTOSW_PERMANENT & answer->flags) {
1000                         if (protocol == answer->protocol)
1001                                 break;
1002                         last_perm = lh;
1003                 }
1004
1005                 answer = NULL;
1006         }
1007         if (answer)
1008                 goto out_permanent;
1009
1010         /* Add the new entry after the last permanent entry if any, so that
1011          * the new entry does not override a permanent entry when matched with
1012          * a wild-card protocol. But it is allowed to override any existing
1013          * non-permanent entry.  This means that when we remove this entry, the
1014          * system automatically returns to the old behavior.
1015          */
1016         list_add_rcu(&p->list, last_perm);
1017 out:
1018         spin_unlock_bh(&inetsw_lock);
1019
1020         return;
1021
1022 out_permanent:
1023         printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
1024                protocol);
1025         goto out;
1026
1027 out_illegal:
1028         printk(KERN_ERR
1029                "Ignoring attempt to register invalid socket type %d.\n",
1030                p->type);
1031         goto out;
1032 }
1033 EXPORT_SYMBOL(inet_register_protosw);
1034
1035 void inet_unregister_protosw(struct inet_protosw *p)
1036 {
1037         if (INET_PROTOSW_PERMANENT & p->flags) {
1038                 printk(KERN_ERR
1039                        "Attempt to unregister permanent protocol %d.\n",
1040                        p->protocol);
1041         } else {
1042                 spin_lock_bh(&inetsw_lock);
1043                 list_del_rcu(&p->list);
1044                 spin_unlock_bh(&inetsw_lock);
1045
1046                 synchronize_net();
1047         }
1048 }
1049 EXPORT_SYMBOL(inet_unregister_protosw);
1050
1051 /*
1052  *      Shall we try to damage output packets if routing dev changes?
1053  */
1054
1055 int sysctl_ip_dynaddr __read_mostly;
1056
1057 static int inet_sk_reselect_saddr(struct sock *sk)
1058 {
1059         struct inet_sock *inet = inet_sk(sk);
1060         int err;
1061         struct rtable *rt;
1062         __be32 old_saddr = inet->saddr;
1063         __be32 new_saddr;
1064         __be32 daddr = inet->daddr;
1065
1066         if (inet->opt && inet->opt->srr)
1067                 daddr = inet->opt->faddr;
1068
1069         /* Query new route. */
1070         err = ip_route_connect(&rt, daddr, 0,
1071                                RT_CONN_FLAGS(sk),
1072                                sk->sk_bound_dev_if,
1073                                sk->sk_protocol,
1074                                inet->sport, inet->dport, sk, 0);
1075         if (err)
1076                 return err;
1077
1078         sk_setup_caps(sk, &rt->u.dst);
1079
1080         new_saddr = rt->rt_src;
1081
1082         if (new_saddr == old_saddr)
1083                 return 0;
1084
1085         if (sysctl_ip_dynaddr > 1) {
1086                 printk(KERN_INFO "%s(): shifting inet->saddr from %pI4 to %pI4\n",
1087                        __func__, &old_saddr, &new_saddr);
1088         }
1089
1090         inet->saddr = inet->rcv_saddr = new_saddr;
1091
1092         /*
1093          * XXX The only one ugly spot where we need to
1094          * XXX really change the sockets identity after
1095          * XXX it has entered the hashes. -DaveM
1096          *
1097          * Besides that, it does not check for connection
1098          * uniqueness. Wait for troubles.
1099          */
1100         __sk_prot_rehash(sk);
1101         return 0;
1102 }
1103
1104 int inet_sk_rebuild_header(struct sock *sk)
1105 {
1106         struct inet_sock *inet = inet_sk(sk);
1107         struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1108         __be32 daddr;
1109         int err;
1110
1111         /* Route is OK, nothing to do. */
1112         if (rt)
1113                 return 0;
1114
1115         /* Reroute. */
1116         daddr = inet->daddr;
1117         if (inet->opt && inet->opt->srr)
1118                 daddr = inet->opt->faddr;
1119 {
1120         struct flowi fl = {
1121                 .oif = sk->sk_bound_dev_if,
1122                 .nl_u = {
1123                         .ip4_u = {
1124                                 .daddr  = daddr,
1125                                 .saddr  = inet->saddr,
1126                                 .tos    = RT_CONN_FLAGS(sk),
1127                         },
1128                 },
1129                 .proto = sk->sk_protocol,
1130                 .flags = inet_sk_flowi_flags(sk),
1131                 .uli_u = {
1132                         .ports = {
1133                                 .sport = inet->sport,
1134                                 .dport = inet->dport,
1135                         },
1136                 },
1137         };
1138
1139         security_sk_classify_flow(sk, &fl);
1140         err = ip_route_output_flow(sock_net(sk), &rt, &fl, sk, 0);
1141 }
1142         if (!err)
1143                 sk_setup_caps(sk, &rt->u.dst);
1144         else {
1145                 /* Routing failed... */
1146                 sk->sk_route_caps = 0;
1147                 /*
1148                  * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1149                  * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1150                  */
1151                 if (!sysctl_ip_dynaddr ||
1152                     sk->sk_state != TCP_SYN_SENT ||
1153                     (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1154                     (err = inet_sk_reselect_saddr(sk)) != 0)
1155                         sk->sk_err_soft = -err;
1156         }
1157
1158         return err;
1159 }
1160 EXPORT_SYMBOL(inet_sk_rebuild_header);
1161
1162 static int inet_gso_send_check(struct sk_buff *skb)
1163 {
1164         struct iphdr *iph;
1165         const struct net_protocol *ops;
1166         int proto;
1167         int ihl;
1168         int err = -EINVAL;
1169
1170         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1171                 goto out;
1172
1173         iph = ip_hdr(skb);
1174         ihl = iph->ihl * 4;
1175         if (ihl < sizeof(*iph))
1176                 goto out;
1177
1178         if (unlikely(!pskb_may_pull(skb, ihl)))
1179                 goto out;
1180
1181         __skb_pull(skb, ihl);
1182         skb_reset_transport_header(skb);
1183         iph = ip_hdr(skb);
1184         proto = iph->protocol & (MAX_INET_PROTOS - 1);
1185         err = -EPROTONOSUPPORT;
1186
1187         rcu_read_lock();
1188         ops = rcu_dereference(inet_protos[proto]);
1189         if (likely(ops && ops->gso_send_check))
1190                 err = ops->gso_send_check(skb);
1191         rcu_read_unlock();
1192
1193 out:
1194         return err;
1195 }
1196
1197 static struct sk_buff *inet_gso_segment(struct sk_buff *skb, int features)
1198 {
1199         struct sk_buff *segs = ERR_PTR(-EINVAL);
1200         struct iphdr *iph;
1201         const struct net_protocol *ops;
1202         int proto;
1203         int ihl;
1204         int id;
1205         unsigned int offset = 0;
1206
1207         if (!(features & NETIF_F_V4_CSUM))
1208                 features &= ~NETIF_F_SG;
1209
1210         if (unlikely(skb_shinfo(skb)->gso_type &
1211                      ~(SKB_GSO_TCPV4 |
1212                        SKB_GSO_UDP |
1213                        SKB_GSO_DODGY |
1214                        SKB_GSO_TCP_ECN |
1215                        0)))
1216                 goto out;
1217
1218         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1219                 goto out;
1220
1221         iph = ip_hdr(skb);
1222         ihl = iph->ihl * 4;
1223         if (ihl < sizeof(*iph))
1224                 goto out;
1225
1226         if (unlikely(!pskb_may_pull(skb, ihl)))
1227                 goto out;
1228
1229         __skb_pull(skb, ihl);
1230         skb_reset_transport_header(skb);
1231         iph = ip_hdr(skb);
1232         id = ntohs(iph->id);
1233         proto = iph->protocol & (MAX_INET_PROTOS - 1);
1234         segs = ERR_PTR(-EPROTONOSUPPORT);
1235
1236         rcu_read_lock();
1237         ops = rcu_dereference(inet_protos[proto]);
1238         if (likely(ops && ops->gso_segment))
1239                 segs = ops->gso_segment(skb, features);
1240         rcu_read_unlock();
1241
1242         if (!segs || IS_ERR(segs))
1243                 goto out;
1244
1245         skb = segs;
1246         do {
1247                 iph = ip_hdr(skb);
1248                 if (proto == IPPROTO_UDP) {
1249                         iph->id = htons(id);
1250                         iph->frag_off = htons(offset >> 3);
1251                         if (skb->next != NULL)
1252                                 iph->frag_off |= htons(IP_MF);
1253                         offset += (skb->len - skb->mac_len - iph->ihl * 4);
1254                 } else
1255                         iph->id = htons(id++);
1256                 iph->tot_len = htons(skb->len - skb->mac_len);
1257                 iph->check = 0;
1258                 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1259         } while ((skb = skb->next));
1260
1261 out:
1262         return segs;
1263 }
1264
1265 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1266                                          struct sk_buff *skb)
1267 {
1268         const struct net_protocol *ops;
1269         struct sk_buff **pp = NULL;
1270         struct sk_buff *p;
1271         struct iphdr *iph;
1272         unsigned int hlen;
1273         unsigned int off;
1274         unsigned int id;
1275         int flush = 1;
1276         int proto;
1277
1278         off = skb_gro_offset(skb);
1279         hlen = off + sizeof(*iph);
1280         iph = skb_gro_header_fast(skb, off);
1281         if (skb_gro_header_hard(skb, hlen)) {
1282                 iph = skb_gro_header_slow(skb, hlen, off);
1283                 if (unlikely(!iph))
1284                         goto out;
1285         }
1286
1287         proto = iph->protocol & (MAX_INET_PROTOS - 1);
1288
1289         rcu_read_lock();
1290         ops = rcu_dereference(inet_protos[proto]);
1291         if (!ops || !ops->gro_receive)
1292                 goto out_unlock;
1293
1294         if (*(u8 *)iph != 0x45)
1295                 goto out_unlock;
1296
1297         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1298                 goto out_unlock;
1299
1300         id = ntohl(*(u32 *)&iph->id);
1301         flush = (u16)((ntohl(*(u32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF));
1302         id >>= 16;
1303
1304         for (p = *head; p; p = p->next) {
1305                 struct iphdr *iph2;
1306
1307                 if (!NAPI_GRO_CB(p)->same_flow)
1308                         continue;
1309
1310                 iph2 = ip_hdr(p);
1311
1312                 if ((iph->protocol ^ iph2->protocol) |
1313                     (iph->tos ^ iph2->tos) |
1314                     (iph->saddr ^ iph2->saddr) |
1315                     (iph->daddr ^ iph2->daddr)) {
1316                         NAPI_GRO_CB(p)->same_flow = 0;
1317                         continue;
1318                 }
1319
1320                 /* All fields must match except length and checksum. */
1321                 NAPI_GRO_CB(p)->flush |=
1322                         (iph->ttl ^ iph2->ttl) |
1323                         ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1324
1325                 NAPI_GRO_CB(p)->flush |= flush;
1326         }
1327
1328         NAPI_GRO_CB(skb)->flush |= flush;
1329         skb_gro_pull(skb, sizeof(*iph));
1330         skb_set_transport_header(skb, skb_gro_offset(skb));
1331
1332         pp = ops->gro_receive(head, skb);
1333
1334 out_unlock:
1335         rcu_read_unlock();
1336
1337 out:
1338         NAPI_GRO_CB(skb)->flush |= flush;
1339
1340         return pp;
1341 }
1342
1343 static int inet_gro_complete(struct sk_buff *skb)
1344 {
1345         const struct net_protocol *ops;
1346         struct iphdr *iph = ip_hdr(skb);
1347         int proto = iph->protocol & (MAX_INET_PROTOS - 1);
1348         int err = -ENOSYS;
1349         __be16 newlen = htons(skb->len - skb_network_offset(skb));
1350
1351         csum_replace2(&iph->check, iph->tot_len, newlen);
1352         iph->tot_len = newlen;
1353
1354         rcu_read_lock();
1355         ops = rcu_dereference(inet_protos[proto]);
1356         if (WARN_ON(!ops || !ops->gro_complete))
1357                 goto out_unlock;
1358
1359         err = ops->gro_complete(skb);
1360
1361 out_unlock:
1362         rcu_read_unlock();
1363
1364         return err;
1365 }
1366
1367 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1368                          unsigned short type, unsigned char protocol,
1369                          struct net *net)
1370 {
1371         struct socket *sock;
1372         int rc = sock_create_kern(family, type, protocol, &sock);
1373
1374         if (rc == 0) {
1375                 *sk = sock->sk;
1376                 (*sk)->sk_allocation = GFP_ATOMIC;
1377                 /*
1378                  * Unhash it so that IP input processing does not even see it,
1379                  * we do not wish this socket to see incoming packets.
1380                  */
1381                 (*sk)->sk_prot->unhash(*sk);
1382
1383                 sk_change_net(*sk, net);
1384         }
1385         return rc;
1386 }
1387 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1388
1389 unsigned long snmp_fold_field(void *mib[], int offt)
1390 {
1391         unsigned long res = 0;
1392         int i;
1393
1394         for_each_possible_cpu(i) {
1395                 res += *(((unsigned long *) per_cpu_ptr(mib[0], i)) + offt);
1396                 res += *(((unsigned long *) per_cpu_ptr(mib[1], i)) + offt);
1397         }
1398         return res;
1399 }
1400 EXPORT_SYMBOL_GPL(snmp_fold_field);
1401
1402 int snmp_mib_init(void *ptr[2], size_t mibsize)
1403 {
1404         BUG_ON(ptr == NULL);
1405         ptr[0] = __alloc_percpu(mibsize, __alignof__(unsigned long long));
1406         if (!ptr[0])
1407                 goto err0;
1408         ptr[1] = __alloc_percpu(mibsize, __alignof__(unsigned long long));
1409         if (!ptr[1])
1410                 goto err1;
1411         return 0;
1412 err1:
1413         free_percpu(ptr[0]);
1414         ptr[0] = NULL;
1415 err0:
1416         return -ENOMEM;
1417 }
1418 EXPORT_SYMBOL_GPL(snmp_mib_init);
1419
1420 void snmp_mib_free(void *ptr[2])
1421 {
1422         BUG_ON(ptr == NULL);
1423         free_percpu(ptr[0]);
1424         free_percpu(ptr[1]);
1425         ptr[0] = ptr[1] = NULL;
1426 }
1427 EXPORT_SYMBOL_GPL(snmp_mib_free);
1428
1429 #ifdef CONFIG_IP_MULTICAST
1430 static const struct net_protocol igmp_protocol = {
1431         .handler =      igmp_rcv,
1432         .netns_ok =     1,
1433 };
1434 #endif
1435
1436 static const struct net_protocol tcp_protocol = {
1437         .handler =      tcp_v4_rcv,
1438         .err_handler =  tcp_v4_err,
1439         .gso_send_check = tcp_v4_gso_send_check,
1440         .gso_segment =  tcp_tso_segment,
1441         .gro_receive =  tcp4_gro_receive,
1442         .gro_complete = tcp4_gro_complete,
1443         .no_policy =    1,
1444         .netns_ok =     1,
1445 };
1446
1447 static const struct net_protocol udp_protocol = {
1448         .handler =      udp_rcv,
1449         .err_handler =  udp_err,
1450         .gso_send_check = udp4_ufo_send_check,
1451         .gso_segment = udp4_ufo_fragment,
1452         .no_policy =    1,
1453         .netns_ok =     1,
1454 };
1455
1456 static const struct net_protocol icmp_protocol = {
1457         .handler =      icmp_rcv,
1458         .no_policy =    1,
1459         .netns_ok =     1,
1460 };
1461
1462 static __net_init int ipv4_mib_init_net(struct net *net)
1463 {
1464         if (snmp_mib_init((void **)net->mib.tcp_statistics,
1465                           sizeof(struct tcp_mib)) < 0)
1466                 goto err_tcp_mib;
1467         if (snmp_mib_init((void **)net->mib.ip_statistics,
1468                           sizeof(struct ipstats_mib)) < 0)
1469                 goto err_ip_mib;
1470         if (snmp_mib_init((void **)net->mib.net_statistics,
1471                           sizeof(struct linux_mib)) < 0)
1472                 goto err_net_mib;
1473         if (snmp_mib_init((void **)net->mib.udp_statistics,
1474                           sizeof(struct udp_mib)) < 0)
1475                 goto err_udp_mib;
1476         if (snmp_mib_init((void **)net->mib.udplite_statistics,
1477                           sizeof(struct udp_mib)) < 0)
1478                 goto err_udplite_mib;
1479         if (snmp_mib_init((void **)net->mib.icmp_statistics,
1480                           sizeof(struct icmp_mib)) < 0)
1481                 goto err_icmp_mib;
1482         if (snmp_mib_init((void **)net->mib.icmpmsg_statistics,
1483                           sizeof(struct icmpmsg_mib)) < 0)
1484                 goto err_icmpmsg_mib;
1485
1486         tcp_mib_init(net);
1487         return 0;
1488
1489 err_icmpmsg_mib:
1490         snmp_mib_free((void **)net->mib.icmp_statistics);
1491 err_icmp_mib:
1492         snmp_mib_free((void **)net->mib.udplite_statistics);
1493 err_udplite_mib:
1494         snmp_mib_free((void **)net->mib.udp_statistics);
1495 err_udp_mib:
1496         snmp_mib_free((void **)net->mib.net_statistics);
1497 err_net_mib:
1498         snmp_mib_free((void **)net->mib.ip_statistics);
1499 err_ip_mib:
1500         snmp_mib_free((void **)net->mib.tcp_statistics);
1501 err_tcp_mib:
1502         return -ENOMEM;
1503 }
1504
1505 static __net_exit void ipv4_mib_exit_net(struct net *net)
1506 {
1507         snmp_mib_free((void **)net->mib.icmpmsg_statistics);
1508         snmp_mib_free((void **)net->mib.icmp_statistics);
1509         snmp_mib_free((void **)net->mib.udplite_statistics);
1510         snmp_mib_free((void **)net->mib.udp_statistics);
1511         snmp_mib_free((void **)net->mib.net_statistics);
1512         snmp_mib_free((void **)net->mib.ip_statistics);
1513         snmp_mib_free((void **)net->mib.tcp_statistics);
1514 }
1515
1516 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1517         .init = ipv4_mib_init_net,
1518         .exit = ipv4_mib_exit_net,
1519 };
1520
1521 static int __init init_ipv4_mibs(void)
1522 {
1523         return register_pernet_subsys(&ipv4_mib_ops);
1524 }
1525
1526 static int ipv4_proc_init(void);
1527
1528 /*
1529  *      IP protocol layer initialiser
1530  */
1531
1532 static struct packet_type ip_packet_type __read_mostly = {
1533         .type = cpu_to_be16(ETH_P_IP),
1534         .func = ip_rcv,
1535         .gso_send_check = inet_gso_send_check,
1536         .gso_segment = inet_gso_segment,
1537         .gro_receive = inet_gro_receive,
1538         .gro_complete = inet_gro_complete,
1539 };
1540
1541 static int __init inet_init(void)
1542 {
1543         struct sk_buff *dummy_skb;
1544         struct inet_protosw *q;
1545         struct list_head *r;
1546         int rc = -EINVAL;
1547
1548         BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1549
1550         rc = proto_register(&tcp_prot, 1);
1551         if (rc)
1552                 goto out;
1553
1554         rc = proto_register(&udp_prot, 1);
1555         if (rc)
1556                 goto out_unregister_tcp_proto;
1557
1558         rc = proto_register(&raw_prot, 1);
1559         if (rc)
1560                 goto out_unregister_udp_proto;
1561
1562         /*
1563          *      Tell SOCKET that we are alive...
1564          */
1565
1566         (void)sock_register(&inet_family_ops);
1567
1568 #ifdef CONFIG_SYSCTL
1569         ip_static_sysctl_init();
1570 #endif
1571
1572         /*
1573          *      Add all the base protocols.
1574          */
1575
1576         if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1577                 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1578         if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1579                 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1580         if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1581                 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1582 #ifdef CONFIG_IP_MULTICAST
1583         if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1584                 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1585 #endif
1586
1587         /* Register the socket-side information for inet_create. */
1588         for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1589                 INIT_LIST_HEAD(r);
1590
1591         for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1592                 inet_register_protosw(q);
1593
1594         /*
1595          *      Set the ARP module up
1596          */
1597
1598         arp_init();
1599
1600         /*
1601          *      Set the IP module up
1602          */
1603
1604         ip_init();
1605
1606         tcp_v4_init();
1607
1608         /* Setup TCP slab cache for open requests. */
1609         tcp_init();
1610
1611         /* Setup UDP memory threshold */
1612         udp_init();
1613
1614         /* Add UDP-Lite (RFC 3828) */
1615         udplite4_register();
1616
1617         /*
1618          *      Set the ICMP layer up
1619          */
1620
1621         if (icmp_init() < 0)
1622                 panic("Failed to create the ICMP control socket.\n");
1623
1624         /*
1625          *      Initialise the multicast router
1626          */
1627 #if defined(CONFIG_IP_MROUTE)
1628         if (ip_mr_init())
1629                 printk(KERN_CRIT "inet_init: Cannot init ipv4 mroute\n");
1630 #endif
1631         /*
1632          *      Initialise per-cpu ipv4 mibs
1633          */
1634
1635         if (init_ipv4_mibs())
1636                 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n");
1637
1638         ipv4_proc_init();
1639
1640         ipfrag_init();
1641
1642         dev_add_pack(&ip_packet_type);
1643
1644         rc = 0;
1645 out:
1646         return rc;
1647 out_unregister_udp_proto:
1648         proto_unregister(&udp_prot);
1649 out_unregister_tcp_proto:
1650         proto_unregister(&tcp_prot);
1651         goto out;
1652 }
1653
1654 fs_initcall(inet_init);
1655
1656 /* ------------------------------------------------------------------------ */
1657
1658 #ifdef CONFIG_PROC_FS
1659 static int __init ipv4_proc_init(void)
1660 {
1661         int rc = 0;
1662
1663         if (raw_proc_init())
1664                 goto out_raw;
1665         if (tcp4_proc_init())
1666                 goto out_tcp;
1667         if (udp4_proc_init())
1668                 goto out_udp;
1669         if (ip_misc_proc_init())
1670                 goto out_misc;
1671 out:
1672         return rc;
1673 out_misc:
1674         udp4_proc_exit();
1675 out_udp:
1676         tcp4_proc_exit();
1677 out_tcp:
1678         raw_proc_exit();
1679 out_raw:
1680         rc = -ENOMEM;
1681         goto out;
1682 }
1683
1684 #else /* CONFIG_PROC_FS */
1685 static int __init ipv4_proc_init(void)
1686 {
1687         return 0;
1688 }
1689 #endif /* CONFIG_PROC_FS */
1690
1691 MODULE_ALIAS_NETPROTO(PF_INET);
1692