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         int try_loading_module = 0;
 258         int err;
 259
 260         sock->state = SS_UNCONNECTED;
 261
 262         /* Look for the requested type/protocol pair. */
 263 lookup_protocol:
 264         err = -ESOCKTNOSUPPORT;
 265         rcu_read_lock();
 266         list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
 267
 268                 err = 0;
 269                 /* Check the non-wild match. */
 270                 if (protocol == answer->protocol) {
 271                         if (protocol != IPPROTO_IP)
 272                                 break;
 273                 } else {
 274                         /* Check for the two wild cases. */
 275                         if (IPPROTO_IP == protocol) {
 276                                 protocol = answer->protocol;
 277                                 break;
 278                         }
 279                         if (IPPROTO_IP == answer->protocol)
 280                                 break;
 281                 }
 282                 err = -EPROTONOSUPPORT;
 283         }
 284
 285         if (unlikely(err)) {
 286                 if (try_loading_module < 2) {
 287                         rcu_read_unlock();
 288                         /*
 289                          * Be more specific, e.g. net-pf-2-proto-132-type-1
 290                          * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
 291                          */
 292                         if (++try_loading_module == 1)
 293                                 request_module("net-pf-%d-proto-%d-type-%d",
 294                                                PF_INET, protocol, sock->type);
 295                         /*
 296                          * Fall back to generic, e.g. net-pf-2-proto-132
 297                          * (net-pf-PF_INET-proto-IPPROTO_SCTP)
 298                          */
 299                         else
 300                                 request_module("net-pf-%d-proto-%d",
 301                                                PF_INET, protocol);
 302                         goto lookup_protocol;
 303                 } else
 304                         goto out_rcu_unlock;
 305         }
 306
 307         err = -EPERM;
 308         if (sock->type == SOCK_RAW && !kern &&
 309             !ns_capable(net->user_ns, CAP_NET_RAW))
 310                 goto out_rcu_unlock;
 311
 312         sock->ops = answer->ops;
 313         answer_prot = answer->prot;
 314         answer_flags = answer->flags;
 315         rcu_read_unlock();
 316
 317         WARN_ON(answer_prot->slab == NULL);
 318
 319         err = -ENOBUFS;
 320         sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
 321         if (sk == NULL)
 322                 goto out;
 323
 324         err = 0;
 325         if (INET_PROTOSW_REUSE & answer_flags)
 326                 sk->sk_reuse = SK_CAN_REUSE;
 327
 328         inet = inet_sk(sk);
 329         inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
 330
 331         inet->nodefrag = 0;
 332
 333         if (SOCK_RAW == sock->type) {
 334                 inet->inet_num = protocol;
 335                 if (IPPROTO_RAW == protocol)
 336                         inet->hdrincl = 1;
 337         }
 338
 339         if (net->ipv4.sysctl_ip_no_pmtu_disc)
 340                 inet->pmtudisc = IP_PMTUDISC_DONT;
 341         else
 342                 inet->pmtudisc = IP_PMTUDISC_WANT;
 343
 344         inet->inet_id = 0;
 345
 346         sock_init_data(sock, sk);
 347
 348         sk->sk_destruct    = inet_sock_destruct;
 349         sk->sk_protocol    = protocol;
 350         sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
 351
 352         inet->uc_ttl    = -1;
 353         inet->mc_loop   = 1;
 354         inet->mc_ttl    = 1;
 355         inet->mc_all    = 1;
 356         inet->mc_index  = 0;
 357         inet->mc_list   = NULL;
 358         inet->rcv_tos   = 0;
 359
 360         sk_refcnt_debug_inc(sk);
 361
 362         if (inet->inet_num) {
 363                 /* It assumes that any protocol which allows
 364                  * the user to assign a number at socket
 365                  * creation time automatically
 366                  * shares.
 367                  */
 368                 inet->inet_sport = htons(inet->inet_num);
 369                 /* Add to protocol hash chains. */
 370                 sk->sk_prot->hash(sk);
 371         }
 372
 373         if (sk->sk_prot->init) {
 374                 err = sk->sk_prot->init(sk);
 375                 if (err)
 376                         sk_common_release(sk);
 377         }
 378 out:
 379         return err;
 380 out_rcu_unlock:
 381         rcu_read_unlock();
 382         goto out;
 383 }
 384
 385
 386 /*
 387  *      The peer socket should always be NULL (or else). When we call this
 388  *      function we are destroying the object and from then on nobody
 389  *      should refer to it.
 390  */
 391 int inet_release(struct socket *sock)
 392 {
 393         struct sock *sk = sock->sk;
 394
 395         if (sk) {
 396                 long timeout;
 397
 398                 sock_rps_reset_flow(sk);
 399
 400                 /* Applications forget to leave groups before exiting */
 401                 ip_mc_drop_socket(sk);
 402
 403                 /* If linger is set, we don't return until the close
 404                  * is complete.  Otherwise we return immediately. The
 405                  * actually closing is done the same either way.
 406                  *
 407                  * If the close is due to the process exiting, we never
 408                  * linger..
 409                  */
 410                 timeout = 0;
 411                 if (sock_flag(sk, SOCK_LINGER) &&
 412                     !(current->flags & PF_EXITING))
 413                         timeout = sk->sk_lingertime;
 414                 sock->sk = NULL;
 415                 sk->sk_prot->close(sk, timeout);
 416         }
 417         return 0;
 418 }
 419 EXPORT_SYMBOL(inet_release);
 420
 421 /* It is off by default, see below. */
 422 int sysctl_ip_nonlocal_bind __read_mostly;
 423 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
 424
 425 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 426 {
 427         struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
 428         struct sock *sk = sock->sk;
 429         struct inet_sock *inet = inet_sk(sk);
 430         struct net *net = sock_net(sk);
 431         unsigned short snum;
 432         int chk_addr_ret;
 433         int err;
 434
 435         /* If the socket has its own bind function then use it. (RAW) */
 436         if (sk->sk_prot->bind) {
 437                 err = sk->sk_prot->bind(sk, uaddr, addr_len);
 438                 goto out;
 439         }
 440         err = -EINVAL;
 441         if (addr_len < sizeof(struct sockaddr_in))
 442                 goto out;
 443
 444         if (addr->sin_family != AF_INET) {
 445                 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
 446                  * only if s_addr is INADDR_ANY.
 447                  */
 448                 err = -EAFNOSUPPORT;
 449                 if (addr->sin_family != AF_UNSPEC ||
 450                     addr->sin_addr.s_addr != htonl(INADDR_ANY))
 451                         goto out;
 452         }
 453
 454         chk_addr_ret = inet_addr_type(net, addr->sin_addr.s_addr);
 455
 456         /* Not specified by any standard per-se, however it breaks too
 457          * many applications when removed.  It is unfortunate since
 458          * allowing applications to make a non-local bind solves
 459          * several problems with systems using dynamic addressing.
 460          * (ie. your servers still start up even if your ISDN link
 461          *  is temporarily down)
 462          */
 463         err = -EADDRNOTAVAIL;
 464         if (!sysctl_ip_nonlocal_bind &&
 465             !(inet->freebind || inet->transparent) &&
 466             addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
 467             chk_addr_ret != RTN_LOCAL &&
 468             chk_addr_ret != RTN_MULTICAST &&
 469             chk_addr_ret != RTN_BROADCAST)
 470                 goto out;
 471
 472         snum = ntohs(addr->sin_port);
 473         err = -EACCES;
 474         if (snum && snum < PROT_SOCK &&
 475             !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
 476                 goto out;
 477
 478         /*      We keep a pair of addresses. rcv_saddr is the one
 479          *      used by hash lookups, and saddr is used for transmit.
 480          *
 481          *      In the BSD API these are the same except where it
 482          *      would be illegal to use them (multicast/broadcast) in
 483          *      which case the sending device address is used.
 484          */
 485         lock_sock(sk);
 486
 487         /* Check these errors (active socket, double bind). */
 488         err = -EINVAL;
 489         if (sk->sk_state != TCP_CLOSE || inet->inet_num)
 490                 goto out_release_sock;
 491
 492         inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
 493         if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
 494                 inet->inet_saddr = 0;  /* Use device */
 495
 496         /* Make sure we are allowed to bind here. */
 497         if (sk->sk_prot->get_port(sk, snum)) {
 498                 inet->inet_saddr = inet->inet_rcv_saddr = 0;
 499                 err = -EADDRINUSE;
 500                 goto out_release_sock;
 501         }
 502
 503         if (inet->inet_rcv_saddr)
 504                 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
 505         if (snum)
 506                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
 507         inet->inet_sport = htons(inet->inet_num);
 508         inet->inet_daddr = 0;
 509         inet->inet_dport = 0;
 510         sk_dst_reset(sk);
 511         err = 0;
 512 out_release_sock:
 513         release_sock(sk);
 514 out:
 515         return err;
 516 }
 517 EXPORT_SYMBOL(inet_bind);
 518
 519 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
 520                        int addr_len, int flags)
 521 {
 522         struct sock *sk = sock->sk;
 523
 524         if (addr_len < sizeof(uaddr->sa_family))
 525                 return -EINVAL;
 526         if (uaddr->sa_family == AF_UNSPEC)
 527                 return sk->sk_prot->disconnect(sk, flags);
 528
 529         if (!inet_sk(sk)->inet_num && inet_autobind(sk))
 530                 return -EAGAIN;
 531         return sk->sk_prot->connect(sk, uaddr, addr_len);
 532 }
 533 EXPORT_SYMBOL(inet_dgram_connect);
 534
 535 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
 536 {
 537         DEFINE_WAIT(wait);
 538
 539         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
 540         sk->sk_write_pending += writebias;
 541
 542         /* Basic assumption: if someone sets sk->sk_err, he _must_
 543          * change state of the socket from TCP_SYN_*.
 544          * Connect() does not allow to get error notifications
 545          * without closing the socket.
 546          */
 547         while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 548                 release_sock(sk);
 549                 timeo = schedule_timeout(timeo);
 550                 lock_sock(sk);
 551                 if (signal_pending(current) || !timeo)
 552                         break;
 553                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
 554         }
 555         finish_wait(sk_sleep(sk), &wait);
 556         sk->sk_write_pending -= writebias;
 557         return timeo;
 558 }
 559
 560 /*
 561  *      Connect to a remote host. There is regrettably still a little
 562  *      TCP 'magic' in here.
 563  */
 564 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 565                           int addr_len, int flags)
 566 {
 567         struct sock *sk = sock->sk;
 568         int err;
 569         long timeo;
 570
 571         if (addr_len < sizeof(uaddr->sa_family))
 572                 return -EINVAL;
 573
 574         if (uaddr->sa_family == AF_UNSPEC) {
 575                 err = sk->sk_prot->disconnect(sk, flags);
 576                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 577                 goto out;
 578         }
 579
 580         switch (sock->state) {
 581         default:
 582                 err = -EINVAL;
 583                 goto out;
 584         case SS_CONNECTED:
 585                 err = -EISCONN;
 586                 goto out;
 587         case SS_CONNECTING:
 588                 err = -EALREADY;
 589                 /* Fall out of switch with err, set for this state */
 590                 break;
 591         case SS_UNCONNECTED:
 592                 err = -EISCONN;
 593                 if (sk->sk_state != TCP_CLOSE)
 594                         goto out;
 595
 596                 err = sk->sk_prot->connect(sk, uaddr, addr_len);
 597                 if (err < 0)
 598                         goto out;
 599
 600                 sock->state = SS_CONNECTING;
 601
 602                 /* Just entered SS_CONNECTING state; the only
 603                  * difference is that return value in non-blocking
 604                  * case is EINPROGRESS, rather than EALREADY.
 605                  */
 606                 err = -EINPROGRESS;
 607                 break;
 608         }
 609
 610         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
 611
 612         if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 613                 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
 614                                 tcp_sk(sk)->fastopen_req &&
 615                                 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
 616
 617                 /* Error code is set above */
 618                 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
 619                         goto out;
 620
 621                 err = sock_intr_errno(timeo);
 622                 if (signal_pending(current))
 623                         goto out;
 624         }
 625
 626         /* Connection was closed by RST, timeout, ICMP error
 627          * or another process disconnected us.
 628          */
 629         if (sk->sk_state == TCP_CLOSE)
 630                 goto sock_error;
 631
 632         /* sk->sk_err may be not zero now, if RECVERR was ordered by user
 633          * and error was received after socket entered established state.
 634          * Hence, it is handled normally after connect() return successfully.
 635          */
 636
 637         sock->state = SS_CONNECTED;
 638         err = 0;
 639 out:
 640         return err;
 641
 642 sock_error:
 643         err = sock_error(sk) ? : -ECONNABORTED;
 644         sock->state = SS_UNCONNECTED;
 645         if (sk->sk_prot->disconnect(sk, flags))
 646                 sock->state = SS_DISCONNECTING;
 647         goto out;
 648 }
 649 EXPORT_SYMBOL(__inet_stream_connect);
 650
 651 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 652                         int addr_len, int flags)
 653 {
 654         int err;
 655
 656         lock_sock(sock->sk);
 657         err = __inet_stream_connect(sock, uaddr, addr_len, flags);
 658         release_sock(sock->sk);
 659         return err;
 660 }
 661 EXPORT_SYMBOL(inet_stream_connect);
 662
 663 /*
 664  *      Accept a pending connection. The TCP layer now gives BSD semantics.
 665  */
 666
 667 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
 668 {
 669         struct sock *sk1 = sock->sk;
 670         int err = -EINVAL;
 671         struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
 672
 673         if (!sk2)
 674                 goto do_err;
 675
 676         lock_sock(sk2);
 677
 678         sock_rps_record_flow(sk2);
 679         WARN_ON(!((1 << sk2->sk_state) &
 680                   (TCPF_ESTABLISHED | TCPF_SYN_RECV |
 681                   TCPF_CLOSE_WAIT | TCPF_CLOSE)));
 682
 683         sock_graft(sk2, newsock);
 684
 685         newsock->state = SS_CONNECTED;
 686         err = 0;
 687         release_sock(sk2);
 688 do_err:
 689         return err;
 690 }
 691 EXPORT_SYMBOL(inet_accept);
 692
 693
 694 /*
 695  *      This does both peername and sockname.
 696  */
 697 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
 698                         int *uaddr_len, int peer)
 699 {
 700         struct sock *sk         = sock->sk;
 701         struct inet_sock *inet  = inet_sk(sk);
 702         DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
 703
 704         sin->sin_family = AF_INET;
 705         if (peer) {
 706                 if (!inet->inet_dport ||
 707                     (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
 708                      peer == 1))
 709                         return -ENOTCONN;
 710                 sin->sin_port = inet->inet_dport;
 711                 sin->sin_addr.s_addr = inet->inet_daddr;
 712         } else {
 713                 __be32 addr = inet->inet_rcv_saddr;
 714                 if (!addr)
 715                         addr = inet->inet_saddr;
 716                 sin->sin_port = inet->inet_sport;
 717                 sin->sin_addr.s_addr = addr;
 718         }
 719         memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 720         *uaddr_len = sizeof(*sin);
 721         return 0;
 722 }
 723 EXPORT_SYMBOL(inet_getname);
 724
 725 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
 726                  size_t size)
 727 {
 728         struct sock *sk = sock->sk;
 729
 730         sock_rps_record_flow(sk);
 731
 732         /* We may need to bind the socket. */
 733         if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
 734             inet_autobind(sk))
 735                 return -EAGAIN;
 736
 737         return sk->sk_prot->sendmsg(iocb, sk, msg, size);
 738 }
 739 EXPORT_SYMBOL(inet_sendmsg);
 740
 741 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
 742                       size_t size, int flags)
 743 {
 744         struct sock *sk = sock->sk;
 745
 746         sock_rps_record_flow(sk);
 747
 748         /* We may need to bind the socket. */
 749         if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
 750             inet_autobind(sk))
 751                 return -EAGAIN;
 752
 753         if (sk->sk_prot->sendpage)
 754                 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
 755         return sock_no_sendpage(sock, page, offset, size, flags);
 756 }
 757 EXPORT_SYMBOL(inet_sendpage);
 758
 759 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
 760                  size_t size, int flags)
 761 {
 762         struct sock *sk = sock->sk;
 763         int addr_len = 0;
 764         int err;
 765
 766         sock_rps_record_flow(sk);
 767
 768         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
 769                                    flags & ~MSG_DONTWAIT, &addr_len);
 770         if (err >= 0)
 771                 msg->msg_namelen = addr_len;
 772         return err;
 773 }
 774 EXPORT_SYMBOL(inet_recvmsg);
 775
 776 int inet_shutdown(struct socket *sock, int how)
 777 {
 778         struct sock *sk = sock->sk;
 779         int err = 0;
 780
 781         /* This should really check to make sure
 782          * the socket is a TCP socket. (WHY AC...)
 783          */
 784         how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
 785                        1->2 bit 2 snds.
 786                        2->3 */
 787         if ((how & ~SHUTDOWN_MASK) || !how)     /* MAXINT->0 */
 788                 return -EINVAL;
 789
 790         lock_sock(sk);
 791         if (sock->state == SS_CONNECTING) {
 792                 if ((1 << sk->sk_state) &
 793                     (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
 794                         sock->state = SS_DISCONNECTING;
 795                 else
 796                         sock->state = SS_CONNECTED;
 797         }
 798
 799         switch (sk->sk_state) {
 800         case TCP_CLOSE:
 801                 err = -ENOTCONN;
 802                 /* Hack to wake up other listeners, who can poll for
 803                    POLLHUP, even on eg. unconnected UDP sockets -- RR */
 804         default:
 805                 sk->sk_shutdown |= how;
 806                 if (sk->sk_prot->shutdown)
 807                         sk->sk_prot->shutdown(sk, how);
 808                 break;
 809
 810         /* Remaining two branches are temporary solution for missing
 811          * close() in multithreaded environment. It is _not_ a good idea,
 812          * but we have no choice until close() is repaired at VFS level.
 813          */
 814         case TCP_LISTEN:
 815                 if (!(how & RCV_SHUTDOWN))
 816                         break;
 817                 /* Fall through */
 818         case TCP_SYN_SENT:
 819                 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
 820                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 821                 break;
 822         }
 823
 824         /* Wake up anyone sleeping in poll. */
 825         sk->sk_state_change(sk);
 826         release_sock(sk);
 827         return err;
 828 }
 829 EXPORT_SYMBOL(inet_shutdown);
 830
 831 /*
 832  *      ioctl() calls you can issue on an INET socket. Most of these are
 833  *      device configuration and stuff and very rarely used. Some ioctls
 834  *      pass on to the socket itself.
 835  *
 836  *      NOTE: I like the idea of a module for the config stuff. ie ifconfig
 837  *      loads the devconfigure module does its configuring and unloads it.
 838  *      There's a good 20K of config code hanging around the kernel.
 839  */
 840
 841 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 842 {
 843         struct sock *sk = sock->sk;
 844         int err = 0;
 845         struct net *net = sock_net(sk);
 846
 847         switch (cmd) {
 848         case SIOCGSTAMP:
 849                 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
 850                 break;
 851         case SIOCGSTAMPNS:
 852                 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
 853                 break;
 854         case SIOCADDRT:
 855         case SIOCDELRT:
 856         case SIOCRTMSG:
 857                 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
 858                 break;
 859         case SIOCDARP:
 860         case SIOCGARP:
 861         case SIOCSARP:
 862                 err = arp_ioctl(net, cmd, (void __user *)arg);
 863                 break;
 864         case SIOCGIFADDR:
 865         case SIOCSIFADDR:
 866         case SIOCGIFBRDADDR:
 867         case SIOCSIFBRDADDR:
 868         case SIOCGIFNETMASK:
 869         case SIOCSIFNETMASK:
 870         case SIOCGIFDSTADDR:
 871         case SIOCSIFDSTADDR:
 872         case SIOCSIFPFLAGS:
 873         case SIOCGIFPFLAGS:
 874         case SIOCSIFFLAGS:
 875                 err = devinet_ioctl(net, cmd, (void __user *)arg);
 876                 break;
 877         default:
 878                 if (sk->sk_prot->ioctl)
 879                         err = sk->sk_prot->ioctl(sk, cmd, arg);
 880                 else
 881                         err = -ENOIOCTLCMD;
 882                 break;
 883         }
 884         return err;
 885 }
 886 EXPORT_SYMBOL(inet_ioctl);
 887
 888 #ifdef CONFIG_COMPAT
 889 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 890 {
 891         struct sock *sk = sock->sk;
 892         int err = -ENOIOCTLCMD;
 893
 894         if (sk->sk_prot->compat_ioctl)
 895                 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
 896
 897         return err;
 898 }
 899 #endif
 900
 901 const struct proto_ops inet_stream_ops = {
 902         .family            = PF_INET,
 903         .owner             = THIS_MODULE,
 904         .release           = inet_release,
 905         .bind              = inet_bind,
 906         .connect           = inet_stream_connect,
 907         .socketpair        = sock_no_socketpair,
 908         .accept            = inet_accept,
 909         .getname           = inet_getname,
 910         .poll              = tcp_poll,
 911         .ioctl             = inet_ioctl,
 912         .listen            = inet_listen,
 913         .shutdown          = inet_shutdown,
 914         .setsockopt        = sock_common_setsockopt,
 915         .getsockopt        = sock_common_getsockopt,
 916         .sendmsg           = inet_sendmsg,
 917         .recvmsg           = inet_recvmsg,
 918         .mmap              = sock_no_mmap,
 919         .sendpage          = inet_sendpage,
 920         .splice_read       = tcp_splice_read,
 921 #ifdef CONFIG_COMPAT
 922         .compat_setsockopt = compat_sock_common_setsockopt,
 923         .compat_getsockopt = compat_sock_common_getsockopt,
 924         .compat_ioctl      = inet_compat_ioctl,
 925 #endif
 926 };
 927 EXPORT_SYMBOL(inet_stream_ops);
 928
 929 const struct proto_ops inet_dgram_ops = {
 930         .family            = PF_INET,
 931         .owner             = THIS_MODULE,
 932         .release           = inet_release,
 933         .bind              = inet_bind,
 934         .connect           = inet_dgram_connect,
 935         .socketpair        = sock_no_socketpair,
 936         .accept            = sock_no_accept,
 937         .getname           = inet_getname,
 938         .poll              = udp_poll,
 939         .ioctl             = inet_ioctl,
 940         .listen            = sock_no_listen,
 941         .shutdown          = inet_shutdown,
 942         .setsockopt        = sock_common_setsockopt,
 943         .getsockopt        = sock_common_getsockopt,
 944         .sendmsg           = inet_sendmsg,
 945         .recvmsg           = inet_recvmsg,
 946         .mmap              = sock_no_mmap,
 947         .sendpage          = inet_sendpage,
 948 #ifdef CONFIG_COMPAT
 949         .compat_setsockopt = compat_sock_common_setsockopt,
 950         .compat_getsockopt = compat_sock_common_getsockopt,
 951         .compat_ioctl      = inet_compat_ioctl,
 952 #endif
 953 };
 954 EXPORT_SYMBOL(inet_dgram_ops);
 955
 956 /*
 957  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
 958  * udp_poll
 959  */
 960 static const struct proto_ops inet_sockraw_ops = {
 961         .family            = PF_INET,
 962         .owner             = THIS_MODULE,
 963         .release           = inet_release,
 964         .bind              = inet_bind,
 965         .connect           = inet_dgram_connect,
 966         .socketpair        = sock_no_socketpair,
 967         .accept            = sock_no_accept,
 968         .getname           = inet_getname,
 969         .poll              = datagram_poll,
 970         .ioctl             = inet_ioctl,
 971         .listen            = sock_no_listen,
 972         .shutdown          = inet_shutdown,
 973         .setsockopt        = sock_common_setsockopt,
 974         .getsockopt        = sock_common_getsockopt,
 975         .sendmsg           = inet_sendmsg,
 976         .recvmsg           = inet_recvmsg,
 977         .mmap              = sock_no_mmap,
 978         .sendpage          = inet_sendpage,
 979 #ifdef CONFIG_COMPAT
 980         .compat_setsockopt = compat_sock_common_setsockopt,
 981         .compat_getsockopt = compat_sock_common_getsockopt,
 982         .compat_ioctl      = inet_compat_ioctl,
 983 #endif
 984 };
 985
 986 static const struct net_proto_family inet_family_ops = {
 987         .family = PF_INET,
 988         .create = inet_create,
 989         .owner  = THIS_MODULE,
 990 };
 991
 992 /* Upon startup we insert all the elements in inetsw_array[] into
 993  * the linked list inetsw.
 994  */
 995 static struct inet_protosw inetsw_array[] =
 996 {
 997         {
 998                 .type =       SOCK_STREAM,
 999                 .protocol =   IPPROTO_TCP,
1000                 .prot =       &tcp_prot,
1001                 .ops =        &inet_stream_ops,
1002                 .flags =      INET_PROTOSW_PERMANENT |
1003                               INET_PROTOSW_ICSK,
1004         },
1005
1006         {
1007                 .type =       SOCK_DGRAM,
1008                 .protocol =   IPPROTO_UDP,
1009                 .prot =       &udp_prot,
1010                 .ops =        &inet_dgram_ops,
1011                 .flags =      INET_PROTOSW_PERMANENT,
1012        },
1013
1014        {
1015                 .type =       SOCK_DGRAM,
1016                 .protocol =   IPPROTO_ICMP,
1017                 .prot =       &ping_prot,
1018                 .ops =        &inet_dgram_ops,
1019                 .flags =      INET_PROTOSW_REUSE,
1020        },
1021
1022        {
1023                .type =       SOCK_RAW,
1024                .protocol =   IPPROTO_IP,        /* wild card */
1025                .prot =       &raw_prot,
1026                .ops =        &inet_sockraw_ops,
1027                .flags =      INET_PROTOSW_REUSE,
1028        }
1029 };
1030
1031 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1032
1033 void inet_register_protosw(struct inet_protosw *p)
1034 {
1035         struct list_head *lh;
1036         struct inet_protosw *answer;
1037         int protocol = p->protocol;
1038         struct list_head *last_perm;
1039
1040         spin_lock_bh(&inetsw_lock);
1041
1042         if (p->type >= SOCK_MAX)
1043                 goto out_illegal;
1044
1045         /* If we are trying to override a permanent protocol, bail. */
1046         answer = NULL;
1047         last_perm = &inetsw[p->type];
1048         list_for_each(lh, &inetsw[p->type]) {
1049                 answer = list_entry(lh, struct inet_protosw, list);
1050
1051                 /* Check only the non-wild match. */
1052                 if (INET_PROTOSW_PERMANENT & answer->flags) {
1053                         if (protocol == answer->protocol)
1054                                 break;
1055                         last_perm = lh;
1056                 }
1057
1058                 answer = NULL;
1059         }
1060         if (answer)
1061                 goto out_permanent;
1062
1063         /* Add the new entry after the last permanent entry if any, so that
1064          * the new entry does not override a permanent entry when matched with
1065          * a wild-card protocol. But it is allowed to override any existing
1066          * non-permanent entry.  This means that when we remove this entry, the
1067          * system automatically returns to the old behavior.
1068          */
1069         list_add_rcu(&p->list, last_perm);
1070 out:
1071         spin_unlock_bh(&inetsw_lock);
1072
1073         return;
1074
1075 out_permanent:
1076         pr_err("Attempt to override permanent protocol %d\n", protocol);
1077         goto out;
1078
1079 out_illegal:
1080         pr_err("Ignoring attempt to register invalid socket type %d\n",
1081                p->type);
1082         goto out;
1083 }
1084 EXPORT_SYMBOL(inet_register_protosw);
1085
1086 void inet_unregister_protosw(struct inet_protosw *p)
1087 {
1088         if (INET_PROTOSW_PERMANENT & p->flags) {
1089                 pr_err("Attempt to unregister permanent protocol %d\n",
1090                        p->protocol);
1091         } else {
1092                 spin_lock_bh(&inetsw_lock);
1093                 list_del_rcu(&p->list);
1094                 spin_unlock_bh(&inetsw_lock);
1095
1096                 synchronize_net();
1097         }
1098 }
1099 EXPORT_SYMBOL(inet_unregister_protosw);
1100
1101 /*
1102  *      Shall we try to damage output packets if routing dev changes?
1103  */
1104
1105 int sysctl_ip_dynaddr __read_mostly;
1106
1107 static int inet_sk_reselect_saddr(struct sock *sk)
1108 {
1109         struct inet_sock *inet = inet_sk(sk);
1110         __be32 old_saddr = inet->inet_saddr;
1111         __be32 daddr = inet->inet_daddr;
1112         struct flowi4 *fl4;
1113         struct rtable *rt;
1114         __be32 new_saddr;
1115         struct ip_options_rcu *inet_opt;
1116
1117         inet_opt = rcu_dereference_protected(inet->inet_opt,
1118                                              sock_owned_by_user(sk));
1119         if (inet_opt && inet_opt->opt.srr)
1120                 daddr = inet_opt->opt.faddr;
1121
1122         /* Query new route. */
1123         fl4 = &inet->cork.fl.u.ip4;
1124         rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1125                               sk->sk_bound_dev_if, sk->sk_protocol,
1126                               inet->inet_sport, inet->inet_dport, sk);
1127         if (IS_ERR(rt))
1128                 return PTR_ERR(rt);
1129
1130         sk_setup_caps(sk, &rt->dst);
1131
1132         new_saddr = fl4->saddr;
1133
1134         if (new_saddr == old_saddr)
1135                 return 0;
1136
1137         if (sysctl_ip_dynaddr > 1) {
1138                 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1139                         __func__, &old_saddr, &new_saddr);
1140         }
1141
1142         inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1143
1144         /*
1145          * XXX The only one ugly spot where we need to
1146          * XXX really change the sockets identity after
1147          * XXX it has entered the hashes. -DaveM
1148          *
1149          * Besides that, it does not check for connection
1150          * uniqueness. Wait for troubles.
1151          */
1152         __sk_prot_rehash(sk);
1153         return 0;
1154 }
1155
1156 int inet_sk_rebuild_header(struct sock *sk)
1157 {
1158         struct inet_sock *inet = inet_sk(sk);
1159         struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1160         __be32 daddr;
1161         struct ip_options_rcu *inet_opt;
1162         struct flowi4 *fl4;
1163         int err;
1164
1165         /* Route is OK, nothing to do. */
1166         if (rt)
1167                 return 0;
1168
1169         /* Reroute. */
1170         rcu_read_lock();
1171         inet_opt = rcu_dereference(inet->inet_opt);
1172         daddr = inet->inet_daddr;
1173         if (inet_opt && inet_opt->opt.srr)
1174                 daddr = inet_opt->opt.faddr;
1175         rcu_read_unlock();
1176         fl4 = &inet->cork.fl.u.ip4;
1177         rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1178                                    inet->inet_dport, inet->inet_sport,
1179                                    sk->sk_protocol, RT_CONN_FLAGS(sk),
1180                                    sk->sk_bound_dev_if);
1181         if (!IS_ERR(rt)) {
1182                 err = 0;
1183                 sk_setup_caps(sk, &rt->dst);
1184         } else {
1185                 err = PTR_ERR(rt);
1186
1187                 /* Routing failed... */
1188                 sk->sk_route_caps = 0;
1189                 /*
1190                  * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1191                  * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1192                  */
1193                 if (!sysctl_ip_dynaddr ||
1194                     sk->sk_state != TCP_SYN_SENT ||
1195                     (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1196                     (err = inet_sk_reselect_saddr(sk)) != 0)
1197                         sk->sk_err_soft = -err;
1198         }
1199
1200         return err;
1201 }
1202 EXPORT_SYMBOL(inet_sk_rebuild_header);
1203
1204 static int inet_gso_send_check(struct sk_buff *skb)
1205 {
1206         const struct net_offload *ops;
1207         const struct iphdr *iph;
1208         int proto;
1209         int ihl;
1210         int err = -EINVAL;
1211
1212         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1213                 goto out;
1214
1215         iph = ip_hdr(skb);
1216         ihl = iph->ihl * 4;
1217         if (ihl < sizeof(*iph))
1218                 goto out;
1219
1220         proto = iph->protocol;
1221
1222         /* Warning: after this point, iph might be no longer valid */
1223         if (unlikely(!pskb_may_pull(skb, ihl)))
1224                 goto out;
1225         __skb_pull(skb, ihl);
1226
1227         skb_reset_transport_header(skb);
1228         err = -EPROTONOSUPPORT;
1229
1230         ops = rcu_dereference(inet_offloads[proto]);
1231         if (likely(ops && ops->callbacks.gso_send_check))
1232                 err = ops->callbacks.gso_send_check(skb);
1233
1234 out:
1235         return err;
1236 }
1237
1238 static struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1239                                         netdev_features_t features)
1240 {
1241         struct sk_buff *segs = ERR_PTR(-EINVAL);
1242         const struct net_offload *ops;
1243         unsigned int offset = 0;
1244         bool udpfrag, encap;
1245         struct iphdr *iph;
1246         int proto;
1247         int nhoff;
1248         int ihl;
1249         int id;
1250
1251         if (unlikely(skb_shinfo(skb)->gso_type &
1252                      ~(SKB_GSO_TCPV4 |
1253                        SKB_GSO_UDP |
1254                        SKB_GSO_DODGY |
1255                        SKB_GSO_TCP_ECN |
1256                        SKB_GSO_GRE |
1257                        SKB_GSO_GRE_CSUM |
1258                        SKB_GSO_IPIP |
1259                        SKB_GSO_SIT |
1260                        SKB_GSO_TCPV6 |
1261                        SKB_GSO_UDP_TUNNEL |
1262                        SKB_GSO_UDP_TUNNEL_CSUM |
1263                        SKB_GSO_MPLS |
1264                        0)))
1265                 goto out;
1266
1267         skb_reset_network_header(skb);
1268         nhoff = skb_network_header(skb) - skb_mac_header(skb);
1269         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1270                 goto out;
1271
1272         iph = ip_hdr(skb);
1273         ihl = iph->ihl * 4;
1274         if (ihl < sizeof(*iph))
1275                 goto out;
1276
1277         id = ntohs(iph->id);
1278         proto = iph->protocol;
1279
1280         /* Warning: after this point, iph might be no longer valid */
1281         if (unlikely(!pskb_may_pull(skb, ihl)))
1282                 goto out;
1283         __skb_pull(skb, ihl);
1284
1285         encap = SKB_GSO_CB(skb)->encap_level > 0;
1286         if (encap)
1287                 features = skb->dev->hw_enc_features & netif_skb_features(skb);
1288         SKB_GSO_CB(skb)->encap_level += ihl;
1289
1290         skb_reset_transport_header(skb);
1291
1292         segs = ERR_PTR(-EPROTONOSUPPORT);
1293
1294         if (skb->encapsulation &&
1295             skb_shinfo(skb)->gso_type & (SKB_GSO_SIT|SKB_GSO_IPIP))
1296                 udpfrag = proto == IPPROTO_UDP && encap;
1297         else
1298                 udpfrag = proto == IPPROTO_UDP && !skb->encapsulation;
1299
1300         ops = rcu_dereference(inet_offloads[proto]);
1301         if (likely(ops && ops->callbacks.gso_segment))
1302                 segs = ops->callbacks.gso_segment(skb, features);
1303
1304         if (IS_ERR_OR_NULL(segs))
1305                 goto out;
1306
1307         skb = segs;
1308         do {
1309                 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1310                 if (udpfrag) {
1311                         iph->id = htons(id);
1312                         iph->frag_off = htons(offset >> 3);
1313                         if (skb->next != NULL)
1314                                 iph->frag_off |= htons(IP_MF);
1315                         offset += skb->len - nhoff - ihl;
1316                 } else {
1317                         iph->id = htons(id++);
1318                 }
1319                 iph->tot_len = htons(skb->len - nhoff);
1320                 ip_send_check(iph);
1321                 if (encap)
1322                         skb_reset_inner_headers(skb);
1323                 skb->network_header = (u8 *)iph - skb->head;
1324         } while ((skb = skb->next));
1325
1326 out:
1327         return segs;
1328 }
1329
1330 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1331                                          struct sk_buff *skb)
1332 {
1333         const struct net_offload *ops;
1334         struct sk_buff **pp = NULL;
1335         struct sk_buff *p;
1336         const struct iphdr *iph;
1337         unsigned int hlen;
1338         unsigned int off;
1339         unsigned int id;
1340         int flush = 1;
1341         int proto;
1342
1343         off = skb_gro_offset(skb);
1344         hlen = off + sizeof(*iph);
1345         iph = skb_gro_header_fast(skb, off);
1346         if (skb_gro_header_hard(skb, hlen)) {
1347                 iph = skb_gro_header_slow(skb, hlen, off);
1348                 if (unlikely(!iph))
1349                         goto out;
1350         }
1351
1352         proto = iph->protocol;
1353
1354         rcu_read_lock();
1355         ops = rcu_dereference(inet_offloads[proto]);
1356         if (!ops || !ops->callbacks.gro_receive)
1357                 goto out_unlock;
1358
1359         if (*(u8 *)iph != 0x45)
1360                 goto out_unlock;
1361
1362         if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1363                 goto out_unlock;
1364
1365         id = ntohl(*(__be32 *)&iph->id);
1366         flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1367         id >>= 16;
1368
1369         for (p = *head; p; p = p->next) {
1370                 struct iphdr *iph2;
1371
1372                 if (!NAPI_GRO_CB(p)->same_flow)
1373                         continue;
1374
1375                 iph2 = (struct iphdr *)(p->data + off);
1376                 /* The above works because, with the exception of the top
1377                  * (inner most) layer, we only aggregate pkts with the same
1378                  * hdr length so all the hdrs we'll need to verify will start
1379                  * at the same offset.
1380                  */
1381                 if ((iph->protocol ^ iph2->protocol) |
1382                     ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1383                     ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1384                         NAPI_GRO_CB(p)->same_flow = 0;
1385                         continue;
1386                 }
1387
1388                 /* All fields must match except length and checksum. */
1389                 NAPI_GRO_CB(p)->flush |=
1390                         (iph->ttl ^ iph2->ttl) |
1391                         (iph->tos ^ iph2->tos) |
1392                         ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1393
1394                 /* Save the IP ID check to be included later when we get to
1395                  * the transport layer so only the inner most IP ID is checked.
1396                  * This is because some GSO/TSO implementations do not
1397                  * correctly increment the IP ID for the outer hdrs.
1398                  */
1399                 NAPI_GRO_CB(p)->flush_id =
1400                             ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1401                 NAPI_GRO_CB(p)->flush |= flush;
1402         }
1403
1404         NAPI_GRO_CB(skb)->flush |= flush;
1405         skb_set_network_header(skb, off);
1406         /* The above will be needed by the transport layer if there is one
1407          * immediately following this IP hdr.
1408          */
1409
1410         skb_gro_pull(skb, sizeof(*iph));
1411         skb_set_transport_header(skb, skb_gro_offset(skb));
1412
1413         pp = ops->callbacks.gro_receive(head, skb);
1414
1415 out_unlock:
1416         rcu_read_unlock();
1417
1418 out:
1419         NAPI_GRO_CB(skb)->flush |= flush;
1420
1421         return pp;
1422 }
1423
1424 static int inet_gro_complete(struct sk_buff *skb, int nhoff)
1425 {
1426         __be16 newlen = htons(skb->len - nhoff);
1427         struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1428         const struct net_offload *ops;
1429         int proto = iph->protocol;
1430         int err = -ENOSYS;
1431
1432         if (skb->encapsulation)
1433                 skb_set_inner_network_header(skb, nhoff);
1434
1435         csum_replace2(&iph->check, iph->tot_len, newlen);
1436         iph->tot_len = newlen;
1437
1438         rcu_read_lock();
1439         ops = rcu_dereference(inet_offloads[proto]);
1440         if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1441                 goto out_unlock;
1442
1443         /* Only need to add sizeof(*iph) to get to the next hdr below
1444          * because any hdr with option will have been flushed in
1445          * inet_gro_receive().
1446          */
1447         err = ops->callbacks.gro_complete(skb, nhoff + sizeof(*iph));
1448
1449 out_unlock:
1450         rcu_read_unlock();
1451
1452         return err;
1453 }
1454
1455 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1456                          unsigned short type, unsigned char protocol,
1457                          struct net *net)
1458 {
1459         struct socket *sock;
1460         int rc = sock_create_kern(family, type, protocol, &sock);
1461
1462         if (rc == 0) {
1463                 *sk = sock->sk;
1464                 (*sk)->sk_allocation = GFP_ATOMIC;
1465                 /*
1466                  * Unhash it so that IP input processing does not even see it,
1467                  * we do not wish this socket to see incoming packets.
1468                  */
1469                 (*sk)->sk_prot->unhash(*sk);
1470
1471                 sk_change_net(*sk, net);
1472         }
1473         return rc;
1474 }
1475 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1476
1477 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1478 {
1479         unsigned long res = 0;
1480         int i;
1481
1482         for_each_possible_cpu(i)
1483                 res += *(((unsigned long *) per_cpu_ptr(mib, i)) + offt);
1484         return res;
1485 }
1486 EXPORT_SYMBOL_GPL(snmp_fold_field);
1487
1488 #if BITS_PER_LONG==32
1489
1490 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1491 {
1492         u64 res = 0;
1493         int cpu;
1494
1495         for_each_possible_cpu(cpu) {
1496                 void *bhptr;
1497                 struct u64_stats_sync *syncp;
1498                 u64 v;
1499                 unsigned int start;
1500
1501                 bhptr = per_cpu_ptr(mib, cpu);
1502                 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1503                 do {
1504                         start = u64_stats_fetch_begin_irq(syncp);
1505                         v = *(((u64 *) bhptr) + offt);
1506                 } while (u64_stats_fetch_retry_irq(syncp, start));
1507
1508                 res += v;
1509         }
1510         return res;
1511 }
1512 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1513 #endif
1514
1515 #ifdef CONFIG_IP_MULTICAST
1516 static const struct net_protocol igmp_protocol = {
1517         .handler =      igmp_rcv,
1518         .netns_ok =     1,
1519 };
1520 #endif
1521
1522 static const struct net_protocol tcp_protocol = {
1523         .early_demux    =       tcp_v4_early_demux,
1524         .handler        =       tcp_v4_rcv,
1525         .err_handler    =       tcp_v4_err,
1526         .no_policy      =       1,
1527         .netns_ok       =       1,
1528         .icmp_strict_tag_validation = 1,
1529 };
1530
1531 static const struct net_protocol udp_protocol = {
1532         .early_demux =  udp_v4_early_demux,
1533         .handler =      udp_rcv,
1534         .err_handler =  udp_err,
1535         .no_policy =    1,
1536         .netns_ok =     1,
1537 };
1538
1539 static const struct net_protocol icmp_protocol = {
1540         .handler =      icmp_rcv,
1541         .err_handler =  icmp_err,
1542         .no_policy =    1,
1543         .netns_ok =     1,
1544 };
1545
1546 static __net_init int ipv4_mib_init_net(struct net *net)
1547 {
1548         int i;
1549
1550         net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1551         if (!net->mib.tcp_statistics)
1552                 goto err_tcp_mib;
1553         net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1554         if (!net->mib.ip_statistics)
1555                 goto err_ip_mib;
1556
1557         for_each_possible_cpu(i) {
1558                 struct ipstats_mib *af_inet_stats;
1559                 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1560                 u64_stats_init(&af_inet_stats->syncp);
1561         }
1562
1563         net->mib.net_statistics = alloc_percpu(struct linux_mib);
1564         if (!net->mib.net_statistics)
1565                 goto err_net_mib;
1566         net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1567         if (!net->mib.udp_statistics)
1568                 goto err_udp_mib;
1569         net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1570         if (!net->mib.udplite_statistics)
1571                 goto err_udplite_mib;
1572         net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1573         if (!net->mib.icmp_statistics)
1574                 goto err_icmp_mib;
1575         net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1576                                               GFP_KERNEL);
1577         if (!net->mib.icmpmsg_statistics)
1578                 goto err_icmpmsg_mib;
1579
1580         tcp_mib_init(net);
1581         return 0;
1582
1583 err_icmpmsg_mib:
1584         free_percpu(net->mib.icmp_statistics);
1585 err_icmp_mib:
1586         free_percpu(net->mib.udplite_statistics);
1587 err_udplite_mib:
1588         free_percpu(net->mib.udp_statistics);
1589 err_udp_mib:
1590         free_percpu(net->mib.net_statistics);
1591 err_net_mib:
1592         free_percpu(net->mib.ip_statistics);
1593 err_ip_mib:
1594         free_percpu(net->mib.tcp_statistics);
1595 err_tcp_mib:
1596         return -ENOMEM;
1597 }
1598
1599 static __net_exit void ipv4_mib_exit_net(struct net *net)
1600 {
1601         kfree(net->mib.icmpmsg_statistics);
1602         free_percpu(net->mib.icmp_statistics);
1603         free_percpu(net->mib.udplite_statistics);
1604         free_percpu(net->mib.udp_statistics);
1605         free_percpu(net->mib.net_statistics);
1606         free_percpu(net->mib.ip_statistics);
1607         free_percpu(net->mib.tcp_statistics);
1608 }
1609
1610 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1611         .init = ipv4_mib_init_net,
1612         .exit = ipv4_mib_exit_net,
1613 };
1614
1615 static int __init init_ipv4_mibs(void)
1616 {
1617         return register_pernet_subsys(&ipv4_mib_ops);
1618 }
1619
1620 static __net_init int inet_init_net(struct net *net)
1621 {
1622         /*
1623          * Set defaults for local port range
1624          */
1625         seqlock_init(&net->ipv4.ip_local_ports.lock);
1626         net->ipv4.ip_local_ports.range[0] =  32768;
1627         net->ipv4.ip_local_ports.range[1] =  61000;
1628
1629         seqlock_init(&net->ipv4.ping_group_range.lock);
1630         /*
1631          * Sane defaults - nobody may create ping sockets.
1632          * Boot scripts should set this to distro-specific group.
1633          */
1634         net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1635         net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1636         return 0;
1637 }
1638
1639 static __net_exit void inet_exit_net(struct net *net)
1640 {
1641 }
1642
1643 static __net_initdata struct pernet_operations af_inet_ops = {
1644         .init = inet_init_net,
1645         .exit = inet_exit_net,
1646 };
1647
1648 static int __init init_inet_pernet_ops(void)
1649 {
1650         return register_pernet_subsys(&af_inet_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         rc = proto_register(&tcp_prot, 1);
1707         if (rc)
1708                 goto out;
1709
1710         rc = proto_register(&udp_prot, 1);
1711         if (rc)
1712                 goto out_unregister_tcp_proto;
1713
1714         rc = proto_register(&raw_prot, 1);
1715         if (rc)
1716                 goto out_unregister_udp_proto;
1717
1718         rc = proto_register(&ping_prot, 1);
1719         if (rc)
1720                 goto out_unregister_raw_proto;
1721
1722         /*
1723          *      Tell SOCKET that we are alive...
1724          */
1725
1726         (void)sock_register(&inet_family_ops);
1727
1728 #ifdef CONFIG_SYSCTL
1729         ip_static_sysctl_init();
1730 #endif
1731
1732         /*
1733          *      Add all the base protocols.
1734          */
1735
1736         if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1737                 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1738         if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1739                 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1740         if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1741                 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1742 #ifdef CONFIG_IP_MULTICAST
1743         if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1744                 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1745 #endif
1746
1747         /* Register the socket-side information for inet_create. */
1748         for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1749                 INIT_LIST_HEAD(r);
1750
1751         for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1752                 inet_register_protosw(q);
1753
1754         /*
1755          *      Set the ARP module up
1756          */
1757
1758         arp_init();
1759
1760         /*
1761          *      Set the IP module up
1762          */
1763
1764         ip_init();
1765
1766         tcp_v4_init();
1767
1768         /* Setup TCP slab cache for open requests. */
1769         tcp_init();
1770
1771         /* Setup UDP memory threshold */
1772         udp_init();
1773
1774         /* Add UDP-Lite (RFC 3828) */
1775         udplite4_register();
1776
1777         ping_init();
1778
1779         /*
1780          *      Set the ICMP layer up
1781          */
1782
1783         if (icmp_init() < 0)
1784                 panic("Failed to create the ICMP control socket.\n");
1785
1786         /*
1787          *      Initialise the multicast router
1788          */
1789 #if defined(CONFIG_IP_MROUTE)
1790         if (ip_mr_init())
1791                 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1792 #endif
1793
1794         if (init_inet_pernet_ops())
1795                 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
1796         /*
1797          *      Initialise per-cpu ipv4 mibs
1798          */
1799
1800         if (init_ipv4_mibs())
1801                 pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1802
1803         ipv4_proc_init();
1804
1805         ipfrag_init();
1806
1807         dev_add_pack(&ip_packet_type);
1808
1809         rc = 0;
1810 out:
1811         return rc;
1812 out_unregister_raw_proto:
1813         proto_unregister(&raw_prot);
1814 out_unregister_udp_proto:
1815         proto_unregister(&udp_prot);
1816 out_unregister_tcp_proto:
1817         proto_unregister(&tcp_prot);
1818         goto out;
1819 }
1820
1821 fs_initcall(inet_init);
1822
1823 /* ------------------------------------------------------------------------ */
1824
1825 #ifdef CONFIG_PROC_FS
1826 static int __init ipv4_proc_init(void)
1827 {
1828         int rc = 0;
1829
1830         if (raw_proc_init())
1831                 goto out_raw;
1832         if (tcp4_proc_init())
1833                 goto out_tcp;
1834         if (udp4_proc_init())
1835                 goto out_udp;
1836         if (ping_proc_init())
1837                 goto out_ping;
1838         if (ip_misc_proc_init())
1839                 goto out_misc;
1840 out:
1841         return rc;
1842 out_misc:
1843         ping_proc_exit();
1844 out_ping:
1845         udp4_proc_exit();
1846 out_udp:
1847         tcp4_proc_exit();
1848 out_tcp:
1849         raw_proc_exit();
1850 out_raw:
1851         rc = -ENOMEM;
1852         goto out;
1853 }
1854
1855 #else /* CONFIG_PROC_FS */
1856 static int __init ipv4_proc_init(void)
1857 {
1858         return 0;
1859 }
1860 #endif /* CONFIG_PROC_FS */
1861
1862 MODULE_ALIAS_NETPROTO(PF_INET);
1863