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