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 <net/ping.h>
 109 #include <linux/skbuff.h>
 110 #include <net/sock.h>
 111 #include <net/raw.h>
 112 #include <net/icmp.h>
 113 #include <net/ipip.h>
 114 #include <net/inet_common.h>
 115 #include <net/xfrm.h>
 116 #include <net/net_namespace.h>
 117 #ifdef CONFIG_IP_MROUTE
 118 #include <linux/mroute.h>
 119 #endif
 120
 121
 122 /* The inetsw table contains everything that inet_create needs to
 123  * build a new socket.
 124  */
 125 static struct list_head inetsw[SOCK_MAX];
 126 static DEFINE_SPINLOCK(inetsw_lock);
 127
 128 struct ipv4_config ipv4_config;
 129 EXPORT_SYMBOL(ipv4_config);
 130
 131 /* New destruction routine */
 132
 133 void inet_sock_destruct(struct sock *sk)
 134 {
 135         struct inet_sock *inet = inet_sk(sk);
 136
 137         __skb_queue_purge(&sk->sk_receive_queue);
 138         __skb_queue_purge(&sk->sk_error_queue);
 139
 140         sk_mem_reclaim(sk);
 141
 142         if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
 143                 pr_err("Attempt to release TCP socket in state %d %p\n",
 144                        sk->sk_state, sk);
 145                 return;
 146         }
 147         if (!sock_flag(sk, SOCK_DEAD)) {
 148                 pr_err("Attempt to release alive inet socket %p\n", sk);
 149                 return;
 150         }
 151
 152         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
 153         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
 154         WARN_ON(sk->sk_wmem_queued);
 155         WARN_ON(sk->sk_forward_alloc);
 156
 157         kfree(rcu_dereference_protected(inet->inet_opt, 1));
 158         dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
 159         sk_refcnt_debug_dec(sk);
 160 }
 161 EXPORT_SYMBOL(inet_sock_destruct);
 162
 163 /*
 164  *      The routines beyond this point handle the behaviour of an AF_INET
 165  *      socket object. Mostly it punts to the subprotocols of IP to do
 166  *      the work.
 167  */
 168
 169 /*
 170  *      Automatically bind an unbound socket.
 171  */
 172
 173 static int inet_autobind(struct sock *sk)
 174 {
 175         struct inet_sock *inet;
 176         /* We may need to bind the socket. */
 177         lock_sock(sk);
 178         inet = inet_sk(sk);
 179         if (!inet->inet_num) {
 180                 if (sk->sk_prot->get_port(sk, 0)) {
 181                         release_sock(sk);
 182                         return -EAGAIN;
 183                 }
 184                 inet->inet_sport = htons(inet->inet_num);
 185         }
 186         release_sock(sk);
 187         return 0;
 188 }
 189
 190 /*
 191  *      Move a socket into listening state.
 192  */
 193 int inet_listen(struct socket *sock, int backlog)
 194 {
 195         struct sock *sk = sock->sk;
 196         unsigned char old_state;
 197         int err;
 198
 199         lock_sock(sk);
 200
 201         err = -EINVAL;
 202         if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
 203                 goto out;
 204
 205         old_state = sk->sk_state;
 206         if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
 207                 goto out;
 208
 209         /* Really, if the socket is already in listen state
 210          * we can only allow the backlog to be adjusted.
 211          */
 212         if (old_state != TCP_LISTEN) {
 213                 err = inet_csk_listen_start(sk, backlog);
 214                 if (err)
 215                         goto out;
 216         }
 217         sk->sk_max_ack_backlog = backlog;
 218         err = 0;
 219
 220 out:
 221         release_sock(sk);
 222         return err;
 223 }
 224 EXPORT_SYMBOL(inet_listen);
 225
 226 u32 inet_ehash_secret __read_mostly;
 227 EXPORT_SYMBOL(inet_ehash_secret);
 228
 229 u32 ipv6_hash_secret __read_mostly;
 230 EXPORT_SYMBOL(ipv6_hash_secret);
 231
 232 /*
 233  * inet_ehash_secret must be set exactly once, and to a non nul value
 234  * ipv6_hash_secret must be set exactly once.
 235  */
 236 void build_ehash_secret(void)
 237 {
 238         u32 rnd;
 239
 240         do {
 241                 get_random_bytes(&rnd, sizeof(rnd));
 242         } while (rnd == 0);
 243
 244         if (cmpxchg(&inet_ehash_secret, 0, rnd) == 0)
 245                 get_random_bytes(&ipv6_hash_secret, sizeof(ipv6_hash_secret));
 246 }
 247 EXPORT_SYMBOL(build_ehash_secret);
 248
 249 static inline int inet_netns_ok(struct net *net, int protocol)
 250 {
 251         int hash;
 252         const struct net_protocol *ipprot;
 253
 254         if (net_eq(net, &init_net))
 255                 return 1;
 256
 257         hash = protocol & (MAX_INET_PROTOS - 1);
 258         ipprot = rcu_dereference(inet_protos[hash]);
 259
 260         if (ipprot == NULL)
 261                 /* raw IP is OK */
 262                 return 1;
 263         return ipprot->netns_ok;
 264 }
 265
 266 /*
 267  *      Create an inet socket.
 268  */
 269
 270 static int inet_create(struct net *net, struct socket *sock, int protocol,
 271                        int kern)
 272 {
 273         struct sock *sk;
 274         struct inet_protosw *answer;
 275         struct inet_sock *inet;
 276         struct proto *answer_prot;
 277         unsigned char answer_flags;
 278         char answer_no_check;
 279         int try_loading_module = 0;
 280         int err;
 281
 282         if (unlikely(!inet_ehash_secret))
 283                 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
 284                         build_ehash_secret();
 285
 286         sock->state = SS_UNCONNECTED;
 287
 288         /* Look for the requested type/protocol pair. */
 289 lookup_protocol:
 290         err = -ESOCKTNOSUPPORT;
 291         rcu_read_lock();
 292         list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
 293
 294                 err = 0;
 295                 /* Check the non-wild match. */
 296                 if (protocol == answer->protocol) {
 297                         if (protocol != IPPROTO_IP)
 298                                 break;
 299                 } else {
 300                         /* Check for the two wild cases. */
 301                         if (IPPROTO_IP == protocol) {
 302                                 protocol = answer->protocol;
 303                                 break;
 304                         }
 305                         if (IPPROTO_IP == answer->protocol)
 306                                 break;
 307                 }
 308                 err = -EPROTONOSUPPORT;
 309         }
 310
 311         if (unlikely(err)) {
 312                 if (try_loading_module < 2) {
 313                         rcu_read_unlock();
 314                         /*
 315                          * Be more specific, e.g. net-pf-2-proto-132-type-1
 316                          * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
 317                          */
 318                         if (++try_loading_module == 1)
 319                                 request_module("net-pf-%d-proto-%d-type-%d",
 320                                                PF_INET, protocol, sock->type);
 321                         /*
 322                          * Fall back to generic, e.g. net-pf-2-proto-132
 323                          * (net-pf-PF_INET-proto-IPPROTO_SCTP)
 324                          */
 325                         else
 326                                 request_module("net-pf-%d-proto-%d",
 327                                                PF_INET, protocol);
 328                         goto lookup_protocol;
 329                 } else
 330                         goto out_rcu_unlock;
 331         }
 332
 333         err = -EPERM;
 334         if (sock->type == SOCK_RAW && !kern && !capable(CAP_NET_RAW))
 335                 goto out_rcu_unlock;
 336
 337         err = -EAFNOSUPPORT;
 338         if (!inet_netns_ok(net, protocol))
 339                 goto out_rcu_unlock;
 340
 341         sock->ops = answer->ops;
 342         answer_prot = answer->prot;
 343         answer_no_check = answer->no_check;
 344         answer_flags = answer->flags;
 345         rcu_read_unlock();
 346
 347         WARN_ON(answer_prot->slab == NULL);
 348
 349         err = -ENOBUFS;
 350         sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
 351         if (sk == NULL)
 352                 goto out;
 353
 354         err = 0;
 355         sk->sk_no_check = answer_no_check;
 356         if (INET_PROTOSW_REUSE & answer_flags)
 357                 sk->sk_reuse = 1;
 358
 359         inet = inet_sk(sk);
 360         inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
 361
 362         inet->nodefrag = 0;
 363
 364         if (SOCK_RAW == sock->type) {
 365                 inet->inet_num = protocol;
 366                 if (IPPROTO_RAW == protocol)
 367                         inet->hdrincl = 1;
 368         }
 369
 370         if (ipv4_config.no_pmtu_disc)
 371                 inet->pmtudisc = IP_PMTUDISC_DONT;
 372         else
 373                 inet->pmtudisc = IP_PMTUDISC_WANT;
 374
 375         inet->inet_id = 0;
 376
 377         sock_init_data(sock, sk);
 378
 379         sk->sk_destruct    = inet_sock_destruct;
 380         sk->sk_protocol    = protocol;
 381         sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
 382
 383         inet->uc_ttl    = -1;
 384         inet->mc_loop   = 1;
 385         inet->mc_ttl    = 1;
 386         inet->mc_all    = 1;
 387         inet->mc_index  = 0;
 388         inet->mc_list   = NULL;
 389
 390         sk_refcnt_debug_inc(sk);
 391
 392         if (inet->inet_num) {
 393                 /* It assumes that any protocol which allows
 394                  * the user to assign a number at socket
 395                  * creation time automatically
 396                  * shares.
 397                  */
 398                 inet->inet_sport = htons(inet->inet_num);
 399                 /* Add to protocol hash chains. */
 400                 sk->sk_prot->hash(sk);
 401         }
 402
 403         if (sk->sk_prot->init) {
 404                 err = sk->sk_prot->init(sk);
 405                 if (err)
 406                         sk_common_release(sk);
 407         }
 408 out:
 409         return err;
 410 out_rcu_unlock:
 411         rcu_read_unlock();
 412         goto out;
 413 }
 414
 415
 416 /*
 417  *      The peer socket should always be NULL (or else). When we call this
 418  *      function we are destroying the object and from then on nobody
 419  *      should refer to it.
 420  */
 421 int inet_release(struct socket *sock)
 422 {
 423         struct sock *sk = sock->sk;
 424
 425         if (sk) {
 426                 long timeout;
 427
 428                 sock_rps_reset_flow(sk);
 429
 430                 /* Applications forget to leave groups before exiting */
 431                 ip_mc_drop_socket(sk);
 432
 433                 /* If linger is set, we don't return until the close
 434                  * is complete.  Otherwise we return immediately. The
 435                  * actually closing is done the same either way.
 436                  *
 437                  * If the close is due to the process exiting, we never
 438                  * linger..
 439                  */
 440                 timeout = 0;
 441                 if (sock_flag(sk, SOCK_LINGER) &&
 442                     !(current->flags & PF_EXITING))
 443                         timeout = sk->sk_lingertime;
 444                 sock->sk = NULL;
 445                 sk->sk_prot->close(sk, timeout);
 446         }
 447         return 0;
 448 }
 449 EXPORT_SYMBOL(inet_release);
 450
 451 /* It is off by default, see below. */
 452 int sysctl_ip_nonlocal_bind __read_mostly;
 453 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
 454
 455 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 456 {
 457         struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
 458         struct sock *sk = sock->sk;
 459         struct inet_sock *inet = inet_sk(sk);
 460         unsigned short snum;
 461         int chk_addr_ret;
 462         int err;
 463
 464         /* If the socket has its own bind function then use it. (RAW) */
 465         if (sk->sk_prot->bind) {
 466                 err = sk->sk_prot->bind(sk, uaddr, addr_len);
 467                 goto out;
 468         }
 469         err = -EINVAL;
 470         if (addr_len < sizeof(struct sockaddr_in))
 471                 goto out;
 472
 473         if (addr->sin_family != AF_INET) {
 474                 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
 475                  * only if s_addr is INADDR_ANY.
 476                  */
 477                 err = -EAFNOSUPPORT;
 478                 if (addr->sin_family != AF_UNSPEC ||
 479                     addr->sin_addr.s_addr != htonl(INADDR_ANY))
 480                         goto out;
 481         }
 482
 483         chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr);
 484
 485         /* Not specified by any standard per-se, however it breaks too
 486          * many applications when removed.  It is unfortunate since
 487          * allowing applications to make a non-local bind solves
 488          * several problems with systems using dynamic addressing.
 489          * (ie. your servers still start up even if your ISDN link
 490          *  is temporarily down)
 491          */
 492         err = -EADDRNOTAVAIL;
 493         if (!sysctl_ip_nonlocal_bind &&
 494             !(inet->freebind || inet->transparent) &&
 495             addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
 496             chk_addr_ret != RTN_LOCAL &&
 497             chk_addr_ret != RTN_MULTICAST &&
 498             chk_addr_ret != RTN_BROADCAST)
 499                 goto out;
 500
 501         snum = ntohs(addr->sin_port);
 502         err = -EACCES;
 503         if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
 504                 goto out;
 505
 506         /*      We keep a pair of addresses. rcv_saddr is the one
 507          *      used by hash lookups, and saddr is used for transmit.
 508          *
 509          *      In the BSD API these are the same except where it
 510          *      would be illegal to use them (multicast/broadcast) in
 511          *      which case the sending device address is used.
 512          */
 513         lock_sock(sk);
 514
 515         /* Check these errors (active socket, double bind). */
 516         err = -EINVAL;
 517         if (sk->sk_state != TCP_CLOSE || inet->inet_num)
 518                 goto out_release_sock;
 519
 520         inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
 521         if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
 522                 inet->inet_saddr = 0;  /* Use device */
 523
 524         /* Make sure we are allowed to bind here. */
 525         if (sk->sk_prot->get_port(sk, snum)) {
 526                 inet->inet_saddr = inet->inet_rcv_saddr = 0;
 527                 err = -EADDRINUSE;
 528                 goto out_release_sock;
 529         }
 530
 531         if (inet->inet_rcv_saddr)
 532                 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
 533         if (snum)
 534                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
 535         inet->inet_sport = htons(inet->inet_num);
 536         inet->inet_daddr = 0;
 537         inet->inet_dport = 0;
 538         sk_dst_reset(sk);
 539         err = 0;
 540 out_release_sock:
 541         release_sock(sk);
 542 out:
 543         return err;
 544 }
 545 EXPORT_SYMBOL(inet_bind);
 546
 547 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
 548                        int addr_len, int flags)
 549 {
 550         struct sock *sk = sock->sk;
 551
 552         if (addr_len < sizeof(uaddr->sa_family))
 553                 return -EINVAL;
 554         if (uaddr->sa_family == AF_UNSPEC)
 555                 return sk->sk_prot->disconnect(sk, flags);
 556
 557         if (!inet_sk(sk)->inet_num && inet_autobind(sk))
 558                 return -EAGAIN;
 559         return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
 560 }
 561 EXPORT_SYMBOL(inet_dgram_connect);
 562
 563 static long inet_wait_for_connect(struct sock *sk, long timeo)
 564 {
 565         DEFINE_WAIT(wait);
 566
 567         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
 568
 569         /* Basic assumption: if someone sets sk->sk_err, he _must_
 570          * change state of the socket from TCP_SYN_*.
 571          * Connect() does not allow to get error notifications
 572          * without closing the socket.
 573          */
 574         while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 575                 release_sock(sk);
 576                 timeo = schedule_timeout(timeo);
 577                 lock_sock(sk);
 578                 if (signal_pending(current) || !timeo)
 579                         break;
 580                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
 581         }
 582         finish_wait(sk_sleep(sk), &wait);
 583         return timeo;
 584 }
 585
 586 /*
 587  *      Connect to a remote host. There is regrettably still a little
 588  *      TCP 'magic' in here.
 589  */
 590 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 591                         int addr_len, int flags)
 592 {
 593         struct sock *sk = sock->sk;
 594         int err;
 595         long timeo;
 596
 597         if (addr_len < sizeof(uaddr->sa_family))
 598                 return -EINVAL;
 599
 600         lock_sock(sk);
 601
 602         if (uaddr->sa_family == AF_UNSPEC) {
 603                 err = sk->sk_prot->disconnect(sk, flags);
 604                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 605                 goto out;
 606         }
 607
 608         switch (sock->state) {
 609         default:
 610                 err = -EINVAL;
 611                 goto out;
 612         case SS_CONNECTED:
 613                 err = -EISCONN;
 614                 goto out;
 615         case SS_CONNECTING:
 616                 err = -EALREADY;
 617                 /* Fall out of switch with err, set for this state */
 618                 break;
 619         case SS_UNCONNECTED:
 620                 err = -EISCONN;
 621                 if (sk->sk_state != TCP_CLOSE)
 622                         goto out;
 623
 624                 err = sk->sk_prot->connect(sk, uaddr, addr_len);
 625                 if (err < 0)
 626                         goto out;
 627
 628                 sock->state = SS_CONNECTING;
 629
 630                 /* Just entered SS_CONNECTING state; the only
 631                  * difference is that return value in non-blocking
 632                  * case is EINPROGRESS, rather than EALREADY.
 633                  */
 634                 err = -EINPROGRESS;
 635                 break;
 636         }
 637
 638         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
 639
 640         if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 641                 /* Error code is set above */
 642                 if (!timeo || !inet_wait_for_connect(sk, timeo))
 643                         goto out;
 644
 645                 err = sock_intr_errno(timeo);
 646                 if (signal_pending(current))
 647                         goto out;
 648         }
 649
 650         /* Connection was closed by RST, timeout, ICMP error
 651          * or another process disconnected us.
 652          */
 653         if (sk->sk_state == TCP_CLOSE)
 654                 goto sock_error;
 655
 656         /* sk->sk_err may be not zero now, if RECVERR was ordered by user
 657          * and error was received after socket entered established state.
 658          * Hence, it is handled normally after connect() return successfully.
 659          */
 660
 661         sock->state = SS_CONNECTED;
 662         err = 0;
 663 out:
 664         release_sock(sk);
 665         return err;
 666
 667 sock_error:
 668         err = sock_error(sk) ? : -ECONNABORTED;
 669         sock->state = SS_UNCONNECTED;
 670         if (sk->sk_prot->disconnect(sk, flags))
 671                 sock->state = SS_DISCONNECTING;
 672         goto out;
 673 }
 674 EXPORT_SYMBOL(inet_stream_connect);
 675
 676 /*
 677  *      Accept a pending connection. The TCP layer now gives BSD semantics.
 678  */
 679
 680 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
 681 {
 682         struct sock *sk1 = sock->sk;
 683         int err = -EINVAL;
 684         struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
 685
 686         if (!sk2)
 687                 goto do_err;
 688
 689         lock_sock(sk2);
 690
 691         sock_rps_record_flow(sk2);
 692         WARN_ON(!((1 << sk2->sk_state) &
 693                   (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE)));
 694
 695         sock_graft(sk2, newsock);
 696
 697         newsock->state = SS_CONNECTED;
 698         err = 0;
 699         release_sock(sk2);
 700 do_err:
 701         return err;
 702 }
 703 EXPORT_SYMBOL(inet_accept);
 704
 705
 706 /*
 707  *      This does both peername and sockname.
 708  */
 709 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
 710                         int *uaddr_len, int peer)
 711 {
 712         struct sock *sk         = sock->sk;
 713         struct inet_sock *inet  = inet_sk(sk);
 714         DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
 715
 716         sin->sin_family = AF_INET;
 717         if (peer) {
 718                 if (!inet->inet_dport ||
 719                     (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
 720                      peer == 1))
 721                         return -ENOTCONN;
 722                 sin->sin_port = inet->inet_dport;
 723                 sin->sin_addr.s_addr = inet->inet_daddr;
 724         } else {
 725                 __be32 addr = inet->inet_rcv_saddr;
 726                 if (!addr)
 727                         addr = inet->inet_saddr;
 728                 sin->sin_port = inet->inet_sport;
 729                 sin->sin_addr.s_addr = addr;
 730         }
 731         memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 732         *uaddr_len = sizeof(*sin);
 733         return 0;
 734 }
 735 EXPORT_SYMBOL(inet_getname);
 736
 737 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
 738                  size_t size)
 739 {
 740         struct sock *sk = sock->sk;
 741
 742         sock_rps_record_flow(sk);
 743
 744         /* We may need to bind the socket. */
 745         if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
 746             inet_autobind(sk))
 747                 return -EAGAIN;
 748
 749         return sk->sk_prot->sendmsg(iocb, sk, msg, size);
 750 }
 751 EXPORT_SYMBOL(inet_sendmsg);
 752
 753 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
 754                       size_t size, int flags)
 755 {
 756         struct sock *sk = sock->sk;
 757
 758         sock_rps_record_flow(sk);
 759
 760         /* We may need to bind the socket. */
 761         if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
 762             inet_autobind(sk))
 763                 return -EAGAIN;
 764
 765         if (sk->sk_prot->sendpage)
 766                 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
 767         return sock_no_sendpage(sock, page, offset, size, flags);
 768 }
 769 EXPORT_SYMBOL(inet_sendpage);
 770
 771 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
 772                  size_t size, int flags)
 773 {
 774         struct sock *sk = sock->sk;
 775         int addr_len = 0;
 776         int err;
 777
 778         sock_rps_record_flow(sk);
 779
 780         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
 781                                    flags & ~MSG_DONTWAIT, &addr_len);
 782         if (err >= 0)
 783                 msg->msg_namelen = addr_len;
 784         return err;
 785 }
 786 EXPORT_SYMBOL(inet_recvmsg);
 787
 788 int inet_shutdown(struct socket *sock, int how)
 789 {
 790         struct sock *sk = sock->sk;
 791         int err = 0;
 792
 793         /* This should really check to make sure
 794          * the socket is a TCP socket. (WHY AC...)
 795          */
 796         how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
 797                        1->2 bit 2 snds.
 798                        2->3 */
 799         if ((how & ~SHUTDOWN_MASK) || !how)     /* MAXINT->0 */
 800                 return -EINVAL;
 801
 802         lock_sock(sk);
 803         if (sock->state == SS_CONNECTING) {
 804                 if ((1 << sk->sk_state) &
 805                     (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
 806                         sock->state = SS_DISCONNECTING;
 807                 else
 808                         sock->state = SS_CONNECTED;
 809         }
 810
 811         switch (sk->sk_state) {
 812         case TCP_CLOSE:
 813                 err = -ENOTCONN;
 814                 /* Hack to wake up other listeners, who can poll for
 815                    POLLHUP, even on eg. unconnected UDP sockets -- RR */
 816         default:
 817                 sk->sk_shutdown |= how;
 818                 if (sk->sk_prot->shutdown)
 819                         sk->sk_prot->shutdown(sk, how);
 820                 break;
 821
 822         /* Remaining two branches are temporary solution for missing
 823          * close() in multithreaded environment. It is _not_ a good idea,
 824          * but we have no choice until close() is repaired at VFS level.
 825          */
 826         case TCP_LISTEN:
 827                 if (!(how & RCV_SHUTDOWN))
 828                         break;
 829                 /* Fall through */
 830         case TCP_SYN_SENT:
 831                 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
 832                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 833                 break;
 834         }
 835
 836         /* Wake up anyone sleeping in poll. */
 837         sk->sk_state_change(sk);
 838         release_sock(sk);
 839         return err;
 840 }
 841 EXPORT_SYMBOL(inet_shutdown);
 842
 843 /*
 844  *      ioctl() calls you can issue on an INET socket. Most of these are
 845  *      device configuration and stuff and very rarely used. Some ioctls
 846  *      pass on to the socket itself.
 847  *
 848  *      NOTE: I like the idea of a module for the config stuff. ie ifconfig
 849  *      loads the devconfigure module does its configuring and unloads it.
 850  *      There's a good 20K of config code hanging around the kernel.
 851  */
 852
 853 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 854 {
 855         struct sock *sk = sock->sk;
 856         int err = 0;
 857         struct net *net = sock_net(sk);
 858
 859         switch (cmd) {
 860         case SIOCGSTAMP:
 861                 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
 862                 break;
 863         case SIOCGSTAMPNS:
 864                 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
 865                 break;
 866         case SIOCADDRT:
 867         case SIOCDELRT:
 868         case SIOCRTMSG:
 869                 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
 870                 break;
 871         case SIOCDARP:
 872         case SIOCGARP:
 873         case SIOCSARP:
 874                 err = arp_ioctl(net, cmd, (void __user *)arg);
 875                 break;
 876         case SIOCGIFADDR:
 877         case SIOCSIFADDR:
 878         case SIOCGIFBRDADDR:
 879         case SIOCSIFBRDADDR:
 880         case SIOCGIFNETMASK:
 881         case SIOCSIFNETMASK:
 882         case SIOCGIFDSTADDR:
 883         case SIOCSIFDSTADDR:
 884         case SIOCSIFPFLAGS:
 885         case SIOCGIFPFLAGS:
 886         case SIOCSIFFLAGS:
 887                 err = devinet_ioctl(net, cmd, (void __user *)arg);
 888                 break;
 889         default:
 890                 if (sk->sk_prot->ioctl)
 891                         err = sk->sk_prot->ioctl(sk, cmd, arg);
 892                 else
 893                         err = -ENOIOCTLCMD;
 894                 break;
 895         }
 896         return err;
 897 }
 898 EXPORT_SYMBOL(inet_ioctl);
 899
 900 #ifdef CONFIG_COMPAT
 901 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 902 {
 903         struct sock *sk = sock->sk;
 904         int err = -ENOIOCTLCMD;
 905
 906         if (sk->sk_prot->compat_ioctl)
 907                 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
 908
 909         return err;
 910 }
 911 #endif
 912
 913 const struct proto_ops inet_stream_ops = {
 914         .family            = PF_INET,
 915         .owner             = THIS_MODULE,
 916         .release           = inet_release,
 917         .bind              = inet_bind,
 918         .connect           = inet_stream_connect,
 919         .socketpair        = sock_no_socketpair,
 920         .accept            = inet_accept,
 921         .getname           = inet_getname,
 922         .poll              = tcp_poll,
 923         .ioctl             = inet_ioctl,
 924         .listen            = inet_listen,
 925         .shutdown          = inet_shutdown,
 926         .setsockopt        = sock_common_setsockopt,
 927         .getsockopt        = sock_common_getsockopt,
 928         .sendmsg           = inet_sendmsg,
 929         .recvmsg           = inet_recvmsg,
 930         .mmap              = sock_no_mmap,
 931         .sendpage          = inet_sendpage,
 932         .splice_read       = tcp_splice_read,
 933 #ifdef CONFIG_COMPAT
 934         .compat_setsockopt = compat_sock_common_setsockopt,
 935         .compat_getsockopt = compat_sock_common_getsockopt,
 936         .compat_ioctl      = inet_compat_ioctl,
 937 #endif
 938 };
 939 EXPORT_SYMBOL(inet_stream_ops);
 940
 941 const struct proto_ops inet_dgram_ops = {
 942         .family            = PF_INET,
 943         .owner             = THIS_MODULE,
 944         .release           = inet_release,
 945         .bind              = inet_bind,
 946         .connect           = inet_dgram_connect,
 947         .socketpair        = sock_no_socketpair,
 948         .accept            = sock_no_accept,
 949         .getname           = inet_getname,
 950         .poll              = udp_poll,
 951         .ioctl             = inet_ioctl,
 952         .listen            = sock_no_listen,
 953         .shutdown          = inet_shutdown,
 954         .setsockopt        = sock_common_setsockopt,
 955         .getsockopt        = sock_common_getsockopt,
 956         .sendmsg           = inet_sendmsg,
 957         .recvmsg           = inet_recvmsg,
 958         .mmap              = sock_no_mmap,
 959         .sendpage          = inet_sendpage,
 960 #ifdef CONFIG_COMPAT
 961         .compat_setsockopt = compat_sock_common_setsockopt,
 962         .compat_getsockopt = compat_sock_common_getsockopt,
 963         .compat_ioctl      = inet_compat_ioctl,
 964 #endif
 965 };
 966 EXPORT_SYMBOL(inet_dgram_ops);
 967
 968 /*
 969  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
 970  * udp_poll
 971  */
 972 static const struct proto_ops inet_sockraw_ops = {
 973         .family            = PF_INET,
 974         .owner             = THIS_MODULE,
 975         .release           = inet_release,
 976         .bind              = inet_bind,
 977         .connect           = inet_dgram_connect,
 978         .socketpair        = sock_no_socketpair,
 979         .accept            = sock_no_accept,
 980         .getname           = inet_getname,
 981         .poll              = datagram_poll,
 982         .ioctl             = inet_ioctl,
 983         .listen            = sock_no_listen,
 984         .shutdown          = inet_shutdown,
 985         .setsockopt        = sock_common_setsockopt,
 986         .getsockopt        = sock_common_getsockopt,
 987         .sendmsg           = inet_sendmsg,
 988         .recvmsg           = inet_recvmsg,
 989         .mmap              = sock_no_mmap,
 990         .sendpage          = inet_sendpage,
 991 #ifdef CONFIG_COMPAT
 992         .compat_setsockopt = compat_sock_common_setsockopt,
 993         .compat_getsockopt = compat_sock_common_getsockopt,
 994         .compat_ioctl      = inet_compat_ioctl,
 995 #endif
 996 };
 997
 998 static const struct net_proto_family inet_family_ops = {
 999         .family = PF_INET,
1000         .create = inet_create,
1001         .owner  = THIS_MODULE,
1002 };
1003
1004 /* Upon startup we insert all the elements in inetsw_array[] into
1005  * the linked list inetsw.
1006  */
1007 static struct inet_protosw inetsw_array[] =
1008 {
1009         {
1010                 .type =       SOCK_STREAM,
1011                 .protocol =   IPPROTO_TCP,
1012                 .prot =       &tcp_prot,
1013                 .ops =        &inet_stream_ops,
1014                 .no_check =   0,
1015                 .flags =      INET_PROTOSW_PERMANENT |
1016                               INET_PROTOSW_ICSK,
1017         },
1018
1019         {
1020                 .type =       SOCK_DGRAM,
1021                 .protocol =   IPPROTO_UDP,
1022                 .prot =       &udp_prot,
1023                 .ops =        &inet_dgram_ops,
1024                 .no_check =   UDP_CSUM_DEFAULT,
1025                 .flags =      INET_PROTOSW_PERMANENT,
1026        },
1027
1028        {
1029                 .type =       SOCK_DGRAM,
1030                 .protocol =   IPPROTO_ICMP,
1031                 .prot =       &ping_prot,
1032                 .ops =        &inet_dgram_ops,
1033                 .no_check =   UDP_CSUM_DEFAULT,
1034                 .flags =      INET_PROTOSW_REUSE,
1035        },
1036
1037        {
1038                .type =       SOCK_RAW,
1039                .protocol =   IPPROTO_IP,        /* wild card */
1040                .prot =       &raw_prot,
1041                .ops =        &inet_sockraw_ops,
1042                .no_check =   UDP_CSUM_DEFAULT,
1043                .flags =      INET_PROTOSW_REUSE,
1044        }
1045 };
1046
1047 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1048
1049 void inet_register_protosw(struct inet_protosw *p)
1050 {
1051         struct list_head *lh;
1052         struct inet_protosw *answer;
1053         int protocol = p->protocol;
1054         struct list_head *last_perm;
1055
1056         spin_lock_bh(&inetsw_lock);
1057
1058         if (p->type >= SOCK_MAX)
1059                 goto out_illegal;
1060
1061         /* If we are trying to override a permanent protocol, bail. */
1062         answer = NULL;
1063         last_perm = &inetsw[p->type];
1064         list_for_each(lh, &inetsw[p->type]) {
1065                 answer = list_entry(lh, struct inet_protosw, list);
1066
1067                 /* Check only the non-wild match. */
1068                 if (INET_PROTOSW_PERMANENT & answer->flags) {
1069                         if (protocol == answer->protocol)
1070                                 break;
1071                         last_perm = lh;
1072                 }
1073
1074                 answer = NULL;
1075         }
1076         if (answer)
1077                 goto out_permanent;
1078
1079         /* Add the new entry after the last permanent entry if any, so that
1080          * the new entry does not override a permanent entry when matched with
1081          * a wild-card protocol. But it is allowed to override any existing
1082          * non-permanent entry.  This means that when we remove this entry, the
1083          * system automatically returns to the old behavior.
1084          */
1085         list_add_rcu(&p->list, last_perm);
1086 out:
1087         spin_unlock_bh(&inetsw_lock);
1088
1089         return;
1090
1091 out_permanent:
1092         printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
1093                protocol);
1094         goto out;
1095
1096 out_illegal:
1097         printk(KERN_ERR
1098                "Ignoring attempt to register invalid socket type %d.\n",
1099                p->type);
1100         goto out;
1101 }
1102 EXPORT_SYMBOL(inet_register_protosw);
1103
1104 void inet_unregister_protosw(struct inet_protosw *p)
1105 {
1106         if (INET_PROTOSW_PERMANENT & p->flags) {
1107                 printk(KERN_ERR
1108                        "Attempt to unregister permanent protocol %d.\n",
1109                        p->protocol);
1110         } else {
1111                 spin_lock_bh(&inetsw_lock);
1112                 list_del_rcu(&p->list);
1113                 spin_unlock_bh(&inetsw_lock);
1114
1115                 synchronize_net();
1116         }
1117 }
1118 EXPORT_SYMBOL(inet_unregister_protosw);
1119
1120 /*
1121  *      Shall we try to damage output packets if routing dev changes?
1122  */
1123
1124 int sysctl_ip_dynaddr __read_mostly;
1125
1126 static int inet_sk_reselect_saddr(struct sock *sk)
1127 {
1128         struct inet_sock *inet = inet_sk(sk);
1129         __be32 old_saddr = inet->inet_saddr;
1130         __be32 daddr = inet->inet_daddr;
1131         struct flowi4 *fl4;
1132         struct rtable *rt;
1133         __be32 new_saddr;
1134         struct ip_options_rcu *inet_opt;
1135
1136         inet_opt = rcu_dereference_protected(inet->inet_opt,
1137                                              sock_owned_by_user(sk));
1138         if (inet_opt && inet_opt->opt.srr)
1139                 daddr = inet_opt->opt.faddr;
1140
1141         /* Query new route. */
1142         fl4 = &inet->cork.fl.u.ip4;
1143         rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1144                               sk->sk_bound_dev_if, sk->sk_protocol,
1145                               inet->inet_sport, inet->inet_dport, sk, false);
1146         if (IS_ERR(rt))
1147                 return PTR_ERR(rt);
1148
1149         sk_setup_caps(sk, &rt->dst);
1150
1151         new_saddr = fl4->saddr;
1152
1153         if (new_saddr == old_saddr)
1154                 return 0;
1155
1156         if (sysctl_ip_dynaddr > 1) {
1157                 printk(KERN_INFO "%s(): shifting inet->saddr from %pI4 to %pI4\n",
1158                        __func__, &old_saddr, &new_saddr);
1159         }
1160
1161         inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1162
1163         /*
1164          * XXX The only one ugly spot where we need to
1165          * XXX really change the sockets identity after
1166          * XXX it has entered the hashes. -DaveM
1167          *
1168          * Besides that, it does not check for connection
1169          * uniqueness. Wait for troubles.
1170          */
1171         __sk_prot_rehash(sk);
1172         return 0;
1173 }
1174
1175 int inet_sk_rebuild_header(struct sock *sk)
1176 {
1177         struct inet_sock *inet = inet_sk(sk);
1178         struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1179         __be32 daddr;
1180         struct ip_options_rcu *inet_opt;
1181         struct flowi4 *fl4;
1182         int err;
1183
1184         /* Route is OK, nothing to do. */
1185         if (rt)
1186                 return 0;
1187
1188         /* Reroute. */
1189         rcu_read_lock();
1190         inet_opt = rcu_dereference(inet->inet_opt);
1191         daddr = inet->inet_daddr;
1192         if (inet_opt && inet_opt->opt.srr)
1193                 daddr = inet_opt->opt.faddr;
1194         rcu_read_unlock();
1195         fl4 = &inet->cork.fl.u.ip4;
1196         rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1197                                    inet->inet_dport, inet->inet_sport,
1198                                    sk->sk_protocol, RT_CONN_FLAGS(sk),
1199                                    sk->sk_bound_dev_if);
1200         if (!IS_ERR(rt)) {
1201                 err = 0;
1202                 sk_setup_caps(sk, &rt->dst);
1203         } else {
1204                 err = PTR_ERR(rt);
1205
1206                 /* Routing failed... */
1207                 sk->sk_route_caps = 0;
1208                 /*
1209                  * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1210                  * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1211                  */
1212                 if (!sysctl_ip_dynaddr ||
1213                     sk->sk_state != TCP_SYN_SENT ||
1214                     (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1215                     (err = inet_sk_reselect_saddr(sk)) != 0)
1216                         sk->sk_err_soft = -err;
1217         }
1218
1219         return err;
1220 }
1221 EXPORT_SYMBOL(inet_sk_rebuild_header);
1222
1223 static int inet_gso_send_check(struct sk_buff *skb)
1224 {
1225         const struct iphdr *iph;
1226         const struct net_protocol *ops;
1227         int proto;
1228         int ihl;
1229         int err = -EINVAL;
1230
1231         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1232                 goto out;
1233
1234         iph = ip_hdr(skb);
1235         ihl = iph->ihl * 4;
1236         if (ihl < sizeof(*iph))
1237                 goto out;
1238
1239         if (unlikely(!pskb_may_pull(skb, ihl)))
1240                 goto out;
1241
1242         __skb_pull(skb, ihl);
1243         skb_reset_transport_header(skb);
1244         iph = ip_hdr(skb);
1245         proto = iph->protocol & (MAX_INET_PROTOS - 1);
1246         err = -EPROTONOSUPPORT;
1247
1248         rcu_read_lock();
1249         ops = rcu_dereference(inet_protos[proto]);
1250         if (likely(ops && ops->gso_send_check))
1251                 err = ops->gso_send_check(skb);
1252         rcu_read_unlock();
1253
1254 out:
1255         return err;
1256 }
1257
1258 static struct sk_buff *inet_gso_segment(struct sk_buff *skb, u32 features)
1259 {
1260         struct sk_buff *segs = ERR_PTR(-EINVAL);
1261         struct iphdr *iph;
1262         const struct net_protocol *ops;
1263         int proto;
1264         int ihl;
1265         int id;
1266         unsigned int offset = 0;
1267
1268         if (!(features & NETIF_F_V4_CSUM))
1269                 features &= ~NETIF_F_SG;
1270
1271         if (unlikely(skb_shinfo(skb)->gso_type &
1272                      ~(SKB_GSO_TCPV4 |
1273                        SKB_GSO_UDP |
1274                        SKB_GSO_DODGY |
1275                        SKB_GSO_TCP_ECN |
1276                        0)))
1277                 goto out;
1278
1279         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1280                 goto out;
1281
1282         iph = ip_hdr(skb);
1283         ihl = iph->ihl * 4;
1284         if (ihl < sizeof(*iph))
1285                 goto out;
1286
1287         if (unlikely(!pskb_may_pull(skb, ihl)))
1288                 goto out;
1289
1290         __skb_pull(skb, ihl);
1291         skb_reset_transport_header(skb);
1292         iph = ip_hdr(skb);
1293         id = ntohs(iph->id);
1294         proto = iph->protocol & (MAX_INET_PROTOS - 1);
1295         segs = ERR_PTR(-EPROTONOSUPPORT);
1296
1297         rcu_read_lock();
1298         ops = rcu_dereference(inet_protos[proto]);
1299         if (likely(ops && ops->gso_segment))
1300                 segs = ops->gso_segment(skb, features);
1301         rcu_read_unlock();
1302
1303         if (!segs || IS_ERR(segs))
1304                 goto out;
1305
1306         skb = segs;
1307         do {
1308                 iph = ip_hdr(skb);
1309                 if (proto == IPPROTO_UDP) {
1310                         iph->id = htons(id);
1311                         iph->frag_off = htons(offset >> 3);
1312                         if (skb->next != NULL)
1313                                 iph->frag_off |= htons(IP_MF);
1314                         offset += (skb->len - skb->mac_len - iph->ihl * 4);
1315                 } else
1316                         iph->id = htons(id++);
1317                 iph->tot_len = htons(skb->len - skb->mac_len);
1318                 iph->check = 0;
1319                 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1320         } while ((skb = skb->next));
1321
1322 out:
1323         return segs;
1324 }
1325
1326 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1327                                          struct sk_buff *skb)
1328 {
1329         const struct net_protocol *ops;
1330         struct sk_buff **pp = NULL;
1331         struct sk_buff *p;
1332         const struct iphdr *iph;
1333         unsigned int hlen;
1334         unsigned int off;
1335         unsigned int id;
1336         int flush = 1;
1337         int proto;
1338
1339         off = skb_gro_offset(skb);
1340         hlen = off + sizeof(*iph);
1341         iph = skb_gro_header_fast(skb, off);
1342         if (skb_gro_header_hard(skb, hlen)) {
1343                 iph = skb_gro_header_slow(skb, hlen, off);
1344                 if (unlikely(!iph))
1345                         goto out;
1346         }
1347
1348         proto = iph->protocol & (MAX_INET_PROTOS - 1);
1349
1350         rcu_read_lock();
1351         ops = rcu_dereference(inet_protos[proto]);
1352         if (!ops || !ops->gro_receive)
1353                 goto out_unlock;
1354
1355         if (*(u8 *)iph != 0x45)
1356                 goto out_unlock;
1357
1358         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1359                 goto out_unlock;
1360
1361         id = ntohl(*(__be32 *)&iph->id);
1362         flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF));
1363         id >>= 16;
1364
1365         for (p = *head; p; p = p->next) {
1366                 struct iphdr *iph2;
1367
1368                 if (!NAPI_GRO_CB(p)->same_flow)
1369                         continue;
1370
1371                 iph2 = ip_hdr(p);
1372
1373                 if ((iph->protocol ^ iph2->protocol) |
1374                     (iph->tos ^ iph2->tos) |
1375                     ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1376                     ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1377                         NAPI_GRO_CB(p)->same_flow = 0;
1378                         continue;
1379                 }
1380
1381                 /* All fields must match except length and checksum. */
1382                 NAPI_GRO_CB(p)->flush |=
1383                         (iph->ttl ^ iph2->ttl) |
1384                         ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1385
1386                 NAPI_GRO_CB(p)->flush |= flush;
1387         }
1388
1389         NAPI_GRO_CB(skb)->flush |= flush;
1390         skb_gro_pull(skb, sizeof(*iph));
1391         skb_set_transport_header(skb, skb_gro_offset(skb));
1392
1393         pp = ops->gro_receive(head, skb);
1394
1395 out_unlock:
1396         rcu_read_unlock();
1397
1398 out:
1399         NAPI_GRO_CB(skb)->flush |= flush;
1400
1401         return pp;
1402 }
1403
1404 static int inet_gro_complete(struct sk_buff *skb)
1405 {
1406         const struct net_protocol *ops;
1407         struct iphdr *iph = ip_hdr(skb);
1408         int proto = iph->protocol & (MAX_INET_PROTOS - 1);
1409         int err = -ENOSYS;
1410         __be16 newlen = htons(skb->len - skb_network_offset(skb));
1411
1412         csum_replace2(&iph->check, iph->tot_len, newlen);
1413         iph->tot_len = newlen;
1414
1415         rcu_read_lock();
1416         ops = rcu_dereference(inet_protos[proto]);
1417         if (WARN_ON(!ops || !ops->gro_complete))
1418                 goto out_unlock;
1419
1420         err = ops->gro_complete(skb);
1421
1422 out_unlock:
1423         rcu_read_unlock();
1424
1425         return err;
1426 }
1427
1428 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1429                          unsigned short type, unsigned char protocol,
1430                          struct net *net)
1431 {
1432         struct socket *sock;
1433         int rc = sock_create_kern(family, type, protocol, &sock);
1434
1435         if (rc == 0) {
1436                 *sk = sock->sk;
1437                 (*sk)->sk_allocation = GFP_ATOMIC;
1438                 /*
1439                  * Unhash it so that IP input processing does not even see it,
1440                  * we do not wish this socket to see incoming packets.
1441                  */
1442                 (*sk)->sk_prot->unhash(*sk);
1443
1444                 sk_change_net(*sk, net);
1445         }
1446         return rc;
1447 }
1448 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1449
1450 unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1451 {
1452         unsigned long res = 0;
1453         int i, j;
1454
1455         for_each_possible_cpu(i) {
1456                 for (j = 0; j < SNMP_ARRAY_SZ; j++)
1457                         res += *(((unsigned long *) per_cpu_ptr(mib[j], i)) + offt);
1458         }
1459         return res;
1460 }
1461 EXPORT_SYMBOL_GPL(snmp_fold_field);
1462
1463 #if BITS_PER_LONG==32
1464
1465 u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset)
1466 {
1467         u64 res = 0;
1468         int cpu;
1469
1470         for_each_possible_cpu(cpu) {
1471                 void *bhptr;
1472                 struct u64_stats_sync *syncp;
1473                 u64 v;
1474                 unsigned int start;
1475
1476                 bhptr = per_cpu_ptr(mib[0], cpu);
1477                 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1478                 do {
1479                         start = u64_stats_fetch_begin_bh(syncp);
1480                         v = *(((u64 *) bhptr) + offt);
1481                 } while (u64_stats_fetch_retry_bh(syncp, start));
1482
1483                 res += v;
1484         }
1485         return res;
1486 }
1487 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1488 #endif
1489
1490 int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align)
1491 {
1492         BUG_ON(ptr == NULL);
1493         ptr[0] = __alloc_percpu(mibsize, align);
1494         if (!ptr[0])
1495                 return -ENOMEM;
1496 #if SNMP_ARRAY_SZ == 2
1497         ptr[1] = __alloc_percpu(mibsize, align);
1498         if (!ptr[1]) {
1499                 free_percpu(ptr[0]);
1500                 ptr[0] = NULL;
1501                 return -ENOMEM;
1502         }
1503 #endif
1504         return 0;
1505 }
1506 EXPORT_SYMBOL_GPL(snmp_mib_init);
1507
1508 void snmp_mib_free(void __percpu *ptr[SNMP_ARRAY_SZ])
1509 {
1510         int i;
1511
1512         BUG_ON(ptr == NULL);
1513         for (i = 0; i < SNMP_ARRAY_SZ; i++) {
1514                 free_percpu(ptr[i]);
1515                 ptr[i] = NULL;
1516         }
1517 }
1518 EXPORT_SYMBOL_GPL(snmp_mib_free);
1519
1520 #ifdef CONFIG_IP_MULTICAST
1521 static const struct net_protocol igmp_protocol = {
1522         .handler =      igmp_rcv,
1523         .netns_ok =     1,
1524 };
1525 #endif
1526
1527 static const struct net_protocol tcp_protocol = {
1528         .handler =      tcp_v4_rcv,
1529         .err_handler =  tcp_v4_err,
1530         .gso_send_check = tcp_v4_gso_send_check,
1531         .gso_segment =  tcp_tso_segment,
1532         .gro_receive =  tcp4_gro_receive,
1533         .gro_complete = tcp4_gro_complete,
1534         .no_policy =    1,
1535         .netns_ok =     1,
1536 };
1537
1538 static const struct net_protocol udp_protocol = {
1539         .handler =      udp_rcv,
1540         .err_handler =  udp_err,
1541         .gso_send_check = udp4_ufo_send_check,
1542         .gso_segment = udp4_ufo_fragment,
1543         .no_policy =    1,
1544         .netns_ok =     1,
1545 };
1546
1547 static const struct net_protocol icmp_protocol = {
1548         .handler =      icmp_rcv,
1549         .err_handler =  ping_err,
1550         .no_policy =    1,
1551         .netns_ok =     1,
1552 };
1553
1554 static __net_init int ipv4_mib_init_net(struct net *net)
1555 {
1556         if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1557                           sizeof(struct tcp_mib),
1558                           __alignof__(struct tcp_mib)) < 0)
1559                 goto err_tcp_mib;
1560         if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1561                           sizeof(struct ipstats_mib),
1562                           __alignof__(struct ipstats_mib)) < 0)
1563                 goto err_ip_mib;
1564         if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1565                           sizeof(struct linux_mib),
1566                           __alignof__(struct linux_mib)) < 0)
1567                 goto err_net_mib;
1568         if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1569                           sizeof(struct udp_mib),
1570                           __alignof__(struct udp_mib)) < 0)
1571                 goto err_udp_mib;
1572         if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1573                           sizeof(struct udp_mib),
1574                           __alignof__(struct udp_mib)) < 0)
1575                 goto err_udplite_mib;
1576         if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1577                           sizeof(struct icmp_mib),
1578                           __alignof__(struct icmp_mib)) < 0)
1579                 goto err_icmp_mib;
1580         if (snmp_mib_init((void __percpu **)net->mib.icmpmsg_statistics,
1581                           sizeof(struct icmpmsg_mib),
1582                           __alignof__(struct icmpmsg_mib)) < 0)
1583                 goto err_icmpmsg_mib;
1584
1585         tcp_mib_init(net);
1586         return 0;
1587
1588 err_icmpmsg_mib:
1589         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1590 err_icmp_mib:
1591         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1592 err_udplite_mib:
1593         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1594 err_udp_mib:
1595         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1596 err_net_mib:
1597         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1598 err_ip_mib:
1599         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1600 err_tcp_mib:
1601         return -ENOMEM;
1602 }
1603
1604 static __net_exit void ipv4_mib_exit_net(struct net *net)
1605 {
1606         snmp_mib_free((void __percpu **)net->mib.icmpmsg_statistics);
1607         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1608         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1609         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1610         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1611         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1612         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1613 }
1614
1615 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1616         .init = ipv4_mib_init_net,
1617         .exit = ipv4_mib_exit_net,
1618 };
1619
1620 static int __init init_ipv4_mibs(void)
1621 {
1622         return register_pernet_subsys(&ipv4_mib_ops);
1623 }
1624
1625 static int ipv4_proc_init(void);
1626
1627 /*
1628  *      IP protocol layer initialiser
1629  */
1630
1631 static struct packet_type ip_packet_type __read_mostly = {
1632         .type = cpu_to_be16(ETH_P_IP),
1633         .func = ip_rcv,
1634         .gso_send_check = inet_gso_send_check,
1635         .gso_segment = inet_gso_segment,
1636         .gro_receive = inet_gro_receive,
1637         .gro_complete = inet_gro_complete,
1638 };
1639
1640 static int __init inet_init(void)
1641 {
1642         struct sk_buff *dummy_skb;
1643         struct inet_protosw *q;
1644         struct list_head *r;
1645         int rc = -EINVAL;
1646
1647         BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1648
1649         sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL);
1650         if (!sysctl_local_reserved_ports)
1651                 goto out;
1652
1653         rc = proto_register(&tcp_prot, 1);
1654         if (rc)
1655                 goto out_free_reserved_ports;
1656
1657         rc = proto_register(&udp_prot, 1);
1658         if (rc)
1659                 goto out_unregister_tcp_proto;
1660
1661         rc = proto_register(&raw_prot, 1);
1662         if (rc)
1663                 goto out_unregister_udp_proto;
1664
1665         rc = proto_register(&ping_prot, 1);
1666         if (rc)
1667                 goto out_unregister_raw_proto;
1668
1669         /*
1670          *      Tell SOCKET that we are alive...
1671          */
1672
1673         (void)sock_register(&inet_family_ops);
1674
1675 #ifdef CONFIG_SYSCTL
1676         ip_static_sysctl_init();
1677 #endif
1678
1679         /*
1680          *      Add all the base protocols.
1681          */
1682
1683         if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1684                 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1685         if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1686                 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1687         if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1688                 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1689 #ifdef CONFIG_IP_MULTICAST
1690         if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1691                 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1692 #endif
1693
1694         /* Register the socket-side information for inet_create. */
1695         for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1696                 INIT_LIST_HEAD(r);
1697
1698         for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1699                 inet_register_protosw(q);
1700
1701         /*
1702          *      Set the ARP module up
1703          */
1704
1705         arp_init();
1706
1707         /*
1708          *      Set the IP module up
1709          */
1710
1711         ip_init();
1712
1713         tcp_v4_init();
1714
1715         /* Setup TCP slab cache for open requests. */
1716         tcp_init();
1717
1718         /* Setup UDP memory threshold */
1719         udp_init();
1720
1721         /* Add UDP-Lite (RFC 3828) */
1722         udplite4_register();
1723
1724         ping_init();
1725
1726         /*
1727          *      Set the ICMP layer up
1728          */
1729
1730         if (icmp_init() < 0)
1731                 panic("Failed to create the ICMP control socket.\n");
1732
1733         /*
1734          *      Initialise the multicast router
1735          */
1736 #if defined(CONFIG_IP_MROUTE)
1737         if (ip_mr_init())
1738                 printk(KERN_CRIT "inet_init: Cannot init ipv4 mroute\n");
1739 #endif
1740         /*
1741          *      Initialise per-cpu ipv4 mibs
1742          */
1743
1744         if (init_ipv4_mibs())
1745                 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n");
1746
1747         ipv4_proc_init();
1748
1749         ipfrag_init();
1750
1751         dev_add_pack(&ip_packet_type);
1752
1753         rc = 0;
1754 out:
1755         return rc;
1756 out_unregister_raw_proto:
1757         proto_unregister(&raw_prot);
1758 out_unregister_udp_proto:
1759         proto_unregister(&udp_prot);
1760 out_unregister_tcp_proto:
1761         proto_unregister(&tcp_prot);
1762 out_free_reserved_ports:
1763         kfree(sysctl_local_reserved_ports);
1764         goto out;
1765 }
1766
1767 fs_initcall(inet_init);
1768
1769 /* ------------------------------------------------------------------------ */
1770
1771 #ifdef CONFIG_PROC_FS
1772 static int __init ipv4_proc_init(void)
1773 {
1774         int rc = 0;
1775
1776         if (raw_proc_init())
1777                 goto out_raw;
1778         if (tcp4_proc_init())
1779                 goto out_tcp;
1780         if (udp4_proc_init())
1781                 goto out_udp;
1782         if (ping_proc_init())
1783                 goto out_ping;
1784         if (ip_misc_proc_init())
1785                 goto out_misc;
1786 out:
1787         return rc;
1788 out_misc:
1789         ping_proc_exit();
1790 out_ping:
1791         udp4_proc_exit();
1792 out_udp:
1793         tcp4_proc_exit();
1794 out_tcp:
1795         raw_proc_exit();
1796 out_raw:
1797         rc = -ENOMEM;
1798         goto out;
1799 }
1800
1801 #else /* CONFIG_PROC_FS */
1802 static int __init ipv4_proc_init(void)
1803 {
1804         return 0;
1805 }
1806 #endif /* CONFIG_PROC_FS */
1807
1808 MODULE_ALIAS_NETPROTO(PF_INET);
1809