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 #ifdef CONFIG_ANDROID_PARANOID_NETWORK
 122 #include <linux/android_aid.h>
 123
 124 static inline int current_has_network(void)
 125 {
 126         return in_egroup_p(AID_INET) || capable(CAP_NET_RAW);
 127 }
 128 #else
 129 static inline int current_has_network(void)
 130 {
 131         return 1;
 132 }
 133 #endif
 134
 135 /* The inetsw table contains everything that inet_create needs to
 136  * build a new socket.
 137  */
 138 static struct list_head inetsw[SOCK_MAX];
 139 static DEFINE_SPINLOCK(inetsw_lock);
 140
 141 struct ipv4_config ipv4_config;
 142 EXPORT_SYMBOL(ipv4_config);
 143
 144 /* New destruction routine */
 145
 146 void inet_sock_destruct(struct sock *sk)
 147 {
 148         struct inet_sock *inet = inet_sk(sk);
 149
 150         __skb_queue_purge(&sk->sk_receive_queue);
 151         __skb_queue_purge(&sk->sk_error_queue);
 152
 153         sk_mem_reclaim(sk);
 154
 155         if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
 156                 pr_err("Attempt to release TCP socket in state %d %p\n",
 157                        sk->sk_state, sk);
 158                 return;
 159         }
 160         if (!sock_flag(sk, SOCK_DEAD)) {
 161                 pr_err("Attempt to release alive inet socket %p\n", sk);
 162                 return;
 163         }
 164
 165         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
 166         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
 167         WARN_ON(sk->sk_wmem_queued);
 168         WARN_ON(sk->sk_forward_alloc);
 169
 170         kfree(rcu_dereference_protected(inet->inet_opt, 1));
 171         dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
 172         sk_refcnt_debug_dec(sk);
 173 }
 174 EXPORT_SYMBOL(inet_sock_destruct);
 175
 176 /*
 177  *      The routines beyond this point handle the behaviour of an AF_INET
 178  *      socket object. Mostly it punts to the subprotocols of IP to do
 179  *      the work.
 180  */
 181
 182 /*
 183  *      Automatically bind an unbound socket.
 184  */
 185
 186 static int inet_autobind(struct sock *sk)
 187 {
 188         struct inet_sock *inet;
 189         /* We may need to bind the socket. */
 190         lock_sock(sk);
 191         inet = inet_sk(sk);
 192         if (!inet->inet_num) {
 193                 if (sk->sk_prot->get_port(sk, 0)) {
 194                         release_sock(sk);
 195                         return -EAGAIN;
 196                 }
 197                 inet->inet_sport = htons(inet->inet_num);
 198         }
 199         release_sock(sk);
 200         return 0;
 201 }
 202
 203 /*
 204  *      Move a socket into listening state.
 205  */
 206 int inet_listen(struct socket *sock, int backlog)
 207 {
 208         struct sock *sk = sock->sk;
 209         unsigned char old_state;
 210         int err;
 211
 212         lock_sock(sk);
 213
 214         err = -EINVAL;
 215         if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
 216                 goto out;
 217
 218         old_state = sk->sk_state;
 219         if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
 220                 goto out;
 221
 222         /* Really, if the socket is already in listen state
 223          * we can only allow the backlog to be adjusted.
 224          */
 225         if (old_state != TCP_LISTEN) {
 226                 err = inet_csk_listen_start(sk, backlog);
 227                 if (err)
 228                         goto out;
 229         }
 230         sk->sk_max_ack_backlog = backlog;
 231         err = 0;
 232
 233 out:
 234         release_sock(sk);
 235         return err;
 236 }
 237 EXPORT_SYMBOL(inet_listen);
 238
 239 u32 inet_ehash_secret __read_mostly;
 240 EXPORT_SYMBOL(inet_ehash_secret);
 241
 242 /*
 243  * inet_ehash_secret must be set exactly once
 244  */
 245 void build_ehash_secret(void)
 246 {
 247         u32 rnd;
 248
 249         do {
 250                 get_random_bytes(&rnd, sizeof(rnd));
 251         } while (rnd == 0);
 252
 253         cmpxchg(&inet_ehash_secret, 0, rnd);
 254 }
 255 EXPORT_SYMBOL(build_ehash_secret);
 256
 257 static inline int inet_netns_ok(struct net *net, int protocol)
 258 {
 259         int hash;
 260         const struct net_protocol *ipprot;
 261
 262         if (net_eq(net, &init_net))
 263                 return 1;
 264
 265         hash = protocol & (MAX_INET_PROTOS - 1);
 266         ipprot = rcu_dereference(inet_protos[hash]);
 267
 268         if (ipprot == NULL)
 269                 /* raw IP is OK */
 270                 return 1;
 271         return ipprot->netns_ok;
 272 }
 273
 274
 275 /*
 276  *      Create an inet socket.
 277  */
 278
 279 static int inet_create(struct net *net, struct socket *sock, int protocol,
 280                        int kern)
 281 {
 282         struct sock *sk;
 283         struct inet_protosw *answer;
 284         struct inet_sock *inet;
 285         struct proto *answer_prot;
 286         unsigned char answer_flags;
 287         char answer_no_check;
 288         int try_loading_module = 0;
 289         int err;
 290
 291         if (!current_has_network())
 292                 return -EACCES;
 293
 294         if (unlikely(!inet_ehash_secret))
 295                 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
 296                         build_ehash_secret();
 297
 298         sock->state = SS_UNCONNECTED;
 299
 300         /* Look for the requested type/protocol pair. */
 301 lookup_protocol:
 302         err = -ESOCKTNOSUPPORT;
 303         rcu_read_lock();
 304         list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
 305
 306                 err = 0;
 307                 /* Check the non-wild match. */
 308                 if (protocol == answer->protocol) {
 309                         if (protocol != IPPROTO_IP)
 310                                 break;
 311                 } else {
 312                         /* Check for the two wild cases. */
 313                         if (IPPROTO_IP == protocol) {
 314                                 protocol = answer->protocol;
 315                                 break;
 316                         }
 317                         if (IPPROTO_IP == answer->protocol)
 318                                 break;
 319                 }
 320                 err = -EPROTONOSUPPORT;
 321         }
 322
 323         if (unlikely(err)) {
 324                 if (try_loading_module < 2) {
 325                         rcu_read_unlock();
 326                         /*
 327                          * Be more specific, e.g. net-pf-2-proto-132-type-1
 328                          * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
 329                          */
 330                         if (++try_loading_module == 1)
 331                                 request_module("net-pf-%d-proto-%d-type-%d",
 332                                                PF_INET, protocol, sock->type);
 333                         /*
 334                          * Fall back to generic, e.g. net-pf-2-proto-132
 335                          * (net-pf-PF_INET-proto-IPPROTO_SCTP)
 336                          */
 337                         else
 338                                 request_module("net-pf-%d-proto-%d",
 339                                                PF_INET, protocol);
 340                         goto lookup_protocol;
 341                 } else
 342                         goto out_rcu_unlock;
 343         }
 344
 345         err = -EPERM;
 346         if (sock->type == SOCK_RAW && !kern && !capable(CAP_NET_RAW))
 347                 goto out_rcu_unlock;
 348
 349         err = -EAFNOSUPPORT;
 350         if (!inet_netns_ok(net, protocol))
 351                 goto out_rcu_unlock;
 352
 353         sock->ops = answer->ops;
 354         answer_prot = answer->prot;
 355         answer_no_check = answer->no_check;
 356         answer_flags = answer->flags;
 357         rcu_read_unlock();
 358
 359         WARN_ON(answer_prot->slab == NULL);
 360
 361         err = -ENOBUFS;
 362         sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
 363         if (sk == NULL)
 364                 goto out;
 365
 366         err = 0;
 367         sk->sk_no_check = answer_no_check;
 368         if (INET_PROTOSW_REUSE & answer_flags)
 369                 sk->sk_reuse = 1;
 370
 371         inet = inet_sk(sk);
 372         inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
 373
 374         inet->nodefrag = 0;
 375
 376         if (SOCK_RAW == sock->type) {
 377                 inet->inet_num = protocol;
 378                 if (IPPROTO_RAW == protocol)
 379                         inet->hdrincl = 1;
 380         }
 381
 382         if (ipv4_config.no_pmtu_disc)
 383                 inet->pmtudisc = IP_PMTUDISC_DONT;
 384         else
 385                 inet->pmtudisc = IP_PMTUDISC_WANT;
 386
 387         inet->inet_id = 0;
 388
 389         sock_init_data(sock, sk);
 390
 391         sk->sk_destruct    = inet_sock_destruct;
 392         sk->sk_protocol    = protocol;
 393         sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
 394
 395         inet->uc_ttl    = -1;
 396         inet->mc_loop   = 1;
 397         inet->mc_ttl    = 1;
 398         inet->mc_all    = 1;
 399         inet->mc_index  = 0;
 400         inet->mc_list   = NULL;
 401
 402         sk_refcnt_debug_inc(sk);
 403
 404         if (inet->inet_num) {
 405                 /* It assumes that any protocol which allows
 406                  * the user to assign a number at socket
 407                  * creation time automatically
 408                  * shares.
 409                  */
 410                 inet->inet_sport = htons(inet->inet_num);
 411                 /* Add to protocol hash chains. */
 412                 sk->sk_prot->hash(sk);
 413         }
 414
 415         if (sk->sk_prot->init) {
 416                 err = sk->sk_prot->init(sk);
 417                 if (err)
 418                         sk_common_release(sk);
 419         }
 420 out:
 421         return err;
 422 out_rcu_unlock:
 423         rcu_read_unlock();
 424         goto out;
 425 }
 426
 427
 428 /*
 429  *      The peer socket should always be NULL (or else). When we call this
 430  *      function we are destroying the object and from then on nobody
 431  *      should refer to it.
 432  */
 433 int inet_release(struct socket *sock)
 434 {
 435         struct sock *sk = sock->sk;
 436
 437         if (sk) {
 438                 long timeout;
 439
 440                 sock_rps_reset_flow(sk);
 441
 442                 /* Applications forget to leave groups before exiting */
 443                 ip_mc_drop_socket(sk);
 444
 445                 /* If linger is set, we don't return until the close
 446                  * is complete.  Otherwise we return immediately. The
 447                  * actually closing is done the same either way.
 448                  *
 449                  * If the close is due to the process exiting, we never
 450                  * linger..
 451                  */
 452                 timeout = 0;
 453                 if (sock_flag(sk, SOCK_LINGER) &&
 454                     !(current->flags & PF_EXITING))
 455                         timeout = sk->sk_lingertime;
 456                 sock->sk = NULL;
 457                 sk->sk_prot->close(sk, timeout);
 458         }
 459         return 0;
 460 }
 461 EXPORT_SYMBOL(inet_release);
 462
 463 /* It is off by default, see below. */
 464 int sysctl_ip_nonlocal_bind __read_mostly;
 465 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
 466
 467 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 468 {
 469         struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
 470         struct sock *sk = sock->sk;
 471         struct inet_sock *inet = inet_sk(sk);
 472         unsigned short snum;
 473         int chk_addr_ret;
 474         int err;
 475
 476         /* If the socket has its own bind function then use it. (RAW) */
 477         if (sk->sk_prot->bind) {
 478                 err = sk->sk_prot->bind(sk, uaddr, addr_len);
 479                 goto out;
 480         }
 481         err = -EINVAL;
 482         if (addr_len < sizeof(struct sockaddr_in))
 483                 goto out;
 484
 485         if (addr->sin_family != AF_INET) {
 486                 err = -EAFNOSUPPORT;
 487                 goto out;
 488         }
 489
 490         chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr);
 491
 492         /* Not specified by any standard per-se, however it breaks too
 493          * many applications when removed.  It is unfortunate since
 494          * allowing applications to make a non-local bind solves
 495          * several problems with systems using dynamic addressing.
 496          * (ie. your servers still start up even if your ISDN link
 497          *  is temporarily down)
 498          */
 499         err = -EADDRNOTAVAIL;
 500         if (!sysctl_ip_nonlocal_bind &&
 501             !(inet->freebind || inet->transparent) &&
 502             addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
 503             chk_addr_ret != RTN_LOCAL &&
 504             chk_addr_ret != RTN_MULTICAST &&
 505             chk_addr_ret != RTN_BROADCAST)
 506                 goto out;
 507
 508         snum = ntohs(addr->sin_port);
 509         err = -EACCES;
 510         if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
 511                 goto out;
 512
 513         /*      We keep a pair of addresses. rcv_saddr is the one
 514          *      used by hash lookups, and saddr is used for transmit.
 515          *
 516          *      In the BSD API these are the same except where it
 517          *      would be illegal to use them (multicast/broadcast) in
 518          *      which case the sending device address is used.
 519          */
 520         lock_sock(sk);
 521
 522         /* Check these errors (active socket, double bind). */
 523         err = -EINVAL;
 524         if (sk->sk_state != TCP_CLOSE || inet->inet_num)
 525                 goto out_release_sock;
 526
 527         inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
 528         if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
 529                 inet->inet_saddr = 0;  /* Use device */
 530
 531         /* Make sure we are allowed to bind here. */
 532         if (sk->sk_prot->get_port(sk, snum)) {
 533                 inet->inet_saddr = inet->inet_rcv_saddr = 0;
 534                 err = -EADDRINUSE;
 535                 goto out_release_sock;
 536         }
 537
 538         if (inet->inet_rcv_saddr)
 539                 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
 540         if (snum)
 541                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
 542         inet->inet_sport = htons(inet->inet_num);
 543         inet->inet_daddr = 0;
 544         inet->inet_dport = 0;
 545         sk_dst_reset(sk);
 546         err = 0;
 547 out_release_sock:
 548         release_sock(sk);
 549 out:
 550         return err;
 551 }
 552 EXPORT_SYMBOL(inet_bind);
 553
 554 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
 555                        int addr_len, int flags)
 556 {
 557         struct sock *sk = sock->sk;
 558
 559         if (addr_len < sizeof(uaddr->sa_family))
 560                 return -EINVAL;
 561         if (uaddr->sa_family == AF_UNSPEC)
 562                 return sk->sk_prot->disconnect(sk, flags);
 563
 564         if (!inet_sk(sk)->inet_num && inet_autobind(sk))
 565                 return -EAGAIN;
 566         return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
 567 }
 568 EXPORT_SYMBOL(inet_dgram_connect);
 569
 570 static long inet_wait_for_connect(struct sock *sk, long timeo)
 571 {
 572         DEFINE_WAIT(wait);
 573
 574         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
 575
 576         /* Basic assumption: if someone sets sk->sk_err, he _must_
 577          * change state of the socket from TCP_SYN_*.
 578          * Connect() does not allow to get error notifications
 579          * without closing the socket.
 580          */
 581         while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 582                 release_sock(sk);
 583                 timeo = schedule_timeout(timeo);
 584                 lock_sock(sk);
 585                 if (signal_pending(current) || !timeo)
 586                         break;
 587                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
 588         }
 589         finish_wait(sk_sleep(sk), &wait);
 590         return timeo;
 591 }
 592
 593 /*
 594  *      Connect to a remote host. There is regrettably still a little
 595  *      TCP 'magic' in here.
 596  */
 597 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 598                         int addr_len, int flags)
 599 {
 600         struct sock *sk = sock->sk;
 601         int err;
 602         long timeo;
 603
 604         if (addr_len < sizeof(uaddr->sa_family))
 605                 return -EINVAL;
 606
 607         lock_sock(sk);
 608
 609         if (uaddr->sa_family == AF_UNSPEC) {
 610                 err = sk->sk_prot->disconnect(sk, flags);
 611                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 612                 goto out;
 613         }
 614
 615         switch (sock->state) {
 616         default:
 617                 err = -EINVAL;
 618                 goto out;
 619         case SS_CONNECTED:
 620                 err = -EISCONN;
 621                 goto out;
 622         case SS_CONNECTING:
 623                 err = -EALREADY;
 624                 /* Fall out of switch with err, set for this state */
 625                 break;
 626         case SS_UNCONNECTED:
 627                 err = -EISCONN;
 628                 if (sk->sk_state != TCP_CLOSE)
 629                         goto out;
 630
 631                 err = sk->sk_prot->connect(sk, uaddr, addr_len);
 632                 if (err < 0)
 633                         goto out;
 634
 635                 sock->state = SS_CONNECTING;
 636
 637                 /* Just entered SS_CONNECTING state; the only
 638                  * difference is that return value in non-blocking
 639                  * case is EINPROGRESS, rather than EALREADY.
 640                  */
 641                 err = -EINPROGRESS;
 642                 break;
 643         }
 644
 645         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
 646
 647         if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 648                 /* Error code is set above */
 649                 if (!timeo || !inet_wait_for_connect(sk, timeo))
 650                         goto out;
 651
 652                 err = sock_intr_errno(timeo);
 653                 if (signal_pending(current))
 654                         goto out;
 655         }
 656
 657         /* Connection was closed by RST, timeout, ICMP error
 658          * or another process disconnected us.
 659          */
 660         if (sk->sk_state == TCP_CLOSE)
 661                 goto sock_error;
 662
 663         /* sk->sk_err may be not zero now, if RECVERR was ordered by user
 664          * and error was received after socket entered established state.
 665          * Hence, it is handled normally after connect() return successfully.
 666          */
 667
 668         sock->state = SS_CONNECTED;
 669         err = 0;
 670 out:
 671         release_sock(sk);
 672         return err;
 673
 674 sock_error:
 675         err = sock_error(sk) ? : -ECONNABORTED;
 676         sock->state = SS_UNCONNECTED;
 677         if (sk->sk_prot->disconnect(sk, flags))
 678                 sock->state = SS_DISCONNECTING;
 679         goto out;
 680 }
 681 EXPORT_SYMBOL(inet_stream_connect);
 682
 683 /*
 684  *      Accept a pending connection. The TCP layer now gives BSD semantics.
 685  */
 686
 687 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
 688 {
 689         struct sock *sk1 = sock->sk;
 690         int err = -EINVAL;
 691         struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
 692
 693         if (!sk2)
 694                 goto do_err;
 695
 696         lock_sock(sk2);
 697
 698         sock_rps_record_flow(sk2);
 699         WARN_ON(!((1 << sk2->sk_state) &
 700                   (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE)));
 701
 702         sock_graft(sk2, newsock);
 703
 704         newsock->state = SS_CONNECTED;
 705         err = 0;
 706         release_sock(sk2);
 707 do_err:
 708         return err;
 709 }
 710 EXPORT_SYMBOL(inet_accept);
 711
 712
 713 /*
 714  *      This does both peername and sockname.
 715  */
 716 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
 717                         int *uaddr_len, int peer)
 718 {
 719         struct sock *sk         = sock->sk;
 720         struct inet_sock *inet  = inet_sk(sk);
 721         DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
 722
 723         sin->sin_family = AF_INET;
 724         if (peer) {
 725                 if (!inet->inet_dport ||
 726                     (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
 727                      peer == 1))
 728                         return -ENOTCONN;
 729                 sin->sin_port = inet->inet_dport;
 730                 sin->sin_addr.s_addr = inet->inet_daddr;
 731         } else {
 732                 __be32 addr = inet->inet_rcv_saddr;
 733                 if (!addr)
 734                         addr = inet->inet_saddr;
 735                 sin->sin_port = inet->inet_sport;
 736                 sin->sin_addr.s_addr = addr;
 737         }
 738         memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 739         *uaddr_len = sizeof(*sin);
 740         return 0;
 741 }
 742 EXPORT_SYMBOL(inet_getname);
 743
 744 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
 745                  size_t size)
 746 {
 747         struct sock *sk = sock->sk;
 748
 749         sock_rps_record_flow(sk);
 750
 751         /* We may need to bind the socket. */
 752         if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
 753             inet_autobind(sk))
 754                 return -EAGAIN;
 755
 756         return sk->sk_prot->sendmsg(iocb, sk, msg, size);
 757 }
 758 EXPORT_SYMBOL(inet_sendmsg);
 759
 760 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
 761                       size_t size, int flags)
 762 {
 763         struct sock *sk = sock->sk;
 764
 765         sock_rps_record_flow(sk);
 766
 767         /* We may need to bind the socket. */
 768         if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
 769             inet_autobind(sk))
 770                 return -EAGAIN;
 771
 772         if (sk->sk_prot->sendpage)
 773                 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
 774         return sock_no_sendpage(sock, page, offset, size, flags);
 775 }
 776 EXPORT_SYMBOL(inet_sendpage);
 777
 778 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
 779                  size_t size, int flags)
 780 {
 781         struct sock *sk = sock->sk;
 782         int addr_len = 0;
 783         int err;
 784
 785         sock_rps_record_flow(sk);
 786
 787         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
 788                                    flags & ~MSG_DONTWAIT, &addr_len);
 789         if (err >= 0)
 790                 msg->msg_namelen = addr_len;
 791         return err;
 792 }
 793 EXPORT_SYMBOL(inet_recvmsg);
 794
 795 int inet_shutdown(struct socket *sock, int how)
 796 {
 797         struct sock *sk = sock->sk;
 798         int err = 0;
 799
 800         /* This should really check to make sure
 801          * the socket is a TCP socket. (WHY AC...)
 802          */
 803         how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
 804                        1->2 bit 2 snds.
 805                        2->3 */
 806         if ((how & ~SHUTDOWN_MASK) || !how)     /* MAXINT->0 */
 807                 return -EINVAL;
 808
 809         lock_sock(sk);
 810         if (sock->state == SS_CONNECTING) {
 811                 if ((1 << sk->sk_state) &
 812                     (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
 813                         sock->state = SS_DISCONNECTING;
 814                 else
 815                         sock->state = SS_CONNECTED;
 816         }
 817
 818         switch (sk->sk_state) {
 819         case TCP_CLOSE:
 820                 err = -ENOTCONN;
 821                 /* Hack to wake up other listeners, who can poll for
 822                    POLLHUP, even on eg. unconnected UDP sockets -- RR */
 823         default:
 824                 sk->sk_shutdown |= how;
 825                 if (sk->sk_prot->shutdown)
 826                         sk->sk_prot->shutdown(sk, how);
 827                 break;
 828
 829         /* Remaining two branches are temporary solution for missing
 830          * close() in multithreaded environment. It is _not_ a good idea,
 831          * but we have no choice until close() is repaired at VFS level.
 832          */
 833         case TCP_LISTEN:
 834                 if (!(how & RCV_SHUTDOWN))
 835                         break;
 836                 /* Fall through */
 837         case TCP_SYN_SENT:
 838                 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
 839                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 840                 break;
 841         }
 842
 843         /* Wake up anyone sleeping in poll. */
 844         sk->sk_state_change(sk);
 845         release_sock(sk);
 846         return err;
 847 }
 848 EXPORT_SYMBOL(inet_shutdown);
 849
 850 /*
 851  *      ioctl() calls you can issue on an INET socket. Most of these are
 852  *      device configuration and stuff and very rarely used. Some ioctls
 853  *      pass on to the socket itself.
 854  *
 855  *      NOTE: I like the idea of a module for the config stuff. ie ifconfig
 856  *      loads the devconfigure module does its configuring and unloads it.
 857  *      There's a good 20K of config code hanging around the kernel.
 858  */
 859
 860 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 861 {
 862         struct sock *sk = sock->sk;
 863         int err = 0;
 864         struct net *net = sock_net(sk);
 865
 866         switch (cmd) {
 867         case SIOCGSTAMP:
 868                 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
 869                 break;
 870         case SIOCGSTAMPNS:
 871                 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
 872                 break;
 873         case SIOCADDRT:
 874         case SIOCDELRT:
 875         case SIOCRTMSG:
 876                 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
 877                 break;
 878         case SIOCDARP:
 879         case SIOCGARP:
 880         case SIOCSARP:
 881                 err = arp_ioctl(net, cmd, (void __user *)arg);
 882                 break;
 883         case SIOCGIFADDR:
 884         case SIOCSIFADDR:
 885         case SIOCGIFBRDADDR:
 886         case SIOCSIFBRDADDR:
 887         case SIOCGIFNETMASK:
 888         case SIOCSIFNETMASK:
 889         case SIOCGIFDSTADDR:
 890         case SIOCSIFDSTADDR:
 891         case SIOCSIFPFLAGS:
 892         case SIOCGIFPFLAGS:
 893         case SIOCSIFFLAGS:
 894         case SIOCKILLADDR:
 895                 err = devinet_ioctl(net, cmd, (void __user *)arg);
 896                 break;
 897         default:
 898                 if (sk->sk_prot->ioctl)
 899                         err = sk->sk_prot->ioctl(sk, cmd, arg);
 900                 else
 901                         err = -ENOIOCTLCMD;
 902                 break;
 903         }
 904         return err;
 905 }
 906 EXPORT_SYMBOL(inet_ioctl);
 907
 908 #ifdef CONFIG_COMPAT
 909 int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 910 {
 911         struct sock *sk = sock->sk;
 912         int err = -ENOIOCTLCMD;
 913
 914         if (sk->sk_prot->compat_ioctl)
 915                 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
 916
 917         return err;
 918 }
 919 #endif
 920
 921 const struct proto_ops inet_stream_ops = {
 922         .family            = PF_INET,
 923         .owner             = THIS_MODULE,
 924         .release           = inet_release,
 925         .bind              = inet_bind,
 926         .connect           = inet_stream_connect,
 927         .socketpair        = sock_no_socketpair,
 928         .accept            = inet_accept,
 929         .getname           = inet_getname,
 930         .poll              = tcp_poll,
 931         .ioctl             = inet_ioctl,
 932         .listen            = inet_listen,
 933         .shutdown          = inet_shutdown,
 934         .setsockopt        = sock_common_setsockopt,
 935         .getsockopt        = sock_common_getsockopt,
 936         .sendmsg           = inet_sendmsg,
 937         .recvmsg           = inet_recvmsg,
 938         .mmap              = sock_no_mmap,
 939         .sendpage          = inet_sendpage,
 940         .splice_read       = tcp_splice_read,
 941 #ifdef CONFIG_COMPAT
 942         .compat_setsockopt = compat_sock_common_setsockopt,
 943         .compat_getsockopt = compat_sock_common_getsockopt,
 944         .compat_ioctl      = inet_compat_ioctl,
 945 #endif
 946 };
 947 EXPORT_SYMBOL(inet_stream_ops);
 948
 949 const struct proto_ops inet_dgram_ops = {
 950         .family            = PF_INET,
 951         .owner             = THIS_MODULE,
 952         .release           = inet_release,
 953         .bind              = inet_bind,
 954         .connect           = inet_dgram_connect,
 955         .socketpair        = sock_no_socketpair,
 956         .accept            = sock_no_accept,
 957         .getname           = inet_getname,
 958         .poll              = udp_poll,
 959         .ioctl             = inet_ioctl,
 960         .listen            = sock_no_listen,
 961         .shutdown          = inet_shutdown,
 962         .setsockopt        = sock_common_setsockopt,
 963         .getsockopt        = sock_common_getsockopt,
 964         .sendmsg           = inet_sendmsg,
 965         .recvmsg           = inet_recvmsg,
 966         .mmap              = sock_no_mmap,
 967         .sendpage          = inet_sendpage,
 968 #ifdef CONFIG_COMPAT
 969         .compat_setsockopt = compat_sock_common_setsockopt,
 970         .compat_getsockopt = compat_sock_common_getsockopt,
 971         .compat_ioctl      = inet_compat_ioctl,
 972 #endif
 973 };
 974 EXPORT_SYMBOL(inet_dgram_ops);
 975
 976 /*
 977  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
 978  * udp_poll
 979  */
 980 static const struct proto_ops inet_sockraw_ops = {
 981         .family            = PF_INET,
 982         .owner             = THIS_MODULE,
 983         .release           = inet_release,
 984         .bind              = inet_bind,
 985         .connect           = inet_dgram_connect,
 986         .socketpair        = sock_no_socketpair,
 987         .accept            = sock_no_accept,
 988         .getname           = inet_getname,
 989         .poll              = datagram_poll,
 990         .ioctl             = inet_ioctl,
 991         .listen            = sock_no_listen,
 992         .shutdown          = inet_shutdown,
 993         .setsockopt        = sock_common_setsockopt,
 994         .getsockopt        = sock_common_getsockopt,
 995         .sendmsg           = inet_sendmsg,
 996         .recvmsg           = inet_recvmsg,
 997         .mmap              = sock_no_mmap,
 998         .sendpage          = inet_sendpage,
 999 #ifdef CONFIG_COMPAT
1000         .compat_setsockopt = compat_sock_common_setsockopt,
1001         .compat_getsockopt = compat_sock_common_getsockopt,
1002         .compat_ioctl      = inet_compat_ioctl,
1003 #endif
1004 };
1005
1006 static const struct net_proto_family inet_family_ops = {
1007         .family = PF_INET,
1008         .create = inet_create,
1009         .owner  = THIS_MODULE,
1010 };
1011
1012 /* Upon startup we insert all the elements in inetsw_array[] into
1013  * the linked list inetsw.
1014  */
1015 static struct inet_protosw inetsw_array[] =
1016 {
1017         {
1018                 .type =       SOCK_STREAM,
1019                 .protocol =   IPPROTO_TCP,
1020                 .prot =       &tcp_prot,
1021                 .ops =        &inet_stream_ops,
1022                 .no_check =   0,
1023                 .flags =      INET_PROTOSW_PERMANENT |
1024                               INET_PROTOSW_ICSK,
1025         },
1026
1027         {
1028                 .type =       SOCK_DGRAM,
1029                 .protocol =   IPPROTO_UDP,
1030                 .prot =       &udp_prot,
1031                 .ops =        &inet_dgram_ops,
1032                 .no_check =   UDP_CSUM_DEFAULT,
1033                 .flags =      INET_PROTOSW_PERMANENT,
1034        },
1035
1036        {
1037                 .type =       SOCK_DGRAM,
1038                 .protocol =   IPPROTO_ICMP,
1039                 .prot =       &ping_prot,
1040                 .ops =        &inet_dgram_ops,
1041                 .no_check =   UDP_CSUM_DEFAULT,
1042                 .flags =      INET_PROTOSW_REUSE,
1043        },
1044
1045        {
1046                .type =       SOCK_RAW,
1047                .protocol =   IPPROTO_IP,        /* wild card */
1048                .prot =       &raw_prot,
1049                .ops =        &inet_sockraw_ops,
1050                .no_check =   UDP_CSUM_DEFAULT,
1051                .flags =      INET_PROTOSW_REUSE,
1052        }
1053 };
1054
1055 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1056
1057 void inet_register_protosw(struct inet_protosw *p)
1058 {
1059         struct list_head *lh;
1060         struct inet_protosw *answer;
1061         int protocol = p->protocol;
1062         struct list_head *last_perm;
1063
1064         spin_lock_bh(&inetsw_lock);
1065
1066         if (p->type >= SOCK_MAX)
1067                 goto out_illegal;
1068
1069         /* If we are trying to override a permanent protocol, bail. */
1070         answer = NULL;
1071         last_perm = &inetsw[p->type];
1072         list_for_each(lh, &inetsw[p->type]) {
1073                 answer = list_entry(lh, struct inet_protosw, list);
1074
1075                 /* Check only the non-wild match. */
1076                 if (INET_PROTOSW_PERMANENT & answer->flags) {
1077                         if (protocol == answer->protocol)
1078                                 break;
1079                         last_perm = lh;
1080                 }
1081
1082                 answer = NULL;
1083         }
1084         if (answer)
1085                 goto out_permanent;
1086
1087         /* Add the new entry after the last permanent entry if any, so that
1088          * the new entry does not override a permanent entry when matched with
1089          * a wild-card protocol. But it is allowed to override any existing
1090          * non-permanent entry.  This means that when we remove this entry, the
1091          * system automatically returns to the old behavior.
1092          */
1093         list_add_rcu(&p->list, last_perm);
1094 out:
1095         spin_unlock_bh(&inetsw_lock);
1096
1097         return;
1098
1099 out_permanent:
1100         printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
1101                protocol);
1102         goto out;
1103
1104 out_illegal:
1105         printk(KERN_ERR
1106                "Ignoring attempt to register invalid socket type %d.\n",
1107                p->type);
1108         goto out;
1109 }
1110 EXPORT_SYMBOL(inet_register_protosw);
1111
1112 void inet_unregister_protosw(struct inet_protosw *p)
1113 {
1114         if (INET_PROTOSW_PERMANENT & p->flags) {
1115                 printk(KERN_ERR
1116                        "Attempt to unregister permanent protocol %d.\n",
1117                        p->protocol);
1118         } else {
1119                 spin_lock_bh(&inetsw_lock);
1120                 list_del_rcu(&p->list);
1121                 spin_unlock_bh(&inetsw_lock);
1122
1123                 synchronize_net();
1124         }
1125 }
1126 EXPORT_SYMBOL(inet_unregister_protosw);
1127
1128 /*
1129  *      Shall we try to damage output packets if routing dev changes?
1130  */
1131
1132 int sysctl_ip_dynaddr __read_mostly;
1133
1134 static int inet_sk_reselect_saddr(struct sock *sk)
1135 {
1136         struct inet_sock *inet = inet_sk(sk);
1137         __be32 old_saddr = inet->inet_saddr;
1138         __be32 daddr = inet->inet_daddr;
1139         struct flowi4 *fl4;
1140         struct rtable *rt;
1141         __be32 new_saddr;
1142         struct ip_options_rcu *inet_opt;
1143
1144         inet_opt = rcu_dereference_protected(inet->inet_opt,
1145                                              sock_owned_by_user(sk));
1146         if (inet_opt && inet_opt->opt.srr)
1147                 daddr = inet_opt->opt.faddr;
1148
1149         /* Query new route. */
1150         fl4 = &inet->cork.fl.u.ip4;
1151         rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1152                               sk->sk_bound_dev_if, sk->sk_protocol,
1153                               inet->inet_sport, inet->inet_dport, sk, false);
1154         if (IS_ERR(rt))
1155                 return PTR_ERR(rt);
1156
1157         sk_setup_caps(sk, &rt->dst);
1158
1159         new_saddr = fl4->saddr;
1160
1161         if (new_saddr == old_saddr)
1162                 return 0;
1163
1164         if (sysctl_ip_dynaddr > 1) {
1165                 printk(KERN_INFO "%s(): shifting inet->saddr from %pI4 to %pI4\n",
1166                        __func__, &old_saddr, &new_saddr);
1167         }
1168
1169         inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1170
1171         /*
1172          * XXX The only one ugly spot where we need to
1173          * XXX really change the sockets identity after
1174          * XXX it has entered the hashes. -DaveM
1175          *
1176          * Besides that, it does not check for connection
1177          * uniqueness. Wait for troubles.
1178          */
1179         __sk_prot_rehash(sk);
1180         return 0;
1181 }
1182
1183 int inet_sk_rebuild_header(struct sock *sk)
1184 {
1185         struct inet_sock *inet = inet_sk(sk);
1186         struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1187         __be32 daddr;
1188         struct ip_options_rcu *inet_opt;
1189         struct flowi4 *fl4;
1190         int err;
1191
1192         /* Route is OK, nothing to do. */
1193         if (rt)
1194                 return 0;
1195
1196         /* Reroute. */
1197         rcu_read_lock();
1198         inet_opt = rcu_dereference(inet->inet_opt);
1199         daddr = inet->inet_daddr;
1200         if (inet_opt && inet_opt->opt.srr)
1201                 daddr = inet_opt->opt.faddr;
1202         rcu_read_unlock();
1203         fl4 = &inet->cork.fl.u.ip4;
1204         rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1205                                    inet->inet_dport, inet->inet_sport,
1206                                    sk->sk_protocol, RT_CONN_FLAGS(sk),
1207                                    sk->sk_bound_dev_if);
1208         if (!IS_ERR(rt)) {
1209                 err = 0;
1210                 sk_setup_caps(sk, &rt->dst);
1211         } else {
1212                 err = PTR_ERR(rt);
1213
1214                 /* Routing failed... */
1215                 sk->sk_route_caps = 0;
1216                 /*
1217                  * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1218                  * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1219                  */
1220                 if (!sysctl_ip_dynaddr ||
1221                     sk->sk_state != TCP_SYN_SENT ||
1222                     (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1223                     (err = inet_sk_reselect_saddr(sk)) != 0)
1224                         sk->sk_err_soft = -err;
1225         }
1226
1227         return err;
1228 }
1229 EXPORT_SYMBOL(inet_sk_rebuild_header);
1230
1231 static int inet_gso_send_check(struct sk_buff *skb)
1232 {
1233         const struct iphdr *iph;
1234         const struct net_protocol *ops;
1235         int proto;
1236         int ihl;
1237         int err = -EINVAL;
1238
1239         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1240                 goto out;
1241
1242         iph = ip_hdr(skb);
1243         ihl = iph->ihl * 4;
1244         if (ihl < sizeof(*iph))
1245                 goto out;
1246
1247         if (unlikely(!pskb_may_pull(skb, ihl)))
1248                 goto out;
1249
1250         __skb_pull(skb, ihl);
1251         skb_reset_transport_header(skb);
1252         iph = ip_hdr(skb);
1253         proto = iph->protocol & (MAX_INET_PROTOS - 1);
1254         err = -EPROTONOSUPPORT;
1255
1256         rcu_read_lock();
1257         ops = rcu_dereference(inet_protos[proto]);
1258         if (likely(ops && ops->gso_send_check))
1259                 err = ops->gso_send_check(skb);
1260         rcu_read_unlock();
1261
1262 out:
1263         return err;
1264 }
1265
1266 static struct sk_buff *inet_gso_segment(struct sk_buff *skb, u32 features)
1267 {
1268         struct sk_buff *segs = ERR_PTR(-EINVAL);
1269         struct iphdr *iph;
1270         const struct net_protocol *ops;
1271         int proto;
1272         int ihl;
1273         int id;
1274         unsigned int offset = 0;
1275
1276         if (!(features & NETIF_F_V4_CSUM))
1277                 features &= ~NETIF_F_SG;
1278
1279         if (unlikely(skb_shinfo(skb)->gso_type &
1280                      ~(SKB_GSO_TCPV4 |
1281                        SKB_GSO_UDP |
1282                        SKB_GSO_DODGY |
1283                        SKB_GSO_TCP_ECN |
1284                        0)))
1285                 goto out;
1286
1287         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1288                 goto out;
1289
1290         iph = ip_hdr(skb);
1291         ihl = iph->ihl * 4;
1292         if (ihl < sizeof(*iph))
1293                 goto out;
1294
1295         if (unlikely(!pskb_may_pull(skb, ihl)))
1296                 goto out;
1297
1298         __skb_pull(skb, ihl);
1299         skb_reset_transport_header(skb);
1300         iph = ip_hdr(skb);
1301         id = ntohs(iph->id);
1302         proto = iph->protocol & (MAX_INET_PROTOS - 1);
1303         segs = ERR_PTR(-EPROTONOSUPPORT);
1304
1305         rcu_read_lock();
1306         ops = rcu_dereference(inet_protos[proto]);
1307         if (likely(ops && ops->gso_segment))
1308                 segs = ops->gso_segment(skb, features);
1309         rcu_read_unlock();
1310
1311         if (!segs || IS_ERR(segs))
1312                 goto out;
1313
1314         skb = segs;
1315         do {
1316                 iph = ip_hdr(skb);
1317                 if (proto == IPPROTO_UDP) {
1318                         iph->id = htons(id);
1319                         iph->frag_off = htons(offset >> 3);
1320                         if (skb->next != NULL)
1321                                 iph->frag_off |= htons(IP_MF);
1322                         offset += (skb->len - skb->mac_len - iph->ihl * 4);
1323                 } else
1324                         iph->id = htons(id++);
1325                 iph->tot_len = htons(skb->len - skb->mac_len);
1326                 iph->check = 0;
1327                 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1328         } while ((skb = skb->next));
1329
1330 out:
1331         return segs;
1332 }
1333
1334 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1335                                          struct sk_buff *skb)
1336 {
1337         const struct net_protocol *ops;
1338         struct sk_buff **pp = NULL;
1339         struct sk_buff *p;
1340         const struct iphdr *iph;
1341         unsigned int hlen;
1342         unsigned int off;
1343         unsigned int id;
1344         int flush = 1;
1345         int proto;
1346
1347         off = skb_gro_offset(skb);
1348         hlen = off + sizeof(*iph);
1349         iph = skb_gro_header_fast(skb, off);
1350         if (skb_gro_header_hard(skb, hlen)) {
1351                 iph = skb_gro_header_slow(skb, hlen, off);
1352                 if (unlikely(!iph))
1353                         goto out;
1354         }
1355
1356         proto = iph->protocol & (MAX_INET_PROTOS - 1);
1357
1358         rcu_read_lock();
1359         ops = rcu_dereference(inet_protos[proto]);
1360         if (!ops || !ops->gro_receive)
1361                 goto out_unlock;
1362
1363         if (*(u8 *)iph != 0x45)
1364                 goto out_unlock;
1365
1366         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1367                 goto out_unlock;
1368
1369         id = ntohl(*(__be32 *)&iph->id);
1370         flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF));
1371         id >>= 16;
1372
1373         for (p = *head; p; p = p->next) {
1374                 struct iphdr *iph2;
1375
1376                 if (!NAPI_GRO_CB(p)->same_flow)
1377                         continue;
1378
1379                 iph2 = ip_hdr(p);
1380
1381                 if ((iph->protocol ^ iph2->protocol) |
1382                     (iph->tos ^ iph2->tos) |
1383                     ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1384                     ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1385                         NAPI_GRO_CB(p)->same_flow = 0;
1386                         continue;
1387                 }
1388
1389                 /* All fields must match except length and checksum. */
1390                 NAPI_GRO_CB(p)->flush |=
1391                         (iph->ttl ^ iph2->ttl) |
1392                         ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1393
1394                 NAPI_GRO_CB(p)->flush |= flush;
1395         }
1396
1397         NAPI_GRO_CB(skb)->flush |= flush;
1398         skb_gro_pull(skb, sizeof(*iph));
1399         skb_set_transport_header(skb, skb_gro_offset(skb));
1400
1401         pp = ops->gro_receive(head, skb);
1402
1403 out_unlock:
1404         rcu_read_unlock();
1405
1406 out:
1407         NAPI_GRO_CB(skb)->flush |= flush;
1408
1409         return pp;
1410 }
1411
1412 static int inet_gro_complete(struct sk_buff *skb)
1413 {
1414         const struct net_protocol *ops;
1415         struct iphdr *iph = ip_hdr(skb);
1416         int proto = iph->protocol & (MAX_INET_PROTOS - 1);
1417         int err = -ENOSYS;
1418         __be16 newlen = htons(skb->len - skb_network_offset(skb));
1419
1420         csum_replace2(&iph->check, iph->tot_len, newlen);
1421         iph->tot_len = newlen;
1422
1423         rcu_read_lock();
1424         ops = rcu_dereference(inet_protos[proto]);
1425         if (WARN_ON(!ops || !ops->gro_complete))
1426                 goto out_unlock;
1427
1428         err = ops->gro_complete(skb);
1429
1430 out_unlock:
1431         rcu_read_unlock();
1432
1433         return err;
1434 }
1435
1436 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1437                          unsigned short type, unsigned char protocol,
1438                          struct net *net)
1439 {
1440         struct socket *sock;
1441         int rc = sock_create_kern(family, type, protocol, &sock);
1442
1443         if (rc == 0) {
1444                 *sk = sock->sk;
1445                 (*sk)->sk_allocation = GFP_ATOMIC;
1446                 /*
1447                  * Unhash it so that IP input processing does not even see it,
1448                  * we do not wish this socket to see incoming packets.
1449                  */
1450                 (*sk)->sk_prot->unhash(*sk);
1451
1452                 sk_change_net(*sk, net);
1453         }
1454         return rc;
1455 }
1456 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1457
1458 unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1459 {
1460         unsigned long res = 0;
1461         int i;
1462
1463         for_each_possible_cpu(i) {
1464                 res += *(((unsigned long *) per_cpu_ptr(mib[0], i)) + offt);
1465                 res += *(((unsigned long *) per_cpu_ptr(mib[1], i)) + offt);
1466         }
1467         return res;
1468 }
1469 EXPORT_SYMBOL_GPL(snmp_fold_field);
1470
1471 #if BITS_PER_LONG==32
1472
1473 u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset)
1474 {
1475         u64 res = 0;
1476         int cpu;
1477
1478         for_each_possible_cpu(cpu) {
1479                 void *bhptr, *userptr;
1480                 struct u64_stats_sync *syncp;
1481                 u64 v_bh, v_user;
1482                 unsigned int start;
1483
1484                 /* first mib used by softirq context, we must use _bh() accessors */
1485                 bhptr = per_cpu_ptr(SNMP_STAT_BHPTR(mib), cpu);
1486                 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1487                 do {
1488                         start = u64_stats_fetch_begin_bh(syncp);
1489                         v_bh = *(((u64 *) bhptr) + offt);
1490                 } while (u64_stats_fetch_retry_bh(syncp, start));
1491
1492                 /* second mib used in USER context */
1493                 userptr = per_cpu_ptr(SNMP_STAT_USRPTR(mib), cpu);
1494                 syncp = (struct u64_stats_sync *)(userptr + syncp_offset);
1495                 do {
1496                         start = u64_stats_fetch_begin(syncp);
1497                         v_user = *(((u64 *) userptr) + offt);
1498                 } while (u64_stats_fetch_retry(syncp, start));
1499
1500                 res += v_bh + v_user;
1501         }
1502         return res;
1503 }
1504 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1505 #endif
1506
1507 int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align)
1508 {
1509         BUG_ON(ptr == NULL);
1510         ptr[0] = __alloc_percpu(mibsize, align);
1511         if (!ptr[0])
1512                 goto err0;
1513         ptr[1] = __alloc_percpu(mibsize, align);
1514         if (!ptr[1])
1515                 goto err1;
1516         return 0;
1517 err1:
1518         free_percpu(ptr[0]);
1519         ptr[0] = NULL;
1520 err0:
1521         return -ENOMEM;
1522 }
1523 EXPORT_SYMBOL_GPL(snmp_mib_init);
1524
1525 void snmp_mib_free(void __percpu *ptr[2])
1526 {
1527         BUG_ON(ptr == NULL);
1528         free_percpu(ptr[0]);
1529         free_percpu(ptr[1]);
1530         ptr[0] = ptr[1] = NULL;
1531 }
1532 EXPORT_SYMBOL_GPL(snmp_mib_free);
1533
1534 #ifdef CONFIG_IP_MULTICAST
1535 static const struct net_protocol igmp_protocol = {
1536         .handler =      igmp_rcv,
1537         .netns_ok =     1,
1538 };
1539 #endif
1540
1541 static const struct net_protocol tcp_protocol = {
1542         .handler =      tcp_v4_rcv,
1543         .err_handler =  tcp_v4_err,
1544         .gso_send_check = tcp_v4_gso_send_check,
1545         .gso_segment =  tcp_tso_segment,
1546         .gro_receive =  tcp4_gro_receive,
1547         .gro_complete = tcp4_gro_complete,
1548         .no_policy =    1,
1549         .netns_ok =     1,
1550 };
1551
1552 static const struct net_protocol udp_protocol = {
1553         .handler =      udp_rcv,
1554         .err_handler =  udp_err,
1555         .gso_send_check = udp4_ufo_send_check,
1556         .gso_segment = udp4_ufo_fragment,
1557         .no_policy =    1,
1558         .netns_ok =     1,
1559 };
1560
1561 static const struct net_protocol icmp_protocol = {
1562         .handler =      icmp_rcv,
1563         .err_handler =  ping_err,
1564         .no_policy =    1,
1565         .netns_ok =     1,
1566 };
1567
1568 static __net_init int ipv4_mib_init_net(struct net *net)
1569 {
1570         if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1571                           sizeof(struct tcp_mib),
1572                           __alignof__(struct tcp_mib)) < 0)
1573                 goto err_tcp_mib;
1574         if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1575                           sizeof(struct ipstats_mib),
1576                           __alignof__(struct ipstats_mib)) < 0)
1577                 goto err_ip_mib;
1578         if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1579                           sizeof(struct linux_mib),
1580                           __alignof__(struct linux_mib)) < 0)
1581                 goto err_net_mib;
1582         if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1583                           sizeof(struct udp_mib),
1584                           __alignof__(struct udp_mib)) < 0)
1585                 goto err_udp_mib;
1586         if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1587                           sizeof(struct udp_mib),
1588                           __alignof__(struct udp_mib)) < 0)
1589                 goto err_udplite_mib;
1590         if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1591                           sizeof(struct icmp_mib),
1592                           __alignof__(struct icmp_mib)) < 0)
1593                 goto err_icmp_mib;
1594         if (snmp_mib_init((void __percpu **)net->mib.icmpmsg_statistics,
1595                           sizeof(struct icmpmsg_mib),
1596                           __alignof__(struct icmpmsg_mib)) < 0)
1597                 goto err_icmpmsg_mib;
1598
1599         tcp_mib_init(net);
1600         return 0;
1601
1602 err_icmpmsg_mib:
1603         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1604 err_icmp_mib:
1605         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1606 err_udplite_mib:
1607         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1608 err_udp_mib:
1609         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1610 err_net_mib:
1611         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1612 err_ip_mib:
1613         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1614 err_tcp_mib:
1615         return -ENOMEM;
1616 }
1617
1618 static __net_exit void ipv4_mib_exit_net(struct net *net)
1619 {
1620         snmp_mib_free((void __percpu **)net->mib.icmpmsg_statistics);
1621         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1622         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1623         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1624         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1625         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1626         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1627 }
1628
1629 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1630         .init = ipv4_mib_init_net,
1631         .exit = ipv4_mib_exit_net,
1632 };
1633
1634 static int __init init_ipv4_mibs(void)
1635 {
1636         return register_pernet_subsys(&ipv4_mib_ops);
1637 }
1638
1639 static int ipv4_proc_init(void);
1640
1641 /*
1642  *      IP protocol layer initialiser
1643  */
1644
1645 static struct packet_type ip_packet_type __read_mostly = {
1646         .type = cpu_to_be16(ETH_P_IP),
1647         .func = ip_rcv,
1648         .gso_send_check = inet_gso_send_check,
1649         .gso_segment = inet_gso_segment,
1650         .gro_receive = inet_gro_receive,
1651         .gro_complete = inet_gro_complete,
1652 };
1653
1654 static int __init inet_init(void)
1655 {
1656         struct sk_buff *dummy_skb;
1657         struct inet_protosw *q;
1658         struct list_head *r;
1659         int rc = -EINVAL;
1660
1661         BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1662
1663         sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL);
1664         if (!sysctl_local_reserved_ports)
1665                 goto out;
1666
1667         rc = proto_register(&tcp_prot, 1);
1668         if (rc)
1669                 goto out_free_reserved_ports;
1670
1671         rc = proto_register(&udp_prot, 1);
1672         if (rc)
1673                 goto out_unregister_tcp_proto;
1674
1675         rc = proto_register(&raw_prot, 1);
1676         if (rc)
1677                 goto out_unregister_udp_proto;
1678
1679         rc = proto_register(&ping_prot, 1);
1680         if (rc)
1681                 goto out_unregister_raw_proto;
1682
1683         /*
1684          *      Tell SOCKET that we are alive...
1685          */
1686
1687         (void)sock_register(&inet_family_ops);
1688
1689 #ifdef CONFIG_SYSCTL
1690         ip_static_sysctl_init();
1691 #endif
1692
1693         /*
1694          *      Add all the base protocols.
1695          */
1696
1697         if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1698                 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1699         if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1700                 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1701         if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1702                 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1703 #ifdef CONFIG_IP_MULTICAST
1704         if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1705                 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1706 #endif
1707
1708         /* Register the socket-side information for inet_create. */
1709         for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1710                 INIT_LIST_HEAD(r);
1711
1712         for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1713                 inet_register_protosw(q);
1714
1715         /*
1716          *      Set the ARP module up
1717          */
1718
1719         arp_init();
1720
1721         /*
1722          *      Set the IP module up
1723          */
1724
1725         ip_init();
1726
1727         tcp_v4_init();
1728
1729         /* Setup TCP slab cache for open requests. */
1730         tcp_init();
1731
1732         /* Setup UDP memory threshold */
1733         udp_init();
1734
1735         /* Add UDP-Lite (RFC 3828) */
1736         udplite4_register();
1737
1738         ping_init();
1739
1740         /*
1741          *      Set the ICMP layer up
1742          */
1743
1744         if (icmp_init() < 0)
1745                 panic("Failed to create the ICMP control socket.\n");
1746
1747         /*
1748          *      Initialise the multicast router
1749          */
1750 #if defined(CONFIG_IP_MROUTE)
1751         if (ip_mr_init())
1752                 printk(KERN_CRIT "inet_init: Cannot init ipv4 mroute\n");
1753 #endif
1754         /*
1755          *      Initialise per-cpu ipv4 mibs
1756          */
1757
1758         if (init_ipv4_mibs())
1759                 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n");
1760
1761         ipv4_proc_init();
1762
1763         ipfrag_init();
1764
1765         dev_add_pack(&ip_packet_type);
1766
1767         rc = 0;
1768 out:
1769         return rc;
1770 out_unregister_raw_proto:
1771         proto_unregister(&raw_prot);
1772 out_unregister_udp_proto:
1773         proto_unregister(&udp_prot);
1774 out_unregister_tcp_proto:
1775         proto_unregister(&tcp_prot);
1776 out_free_reserved_ports:
1777         kfree(sysctl_local_reserved_ports);
1778         goto out;
1779 }
1780
1781 fs_initcall(inet_init);
1782
1783 /* ------------------------------------------------------------------------ */
1784
1785 #ifdef CONFIG_PROC_FS
1786 static int __init ipv4_proc_init(void)
1787 {
1788         int rc = 0;
1789
1790         if (raw_proc_init())
1791                 goto out_raw;
1792         if (tcp4_proc_init())
1793                 goto out_tcp;
1794         if (udp4_proc_init())
1795                 goto out_udp;
1796         if (ping_proc_init())
1797                 goto out_ping;
1798         if (ip_misc_proc_init())
1799                 goto out_misc;
1800 out:
1801         return rc;
1802 out_misc:
1803         ping_proc_exit();
1804 out_ping:
1805         udp4_proc_exit();
1806 out_udp:
1807         tcp4_proc_exit();
1808 out_tcp:
1809         raw_proc_exit();
1810 out_raw:
1811         rc = -ENOMEM;
1812         goto out;
1813 }
1814
1815 #else /* CONFIG_PROC_FS */
1816 static int __init ipv4_proc_init(void)
1817 {
1818         return 0;
1819 }
1820 #endif /* CONFIG_PROC_FS */
1821
1822 MODULE_ALIAS_NETPROTO(PF_INET);
1823