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 struct ipv4_config ipv4_config;
 130 EXPORT_SYMBOL(ipv4_config);
 131
 132 /* New destruction routine */
 133
 134 void inet_sock_destruct(struct sock *sk)
 135 {
 136         struct inet_sock *inet = inet_sk(sk);
 137
 138         __skb_queue_purge(&sk->sk_receive_queue);
 139         __skb_queue_purge(&sk->sk_error_queue);
 140
 141         sk_mem_reclaim(sk);
 142
 143         if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
 144                 pr_err("Attempt to release TCP socket in state %d %p\n",
 145                        sk->sk_state, sk);
 146                 return;
 147         }
 148         if (!sock_flag(sk, SOCK_DEAD)) {
 149                 pr_err("Attempt to release alive inet socket %p\n", sk);
 150                 return;
 151         }
 152
 153         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
 154         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
 155         WARN_ON(sk->sk_wmem_queued);
 156         WARN_ON(sk->sk_forward_alloc);
 157
 158         kfree(rcu_dereference_protected(inet->inet_opt, 1));
 159         dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
 160         dst_release(sk->sk_rx_dst);
 161         sk_refcnt_debug_dec(sk);
 162 }
 163 EXPORT_SYMBOL(inet_sock_destruct);
 164
 165 /*
 166  *      The routines beyond this point handle the behaviour of an AF_INET
 167  *      socket object. Mostly it punts to the subprotocols of IP to do
 168  *      the work.
 169  */
 170
 171 /*
 172  *      Automatically bind an unbound socket.
 173  */
 174
 175 static int inet_autobind(struct sock *sk)
 176 {
 177         struct inet_sock *inet;
 178         /* We may need to bind the socket. */
 179         lock_sock(sk);
 180         inet = inet_sk(sk);
 181         if (!inet->inet_num) {
 182                 if (sk->sk_prot->get_port(sk, 0)) {
 183                         release_sock(sk);
 184                         return -EAGAIN;
 185                 }
 186                 inet->inet_sport = htons(inet->inet_num);
 187         }
 188         release_sock(sk);
 189         return 0;
 190 }
 191
 192 /*
 193  *      Move a socket into listening state.
 194  */
 195 int inet_listen(struct socket *sock, int backlog)
 196 {
 197         struct sock *sk = sock->sk;
 198         unsigned char old_state;
 199         int err;
 200
 201         lock_sock(sk);
 202
 203         err = -EINVAL;
 204         if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
 205                 goto out;
 206
 207         old_state = sk->sk_state;
 208         if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
 209                 goto out;
 210
 211         /* Really, if the socket is already in listen state
 212          * we can only allow the backlog to be adjusted.
 213          */
 214         if (old_state != TCP_LISTEN) {
 215                 /* Check special setups for testing purpose to enable TFO w/o
 216                  * requiring TCP_FASTOPEN sockopt.
 217                  * Note that only TCP sockets (SOCK_STREAM) will reach here.
 218                  * Also fastopenq may already been allocated because this
 219                  * socket was in TCP_LISTEN state previously but was
 220                  * shutdown() (rather than close()).
 221                  */
 222                 if ((sysctl_tcp_fastopen & TFO_SERVER_ENABLE) != 0 &&
 223                     inet_csk(sk)->icsk_accept_queue.fastopenq == NULL) {
 224                         if ((sysctl_tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) != 0)
 225                                 err = fastopen_init_queue(sk, backlog);
 226                         else if ((sysctl_tcp_fastopen &
 227                                   TFO_SERVER_WO_SOCKOPT2) != 0)
 228                                 err = fastopen_init_queue(sk,
 229                                     ((uint)sysctl_tcp_fastopen) >> 16);
 230                         else
 231                                 err = 0;
 232                         if (err)
 233                                 goto out;
 234                 }
 235                 err = inet_csk_listen_start(sk, backlog);
 236                 if (err)
 237                         goto out;
 238         }
 239         sk->sk_max_ack_backlog = backlog;
 240         err = 0;
 241
 242 out:
 243         release_sock(sk);
 244         return err;
 245 }
 246 EXPORT_SYMBOL(inet_listen);
 247
 248 /*
 249  *      Create an inet socket.
 250  */
 251
 252 static int inet_create(struct net *net, struct socket *sock, int protocol,
 253                        int kern)
 254 {
 255         struct sock *sk;
 256         struct inet_protosw *answer;
 257         struct inet_sock *inet;
 258         struct proto *answer_prot;
 259         unsigned char answer_flags;
 260         char answer_no_check;
 261         int try_loading_module = 0;
 262         int err;
 263
 264         sock->state = SS_UNCONNECTED;
 265
 266         /* Look for the requested type/protocol pair. */
 267 lookup_protocol:
 268         err = -ESOCKTNOSUPPORT;
 269         rcu_read_lock();
 270         list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
 271
 272                 err = 0;
 273                 /* Check the non-wild match. */
 274                 if (protocol == answer->protocol) {
 275                         if (protocol != IPPROTO_IP)
 276                                 break;
 277                 } else {
 278                         /* Check for the two wild cases. */
 279                         if (IPPROTO_IP == protocol) {
 280                                 protocol = answer->protocol;
 281                                 break;
 282                         }
 283                         if (IPPROTO_IP == answer->protocol)
 284                                 break;
 285                 }
 286                 err = -EPROTONOSUPPORT;
 287         }
 288
 289         if (unlikely(err)) {
 290                 if (try_loading_module < 2) {
 291                         rcu_read_unlock();
 292                         /*
 293                          * Be more specific, e.g. net-pf-2-proto-132-type-1
 294                          * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
 295                          */
 296                         if (++try_loading_module == 1)
 297                                 request_module("net-pf-%d-proto-%d-type-%d",
 298                                                PF_INET, protocol, sock->type);
 299                         /*
 300                          * Fall back to generic, e.g. net-pf-2-proto-132
 301                          * (net-pf-PF_INET-proto-IPPROTO_SCTP)
 302                          */
 303                         else
 304                                 request_module("net-pf-%d-proto-%d",
 305                                                PF_INET, protocol);
 306                         goto lookup_protocol;
 307                 } else
 308                         goto out_rcu_unlock;
 309         }
 310
 311         err = -EPERM;
 312         if (sock->type == SOCK_RAW && !kern &&
 313             !ns_capable(net->user_ns, CAP_NET_RAW))
 314                 goto out_rcu_unlock;
 315
 316         sock->ops = answer->ops;
 317         answer_prot = answer->prot;
 318         answer_no_check = answer->no_check;
 319         answer_flags = answer->flags;
 320         rcu_read_unlock();
 321
 322         WARN_ON(answer_prot->slab == NULL);
 323
 324         err = -ENOBUFS;
 325         sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
 326         if (sk == NULL)
 327                 goto out;
 328
 329         err = 0;
 330         sk->sk_no_check = answer_no_check;
 331         if (INET_PROTOSW_REUSE & answer_flags)
 332                 sk->sk_reuse = SK_CAN_REUSE;
 333
 334         inet = inet_sk(sk);
 335         inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
 336
 337         inet->nodefrag = 0;
 338
 339         if (SOCK_RAW == sock->type) {
 340                 inet->inet_num = protocol;
 341                 if (IPPROTO_RAW == protocol)
 342                         inet->hdrincl = 1;
 343         }
 344
 345         if (ipv4_config.no_pmtu_disc)
 346                 inet->pmtudisc = IP_PMTUDISC_DONT;
 347         else
 348                 inet->pmtudisc = IP_PMTUDISC_WANT;
 349
 350         inet->inet_id = 0;
 351
 352         sock_init_data(sock, sk);
 353
 354         sk->sk_destruct    = inet_sock_destruct;
 355         sk->sk_protocol    = protocol;
 356         sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
 357
 358         inet->uc_ttl    = -1;
 359         inet->mc_loop   = 1;
 360         inet->mc_ttl    = 1;
 361         inet->mc_all    = 1;
 362         inet->mc_index  = 0;
 363         inet->mc_list   = NULL;
 364         inet->rcv_tos   = 0;
 365
 366         sk_refcnt_debug_inc(sk);
 367
 368         if (inet->inet_num) {
 369                 /* It assumes that any protocol which allows
 370                  * the user to assign a number at socket
 371                  * creation time automatically
 372                  * shares.
 373                  */
 374                 inet->inet_sport = htons(inet->inet_num);
 375                 /* Add to protocol hash chains. */
 376                 sk->sk_prot->hash(sk);
 377         }
 378
 379         if (sk->sk_prot->init) {
 380                 err = sk->sk_prot->init(sk);
 381                 if (err)
 382                         sk_common_release(sk);
 383         }
 384 out:
 385         return err;
 386 out_rcu_unlock:
 387         rcu_read_unlock();
 388         goto out;
 389 }
 390
 391
 392 /*
 393  *      The peer socket should always be NULL (or else). When we call this
 394  *      function we are destroying the object and from then on nobody
 395  *      should refer to it.
 396  */
 397 int inet_release(struct socket *sock)
 398 {
 399         struct sock *sk = sock->sk;
 400
 401         if (sk) {
 402                 long timeout;
 403
 404                 sock_rps_reset_flow(sk);
 405
 406                 /* Applications forget to leave groups before exiting */
 407                 ip_mc_drop_socket(sk);
 408
 409                 /* If linger is set, we don't return until the close
 410                  * is complete.  Otherwise we return immediately. The
 411                  * actually closing is done the same either way.
 412                  *
 413                  * If the close is due to the process exiting, we never
 414                  * linger..
 415                  */
 416                 timeout = 0;
 417                 if (sock_flag(sk, SOCK_LINGER) &&
 418                     !(current->flags & PF_EXITING))
 419                         timeout = sk->sk_lingertime;
 420                 sock->sk = NULL;
 421                 sk->sk_prot->close(sk, timeout);
 422         }
 423         return 0;
 424 }
 425 EXPORT_SYMBOL(inet_release);
 426
 427 /* It is off by default, see below. */
 428 int sysctl_ip_nonlocal_bind __read_mostly;
 429 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
 430
 431 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 432 {
 433         struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
 434         struct sock *sk = sock->sk;
 435         struct inet_sock *inet = inet_sk(sk);
 436         struct net *net = sock_net(sk);
 437         unsigned short snum;
 438         int chk_addr_ret;
 439         int err;
 440
 441         /* If the socket has its own bind function then use it. (RAW) */
 442         if (sk->sk_prot->bind) {
 443                 err = sk->sk_prot->bind(sk, uaddr, addr_len);
 444                 goto out;
 445         }
 446         err = -EINVAL;
 447         if (addr_len < sizeof(struct sockaddr_in))
 448                 goto out;
 449
 450         if (addr->sin_family != AF_INET) {
 451                 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
 452                  * only if s_addr is INADDR_ANY.
 453                  */
 454                 err = -EAFNOSUPPORT;
 455                 if (addr->sin_family != AF_UNSPEC ||
 456                     addr->sin_addr.s_addr != htonl(INADDR_ANY))
 457                         goto out;
 458         }
 459
 460         chk_addr_ret = inet_addr_type(net, addr->sin_addr.s_addr);
 461
 462         /* Not specified by any standard per-se, however it breaks too
 463          * many applications when removed.  It is unfortunate since
 464          * allowing applications to make a non-local bind solves
 465          * several problems with systems using dynamic addressing.
 466          * (ie. your servers still start up even if your ISDN link
 467          *  is temporarily down)
 468          */
 469         err = -EADDRNOTAVAIL;
 470         if (!sysctl_ip_nonlocal_bind &&
 471             !(inet->freebind || inet->transparent) &&
 472             addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
 473             chk_addr_ret != RTN_LOCAL &&
 474             chk_addr_ret != RTN_MULTICAST &&
 475             chk_addr_ret != RTN_BROADCAST)
 476                 goto out;
 477
 478         snum = ntohs(addr->sin_port);
 479         err = -EACCES;
 480         if (snum && snum < PROT_SOCK &&
 481             !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
 482                 goto out;
 483
 484         /*      We keep a pair of addresses. rcv_saddr is the one
 485          *      used by hash lookups, and saddr is used for transmit.
 486          *
 487          *      In the BSD API these are the same except where it
 488          *      would be illegal to use them (multicast/broadcast) in
 489          *      which case the sending device address is used.
 490          */
 491         lock_sock(sk);
 492
 493         /* Check these errors (active socket, double bind). */
 494         err = -EINVAL;
 495         if (sk->sk_state != TCP_CLOSE || inet->inet_num)
 496                 goto out_release_sock;
 497
 498         inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
 499         if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
 500                 inet->inet_saddr = 0;  /* Use device */
 501
 502         /* Make sure we are allowed to bind here. */
 503         if (sk->sk_prot->get_port(sk, snum)) {
 504                 inet->inet_saddr = inet->inet_rcv_saddr = 0;
 505                 err = -EADDRINUSE;
 506                 goto out_release_sock;
 507         }
 508
 509         if (inet->inet_rcv_saddr)
 510                 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
 511         if (snum)
 512                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
 513         inet->inet_sport = htons(inet->inet_num);
 514         inet->inet_daddr = 0;
 515         inet->inet_dport = 0;
 516         sk_dst_reset(sk);
 517         err = 0;
 518 out_release_sock:
 519         release_sock(sk);
 520 out:
 521         return err;
 522 }
 523 EXPORT_SYMBOL(inet_bind);
 524
 525 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
 526                        int addr_len, int flags)
 527 {
 528         struct sock *sk = sock->sk;
 529
 530         if (addr_len < sizeof(uaddr->sa_family))
 531                 return -EINVAL;
 532         if (uaddr->sa_family == AF_UNSPEC)
 533                 return sk->sk_prot->disconnect(sk, flags);
 534
 535         if (!inet_sk(sk)->inet_num && inet_autobind(sk))
 536                 return -EAGAIN;
 537         return sk->sk_prot->connect(sk, uaddr, addr_len);
 538 }
 539 EXPORT_SYMBOL(inet_dgram_connect);
 540
 541 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
 542 {
 543         DEFINE_WAIT(wait);
 544
 545         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
 546         sk->sk_write_pending += writebias;
 547
 548         /* Basic assumption: if someone sets sk->sk_err, he _must_
 549          * change state of the socket from TCP_SYN_*.
 550          * Connect() does not allow to get error notifications
 551          * without closing the socket.
 552          */
 553         while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 554                 release_sock(sk);
 555                 timeo = schedule_timeout(timeo);
 556                 lock_sock(sk);
 557                 if (signal_pending(current) || !timeo)
 558                         break;
 559                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
 560         }
 561         finish_wait(sk_sleep(sk), &wait);
 562         sk->sk_write_pending -= writebias;
 563         return timeo;
 564 }
 565
 566 /*
 567  *      Connect to a remote host. There is regrettably still a little
 568  *      TCP 'magic' in here.
 569  */
 570 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 571                           int addr_len, int flags)
 572 {
 573         struct sock *sk = sock->sk;
 574         int err;
 575         long timeo;
 576
 577         if (addr_len < sizeof(uaddr->sa_family))
 578                 return -EINVAL;
 579
 580         if (uaddr->sa_family == AF_UNSPEC) {
 581                 err = sk->sk_prot->disconnect(sk, flags);
 582                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 583                 goto out;
 584         }
 585
 586         switch (sock->state) {
 587         default:
 588                 err = -EINVAL;
 589                 goto out;
 590         case SS_CONNECTED:
 591                 err = -EISCONN;
 592                 goto out;
 593         case SS_CONNECTING:
 594                 err = -EALREADY;
 595                 /* Fall out of switch with err, set for this state */
 596                 break;
 597         case SS_UNCONNECTED:
 598                 err = -EISCONN;
 599                 if (sk->sk_state != TCP_CLOSE)
 600                         goto out;
 601
 602                 err = sk->sk_prot->connect(sk, uaddr, addr_len);
 603                 if (err < 0)
 604                         goto out;
 605
 606                 sock->state = SS_CONNECTING;
 607
 608                 /* Just entered SS_CONNECTING state; the only
 609                  * difference is that return value in non-blocking
 610                  * case is EINPROGRESS, rather than EALREADY.
 611                  */
 612                 err = -EINPROGRESS;
 613                 break;
 614         }
 615
 616         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
 617
 618         if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 619                 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
 620                                 tcp_sk(sk)->fastopen_req &&
 621                                 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
 622
 623                 /* Error code is set above */
 624                 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
 625                         goto out;
 626
 627                 err = sock_intr_errno(timeo);
 628                 if (signal_pending(current))
 629                         goto out;
 630         }
 631
 632         /* Connection was closed by RST, timeout, ICMP error
 633          * or another process disconnected us.
 634          */
 635         if (sk->sk_state == TCP_CLOSE)
 636                 goto sock_error;
 637
 638         /* sk->sk_err may be not zero now, if RECVERR was ordered by user
 639          * and error was received after socket entered established state.
 640          * Hence, it is handled normally after connect() return successfully.
 641          */
 642
 643         sock->state = SS_CONNECTED;
 644         err = 0;
 645 out:
 646         return err;
 647
 648 sock_error:
 649         err = sock_error(sk) ? : -ECONNABORTED;
 650         sock->state = SS_UNCONNECTED;
 651         if (sk->sk_prot->disconnect(sk, flags))
 652                 sock->state = SS_DISCONNECTING;
 653         goto out;
 654 }
 655 EXPORT_SYMBOL(__inet_stream_connect);
 656
 657 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 658                         int addr_len, int flags)
 659 {
 660         int err;
 661
 662         lock_sock(sock->sk);
 663         err = __inet_stream_connect(sock, uaddr, addr_len, flags);
 664         release_sock(sock->sk);
 665         return err;
 666 }
 667 EXPORT_SYMBOL(inet_stream_connect);
 668
 669 /*
 670  *      Accept a pending connection. The TCP layer now gives BSD semantics.
 671  */
 672
 673 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
 674 {
 675         struct sock *sk1 = sock->sk;
 676         int err = -EINVAL;
 677         struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
 678
 679         if (!sk2)
 680                 goto do_err;
 681
 682         lock_sock(sk2);
 683
 684         sock_rps_record_flow(sk2);
 685         WARN_ON(!((1 << sk2->sk_state) &
 686                   (TCPF_ESTABLISHED | TCPF_SYN_RECV |
 687                   TCPF_CLOSE_WAIT | TCPF_CLOSE)));
 688
 689         sock_graft(sk2, newsock);
 690
 691         newsock->state = SS_CONNECTED;
 692         err = 0;
 693         release_sock(sk2);
 694 do_err:
 695         return err;
 696 }
 697 EXPORT_SYMBOL(inet_accept);
 698
 699
 700 /*
 701  *      This does both peername and sockname.
 702  */
 703 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
 704                         int *uaddr_len, int peer)
 705 {
 706         struct sock *sk         = sock->sk;
 707         struct inet_sock *inet  = inet_sk(sk);
 708         DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
 709
 710         sin->sin_family = AF_INET;
 711         if (peer) {
 712                 if (!inet->inet_dport ||
 713                     (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
 714                      peer == 1))
 715                         return -ENOTCONN;
 716                 sin->sin_port = inet->inet_dport;
 717                 sin->sin_addr.s_addr = inet->inet_daddr;
 718         } else {
 719                 __be32 addr = inet->inet_rcv_saddr;
 720                 if (!addr)
 721                         addr = inet->inet_saddr;
 722                 sin->sin_port = inet->inet_sport;
 723                 sin->sin_addr.s_addr = addr;
 724         }
 725         memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 726         *uaddr_len = sizeof(*sin);
 727         return 0;
 728 }
 729 EXPORT_SYMBOL(inet_getname);
 730
 731 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
 732                  size_t size)
 733 {
 734         struct sock *sk = sock->sk;
 735
 736         sock_rps_record_flow(sk);
 737
 738         /* We may need to bind the socket. */
 739         if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
 740             inet_autobind(sk))
 741                 return -EAGAIN;
 742
 743         return sk->sk_prot->sendmsg(iocb, sk, msg, size);
 744 }
 745 EXPORT_SYMBOL(inet_sendmsg);
 746
 747 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
 748                       size_t size, int flags)
 749 {
 750         struct sock *sk = sock->sk;
 751
 752         sock_rps_record_flow(sk);
 753
 754         /* We may need to bind the socket. */
 755         if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
 756             inet_autobind(sk))
 757                 return -EAGAIN;
 758
 759         if (sk->sk_prot->sendpage)
 760                 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
 761         return sock_no_sendpage(sock, page, offset, size, flags);
 762 }
 763 EXPORT_SYMBOL(inet_sendpage);
 764
 765 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
 766                  size_t size, int flags)
 767 {
 768         struct sock *sk = sock->sk;
 769         int addr_len = 0;
 770         int err;
 771
 772         sock_rps_record_flow(sk);
 773
 774         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
 775                                    flags & ~MSG_DONTWAIT, &addr_len);
 776         if (err >= 0)
 777                 msg->msg_namelen = addr_len;
 778         return err;
 779 }
 780 EXPORT_SYMBOL(inet_recvmsg);
 781
 782 int inet_shutdown(struct socket *sock, int how)
 783 {
 784         struct sock *sk = sock->sk;
 785         int err = 0;
 786
 787         /* This should really check to make sure
 788          * the socket is a TCP socket. (WHY AC...)
 789          */
 790         how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
 791                        1->2 bit 2 snds.
 792                        2->3 */
 793         if ((how & ~SHUTDOWN_MASK) || !how)     /* MAXINT->0 */
 794                 return -EINVAL;
 795
 796         lock_sock(sk);
 797         if (sock->state == SS_CONNECTING) {
 798                 if ((1 << sk->sk_state) &
 799                     (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
 800                         sock->state = SS_DISCONNECTING;
 801                 else
 802                         sock->state = SS_CONNECTED;
 803         }
 804
 805         switch (sk->sk_state) {
 806         case TCP_CLOSE:
 807                 err = -ENOTCONN;
 808                 /* Hack to wake up other listeners, who can poll for
 809                    POLLHUP, even on eg. unconnected UDP sockets -- RR */
 810         default:
 811                 sk->sk_shutdown |= how;
 812                 if (sk->sk_prot->shutdown)
 813                         sk->sk_prot->shutdown(sk, how);
 814                 break;
 815
 816         /* Remaining two branches are temporary solution for missing
 817          * close() in multithreaded environment. It is _not_ a good idea,
 818          * but we have no choice until close() is repaired at VFS level.
 819          */
 820         case TCP_LISTEN:
 821                 if (!(how & RCV_SHUTDOWN))
 822                         break;
 823                 /* Fall through */
 824         case TCP_SYN_SENT:
 825                 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
 826                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 827                 break;
 828         }
 829
 830         /* Wake up anyone sleeping in poll. */
 831         sk->sk_state_change(sk);
 832         release_sock(sk);
 833         return err;
 834 }
 835 EXPORT_SYMBOL(inet_shutdown);
 836
 837 /*
 838  *      ioctl() calls you can issue on an INET socket. Most of these are
 839  *      device configuration and stuff and very rarely used. Some ioctls
 840  *      pass on to the socket itself.
 841  *
 842  *      NOTE: I like the idea of a module for the config stuff. ie ifconfig
 843  *      loads the devconfigure module does its configuring and unloads it.
 844  *      There's a good 20K of config code hanging around the kernel.
 845  */
 846
 847 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 848 {
 849         struct sock *sk = sock->sk;
 850         int err = 0;
 851         struct net *net = sock_net(sk);
 852
 853         switch (cmd) {
 854         case SIOCGSTAMP:
 855                 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
 856                 break;
 857         case SIOCGSTAMPNS:
 858                 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
 859                 break;
 860         case SIOCADDRT:
 861         case SIOCDELRT:
 862         case SIOCRTMSG:
 863                 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
 864                 break;
 865         case SIOCDARP:
 866         case SIOCGARP:
 867         case SIOCSARP:
 868                 err = arp_ioctl(net, cmd, (void __user *)arg);
 869                 break;
 870         case SIOCGIFADDR:
 871         case SIOCSIFADDR:
 872         case SIOCGIFBRDADDR:
 873         case SIOCSIFBRDADDR:
 874         case SIOCGIFNETMASK:
 875         case SIOCSIFNETMASK:
 876         case SIOCGIFDSTADDR:
 877         case SIOCSIFDSTADDR:
 878         case SIOCSIFPFLAGS:
 879         case SIOCGIFPFLAGS:
 880         case SIOCSIFFLAGS:
 881                 err = devinet_ioctl(net, cmd, (void __user *)arg);
 882                 break;
 883         default:
 884                 if (sk->sk_prot->ioctl)
 885                         err = sk->sk_prot->ioctl(sk, cmd, arg);
 886                 else
 887                         err = -ENOIOCTLCMD;
 888                 break;
 889         }
 890         return err;
 891 }
 892 EXPORT_SYMBOL(inet_ioctl);
 893
 894 #ifdef CONFIG_COMPAT
 895 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 896 {
 897         struct sock *sk = sock->sk;
 898         int err = -ENOIOCTLCMD;
 899
 900         if (sk->sk_prot->compat_ioctl)
 901                 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
 902
 903         return err;
 904 }
 905 #endif
 906
 907 const struct proto_ops inet_stream_ops = {
 908         .family            = PF_INET,
 909         .owner             = THIS_MODULE,
 910         .release           = inet_release,
 911         .bind              = inet_bind,
 912         .connect           = inet_stream_connect,
 913         .socketpair        = sock_no_socketpair,
 914         .accept            = inet_accept,
 915         .getname           = inet_getname,
 916         .poll              = tcp_poll,
 917         .ioctl             = inet_ioctl,
 918         .listen            = inet_listen,
 919         .shutdown          = inet_shutdown,
 920         .setsockopt        = sock_common_setsockopt,
 921         .getsockopt        = sock_common_getsockopt,
 922         .sendmsg           = inet_sendmsg,
 923         .recvmsg           = inet_recvmsg,
 924         .mmap              = sock_no_mmap,
 925         .sendpage          = inet_sendpage,
 926         .splice_read       = tcp_splice_read,
 927 #ifdef CONFIG_COMPAT
 928         .compat_setsockopt = compat_sock_common_setsockopt,
 929         .compat_getsockopt = compat_sock_common_getsockopt,
 930         .compat_ioctl      = inet_compat_ioctl,
 931 #endif
 932 };
 933 EXPORT_SYMBOL(inet_stream_ops);
 934
 935 const struct proto_ops inet_dgram_ops = {
 936         .family            = PF_INET,
 937         .owner             = THIS_MODULE,
 938         .release           = inet_release,
 939         .bind              = inet_bind,
 940         .connect           = inet_dgram_connect,
 941         .socketpair        = sock_no_socketpair,
 942         .accept            = sock_no_accept,
 943         .getname           = inet_getname,
 944         .poll              = udp_poll,
 945         .ioctl             = inet_ioctl,
 946         .listen            = sock_no_listen,
 947         .shutdown          = inet_shutdown,
 948         .setsockopt        = sock_common_setsockopt,
 949         .getsockopt        = sock_common_getsockopt,
 950         .sendmsg           = inet_sendmsg,
 951         .recvmsg           = inet_recvmsg,
 952         .mmap              = sock_no_mmap,
 953         .sendpage          = inet_sendpage,
 954 #ifdef CONFIG_COMPAT
 955         .compat_setsockopt = compat_sock_common_setsockopt,
 956         .compat_getsockopt = compat_sock_common_getsockopt,
 957         .compat_ioctl      = inet_compat_ioctl,
 958 #endif
 959 };
 960 EXPORT_SYMBOL(inet_dgram_ops);
 961
 962 /*
 963  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
 964  * udp_poll
 965  */
 966 static const struct proto_ops inet_sockraw_ops = {
 967         .family            = PF_INET,
 968         .owner             = THIS_MODULE,
 969         .release           = inet_release,
 970         .bind              = inet_bind,
 971         .connect           = inet_dgram_connect,
 972         .socketpair        = sock_no_socketpair,
 973         .accept            = sock_no_accept,
 974         .getname           = inet_getname,
 975         .poll              = datagram_poll,
 976         .ioctl             = inet_ioctl,
 977         .listen            = sock_no_listen,
 978         .shutdown          = inet_shutdown,
 979         .setsockopt        = sock_common_setsockopt,
 980         .getsockopt        = sock_common_getsockopt,
 981         .sendmsg           = inet_sendmsg,
 982         .recvmsg           = inet_recvmsg,
 983         .mmap              = sock_no_mmap,
 984         .sendpage          = inet_sendpage,
 985 #ifdef CONFIG_COMPAT
 986         .compat_setsockopt = compat_sock_common_setsockopt,
 987         .compat_getsockopt = compat_sock_common_getsockopt,
 988         .compat_ioctl      = inet_compat_ioctl,
 989 #endif
 990 };
 991
 992 static const struct net_proto_family inet_family_ops = {
 993         .family = PF_INET,
 994         .create = inet_create,
 995         .owner  = THIS_MODULE,
 996 };
 997
 998 /* Upon startup we insert all the elements in inetsw_array[] into
 999  * the linked list inetsw.
1000  */
1001 static struct inet_protosw inetsw_array[] =
1002 {
1003         {
1004                 .type =       SOCK_STREAM,
1005                 .protocol =   IPPROTO_TCP,
1006                 .prot =       &tcp_prot,
1007                 .ops =        &inet_stream_ops,
1008                 .no_check =   0,
1009                 .flags =      INET_PROTOSW_PERMANENT |
1010                               INET_PROTOSW_ICSK,
1011         },
1012
1013         {
1014                 .type =       SOCK_DGRAM,
1015                 .protocol =   IPPROTO_UDP,
1016                 .prot =       &udp_prot,
1017                 .ops =        &inet_dgram_ops,
1018                 .no_check =   UDP_CSUM_DEFAULT,
1019                 .flags =      INET_PROTOSW_PERMANENT,
1020        },
1021
1022        {
1023                 .type =       SOCK_DGRAM,
1024                 .protocol =   IPPROTO_ICMP,
1025                 .prot =       &ping_prot,
1026                 .ops =        &inet_dgram_ops,
1027                 .no_check =   UDP_CSUM_DEFAULT,
1028                 .flags =      INET_PROTOSW_REUSE,
1029        },
1030
1031        {
1032                .type =       SOCK_RAW,
1033                .protocol =   IPPROTO_IP,        /* wild card */
1034                .prot =       &raw_prot,
1035                .ops =        &inet_sockraw_ops,
1036                .no_check =   UDP_CSUM_DEFAULT,
1037                .flags =      INET_PROTOSW_REUSE,
1038        }
1039 };
1040
1041 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1042
1043 void inet_register_protosw(struct inet_protosw *p)
1044 {
1045         struct list_head *lh;
1046         struct inet_protosw *answer;
1047         int protocol = p->protocol;
1048         struct list_head *last_perm;
1049
1050         spin_lock_bh(&inetsw_lock);
1051
1052         if (p->type >= SOCK_MAX)
1053                 goto out_illegal;
1054
1055         /* If we are trying to override a permanent protocol, bail. */
1056         answer = NULL;
1057         last_perm = &inetsw[p->type];
1058         list_for_each(lh, &inetsw[p->type]) {
1059                 answer = list_entry(lh, struct inet_protosw, list);
1060
1061                 /* Check only the non-wild match. */
1062                 if (INET_PROTOSW_PERMANENT & answer->flags) {
1063                         if (protocol == answer->protocol)
1064                                 break;
1065                         last_perm = lh;
1066                 }
1067
1068                 answer = NULL;
1069         }
1070         if (answer)
1071                 goto out_permanent;
1072
1073         /* Add the new entry after the last permanent entry if any, so that
1074          * the new entry does not override a permanent entry when matched with
1075          * a wild-card protocol. But it is allowed to override any existing
1076          * non-permanent entry.  This means that when we remove this entry, the
1077          * system automatically returns to the old behavior.
1078          */
1079         list_add_rcu(&p->list, last_perm);
1080 out:
1081         spin_unlock_bh(&inetsw_lock);
1082
1083         return;
1084
1085 out_permanent:
1086         pr_err("Attempt to override permanent protocol %d\n", protocol);
1087         goto out;
1088
1089 out_illegal:
1090         pr_err("Ignoring attempt to register invalid socket type %d\n",
1091                p->type);
1092         goto out;
1093 }
1094 EXPORT_SYMBOL(inet_register_protosw);
1095
1096 void inet_unregister_protosw(struct inet_protosw *p)
1097 {
1098         if (INET_PROTOSW_PERMANENT & p->flags) {
1099                 pr_err("Attempt to unregister permanent protocol %d\n",
1100                        p->protocol);
1101         } else {
1102                 spin_lock_bh(&inetsw_lock);
1103                 list_del_rcu(&p->list);
1104                 spin_unlock_bh(&inetsw_lock);
1105
1106                 synchronize_net();
1107         }
1108 }
1109 EXPORT_SYMBOL(inet_unregister_protosw);
1110
1111 /*
1112  *      Shall we try to damage output packets if routing dev changes?
1113  */
1114
1115 int sysctl_ip_dynaddr __read_mostly;
1116
1117 static int inet_sk_reselect_saddr(struct sock *sk)
1118 {
1119         struct inet_sock *inet = inet_sk(sk);
1120         __be32 old_saddr = inet->inet_saddr;
1121         __be32 daddr = inet->inet_daddr;
1122         struct flowi4 *fl4;
1123         struct rtable *rt;
1124         __be32 new_saddr;
1125         struct ip_options_rcu *inet_opt;
1126
1127         inet_opt = rcu_dereference_protected(inet->inet_opt,
1128                                              sock_owned_by_user(sk));
1129         if (inet_opt && inet_opt->opt.srr)
1130                 daddr = inet_opt->opt.faddr;
1131
1132         /* Query new route. */
1133         fl4 = &inet->cork.fl.u.ip4;
1134         rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1135                               sk->sk_bound_dev_if, sk->sk_protocol,
1136                               inet->inet_sport, inet->inet_dport, sk, false);
1137         if (IS_ERR(rt))
1138                 return PTR_ERR(rt);
1139
1140         sk_setup_caps(sk, &rt->dst);
1141
1142         new_saddr = fl4->saddr;
1143
1144         if (new_saddr == old_saddr)
1145                 return 0;
1146
1147         if (sysctl_ip_dynaddr > 1) {
1148                 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1149                         __func__, &old_saddr, &new_saddr);
1150         }
1151
1152         inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1153
1154         /*
1155          * XXX The only one ugly spot where we need to
1156          * XXX really change the sockets identity after
1157          * XXX it has entered the hashes. -DaveM
1158          *
1159          * Besides that, it does not check for connection
1160          * uniqueness. Wait for troubles.
1161          */
1162         __sk_prot_rehash(sk);
1163         return 0;
1164 }
1165
1166 int inet_sk_rebuild_header(struct sock *sk)
1167 {
1168         struct inet_sock *inet = inet_sk(sk);
1169         struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1170         __be32 daddr;
1171         struct ip_options_rcu *inet_opt;
1172         struct flowi4 *fl4;
1173         int err;
1174
1175         /* Route is OK, nothing to do. */
1176         if (rt)
1177                 return 0;
1178
1179         /* Reroute. */
1180         rcu_read_lock();
1181         inet_opt = rcu_dereference(inet->inet_opt);
1182         daddr = inet->inet_daddr;
1183         if (inet_opt && inet_opt->opt.srr)
1184                 daddr = inet_opt->opt.faddr;
1185         rcu_read_unlock();
1186         fl4 = &inet->cork.fl.u.ip4;
1187         rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1188                                    inet->inet_dport, inet->inet_sport,
1189                                    sk->sk_protocol, RT_CONN_FLAGS(sk),
1190                                    sk->sk_bound_dev_if);
1191         if (!IS_ERR(rt)) {
1192                 err = 0;
1193                 sk_setup_caps(sk, &rt->dst);
1194         } else {
1195                 err = PTR_ERR(rt);
1196
1197                 /* Routing failed... */
1198                 sk->sk_route_caps = 0;
1199                 /*
1200                  * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1201                  * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1202                  */
1203                 if (!sysctl_ip_dynaddr ||
1204                     sk->sk_state != TCP_SYN_SENT ||
1205                     (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1206                     (err = inet_sk_reselect_saddr(sk)) != 0)
1207                         sk->sk_err_soft = -err;
1208         }
1209
1210         return err;
1211 }
1212 EXPORT_SYMBOL(inet_sk_rebuild_header);
1213
1214 static int inet_gso_send_check(struct sk_buff *skb)
1215 {
1216         const struct net_offload *ops;
1217         const struct iphdr *iph;
1218         int proto;
1219         int ihl;
1220         int err = -EINVAL;
1221
1222         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1223                 goto out;
1224
1225         iph = ip_hdr(skb);
1226         ihl = iph->ihl * 4;
1227         if (ihl < sizeof(*iph))
1228                 goto out;
1229
1230         proto = iph->protocol;
1231
1232         /* Warning: after this point, iph might be no longer valid */
1233         if (unlikely(!pskb_may_pull(skb, ihl)))
1234                 goto out;
1235         __skb_pull(skb, ihl);
1236
1237         skb_reset_transport_header(skb);
1238         err = -EPROTONOSUPPORT;
1239
1240         ops = rcu_dereference(inet_offloads[proto]);
1241         if (likely(ops && ops->callbacks.gso_send_check))
1242                 err = ops->callbacks.gso_send_check(skb);
1243
1244 out:
1245         return err;
1246 }
1247
1248 static struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1249                                         netdev_features_t features)
1250 {
1251         struct sk_buff *segs = ERR_PTR(-EINVAL);
1252         const struct net_offload *ops;
1253         unsigned int offset = 0;
1254         bool udpfrag, encap;
1255         struct iphdr *iph;
1256         int proto;
1257         int nhoff;
1258         int ihl;
1259         int id;
1260
1261         if (unlikely(skb_shinfo(skb)->gso_type &
1262                      ~(SKB_GSO_TCPV4 |
1263                        SKB_GSO_UDP |
1264                        SKB_GSO_DODGY |
1265                        SKB_GSO_TCP_ECN |
1266                        SKB_GSO_GRE |
1267                        SKB_GSO_IPIP |
1268                        SKB_GSO_SIT |
1269                        SKB_GSO_TCPV6 |
1270                        SKB_GSO_UDP_TUNNEL |
1271                        SKB_GSO_MPLS |
1272                        0)))
1273                 goto out;
1274
1275         skb_reset_network_header(skb);
1276         nhoff = skb_network_header(skb) - skb_mac_header(skb);
1277         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1278                 goto out;
1279
1280         iph = ip_hdr(skb);
1281         ihl = iph->ihl * 4;
1282         if (ihl < sizeof(*iph))
1283                 goto out;
1284
1285         id = ntohs(iph->id);
1286         proto = iph->protocol;
1287
1288         /* Warning: after this point, iph might be no longer valid */
1289         if (unlikely(!pskb_may_pull(skb, ihl)))
1290                 goto out;
1291         __skb_pull(skb, ihl);
1292
1293         encap = SKB_GSO_CB(skb)->encap_level > 0;
1294         if (encap)
1295                 features = skb->dev->hw_enc_features & netif_skb_features(skb);
1296         SKB_GSO_CB(skb)->encap_level += ihl;
1297
1298         skb_reset_transport_header(skb);
1299
1300         segs = ERR_PTR(-EPROTONOSUPPORT);
1301
1302         /* Note : following gso_segment() might change skb->encapsulation */
1303         udpfrag = !skb->encapsulation && proto == IPPROTO_UDP;
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                         (__force int)((iph->frag_off ^ iph2->frag_off) & htons(IP_DF)) |
1398                         ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1399
1400                 NAPI_GRO_CB(p)->flush |= flush;
1401         }
1402
1403         NAPI_GRO_CB(skb)->flush |= flush;
1404         skb_set_network_header(skb, off);
1405         /* The above will be needed by the transport layer if there is one
1406          * immediately following this IP hdr.
1407          */
1408
1409         skb_gro_pull(skb, sizeof(*iph));
1410         skb_set_transport_header(skb, skb_gro_offset(skb));
1411
1412         pp = ops->callbacks.gro_receive(head, skb);
1413
1414 out_unlock:
1415         rcu_read_unlock();
1416
1417 out:
1418         NAPI_GRO_CB(skb)->flush |= flush;
1419
1420         return pp;
1421 }
1422
1423 static int inet_gro_complete(struct sk_buff *skb, int nhoff)
1424 {
1425         __be16 newlen = htons(skb->len - nhoff);
1426         struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1427         const struct net_offload *ops;
1428         int proto = iph->protocol;
1429         int err = -ENOSYS;
1430
1431         csum_replace2(&iph->check, iph->tot_len, newlen);
1432         iph->tot_len = newlen;
1433
1434         rcu_read_lock();
1435         ops = rcu_dereference(inet_offloads[proto]);
1436         if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1437                 goto out_unlock;
1438
1439         /* Only need to add sizeof(*iph) to get to the next hdr below
1440          * because any hdr with option will have been flushed in
1441          * inet_gro_receive().
1442          */
1443         err = ops->callbacks.gro_complete(skb, nhoff + sizeof(*iph));
1444
1445 out_unlock:
1446         rcu_read_unlock();
1447
1448         return err;
1449 }
1450
1451 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1452                          unsigned short type, unsigned char protocol,
1453                          struct net *net)
1454 {
1455         struct socket *sock;
1456         int rc = sock_create_kern(family, type, protocol, &sock);
1457
1458         if (rc == 0) {
1459                 *sk = sock->sk;
1460                 (*sk)->sk_allocation = GFP_ATOMIC;
1461                 /*
1462                  * Unhash it so that IP input processing does not even see it,
1463                  * we do not wish this socket to see incoming packets.
1464                  */
1465                 (*sk)->sk_prot->unhash(*sk);
1466
1467                 sk_change_net(*sk, net);
1468         }
1469         return rc;
1470 }
1471 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1472
1473 unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1474 {
1475         unsigned long res = 0;
1476         int i, j;
1477
1478         for_each_possible_cpu(i) {
1479                 for (j = 0; j < SNMP_ARRAY_SZ; j++)
1480                         res += *(((unsigned long *) per_cpu_ptr(mib[j], i)) + offt);
1481         }
1482         return res;
1483 }
1484 EXPORT_SYMBOL_GPL(snmp_fold_field);
1485
1486 #if BITS_PER_LONG==32
1487
1488 u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset)
1489 {
1490         u64 res = 0;
1491         int cpu;
1492
1493         for_each_possible_cpu(cpu) {
1494                 void *bhptr;
1495                 struct u64_stats_sync *syncp;
1496                 u64 v;
1497                 unsigned int start;
1498
1499                 bhptr = per_cpu_ptr(mib[0], cpu);
1500                 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1501                 do {
1502                         start = u64_stats_fetch_begin_bh(syncp);
1503                         v = *(((u64 *) bhptr) + offt);
1504                 } while (u64_stats_fetch_retry_bh(syncp, start));
1505
1506                 res += v;
1507         }
1508         return res;
1509 }
1510 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1511 #endif
1512
1513 int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align)
1514 {
1515         BUG_ON(ptr == NULL);
1516         ptr[0] = __alloc_percpu(mibsize, align);
1517         if (!ptr[0])
1518                 return -ENOMEM;
1519
1520 #if SNMP_ARRAY_SZ == 2
1521         ptr[1] = __alloc_percpu(mibsize, align);
1522         if (!ptr[1]) {
1523                 free_percpu(ptr[0]);
1524                 ptr[0] = NULL;
1525                 return -ENOMEM;
1526         }
1527 #endif
1528         return 0;
1529 }
1530 EXPORT_SYMBOL_GPL(snmp_mib_init);
1531
1532 #ifdef CONFIG_IP_MULTICAST
1533 static const struct net_protocol igmp_protocol = {
1534         .handler =      igmp_rcv,
1535         .netns_ok =     1,
1536 };
1537 #endif
1538
1539 static const struct net_protocol tcp_protocol = {
1540         .early_demux    =       tcp_v4_early_demux,
1541         .handler        =       tcp_v4_rcv,
1542         .err_handler    =       tcp_v4_err,
1543         .no_policy      =       1,
1544         .netns_ok       =       1,
1545 };
1546
1547 static const struct net_protocol udp_protocol = {
1548         .early_demux =  udp_v4_early_demux,
1549         .handler =      udp_rcv,
1550         .err_handler =  udp_err,
1551         .no_policy =    1,
1552         .netns_ok =     1,
1553 };
1554
1555 static const struct net_protocol icmp_protocol = {
1556         .handler =      icmp_rcv,
1557         .err_handler =  icmp_err,
1558         .no_policy =    1,
1559         .netns_ok =     1,
1560 };
1561
1562 static __net_init int ipv4_mib_init_net(struct net *net)
1563 {
1564         int i;
1565
1566         if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1567                           sizeof(struct tcp_mib),
1568                           __alignof__(struct tcp_mib)) < 0)
1569                 goto err_tcp_mib;
1570         if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1571                           sizeof(struct ipstats_mib),
1572                           __alignof__(struct ipstats_mib)) < 0)
1573                 goto err_ip_mib;
1574
1575         for_each_possible_cpu(i) {
1576                 struct ipstats_mib *af_inet_stats;
1577                 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics[0], i);
1578                 u64_stats_init(&af_inet_stats->syncp);
1579 #if SNMP_ARRAY_SZ == 2
1580                 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics[1], i);
1581                 u64_stats_init(&af_inet_stats->syncp);
1582 #endif
1583         }
1584
1585         if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1586                           sizeof(struct linux_mib),
1587                           __alignof__(struct linux_mib)) < 0)
1588                 goto err_net_mib;
1589         if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1590                           sizeof(struct udp_mib),
1591                           __alignof__(struct udp_mib)) < 0)
1592                 goto err_udp_mib;
1593         if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1594                           sizeof(struct udp_mib),
1595                           __alignof__(struct udp_mib)) < 0)
1596                 goto err_udplite_mib;
1597         if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1598                           sizeof(struct icmp_mib),
1599                           __alignof__(struct icmp_mib)) < 0)
1600                 goto err_icmp_mib;
1601         net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1602                                               GFP_KERNEL);
1603         if (!net->mib.icmpmsg_statistics)
1604                 goto err_icmpmsg_mib;
1605
1606         tcp_mib_init(net);
1607         return 0;
1608
1609 err_icmpmsg_mib:
1610         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1611 err_icmp_mib:
1612         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1613 err_udplite_mib:
1614         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1615 err_udp_mib:
1616         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1617 err_net_mib:
1618         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1619 err_ip_mib:
1620         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1621 err_tcp_mib:
1622         return -ENOMEM;
1623 }
1624
1625 static __net_exit void ipv4_mib_exit_net(struct net *net)
1626 {
1627         kfree(net->mib.icmpmsg_statistics);
1628         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1629         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1630         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1631         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1632         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1633         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1634 }
1635
1636 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1637         .init = ipv4_mib_init_net,
1638         .exit = ipv4_mib_exit_net,
1639 };
1640
1641 static int __init init_ipv4_mibs(void)
1642 {
1643         return register_pernet_subsys(&ipv4_mib_ops);
1644 }
1645
1646 static int ipv4_proc_init(void);
1647
1648 /*
1649  *      IP protocol layer initialiser
1650  */
1651
1652 static struct packet_offload ip_packet_offload __read_mostly = {
1653         .type = cpu_to_be16(ETH_P_IP),
1654         .callbacks = {
1655                 .gso_send_check = inet_gso_send_check,
1656                 .gso_segment = inet_gso_segment,
1657                 .gro_receive = inet_gro_receive,
1658                 .gro_complete = inet_gro_complete,
1659         },
1660 };
1661
1662 static const struct net_offload ipip_offload = {
1663         .callbacks = {
1664                 .gso_send_check = inet_gso_send_check,
1665                 .gso_segment    = inet_gso_segment,
1666         },
1667 };
1668
1669 static int __init ipv4_offload_init(void)
1670 {
1671         /*
1672          * Add offloads
1673          */
1674         if (udpv4_offload_init() < 0)
1675                 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1676         if (tcpv4_offload_init() < 0)
1677                 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1678
1679         dev_add_offload(&ip_packet_offload);
1680         inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1681         return 0;
1682 }
1683
1684 fs_initcall(ipv4_offload_init);
1685
1686 static struct packet_type ip_packet_type __read_mostly = {
1687         .type = cpu_to_be16(ETH_P_IP),
1688         .func = ip_rcv,
1689 };
1690
1691 static int __init inet_init(void)
1692 {
1693         struct inet_protosw *q;
1694         struct list_head *r;
1695         int rc = -EINVAL;
1696
1697         BUILD_BUG_ON(sizeof(struct inet_skb_parm) > FIELD_SIZEOF(struct sk_buff, cb));
1698
1699         sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL);
1700         if (!sysctl_local_reserved_ports)
1701                 goto out;
1702
1703         rc = proto_register(&tcp_prot, 1);
1704         if (rc)
1705                 goto out_free_reserved_ports;
1706
1707         rc = proto_register(&udp_prot, 1);
1708         if (rc)
1709                 goto out_unregister_tcp_proto;
1710
1711         rc = proto_register(&raw_prot, 1);
1712         if (rc)
1713                 goto out_unregister_udp_proto;
1714
1715         rc = proto_register(&ping_prot, 1);
1716         if (rc)
1717                 goto out_unregister_raw_proto;
1718
1719         /*
1720          *      Tell SOCKET that we are alive...
1721          */
1722
1723         (void)sock_register(&inet_family_ops);
1724
1725 #ifdef CONFIG_SYSCTL
1726         ip_static_sysctl_init();
1727 #endif
1728
1729         /*
1730          *      Add all the base protocols.
1731          */
1732
1733         if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1734                 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1735         if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1736                 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1737         if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1738                 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1739 #ifdef CONFIG_IP_MULTICAST
1740         if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1741                 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1742 #endif
1743
1744         /* Register the socket-side information for inet_create. */
1745         for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1746                 INIT_LIST_HEAD(r);
1747
1748         for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1749                 inet_register_protosw(q);
1750
1751         /*
1752          *      Set the ARP module up
1753          */
1754
1755         arp_init();
1756
1757         /*
1758          *      Set the IP module up
1759          */
1760
1761         ip_init();
1762
1763         tcp_v4_init();
1764
1765         /* Setup TCP slab cache for open requests. */
1766         tcp_init();
1767
1768         /* Setup UDP memory threshold */
1769         udp_init();
1770
1771         /* Add UDP-Lite (RFC 3828) */
1772         udplite4_register();
1773
1774         ping_init();
1775
1776         /*
1777          *      Set the ICMP layer up
1778          */
1779
1780         if (icmp_init() < 0)
1781                 panic("Failed to create the ICMP control socket.\n");
1782
1783         /*
1784          *      Initialise the multicast router
1785          */
1786 #if defined(CONFIG_IP_MROUTE)
1787         if (ip_mr_init())
1788                 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1789 #endif
1790         /*
1791          *      Initialise per-cpu ipv4 mibs
1792          */
1793
1794         if (init_ipv4_mibs())
1795                 pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1796
1797         ipv4_proc_init();
1798
1799         ipfrag_init();
1800
1801         dev_add_pack(&ip_packet_type);
1802
1803         rc = 0;
1804 out:
1805         return rc;
1806 out_unregister_raw_proto:
1807         proto_unregister(&raw_prot);
1808 out_unregister_udp_proto:
1809         proto_unregister(&udp_prot);
1810 out_unregister_tcp_proto:
1811         proto_unregister(&tcp_prot);
1812 out_free_reserved_ports:
1813         kfree(sysctl_local_reserved_ports);
1814         goto out;
1815 }
1816
1817 fs_initcall(inet_init);
1818
1819 /* ------------------------------------------------------------------------ */
1820
1821 #ifdef CONFIG_PROC_FS
1822 static int __init ipv4_proc_init(void)
1823 {
1824         int rc = 0;
1825
1826         if (raw_proc_init())
1827                 goto out_raw;
1828         if (tcp4_proc_init())
1829                 goto out_tcp;
1830         if (udp4_proc_init())
1831                 goto out_udp;
1832         if (ping_proc_init())
1833                 goto out_ping;
1834         if (ip_misc_proc_init())
1835                 goto out_misc;
1836 out:
1837         return rc;
1838 out_misc:
1839         ping_proc_exit();
1840 out_ping:
1841         udp4_proc_exit();
1842 out_udp:
1843         tcp4_proc_exit();
1844 out_tcp:
1845         raw_proc_exit();
1846 out_raw:
1847         rc = -ENOMEM;
1848         goto out;
1849 }
1850
1851 #else /* CONFIG_PROC_FS */
1852 static int __init ipv4_proc_init(void)
1853 {
1854         return 0;
1855 }
1856 #endif /* CONFIG_PROC_FS */
1857
1858 MODULE_ALIAS_NETPROTO(PF_INET);
1859