Pileus Git - ~andy/linux/blob - net/ipv4/af_inet.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              PF_INET protocol family socket handler.
   7  *
   8  * Authors:     Ross Biro
   9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  10  *              Florian La Roche, <flla@stud.uni-sb.de>
  11  *              Alan Cox, <A.Cox@swansea.ac.uk>
  12  *
  13  * Changes (see also sock.c)
  14  *
  15  *              piggy,
  16  *              Karl Knutson    :       Socket protocol table
  17  *              A.N.Kuznetsov   :       Socket death error in accept().
  18  *              John Richardson :       Fix non blocking error in connect()
  19  *                                      so sockets that fail to connect
  20  *                                      don't return -EINPROGRESS.
  21  *              Alan Cox        :       Asynchronous I/O support
  22  *              Alan Cox        :       Keep correct socket pointer on sock
  23  *                                      structures
  24  *                                      when accept() ed
  25  *              Alan Cox        :       Semantics of SO_LINGER aren't state
  26  *                                      moved to close when you look carefully.
  27  *                                      With this fixed and the accept bug fixed
  28  *                                      some RPC stuff seems happier.
  29  *              Niibe Yutaka    :       4.4BSD style write async I/O
  30  *              Alan Cox,
  31  *              Tony Gale       :       Fixed reuse semantics.
  32  *              Alan Cox        :       bind() shouldn't abort existing but dead
  33  *                                      sockets. Stops FTP netin:.. I hope.
  34  *              Alan Cox        :       bind() works correctly for RAW sockets.
  35  *                                      Note that FreeBSD at least was broken
  36  *                                      in this respect so be careful with
  37  *                                      compatibility tests...
  38  *              Alan Cox        :       routing cache support
  39  *              Alan Cox        :       memzero the socket structure for
  40  *                                      compactness.
  41  *              Matt Day        :       nonblock connect error handler
  42  *              Alan Cox        :       Allow large numbers of pending sockets
  43  *                                      (eg for big web sites), but only if
  44  *                                      specifically application requested.
  45  *              Alan Cox        :       New buffering throughout IP. Used
  46  *                                      dumbly.
  47  *              Alan Cox        :       New buffering now used smartly.
  48  *              Alan Cox        :       BSD rather than common sense
  49  *                                      interpretation of listen.
  50  *              Germano Caronni :       Assorted small races.
  51  *              Alan Cox        :       sendmsg/recvmsg basic support.
  52  *              Alan Cox        :       Only sendmsg/recvmsg now supported.
  53  *              Alan Cox        :       Locked down bind (see security list).
  54  *              Alan Cox        :       Loosened bind a little.
  55  *              Mike McLagan    :       ADD/DEL DLCI Ioctls
  56  *      Willy Konynenberg       :       Transparent proxying support.
  57  *              David S. Miller :       New socket lookup architecture.
  58  *                                      Some other random speedups.
  59  *              Cyrus Durgin    :       Cleaned up file for kmod hacks.
  60  *              Andi Kleen      :       Fix inet_stream_connect TCP race.
  61  *
  62  *              This program is free software; you can redistribute it and/or
  63  *              modify it under the terms of the GNU General Public License
  64  *              as published by the Free Software Foundation; either version
  65  *              2 of the License, or (at your option) any later version.
  66  */
  67
  68 #define pr_fmt(fmt) "IPv4: " fmt
  69
  70 #include <linux/err.h>
  71 #include <linux/errno.h>
  72 #include <linux/types.h>
  73 #include <linux/socket.h>
  74 #include <linux/in.h>
  75 #include <linux/kernel.h>
  76 #include <linux/module.h>
  77 #include <linux/sched.h>
  78 #include <linux/timer.h>
  79 #include <linux/string.h>
  80 #include <linux/sockios.h>
  81 #include <linux/net.h>
  82 #include <linux/capability.h>
  83 #include <linux/fcntl.h>
  84 #include <linux/mm.h>
  85 #include <linux/interrupt.h>
  86 #include <linux/stat.h>
  87 #include <linux/init.h>
  88 #include <linux/poll.h>
  89 #include <linux/netfilter_ipv4.h>
  90 #include <linux/random.h>
  91 #include <linux/slab.h>
  92
  93 #include <asm/uaccess.h>
  94
  95 #include <linux/inet.h>
  96 #include <linux/igmp.h>
  97 #include <linux/inetdevice.h>
  98 #include <linux/netdevice.h>
  99 #include <net/checksum.h>
 100 #include <net/ip.h>
 101 #include <net/protocol.h>
 102 #include <net/arp.h>
 103 #include <net/route.h>
 104 #include <net/ip_fib.h>
 105 #include <net/inet_connection_sock.h>
 106 #include <net/tcp.h>
 107 #include <net/udp.h>
 108 #include <net/udplite.h>
 109 #include <net/ping.h>
 110 #include <linux/skbuff.h>
 111 #include <net/sock.h>
 112 #include <net/raw.h>
 113 #include <net/icmp.h>
 114 #include <net/inet_common.h>
 115 #include <net/xfrm.h>
 116 #include <net/net_namespace.h>
 117 #include <net/secure_seq.h>
 118 #ifdef CONFIG_IP_MROUTE
 119 #include <linux/mroute.h>
 120 #endif
 121
 122
 123 /* The inetsw table contains everything that inet_create needs to
 124  * build a new socket.
 125  */
 126 static struct list_head inetsw[SOCK_MAX];
 127 static DEFINE_SPINLOCK(inetsw_lock);
 128
 129 /* New destruction routine */
 130
 131 void inet_sock_destruct(struct sock *sk)
 132 {
 133         struct inet_sock *inet = inet_sk(sk);
 134
 135         __skb_queue_purge(&sk->sk_receive_queue);
 136         __skb_queue_purge(&sk->sk_error_queue);
 137
 138         sk_mem_reclaim(sk);
 139
 140         if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
 141                 pr_err("Attempt to release TCP socket in state %d %p\n",
 142                        sk->sk_state, sk);
 143                 return;
 144         }
 145         if (!sock_flag(sk, SOCK_DEAD)) {
 146                 pr_err("Attempt to release alive inet socket %p\n", sk);
 147                 return;
 148         }
 149
 150         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
 151         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
 152         WARN_ON(sk->sk_wmem_queued);
 153         WARN_ON(sk->sk_forward_alloc);
 154
 155         kfree(rcu_dereference_protected(inet->inet_opt, 1));
 156         dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
 157         dst_release(sk->sk_rx_dst);
 158         sk_refcnt_debug_dec(sk);
 159 }
 160 EXPORT_SYMBOL(inet_sock_destruct);
 161
 162 /*
 163  *      The routines beyond this point handle the behaviour of an AF_INET
 164  *      socket object. Mostly it punts to the subprotocols of IP to do
 165  *      the work.
 166  */
 167
 168 /*
 169  *      Automatically bind an unbound socket.
 170  */
 171
 172 static int inet_autobind(struct sock *sk)
 173 {
 174         struct inet_sock *inet;
 175         /* We may need to bind the socket. */
 176         lock_sock(sk);
 177         inet = inet_sk(sk);
 178         if (!inet->inet_num) {
 179                 if (sk->sk_prot->get_port(sk, 0)) {
 180                         release_sock(sk);
 181                         return -EAGAIN;
 182                 }
 183                 inet->inet_sport = htons(inet->inet_num);
 184         }
 185         release_sock(sk);
 186         return 0;
 187 }
 188
 189 /*
 190  *      Move a socket into listening state.
 191  */
 192 int inet_listen(struct socket *sock, int backlog)
 193 {
 194         struct sock *sk = sock->sk;
 195         unsigned char old_state;
 196         int err;
 197
 198         lock_sock(sk);
 199
 200         err = -EINVAL;
 201         if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
 202                 goto out;
 203
 204         old_state = sk->sk_state;
 205         if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
 206                 goto out;
 207
 208         /* Really, if the socket is already in listen state
 209          * we can only allow the backlog to be adjusted.
 210          */
 211         if (old_state != TCP_LISTEN) {
 212                 /* Check special setups for testing purpose to enable TFO w/o
 213                  * requiring TCP_FASTOPEN sockopt.
 214                  * Note that only TCP sockets (SOCK_STREAM) will reach here.
 215                  * Also fastopenq may already been allocated because this
 216                  * socket was in TCP_LISTEN state previously but was
 217                  * shutdown() (rather than close()).
 218                  */
 219                 if ((sysctl_tcp_fastopen & TFO_SERVER_ENABLE) != 0 &&
 220                     inet_csk(sk)->icsk_accept_queue.fastopenq == NULL) {
 221                         if ((sysctl_tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) != 0)
 222                                 err = fastopen_init_queue(sk, backlog);
 223                         else if ((sysctl_tcp_fastopen &
 224                                   TFO_SERVER_WO_SOCKOPT2) != 0)
 225                                 err = fastopen_init_queue(sk,
 226                                     ((uint)sysctl_tcp_fastopen) >> 16);
 227                         else
 228                                 err = 0;
 229                         if (err)
 230                                 goto out;
 231                 }
 232                 err = inet_csk_listen_start(sk, backlog);
 233                 if (err)
 234                         goto out;
 235         }
 236         sk->sk_max_ack_backlog = backlog;
 237         err = 0;
 238
 239 out:
 240         release_sock(sk);
 241         return err;
 242 }
 243 EXPORT_SYMBOL(inet_listen);
 244
 245 /*
 246  *      Create an inet socket.
 247  */
 248
 249 static int inet_create(struct net *net, struct socket *sock, int protocol,
 250                        int kern)
 251 {
 252         struct sock *sk;
 253         struct inet_protosw *answer;
 254         struct inet_sock *inet;
 255         struct proto *answer_prot;
 256         unsigned char answer_flags;
 257         char answer_no_check;
 258         int try_loading_module = 0;
 259         int err;
 260
 261         sock->state = SS_UNCONNECTED;
 262
 263         /* Look for the requested type/protocol pair. */
 264 lookup_protocol:
 265         err = -ESOCKTNOSUPPORT;
 266         rcu_read_lock();
 267         list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
 268
 269                 err = 0;
 270                 /* Check the non-wild match. */
 271                 if (protocol == answer->protocol) {
 272                         if (protocol != IPPROTO_IP)
 273                                 break;
 274                 } else {
 275                         /* Check for the two wild cases. */
 276                         if (IPPROTO_IP == protocol) {
 277                                 protocol = answer->protocol;
 278                                 break;
 279                         }
 280                         if (IPPROTO_IP == answer->protocol)
 281                                 break;
 282                 }
 283                 err = -EPROTONOSUPPORT;
 284         }
 285
 286         if (unlikely(err)) {
 287                 if (try_loading_module < 2) {
 288                         rcu_read_unlock();
 289                         /*
 290                          * Be more specific, e.g. net-pf-2-proto-132-type-1
 291                          * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
 292                          */
 293                         if (++try_loading_module == 1)
 294                                 request_module("net-pf-%d-proto-%d-type-%d",
 295                                                PF_INET, protocol, sock->type);
 296                         /*
 297                          * Fall back to generic, e.g. net-pf-2-proto-132
 298                          * (net-pf-PF_INET-proto-IPPROTO_SCTP)
 299                          */
 300                         else
 301                                 request_module("net-pf-%d-proto-%d",
 302                                                PF_INET, protocol);
 303                         goto lookup_protocol;
 304                 } else
 305                         goto out_rcu_unlock;
 306         }
 307
 308         err = -EPERM;
 309         if (sock->type == SOCK_RAW && !kern &&
 310             !ns_capable(net->user_ns, CAP_NET_RAW))
 311                 goto out_rcu_unlock;
 312
 313         sock->ops = answer->ops;
 314         answer_prot = answer->prot;
 315         answer_no_check = answer->no_check;
 316         answer_flags = answer->flags;
 317         rcu_read_unlock();
 318
 319         WARN_ON(answer_prot->slab == NULL);
 320
 321         err = -ENOBUFS;
 322         sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
 323         if (sk == NULL)
 324                 goto out;
 325
 326         err = 0;
 327         sk->sk_no_check = answer_no_check;
 328         if (INET_PROTOSW_REUSE & answer_flags)
 329                 sk->sk_reuse = SK_CAN_REUSE;
 330
 331         inet = inet_sk(sk);
 332         inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
 333
 334         inet->nodefrag = 0;
 335
 336         if (SOCK_RAW == sock->type) {
 337                 inet->inet_num = protocol;
 338                 if (IPPROTO_RAW == protocol)
 339                         inet->hdrincl = 1;
 340         }
 341
 342         if (net->ipv4.sysctl_ip_no_pmtu_disc)
 343                 inet->pmtudisc = IP_PMTUDISC_DONT;
 344         else
 345                 inet->pmtudisc = IP_PMTUDISC_WANT;
 346
 347         inet->inet_id = 0;
 348
 349         sock_init_data(sock, sk);
 350
 351         sk->sk_destruct    = inet_sock_destruct;
 352         sk->sk_protocol    = protocol;
 353         sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
 354
 355         inet->uc_ttl    = -1;
 356         inet->mc_loop   = 1;
 357         inet->mc_ttl    = 1;
 358         inet->mc_all    = 1;
 359         inet->mc_index  = 0;
 360         inet->mc_list   = NULL;
 361         inet->rcv_tos   = 0;
 362
 363         sk_refcnt_debug_inc(sk);
 364
 365         if (inet->inet_num) {
 366                 /* It assumes that any protocol which allows
 367                  * the user to assign a number at socket
 368                  * creation time automatically
 369                  * shares.
 370                  */
 371                 inet->inet_sport = htons(inet->inet_num);
 372                 /* Add to protocol hash chains. */
 373                 sk->sk_prot->hash(sk);
 374         }
 375
 376         if (sk->sk_prot->init) {
 377                 err = sk->sk_prot->init(sk);
 378                 if (err)
 379                         sk_common_release(sk);
 380         }
 381 out:
 382         return err;
 383 out_rcu_unlock:
 384         rcu_read_unlock();
 385         goto out;
 386 }
 387
 388
 389 /*
 390  *      The peer socket should always be NULL (or else). When we call this
 391  *      function we are destroying the object and from then on nobody
 392  *      should refer to it.
 393  */
 394 int inet_release(struct socket *sock)
 395 {
 396         struct sock *sk = sock->sk;
 397
 398         if (sk) {
 399                 long timeout;
 400
 401                 sock_rps_reset_flow(sk);
 402
 403                 /* Applications forget to leave groups before exiting */
 404                 ip_mc_drop_socket(sk);
 405
 406                 /* If linger is set, we don't return until the close
 407                  * is complete.  Otherwise we return immediately. The
 408                  * actually closing is done the same either way.
 409                  *
 410                  * If the close is due to the process exiting, we never
 411                  * linger..
 412                  */
 413                 timeout = 0;
 414                 if (sock_flag(sk, SOCK_LINGER) &&
 415                     !(current->flags & PF_EXITING))
 416                         timeout = sk->sk_lingertime;
 417                 sock->sk = NULL;
 418                 sk->sk_prot->close(sk, timeout);
 419         }
 420         return 0;
 421 }
 422 EXPORT_SYMBOL(inet_release);
 423
 424 /* It is off by default, see below. */
 425 int sysctl_ip_nonlocal_bind __read_mostly;
 426 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
 427
 428 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 429 {
 430         struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
 431         struct sock *sk = sock->sk;
 432         struct inet_sock *inet = inet_sk(sk);
 433         struct net *net = sock_net(sk);
 434         unsigned short snum;
 435         int chk_addr_ret;
 436         int err;
 437
 438         /* If the socket has its own bind function then use it. (RAW) */
 439         if (sk->sk_prot->bind) {
 440                 err = sk->sk_prot->bind(sk, uaddr, addr_len);
 441                 goto out;
 442         }
 443         err = -EINVAL;
 444         if (addr_len < sizeof(struct sockaddr_in))
 445                 goto out;
 446
 447         if (addr->sin_family != AF_INET) {
 448                 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
 449                  * only if s_addr is INADDR_ANY.
 450                  */
 451                 err = -EAFNOSUPPORT;
 452                 if (addr->sin_family != AF_UNSPEC ||
 453                     addr->sin_addr.s_addr != htonl(INADDR_ANY))
 454                         goto out;
 455         }
 456
 457         chk_addr_ret = inet_addr_type(net, addr->sin_addr.s_addr);
 458
 459         /* Not specified by any standard per-se, however it breaks too
 460          * many applications when removed.  It is unfortunate since
 461          * allowing applications to make a non-local bind solves
 462          * several problems with systems using dynamic addressing.
 463          * (ie. your servers still start up even if your ISDN link
 464          *  is temporarily down)
 465          */
 466         err = -EADDRNOTAVAIL;
 467         if (!sysctl_ip_nonlocal_bind &&
 468             !(inet->freebind || inet->transparent) &&
 469             addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
 470             chk_addr_ret != RTN_LOCAL &&
 471             chk_addr_ret != RTN_MULTICAST &&
 472             chk_addr_ret != RTN_BROADCAST)
 473                 goto out;
 474
 475         snum = ntohs(addr->sin_port);
 476         err = -EACCES;
 477         if (snum && snum < PROT_SOCK &&
 478             !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
 479                 goto out;
 480
 481         /*      We keep a pair of addresses. rcv_saddr is the one
 482          *      used by hash lookups, and saddr is used for transmit.
 483          *
 484          *      In the BSD API these are the same except where it
 485          *      would be illegal to use them (multicast/broadcast) in
 486          *      which case the sending device address is used.
 487          */
 488         lock_sock(sk);
 489
 490         /* Check these errors (active socket, double bind). */
 491         err = -EINVAL;
 492         if (sk->sk_state != TCP_CLOSE || inet->inet_num)
 493                 goto out_release_sock;
 494
 495         inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
 496         if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
 497                 inet->inet_saddr = 0;  /* Use device */
 498
 499         /* Make sure we are allowed to bind here. */
 500         if (sk->sk_prot->get_port(sk, snum)) {
 501                 inet->inet_saddr = inet->inet_rcv_saddr = 0;
 502                 err = -EADDRINUSE;
 503                 goto out_release_sock;
 504         }
 505
 506         if (inet->inet_rcv_saddr)
 507                 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
 508         if (snum)
 509                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
 510         inet->inet_sport = htons(inet->inet_num);
 511         inet->inet_daddr = 0;
 512         inet->inet_dport = 0;
 513         sk_dst_reset(sk);
 514         err = 0;
 515 out_release_sock:
 516         release_sock(sk);
 517 out:
 518         return err;
 519 }
 520 EXPORT_SYMBOL(inet_bind);
 521
 522 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
 523                        int addr_len, int flags)
 524 {
 525         struct sock *sk = sock->sk;
 526
 527         if (addr_len < sizeof(uaddr->sa_family))
 528                 return -EINVAL;
 529         if (uaddr->sa_family == AF_UNSPEC)
 530                 return sk->sk_prot->disconnect(sk, flags);
 531
 532         if (!inet_sk(sk)->inet_num && inet_autobind(sk))
 533                 return -EAGAIN;
 534         return sk->sk_prot->connect(sk, uaddr, addr_len);
 535 }
 536 EXPORT_SYMBOL(inet_dgram_connect);
 537
 538 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
 539 {
 540         DEFINE_WAIT(wait);
 541
 542         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
 543         sk->sk_write_pending += writebias;
 544
 545         /* Basic assumption: if someone sets sk->sk_err, he _must_
 546          * change state of the socket from TCP_SYN_*.
 547          * Connect() does not allow to get error notifications
 548          * without closing the socket.
 549          */
 550         while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 551                 release_sock(sk);
 552                 timeo = schedule_timeout(timeo);
 553                 lock_sock(sk);
 554                 if (signal_pending(current) || !timeo)
 555                         break;
 556                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
 557         }
 558         finish_wait(sk_sleep(sk), &wait);
 559         sk->sk_write_pending -= writebias;
 560         return timeo;
 561 }
 562
 563 /*
 564  *      Connect to a remote host. There is regrettably still a little
 565  *      TCP 'magic' in here.
 566  */
 567 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 568                           int addr_len, int flags)
 569 {
 570         struct sock *sk = sock->sk;
 571         int err;
 572         long timeo;
 573
 574         if (addr_len < sizeof(uaddr->sa_family))
 575                 return -EINVAL;
 576
 577         if (uaddr->sa_family == AF_UNSPEC) {
 578                 err = sk->sk_prot->disconnect(sk, flags);
 579                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 580                 goto out;
 581         }
 582
 583         switch (sock->state) {
 584         default:
 585                 err = -EINVAL;
 586                 goto out;
 587         case SS_CONNECTED:
 588                 err = -EISCONN;
 589                 goto out;
 590         case SS_CONNECTING:
 591                 err = -EALREADY;
 592                 /* Fall out of switch with err, set for this state */
 593                 break;
 594         case SS_UNCONNECTED:
 595                 err = -EISCONN;
 596                 if (sk->sk_state != TCP_CLOSE)
 597                         goto out;
 598
 599                 err = sk->sk_prot->connect(sk, uaddr, addr_len);
 600                 if (err < 0)
 601                         goto out;
 602
 603                 sock->state = SS_CONNECTING;
 604
 605                 /* Just entered SS_CONNECTING state; the only
 606                  * difference is that return value in non-blocking
 607                  * case is EINPROGRESS, rather than EALREADY.
 608                  */
 609                 err = -EINPROGRESS;
 610                 break;
 611         }
 612
 613         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
 614
 615         if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 616                 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
 617                                 tcp_sk(sk)->fastopen_req &&
 618                                 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
 619
 620                 /* Error code is set above */
 621                 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
 622                         goto out;
 623
 624                 err = sock_intr_errno(timeo);
 625                 if (signal_pending(current))
 626                         goto out;
 627         }
 628
 629         /* Connection was closed by RST, timeout, ICMP error
 630          * or another process disconnected us.
 631          */
 632         if (sk->sk_state == TCP_CLOSE)
 633                 goto sock_error;
 634
 635         /* sk->sk_err may be not zero now, if RECVERR was ordered by user
 636          * and error was received after socket entered established state.
 637          * Hence, it is handled normally after connect() return successfully.
 638          */
 639
 640         sock->state = SS_CONNECTED;
 641         err = 0;
 642 out:
 643         return err;
 644
 645 sock_error:
 646         err = sock_error(sk) ? : -ECONNABORTED;
 647         sock->state = SS_UNCONNECTED;
 648         if (sk->sk_prot->disconnect(sk, flags))
 649                 sock->state = SS_DISCONNECTING;
 650         goto out;
 651 }
 652 EXPORT_SYMBOL(__inet_stream_connect);
 653
 654 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 655                         int addr_len, int flags)
 656 {
 657         int err;
 658
 659         lock_sock(sock->sk);
 660         err = __inet_stream_connect(sock, uaddr, addr_len, flags);
 661         release_sock(sock->sk);
 662         return err;
 663 }
 664 EXPORT_SYMBOL(inet_stream_connect);
 665
 666 /*
 667  *      Accept a pending connection. The TCP layer now gives BSD semantics.
 668  */
 669
 670 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
 671 {
 672         struct sock *sk1 = sock->sk;
 673         int err = -EINVAL;
 674         struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
 675
 676         if (!sk2)
 677                 goto do_err;
 678
 679         lock_sock(sk2);
 680
 681         sock_rps_record_flow(sk2);
 682         WARN_ON(!((1 << sk2->sk_state) &
 683                   (TCPF_ESTABLISHED | TCPF_SYN_RECV |
 684                   TCPF_CLOSE_WAIT | TCPF_CLOSE)));
 685
 686         sock_graft(sk2, newsock);
 687
 688         newsock->state = SS_CONNECTED;
 689         err = 0;
 690         release_sock(sk2);
 691 do_err:
 692         return err;
 693 }
 694 EXPORT_SYMBOL(inet_accept);
 695
 696
 697 /*
 698  *      This does both peername and sockname.
 699  */
 700 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
 701                         int *uaddr_len, int peer)
 702 {
 703         struct sock *sk         = sock->sk;
 704         struct inet_sock *inet  = inet_sk(sk);
 705         DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
 706
 707         sin->sin_family = AF_INET;
 708         if (peer) {
 709                 if (!inet->inet_dport ||
 710                     (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
 711                      peer == 1))
 712                         return -ENOTCONN;
 713                 sin->sin_port = inet->inet_dport;
 714                 sin->sin_addr.s_addr = inet->inet_daddr;
 715         } else {
 716                 __be32 addr = inet->inet_rcv_saddr;
 717                 if (!addr)
 718                         addr = inet->inet_saddr;
 719                 sin->sin_port = inet->inet_sport;
 720                 sin->sin_addr.s_addr = addr;
 721         }
 722         memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 723         *uaddr_len = sizeof(*sin);
 724         return 0;
 725 }
 726 EXPORT_SYMBOL(inet_getname);
 727
 728 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
 729                  size_t size)
 730 {
 731         struct sock *sk = sock->sk;
 732
 733         sock_rps_record_flow(sk);
 734
 735         /* We may need to bind the socket. */
 736         if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
 737             inet_autobind(sk))
 738                 return -EAGAIN;
 739
 740         return sk->sk_prot->sendmsg(iocb, sk, msg, size);
 741 }
 742 EXPORT_SYMBOL(inet_sendmsg);
 743
 744 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
 745                       size_t size, int flags)
 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         if (sk->sk_prot->sendpage)
 757                 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
 758         return sock_no_sendpage(sock, page, offset, size, flags);
 759 }
 760 EXPORT_SYMBOL(inet_sendpage);
 761
 762 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
 763                  size_t size, int flags)
 764 {
 765         struct sock *sk = sock->sk;
 766         int addr_len = 0;
 767         int err;
 768
 769         sock_rps_record_flow(sk);
 770
 771         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
 772                                    flags & ~MSG_DONTWAIT, &addr_len);
 773         if (err >= 0)
 774                 msg->msg_namelen = addr_len;
 775         return err;
 776 }
 777 EXPORT_SYMBOL(inet_recvmsg);
 778
 779 int inet_shutdown(struct socket *sock, int how)
 780 {
 781         struct sock *sk = sock->sk;
 782         int err = 0;
 783
 784         /* This should really check to make sure
 785          * the socket is a TCP socket. (WHY AC...)
 786          */
 787         how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
 788                        1->2 bit 2 snds.
 789                        2->3 */
 790         if ((how & ~SHUTDOWN_MASK) || !how)     /* MAXINT->0 */
 791                 return -EINVAL;
 792
 793         lock_sock(sk);
 794         if (sock->state == SS_CONNECTING) {
 795                 if ((1 << sk->sk_state) &
 796                     (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
 797                         sock->state = SS_DISCONNECTING;
 798                 else
 799                         sock->state = SS_CONNECTED;
 800         }
 801
 802         switch (sk->sk_state) {
 803         case TCP_CLOSE:
 804                 err = -ENOTCONN;
 805                 /* Hack to wake up other listeners, who can poll for
 806                    POLLHUP, even on eg. unconnected UDP sockets -- RR */
 807         default:
 808                 sk->sk_shutdown |= how;
 809                 if (sk->sk_prot->shutdown)
 810                         sk->sk_prot->shutdown(sk, how);
 811                 break;
 812
 813         /* Remaining two branches are temporary solution for missing
 814          * close() in multithreaded environment. It is _not_ a good idea,
 815          * but we have no choice until close() is repaired at VFS level.
 816          */
 817         case TCP_LISTEN:
 818                 if (!(how & RCV_SHUTDOWN))
 819                         break;
 820                 /* Fall through */
 821         case TCP_SYN_SENT:
 822                 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
 823                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 824                 break;
 825         }
 826
 827         /* Wake up anyone sleeping in poll. */
 828         sk->sk_state_change(sk);
 829         release_sock(sk);
 830         return err;
 831 }
 832 EXPORT_SYMBOL(inet_shutdown);
 833
 834 /*
 835  *      ioctl() calls you can issue on an INET socket. Most of these are
 836  *      device configuration and stuff and very rarely used. Some ioctls
 837  *      pass on to the socket itself.
 838  *
 839  *      NOTE: I like the idea of a module for the config stuff. ie ifconfig
 840  *      loads the devconfigure module does its configuring and unloads it.
 841  *      There's a good 20K of config code hanging around the kernel.
 842  */
 843
 844 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 845 {
 846         struct sock *sk = sock->sk;
 847         int err = 0;
 848         struct net *net = sock_net(sk);
 849
 850         switch (cmd) {
 851         case SIOCGSTAMP:
 852                 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
 853                 break;
 854         case SIOCGSTAMPNS:
 855                 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
 856                 break;
 857         case SIOCADDRT:
 858         case SIOCDELRT:
 859         case SIOCRTMSG:
 860                 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
 861                 break;
 862         case SIOCDARP:
 863         case SIOCGARP:
 864         case SIOCSARP:
 865                 err = arp_ioctl(net, cmd, (void __user *)arg);
 866                 break;
 867         case SIOCGIFADDR:
 868         case SIOCSIFADDR:
 869         case SIOCGIFBRDADDR:
 870         case SIOCSIFBRDADDR:
 871         case SIOCGIFNETMASK:
 872         case SIOCSIFNETMASK:
 873         case SIOCGIFDSTADDR:
 874         case SIOCSIFDSTADDR:
 875         case SIOCSIFPFLAGS:
 876         case SIOCGIFPFLAGS:
 877         case SIOCSIFFLAGS:
 878                 err = devinet_ioctl(net, cmd, (void __user *)arg);
 879                 break;
 880         default:
 881                 if (sk->sk_prot->ioctl)
 882                         err = sk->sk_prot->ioctl(sk, cmd, arg);
 883                 else
 884                         err = -ENOIOCTLCMD;
 885                 break;
 886         }
 887         return err;
 888 }
 889 EXPORT_SYMBOL(inet_ioctl);
 890
 891 #ifdef CONFIG_COMPAT
 892 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 893 {
 894         struct sock *sk = sock->sk;
 895         int err = -ENOIOCTLCMD;
 896
 897         if (sk->sk_prot->compat_ioctl)
 898                 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
 899
 900         return err;
 901 }
 902 #endif
 903
 904 const struct proto_ops inet_stream_ops = {
 905         .family            = PF_INET,
 906         .owner             = THIS_MODULE,
 907         .release           = inet_release,
 908         .bind              = inet_bind,
 909         .connect           = inet_stream_connect,
 910         .socketpair        = sock_no_socketpair,
 911         .accept            = inet_accept,
 912         .getname           = inet_getname,
 913         .poll              = tcp_poll,
 914         .ioctl             = inet_ioctl,
 915         .listen            = inet_listen,
 916         .shutdown          = inet_shutdown,
 917         .setsockopt        = sock_common_setsockopt,
 918         .getsockopt        = sock_common_getsockopt,
 919         .sendmsg           = inet_sendmsg,
 920         .recvmsg           = inet_recvmsg,
 921         .mmap              = sock_no_mmap,
 922         .sendpage          = inet_sendpage,
 923         .splice_read       = tcp_splice_read,
 924 #ifdef CONFIG_COMPAT
 925         .compat_setsockopt = compat_sock_common_setsockopt,
 926         .compat_getsockopt = compat_sock_common_getsockopt,
 927         .compat_ioctl      = inet_compat_ioctl,
 928 #endif
 929 };
 930 EXPORT_SYMBOL(inet_stream_ops);
 931
 932 const struct proto_ops inet_dgram_ops = {
 933         .family            = PF_INET,
 934         .owner             = THIS_MODULE,
 935         .release           = inet_release,
 936         .bind              = inet_bind,
 937         .connect           = inet_dgram_connect,
 938         .socketpair        = sock_no_socketpair,
 939         .accept            = sock_no_accept,
 940         .getname           = inet_getname,
 941         .poll              = udp_poll,
 942         .ioctl             = inet_ioctl,
 943         .listen            = sock_no_listen,
 944         .shutdown          = inet_shutdown,
 945         .setsockopt        = sock_common_setsockopt,
 946         .getsockopt        = sock_common_getsockopt,
 947         .sendmsg           = inet_sendmsg,
 948         .recvmsg           = inet_recvmsg,
 949         .mmap              = sock_no_mmap,
 950         .sendpage          = inet_sendpage,
 951 #ifdef CONFIG_COMPAT
 952         .compat_setsockopt = compat_sock_common_setsockopt,
 953         .compat_getsockopt = compat_sock_common_getsockopt,
 954         .compat_ioctl      = inet_compat_ioctl,
 955 #endif
 956 };
 957 EXPORT_SYMBOL(inet_dgram_ops);
 958
 959 /*
 960  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
 961  * udp_poll
 962  */
 963 static const struct proto_ops inet_sockraw_ops = {
 964         .family            = PF_INET,
 965         .owner             = THIS_MODULE,
 966         .release           = inet_release,
 967         .bind              = inet_bind,
 968         .connect           = inet_dgram_connect,
 969         .socketpair        = sock_no_socketpair,
 970         .accept            = sock_no_accept,
 971         .getname           = inet_getname,
 972         .poll              = datagram_poll,
 973         .ioctl             = inet_ioctl,
 974         .listen            = sock_no_listen,
 975         .shutdown          = inet_shutdown,
 976         .setsockopt        = sock_common_setsockopt,
 977         .getsockopt        = sock_common_getsockopt,
 978         .sendmsg           = inet_sendmsg,
 979         .recvmsg           = inet_recvmsg,
 980         .mmap              = sock_no_mmap,
 981         .sendpage          = inet_sendpage,
 982 #ifdef CONFIG_COMPAT
 983         .compat_setsockopt = compat_sock_common_setsockopt,
 984         .compat_getsockopt = compat_sock_common_getsockopt,
 985         .compat_ioctl      = inet_compat_ioctl,
 986 #endif
 987 };
 988
 989 static const struct net_proto_family inet_family_ops = {
 990         .family = PF_INET,
 991         .create = inet_create,
 992         .owner  = THIS_MODULE,
 993 };
 994
 995 /* Upon startup we insert all the elements in inetsw_array[] into
 996  * the linked list inetsw.
 997  */
 998 static struct inet_protosw inetsw_array[] =
 999 {
1000         {
1001                 .type =       SOCK_STREAM,
1002                 .protocol =   IPPROTO_TCP,
1003                 .prot =       &tcp_prot,
1004                 .ops =        &inet_stream_ops,
1005                 .no_check =   0,
1006                 .flags =      INET_PROTOSW_PERMANENT |
1007                               INET_PROTOSW_ICSK,
1008         },
1009
1010         {
1011                 .type =       SOCK_DGRAM,
1012                 .protocol =   IPPROTO_UDP,
1013                 .prot =       &udp_prot,
1014                 .ops =        &inet_dgram_ops,
1015                 .no_check =   UDP_CSUM_DEFAULT,
1016                 .flags =      INET_PROTOSW_PERMANENT,
1017        },
1018
1019        {
1020                 .type =       SOCK_DGRAM,
1021                 .protocol =   IPPROTO_ICMP,
1022                 .prot =       &ping_prot,
1023                 .ops =        &inet_dgram_ops,
1024                 .no_check =   UDP_CSUM_DEFAULT,
1025                 .flags =      INET_PROTOSW_REUSE,
1026        },
1027
1028        {
1029                .type =       SOCK_RAW,
1030                .protocol =   IPPROTO_IP,        /* wild card */
1031                .prot =       &raw_prot,
1032                .ops =        &inet_sockraw_ops,
1033                .no_check =   UDP_CSUM_DEFAULT,
1034                .flags =      INET_PROTOSW_REUSE,
1035        }
1036 };
1037
1038 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1039
1040 void inet_register_protosw(struct inet_protosw *p)
1041 {
1042         struct list_head *lh;
1043         struct inet_protosw *answer;
1044         int protocol = p->protocol;
1045         struct list_head *last_perm;
1046
1047         spin_lock_bh(&inetsw_lock);
1048
1049         if (p->type >= SOCK_MAX)
1050                 goto out_illegal;
1051
1052         /* If we are trying to override a permanent protocol, bail. */
1053         answer = NULL;
1054         last_perm = &inetsw[p->type];
1055         list_for_each(lh, &inetsw[p->type]) {
1056                 answer = list_entry(lh, struct inet_protosw, list);
1057
1058                 /* Check only the non-wild match. */
1059                 if (INET_PROTOSW_PERMANENT & answer->flags) {
1060                         if (protocol == answer->protocol)
1061                                 break;
1062                         last_perm = lh;
1063                 }
1064
1065                 answer = NULL;
1066         }
1067         if (answer)
1068                 goto out_permanent;
1069
1070         /* Add the new entry after the last permanent entry if any, so that
1071          * the new entry does not override a permanent entry when matched with
1072          * a wild-card protocol. But it is allowed to override any existing
1073          * non-permanent entry.  This means that when we remove this entry, the
1074          * system automatically returns to the old behavior.
1075          */
1076         list_add_rcu(&p->list, last_perm);
1077 out:
1078         spin_unlock_bh(&inetsw_lock);
1079
1080         return;
1081
1082 out_permanent:
1083         pr_err("Attempt to override permanent protocol %d\n", protocol);
1084         goto out;
1085
1086 out_illegal:
1087         pr_err("Ignoring attempt to register invalid socket type %d\n",
1088                p->type);
1089         goto out;
1090 }
1091 EXPORT_SYMBOL(inet_register_protosw);
1092
1093 void inet_unregister_protosw(struct inet_protosw *p)
1094 {
1095         if (INET_PROTOSW_PERMANENT & p->flags) {
1096                 pr_err("Attempt to unregister permanent protocol %d\n",
1097                        p->protocol);
1098         } else {
1099                 spin_lock_bh(&inetsw_lock);
1100                 list_del_rcu(&p->list);
1101                 spin_unlock_bh(&inetsw_lock);
1102
1103                 synchronize_net();
1104         }
1105 }
1106 EXPORT_SYMBOL(inet_unregister_protosw);
1107
1108 /*
1109  *      Shall we try to damage output packets if routing dev changes?
1110  */
1111
1112 int sysctl_ip_dynaddr __read_mostly;
1113
1114 static int inet_sk_reselect_saddr(struct sock *sk)
1115 {
1116         struct inet_sock *inet = inet_sk(sk);
1117         __be32 old_saddr = inet->inet_saddr;
1118         __be32 daddr = inet->inet_daddr;
1119         struct flowi4 *fl4;
1120         struct rtable *rt;
1121         __be32 new_saddr;
1122         struct ip_options_rcu *inet_opt;
1123
1124         inet_opt = rcu_dereference_protected(inet->inet_opt,
1125                                              sock_owned_by_user(sk));
1126         if (inet_opt && inet_opt->opt.srr)
1127                 daddr = inet_opt->opt.faddr;
1128
1129         /* Query new route. */
1130         fl4 = &inet->cork.fl.u.ip4;
1131         rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1132                               sk->sk_bound_dev_if, sk->sk_protocol,
1133                               inet->inet_sport, inet->inet_dport, sk);
1134         if (IS_ERR(rt))
1135                 return PTR_ERR(rt);
1136
1137         sk_setup_caps(sk, &rt->dst);
1138
1139         new_saddr = fl4->saddr;
1140
1141         if (new_saddr == old_saddr)
1142                 return 0;
1143
1144         if (sysctl_ip_dynaddr > 1) {
1145                 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1146                         __func__, &old_saddr, &new_saddr);
1147         }
1148
1149         inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1150
1151         /*
1152          * XXX The only one ugly spot where we need to
1153          * XXX really change the sockets identity after
1154          * XXX it has entered the hashes. -DaveM
1155          *
1156          * Besides that, it does not check for connection
1157          * uniqueness. Wait for troubles.
1158          */
1159         __sk_prot_rehash(sk);
1160         return 0;
1161 }
1162
1163 int inet_sk_rebuild_header(struct sock *sk)
1164 {
1165         struct inet_sock *inet = inet_sk(sk);
1166         struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1167         __be32 daddr;
1168         struct ip_options_rcu *inet_opt;
1169         struct flowi4 *fl4;
1170         int err;
1171
1172         /* Route is OK, nothing to do. */
1173         if (rt)
1174                 return 0;
1175
1176         /* Reroute. */
1177         rcu_read_lock();
1178         inet_opt = rcu_dereference(inet->inet_opt);
1179         daddr = inet->inet_daddr;
1180         if (inet_opt && inet_opt->opt.srr)
1181                 daddr = inet_opt->opt.faddr;
1182         rcu_read_unlock();
1183         fl4 = &inet->cork.fl.u.ip4;
1184         rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1185                                    inet->inet_dport, inet->inet_sport,
1186                                    sk->sk_protocol, RT_CONN_FLAGS(sk),
1187                                    sk->sk_bound_dev_if);
1188         if (!IS_ERR(rt)) {
1189                 err = 0;
1190                 sk_setup_caps(sk, &rt->dst);
1191         } else {
1192                 err = PTR_ERR(rt);
1193
1194                 /* Routing failed... */
1195                 sk->sk_route_caps = 0;
1196                 /*
1197                  * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1198                  * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1199                  */
1200                 if (!sysctl_ip_dynaddr ||
1201                     sk->sk_state != TCP_SYN_SENT ||
1202                     (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1203                     (err = inet_sk_reselect_saddr(sk)) != 0)
1204                         sk->sk_err_soft = -err;
1205         }
1206
1207         return err;
1208 }
1209 EXPORT_SYMBOL(inet_sk_rebuild_header);
1210
1211 static int inet_gso_send_check(struct sk_buff *skb)
1212 {
1213         const struct net_offload *ops;
1214         const struct iphdr *iph;
1215         int proto;
1216         int ihl;
1217         int err = -EINVAL;
1218
1219         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1220                 goto out;
1221
1222         iph = ip_hdr(skb);
1223         ihl = iph->ihl * 4;
1224         if (ihl < sizeof(*iph))
1225                 goto out;
1226
1227         proto = iph->protocol;
1228
1229         /* Warning: after this point, iph might be no longer valid */
1230         if (unlikely(!pskb_may_pull(skb, ihl)))
1231                 goto out;
1232         __skb_pull(skb, ihl);
1233
1234         skb_reset_transport_header(skb);
1235         err = -EPROTONOSUPPORT;
1236
1237         ops = rcu_dereference(inet_offloads[proto]);
1238         if (likely(ops && ops->callbacks.gso_send_check))
1239                 err = ops->callbacks.gso_send_check(skb);
1240
1241 out:
1242         return err;
1243 }
1244
1245 static struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1246                                         netdev_features_t features)
1247 {
1248         struct sk_buff *segs = ERR_PTR(-EINVAL);
1249         const struct net_offload *ops;
1250         unsigned int offset = 0;
1251         bool udpfrag, encap;
1252         struct iphdr *iph;
1253         int proto;
1254         int nhoff;
1255         int ihl;
1256         int id;
1257
1258         if (unlikely(skb_shinfo(skb)->gso_type &
1259                      ~(SKB_GSO_TCPV4 |
1260                        SKB_GSO_UDP |
1261                        SKB_GSO_DODGY |
1262                        SKB_GSO_TCP_ECN |
1263                        SKB_GSO_GRE |
1264                        SKB_GSO_IPIP |
1265                        SKB_GSO_SIT |
1266                        SKB_GSO_TCPV6 |
1267                        SKB_GSO_UDP_TUNNEL |
1268                        SKB_GSO_MPLS |
1269                        0)))
1270                 goto out;
1271
1272         skb_reset_network_header(skb);
1273         nhoff = skb_network_header(skb) - skb_mac_header(skb);
1274         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1275                 goto out;
1276
1277         iph = ip_hdr(skb);
1278         ihl = iph->ihl * 4;
1279         if (ihl < sizeof(*iph))
1280                 goto out;
1281
1282         id = ntohs(iph->id);
1283         proto = iph->protocol;
1284
1285         /* Warning: after this point, iph might be no longer valid */
1286         if (unlikely(!pskb_may_pull(skb, ihl)))
1287                 goto out;
1288         __skb_pull(skb, ihl);
1289
1290         encap = SKB_GSO_CB(skb)->encap_level > 0;
1291         if (encap)
1292                 features = skb->dev->hw_enc_features & netif_skb_features(skb);
1293         SKB_GSO_CB(skb)->encap_level += ihl;
1294
1295         skb_reset_transport_header(skb);
1296
1297         segs = ERR_PTR(-EPROTONOSUPPORT);
1298
1299         if (skb->encapsulation &&
1300             skb_shinfo(skb)->gso_type & (SKB_GSO_SIT|SKB_GSO_IPIP))
1301                 udpfrag = proto == IPPROTO_UDP && encap;
1302         else
1303                 udpfrag = proto == IPPROTO_UDP && !skb->encapsulation;
1304
1305         ops = rcu_dereference(inet_offloads[proto]);
1306         if (likely(ops && ops->callbacks.gso_segment))
1307                 segs = ops->callbacks.gso_segment(skb, features);
1308
1309         if (IS_ERR_OR_NULL(segs))
1310                 goto out;
1311
1312         skb = segs;
1313         do {
1314                 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1315                 if (udpfrag) {
1316                         iph->id = htons(id);
1317                         iph->frag_off = htons(offset >> 3);
1318                         if (skb->next != NULL)
1319                                 iph->frag_off |= htons(IP_MF);
1320                         offset += skb->len - nhoff - ihl;
1321                 } else {
1322                         iph->id = htons(id++);
1323                 }
1324                 iph->tot_len = htons(skb->len - nhoff);
1325                 ip_send_check(iph);
1326                 if (encap)
1327                         skb_reset_inner_headers(skb);
1328                 skb->network_header = (u8 *)iph - skb->head;
1329         } while ((skb = skb->next));
1330
1331 out:
1332         return segs;
1333 }
1334
1335 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1336                                          struct sk_buff *skb)
1337 {
1338         const struct net_offload *ops;
1339         struct sk_buff **pp = NULL;
1340         struct sk_buff *p;
1341         const struct iphdr *iph;
1342         unsigned int hlen;
1343         unsigned int off;
1344         unsigned int id;
1345         int flush = 1;
1346         int proto;
1347
1348         off = skb_gro_offset(skb);
1349         hlen = off + sizeof(*iph);
1350         iph = skb_gro_header_fast(skb, off);
1351         if (skb_gro_header_hard(skb, hlen)) {
1352                 iph = skb_gro_header_slow(skb, hlen, off);
1353                 if (unlikely(!iph))
1354                         goto out;
1355         }
1356
1357         proto = iph->protocol;
1358
1359         rcu_read_lock();
1360         ops = rcu_dereference(inet_offloads[proto]);
1361         if (!ops || !ops->callbacks.gro_receive)
1362                 goto out_unlock;
1363
1364         if (*(u8 *)iph != 0x45)
1365                 goto out_unlock;
1366
1367         if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1368                 goto out_unlock;
1369
1370         id = ntohl(*(__be32 *)&iph->id);
1371         flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1372         id >>= 16;
1373
1374         for (p = *head; p; p = p->next) {
1375                 struct iphdr *iph2;
1376
1377                 if (!NAPI_GRO_CB(p)->same_flow)
1378                         continue;
1379
1380                 iph2 = (struct iphdr *)(p->data + off);
1381                 /* The above works because, with the exception of the top
1382                  * (inner most) layer, we only aggregate pkts with the same
1383                  * hdr length so all the hdrs we'll need to verify will start
1384                  * at the same offset.
1385                  */
1386                 if ((iph->protocol ^ iph2->protocol) |
1387                     ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1388                     ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1389                         NAPI_GRO_CB(p)->same_flow = 0;
1390                         continue;
1391                 }
1392
1393                 /* All fields must match except length and checksum. */
1394                 NAPI_GRO_CB(p)->flush |=
1395                         (iph->ttl ^ iph2->ttl) |
1396                         (iph->tos ^ iph2->tos) |
1397                         ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1398
1399                 /* Save the IP ID check to be included later when we get to
1400                  * the transport layer so only the inner most IP ID is checked.
1401                  * This is because some GSO/TSO implementations do not
1402                  * correctly increment the IP ID for the outer hdrs.
1403                  */
1404                 NAPI_GRO_CB(p)->flush_id =
1405                             ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1406                 NAPI_GRO_CB(p)->flush |= flush;
1407         }
1408
1409         NAPI_GRO_CB(skb)->flush |= flush;
1410         skb_set_network_header(skb, off);
1411         /* The above will be needed by the transport layer if there is one
1412          * immediately following this IP hdr.
1413          */
1414
1415         skb_gro_pull(skb, sizeof(*iph));
1416         skb_set_transport_header(skb, skb_gro_offset(skb));
1417
1418         pp = ops->callbacks.gro_receive(head, skb);
1419
1420 out_unlock:
1421         rcu_read_unlock();
1422
1423 out:
1424         NAPI_GRO_CB(skb)->flush |= flush;
1425
1426         return pp;
1427 }
1428
1429 static int inet_gro_complete(struct sk_buff *skb, int nhoff)
1430 {
1431         __be16 newlen = htons(skb->len - nhoff);
1432         struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1433         const struct net_offload *ops;
1434         int proto = iph->protocol;
1435         int err = -ENOSYS;
1436
1437         csum_replace2(&iph->check, iph->tot_len, newlen);
1438         iph->tot_len = newlen;
1439
1440         rcu_read_lock();
1441         ops = rcu_dereference(inet_offloads[proto]);
1442         if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1443                 goto out_unlock;
1444
1445         /* Only need to add sizeof(*iph) to get to the next hdr below
1446          * because any hdr with option will have been flushed in
1447          * inet_gro_receive().
1448          */
1449         err = ops->callbacks.gro_complete(skb, nhoff + sizeof(*iph));
1450
1451 out_unlock:
1452         rcu_read_unlock();
1453
1454         return err;
1455 }
1456
1457 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1458                          unsigned short type, unsigned char protocol,
1459                          struct net *net)
1460 {
1461         struct socket *sock;
1462         int rc = sock_create_kern(family, type, protocol, &sock);
1463
1464         if (rc == 0) {
1465                 *sk = sock->sk;
1466                 (*sk)->sk_allocation = GFP_ATOMIC;
1467                 /*
1468                  * Unhash it so that IP input processing does not even see it,
1469                  * we do not wish this socket to see incoming packets.
1470                  */
1471                 (*sk)->sk_prot->unhash(*sk);
1472
1473                 sk_change_net(*sk, net);
1474         }
1475         return rc;
1476 }
1477 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1478
1479 unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1480 {
1481         unsigned long res = 0;
1482         int i, j;
1483
1484         for_each_possible_cpu(i) {
1485                 for (j = 0; j < SNMP_ARRAY_SZ; j++)
1486                         res += *(((unsigned long *) per_cpu_ptr(mib[j], i)) + offt);
1487         }
1488         return res;
1489 }
1490 EXPORT_SYMBOL_GPL(snmp_fold_field);
1491
1492 #if BITS_PER_LONG==32
1493
1494 u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset)
1495 {
1496         u64 res = 0;
1497         int cpu;
1498
1499         for_each_possible_cpu(cpu) {
1500                 void *bhptr;
1501                 struct u64_stats_sync *syncp;
1502                 u64 v;
1503                 unsigned int start;
1504
1505                 bhptr = per_cpu_ptr(mib[0], cpu);
1506                 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1507                 do {
1508                         start = u64_stats_fetch_begin_bh(syncp);
1509                         v = *(((u64 *) bhptr) + offt);
1510                 } while (u64_stats_fetch_retry_bh(syncp, start));
1511
1512                 res += v;
1513         }
1514         return res;
1515 }
1516 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1517 #endif
1518
1519 int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align)
1520 {
1521         BUG_ON(ptr == NULL);
1522         ptr[0] = __alloc_percpu(mibsize, align);
1523         if (!ptr[0])
1524                 return -ENOMEM;
1525
1526 #if SNMP_ARRAY_SZ == 2
1527         ptr[1] = __alloc_percpu(mibsize, align);
1528         if (!ptr[1]) {
1529                 free_percpu(ptr[0]);
1530                 ptr[0] = NULL;
1531                 return -ENOMEM;
1532         }
1533 #endif
1534         return 0;
1535 }
1536 EXPORT_SYMBOL_GPL(snmp_mib_init);
1537
1538 #ifdef CONFIG_IP_MULTICAST
1539 static const struct net_protocol igmp_protocol = {
1540         .handler =      igmp_rcv,
1541         .netns_ok =     1,
1542 };
1543 #endif
1544
1545 static const struct net_protocol tcp_protocol = {
1546         .early_demux    =       tcp_v4_early_demux,
1547         .handler        =       tcp_v4_rcv,
1548         .err_handler    =       tcp_v4_err,
1549         .no_policy      =       1,
1550         .netns_ok       =       1,
1551         .icmp_strict_tag_validation = 1,
1552 };
1553
1554 static const struct net_protocol udp_protocol = {
1555         .early_demux =  udp_v4_early_demux,
1556         .handler =      udp_rcv,
1557         .err_handler =  udp_err,
1558         .no_policy =    1,
1559         .netns_ok =     1,
1560 };
1561
1562 static const struct net_protocol icmp_protocol = {
1563         .handler =      icmp_rcv,
1564         .err_handler =  icmp_err,
1565         .no_policy =    1,
1566         .netns_ok =     1,
1567 };
1568
1569 static __net_init int ipv4_mib_init_net(struct net *net)
1570 {
1571         int i;
1572
1573         if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1574                           sizeof(struct tcp_mib),
1575                           __alignof__(struct tcp_mib)) < 0)
1576                 goto err_tcp_mib;
1577         if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1578                           sizeof(struct ipstats_mib),
1579                           __alignof__(struct ipstats_mib)) < 0)
1580                 goto err_ip_mib;
1581
1582         for_each_possible_cpu(i) {
1583                 struct ipstats_mib *af_inet_stats;
1584                 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics[0], i);
1585                 u64_stats_init(&af_inet_stats->syncp);
1586 #if SNMP_ARRAY_SZ == 2
1587                 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics[1], i);
1588                 u64_stats_init(&af_inet_stats->syncp);
1589 #endif
1590         }
1591
1592         if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1593                           sizeof(struct linux_mib),
1594                           __alignof__(struct linux_mib)) < 0)
1595                 goto err_net_mib;
1596         if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1597                           sizeof(struct udp_mib),
1598                           __alignof__(struct udp_mib)) < 0)
1599                 goto err_udp_mib;
1600         if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1601                           sizeof(struct udp_mib),
1602                           __alignof__(struct udp_mib)) < 0)
1603                 goto err_udplite_mib;
1604         if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1605                           sizeof(struct icmp_mib),
1606                           __alignof__(struct icmp_mib)) < 0)
1607                 goto err_icmp_mib;
1608         net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1609                                               GFP_KERNEL);
1610         if (!net->mib.icmpmsg_statistics)
1611                 goto err_icmpmsg_mib;
1612
1613         tcp_mib_init(net);
1614         return 0;
1615
1616 err_icmpmsg_mib:
1617         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1618 err_icmp_mib:
1619         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1620 err_udplite_mib:
1621         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1622 err_udp_mib:
1623         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1624 err_net_mib:
1625         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1626 err_ip_mib:
1627         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1628 err_tcp_mib:
1629         return -ENOMEM;
1630 }
1631
1632 static __net_exit void ipv4_mib_exit_net(struct net *net)
1633 {
1634         kfree(net->mib.icmpmsg_statistics);
1635         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1636         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1637         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1638         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1639         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1640         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1641 }
1642
1643 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1644         .init = ipv4_mib_init_net,
1645         .exit = ipv4_mib_exit_net,
1646 };
1647
1648 static int __init init_ipv4_mibs(void)
1649 {
1650         return register_pernet_subsys(&ipv4_mib_ops);
1651 }
1652
1653 static int ipv4_proc_init(void);
1654
1655 /*
1656  *      IP protocol layer initialiser
1657  */
1658
1659 static struct packet_offload ip_packet_offload __read_mostly = {
1660         .type = cpu_to_be16(ETH_P_IP),
1661         .callbacks = {
1662                 .gso_send_check = inet_gso_send_check,
1663                 .gso_segment = inet_gso_segment,
1664                 .gro_receive = inet_gro_receive,
1665                 .gro_complete = inet_gro_complete,
1666         },
1667 };
1668
1669 static const struct net_offload ipip_offload = {
1670         .callbacks = {
1671                 .gso_send_check = inet_gso_send_check,
1672                 .gso_segment    = inet_gso_segment,
1673         },
1674 };
1675
1676 static int __init ipv4_offload_init(void)
1677 {
1678         /*
1679          * Add offloads
1680          */
1681         if (udpv4_offload_init() < 0)
1682                 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1683         if (tcpv4_offload_init() < 0)
1684                 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1685
1686         dev_add_offload(&ip_packet_offload);
1687         inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1688         return 0;
1689 }
1690
1691 fs_initcall(ipv4_offload_init);
1692
1693 static struct packet_type ip_packet_type __read_mostly = {
1694         .type = cpu_to_be16(ETH_P_IP),
1695         .func = ip_rcv,
1696 };
1697
1698 static int __init inet_init(void)
1699 {
1700         struct inet_protosw *q;
1701         struct list_head *r;
1702         int rc = -EINVAL;
1703
1704         BUILD_BUG_ON(sizeof(struct inet_skb_parm) > FIELD_SIZEOF(struct sk_buff, cb));
1705
1706         sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL);
1707         if (!sysctl_local_reserved_ports)
1708                 goto out;
1709
1710         rc = proto_register(&tcp_prot, 1);
1711         if (rc)
1712                 goto out_free_reserved_ports;
1713
1714         rc = proto_register(&udp_prot, 1);
1715         if (rc)
1716                 goto out_unregister_tcp_proto;
1717
1718         rc = proto_register(&raw_prot, 1);
1719         if (rc)
1720                 goto out_unregister_udp_proto;
1721
1722         rc = proto_register(&ping_prot, 1);
1723         if (rc)
1724                 goto out_unregister_raw_proto;
1725
1726         /*
1727          *      Tell SOCKET that we are alive...
1728          */
1729
1730         (void)sock_register(&inet_family_ops);
1731
1732 #ifdef CONFIG_SYSCTL
1733         ip_static_sysctl_init();
1734 #endif
1735
1736         /*
1737          *      Add all the base protocols.
1738          */
1739
1740         if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1741                 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1742         if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1743                 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1744         if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1745                 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1746 #ifdef CONFIG_IP_MULTICAST
1747         if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1748                 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1749 #endif
1750
1751         /* Register the socket-side information for inet_create. */
1752         for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1753                 INIT_LIST_HEAD(r);
1754
1755         for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1756                 inet_register_protosw(q);
1757
1758         /*
1759          *      Set the ARP module up
1760          */
1761
1762         arp_init();
1763
1764         /*
1765          *      Set the IP module up
1766          */
1767
1768         ip_init();
1769
1770         tcp_v4_init();
1771
1772         /* Setup TCP slab cache for open requests. */
1773         tcp_init();
1774
1775         /* Setup UDP memory threshold */
1776         udp_init();
1777
1778         /* Add UDP-Lite (RFC 3828) */
1779         udplite4_register();
1780
1781         ping_init();
1782
1783         /*
1784          *      Set the ICMP layer up
1785          */
1786
1787         if (icmp_init() < 0)
1788                 panic("Failed to create the ICMP control socket.\n");
1789
1790         /*
1791          *      Initialise the multicast router
1792          */
1793 #if defined(CONFIG_IP_MROUTE)
1794         if (ip_mr_init())
1795                 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1796 #endif
1797         /*
1798          *      Initialise per-cpu ipv4 mibs
1799          */
1800
1801         if (init_ipv4_mibs())
1802                 pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1803
1804         ipv4_proc_init();
1805
1806         ipfrag_init();
1807
1808         dev_add_pack(&ip_packet_type);
1809
1810         rc = 0;
1811 out:
1812         return rc;
1813 out_unregister_raw_proto:
1814         proto_unregister(&raw_prot);
1815 out_unregister_udp_proto:
1816         proto_unregister(&udp_prot);
1817 out_unregister_tcp_proto:
1818         proto_unregister(&tcp_prot);
1819 out_free_reserved_ports:
1820         kfree(sysctl_local_reserved_ports);
1821         goto out;
1822 }
1823
1824 fs_initcall(inet_init);
1825
1826 /* ------------------------------------------------------------------------ */
1827
1828 #ifdef CONFIG_PROC_FS
1829 static int __init ipv4_proc_init(void)
1830 {
1831         int rc = 0;
1832
1833         if (raw_proc_init())
1834                 goto out_raw;
1835         if (tcp4_proc_init())
1836                 goto out_tcp;
1837         if (udp4_proc_init())
1838                 goto out_udp;
1839         if (ping_proc_init())
1840                 goto out_ping;
1841         if (ip_misc_proc_init())
1842                 goto out_misc;
1843 out:
1844         return rc;
1845 out_misc:
1846         ping_proc_exit();
1847 out_ping:
1848         udp4_proc_exit();
1849 out_udp:
1850         tcp4_proc_exit();
1851 out_tcp:
1852         raw_proc_exit();
1853 out_raw:
1854         rc = -ENOMEM;
1855         goto out;
1856 }
1857
1858 #else /* CONFIG_PROC_FS */
1859 static int __init ipv4_proc_init(void)
1860 {
1861         return 0;
1862 }
1863 #endif /* CONFIG_PROC_FS */
1864
1865 MODULE_ALIAS_NETPROTO(PF_INET);
1866