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 u32 ipv6_hash_secret __read_mostly;
 252 EXPORT_SYMBOL(ipv6_hash_secret);
 253
 254 /*
 255  * inet_ehash_secret must be set exactly once, and to a non nul value
 256  * ipv6_hash_secret must be set exactly once.
 257  */
 258 void build_ehash_secret(void)
 259 {
 260         u32 rnd;
 261
 262         do {
 263                 get_random_bytes(&rnd, sizeof(rnd));
 264         } while (rnd == 0);
 265
 266         if (cmpxchg(&inet_ehash_secret, 0, rnd) == 0)
 267                 get_random_bytes(&ipv6_hash_secret, sizeof(ipv6_hash_secret));
 268 }
 269 EXPORT_SYMBOL(build_ehash_secret);
 270
 271 /*
 272  *      Create an inet socket.
 273  */
 274
 275 static int inet_create(struct net *net, struct socket *sock, int protocol,
 276                        int kern)
 277 {
 278         struct sock *sk;
 279         struct inet_protosw *answer;
 280         struct inet_sock *inet;
 281         struct proto *answer_prot;
 282         unsigned char answer_flags;
 283         char answer_no_check;
 284         int try_loading_module = 0;
 285         int err;
 286
 287         if (unlikely(!inet_ehash_secret))
 288                 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
 289                         build_ehash_secret();
 290
 291         sock->state = SS_UNCONNECTED;
 292
 293         /* Look for the requested type/protocol pair. */
 294 lookup_protocol:
 295         err = -ESOCKTNOSUPPORT;
 296         rcu_read_lock();
 297         list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
 298
 299                 err = 0;
 300                 /* Check the non-wild match. */
 301                 if (protocol == answer->protocol) {
 302                         if (protocol != IPPROTO_IP)
 303                                 break;
 304                 } else {
 305                         /* Check for the two wild cases. */
 306                         if (IPPROTO_IP == protocol) {
 307                                 protocol = answer->protocol;
 308                                 break;
 309                         }
 310                         if (IPPROTO_IP == answer->protocol)
 311                                 break;
 312                 }
 313                 err = -EPROTONOSUPPORT;
 314         }
 315
 316         if (unlikely(err)) {
 317                 if (try_loading_module < 2) {
 318                         rcu_read_unlock();
 319                         /*
 320                          * Be more specific, e.g. net-pf-2-proto-132-type-1
 321                          * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
 322                          */
 323                         if (++try_loading_module == 1)
 324                                 request_module("net-pf-%d-proto-%d-type-%d",
 325                                                PF_INET, protocol, sock->type);
 326                         /*
 327                          * Fall back to generic, e.g. net-pf-2-proto-132
 328                          * (net-pf-PF_INET-proto-IPPROTO_SCTP)
 329                          */
 330                         else
 331                                 request_module("net-pf-%d-proto-%d",
 332                                                PF_INET, protocol);
 333                         goto lookup_protocol;
 334                 } else
 335                         goto out_rcu_unlock;
 336         }
 337
 338         err = -EPERM;
 339         if (sock->type == SOCK_RAW && !kern &&
 340             !ns_capable(net->user_ns, CAP_NET_RAW))
 341                 goto out_rcu_unlock;
 342
 343         sock->ops = answer->ops;
 344         answer_prot = answer->prot;
 345         answer_no_check = answer->no_check;
 346         answer_flags = answer->flags;
 347         rcu_read_unlock();
 348
 349         WARN_ON(answer_prot->slab == NULL);
 350
 351         err = -ENOBUFS;
 352         sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
 353         if (sk == NULL)
 354                 goto out;
 355
 356         err = 0;
 357         sk->sk_no_check = answer_no_check;
 358         if (INET_PROTOSW_REUSE & answer_flags)
 359                 sk->sk_reuse = SK_CAN_REUSE;
 360
 361         inet = inet_sk(sk);
 362         inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
 363
 364         inet->nodefrag = 0;
 365
 366         if (SOCK_RAW == sock->type) {
 367                 inet->inet_num = protocol;
 368                 if (IPPROTO_RAW == protocol)
 369                         inet->hdrincl = 1;
 370         }
 371
 372         if (ipv4_config.no_pmtu_disc)
 373                 inet->pmtudisc = IP_PMTUDISC_DONT;
 374         else
 375                 inet->pmtudisc = IP_PMTUDISC_WANT;
 376
 377         inet->inet_id = 0;
 378
 379         sock_init_data(sock, sk);
 380
 381         sk->sk_destruct    = inet_sock_destruct;
 382         sk->sk_protocol    = protocol;
 383         sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
 384
 385         inet->uc_ttl    = -1;
 386         inet->mc_loop   = 1;
 387         inet->mc_ttl    = 1;
 388         inet->mc_all    = 1;
 389         inet->mc_index  = 0;
 390         inet->mc_list   = NULL;
 391         inet->rcv_tos   = 0;
 392
 393         sk_refcnt_debug_inc(sk);
 394
 395         if (inet->inet_num) {
 396                 /* It assumes that any protocol which allows
 397                  * the user to assign a number at socket
 398                  * creation time automatically
 399                  * shares.
 400                  */
 401                 inet->inet_sport = htons(inet->inet_num);
 402                 /* Add to protocol hash chains. */
 403                 sk->sk_prot->hash(sk);
 404         }
 405
 406         if (sk->sk_prot->init) {
 407                 err = sk->sk_prot->init(sk);
 408                 if (err)
 409                         sk_common_release(sk);
 410         }
 411 out:
 412         return err;
 413 out_rcu_unlock:
 414         rcu_read_unlock();
 415         goto out;
 416 }
 417
 418
 419 /*
 420  *      The peer socket should always be NULL (or else). When we call this
 421  *      function we are destroying the object and from then on nobody
 422  *      should refer to it.
 423  */
 424 int inet_release(struct socket *sock)
 425 {
 426         struct sock *sk = sock->sk;
 427
 428         if (sk) {
 429                 long timeout;
 430
 431                 sock_rps_reset_flow(sk);
 432
 433                 /* Applications forget to leave groups before exiting */
 434                 ip_mc_drop_socket(sk);
 435
 436                 /* If linger is set, we don't return until the close
 437                  * is complete.  Otherwise we return immediately. The
 438                  * actually closing is done the same either way.
 439                  *
 440                  * If the close is due to the process exiting, we never
 441                  * linger..
 442                  */
 443                 timeout = 0;
 444                 if (sock_flag(sk, SOCK_LINGER) &&
 445                     !(current->flags & PF_EXITING))
 446                         timeout = sk->sk_lingertime;
 447                 sock->sk = NULL;
 448                 sk->sk_prot->close(sk, timeout);
 449         }
 450         return 0;
 451 }
 452 EXPORT_SYMBOL(inet_release);
 453
 454 /* It is off by default, see below. */
 455 int sysctl_ip_nonlocal_bind __read_mostly;
 456 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
 457
 458 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 459 {
 460         struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
 461         struct sock *sk = sock->sk;
 462         struct inet_sock *inet = inet_sk(sk);
 463         struct net *net = sock_net(sk);
 464         unsigned short snum;
 465         int chk_addr_ret;
 466         int err;
 467
 468         /* If the socket has its own bind function then use it. (RAW) */
 469         if (sk->sk_prot->bind) {
 470                 err = sk->sk_prot->bind(sk, uaddr, addr_len);
 471                 goto out;
 472         }
 473         err = -EINVAL;
 474         if (addr_len < sizeof(struct sockaddr_in))
 475                 goto out;
 476
 477         if (addr->sin_family != AF_INET) {
 478                 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
 479                  * only if s_addr is INADDR_ANY.
 480                  */
 481                 err = -EAFNOSUPPORT;
 482                 if (addr->sin_family != AF_UNSPEC ||
 483                     addr->sin_addr.s_addr != htonl(INADDR_ANY))
 484                         goto out;
 485         }
 486
 487         chk_addr_ret = inet_addr_type(net, addr->sin_addr.s_addr);
 488
 489         /* Not specified by any standard per-se, however it breaks too
 490          * many applications when removed.  It is unfortunate since
 491          * allowing applications to make a non-local bind solves
 492          * several problems with systems using dynamic addressing.
 493          * (ie. your servers still start up even if your ISDN link
 494          *  is temporarily down)
 495          */
 496         err = -EADDRNOTAVAIL;
 497         if (!sysctl_ip_nonlocal_bind &&
 498             !(inet->freebind || inet->transparent) &&
 499             addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
 500             chk_addr_ret != RTN_LOCAL &&
 501             chk_addr_ret != RTN_MULTICAST &&
 502             chk_addr_ret != RTN_BROADCAST)
 503                 goto out;
 504
 505         snum = ntohs(addr->sin_port);
 506         err = -EACCES;
 507         if (snum && snum < PROT_SOCK &&
 508             !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
 509                 goto out;
 510
 511         /*      We keep a pair of addresses. rcv_saddr is the one
 512          *      used by hash lookups, and saddr is used for transmit.
 513          *
 514          *      In the BSD API these are the same except where it
 515          *      would be illegal to use them (multicast/broadcast) in
 516          *      which case the sending device address is used.
 517          */
 518         lock_sock(sk);
 519
 520         /* Check these errors (active socket, double bind). */
 521         err = -EINVAL;
 522         if (sk->sk_state != TCP_CLOSE || inet->inet_num)
 523                 goto out_release_sock;
 524
 525         inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
 526         if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
 527                 inet->inet_saddr = 0;  /* Use device */
 528
 529         /* Make sure we are allowed to bind here. */
 530         if (sk->sk_prot->get_port(sk, snum)) {
 531                 inet->inet_saddr = inet->inet_rcv_saddr = 0;
 532                 err = -EADDRINUSE;
 533                 goto out_release_sock;
 534         }
 535
 536         if (inet->inet_rcv_saddr)
 537                 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
 538         if (snum)
 539                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
 540         inet->inet_sport = htons(inet->inet_num);
 541         inet->inet_daddr = 0;
 542         inet->inet_dport = 0;
 543         sk_dst_reset(sk);
 544         err = 0;
 545 out_release_sock:
 546         release_sock(sk);
 547 out:
 548         return err;
 549 }
 550 EXPORT_SYMBOL(inet_bind);
 551
 552 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
 553                        int addr_len, int flags)
 554 {
 555         struct sock *sk = sock->sk;
 556
 557         if (addr_len < sizeof(uaddr->sa_family))
 558                 return -EINVAL;
 559         if (uaddr->sa_family == AF_UNSPEC)
 560                 return sk->sk_prot->disconnect(sk, flags);
 561
 562         if (!inet_sk(sk)->inet_num && inet_autobind(sk))
 563                 return -EAGAIN;
 564         return sk->sk_prot->connect(sk, uaddr, addr_len);
 565 }
 566 EXPORT_SYMBOL(inet_dgram_connect);
 567
 568 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
 569 {
 570         DEFINE_WAIT(wait);
 571
 572         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
 573         sk->sk_write_pending += writebias;
 574
 575         /* Basic assumption: if someone sets sk->sk_err, he _must_
 576          * change state of the socket from TCP_SYN_*.
 577          * Connect() does not allow to get error notifications
 578          * without closing the socket.
 579          */
 580         while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 581                 release_sock(sk);
 582                 timeo = schedule_timeout(timeo);
 583                 lock_sock(sk);
 584                 if (signal_pending(current) || !timeo)
 585                         break;
 586                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
 587         }
 588         finish_wait(sk_sleep(sk), &wait);
 589         sk->sk_write_pending -= writebias;
 590         return timeo;
 591 }
 592
 593 /*
 594  *      Connect to a remote host. There is regrettably still a little
 595  *      TCP 'magic' in here.
 596  */
 597 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 598                           int addr_len, int flags)
 599 {
 600         struct sock *sk = sock->sk;
 601         int err;
 602         long timeo;
 603
 604         if (addr_len < sizeof(uaddr->sa_family))
 605                 return -EINVAL;
 606
 607         if (uaddr->sa_family == AF_UNSPEC) {
 608                 err = sk->sk_prot->disconnect(sk, flags);
 609                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 610                 goto out;
 611         }
 612
 613         switch (sock->state) {
 614         default:
 615                 err = -EINVAL;
 616                 goto out;
 617         case SS_CONNECTED:
 618                 err = -EISCONN;
 619                 goto out;
 620         case SS_CONNECTING:
 621                 err = -EALREADY;
 622                 /* Fall out of switch with err, set for this state */
 623                 break;
 624         case SS_UNCONNECTED:
 625                 err = -EISCONN;
 626                 if (sk->sk_state != TCP_CLOSE)
 627                         goto out;
 628
 629                 err = sk->sk_prot->connect(sk, uaddr, addr_len);
 630                 if (err < 0)
 631                         goto out;
 632
 633                 sock->state = SS_CONNECTING;
 634
 635                 /* Just entered SS_CONNECTING state; the only
 636                  * difference is that return value in non-blocking
 637                  * case is EINPROGRESS, rather than EALREADY.
 638                  */
 639                 err = -EINPROGRESS;
 640                 break;
 641         }
 642
 643         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
 644
 645         if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 646                 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
 647                                 tcp_sk(sk)->fastopen_req &&
 648                                 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
 649
 650                 /* Error code is set above */
 651                 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
 652                         goto out;
 653
 654                 err = sock_intr_errno(timeo);
 655                 if (signal_pending(current))
 656                         goto out;
 657         }
 658
 659         /* Connection was closed by RST, timeout, ICMP error
 660          * or another process disconnected us.
 661          */
 662         if (sk->sk_state == TCP_CLOSE)
 663                 goto sock_error;
 664
 665         /* sk->sk_err may be not zero now, if RECVERR was ordered by user
 666          * and error was received after socket entered established state.
 667          * Hence, it is handled normally after connect() return successfully.
 668          */
 669
 670         sock->state = SS_CONNECTED;
 671         err = 0;
 672 out:
 673         return err;
 674
 675 sock_error:
 676         err = sock_error(sk) ? : -ECONNABORTED;
 677         sock->state = SS_UNCONNECTED;
 678         if (sk->sk_prot->disconnect(sk, flags))
 679                 sock->state = SS_DISCONNECTING;
 680         goto out;
 681 }
 682 EXPORT_SYMBOL(__inet_stream_connect);
 683
 684 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 685                         int addr_len, int flags)
 686 {
 687         int err;
 688
 689         lock_sock(sock->sk);
 690         err = __inet_stream_connect(sock, uaddr, addr_len, flags);
 691         release_sock(sock->sk);
 692         return err;
 693 }
 694 EXPORT_SYMBOL(inet_stream_connect);
 695
 696 /*
 697  *      Accept a pending connection. The TCP layer now gives BSD semantics.
 698  */
 699
 700 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
 701 {
 702         struct sock *sk1 = sock->sk;
 703         int err = -EINVAL;
 704         struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
 705
 706         if (!sk2)
 707                 goto do_err;
 708
 709         lock_sock(sk2);
 710
 711         sock_rps_record_flow(sk2);
 712         WARN_ON(!((1 << sk2->sk_state) &
 713                   (TCPF_ESTABLISHED | TCPF_SYN_RECV |
 714                   TCPF_CLOSE_WAIT | TCPF_CLOSE)));
 715
 716         sock_graft(sk2, newsock);
 717
 718         newsock->state = SS_CONNECTED;
 719         err = 0;
 720         release_sock(sk2);
 721 do_err:
 722         return err;
 723 }
 724 EXPORT_SYMBOL(inet_accept);
 725
 726
 727 /*
 728  *      This does both peername and sockname.
 729  */
 730 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
 731                         int *uaddr_len, int peer)
 732 {
 733         struct sock *sk         = sock->sk;
 734         struct inet_sock *inet  = inet_sk(sk);
 735         DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
 736
 737         sin->sin_family = AF_INET;
 738         if (peer) {
 739                 if (!inet->inet_dport ||
 740                     (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
 741                      peer == 1))
 742                         return -ENOTCONN;
 743                 sin->sin_port = inet->inet_dport;
 744                 sin->sin_addr.s_addr = inet->inet_daddr;
 745         } else {
 746                 __be32 addr = inet->inet_rcv_saddr;
 747                 if (!addr)
 748                         addr = inet->inet_saddr;
 749                 sin->sin_port = inet->inet_sport;
 750                 sin->sin_addr.s_addr = addr;
 751         }
 752         memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 753         *uaddr_len = sizeof(*sin);
 754         return 0;
 755 }
 756 EXPORT_SYMBOL(inet_getname);
 757
 758 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
 759                  size_t size)
 760 {
 761         struct sock *sk = sock->sk;
 762
 763         sock_rps_record_flow(sk);
 764
 765         /* We may need to bind the socket. */
 766         if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
 767             inet_autobind(sk))
 768                 return -EAGAIN;
 769
 770         return sk->sk_prot->sendmsg(iocb, sk, msg, size);
 771 }
 772 EXPORT_SYMBOL(inet_sendmsg);
 773
 774 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
 775                       size_t size, int flags)
 776 {
 777         struct sock *sk = sock->sk;
 778
 779         sock_rps_record_flow(sk);
 780
 781         /* We may need to bind the socket. */
 782         if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
 783             inet_autobind(sk))
 784                 return -EAGAIN;
 785
 786         if (sk->sk_prot->sendpage)
 787                 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
 788         return sock_no_sendpage(sock, page, offset, size, flags);
 789 }
 790 EXPORT_SYMBOL(inet_sendpage);
 791
 792 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
 793                  size_t size, int flags)
 794 {
 795         struct sock *sk = sock->sk;
 796         int addr_len = 0;
 797         int err;
 798
 799         sock_rps_record_flow(sk);
 800
 801         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
 802                                    flags & ~MSG_DONTWAIT, &addr_len);
 803         if (err >= 0)
 804                 msg->msg_namelen = addr_len;
 805         return err;
 806 }
 807 EXPORT_SYMBOL(inet_recvmsg);
 808
 809 int inet_shutdown(struct socket *sock, int how)
 810 {
 811         struct sock *sk = sock->sk;
 812         int err = 0;
 813
 814         /* This should really check to make sure
 815          * the socket is a TCP socket. (WHY AC...)
 816          */
 817         how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
 818                        1->2 bit 2 snds.
 819                        2->3 */
 820         if ((how & ~SHUTDOWN_MASK) || !how)     /* MAXINT->0 */
 821                 return -EINVAL;
 822
 823         lock_sock(sk);
 824         if (sock->state == SS_CONNECTING) {
 825                 if ((1 << sk->sk_state) &
 826                     (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
 827                         sock->state = SS_DISCONNECTING;
 828                 else
 829                         sock->state = SS_CONNECTED;
 830         }
 831
 832         switch (sk->sk_state) {
 833         case TCP_CLOSE:
 834                 err = -ENOTCONN;
 835                 /* Hack to wake up other listeners, who can poll for
 836                    POLLHUP, even on eg. unconnected UDP sockets -- RR */
 837         default:
 838                 sk->sk_shutdown |= how;
 839                 if (sk->sk_prot->shutdown)
 840                         sk->sk_prot->shutdown(sk, how);
 841                 break;
 842
 843         /* Remaining two branches are temporary solution for missing
 844          * close() in multithreaded environment. It is _not_ a good idea,
 845          * but we have no choice until close() is repaired at VFS level.
 846          */
 847         case TCP_LISTEN:
 848                 if (!(how & RCV_SHUTDOWN))
 849                         break;
 850                 /* Fall through */
 851         case TCP_SYN_SENT:
 852                 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
 853                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 854                 break;
 855         }
 856
 857         /* Wake up anyone sleeping in poll. */
 858         sk->sk_state_change(sk);
 859         release_sock(sk);
 860         return err;
 861 }
 862 EXPORT_SYMBOL(inet_shutdown);
 863
 864 /*
 865  *      ioctl() calls you can issue on an INET socket. Most of these are
 866  *      device configuration and stuff and very rarely used. Some ioctls
 867  *      pass on to the socket itself.
 868  *
 869  *      NOTE: I like the idea of a module for the config stuff. ie ifconfig
 870  *      loads the devconfigure module does its configuring and unloads it.
 871  *      There's a good 20K of config code hanging around the kernel.
 872  */
 873
 874 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 875 {
 876         struct sock *sk = sock->sk;
 877         int err = 0;
 878         struct net *net = sock_net(sk);
 879
 880         switch (cmd) {
 881         case SIOCGSTAMP:
 882                 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
 883                 break;
 884         case SIOCGSTAMPNS:
 885                 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
 886                 break;
 887         case SIOCADDRT:
 888         case SIOCDELRT:
 889         case SIOCRTMSG:
 890                 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
 891                 break;
 892         case SIOCDARP:
 893         case SIOCGARP:
 894         case SIOCSARP:
 895                 err = arp_ioctl(net, cmd, (void __user *)arg);
 896                 break;
 897         case SIOCGIFADDR:
 898         case SIOCSIFADDR:
 899         case SIOCGIFBRDADDR:
 900         case SIOCSIFBRDADDR:
 901         case SIOCGIFNETMASK:
 902         case SIOCSIFNETMASK:
 903         case SIOCGIFDSTADDR:
 904         case SIOCSIFDSTADDR:
 905         case SIOCSIFPFLAGS:
 906         case SIOCGIFPFLAGS:
 907         case SIOCSIFFLAGS:
 908                 err = devinet_ioctl(net, cmd, (void __user *)arg);
 909                 break;
 910         default:
 911                 if (sk->sk_prot->ioctl)
 912                         err = sk->sk_prot->ioctl(sk, cmd, arg);
 913                 else
 914                         err = -ENOIOCTLCMD;
 915                 break;
 916         }
 917         return err;
 918 }
 919 EXPORT_SYMBOL(inet_ioctl);
 920
 921 #ifdef CONFIG_COMPAT
 922 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 923 {
 924         struct sock *sk = sock->sk;
 925         int err = -ENOIOCTLCMD;
 926
 927         if (sk->sk_prot->compat_ioctl)
 928                 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
 929
 930         return err;
 931 }
 932 #endif
 933
 934 const struct proto_ops inet_stream_ops = {
 935         .family            = PF_INET,
 936         .owner             = THIS_MODULE,
 937         .release           = inet_release,
 938         .bind              = inet_bind,
 939         .connect           = inet_stream_connect,
 940         .socketpair        = sock_no_socketpair,
 941         .accept            = inet_accept,
 942         .getname           = inet_getname,
 943         .poll              = tcp_poll,
 944         .ioctl             = inet_ioctl,
 945         .listen            = inet_listen,
 946         .shutdown          = inet_shutdown,
 947         .setsockopt        = sock_common_setsockopt,
 948         .getsockopt        = sock_common_getsockopt,
 949         .sendmsg           = inet_sendmsg,
 950         .recvmsg           = inet_recvmsg,
 951         .mmap              = sock_no_mmap,
 952         .sendpage          = inet_sendpage,
 953         .splice_read       = tcp_splice_read,
 954 #ifdef CONFIG_COMPAT
 955         .compat_setsockopt = compat_sock_common_setsockopt,
 956         .compat_getsockopt = compat_sock_common_getsockopt,
 957         .compat_ioctl      = inet_compat_ioctl,
 958 #endif
 959 };
 960 EXPORT_SYMBOL(inet_stream_ops);
 961
 962 const struct proto_ops inet_dgram_ops = {
 963         .family            = PF_INET,
 964         .owner             = THIS_MODULE,
 965         .release           = inet_release,
 966         .bind              = inet_bind,
 967         .connect           = inet_dgram_connect,
 968         .socketpair        = sock_no_socketpair,
 969         .accept            = sock_no_accept,
 970         .getname           = inet_getname,
 971         .poll              = udp_poll,
 972         .ioctl             = inet_ioctl,
 973         .listen            = sock_no_listen,
 974         .shutdown          = inet_shutdown,
 975         .setsockopt        = sock_common_setsockopt,
 976         .getsockopt        = sock_common_getsockopt,
 977         .sendmsg           = inet_sendmsg,
 978         .recvmsg           = inet_recvmsg,
 979         .mmap              = sock_no_mmap,
 980         .sendpage          = inet_sendpage,
 981 #ifdef CONFIG_COMPAT
 982         .compat_setsockopt = compat_sock_common_setsockopt,
 983         .compat_getsockopt = compat_sock_common_getsockopt,
 984         .compat_ioctl      = inet_compat_ioctl,
 985 #endif
 986 };
 987 EXPORT_SYMBOL(inet_dgram_ops);
 988
 989 /*
 990  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
 991  * udp_poll
 992  */
 993 static const struct proto_ops inet_sockraw_ops = {
 994         .family            = PF_INET,
 995         .owner             = THIS_MODULE,
 996         .release           = inet_release,
 997         .bind              = inet_bind,
 998         .connect           = inet_dgram_connect,
 999         .socketpair        = sock_no_socketpair,
1000         .accept            = sock_no_accept,
1001         .getname           = inet_getname,
1002         .poll              = datagram_poll,
1003         .ioctl             = inet_ioctl,
1004         .listen            = sock_no_listen,
1005         .shutdown          = inet_shutdown,
1006         .setsockopt        = sock_common_setsockopt,
1007         .getsockopt        = sock_common_getsockopt,
1008         .sendmsg           = inet_sendmsg,
1009         .recvmsg           = inet_recvmsg,
1010         .mmap              = sock_no_mmap,
1011         .sendpage          = inet_sendpage,
1012 #ifdef CONFIG_COMPAT
1013         .compat_setsockopt = compat_sock_common_setsockopt,
1014         .compat_getsockopt = compat_sock_common_getsockopt,
1015         .compat_ioctl      = inet_compat_ioctl,
1016 #endif
1017 };
1018
1019 static const struct net_proto_family inet_family_ops = {
1020         .family = PF_INET,
1021         .create = inet_create,
1022         .owner  = THIS_MODULE,
1023 };
1024
1025 /* Upon startup we insert all the elements in inetsw_array[] into
1026  * the linked list inetsw.
1027  */
1028 static struct inet_protosw inetsw_array[] =
1029 {
1030         {
1031                 .type =       SOCK_STREAM,
1032                 .protocol =   IPPROTO_TCP,
1033                 .prot =       &tcp_prot,
1034                 .ops =        &inet_stream_ops,
1035                 .no_check =   0,
1036                 .flags =      INET_PROTOSW_PERMANENT |
1037                               INET_PROTOSW_ICSK,
1038         },
1039
1040         {
1041                 .type =       SOCK_DGRAM,
1042                 .protocol =   IPPROTO_UDP,
1043                 .prot =       &udp_prot,
1044                 .ops =        &inet_dgram_ops,
1045                 .no_check =   UDP_CSUM_DEFAULT,
1046                 .flags =      INET_PROTOSW_PERMANENT,
1047        },
1048
1049        {
1050                 .type =       SOCK_DGRAM,
1051                 .protocol =   IPPROTO_ICMP,
1052                 .prot =       &ping_prot,
1053                 .ops =        &inet_dgram_ops,
1054                 .no_check =   UDP_CSUM_DEFAULT,
1055                 .flags =      INET_PROTOSW_REUSE,
1056        },
1057
1058        {
1059                .type =       SOCK_RAW,
1060                .protocol =   IPPROTO_IP,        /* wild card */
1061                .prot =       &raw_prot,
1062                .ops =        &inet_sockraw_ops,
1063                .no_check =   UDP_CSUM_DEFAULT,
1064                .flags =      INET_PROTOSW_REUSE,
1065        }
1066 };
1067
1068 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1069
1070 void inet_register_protosw(struct inet_protosw *p)
1071 {
1072         struct list_head *lh;
1073         struct inet_protosw *answer;
1074         int protocol = p->protocol;
1075         struct list_head *last_perm;
1076
1077         spin_lock_bh(&inetsw_lock);
1078
1079         if (p->type >= SOCK_MAX)
1080                 goto out_illegal;
1081
1082         /* If we are trying to override a permanent protocol, bail. */
1083         answer = NULL;
1084         last_perm = &inetsw[p->type];
1085         list_for_each(lh, &inetsw[p->type]) {
1086                 answer = list_entry(lh, struct inet_protosw, list);
1087
1088                 /* Check only the non-wild match. */
1089                 if (INET_PROTOSW_PERMANENT & answer->flags) {
1090                         if (protocol == answer->protocol)
1091                                 break;
1092                         last_perm = lh;
1093                 }
1094
1095                 answer = NULL;
1096         }
1097         if (answer)
1098                 goto out_permanent;
1099
1100         /* Add the new entry after the last permanent entry if any, so that
1101          * the new entry does not override a permanent entry when matched with
1102          * a wild-card protocol. But it is allowed to override any existing
1103          * non-permanent entry.  This means that when we remove this entry, the
1104          * system automatically returns to the old behavior.
1105          */
1106         list_add_rcu(&p->list, last_perm);
1107 out:
1108         spin_unlock_bh(&inetsw_lock);
1109
1110         return;
1111
1112 out_permanent:
1113         pr_err("Attempt to override permanent protocol %d\n", protocol);
1114         goto out;
1115
1116 out_illegal:
1117         pr_err("Ignoring attempt to register invalid socket type %d\n",
1118                p->type);
1119         goto out;
1120 }
1121 EXPORT_SYMBOL(inet_register_protosw);
1122
1123 void inet_unregister_protosw(struct inet_protosw *p)
1124 {
1125         if (INET_PROTOSW_PERMANENT & p->flags) {
1126                 pr_err("Attempt to unregister permanent protocol %d\n",
1127                        p->protocol);
1128         } else {
1129                 spin_lock_bh(&inetsw_lock);
1130                 list_del_rcu(&p->list);
1131                 spin_unlock_bh(&inetsw_lock);
1132
1133                 synchronize_net();
1134         }
1135 }
1136 EXPORT_SYMBOL(inet_unregister_protosw);
1137
1138 /*
1139  *      Shall we try to damage output packets if routing dev changes?
1140  */
1141
1142 int sysctl_ip_dynaddr __read_mostly;
1143
1144 static int inet_sk_reselect_saddr(struct sock *sk)
1145 {
1146         struct inet_sock *inet = inet_sk(sk);
1147         __be32 old_saddr = inet->inet_saddr;
1148         __be32 daddr = inet->inet_daddr;
1149         struct flowi4 *fl4;
1150         struct rtable *rt;
1151         __be32 new_saddr;
1152         struct ip_options_rcu *inet_opt;
1153
1154         inet_opt = rcu_dereference_protected(inet->inet_opt,
1155                                              sock_owned_by_user(sk));
1156         if (inet_opt && inet_opt->opt.srr)
1157                 daddr = inet_opt->opt.faddr;
1158
1159         /* Query new route. */
1160         fl4 = &inet->cork.fl.u.ip4;
1161         rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1162                               sk->sk_bound_dev_if, sk->sk_protocol,
1163                               inet->inet_sport, inet->inet_dport, sk, false);
1164         if (IS_ERR(rt))
1165                 return PTR_ERR(rt);
1166
1167         sk_setup_caps(sk, &rt->dst);
1168
1169         new_saddr = fl4->saddr;
1170
1171         if (new_saddr == old_saddr)
1172                 return 0;
1173
1174         if (sysctl_ip_dynaddr > 1) {
1175                 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1176                         __func__, &old_saddr, &new_saddr);
1177         }
1178
1179         inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1180
1181         /*
1182          * XXX The only one ugly spot where we need to
1183          * XXX really change the sockets identity after
1184          * XXX it has entered the hashes. -DaveM
1185          *
1186          * Besides that, it does not check for connection
1187          * uniqueness. Wait for troubles.
1188          */
1189         __sk_prot_rehash(sk);
1190         return 0;
1191 }
1192
1193 int inet_sk_rebuild_header(struct sock *sk)
1194 {
1195         struct inet_sock *inet = inet_sk(sk);
1196         struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1197         __be32 daddr;
1198         struct ip_options_rcu *inet_opt;
1199         struct flowi4 *fl4;
1200         int err;
1201
1202         /* Route is OK, nothing to do. */
1203         if (rt)
1204                 return 0;
1205
1206         /* Reroute. */
1207         rcu_read_lock();
1208         inet_opt = rcu_dereference(inet->inet_opt);
1209         daddr = inet->inet_daddr;
1210         if (inet_opt && inet_opt->opt.srr)
1211                 daddr = inet_opt->opt.faddr;
1212         rcu_read_unlock();
1213         fl4 = &inet->cork.fl.u.ip4;
1214         rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1215                                    inet->inet_dport, inet->inet_sport,
1216                                    sk->sk_protocol, RT_CONN_FLAGS(sk),
1217                                    sk->sk_bound_dev_if);
1218         if (!IS_ERR(rt)) {
1219                 err = 0;
1220                 sk_setup_caps(sk, &rt->dst);
1221         } else {
1222                 err = PTR_ERR(rt);
1223
1224                 /* Routing failed... */
1225                 sk->sk_route_caps = 0;
1226                 /*
1227                  * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1228                  * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1229                  */
1230                 if (!sysctl_ip_dynaddr ||
1231                     sk->sk_state != TCP_SYN_SENT ||
1232                     (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1233                     (err = inet_sk_reselect_saddr(sk)) != 0)
1234                         sk->sk_err_soft = -err;
1235         }
1236
1237         return err;
1238 }
1239 EXPORT_SYMBOL(inet_sk_rebuild_header);
1240
1241 static int inet_gso_send_check(struct sk_buff *skb)
1242 {
1243         const struct net_offload *ops;
1244         const struct iphdr *iph;
1245         int proto;
1246         int ihl;
1247         int err = -EINVAL;
1248
1249         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1250                 goto out;
1251
1252         iph = ip_hdr(skb);
1253         ihl = iph->ihl * 4;
1254         if (ihl < sizeof(*iph))
1255                 goto out;
1256
1257         if (unlikely(!pskb_may_pull(skb, ihl)))
1258                 goto out;
1259
1260         __skb_pull(skb, ihl);
1261         skb_reset_transport_header(skb);
1262         iph = ip_hdr(skb);
1263         proto = iph->protocol;
1264         err = -EPROTONOSUPPORT;
1265
1266         rcu_read_lock();
1267         ops = rcu_dereference(inet_offloads[proto]);
1268         if (likely(ops && ops->callbacks.gso_send_check))
1269                 err = ops->callbacks.gso_send_check(skb);
1270         rcu_read_unlock();
1271
1272 out:
1273         return err;
1274 }
1275
1276 static struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1277         netdev_features_t features)
1278 {
1279         struct sk_buff *segs = ERR_PTR(-EINVAL);
1280         const struct net_offload *ops;
1281         struct iphdr *iph;
1282         int proto;
1283         int ihl;
1284         int id;
1285         unsigned int offset = 0;
1286
1287         if (!(features & NETIF_F_V4_CSUM))
1288                 features &= ~NETIF_F_SG;
1289
1290         if (unlikely(skb_shinfo(skb)->gso_type &
1291                      ~(SKB_GSO_TCPV4 |
1292                        SKB_GSO_UDP |
1293                        SKB_GSO_DODGY |
1294                        SKB_GSO_TCP_ECN |
1295                        SKB_GSO_GRE |
1296                        0)))
1297                 goto out;
1298
1299         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1300                 goto out;
1301
1302         iph = ip_hdr(skb);
1303         ihl = iph->ihl * 4;
1304         if (ihl < sizeof(*iph))
1305                 goto out;
1306
1307         if (unlikely(!pskb_may_pull(skb, ihl)))
1308                 goto out;
1309
1310         __skb_pull(skb, ihl);
1311         skb_reset_transport_header(skb);
1312         iph = ip_hdr(skb);
1313         id = ntohs(iph->id);
1314         proto = iph->protocol;
1315         segs = ERR_PTR(-EPROTONOSUPPORT);
1316
1317         rcu_read_lock();
1318         ops = rcu_dereference(inet_offloads[proto]);
1319         if (likely(ops && ops->callbacks.gso_segment))
1320                 segs = ops->callbacks.gso_segment(skb, features);
1321         rcu_read_unlock();
1322
1323         if (IS_ERR_OR_NULL(segs))
1324                 goto out;
1325
1326         skb = segs;
1327         do {
1328                 iph = ip_hdr(skb);
1329                 if (proto == IPPROTO_UDP) {
1330                         iph->id = htons(id);
1331                         iph->frag_off = htons(offset >> 3);
1332                         if (skb->next != NULL)
1333                                 iph->frag_off |= htons(IP_MF);
1334                         offset += (skb->len - skb->mac_len - iph->ihl * 4);
1335                 } else  {
1336                         if (!(iph->frag_off & htons(IP_DF)))
1337                                 iph->id = htons(id++);
1338                 }
1339                 iph->tot_len = htons(skb->len - skb->mac_len);
1340                 iph->check = 0;
1341                 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1342         } while ((skb = skb->next));
1343
1344 out:
1345         return segs;
1346 }
1347
1348 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1349                                          struct sk_buff *skb)
1350 {
1351         const struct net_offload *ops;
1352         struct sk_buff **pp = NULL;
1353         struct sk_buff *p;
1354         const struct iphdr *iph;
1355         unsigned int hlen;
1356         unsigned int off;
1357         unsigned int id;
1358         int flush = 1;
1359         int proto;
1360
1361         off = skb_gro_offset(skb);
1362         hlen = off + sizeof(*iph);
1363         iph = skb_gro_header_fast(skb, off);
1364         if (skb_gro_header_hard(skb, hlen)) {
1365                 iph = skb_gro_header_slow(skb, hlen, off);
1366                 if (unlikely(!iph))
1367                         goto out;
1368         }
1369
1370         proto = iph->protocol;
1371
1372         rcu_read_lock();
1373         ops = rcu_dereference(inet_offloads[proto]);
1374         if (!ops || !ops->callbacks.gro_receive)
1375                 goto out_unlock;
1376
1377         if (*(u8 *)iph != 0x45)
1378                 goto out_unlock;
1379
1380         if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1381                 goto out_unlock;
1382
1383         id = ntohl(*(__be32 *)&iph->id);
1384         flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF));
1385         id >>= 16;
1386
1387         for (p = *head; p; p = p->next) {
1388                 struct iphdr *iph2;
1389
1390                 if (!NAPI_GRO_CB(p)->same_flow)
1391                         continue;
1392
1393                 iph2 = ip_hdr(p);
1394
1395                 if ((iph->protocol ^ iph2->protocol) |
1396                     ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1397                     ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1398                         NAPI_GRO_CB(p)->same_flow = 0;
1399                         continue;
1400                 }
1401
1402                 /* All fields must match except length and checksum. */
1403                 NAPI_GRO_CB(p)->flush |=
1404                         (iph->ttl ^ iph2->ttl) |
1405                         (iph->tos ^ iph2->tos) |
1406                         ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1407
1408                 NAPI_GRO_CB(p)->flush |= flush;
1409         }
1410
1411         NAPI_GRO_CB(skb)->flush |= flush;
1412         skb_gro_pull(skb, sizeof(*iph));
1413         skb_set_transport_header(skb, skb_gro_offset(skb));
1414
1415         pp = ops->callbacks.gro_receive(head, skb);
1416
1417 out_unlock:
1418         rcu_read_unlock();
1419
1420 out:
1421         NAPI_GRO_CB(skb)->flush |= flush;
1422
1423         return pp;
1424 }
1425
1426 static int inet_gro_complete(struct sk_buff *skb)
1427 {
1428         __be16 newlen = htons(skb->len - skb_network_offset(skb));
1429         struct iphdr *iph = ip_hdr(skb);
1430         const struct net_offload *ops;
1431         int proto = iph->protocol;
1432         int err = -ENOSYS;
1433
1434         csum_replace2(&iph->check, iph->tot_len, newlen);
1435         iph->tot_len = newlen;
1436
1437         rcu_read_lock();
1438         ops = rcu_dereference(inet_offloads[proto]);
1439         if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1440                 goto out_unlock;
1441
1442         err = ops->callbacks.gro_complete(skb);
1443
1444 out_unlock:
1445         rcu_read_unlock();
1446
1447         return err;
1448 }
1449
1450 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1451                          unsigned short type, unsigned char protocol,
1452                          struct net *net)
1453 {
1454         struct socket *sock;
1455         int rc = sock_create_kern(family, type, protocol, &sock);
1456
1457         if (rc == 0) {
1458                 *sk = sock->sk;
1459                 (*sk)->sk_allocation = GFP_ATOMIC;
1460                 /*
1461                  * Unhash it so that IP input processing does not even see it,
1462                  * we do not wish this socket to see incoming packets.
1463                  */
1464                 (*sk)->sk_prot->unhash(*sk);
1465
1466                 sk_change_net(*sk, net);
1467         }
1468         return rc;
1469 }
1470 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1471
1472 unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1473 {
1474         unsigned long res = 0;
1475         int i, j;
1476
1477         for_each_possible_cpu(i) {
1478                 for (j = 0; j < SNMP_ARRAY_SZ; j++)
1479                         res += *(((unsigned long *) per_cpu_ptr(mib[j], i)) + offt);
1480         }
1481         return res;
1482 }
1483 EXPORT_SYMBOL_GPL(snmp_fold_field);
1484
1485 #if BITS_PER_LONG==32
1486
1487 u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset)
1488 {
1489         u64 res = 0;
1490         int cpu;
1491
1492         for_each_possible_cpu(cpu) {
1493                 void *bhptr;
1494                 struct u64_stats_sync *syncp;
1495                 u64 v;
1496                 unsigned int start;
1497
1498                 bhptr = per_cpu_ptr(mib[0], cpu);
1499                 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1500                 do {
1501                         start = u64_stats_fetch_begin_bh(syncp);
1502                         v = *(((u64 *) bhptr) + offt);
1503                 } while (u64_stats_fetch_retry_bh(syncp, start));
1504
1505                 res += v;
1506         }
1507         return res;
1508 }
1509 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1510 #endif
1511
1512 int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align)
1513 {
1514         BUG_ON(ptr == NULL);
1515         ptr[0] = __alloc_percpu(mibsize, align);
1516         if (!ptr[0])
1517                 return -ENOMEM;
1518 #if SNMP_ARRAY_SZ == 2
1519         ptr[1] = __alloc_percpu(mibsize, align);
1520         if (!ptr[1]) {
1521                 free_percpu(ptr[0]);
1522                 ptr[0] = NULL;
1523                 return -ENOMEM;
1524         }
1525 #endif
1526         return 0;
1527 }
1528 EXPORT_SYMBOL_GPL(snmp_mib_init);
1529
1530 void snmp_mib_free(void __percpu *ptr[SNMP_ARRAY_SZ])
1531 {
1532         int i;
1533
1534         BUG_ON(ptr == NULL);
1535         for (i = 0; i < SNMP_ARRAY_SZ; i++) {
1536                 free_percpu(ptr[i]);
1537                 ptr[i] = NULL;
1538         }
1539 }
1540 EXPORT_SYMBOL_GPL(snmp_mib_free);
1541
1542 #ifdef CONFIG_IP_MULTICAST
1543 static const struct net_protocol igmp_protocol = {
1544         .handler =      igmp_rcv,
1545         .netns_ok =     1,
1546 };
1547 #endif
1548
1549 static const struct net_protocol tcp_protocol = {
1550         .early_demux    =       tcp_v4_early_demux,
1551         .handler        =       tcp_v4_rcv,
1552         .err_handler    =       tcp_v4_err,
1553         .no_policy      =       1,
1554         .netns_ok       =       1,
1555 };
1556
1557 static const struct net_offload tcp_offload = {
1558         .callbacks = {
1559                 .gso_send_check =       tcp_v4_gso_send_check,
1560                 .gso_segment    =       tcp_tso_segment,
1561                 .gro_receive    =       tcp4_gro_receive,
1562                 .gro_complete   =       tcp4_gro_complete,
1563         },
1564 };
1565
1566 static const struct net_protocol udp_protocol = {
1567         .handler =      udp_rcv,
1568         .err_handler =  udp_err,
1569         .no_policy =    1,
1570         .netns_ok =     1,
1571 };
1572
1573 static const struct net_offload udp_offload = {
1574         .callbacks = {
1575                 .gso_send_check = udp4_ufo_send_check,
1576                 .gso_segment = udp4_ufo_fragment,
1577         },
1578 };
1579
1580 static const struct net_protocol icmp_protocol = {
1581         .handler =      icmp_rcv,
1582         .err_handler =  icmp_err,
1583         .no_policy =    1,
1584         .netns_ok =     1,
1585 };
1586
1587 static __net_init int ipv4_mib_init_net(struct net *net)
1588 {
1589         if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1590                           sizeof(struct tcp_mib),
1591                           __alignof__(struct tcp_mib)) < 0)
1592                 goto err_tcp_mib;
1593         if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1594                           sizeof(struct ipstats_mib),
1595                           __alignof__(struct ipstats_mib)) < 0)
1596                 goto err_ip_mib;
1597         if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1598                           sizeof(struct linux_mib),
1599                           __alignof__(struct linux_mib)) < 0)
1600                 goto err_net_mib;
1601         if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1602                           sizeof(struct udp_mib),
1603                           __alignof__(struct udp_mib)) < 0)
1604                 goto err_udp_mib;
1605         if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1606                           sizeof(struct udp_mib),
1607                           __alignof__(struct udp_mib)) < 0)
1608                 goto err_udplite_mib;
1609         if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1610                           sizeof(struct icmp_mib),
1611                           __alignof__(struct icmp_mib)) < 0)
1612                 goto err_icmp_mib;
1613         net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1614                                               GFP_KERNEL);
1615         if (!net->mib.icmpmsg_statistics)
1616                 goto err_icmpmsg_mib;
1617
1618         tcp_mib_init(net);
1619         return 0;
1620
1621 err_icmpmsg_mib:
1622         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1623 err_icmp_mib:
1624         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1625 err_udplite_mib:
1626         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1627 err_udp_mib:
1628         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1629 err_net_mib:
1630         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1631 err_ip_mib:
1632         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1633 err_tcp_mib:
1634         return -ENOMEM;
1635 }
1636
1637 static __net_exit void ipv4_mib_exit_net(struct net *net)
1638 {
1639         kfree(net->mib.icmpmsg_statistics);
1640         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1641         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1642         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1643         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1644         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1645         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1646 }
1647
1648 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1649         .init = ipv4_mib_init_net,
1650         .exit = ipv4_mib_exit_net,
1651 };
1652
1653 static int __init init_ipv4_mibs(void)
1654 {
1655         return register_pernet_subsys(&ipv4_mib_ops);
1656 }
1657
1658 static int ipv4_proc_init(void);
1659
1660 /*
1661  *      IP protocol layer initialiser
1662  */
1663
1664 static struct packet_offload ip_packet_offload __read_mostly = {
1665         .type = cpu_to_be16(ETH_P_IP),
1666         .callbacks = {
1667                 .gso_send_check = inet_gso_send_check,
1668                 .gso_segment = inet_gso_segment,
1669                 .gro_receive = inet_gro_receive,
1670                 .gro_complete = inet_gro_complete,
1671         },
1672 };
1673
1674 static int __init ipv4_offload_init(void)
1675 {
1676         /*
1677          * Add offloads
1678          */
1679         if (inet_add_offload(&udp_offload, IPPROTO_UDP) < 0)
1680                 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1681         if (inet_add_offload(&tcp_offload, IPPROTO_TCP) < 0)
1682                 pr_crit("%s: Cannot add TCP protocol offlaod\n", __func__);
1683
1684         dev_add_offload(&ip_packet_offload);
1685         return 0;
1686 }
1687
1688 fs_initcall(ipv4_offload_init);
1689
1690 static struct packet_type ip_packet_type __read_mostly = {
1691         .type = cpu_to_be16(ETH_P_IP),
1692         .func = ip_rcv,
1693 };
1694
1695 static int __init inet_init(void)
1696 {
1697         struct inet_protosw *q;
1698         struct list_head *r;
1699         int rc = -EINVAL;
1700
1701         BUILD_BUG_ON(sizeof(struct inet_skb_parm) > FIELD_SIZEOF(struct sk_buff, cb));
1702
1703         sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL);
1704         if (!sysctl_local_reserved_ports)
1705                 goto out;
1706
1707         rc = proto_register(&tcp_prot, 1);
1708         if (rc)
1709                 goto out_free_reserved_ports;
1710
1711         rc = proto_register(&udp_prot, 1);
1712         if (rc)
1713                 goto out_unregister_tcp_proto;
1714
1715         rc = proto_register(&raw_prot, 1);
1716         if (rc)
1717                 goto out_unregister_udp_proto;
1718
1719         rc = proto_register(&ping_prot, 1);
1720         if (rc)
1721                 goto out_unregister_raw_proto;
1722
1723         /*
1724          *      Tell SOCKET that we are alive...
1725          */
1726
1727         (void)sock_register(&inet_family_ops);
1728
1729 #ifdef CONFIG_SYSCTL
1730         ip_static_sysctl_init();
1731 #endif
1732
1733         tcp_prot.sysctl_mem = init_net.ipv4.sysctl_tcp_mem;
1734
1735         /*
1736          *      Add all the base protocols.
1737          */
1738
1739         if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1740                 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1741         if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1742                 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1743         if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1744                 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1745 #ifdef CONFIG_IP_MULTICAST
1746         if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1747                 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1748 #endif
1749
1750         /* Register the socket-side information for inet_create. */
1751         for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1752                 INIT_LIST_HEAD(r);
1753
1754         for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1755                 inet_register_protosw(q);
1756
1757         /*
1758          *      Set the ARP module up
1759          */
1760
1761         arp_init();
1762
1763         /*
1764          *      Set the IP module up
1765          */
1766
1767         ip_init();
1768
1769         tcp_v4_init();
1770
1771         /* Setup TCP slab cache for open requests. */
1772         tcp_init();
1773
1774         /* Setup UDP memory threshold */
1775         udp_init();
1776
1777         /* Add UDP-Lite (RFC 3828) */
1778         udplite4_register();
1779
1780         ping_init();
1781
1782         /*
1783          *      Set the ICMP layer up
1784          */
1785
1786         if (icmp_init() < 0)
1787                 panic("Failed to create the ICMP control socket.\n");
1788
1789         /*
1790          *      Initialise the multicast router
1791          */
1792 #if defined(CONFIG_IP_MROUTE)
1793         if (ip_mr_init())
1794                 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1795 #endif
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 out_free_reserved_ports:
1819         kfree(sysctl_local_reserved_ports);
1820         goto out;
1821 }
1822
1823 fs_initcall(inet_init);
1824
1825 /* ------------------------------------------------------------------------ */
1826
1827 #ifdef CONFIG_PROC_FS
1828 static int __init ipv4_proc_init(void)
1829 {
1830         int rc = 0;
1831
1832         if (raw_proc_init())
1833                 goto out_raw;
1834         if (tcp4_proc_init())
1835                 goto out_tcp;
1836         if (udp4_proc_init())
1837                 goto out_udp;
1838         if (ping_proc_init())
1839                 goto out_ping;
1840         if (ip_misc_proc_init())
1841                 goto out_misc;
1842 out:
1843         return rc;
1844 out_misc:
1845         ping_proc_exit();
1846 out_ping:
1847         udp4_proc_exit();
1848 out_udp:
1849         tcp4_proc_exit();
1850 out_tcp:
1851         raw_proc_exit();
1852 out_raw:
1853         rc = -ENOMEM;
1854         goto out;
1855 }
1856
1857 #else /* CONFIG_PROC_FS */
1858 static int __init ipv4_proc_init(void)
1859 {
1860         return 0;
1861 }
1862 #endif /* CONFIG_PROC_FS */
1863
1864 MODULE_ALIAS_NETPROTO(PF_INET);
1865