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/ipip.h>
 115 #include <net/inet_common.h>
 116 #include <net/xfrm.h>
 117 #include <net/net_namespace.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 u32 inet_ehash_secret __read_mostly;
 249 EXPORT_SYMBOL(inet_ehash_secret);
 250
 251 /*
 252  * inet_ehash_secret must be set exactly once
 253  */
 254 void build_ehash_secret(void)
 255 {
 256         u32 rnd;
 257
 258         do {
 259                 get_random_bytes(&rnd, sizeof(rnd));
 260         } while (rnd == 0);
 261
 262         cmpxchg(&inet_ehash_secret, 0, rnd);
 263 }
 264 EXPORT_SYMBOL(build_ehash_secret);
 265
 266 static inline int inet_netns_ok(struct net *net, __u8 protocol)
 267 {
 268         const struct net_protocol *ipprot;
 269
 270         if (net_eq(net, &init_net))
 271                 return 1;
 272
 273         ipprot = rcu_dereference(inet_protos[protocol]);
 274         if (ipprot == NULL) {
 275                 /* raw IP is OK */
 276                 return 1;
 277         }
 278         return ipprot->netns_ok;
 279 }
 280
 281 /*
 282  *      Create an inet socket.
 283  */
 284
 285 static int inet_create(struct net *net, struct socket *sock, int protocol,
 286                        int kern)
 287 {
 288         struct sock *sk;
 289         struct inet_protosw *answer;
 290         struct inet_sock *inet;
 291         struct proto *answer_prot;
 292         unsigned char answer_flags;
 293         char answer_no_check;
 294         int try_loading_module = 0;
 295         int err;
 296
 297         if (unlikely(!inet_ehash_secret))
 298                 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
 299                         build_ehash_secret();
 300
 301         sock->state = SS_UNCONNECTED;
 302
 303         /* Look for the requested type/protocol pair. */
 304 lookup_protocol:
 305         err = -ESOCKTNOSUPPORT;
 306         rcu_read_lock();
 307         list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
 308
 309                 err = 0;
 310                 /* Check the non-wild match. */
 311                 if (protocol == answer->protocol) {
 312                         if (protocol != IPPROTO_IP)
 313                                 break;
 314                 } else {
 315                         /* Check for the two wild cases. */
 316                         if (IPPROTO_IP == protocol) {
 317                                 protocol = answer->protocol;
 318                                 break;
 319                         }
 320                         if (IPPROTO_IP == answer->protocol)
 321                                 break;
 322                 }
 323                 err = -EPROTONOSUPPORT;
 324         }
 325
 326         if (unlikely(err)) {
 327                 if (try_loading_module < 2) {
 328                         rcu_read_unlock();
 329                         /*
 330                          * Be more specific, e.g. net-pf-2-proto-132-type-1
 331                          * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
 332                          */
 333                         if (++try_loading_module == 1)
 334                                 request_module("net-pf-%d-proto-%d-type-%d",
 335                                                PF_INET, protocol, sock->type);
 336                         /*
 337                          * Fall back to generic, e.g. net-pf-2-proto-132
 338                          * (net-pf-PF_INET-proto-IPPROTO_SCTP)
 339                          */
 340                         else
 341                                 request_module("net-pf-%d-proto-%d",
 342                                                PF_INET, protocol);
 343                         goto lookup_protocol;
 344                 } else
 345                         goto out_rcu_unlock;
 346         }
 347
 348         err = -EPERM;
 349         if (sock->type == SOCK_RAW && !kern &&
 350             !ns_capable(net->user_ns, CAP_NET_RAW))
 351                 goto out_rcu_unlock;
 352
 353         err = -EAFNOSUPPORT;
 354         if (!inet_netns_ok(net, protocol))
 355                 goto out_rcu_unlock;
 356
 357         sock->ops = answer->ops;
 358         answer_prot = answer->prot;
 359         answer_no_check = answer->no_check;
 360         answer_flags = answer->flags;
 361         rcu_read_unlock();
 362
 363         WARN_ON(answer_prot->slab == NULL);
 364
 365         err = -ENOBUFS;
 366         sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
 367         if (sk == NULL)
 368                 goto out;
 369
 370         err = 0;
 371         sk->sk_no_check = answer_no_check;
 372         if (INET_PROTOSW_REUSE & answer_flags)
 373                 sk->sk_reuse = SK_CAN_REUSE;
 374
 375         inet = inet_sk(sk);
 376         inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
 377
 378         inet->nodefrag = 0;
 379
 380         if (SOCK_RAW == sock->type) {
 381                 inet->inet_num = protocol;
 382                 if (IPPROTO_RAW == protocol)
 383                         inet->hdrincl = 1;
 384         }
 385
 386         if (ipv4_config.no_pmtu_disc)
 387                 inet->pmtudisc = IP_PMTUDISC_DONT;
 388         else
 389                 inet->pmtudisc = IP_PMTUDISC_WANT;
 390
 391         inet->inet_id = 0;
 392
 393         sock_init_data(sock, sk);
 394
 395         sk->sk_destruct    = inet_sock_destruct;
 396         sk->sk_protocol    = protocol;
 397         sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
 398
 399         inet->uc_ttl    = -1;
 400         inet->mc_loop   = 1;
 401         inet->mc_ttl    = 1;
 402         inet->mc_all    = 1;
 403         inet->mc_index  = 0;
 404         inet->mc_list   = NULL;
 405         inet->rcv_tos   = 0;
 406
 407         sk_refcnt_debug_inc(sk);
 408
 409         if (inet->inet_num) {
 410                 /* It assumes that any protocol which allows
 411                  * the user to assign a number at socket
 412                  * creation time automatically
 413                  * shares.
 414                  */
 415                 inet->inet_sport = htons(inet->inet_num);
 416                 /* Add to protocol hash chains. */
 417                 sk->sk_prot->hash(sk);
 418         }
 419
 420         if (sk->sk_prot->init) {
 421                 err = sk->sk_prot->init(sk);
 422                 if (err)
 423                         sk_common_release(sk);
 424         }
 425 out:
 426         return err;
 427 out_rcu_unlock:
 428         rcu_read_unlock();
 429         goto out;
 430 }
 431
 432
 433 /*
 434  *      The peer socket should always be NULL (or else). When we call this
 435  *      function we are destroying the object and from then on nobody
 436  *      should refer to it.
 437  */
 438 int inet_release(struct socket *sock)
 439 {
 440         struct sock *sk = sock->sk;
 441
 442         if (sk) {
 443                 long timeout;
 444
 445                 sock_rps_reset_flow(sk);
 446
 447                 /* Applications forget to leave groups before exiting */
 448                 ip_mc_drop_socket(sk);
 449
 450                 /* If linger is set, we don't return until the close
 451                  * is complete.  Otherwise we return immediately. The
 452                  * actually closing is done the same either way.
 453                  *
 454                  * If the close is due to the process exiting, we never
 455                  * linger..
 456                  */
 457                 timeout = 0;
 458                 if (sock_flag(sk, SOCK_LINGER) &&
 459                     !(current->flags & PF_EXITING))
 460                         timeout = sk->sk_lingertime;
 461                 sock->sk = NULL;
 462                 sk->sk_prot->close(sk, timeout);
 463         }
 464         return 0;
 465 }
 466 EXPORT_SYMBOL(inet_release);
 467
 468 /* It is off by default, see below. */
 469 int sysctl_ip_nonlocal_bind __read_mostly;
 470 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
 471
 472 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 473 {
 474         struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
 475         struct sock *sk = sock->sk;
 476         struct inet_sock *inet = inet_sk(sk);
 477         struct net *net = sock_net(sk);
 478         unsigned short snum;
 479         int chk_addr_ret;
 480         int err;
 481
 482         /* If the socket has its own bind function then use it. (RAW) */
 483         if (sk->sk_prot->bind) {
 484                 err = sk->sk_prot->bind(sk, uaddr, addr_len);
 485                 goto out;
 486         }
 487         err = -EINVAL;
 488         if (addr_len < sizeof(struct sockaddr_in))
 489                 goto out;
 490
 491         if (addr->sin_family != AF_INET) {
 492                 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
 493                  * only if s_addr is INADDR_ANY.
 494                  */
 495                 err = -EAFNOSUPPORT;
 496                 if (addr->sin_family != AF_UNSPEC ||
 497                     addr->sin_addr.s_addr != htonl(INADDR_ANY))
 498                         goto out;
 499         }
 500
 501         chk_addr_ret = inet_addr_type(net, addr->sin_addr.s_addr);
 502
 503         /* Not specified by any standard per-se, however it breaks too
 504          * many applications when removed.  It is unfortunate since
 505          * allowing applications to make a non-local bind solves
 506          * several problems with systems using dynamic addressing.
 507          * (ie. your servers still start up even if your ISDN link
 508          *  is temporarily down)
 509          */
 510         err = -EADDRNOTAVAIL;
 511         if (!sysctl_ip_nonlocal_bind &&
 512             !(inet->freebind || inet->transparent) &&
 513             addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
 514             chk_addr_ret != RTN_LOCAL &&
 515             chk_addr_ret != RTN_MULTICAST &&
 516             chk_addr_ret != RTN_BROADCAST)
 517                 goto out;
 518
 519         snum = ntohs(addr->sin_port);
 520         err = -EACCES;
 521         if (snum && snum < PROT_SOCK &&
 522             !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
 523                 goto out;
 524
 525         /*      We keep a pair of addresses. rcv_saddr is the one
 526          *      used by hash lookups, and saddr is used for transmit.
 527          *
 528          *      In the BSD API these are the same except where it
 529          *      would be illegal to use them (multicast/broadcast) in
 530          *      which case the sending device address is used.
 531          */
 532         lock_sock(sk);
 533
 534         /* Check these errors (active socket, double bind). */
 535         err = -EINVAL;
 536         if (sk->sk_state != TCP_CLOSE || inet->inet_num)
 537                 goto out_release_sock;
 538
 539         inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
 540         if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
 541                 inet->inet_saddr = 0;  /* Use device */
 542
 543         /* Make sure we are allowed to bind here. */
 544         if (sk->sk_prot->get_port(sk, snum)) {
 545                 inet->inet_saddr = inet->inet_rcv_saddr = 0;
 546                 err = -EADDRINUSE;
 547                 goto out_release_sock;
 548         }
 549
 550         if (inet->inet_rcv_saddr)
 551                 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
 552         if (snum)
 553                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
 554         inet->inet_sport = htons(inet->inet_num);
 555         inet->inet_daddr = 0;
 556         inet->inet_dport = 0;
 557         sk_dst_reset(sk);
 558         err = 0;
 559 out_release_sock:
 560         release_sock(sk);
 561 out:
 562         return err;
 563 }
 564 EXPORT_SYMBOL(inet_bind);
 565
 566 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
 567                        int addr_len, int flags)
 568 {
 569         struct sock *sk = sock->sk;
 570
 571         if (addr_len < sizeof(uaddr->sa_family))
 572                 return -EINVAL;
 573         if (uaddr->sa_family == AF_UNSPEC)
 574                 return sk->sk_prot->disconnect(sk, flags);
 575
 576         if (!inet_sk(sk)->inet_num && inet_autobind(sk))
 577                 return -EAGAIN;
 578         return sk->sk_prot->connect(sk, uaddr, addr_len);
 579 }
 580 EXPORT_SYMBOL(inet_dgram_connect);
 581
 582 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
 583 {
 584         DEFINE_WAIT(wait);
 585
 586         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
 587         sk->sk_write_pending += writebias;
 588
 589         /* Basic assumption: if someone sets sk->sk_err, he _must_
 590          * change state of the socket from TCP_SYN_*.
 591          * Connect() does not allow to get error notifications
 592          * without closing the socket.
 593          */
 594         while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 595                 release_sock(sk);
 596                 timeo = schedule_timeout(timeo);
 597                 lock_sock(sk);
 598                 if (signal_pending(current) || !timeo)
 599                         break;
 600                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
 601         }
 602         finish_wait(sk_sleep(sk), &wait);
 603         sk->sk_write_pending -= writebias;
 604         return timeo;
 605 }
 606
 607 /*
 608  *      Connect to a remote host. There is regrettably still a little
 609  *      TCP 'magic' in here.
 610  */
 611 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 612                           int addr_len, int flags)
 613 {
 614         struct sock *sk = sock->sk;
 615         int err;
 616         long timeo;
 617
 618         if (addr_len < sizeof(uaddr->sa_family))
 619                 return -EINVAL;
 620
 621         if (uaddr->sa_family == AF_UNSPEC) {
 622                 err = sk->sk_prot->disconnect(sk, flags);
 623                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 624                 goto out;
 625         }
 626
 627         switch (sock->state) {
 628         default:
 629                 err = -EINVAL;
 630                 goto out;
 631         case SS_CONNECTED:
 632                 err = -EISCONN;
 633                 goto out;
 634         case SS_CONNECTING:
 635                 err = -EALREADY;
 636                 /* Fall out of switch with err, set for this state */
 637                 break;
 638         case SS_UNCONNECTED:
 639                 err = -EISCONN;
 640                 if (sk->sk_state != TCP_CLOSE)
 641                         goto out;
 642
 643                 err = sk->sk_prot->connect(sk, uaddr, addr_len);
 644                 if (err < 0)
 645                         goto out;
 646
 647                 sock->state = SS_CONNECTING;
 648
 649                 /* Just entered SS_CONNECTING state; the only
 650                  * difference is that return value in non-blocking
 651                  * case is EINPROGRESS, rather than EALREADY.
 652                  */
 653                 err = -EINPROGRESS;
 654                 break;
 655         }
 656
 657         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
 658
 659         if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 660                 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
 661                                 tcp_sk(sk)->fastopen_req &&
 662                                 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
 663
 664                 /* Error code is set above */
 665                 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
 666                         goto out;
 667
 668                 err = sock_intr_errno(timeo);
 669                 if (signal_pending(current))
 670                         goto out;
 671         }
 672
 673         /* Connection was closed by RST, timeout, ICMP error
 674          * or another process disconnected us.
 675          */
 676         if (sk->sk_state == TCP_CLOSE)
 677                 goto sock_error;
 678
 679         /* sk->sk_err may be not zero now, if RECVERR was ordered by user
 680          * and error was received after socket entered established state.
 681          * Hence, it is handled normally after connect() return successfully.
 682          */
 683
 684         sock->state = SS_CONNECTED;
 685         err = 0;
 686 out:
 687         return err;
 688
 689 sock_error:
 690         err = sock_error(sk) ? : -ECONNABORTED;
 691         sock->state = SS_UNCONNECTED;
 692         if (sk->sk_prot->disconnect(sk, flags))
 693                 sock->state = SS_DISCONNECTING;
 694         goto out;
 695 }
 696 EXPORT_SYMBOL(__inet_stream_connect);
 697
 698 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 699                         int addr_len, int flags)
 700 {
 701         int err;
 702
 703         lock_sock(sock->sk);
 704         err = __inet_stream_connect(sock, uaddr, addr_len, flags);
 705         release_sock(sock->sk);
 706         return err;
 707 }
 708 EXPORT_SYMBOL(inet_stream_connect);
 709
 710 /*
 711  *      Accept a pending connection. The TCP layer now gives BSD semantics.
 712  */
 713
 714 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
 715 {
 716         struct sock *sk1 = sock->sk;
 717         int err = -EINVAL;
 718         struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
 719
 720         if (!sk2)
 721                 goto do_err;
 722
 723         lock_sock(sk2);
 724
 725         sock_rps_record_flow(sk2);
 726         WARN_ON(!((1 << sk2->sk_state) &
 727                   (TCPF_ESTABLISHED | TCPF_SYN_RECV |
 728                   TCPF_CLOSE_WAIT | TCPF_CLOSE)));
 729
 730         sock_graft(sk2, newsock);
 731
 732         newsock->state = SS_CONNECTED;
 733         err = 0;
 734         release_sock(sk2);
 735 do_err:
 736         return err;
 737 }
 738 EXPORT_SYMBOL(inet_accept);
 739
 740
 741 /*
 742  *      This does both peername and sockname.
 743  */
 744 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
 745                         int *uaddr_len, int peer)
 746 {
 747         struct sock *sk         = sock->sk;
 748         struct inet_sock *inet  = inet_sk(sk);
 749         DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
 750
 751         sin->sin_family = AF_INET;
 752         if (peer) {
 753                 if (!inet->inet_dport ||
 754                     (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
 755                      peer == 1))
 756                         return -ENOTCONN;
 757                 sin->sin_port = inet->inet_dport;
 758                 sin->sin_addr.s_addr = inet->inet_daddr;
 759         } else {
 760                 __be32 addr = inet->inet_rcv_saddr;
 761                 if (!addr)
 762                         addr = inet->inet_saddr;
 763                 sin->sin_port = inet->inet_sport;
 764                 sin->sin_addr.s_addr = addr;
 765         }
 766         memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 767         *uaddr_len = sizeof(*sin);
 768         return 0;
 769 }
 770 EXPORT_SYMBOL(inet_getname);
 771
 772 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
 773                  size_t size)
 774 {
 775         struct sock *sk = sock->sk;
 776
 777         sock_rps_record_flow(sk);
 778
 779         /* We may need to bind the socket. */
 780         if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
 781             inet_autobind(sk))
 782                 return -EAGAIN;
 783
 784         return sk->sk_prot->sendmsg(iocb, sk, msg, size);
 785 }
 786 EXPORT_SYMBOL(inet_sendmsg);
 787
 788 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
 789                       size_t size, int flags)
 790 {
 791         struct sock *sk = sock->sk;
 792
 793         sock_rps_record_flow(sk);
 794
 795         /* We may need to bind the socket. */
 796         if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
 797             inet_autobind(sk))
 798                 return -EAGAIN;
 799
 800         if (sk->sk_prot->sendpage)
 801                 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
 802         return sock_no_sendpage(sock, page, offset, size, flags);
 803 }
 804 EXPORT_SYMBOL(inet_sendpage);
 805
 806 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
 807                  size_t size, int flags)
 808 {
 809         struct sock *sk = sock->sk;
 810         int addr_len = 0;
 811         int err;
 812
 813         sock_rps_record_flow(sk);
 814
 815         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
 816                                    flags & ~MSG_DONTWAIT, &addr_len);
 817         if (err >= 0)
 818                 msg->msg_namelen = addr_len;
 819         return err;
 820 }
 821 EXPORT_SYMBOL(inet_recvmsg);
 822
 823 int inet_shutdown(struct socket *sock, int how)
 824 {
 825         struct sock *sk = sock->sk;
 826         int err = 0;
 827
 828         /* This should really check to make sure
 829          * the socket is a TCP socket. (WHY AC...)
 830          */
 831         how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
 832                        1->2 bit 2 snds.
 833                        2->3 */
 834         if ((how & ~SHUTDOWN_MASK) || !how)     /* MAXINT->0 */
 835                 return -EINVAL;
 836
 837         lock_sock(sk);
 838         if (sock->state == SS_CONNECTING) {
 839                 if ((1 << sk->sk_state) &
 840                     (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
 841                         sock->state = SS_DISCONNECTING;
 842                 else
 843                         sock->state = SS_CONNECTED;
 844         }
 845
 846         switch (sk->sk_state) {
 847         case TCP_CLOSE:
 848                 err = -ENOTCONN;
 849                 /* Hack to wake up other listeners, who can poll for
 850                    POLLHUP, even on eg. unconnected UDP sockets -- RR */
 851         default:
 852                 sk->sk_shutdown |= how;
 853                 if (sk->sk_prot->shutdown)
 854                         sk->sk_prot->shutdown(sk, how);
 855                 break;
 856
 857         /* Remaining two branches are temporary solution for missing
 858          * close() in multithreaded environment. It is _not_ a good idea,
 859          * but we have no choice until close() is repaired at VFS level.
 860          */
 861         case TCP_LISTEN:
 862                 if (!(how & RCV_SHUTDOWN))
 863                         break;
 864                 /* Fall through */
 865         case TCP_SYN_SENT:
 866                 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
 867                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 868                 break;
 869         }
 870
 871         /* Wake up anyone sleeping in poll. */
 872         sk->sk_state_change(sk);
 873         release_sock(sk);
 874         return err;
 875 }
 876 EXPORT_SYMBOL(inet_shutdown);
 877
 878 /*
 879  *      ioctl() calls you can issue on an INET socket. Most of these are
 880  *      device configuration and stuff and very rarely used. Some ioctls
 881  *      pass on to the socket itself.
 882  *
 883  *      NOTE: I like the idea of a module for the config stuff. ie ifconfig
 884  *      loads the devconfigure module does its configuring and unloads it.
 885  *      There's a good 20K of config code hanging around the kernel.
 886  */
 887
 888 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 889 {
 890         struct sock *sk = sock->sk;
 891         int err = 0;
 892         struct net *net = sock_net(sk);
 893
 894         switch (cmd) {
 895         case SIOCGSTAMP:
 896                 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
 897                 break;
 898         case SIOCGSTAMPNS:
 899                 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
 900                 break;
 901         case SIOCADDRT:
 902         case SIOCDELRT:
 903         case SIOCRTMSG:
 904                 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
 905                 break;
 906         case SIOCDARP:
 907         case SIOCGARP:
 908         case SIOCSARP:
 909                 err = arp_ioctl(net, cmd, (void __user *)arg);
 910                 break;
 911         case SIOCGIFADDR:
 912         case SIOCSIFADDR:
 913         case SIOCGIFBRDADDR:
 914         case SIOCSIFBRDADDR:
 915         case SIOCGIFNETMASK:
 916         case SIOCSIFNETMASK:
 917         case SIOCGIFDSTADDR:
 918         case SIOCSIFDSTADDR:
 919         case SIOCSIFPFLAGS:
 920         case SIOCGIFPFLAGS:
 921         case SIOCSIFFLAGS:
 922                 err = devinet_ioctl(net, cmd, (void __user *)arg);
 923                 break;
 924         default:
 925                 if (sk->sk_prot->ioctl)
 926                         err = sk->sk_prot->ioctl(sk, cmd, arg);
 927                 else
 928                         err = -ENOIOCTLCMD;
 929                 break;
 930         }
 931         return err;
 932 }
 933 EXPORT_SYMBOL(inet_ioctl);
 934
 935 #ifdef CONFIG_COMPAT
 936 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 937 {
 938         struct sock *sk = sock->sk;
 939         int err = -ENOIOCTLCMD;
 940
 941         if (sk->sk_prot->compat_ioctl)
 942                 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
 943
 944         return err;
 945 }
 946 #endif
 947
 948 const struct proto_ops inet_stream_ops = {
 949         .family            = PF_INET,
 950         .owner             = THIS_MODULE,
 951         .release           = inet_release,
 952         .bind              = inet_bind,
 953         .connect           = inet_stream_connect,
 954         .socketpair        = sock_no_socketpair,
 955         .accept            = inet_accept,
 956         .getname           = inet_getname,
 957         .poll              = tcp_poll,
 958         .ioctl             = inet_ioctl,
 959         .listen            = inet_listen,
 960         .shutdown          = inet_shutdown,
 961         .setsockopt        = sock_common_setsockopt,
 962         .getsockopt        = sock_common_getsockopt,
 963         .sendmsg           = inet_sendmsg,
 964         .recvmsg           = inet_recvmsg,
 965         .mmap              = sock_no_mmap,
 966         .sendpage          = inet_sendpage,
 967         .splice_read       = tcp_splice_read,
 968 #ifdef CONFIG_COMPAT
 969         .compat_setsockopt = compat_sock_common_setsockopt,
 970         .compat_getsockopt = compat_sock_common_getsockopt,
 971         .compat_ioctl      = inet_compat_ioctl,
 972 #endif
 973 };
 974 EXPORT_SYMBOL(inet_stream_ops);
 975
 976 const struct proto_ops inet_dgram_ops = {
 977         .family            = PF_INET,
 978         .owner             = THIS_MODULE,
 979         .release           = inet_release,
 980         .bind              = inet_bind,
 981         .connect           = inet_dgram_connect,
 982         .socketpair        = sock_no_socketpair,
 983         .accept            = sock_no_accept,
 984         .getname           = inet_getname,
 985         .poll              = udp_poll,
 986         .ioctl             = inet_ioctl,
 987         .listen            = sock_no_listen,
 988         .shutdown          = inet_shutdown,
 989         .setsockopt        = sock_common_setsockopt,
 990         .getsockopt        = sock_common_getsockopt,
 991         .sendmsg           = inet_sendmsg,
 992         .recvmsg           = inet_recvmsg,
 993         .mmap              = sock_no_mmap,
 994         .sendpage          = inet_sendpage,
 995 #ifdef CONFIG_COMPAT
 996         .compat_setsockopt = compat_sock_common_setsockopt,
 997         .compat_getsockopt = compat_sock_common_getsockopt,
 998         .compat_ioctl      = inet_compat_ioctl,
 999 #endif
1000 };
1001 EXPORT_SYMBOL(inet_dgram_ops);
1002
1003 /*
1004  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1005  * udp_poll
1006  */
1007 static const struct proto_ops inet_sockraw_ops = {
1008         .family            = PF_INET,
1009         .owner             = THIS_MODULE,
1010         .release           = inet_release,
1011         .bind              = inet_bind,
1012         .connect           = inet_dgram_connect,
1013         .socketpair        = sock_no_socketpair,
1014         .accept            = sock_no_accept,
1015         .getname           = inet_getname,
1016         .poll              = datagram_poll,
1017         .ioctl             = inet_ioctl,
1018         .listen            = sock_no_listen,
1019         .shutdown          = inet_shutdown,
1020         .setsockopt        = sock_common_setsockopt,
1021         .getsockopt        = sock_common_getsockopt,
1022         .sendmsg           = inet_sendmsg,
1023         .recvmsg           = inet_recvmsg,
1024         .mmap              = sock_no_mmap,
1025         .sendpage          = inet_sendpage,
1026 #ifdef CONFIG_COMPAT
1027         .compat_setsockopt = compat_sock_common_setsockopt,
1028         .compat_getsockopt = compat_sock_common_getsockopt,
1029         .compat_ioctl      = inet_compat_ioctl,
1030 #endif
1031 };
1032
1033 static const struct net_proto_family inet_family_ops = {
1034         .family = PF_INET,
1035         .create = inet_create,
1036         .owner  = THIS_MODULE,
1037 };
1038
1039 /* Upon startup we insert all the elements in inetsw_array[] into
1040  * the linked list inetsw.
1041  */
1042 static struct inet_protosw inetsw_array[] =
1043 {
1044         {
1045                 .type =       SOCK_STREAM,
1046                 .protocol =   IPPROTO_TCP,
1047                 .prot =       &tcp_prot,
1048                 .ops =        &inet_stream_ops,
1049                 .no_check =   0,
1050                 .flags =      INET_PROTOSW_PERMANENT |
1051                               INET_PROTOSW_ICSK,
1052         },
1053
1054         {
1055                 .type =       SOCK_DGRAM,
1056                 .protocol =   IPPROTO_UDP,
1057                 .prot =       &udp_prot,
1058                 .ops =        &inet_dgram_ops,
1059                 .no_check =   UDP_CSUM_DEFAULT,
1060                 .flags =      INET_PROTOSW_PERMANENT,
1061        },
1062
1063        {
1064                 .type =       SOCK_DGRAM,
1065                 .protocol =   IPPROTO_ICMP,
1066                 .prot =       &ping_prot,
1067                 .ops =        &inet_dgram_ops,
1068                 .no_check =   UDP_CSUM_DEFAULT,
1069                 .flags =      INET_PROTOSW_REUSE,
1070        },
1071
1072        {
1073                .type =       SOCK_RAW,
1074                .protocol =   IPPROTO_IP,        /* wild card */
1075                .prot =       &raw_prot,
1076                .ops =        &inet_sockraw_ops,
1077                .no_check =   UDP_CSUM_DEFAULT,
1078                .flags =      INET_PROTOSW_REUSE,
1079        }
1080 };
1081
1082 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1083
1084 void inet_register_protosw(struct inet_protosw *p)
1085 {
1086         struct list_head *lh;
1087         struct inet_protosw *answer;
1088         int protocol = p->protocol;
1089         struct list_head *last_perm;
1090
1091         spin_lock_bh(&inetsw_lock);
1092
1093         if (p->type >= SOCK_MAX)
1094                 goto out_illegal;
1095
1096         /* If we are trying to override a permanent protocol, bail. */
1097         answer = NULL;
1098         last_perm = &inetsw[p->type];
1099         list_for_each(lh, &inetsw[p->type]) {
1100                 answer = list_entry(lh, struct inet_protosw, list);
1101
1102                 /* Check only the non-wild match. */
1103                 if (INET_PROTOSW_PERMANENT & answer->flags) {
1104                         if (protocol == answer->protocol)
1105                                 break;
1106                         last_perm = lh;
1107                 }
1108
1109                 answer = NULL;
1110         }
1111         if (answer)
1112                 goto out_permanent;
1113
1114         /* Add the new entry after the last permanent entry if any, so that
1115          * the new entry does not override a permanent entry when matched with
1116          * a wild-card protocol. But it is allowed to override any existing
1117          * non-permanent entry.  This means that when we remove this entry, the
1118          * system automatically returns to the old behavior.
1119          */
1120         list_add_rcu(&p->list, last_perm);
1121 out:
1122         spin_unlock_bh(&inetsw_lock);
1123
1124         return;
1125
1126 out_permanent:
1127         pr_err("Attempt to override permanent protocol %d\n", protocol);
1128         goto out;
1129
1130 out_illegal:
1131         pr_err("Ignoring attempt to register invalid socket type %d\n",
1132                p->type);
1133         goto out;
1134 }
1135 EXPORT_SYMBOL(inet_register_protosw);
1136
1137 void inet_unregister_protosw(struct inet_protosw *p)
1138 {
1139         if (INET_PROTOSW_PERMANENT & p->flags) {
1140                 pr_err("Attempt to unregister permanent protocol %d\n",
1141                        p->protocol);
1142         } else {
1143                 spin_lock_bh(&inetsw_lock);
1144                 list_del_rcu(&p->list);
1145                 spin_unlock_bh(&inetsw_lock);
1146
1147                 synchronize_net();
1148         }
1149 }
1150 EXPORT_SYMBOL(inet_unregister_protosw);
1151
1152 /*
1153  *      Shall we try to damage output packets if routing dev changes?
1154  */
1155
1156 int sysctl_ip_dynaddr __read_mostly;
1157
1158 static int inet_sk_reselect_saddr(struct sock *sk)
1159 {
1160         struct inet_sock *inet = inet_sk(sk);
1161         __be32 old_saddr = inet->inet_saddr;
1162         __be32 daddr = inet->inet_daddr;
1163         struct flowi4 *fl4;
1164         struct rtable *rt;
1165         __be32 new_saddr;
1166         struct ip_options_rcu *inet_opt;
1167
1168         inet_opt = rcu_dereference_protected(inet->inet_opt,
1169                                              sock_owned_by_user(sk));
1170         if (inet_opt && inet_opt->opt.srr)
1171                 daddr = inet_opt->opt.faddr;
1172
1173         /* Query new route. */
1174         fl4 = &inet->cork.fl.u.ip4;
1175         rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1176                               sk->sk_bound_dev_if, sk->sk_protocol,
1177                               inet->inet_sport, inet->inet_dport, sk, false);
1178         if (IS_ERR(rt))
1179                 return PTR_ERR(rt);
1180
1181         sk_setup_caps(sk, &rt->dst);
1182
1183         new_saddr = fl4->saddr;
1184
1185         if (new_saddr == old_saddr)
1186                 return 0;
1187
1188         if (sysctl_ip_dynaddr > 1) {
1189                 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1190                         __func__, &old_saddr, &new_saddr);
1191         }
1192
1193         inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1194
1195         /*
1196          * XXX The only one ugly spot where we need to
1197          * XXX really change the sockets identity after
1198          * XXX it has entered the hashes. -DaveM
1199          *
1200          * Besides that, it does not check for connection
1201          * uniqueness. Wait for troubles.
1202          */
1203         __sk_prot_rehash(sk);
1204         return 0;
1205 }
1206
1207 int inet_sk_rebuild_header(struct sock *sk)
1208 {
1209         struct inet_sock *inet = inet_sk(sk);
1210         struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1211         __be32 daddr;
1212         struct ip_options_rcu *inet_opt;
1213         struct flowi4 *fl4;
1214         int err;
1215
1216         /* Route is OK, nothing to do. */
1217         if (rt)
1218                 return 0;
1219
1220         /* Reroute. */
1221         rcu_read_lock();
1222         inet_opt = rcu_dereference(inet->inet_opt);
1223         daddr = inet->inet_daddr;
1224         if (inet_opt && inet_opt->opt.srr)
1225                 daddr = inet_opt->opt.faddr;
1226         rcu_read_unlock();
1227         fl4 = &inet->cork.fl.u.ip4;
1228         rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1229                                    inet->inet_dport, inet->inet_sport,
1230                                    sk->sk_protocol, RT_CONN_FLAGS(sk),
1231                                    sk->sk_bound_dev_if);
1232         if (!IS_ERR(rt)) {
1233                 err = 0;
1234                 sk_setup_caps(sk, &rt->dst);
1235         } else {
1236                 err = PTR_ERR(rt);
1237
1238                 /* Routing failed... */
1239                 sk->sk_route_caps = 0;
1240                 /*
1241                  * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1242                  * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1243                  */
1244                 if (!sysctl_ip_dynaddr ||
1245                     sk->sk_state != TCP_SYN_SENT ||
1246                     (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1247                     (err = inet_sk_reselect_saddr(sk)) != 0)
1248                         sk->sk_err_soft = -err;
1249         }
1250
1251         return err;
1252 }
1253 EXPORT_SYMBOL(inet_sk_rebuild_header);
1254
1255 static int inet_gso_send_check(struct sk_buff *skb)
1256 {
1257         const struct net_offload *ops;
1258         const struct iphdr *iph;
1259         int proto;
1260         int ihl;
1261         int err = -EINVAL;
1262
1263         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1264                 goto out;
1265
1266         iph = ip_hdr(skb);
1267         ihl = iph->ihl * 4;
1268         if (ihl < sizeof(*iph))
1269                 goto out;
1270
1271         if (unlikely(!pskb_may_pull(skb, ihl)))
1272                 goto out;
1273
1274         __skb_pull(skb, ihl);
1275         skb_reset_transport_header(skb);
1276         iph = ip_hdr(skb);
1277         proto = iph->protocol;
1278         err = -EPROTONOSUPPORT;
1279
1280         rcu_read_lock();
1281         ops = rcu_dereference(inet_offloads[proto]);
1282         if (likely(ops && ops->callbacks.gso_send_check))
1283                 err = ops->callbacks.gso_send_check(skb);
1284         rcu_read_unlock();
1285
1286 out:
1287         return err;
1288 }
1289
1290 static struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1291         netdev_features_t features)
1292 {
1293         struct sk_buff *segs = ERR_PTR(-EINVAL);
1294         const struct net_offload *ops;
1295         struct iphdr *iph;
1296         int proto;
1297         int ihl;
1298         int id;
1299         unsigned int offset = 0;
1300
1301         if (!(features & NETIF_F_V4_CSUM))
1302                 features &= ~NETIF_F_SG;
1303
1304         if (unlikely(skb_shinfo(skb)->gso_type &
1305                      ~(SKB_GSO_TCPV4 |
1306                        SKB_GSO_UDP |
1307                        SKB_GSO_DODGY |
1308                        SKB_GSO_TCP_ECN |
1309                        0)))
1310                 goto out;
1311
1312         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1313                 goto out;
1314
1315         iph = ip_hdr(skb);
1316         ihl = iph->ihl * 4;
1317         if (ihl < sizeof(*iph))
1318                 goto out;
1319
1320         if (unlikely(!pskb_may_pull(skb, ihl)))
1321                 goto out;
1322
1323         __skb_pull(skb, ihl);
1324         skb_reset_transport_header(skb);
1325         iph = ip_hdr(skb);
1326         id = ntohs(iph->id);
1327         proto = iph->protocol;
1328         segs = ERR_PTR(-EPROTONOSUPPORT);
1329
1330         rcu_read_lock();
1331         ops = rcu_dereference(inet_offloads[proto]);
1332         if (likely(ops && ops->callbacks.gso_segment))
1333                 segs = ops->callbacks.gso_segment(skb, features);
1334         rcu_read_unlock();
1335
1336         if (!segs || IS_ERR(segs))
1337                 goto out;
1338
1339         skb = segs;
1340         do {
1341                 iph = ip_hdr(skb);
1342                 if (proto == IPPROTO_UDP) {
1343                         iph->id = htons(id);
1344                         iph->frag_off = htons(offset >> 3);
1345                         if (skb->next != NULL)
1346                                 iph->frag_off |= htons(IP_MF);
1347                         offset += (skb->len - skb->mac_len - iph->ihl * 4);
1348                 } else
1349                         iph->id = htons(id++);
1350                 iph->tot_len = htons(skb->len - skb->mac_len);
1351                 iph->check = 0;
1352                 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1353         } while ((skb = skb->next));
1354
1355 out:
1356         return segs;
1357 }
1358
1359 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1360                                          struct sk_buff *skb)
1361 {
1362         const struct net_offload *ops;
1363         struct sk_buff **pp = NULL;
1364         struct sk_buff *p;
1365         const struct iphdr *iph;
1366         unsigned int hlen;
1367         unsigned int off;
1368         unsigned int id;
1369         int flush = 1;
1370         int proto;
1371
1372         off = skb_gro_offset(skb);
1373         hlen = off + sizeof(*iph);
1374         iph = skb_gro_header_fast(skb, off);
1375         if (skb_gro_header_hard(skb, hlen)) {
1376                 iph = skb_gro_header_slow(skb, hlen, off);
1377                 if (unlikely(!iph))
1378                         goto out;
1379         }
1380
1381         proto = iph->protocol;
1382
1383         rcu_read_lock();
1384         ops = rcu_dereference(inet_offloads[proto]);
1385         if (!ops || !ops->callbacks.gro_receive)
1386                 goto out_unlock;
1387
1388         if (*(u8 *)iph != 0x45)
1389                 goto out_unlock;
1390
1391         if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1392                 goto out_unlock;
1393
1394         id = ntohl(*(__be32 *)&iph->id);
1395         flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF));
1396         id >>= 16;
1397
1398         for (p = *head; p; p = p->next) {
1399                 struct iphdr *iph2;
1400
1401                 if (!NAPI_GRO_CB(p)->same_flow)
1402                         continue;
1403
1404                 iph2 = ip_hdr(p);
1405
1406                 if ((iph->protocol ^ iph2->protocol) |
1407                     ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1408                     ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1409                         NAPI_GRO_CB(p)->same_flow = 0;
1410                         continue;
1411                 }
1412
1413                 /* All fields must match except length and checksum. */
1414                 NAPI_GRO_CB(p)->flush |=
1415                         (iph->ttl ^ iph2->ttl) |
1416                         (iph->tos ^ iph2->tos) |
1417                         ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1418
1419                 NAPI_GRO_CB(p)->flush |= flush;
1420         }
1421
1422         NAPI_GRO_CB(skb)->flush |= flush;
1423         skb_gro_pull(skb, sizeof(*iph));
1424         skb_set_transport_header(skb, skb_gro_offset(skb));
1425
1426         pp = ops->callbacks.gro_receive(head, skb);
1427
1428 out_unlock:
1429         rcu_read_unlock();
1430
1431 out:
1432         NAPI_GRO_CB(skb)->flush |= flush;
1433
1434         return pp;
1435 }
1436
1437 static int inet_gro_complete(struct sk_buff *skb)
1438 {
1439         __be16 newlen = htons(skb->len - skb_network_offset(skb));
1440         struct iphdr *iph = ip_hdr(skb);
1441         const struct net_offload *ops;
1442         int proto = iph->protocol;
1443         int err = -ENOSYS;
1444
1445         csum_replace2(&iph->check, iph->tot_len, newlen);
1446         iph->tot_len = newlen;
1447
1448         rcu_read_lock();
1449         ops = rcu_dereference(inet_offloads[proto]);
1450         if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1451                 goto out_unlock;
1452
1453         err = ops->callbacks.gro_complete(skb);
1454
1455 out_unlock:
1456         rcu_read_unlock();
1457
1458         return err;
1459 }
1460
1461 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1462                          unsigned short type, unsigned char protocol,
1463                          struct net *net)
1464 {
1465         struct socket *sock;
1466         int rc = sock_create_kern(family, type, protocol, &sock);
1467
1468         if (rc == 0) {
1469                 *sk = sock->sk;
1470                 (*sk)->sk_allocation = GFP_ATOMIC;
1471                 /*
1472                  * Unhash it so that IP input processing does not even see it,
1473                  * we do not wish this socket to see incoming packets.
1474                  */
1475                 (*sk)->sk_prot->unhash(*sk);
1476
1477                 sk_change_net(*sk, net);
1478         }
1479         return rc;
1480 }
1481 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1482
1483 unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1484 {
1485         unsigned long res = 0;
1486         int i, j;
1487
1488         for_each_possible_cpu(i) {
1489                 for (j = 0; j < SNMP_ARRAY_SZ; j++)
1490                         res += *(((unsigned long *) per_cpu_ptr(mib[j], i)) + offt);
1491         }
1492         return res;
1493 }
1494 EXPORT_SYMBOL_GPL(snmp_fold_field);
1495
1496 #if BITS_PER_LONG==32
1497
1498 u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset)
1499 {
1500         u64 res = 0;
1501         int cpu;
1502
1503         for_each_possible_cpu(cpu) {
1504                 void *bhptr;
1505                 struct u64_stats_sync *syncp;
1506                 u64 v;
1507                 unsigned int start;
1508
1509                 bhptr = per_cpu_ptr(mib[0], cpu);
1510                 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1511                 do {
1512                         start = u64_stats_fetch_begin_bh(syncp);
1513                         v = *(((u64 *) bhptr) + offt);
1514                 } while (u64_stats_fetch_retry_bh(syncp, start));
1515
1516                 res += v;
1517         }
1518         return res;
1519 }
1520 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1521 #endif
1522
1523 int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align)
1524 {
1525         BUG_ON(ptr == NULL);
1526         ptr[0] = __alloc_percpu(mibsize, align);
1527         if (!ptr[0])
1528                 return -ENOMEM;
1529 #if SNMP_ARRAY_SZ == 2
1530         ptr[1] = __alloc_percpu(mibsize, align);
1531         if (!ptr[1]) {
1532                 free_percpu(ptr[0]);
1533                 ptr[0] = NULL;
1534                 return -ENOMEM;
1535         }
1536 #endif
1537         return 0;
1538 }
1539 EXPORT_SYMBOL_GPL(snmp_mib_init);
1540
1541 void snmp_mib_free(void __percpu *ptr[SNMP_ARRAY_SZ])
1542 {
1543         int i;
1544
1545         BUG_ON(ptr == NULL);
1546         for (i = 0; i < SNMP_ARRAY_SZ; i++) {
1547                 free_percpu(ptr[i]);
1548                 ptr[i] = NULL;
1549         }
1550 }
1551 EXPORT_SYMBOL_GPL(snmp_mib_free);
1552
1553 #ifdef CONFIG_IP_MULTICAST
1554 static const struct net_protocol igmp_protocol = {
1555         .handler =      igmp_rcv,
1556         .netns_ok =     1,
1557 };
1558 #endif
1559
1560 static const struct net_protocol tcp_protocol = {
1561         .early_demux    =       tcp_v4_early_demux,
1562         .handler        =       tcp_v4_rcv,
1563         .err_handler    =       tcp_v4_err,
1564         .no_policy      =       1,
1565         .netns_ok       =       1,
1566 };
1567
1568 static const struct net_offload tcp_offload = {
1569         .callbacks = {
1570                 .gso_send_check =       tcp_v4_gso_send_check,
1571                 .gso_segment    =       tcp_tso_segment,
1572                 .gro_receive    =       tcp4_gro_receive,
1573                 .gro_complete   =       tcp4_gro_complete,
1574         },
1575 };
1576
1577 static const struct net_protocol udp_protocol = {
1578         .handler =      udp_rcv,
1579         .err_handler =  udp_err,
1580         .no_policy =    1,
1581         .netns_ok =     1,
1582 };
1583
1584 static const struct net_offload udp_offload = {
1585         .callbacks = {
1586                 .gso_send_check = udp4_ufo_send_check,
1587                 .gso_segment = udp4_ufo_fragment,
1588         },
1589 };
1590
1591 static const struct net_protocol icmp_protocol = {
1592         .handler =      icmp_rcv,
1593         .err_handler =  ping_err,
1594         .no_policy =    1,
1595         .netns_ok =     1,
1596 };
1597
1598 static __net_init int ipv4_mib_init_net(struct net *net)
1599 {
1600         if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1601                           sizeof(struct tcp_mib),
1602                           __alignof__(struct tcp_mib)) < 0)
1603                 goto err_tcp_mib;
1604         if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1605                           sizeof(struct ipstats_mib),
1606                           __alignof__(struct ipstats_mib)) < 0)
1607                 goto err_ip_mib;
1608         if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1609                           sizeof(struct linux_mib),
1610                           __alignof__(struct linux_mib)) < 0)
1611                 goto err_net_mib;
1612         if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1613                           sizeof(struct udp_mib),
1614                           __alignof__(struct udp_mib)) < 0)
1615                 goto err_udp_mib;
1616         if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1617                           sizeof(struct udp_mib),
1618                           __alignof__(struct udp_mib)) < 0)
1619                 goto err_udplite_mib;
1620         if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1621                           sizeof(struct icmp_mib),
1622                           __alignof__(struct icmp_mib)) < 0)
1623                 goto err_icmp_mib;
1624         net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1625                                               GFP_KERNEL);
1626         if (!net->mib.icmpmsg_statistics)
1627                 goto err_icmpmsg_mib;
1628
1629         tcp_mib_init(net);
1630         return 0;
1631
1632 err_icmpmsg_mib:
1633         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1634 err_icmp_mib:
1635         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1636 err_udplite_mib:
1637         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1638 err_udp_mib:
1639         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1640 err_net_mib:
1641         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1642 err_ip_mib:
1643         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1644 err_tcp_mib:
1645         return -ENOMEM;
1646 }
1647
1648 static __net_exit void ipv4_mib_exit_net(struct net *net)
1649 {
1650         kfree(net->mib.icmpmsg_statistics);
1651         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1652         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1653         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1654         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1655         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1656         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1657 }
1658
1659 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1660         .init = ipv4_mib_init_net,
1661         .exit = ipv4_mib_exit_net,
1662 };
1663
1664 static int __init init_ipv4_mibs(void)
1665 {
1666         return register_pernet_subsys(&ipv4_mib_ops);
1667 }
1668
1669 static int ipv4_proc_init(void);
1670
1671 /*
1672  *      IP protocol layer initialiser
1673  */
1674
1675 static struct packet_offload ip_packet_offload __read_mostly = {
1676         .type = cpu_to_be16(ETH_P_IP),
1677         .callbacks = {
1678                 .gso_send_check = inet_gso_send_check,
1679                 .gso_segment = inet_gso_segment,
1680                 .gro_receive = inet_gro_receive,
1681                 .gro_complete = inet_gro_complete,
1682         },
1683 };
1684
1685 static int __init ipv4_offload_init(void)
1686 {
1687         /*
1688          * Add offloads
1689          */
1690         if (inet_add_offload(&udp_offload, IPPROTO_UDP) < 0)
1691                 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1692         if (inet_add_offload(&tcp_offload, IPPROTO_TCP) < 0)
1693                 pr_crit("%s: Cannot add TCP protocol offlaod\n", __func__);
1694
1695         dev_add_offload(&ip_packet_offload);
1696         return 0;
1697 }
1698
1699 fs_initcall(ipv4_offload_init);
1700
1701 static struct packet_type ip_packet_type __read_mostly = {
1702         .type = cpu_to_be16(ETH_P_IP),
1703         .func = ip_rcv,
1704 };
1705
1706 static int __init inet_init(void)
1707 {
1708         struct sk_buff *dummy_skb;
1709         struct inet_protosw *q;
1710         struct list_head *r;
1711         int rc = -EINVAL;
1712
1713         BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1714
1715         sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL);
1716         if (!sysctl_local_reserved_ports)
1717                 goto out;
1718
1719         rc = proto_register(&tcp_prot, 1);
1720         if (rc)
1721                 goto out_free_reserved_ports;
1722
1723         rc = proto_register(&udp_prot, 1);
1724         if (rc)
1725                 goto out_unregister_tcp_proto;
1726
1727         rc = proto_register(&raw_prot, 1);
1728         if (rc)
1729                 goto out_unregister_udp_proto;
1730
1731         rc = proto_register(&ping_prot, 1);
1732         if (rc)
1733                 goto out_unregister_raw_proto;
1734
1735         /*
1736          *      Tell SOCKET that we are alive...
1737          */
1738
1739         (void)sock_register(&inet_family_ops);
1740
1741 #ifdef CONFIG_SYSCTL
1742         ip_static_sysctl_init();
1743 #endif
1744
1745         tcp_prot.sysctl_mem = init_net.ipv4.sysctl_tcp_mem;
1746
1747         /*
1748          *      Add all the base protocols.
1749          */
1750
1751         if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1752                 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1753         if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1754                 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1755         if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1756                 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1757 #ifdef CONFIG_IP_MULTICAST
1758         if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1759                 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1760 #endif
1761
1762         /* Register the socket-side information for inet_create. */
1763         for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1764                 INIT_LIST_HEAD(r);
1765
1766         for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1767                 inet_register_protosw(q);
1768
1769         /*
1770          *      Set the ARP module up
1771          */
1772
1773         arp_init();
1774
1775         /*
1776          *      Set the IP module up
1777          */
1778
1779         ip_init();
1780
1781         tcp_v4_init();
1782
1783         /* Setup TCP slab cache for open requests. */
1784         tcp_init();
1785
1786         /* Setup UDP memory threshold */
1787         udp_init();
1788
1789         /* Add UDP-Lite (RFC 3828) */
1790         udplite4_register();
1791
1792         ping_init();
1793
1794         /*
1795          *      Set the ICMP layer up
1796          */
1797
1798         if (icmp_init() < 0)
1799                 panic("Failed to create the ICMP control socket.\n");
1800
1801         /*
1802          *      Initialise the multicast router
1803          */
1804 #if defined(CONFIG_IP_MROUTE)
1805         if (ip_mr_init())
1806                 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1807 #endif
1808         /*
1809          *      Initialise per-cpu ipv4 mibs
1810          */
1811
1812         if (init_ipv4_mibs())
1813                 pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1814
1815         ipv4_proc_init();
1816
1817         ipfrag_init();
1818
1819         dev_add_pack(&ip_packet_type);
1820
1821         rc = 0;
1822 out:
1823         return rc;
1824 out_unregister_raw_proto:
1825         proto_unregister(&raw_prot);
1826 out_unregister_udp_proto:
1827         proto_unregister(&udp_prot);
1828 out_unregister_tcp_proto:
1829         proto_unregister(&tcp_prot);
1830 out_free_reserved_ports:
1831         kfree(sysctl_local_reserved_ports);
1832         goto out;
1833 }
1834
1835 fs_initcall(inet_init);
1836
1837 /* ------------------------------------------------------------------------ */
1838
1839 #ifdef CONFIG_PROC_FS
1840 static int __init ipv4_proc_init(void)
1841 {
1842         int rc = 0;
1843
1844         if (raw_proc_init())
1845                 goto out_raw;
1846         if (tcp4_proc_init())
1847                 goto out_tcp;
1848         if (udp4_proc_init())
1849                 goto out_udp;
1850         if (ping_proc_init())
1851                 goto out_ping;
1852         if (ip_misc_proc_init())
1853                 goto out_misc;
1854 out:
1855         return rc;
1856 out_misc:
1857         ping_proc_exit();
1858 out_ping:
1859         udp4_proc_exit();
1860 out_udp:
1861         tcp4_proc_exit();
1862 out_tcp:
1863         raw_proc_exit();
1864 out_raw:
1865         rc = -ENOMEM;
1866         goto out;
1867 }
1868
1869 #else /* CONFIG_PROC_FS */
1870 static int __init ipv4_proc_init(void)
1871 {
1872         return 0;
1873 }
1874 #endif /* CONFIG_PROC_FS */
1875
1876 MODULE_ALIAS_NETPROTO(PF_INET);
1877