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