SAFE public projects git trees. - safe/jmp/linux-2.6/blob - net/ipv4/af_inet.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              PF_INET protocol family socket handler.
   7  *
   8  * Authors:     Ross Biro
   9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  10  *              Florian La Roche, <flla@stud.uni-sb.de>
  11  *              Alan Cox, <A.Cox@swansea.ac.uk>
  12  *
  13  * Changes (see also sock.c)
  14  *
  15  *              piggy,
  16  *              Karl Knutson    :       Socket protocol table
  17  *              A.N.Kuznetsov   :       Socket death error in accept().
  18  *              John Richardson :       Fix non blocking error in connect()
  19  *                                      so sockets that fail to connect
  20  *                                      don't return -EINPROGRESS.
  21  *              Alan Cox        :       Asynchronous I/O support
  22  *              Alan Cox        :       Keep correct socket pointer on sock
  23  *                                      structures
  24  *                                      when accept() ed
  25  *              Alan Cox        :       Semantics of SO_LINGER aren't state
  26  *                                      moved to close when you look carefully.
  27  *                                      With this fixed and the accept bug fixed
  28  *                                      some RPC stuff seems happier.
  29  *              Niibe Yutaka    :       4.4BSD style write async I/O
  30  *              Alan Cox,
  31  *              Tony Gale       :       Fixed reuse semantics.
  32  *              Alan Cox        :       bind() shouldn't abort existing but dead
  33  *                                      sockets. Stops FTP netin:.. I hope.
  34  *              Alan Cox        :       bind() works correctly for RAW sockets.
  35  *                                      Note that FreeBSD at least was broken
  36  *                                      in this respect so be careful with
  37  *                                      compatibility tests...
  38  *              Alan Cox        :       routing cache support
  39  *              Alan Cox        :       memzero the socket structure for
  40  *                                      compactness.
  41  *              Matt Day        :       nonblock connect error handler
  42  *              Alan Cox        :       Allow large numbers of pending sockets
  43  *                                      (eg for big web sites), but only if
  44  *                                      specifically application requested.
  45  *              Alan Cox        :       New buffering throughout IP. Used
  46  *                                      dumbly.
  47  *              Alan Cox        :       New buffering now used smartly.
  48  *              Alan Cox        :       BSD rather than common sense
  49  *                                      interpretation of listen.
  50  *              Germano Caronni :       Assorted small races.
  51  *              Alan Cox        :       sendmsg/recvmsg basic support.
  52  *              Alan Cox        :       Only sendmsg/recvmsg now supported.
  53  *              Alan Cox        :       Locked down bind (see security list).
  54  *              Alan Cox        :       Loosened bind a little.
  55  *              Mike McLagan    :       ADD/DEL DLCI Ioctls
  56  *      Willy Konynenberg       :       Transparent proxying support.
  57  *              David S. Miller :       New socket lookup architecture.
  58  *                                      Some other random speedups.
  59  *              Cyrus Durgin    :       Cleaned up file for kmod hacks.
  60  *              Andi Kleen      :       Fix inet_stream_connect TCP race.
  61  *
  62  *              This program is free software; you can redistribute it and/or
  63  *              modify it under the terms of the GNU General Public License
  64  *              as published by the Free Software Foundation; either version
  65  *              2 of the License, or (at your option) any later version.
  66  */
  67
  68 #include <linux/err.h>
  69 #include <linux/errno.h>
  70 #include <linux/types.h>
  71 #include <linux/socket.h>
  72 #include <linux/in.h>
  73 #include <linux/kernel.h>
  74 #include <linux/module.h>
  75 #include <linux/sched.h>
  76 #include <linux/timer.h>
  77 #include <linux/string.h>
  78 #include <linux/sockios.h>
  79 #include <linux/net.h>
  80 #include <linux/capability.h>
  81 #include <linux/fcntl.h>
  82 #include <linux/mm.h>
  83 #include <linux/interrupt.h>
  84 #include <linux/stat.h>
  85 #include <linux/init.h>
  86 #include <linux/poll.h>
  87 #include <linux/netfilter_ipv4.h>
  88 #include <linux/random.h>
  89 #include <linux/slab.h>
  90
  91 #include <asm/uaccess.h>
  92 #include <asm/system.h>
  93
  94 #include <linux/inet.h>
  95 #include <linux/igmp.h>
  96 #include <linux/inetdevice.h>
  97 #include <linux/netdevice.h>
  98 #include <net/checksum.h>
  99 #include <net/ip.h>
 100 #include <net/protocol.h>
 101 #include <net/arp.h>
 102 #include <net/route.h>
 103 #include <net/ip_fib.h>
 104 #include <net/inet_connection_sock.h>
 105 #include <net/tcp.h>
 106 #include <net/udp.h>
 107 #include <net/udplite.h>
 108 #include <linux/skbuff.h>
 109 #include <net/sock.h>
 110 #include <net/raw.h>
 111 #include <net/icmp.h>
 112 #include <net/ipip.h>
 113 #include <net/inet_common.h>
 114 #include <net/xfrm.h>
 115 #include <net/net_namespace.h>
 116 #ifdef CONFIG_IP_MROUTE
 117 #include <linux/mroute.h>
 118 #endif
 119
 120
 121 /* The inetsw table contains everything that inet_create needs to
 122  * build a new socket.
 123  */
 124 static struct list_head inetsw[SOCK_MAX];
 125 static DEFINE_SPINLOCK(inetsw_lock);
 126
 127 struct ipv4_config ipv4_config;
 128 EXPORT_SYMBOL(ipv4_config);
 129
 130 /* New destruction routine */
 131
 132 void inet_sock_destruct(struct sock *sk)
 133 {
 134         struct inet_sock *inet = inet_sk(sk);
 135
 136         __skb_queue_purge(&sk->sk_receive_queue);
 137         __skb_queue_purge(&sk->sk_error_queue);
 138
 139         sk_mem_reclaim(sk);
 140
 141         if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
 142                 pr_err("Attempt to release TCP socket in state %d %p\n",
 143                        sk->sk_state, sk);
 144                 return;
 145         }
 146         if (!sock_flag(sk, SOCK_DEAD)) {
 147                 pr_err("Attempt to release alive inet socket %p\n", sk);
 148                 return;
 149         }
 150
 151         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
 152         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
 153         WARN_ON(sk->sk_wmem_queued);
 154         WARN_ON(sk->sk_forward_alloc);
 155
 156         kfree(inet->opt);
 157         dst_release(sk->sk_dst_cache);
 158         sk_refcnt_debug_dec(sk);
 159 }
 160 EXPORT_SYMBOL(inet_sock_destruct);
 161
 162 /*
 163  *      The routines beyond this point handle the behaviour of an AF_INET
 164  *      socket object. Mostly it punts to the subprotocols of IP to do
 165  *      the work.
 166  */
 167
 168 /*
 169  *      Automatically bind an unbound socket.
 170  */
 171
 172 static int inet_autobind(struct sock *sk)
 173 {
 174         struct inet_sock *inet;
 175         /* We may need to bind the socket. */
 176         lock_sock(sk);
 177         inet = inet_sk(sk);
 178         if (!inet->inet_num) {
 179                 if (sk->sk_prot->get_port(sk, 0)) {
 180                         release_sock(sk);
 181                         return -EAGAIN;
 182                 }
 183                 inet->inet_sport = htons(inet->inet_num);
 184         }
 185         release_sock(sk);
 186         return 0;
 187 }
 188
 189 /*
 190  *      Move a socket into listening state.
 191  */
 192 int inet_listen(struct socket *sock, int backlog)
 193 {
 194         struct sock *sk = sock->sk;
 195         unsigned char old_state;
 196         int err;
 197
 198         lock_sock(sk);
 199
 200         err = -EINVAL;
 201         if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
 202                 goto out;
 203
 204         old_state = sk->sk_state;
 205         if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
 206                 goto out;
 207
 208         /* Really, if the socket is already in listen state
 209          * we can only allow the backlog to be adjusted.
 210          */
 211         if (old_state != TCP_LISTEN) {
 212                 err = inet_csk_listen_start(sk, backlog);
 213                 if (err)
 214                         goto out;
 215         }
 216         sk->sk_max_ack_backlog = backlog;
 217         err = 0;
 218
 219 out:
 220         release_sock(sk);
 221         return err;
 222 }
 223 EXPORT_SYMBOL(inet_listen);
 224
 225 u32 inet_ehash_secret __read_mostly;
 226 EXPORT_SYMBOL(inet_ehash_secret);
 227
 228 /*
 229  * inet_ehash_secret must be set exactly once
 230  * Instead of using a dedicated spinlock, we (ab)use inetsw_lock
 231  */
 232 void build_ehash_secret(void)
 233 {
 234         u32 rnd;
 235         do {
 236                 get_random_bytes(&rnd, sizeof(rnd));
 237         } while (rnd == 0);
 238         spin_lock_bh(&inetsw_lock);
 239         if (!inet_ehash_secret)
 240                 inet_ehash_secret = rnd;
 241         spin_unlock_bh(&inetsw_lock);
 242 }
 243 EXPORT_SYMBOL(build_ehash_secret);
 244
 245 static inline int inet_netns_ok(struct net *net, int protocol)
 246 {
 247         int hash;
 248         const struct net_protocol *ipprot;
 249
 250         if (net_eq(net, &init_net))
 251                 return 1;
 252
 253         hash = protocol & (MAX_INET_PROTOS - 1);
 254         ipprot = rcu_dereference(inet_protos[hash]);
 255
 256         if (ipprot == NULL)
 257                 /* raw IP is OK */
 258                 return 1;
 259         return ipprot->netns_ok;
 260 }
 261
 262 /*
 263  *      Create an inet socket.
 264  */
 265
 266 static int inet_create(struct net *net, struct socket *sock, int protocol,
 267                        int kern)
 268 {
 269         struct sock *sk;
 270         struct inet_protosw *answer;
 271         struct inet_sock *inet;
 272         struct proto *answer_prot;
 273         unsigned char answer_flags;
 274         char answer_no_check;
 275         int try_loading_module = 0;
 276         int err;
 277
 278         if (unlikely(!inet_ehash_secret))
 279                 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
 280                         build_ehash_secret();
 281
 282         sock->state = SS_UNCONNECTED;
 283
 284         /* Look for the requested type/protocol pair. */
 285 lookup_protocol:
 286         err = -ESOCKTNOSUPPORT;
 287         rcu_read_lock();
 288         list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
 289
 290                 err = 0;
 291                 /* Check the non-wild match. */
 292                 if (protocol == answer->protocol) {
 293                         if (protocol != IPPROTO_IP)
 294                                 break;
 295                 } else {
 296                         /* Check for the two wild cases. */
 297                         if (IPPROTO_IP == protocol) {
 298                                 protocol = answer->protocol;
 299                                 break;
 300                         }
 301                         if (IPPROTO_IP == answer->protocol)
 302                                 break;
 303                 }
 304                 err = -EPROTONOSUPPORT;
 305         }
 306
 307         if (unlikely(err)) {
 308                 if (try_loading_module < 2) {
 309                         rcu_read_unlock();
 310                         /*
 311                          * Be more specific, e.g. net-pf-2-proto-132-type-1
 312                          * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
 313                          */
 314                         if (++try_loading_module == 1)
 315                                 request_module("net-pf-%d-proto-%d-type-%d",
 316                                                PF_INET, protocol, sock->type);
 317                         /*
 318                          * Fall back to generic, e.g. net-pf-2-proto-132
 319                          * (net-pf-PF_INET-proto-IPPROTO_SCTP)
 320                          */
 321                         else
 322                                 request_module("net-pf-%d-proto-%d",
 323                                                PF_INET, protocol);
 324                         goto lookup_protocol;
 325                 } else
 326                         goto out_rcu_unlock;
 327         }
 328
 329         err = -EPERM;
 330         if (sock->type == SOCK_RAW && !kern && !capable(CAP_NET_RAW))
 331                 goto out_rcu_unlock;
 332
 333         err = -EAFNOSUPPORT;
 334         if (!inet_netns_ok(net, protocol))
 335                 goto out_rcu_unlock;
 336
 337         sock->ops = answer->ops;
 338         answer_prot = answer->prot;
 339         answer_no_check = answer->no_check;
 340         answer_flags = answer->flags;
 341         rcu_read_unlock();
 342
 343         WARN_ON(answer_prot->slab == NULL);
 344
 345         err = -ENOBUFS;
 346         sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
 347         if (sk == NULL)
 348                 goto out;
 349
 350         err = 0;
 351         sk->sk_no_check = answer_no_check;
 352         if (INET_PROTOSW_REUSE & answer_flags)
 353                 sk->sk_reuse = 1;
 354
 355         inet = inet_sk(sk);
 356         inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
 357
 358         if (SOCK_RAW == sock->type) {
 359                 inet->inet_num = protocol;
 360                 if (IPPROTO_RAW == protocol)
 361                         inet->hdrincl = 1;
 362         }
 363
 364         if (ipv4_config.no_pmtu_disc)
 365                 inet->pmtudisc = IP_PMTUDISC_DONT;
 366         else
 367                 inet->pmtudisc = IP_PMTUDISC_WANT;
 368
 369         inet->inet_id = 0;
 370
 371         sock_init_data(sock, sk);
 372
 373         sk->sk_destruct    = inet_sock_destruct;
 374         sk->sk_protocol    = protocol;
 375         sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
 376
 377         inet->uc_ttl    = -1;
 378         inet->mc_loop   = 1;
 379         inet->mc_ttl    = 1;
 380         inet->mc_all    = 1;
 381         inet->mc_index  = 0;
 382         inet->mc_list   = NULL;
 383
 384         sk_refcnt_debug_inc(sk);
 385
 386         if (inet->inet_num) {
 387                 /* It assumes that any protocol which allows
 388                  * the user to assign a number at socket
 389                  * creation time automatically
 390                  * shares.
 391                  */
 392                 inet->inet_sport = htons(inet->inet_num);
 393                 /* Add to protocol hash chains. */
 394                 sk->sk_prot->hash(sk);
 395         }
 396
 397         if (sk->sk_prot->init) {
 398                 err = sk->sk_prot->init(sk);
 399                 if (err)
 400                         sk_common_release(sk);
 401         }
 402 out:
 403         return err;
 404 out_rcu_unlock:
 405         rcu_read_unlock();
 406         goto out;
 407 }
 408
 409
 410 /*
 411  *      The peer socket should always be NULL (or else). When we call this
 412  *      function we are destroying the object and from then on nobody
 413  *      should refer to it.
 414  */
 415 int inet_release(struct socket *sock)
 416 {
 417         struct sock *sk = sock->sk;
 418
 419         if (sk) {
 420                 long timeout;
 421
 422                 /* Applications forget to leave groups before exiting */
 423                 ip_mc_drop_socket(sk);
 424
 425                 /* If linger is set, we don't return until the close
 426                  * is complete.  Otherwise we return immediately. The
 427                  * actually closing is done the same either way.
 428                  *
 429                  * If the close is due to the process exiting, we never
 430                  * linger..
 431                  */
 432                 timeout = 0;
 433                 if (sock_flag(sk, SOCK_LINGER) &&
 434                     !(current->flags & PF_EXITING))
 435                         timeout = sk->sk_lingertime;
 436                 sock->sk = NULL;
 437                 sk->sk_prot->close(sk, timeout);
 438         }
 439         return 0;
 440 }
 441 EXPORT_SYMBOL(inet_release);
 442
 443 /* It is off by default, see below. */
 444 int sysctl_ip_nonlocal_bind __read_mostly;
 445 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
 446
 447 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 448 {
 449         struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
 450         struct sock *sk = sock->sk;
 451         struct inet_sock *inet = inet_sk(sk);
 452         unsigned short snum;
 453         int chk_addr_ret;
 454         int err;
 455
 456         /* If the socket has its own bind function then use it. (RAW) */
 457         if (sk->sk_prot->bind) {
 458                 err = sk->sk_prot->bind(sk, uaddr, addr_len);
 459                 goto out;
 460         }
 461         err = -EINVAL;
 462         if (addr_len < sizeof(struct sockaddr_in))
 463                 goto out;
 464
 465         chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr);
 466
 467         /* Not specified by any standard per-se, however it breaks too
 468          * many applications when removed.  It is unfortunate since
 469          * allowing applications to make a non-local bind solves
 470          * several problems with systems using dynamic addressing.
 471          * (ie. your servers still start up even if your ISDN link
 472          *  is temporarily down)
 473          */
 474         err = -EADDRNOTAVAIL;
 475         if (!sysctl_ip_nonlocal_bind &&
 476             !(inet->freebind || inet->transparent) &&
 477             addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
 478             chk_addr_ret != RTN_LOCAL &&
 479             chk_addr_ret != RTN_MULTICAST &&
 480             chk_addr_ret != RTN_BROADCAST)
 481                 goto out;
 482
 483         snum = ntohs(addr->sin_port);
 484         err = -EACCES;
 485         if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
 486                 goto out;
 487
 488         /*      We keep a pair of addresses. rcv_saddr is the one
 489          *      used by hash lookups, and saddr is used for transmit.
 490          *
 491          *      In the BSD API these are the same except where it
 492          *      would be illegal to use them (multicast/broadcast) in
 493          *      which case the sending device address is used.
 494          */
 495         lock_sock(sk);
 496
 497         /* Check these errors (active socket, double bind). */
 498         err = -EINVAL;
 499         if (sk->sk_state != TCP_CLOSE || inet->inet_num)
 500                 goto out_release_sock;
 501
 502         inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
 503         if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
 504                 inet->inet_saddr = 0;  /* Use device */
 505
 506         /* Make sure we are allowed to bind here. */
 507         if (sk->sk_prot->get_port(sk, snum)) {
 508                 inet->inet_saddr = inet->inet_rcv_saddr = 0;
 509                 err = -EADDRINUSE;
 510                 goto out_release_sock;
 511         }
 512
 513         if (inet->inet_rcv_saddr)
 514                 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
 515         if (snum)
 516                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
 517         inet->inet_sport = htons(inet->inet_num);
 518         inet->inet_daddr = 0;
 519         inet->inet_dport = 0;
 520         sk_dst_reset(sk);
 521         err = 0;
 522 out_release_sock:
 523         release_sock(sk);
 524 out:
 525         return err;
 526 }
 527 EXPORT_SYMBOL(inet_bind);
 528
 529 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
 530                        int addr_len, int flags)
 531 {
 532         struct sock *sk = sock->sk;
 533
 534         if (uaddr->sa_family == AF_UNSPEC)
 535                 return sk->sk_prot->disconnect(sk, flags);
 536
 537         if (!inet_sk(sk)->inet_num && inet_autobind(sk))
 538                 return -EAGAIN;
 539         return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
 540 }
 541 EXPORT_SYMBOL(inet_dgram_connect);
 542
 543 static long inet_wait_for_connect(struct sock *sk, long timeo)
 544 {
 545         DEFINE_WAIT(wait);
 546
 547         prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 548
 549         /* Basic assumption: if someone sets sk->sk_err, he _must_
 550          * change state of the socket from TCP_SYN_*.
 551          * Connect() does not allow to get error notifications
 552          * without closing the socket.
 553          */
 554         while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 555                 release_sock(sk);
 556                 timeo = schedule_timeout(timeo);
 557                 lock_sock(sk);
 558                 if (signal_pending(current) || !timeo)
 559                         break;
 560                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 561         }
 562         finish_wait(sk->sk_sleep, &wait);
 563         return timeo;
 564 }
 565
 566 /*
 567  *      Connect to a remote host. There is regrettably still a little
 568  *      TCP 'magic' in here.
 569  */
 570 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 571                         int addr_len, int flags)
 572 {
 573         struct sock *sk = sock->sk;
 574         int err;
 575         long timeo;
 576
 577         lock_sock(sk);
 578
 579         if (uaddr->sa_family == AF_UNSPEC) {
 580                 err = sk->sk_prot->disconnect(sk, flags);
 581                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 582                 goto out;
 583         }
 584
 585         switch (sock->state) {
 586         default:
 587                 err = -EINVAL;
 588                 goto out;
 589         case SS_CONNECTED:
 590                 err = -EISCONN;
 591                 goto out;
 592         case SS_CONNECTING:
 593                 err = -EALREADY;
 594                 /* Fall out of switch with err, set for this state */
 595                 break;
 596         case SS_UNCONNECTED:
 597                 err = -EISCONN;
 598                 if (sk->sk_state != TCP_CLOSE)
 599                         goto out;
 600
 601                 err = sk->sk_prot->connect(sk, uaddr, addr_len);
 602                 if (err < 0)
 603                         goto out;
 604
 605                 sock->state = SS_CONNECTING;
 606
 607                 /* Just entered SS_CONNECTING state; the only
 608                  * difference is that return value in non-blocking
 609                  * case is EINPROGRESS, rather than EALREADY.
 610                  */
 611                 err = -EINPROGRESS;
 612                 break;
 613         }
 614
 615         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
 616
 617         if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 618                 /* Error code is set above */
 619                 if (!timeo || !inet_wait_for_connect(sk, timeo))
 620                         goto out;
 621
 622                 err = sock_intr_errno(timeo);
 623                 if (signal_pending(current))
 624                         goto out;
 625         }
 626
 627         /* Connection was closed by RST, timeout, ICMP error
 628          * or another process disconnected us.
 629          */
 630         if (sk->sk_state == TCP_CLOSE)
 631                 goto sock_error;
 632
 633         /* sk->sk_err may be not zero now, if RECVERR was ordered by user
 634          * and error was received after socket entered established state.
 635          * Hence, it is handled normally after connect() return successfully.
 636          */
 637
 638         sock->state = SS_CONNECTED;
 639         err = 0;
 640 out:
 641         release_sock(sk);
 642         return err;
 643
 644 sock_error:
 645         err = sock_error(sk) ? : -ECONNABORTED;
 646         sock->state = SS_UNCONNECTED;
 647         if (sk->sk_prot->disconnect(sk, flags))
 648                 sock->state = SS_DISCONNECTING;
 649         goto out;
 650 }
 651 EXPORT_SYMBOL(inet_stream_connect);
 652
 653 /*
 654  *      Accept a pending connection. The TCP layer now gives BSD semantics.
 655  */
 656
 657 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
 658 {
 659         struct sock *sk1 = sock->sk;
 660         int err = -EINVAL;
 661         struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
 662
 663         if (!sk2)
 664                 goto do_err;
 665
 666         lock_sock(sk2);
 667
 668         WARN_ON(!((1 << sk2->sk_state) &
 669                   (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE)));
 670
 671         sock_graft(sk2, newsock);
 672
 673         newsock->state = SS_CONNECTED;
 674         err = 0;
 675         release_sock(sk2);
 676 do_err:
 677         return err;
 678 }
 679 EXPORT_SYMBOL(inet_accept);
 680
 681
 682 /*
 683  *      This does both peername and sockname.
 684  */
 685 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
 686                         int *uaddr_len, int peer)
 687 {
 688         struct sock *sk         = sock->sk;
 689         struct inet_sock *inet  = inet_sk(sk);
 690         DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
 691
 692         sin->sin_family = AF_INET;
 693         if (peer) {
 694                 if (!inet->inet_dport ||
 695                     (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
 696                      peer == 1))
 697                         return -ENOTCONN;
 698                 sin->sin_port = inet->inet_dport;
 699                 sin->sin_addr.s_addr = inet->inet_daddr;
 700         } else {
 701                 __be32 addr = inet->inet_rcv_saddr;
 702                 if (!addr)
 703                         addr = inet->inet_saddr;
 704                 sin->sin_port = inet->inet_sport;
 705                 sin->sin_addr.s_addr = addr;
 706         }
 707         memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 708         *uaddr_len = sizeof(*sin);
 709         return 0;
 710 }
 711 EXPORT_SYMBOL(inet_getname);
 712
 713 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
 714                  size_t size)
 715 {
 716         struct sock *sk = sock->sk;
 717
 718         /* We may need to bind the socket. */
 719         if (!inet_sk(sk)->inet_num && inet_autobind(sk))
 720                 return -EAGAIN;
 721
 722         return sk->sk_prot->sendmsg(iocb, sk, msg, size);
 723 }
 724 EXPORT_SYMBOL(inet_sendmsg);
 725
 726
 727 static ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
 728                              size_t size, int flags)
 729 {
 730         struct sock *sk = sock->sk;
 731
 732         /* We may need to bind the socket. */
 733         if (!inet_sk(sk)->inet_num && inet_autobind(sk))
 734                 return -EAGAIN;
 735
 736         if (sk->sk_prot->sendpage)
 737                 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
 738         return sock_no_sendpage(sock, page, offset, size, flags);
 739 }
 740
 741
 742 int inet_shutdown(struct socket *sock, int how)
 743 {
 744         struct sock *sk = sock->sk;
 745         int err = 0;
 746
 747         /* This should really check to make sure
 748          * the socket is a TCP socket. (WHY AC...)
 749          */
 750         how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
 751                        1->2 bit 2 snds.
 752                        2->3 */
 753         if ((how & ~SHUTDOWN_MASK) || !how)     /* MAXINT->0 */
 754                 return -EINVAL;
 755
 756         lock_sock(sk);
 757         if (sock->state == SS_CONNECTING) {
 758                 if ((1 << sk->sk_state) &
 759                     (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
 760                         sock->state = SS_DISCONNECTING;
 761                 else
 762                         sock->state = SS_CONNECTED;
 763         }
 764
 765         switch (sk->sk_state) {
 766         case TCP_CLOSE:
 767                 err = -ENOTCONN;
 768                 /* Hack to wake up other listeners, who can poll for
 769                    POLLHUP, even on eg. unconnected UDP sockets -- RR */
 770         default:
 771                 sk->sk_shutdown |= how;
 772                 if (sk->sk_prot->shutdown)
 773                         sk->sk_prot->shutdown(sk, how);
 774                 break;
 775
 776         /* Remaining two branches are temporary solution for missing
 777          * close() in multithreaded environment. It is _not_ a good idea,
 778          * but we have no choice until close() is repaired at VFS level.
 779          */
 780         case TCP_LISTEN:
 781                 if (!(how & RCV_SHUTDOWN))
 782                         break;
 783                 /* Fall through */
 784         case TCP_SYN_SENT:
 785                 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
 786                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 787                 break;
 788         }
 789
 790         /* Wake up anyone sleeping in poll. */
 791         sk->sk_state_change(sk);
 792         release_sock(sk);
 793         return err;
 794 }
 795 EXPORT_SYMBOL(inet_shutdown);
 796
 797 /*
 798  *      ioctl() calls you can issue on an INET socket. Most of these are
 799  *      device configuration and stuff and very rarely used. Some ioctls
 800  *      pass on to the socket itself.
 801  *
 802  *      NOTE: I like the idea of a module for the config stuff. ie ifconfig
 803  *      loads the devconfigure module does its configuring and unloads it.
 804  *      There's a good 20K of config code hanging around the kernel.
 805  */
 806
 807 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 808 {
 809         struct sock *sk = sock->sk;
 810         int err = 0;
 811         struct net *net = sock_net(sk);
 812
 813         switch (cmd) {
 814         case SIOCGSTAMP:
 815                 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
 816                 break;
 817         case SIOCGSTAMPNS:
 818                 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
 819                 break;
 820         case SIOCADDRT:
 821         case SIOCDELRT:
 822         case SIOCRTMSG:
 823                 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
 824                 break;
 825         case SIOCDARP:
 826         case SIOCGARP:
 827         case SIOCSARP:
 828                 err = arp_ioctl(net, cmd, (void __user *)arg);
 829                 break;
 830         case SIOCGIFADDR:
 831         case SIOCSIFADDR:
 832         case SIOCGIFBRDADDR:
 833         case SIOCSIFBRDADDR:
 834         case SIOCGIFNETMASK:
 835         case SIOCSIFNETMASK:
 836         case SIOCGIFDSTADDR:
 837         case SIOCSIFDSTADDR:
 838         case SIOCSIFPFLAGS:
 839         case SIOCGIFPFLAGS:
 840         case SIOCSIFFLAGS:
 841                 err = devinet_ioctl(net, cmd, (void __user *)arg);
 842                 break;
 843         default:
 844                 if (sk->sk_prot->ioctl)
 845                         err = sk->sk_prot->ioctl(sk, cmd, arg);
 846                 else
 847                         err = -ENOIOCTLCMD;
 848                 break;
 849         }
 850         return err;
 851 }
 852 EXPORT_SYMBOL(inet_ioctl);
 853
 854 const struct proto_ops inet_stream_ops = {
 855         .family            = PF_INET,
 856         .owner             = THIS_MODULE,
 857         .release           = inet_release,
 858         .bind              = inet_bind,
 859         .connect           = inet_stream_connect,
 860         .socketpair        = sock_no_socketpair,
 861         .accept            = inet_accept,
 862         .getname           = inet_getname,
 863         .poll              = tcp_poll,
 864         .ioctl             = inet_ioctl,
 865         .listen            = inet_listen,
 866         .shutdown          = inet_shutdown,
 867         .setsockopt        = sock_common_setsockopt,
 868         .getsockopt        = sock_common_getsockopt,
 869         .sendmsg           = tcp_sendmsg,
 870         .recvmsg           = sock_common_recvmsg,
 871         .mmap              = sock_no_mmap,
 872         .sendpage          = tcp_sendpage,
 873         .splice_read       = tcp_splice_read,
 874 #ifdef CONFIG_COMPAT
 875         .compat_setsockopt = compat_sock_common_setsockopt,
 876         .compat_getsockopt = compat_sock_common_getsockopt,
 877 #endif
 878 };
 879 EXPORT_SYMBOL(inet_stream_ops);
 880
 881 const struct proto_ops inet_dgram_ops = {
 882         .family            = PF_INET,
 883         .owner             = THIS_MODULE,
 884         .release           = inet_release,
 885         .bind              = inet_bind,
 886         .connect           = inet_dgram_connect,
 887         .socketpair        = sock_no_socketpair,
 888         .accept            = sock_no_accept,
 889         .getname           = inet_getname,
 890         .poll              = udp_poll,
 891         .ioctl             = inet_ioctl,
 892         .listen            = sock_no_listen,
 893         .shutdown          = inet_shutdown,
 894         .setsockopt        = sock_common_setsockopt,
 895         .getsockopt        = sock_common_getsockopt,
 896         .sendmsg           = inet_sendmsg,
 897         .recvmsg           = sock_common_recvmsg,
 898         .mmap              = sock_no_mmap,
 899         .sendpage          = inet_sendpage,
 900 #ifdef CONFIG_COMPAT
 901         .compat_setsockopt = compat_sock_common_setsockopt,
 902         .compat_getsockopt = compat_sock_common_getsockopt,
 903 #endif
 904 };
 905 EXPORT_SYMBOL(inet_dgram_ops);
 906
 907 /*
 908  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
 909  * udp_poll
 910  */
 911 static const struct proto_ops inet_sockraw_ops = {
 912         .family            = PF_INET,
 913         .owner             = THIS_MODULE,
 914         .release           = inet_release,
 915         .bind              = inet_bind,
 916         .connect           = inet_dgram_connect,
 917         .socketpair        = sock_no_socketpair,
 918         .accept            = sock_no_accept,
 919         .getname           = inet_getname,
 920         .poll              = datagram_poll,
 921         .ioctl             = inet_ioctl,
 922         .listen            = sock_no_listen,
 923         .shutdown          = inet_shutdown,
 924         .setsockopt        = sock_common_setsockopt,
 925         .getsockopt        = sock_common_getsockopt,
 926         .sendmsg           = inet_sendmsg,
 927         .recvmsg           = sock_common_recvmsg,
 928         .mmap              = sock_no_mmap,
 929         .sendpage          = inet_sendpage,
 930 #ifdef CONFIG_COMPAT
 931         .compat_setsockopt = compat_sock_common_setsockopt,
 932         .compat_getsockopt = compat_sock_common_getsockopt,
 933 #endif
 934 };
 935
 936 static const struct net_proto_family inet_family_ops = {
 937         .family = PF_INET,
 938         .create = inet_create,
 939         .owner  = THIS_MODULE,
 940 };
 941
 942 /* Upon startup we insert all the elements in inetsw_array[] into
 943  * the linked list inetsw.
 944  */
 945 static struct inet_protosw inetsw_array[] =
 946 {
 947         {
 948                 .type =       SOCK_STREAM,
 949                 .protocol =   IPPROTO_TCP,
 950                 .prot =       &tcp_prot,
 951                 .ops =        &inet_stream_ops,
 952                 .no_check =   0,
 953                 .flags =      INET_PROTOSW_PERMANENT |
 954                               INET_PROTOSW_ICSK,
 955         },
 956
 957         {
 958                 .type =       SOCK_DGRAM,
 959                 .protocol =   IPPROTO_UDP,
 960                 .prot =       &udp_prot,
 961                 .ops =        &inet_dgram_ops,
 962                 .no_check =   UDP_CSUM_DEFAULT,
 963                 .flags =      INET_PROTOSW_PERMANENT,
 964        },
 965
 966
 967        {
 968                .type =       SOCK_RAW,
 969                .protocol =   IPPROTO_IP,        /* wild card */
 970                .prot =       &raw_prot,
 971                .ops =        &inet_sockraw_ops,
 972                .no_check =   UDP_CSUM_DEFAULT,
 973                .flags =      INET_PROTOSW_REUSE,
 974        }
 975 };
 976
 977 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
 978
 979 void inet_register_protosw(struct inet_protosw *p)
 980 {
 981         struct list_head *lh;
 982         struct inet_protosw *answer;
 983         int protocol = p->protocol;
 984         struct list_head *last_perm;
 985
 986         spin_lock_bh(&inetsw_lock);
 987
 988         if (p->type >= SOCK_MAX)
 989                 goto out_illegal;
 990
 991         /* If we are trying to override a permanent protocol, bail. */
 992         answer = NULL;
 993         last_perm = &inetsw[p->type];
 994         list_for_each(lh, &inetsw[p->type]) {
 995                 answer = list_entry(lh, struct inet_protosw, list);
 996
 997                 /* Check only the non-wild match. */
 998                 if (INET_PROTOSW_PERMANENT & answer->flags) {
 999                         if (protocol == answer->protocol)
1000                                 break;
1001                         last_perm = lh;
1002                 }
1003
1004                 answer = NULL;
1005         }
1006         if (answer)
1007                 goto out_permanent;
1008
1009         /* Add the new entry after the last permanent entry if any, so that
1010          * the new entry does not override a permanent entry when matched with
1011          * a wild-card protocol. But it is allowed to override any existing
1012          * non-permanent entry.  This means that when we remove this entry, the
1013          * system automatically returns to the old behavior.
1014          */
1015         list_add_rcu(&p->list, last_perm);
1016 out:
1017         spin_unlock_bh(&inetsw_lock);
1018
1019         return;
1020
1021 out_permanent:
1022         printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
1023                protocol);
1024         goto out;
1025
1026 out_illegal:
1027         printk(KERN_ERR
1028                "Ignoring attempt to register invalid socket type %d.\n",
1029                p->type);
1030         goto out;
1031 }
1032 EXPORT_SYMBOL(inet_register_protosw);
1033
1034 void inet_unregister_protosw(struct inet_protosw *p)
1035 {
1036         if (INET_PROTOSW_PERMANENT & p->flags) {
1037                 printk(KERN_ERR
1038                        "Attempt to unregister permanent protocol %d.\n",
1039                        p->protocol);
1040         } else {
1041                 spin_lock_bh(&inetsw_lock);
1042                 list_del_rcu(&p->list);
1043                 spin_unlock_bh(&inetsw_lock);
1044
1045                 synchronize_net();
1046         }
1047 }
1048 EXPORT_SYMBOL(inet_unregister_protosw);
1049
1050 /*
1051  *      Shall we try to damage output packets if routing dev changes?
1052  */
1053
1054 int sysctl_ip_dynaddr __read_mostly;
1055
1056 static int inet_sk_reselect_saddr(struct sock *sk)
1057 {
1058         struct inet_sock *inet = inet_sk(sk);
1059         int err;
1060         struct rtable *rt;
1061         __be32 old_saddr = inet->inet_saddr;
1062         __be32 new_saddr;
1063         __be32 daddr = inet->inet_daddr;
1064
1065         if (inet->opt && inet->opt->srr)
1066                 daddr = inet->opt->faddr;
1067
1068         /* Query new route. */
1069         err = ip_route_connect(&rt, daddr, 0,
1070                                RT_CONN_FLAGS(sk),
1071                                sk->sk_bound_dev_if,
1072                                sk->sk_protocol,
1073                                inet->inet_sport, inet->inet_dport, sk, 0);
1074         if (err)
1075                 return err;
1076
1077         sk_setup_caps(sk, &rt->u.dst);
1078
1079         new_saddr = rt->rt_src;
1080
1081         if (new_saddr == old_saddr)
1082                 return 0;
1083
1084         if (sysctl_ip_dynaddr > 1) {
1085                 printk(KERN_INFO "%s(): shifting inet->saddr from %pI4 to %pI4\n",
1086                        __func__, &old_saddr, &new_saddr);
1087         }
1088
1089         inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1090
1091         /*
1092          * XXX The only one ugly spot where we need to
1093          * XXX really change the sockets identity after
1094          * XXX it has entered the hashes. -DaveM
1095          *
1096          * Besides that, it does not check for connection
1097          * uniqueness. Wait for troubles.
1098          */
1099         __sk_prot_rehash(sk);
1100         return 0;
1101 }
1102
1103 int inet_sk_rebuild_header(struct sock *sk)
1104 {
1105         struct inet_sock *inet = inet_sk(sk);
1106         struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1107         __be32 daddr;
1108         int err;
1109
1110         /* Route is OK, nothing to do. */
1111         if (rt)
1112                 return 0;
1113
1114         /* Reroute. */
1115         daddr = inet->inet_daddr;
1116         if (inet->opt && inet->opt->srr)
1117                 daddr = inet->opt->faddr;
1118 {
1119         struct flowi fl = {
1120                 .oif = sk->sk_bound_dev_if,
1121                 .mark = sk->sk_mark,
1122                 .nl_u = {
1123                         .ip4_u = {
1124                                 .daddr  = daddr,
1125                                 .saddr  = inet->inet_saddr,
1126                                 .tos    = RT_CONN_FLAGS(sk),
1127                         },
1128                 },
1129                 .proto = sk->sk_protocol,
1130                 .flags = inet_sk_flowi_flags(sk),
1131                 .uli_u = {
1132                         .ports = {
1133                                 .sport = inet->inet_sport,
1134                                 .dport = inet->inet_dport,
1135                         },
1136                 },
1137         };
1138
1139         security_sk_classify_flow(sk, &fl);
1140         err = ip_route_output_flow(sock_net(sk), &rt, &fl, sk, 0);
1141 }
1142         if (!err)
1143                 sk_setup_caps(sk, &rt->u.dst);
1144         else {
1145                 /* Routing failed... */
1146                 sk->sk_route_caps = 0;
1147                 /*
1148                  * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1149                  * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1150                  */
1151                 if (!sysctl_ip_dynaddr ||
1152                     sk->sk_state != TCP_SYN_SENT ||
1153                     (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1154                     (err = inet_sk_reselect_saddr(sk)) != 0)
1155                         sk->sk_err_soft = -err;
1156         }
1157
1158         return err;
1159 }
1160 EXPORT_SYMBOL(inet_sk_rebuild_header);
1161
1162 static int inet_gso_send_check(struct sk_buff *skb)
1163 {
1164         struct iphdr *iph;
1165         const struct net_protocol *ops;
1166         int proto;
1167         int ihl;
1168         int err = -EINVAL;
1169
1170         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1171                 goto out;
1172
1173         iph = ip_hdr(skb);
1174         ihl = iph->ihl * 4;
1175         if (ihl < sizeof(*iph))
1176                 goto out;
1177
1178         if (unlikely(!pskb_may_pull(skb, ihl)))
1179                 goto out;
1180
1181         __skb_pull(skb, ihl);
1182         skb_reset_transport_header(skb);
1183         iph = ip_hdr(skb);
1184         proto = iph->protocol & (MAX_INET_PROTOS - 1);
1185         err = -EPROTONOSUPPORT;
1186
1187         rcu_read_lock();
1188         ops = rcu_dereference(inet_protos[proto]);
1189         if (likely(ops && ops->gso_send_check))
1190                 err = ops->gso_send_check(skb);
1191         rcu_read_unlock();
1192
1193 out:
1194         return err;
1195 }
1196
1197 static struct sk_buff *inet_gso_segment(struct sk_buff *skb, int features)
1198 {
1199         struct sk_buff *segs = ERR_PTR(-EINVAL);
1200         struct iphdr *iph;
1201         const struct net_protocol *ops;
1202         int proto;
1203         int ihl;
1204         int id;
1205         unsigned int offset = 0;
1206
1207         if (!(features & NETIF_F_V4_CSUM))
1208                 features &= ~NETIF_F_SG;
1209
1210         if (unlikely(skb_shinfo(skb)->gso_type &
1211                      ~(SKB_GSO_TCPV4 |
1212                        SKB_GSO_UDP |
1213                        SKB_GSO_DODGY |
1214                        SKB_GSO_TCP_ECN |
1215                        0)))
1216                 goto out;
1217
1218         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1219                 goto out;
1220
1221         iph = ip_hdr(skb);
1222         ihl = iph->ihl * 4;
1223         if (ihl < sizeof(*iph))
1224                 goto out;
1225
1226         if (unlikely(!pskb_may_pull(skb, ihl)))
1227                 goto out;
1228
1229         __skb_pull(skb, ihl);
1230         skb_reset_transport_header(skb);
1231         iph = ip_hdr(skb);
1232         id = ntohs(iph->id);
1233         proto = iph->protocol & (MAX_INET_PROTOS - 1);
1234         segs = ERR_PTR(-EPROTONOSUPPORT);
1235
1236         rcu_read_lock();
1237         ops = rcu_dereference(inet_protos[proto]);
1238         if (likely(ops && ops->gso_segment))
1239                 segs = ops->gso_segment(skb, features);
1240         rcu_read_unlock();
1241
1242         if (!segs || IS_ERR(segs))
1243                 goto out;
1244
1245         skb = segs;
1246         do {
1247                 iph = ip_hdr(skb);
1248                 if (proto == IPPROTO_UDP) {
1249                         iph->id = htons(id);
1250                         iph->frag_off = htons(offset >> 3);
1251                         if (skb->next != NULL)
1252                                 iph->frag_off |= htons(IP_MF);
1253                         offset += (skb->len - skb->mac_len - iph->ihl * 4);
1254                 } else
1255                         iph->id = htons(id++);
1256                 iph->tot_len = htons(skb->len - skb->mac_len);
1257                 iph->check = 0;
1258                 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1259         } while ((skb = skb->next));
1260
1261 out:
1262         return segs;
1263 }
1264
1265 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1266                                          struct sk_buff *skb)
1267 {
1268         const struct net_protocol *ops;
1269         struct sk_buff **pp = NULL;
1270         struct sk_buff *p;
1271         struct iphdr *iph;
1272         unsigned int hlen;
1273         unsigned int off;
1274         unsigned int id;
1275         int flush = 1;
1276         int proto;
1277
1278         off = skb_gro_offset(skb);
1279         hlen = off + sizeof(*iph);
1280         iph = skb_gro_header_fast(skb, off);
1281         if (skb_gro_header_hard(skb, hlen)) {
1282                 iph = skb_gro_header_slow(skb, hlen, off);
1283                 if (unlikely(!iph))
1284                         goto out;
1285         }
1286
1287         proto = iph->protocol & (MAX_INET_PROTOS - 1);
1288
1289         rcu_read_lock();
1290         ops = rcu_dereference(inet_protos[proto]);
1291         if (!ops || !ops->gro_receive)
1292                 goto out_unlock;
1293
1294         if (*(u8 *)iph != 0x45)
1295                 goto out_unlock;
1296
1297         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1298                 goto out_unlock;
1299
1300         id = ntohl(*(u32 *)&iph->id);
1301         flush = (u16)((ntohl(*(u32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF));
1302         id >>= 16;
1303
1304         for (p = *head; p; p = p->next) {
1305                 struct iphdr *iph2;
1306
1307                 if (!NAPI_GRO_CB(p)->same_flow)
1308                         continue;
1309
1310                 iph2 = ip_hdr(p);
1311
1312                 if ((iph->protocol ^ iph2->protocol) |
1313                     (iph->tos ^ iph2->tos) |
1314                     (iph->saddr ^ iph2->saddr) |
1315                     (iph->daddr ^ iph2->daddr)) {
1316                         NAPI_GRO_CB(p)->same_flow = 0;
1317                         continue;
1318                 }
1319
1320                 /* All fields must match except length and checksum. */
1321                 NAPI_GRO_CB(p)->flush |=
1322                         (iph->ttl ^ iph2->ttl) |
1323                         ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1324
1325                 NAPI_GRO_CB(p)->flush |= flush;
1326         }
1327
1328         NAPI_GRO_CB(skb)->flush |= flush;
1329         skb_gro_pull(skb, sizeof(*iph));
1330         skb_set_transport_header(skb, skb_gro_offset(skb));
1331
1332         pp = ops->gro_receive(head, skb);
1333
1334 out_unlock:
1335         rcu_read_unlock();
1336
1337 out:
1338         NAPI_GRO_CB(skb)->flush |= flush;
1339
1340         return pp;
1341 }
1342
1343 static int inet_gro_complete(struct sk_buff *skb)
1344 {
1345         const struct net_protocol *ops;
1346         struct iphdr *iph = ip_hdr(skb);
1347         int proto = iph->protocol & (MAX_INET_PROTOS - 1);
1348         int err = -ENOSYS;
1349         __be16 newlen = htons(skb->len - skb_network_offset(skb));
1350
1351         csum_replace2(&iph->check, iph->tot_len, newlen);
1352         iph->tot_len = newlen;
1353
1354         rcu_read_lock();
1355         ops = rcu_dereference(inet_protos[proto]);
1356         if (WARN_ON(!ops || !ops->gro_complete))
1357                 goto out_unlock;
1358
1359         err = ops->gro_complete(skb);
1360
1361 out_unlock:
1362         rcu_read_unlock();
1363
1364         return err;
1365 }
1366
1367 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1368                          unsigned short type, unsigned char protocol,
1369                          struct net *net)
1370 {
1371         struct socket *sock;
1372         int rc = sock_create_kern(family, type, protocol, &sock);
1373
1374         if (rc == 0) {
1375                 *sk = sock->sk;
1376                 (*sk)->sk_allocation = GFP_ATOMIC;
1377                 /*
1378                  * Unhash it so that IP input processing does not even see it,
1379                  * we do not wish this socket to see incoming packets.
1380                  */
1381                 (*sk)->sk_prot->unhash(*sk);
1382
1383                 sk_change_net(*sk, net);
1384         }
1385         return rc;
1386 }
1387 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1388
1389 unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1390 {
1391         unsigned long res = 0;
1392         int i;
1393
1394         for_each_possible_cpu(i) {
1395                 res += *(((unsigned long *) per_cpu_ptr(mib[0], i)) + offt);
1396                 res += *(((unsigned long *) per_cpu_ptr(mib[1], i)) + offt);
1397         }
1398         return res;
1399 }
1400 EXPORT_SYMBOL_GPL(snmp_fold_field);
1401
1402 int snmp_mib_init(void __percpu *ptr[2], size_t mibsize)
1403 {
1404         BUG_ON(ptr == NULL);
1405         ptr[0] = __alloc_percpu(mibsize, __alignof__(unsigned long long));
1406         if (!ptr[0])
1407                 goto err0;
1408         ptr[1] = __alloc_percpu(mibsize, __alignof__(unsigned long long));
1409         if (!ptr[1])
1410                 goto err1;
1411         return 0;
1412 err1:
1413         free_percpu(ptr[0]);
1414         ptr[0] = NULL;
1415 err0:
1416         return -ENOMEM;
1417 }
1418 EXPORT_SYMBOL_GPL(snmp_mib_init);
1419
1420 void snmp_mib_free(void __percpu *ptr[2])
1421 {
1422         BUG_ON(ptr == NULL);
1423         free_percpu(ptr[0]);
1424         free_percpu(ptr[1]);
1425         ptr[0] = ptr[1] = NULL;
1426 }
1427 EXPORT_SYMBOL_GPL(snmp_mib_free);
1428
1429 #ifdef CONFIG_IP_MULTICAST
1430 static const struct net_protocol igmp_protocol = {
1431         .handler =      igmp_rcv,
1432         .netns_ok =     1,
1433 };
1434 #endif
1435
1436 static const struct net_protocol tcp_protocol = {
1437         .handler =      tcp_v4_rcv,
1438         .err_handler =  tcp_v4_err,
1439         .gso_send_check = tcp_v4_gso_send_check,
1440         .gso_segment =  tcp_tso_segment,
1441         .gro_receive =  tcp4_gro_receive,
1442         .gro_complete = tcp4_gro_complete,
1443         .no_policy =    1,
1444         .netns_ok =     1,
1445 };
1446
1447 static const struct net_protocol udp_protocol = {
1448         .handler =      udp_rcv,
1449         .err_handler =  udp_err,
1450         .gso_send_check = udp4_ufo_send_check,
1451         .gso_segment = udp4_ufo_fragment,
1452         .no_policy =    1,
1453         .netns_ok =     1,
1454 };
1455
1456 static const struct net_protocol icmp_protocol = {
1457         .handler =      icmp_rcv,
1458         .no_policy =    1,
1459         .netns_ok =     1,
1460 };
1461
1462 static __net_init int ipv4_mib_init_net(struct net *net)
1463 {
1464         if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1465                           sizeof(struct tcp_mib)) < 0)
1466                 goto err_tcp_mib;
1467         if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1468                           sizeof(struct ipstats_mib)) < 0)
1469                 goto err_ip_mib;
1470         if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1471                           sizeof(struct linux_mib)) < 0)
1472                 goto err_net_mib;
1473         if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1474                           sizeof(struct udp_mib)) < 0)
1475                 goto err_udp_mib;
1476         if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1477                           sizeof(struct udp_mib)) < 0)
1478                 goto err_udplite_mib;
1479         if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1480                           sizeof(struct icmp_mib)) < 0)
1481                 goto err_icmp_mib;
1482         if (snmp_mib_init((void __percpu **)net->mib.icmpmsg_statistics,
1483                           sizeof(struct icmpmsg_mib)) < 0)
1484                 goto err_icmpmsg_mib;
1485
1486         tcp_mib_init(net);
1487         return 0;
1488
1489 err_icmpmsg_mib:
1490         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1491 err_icmp_mib:
1492         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1493 err_udplite_mib:
1494         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1495 err_udp_mib:
1496         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1497 err_net_mib:
1498         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1499 err_ip_mib:
1500         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1501 err_tcp_mib:
1502         return -ENOMEM;
1503 }
1504
1505 static __net_exit void ipv4_mib_exit_net(struct net *net)
1506 {
1507         snmp_mib_free((void __percpu **)net->mib.icmpmsg_statistics);
1508         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1509         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1510         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1511         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1512         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1513         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1514 }
1515
1516 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1517         .init = ipv4_mib_init_net,
1518         .exit = ipv4_mib_exit_net,
1519 };
1520
1521 static int __init init_ipv4_mibs(void)
1522 {
1523         return register_pernet_subsys(&ipv4_mib_ops);
1524 }
1525
1526 static int ipv4_proc_init(void);
1527
1528 /*
1529  *      IP protocol layer initialiser
1530  */
1531
1532 static struct packet_type ip_packet_type __read_mostly = {
1533         .type = cpu_to_be16(ETH_P_IP),
1534         .func = ip_rcv,
1535         .gso_send_check = inet_gso_send_check,
1536         .gso_segment = inet_gso_segment,
1537         .gro_receive = inet_gro_receive,
1538         .gro_complete = inet_gro_complete,
1539 };
1540
1541 static int __init inet_init(void)
1542 {
1543         struct sk_buff *dummy_skb;
1544         struct inet_protosw *q;
1545         struct list_head *r;
1546         int rc = -EINVAL;
1547
1548         BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1549
1550         rc = proto_register(&tcp_prot, 1);
1551         if (rc)
1552                 goto out;
1553
1554         rc = proto_register(&udp_prot, 1);
1555         if (rc)
1556                 goto out_unregister_tcp_proto;
1557
1558         rc = proto_register(&raw_prot, 1);
1559         if (rc)
1560                 goto out_unregister_udp_proto;
1561
1562         /*
1563          *      Tell SOCKET that we are alive...
1564          */
1565
1566         (void)sock_register(&inet_family_ops);
1567
1568 #ifdef CONFIG_SYSCTL
1569         ip_static_sysctl_init();
1570 #endif
1571
1572         /*
1573          *      Add all the base protocols.
1574          */
1575
1576         if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1577                 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1578         if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1579                 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1580         if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1581                 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1582 #ifdef CONFIG_IP_MULTICAST
1583         if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1584                 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1585 #endif
1586
1587         /* Register the socket-side information for inet_create. */
1588         for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1589                 INIT_LIST_HEAD(r);
1590
1591         for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1592                 inet_register_protosw(q);
1593
1594         /*
1595          *      Set the ARP module up
1596          */
1597
1598         arp_init();
1599
1600         /*
1601          *      Set the IP module up
1602          */
1603
1604         ip_init();
1605
1606         tcp_v4_init();
1607
1608         /* Setup TCP slab cache for open requests. */
1609         tcp_init();
1610
1611         /* Setup UDP memory threshold */
1612         udp_init();
1613
1614         /* Add UDP-Lite (RFC 3828) */
1615         udplite4_register();
1616
1617         /*
1618          *      Set the ICMP layer up
1619          */
1620
1621         if (icmp_init() < 0)
1622                 panic("Failed to create the ICMP control socket.\n");
1623
1624         /*
1625          *      Initialise the multicast router
1626          */
1627 #if defined(CONFIG_IP_MROUTE)
1628         if (ip_mr_init())
1629                 printk(KERN_CRIT "inet_init: Cannot init ipv4 mroute\n");
1630 #endif
1631         /*
1632          *      Initialise per-cpu ipv4 mibs
1633          */
1634
1635         if (init_ipv4_mibs())
1636                 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n");
1637
1638         ipv4_proc_init();
1639
1640         ipfrag_init();
1641
1642         dev_add_pack(&ip_packet_type);
1643
1644         rc = 0;
1645 out:
1646         return rc;
1647 out_unregister_udp_proto:
1648         proto_unregister(&udp_prot);
1649 out_unregister_tcp_proto:
1650         proto_unregister(&tcp_prot);
1651         goto out;
1652 }
1653
1654 fs_initcall(inet_init);
1655
1656 /* ------------------------------------------------------------------------ */
1657
1658 #ifdef CONFIG_PROC_FS
1659 static int __init ipv4_proc_init(void)
1660 {
1661         int rc = 0;
1662
1663         if (raw_proc_init())
1664                 goto out_raw;
1665         if (tcp4_proc_init())
1666                 goto out_tcp;
1667         if (udp4_proc_init())
1668                 goto out_udp;
1669         if (ip_misc_proc_init())
1670                 goto out_misc;
1671 out:
1672         return rc;
1673 out_misc:
1674         udp4_proc_exit();
1675 out_udp:
1676         tcp4_proc_exit();
1677 out_tcp:
1678         raw_proc_exit();
1679 out_raw:
1680         rc = -ENOMEM;
1681         goto out;
1682 }
1683
1684 #else /* CONFIG_PROC_FS */
1685 static int __init ipv4_proc_init(void)
1686 {
1687         return 0;
1688 }
1689 #endif /* CONFIG_PROC_FS */
1690
1691 MODULE_ALIAS_NETPROTO(PF_INET);
1692