net/ipv4/af_inet.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              PF_INET protocol family socket handler.
   7  *
   8  * Authors:     Ross Biro
   9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  10  *              Florian La Roche, <flla@stud.uni-sb.de>
  11  *              Alan Cox, <A.Cox@swansea.ac.uk>
  12  *
  13  * Changes (see also sock.c)
  14  *
  15  *              piggy,
  16  *              Karl Knutson    :       Socket protocol table
  17  *              A.N.Kuznetsov   :       Socket death error in accept().
  18  *              John Richardson :       Fix non blocking error in connect()
  19  *                                      so sockets that fail to connect
  20  *                                      don't return -EINPROGRESS.
  21  *              Alan Cox        :       Asynchronous I/O support
  22  *              Alan Cox        :       Keep correct socket pointer on sock
  23  *                                      structures
  24  *                                      when accept() ed
  25  *              Alan Cox        :       Semantics of SO_LINGER aren't state
  26  *                                      moved to close when you look carefully.
  27  *                                      With this fixed and the accept bug fixed
  28  *                                      some RPC stuff seems happier.
  29  *              Niibe Yutaka    :       4.4BSD style write async I/O
  30  *              Alan Cox,
  31  *              Tony Gale       :       Fixed reuse semantics.
  32  *              Alan Cox        :       bind() shouldn't abort existing but dead
  33  *                                      sockets. Stops FTP netin:.. I hope.
  34  *              Alan Cox        :       bind() works correctly for RAW sockets.
  35  *                                      Note that FreeBSD at least was broken
  36  *                                      in this respect so be careful with
  37  *                                      compatibility tests...
  38  *              Alan Cox        :       routing cache support
  39  *              Alan Cox        :       memzero the socket structure for
  40  *                                      compactness.
  41  *              Matt Day        :       nonblock connect error handler
  42  *              Alan Cox        :       Allow large numbers of pending sockets
  43  *                                      (eg for big web sites), but only if
  44  *                                      specifically application requested.
  45  *              Alan Cox        :       New buffering throughout IP. Used
  46  *                                      dumbly.
  47  *              Alan Cox        :       New buffering now used smartly.
  48  *              Alan Cox        :       BSD rather than common sense
  49  *                                      interpretation of listen.
  50  *              Germano Caronni :       Assorted small races.
  51  *              Alan Cox        :       sendmsg/recvmsg basic support.
  52  *              Alan Cox        :       Only sendmsg/recvmsg now supported.
  53  *              Alan Cox        :       Locked down bind (see security list).
  54  *              Alan Cox        :       Loosened bind a little.
  55  *              Mike McLagan    :       ADD/DEL DLCI Ioctls
  56  *      Willy Konynenberg       :       Transparent proxying support.
  57  *              David S. Miller :       New socket lookup architecture.
  58  *                                      Some other random speedups.
  59  *              Cyrus Durgin    :       Cleaned up file for kmod hacks.
  60  *              Andi Kleen      :       Fix inet_stream_connect TCP race.
  61  *
  62  *              This program is free software; you can redistribute it and/or
  63  *              modify it under the terms of the GNU General Public License
  64  *              as published by the Free Software Foundation; either version
  65  *              2 of the License, or (at your option) any later version.
  66  */
  67
  68 #define pr_fmt(fmt) "IPv4: " fmt
  69
  70 #include <linux/err.h>
  71 #include <linux/errno.h>
  72 #include <linux/types.h>
  73 #include <linux/socket.h>
  74 #include <linux/in.h>
  75 #include <linux/kernel.h>
  76 #include <linux/module.h>
  77 #include <linux/sched.h>
  78 #include <linux/timer.h>
  79 #include <linux/string.h>
  80 #include <linux/sockios.h>
  81 #include <linux/net.h>
  82 #include <linux/capability.h>
  83 #include <linux/fcntl.h>
  84 #include <linux/mm.h>
  85 #include <linux/interrupt.h>
  86 #include <linux/stat.h>
  87 #include <linux/init.h>
  88 #include <linux/poll.h>
  89 #include <linux/netfilter_ipv4.h>
  90 #include <linux/random.h>
  91 #include <linux/slab.h>
  92
  93 #include <asm/uaccess.h>
  94
  95 #include <linux/inet.h>
  96 #include <linux/igmp.h>
  97 #include <linux/inetdevice.h>
  98 #include <linux/netdevice.h>
  99 #include <net/checksum.h>
 100 #include <net/ip.h>
 101 #include <net/protocol.h>
 102 #include <net/arp.h>
 103 #include <net/route.h>
 104 #include <net/ip_fib.h>
 105 #include <net/inet_connection_sock.h>
 106 #include <net/tcp.h>
 107 #include <net/udp.h>
 108 #include <net/udplite.h>
 109 #include <net/ping.h>
 110 #include <linux/skbuff.h>
 111 #include <net/sock.h>
 112 #include <net/raw.h>
 113 #include <net/icmp.h>
 114 #include <net/inet_common.h>
 115 #include <net/xfrm.h>
 116 #include <net/net_namespace.h>
 117 #include <net/secure_seq.h>
 118 #ifdef CONFIG_IP_MROUTE
 119 #include <linux/mroute.h>
 120 #endif
 121
 122
 123 /* The inetsw table contains everything that inet_create needs to
 124  * build a new socket.
 125  */
 126 static struct list_head inetsw[SOCK_MAX];
 127 static DEFINE_SPINLOCK(inetsw_lock);
 128
 129 struct ipv4_config ipv4_config;
 130 EXPORT_SYMBOL(ipv4_config);
 131
 132 /* New destruction routine */
 133
 134 void inet_sock_destruct(struct sock *sk)
 135 {
 136         struct inet_sock *inet = inet_sk(sk);
 137
 138         __skb_queue_purge(&sk->sk_receive_queue);
 139         __skb_queue_purge(&sk->sk_error_queue);
 140
 141         sk_mem_reclaim(sk);
 142
 143         if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
 144                 pr_err("Attempt to release TCP socket in state %d %p\n",
 145                        sk->sk_state, sk);
 146                 return;
 147         }
 148         if (!sock_flag(sk, SOCK_DEAD)) {
 149                 pr_err("Attempt to release alive inet socket %p\n", sk);
 150                 return;
 151         }
 152
 153         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
 154         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
 155         WARN_ON(sk->sk_wmem_queued);
 156         WARN_ON(sk->sk_forward_alloc);
 157
 158         kfree(rcu_dereference_protected(inet->inet_opt, 1));
 159         dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
 160         dst_release(sk->sk_rx_dst);
 161         sk_refcnt_debug_dec(sk);
 162 }
 163 EXPORT_SYMBOL(inet_sock_destruct);
 164
 165 /*
 166  *      The routines beyond this point handle the behaviour of an AF_INET
 167  *      socket object. Mostly it punts to the subprotocols of IP to do
 168  *      the work.
 169  */
 170
 171 /*
 172  *      Automatically bind an unbound socket.
 173  */
 174
 175 static int inet_autobind(struct sock *sk)
 176 {
 177         struct inet_sock *inet;
 178         /* We may need to bind the socket. */
 179         lock_sock(sk);
 180         inet = inet_sk(sk);
 181         if (!inet->inet_num) {
 182                 if (sk->sk_prot->get_port(sk, 0)) {
 183                         release_sock(sk);
 184                         return -EAGAIN;
 185                 }
 186                 inet->inet_sport = htons(inet->inet_num);
 187         }
 188         release_sock(sk);
 189         return 0;
 190 }
 191
 192 /*
 193  *      Move a socket into listening state.
 194  */
 195 int inet_listen(struct socket *sock, int backlog)
 196 {
 197         struct sock *sk = sock->sk;
 198         unsigned char old_state;
 199         int err;
 200
 201         lock_sock(sk);
 202
 203         err = -EINVAL;
 204         if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
 205                 goto out;
 206
 207         old_state = sk->sk_state;
 208         if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
 209                 goto out;
 210
 211         /* Really, if the socket is already in listen state
 212          * we can only allow the backlog to be adjusted.
 213          */
 214         if (old_state != TCP_LISTEN) {
 215                 /* Check special setups for testing purpose to enable TFO w/o
 216                  * requiring TCP_FASTOPEN sockopt.
 217                  * Note that only TCP sockets (SOCK_STREAM) will reach here.
 218                  * Also fastopenq may already been allocated because this
 219                  * socket was in TCP_LISTEN state previously but was
 220                  * shutdown() (rather than close()).
 221                  */
 222                 if ((sysctl_tcp_fastopen & TFO_SERVER_ENABLE) != 0 &&
 223                     inet_csk(sk)->icsk_accept_queue.fastopenq == NULL) {
 224                         if ((sysctl_tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) != 0)
 225                                 err = fastopen_init_queue(sk, backlog);
 226                         else if ((sysctl_tcp_fastopen &
 227                                   TFO_SERVER_WO_SOCKOPT2) != 0)
 228                                 err = fastopen_init_queue(sk,
 229                                     ((uint)sysctl_tcp_fastopen) >> 16);
 230                         else
 231                                 err = 0;
 232                         if (err)
 233                                 goto out;
 234                 }
 235                 err = inet_csk_listen_start(sk, backlog);
 236                 if (err)
 237                         goto out;
 238         }
 239         sk->sk_max_ack_backlog = backlog;
 240         err = 0;
 241
 242 out:
 243         release_sock(sk);
 244         return err;
 245 }
 246 EXPORT_SYMBOL(inet_listen);
 247
 248 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                 net_secret_init();
 269         }
 270 }
 271 EXPORT_SYMBOL(build_ehash_secret);
 272
 273 /*
 274  *      Create an inet socket.
 275  */
 276
 277 static int inet_create(struct net *net, struct socket *sock, int protocol,
 278                        int kern)
 279 {
 280         struct sock *sk;
 281         struct inet_protosw *answer;
 282         struct inet_sock *inet;
 283         struct proto *answer_prot;
 284         unsigned char answer_flags;
 285         char answer_no_check;
 286         int try_loading_module = 0;
 287         int err;
 288
 289         if (unlikely(!inet_ehash_secret))
 290                 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
 291                         build_ehash_secret();
 292
 293         sock->state = SS_UNCONNECTED;
 294
 295         /* Look for the requested type/protocol pair. */
 296 lookup_protocol:
 297         err = -ESOCKTNOSUPPORT;
 298         rcu_read_lock();
 299         list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
 300
 301                 err = 0;
 302                 /* Check the non-wild match. */
 303                 if (protocol == answer->protocol) {
 304                         if (protocol != IPPROTO_IP)
 305                                 break;
 306                 } else {
 307                         /* Check for the two wild cases. */
 308                         if (IPPROTO_IP == protocol) {
 309                                 protocol = answer->protocol;
 310                                 break;
 311                         }
 312                         if (IPPROTO_IP == answer->protocol)
 313                                 break;
 314                 }
 315                 err = -EPROTONOSUPPORT;
 316         }
 317
 318         if (unlikely(err)) {
 319                 if (try_loading_module < 2) {
 320                         rcu_read_unlock();
 321                         /*
 322                          * Be more specific, e.g. net-pf-2-proto-132-type-1
 323                          * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
 324                          */
 325                         if (++try_loading_module == 1)
 326                                 request_module("net-pf-%d-proto-%d-type-%d",
 327                                                PF_INET, protocol, sock->type);
 328                         /*
 329                          * Fall back to generic, e.g. net-pf-2-proto-132
 330                          * (net-pf-PF_INET-proto-IPPROTO_SCTP)
 331                          */
 332                         else
 333                                 request_module("net-pf-%d-proto-%d",
 334                                                PF_INET, protocol);
 335                         goto lookup_protocol;
 336                 } else
 337                         goto out_rcu_unlock;
 338         }
 339
 340         err = -EPERM;
 341         if (sock->type == SOCK_RAW && !kern &&
 342             !ns_capable(net->user_ns, CAP_NET_RAW))
 343                 goto out_rcu_unlock;
 344
 345         sock->ops = answer->ops;
 346         answer_prot = answer->prot;
 347         answer_no_check = answer->no_check;
 348         answer_flags = answer->flags;
 349         rcu_read_unlock();
 350
 351         WARN_ON(answer_prot->slab == NULL);
 352
 353         err = -ENOBUFS;
 354         sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
 355         if (sk == NULL)
 356                 goto out;
 357
 358         err = 0;
 359         sk->sk_no_check = answer_no_check;
 360         if (INET_PROTOSW_REUSE & answer_flags)
 361                 sk->sk_reuse = SK_CAN_REUSE;
 362
 363         inet = inet_sk(sk);
 364         inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
 365
 366         inet->nodefrag = 0;
 367
 368         if (SOCK_RAW == sock->type) {
 369                 inet->inet_num = protocol;
 370                 if (IPPROTO_RAW == protocol)
 371                         inet->hdrincl = 1;
 372         }
 373
 374         if (ipv4_config.no_pmtu_disc)
 375                 inet->pmtudisc = IP_PMTUDISC_DONT;
 376         else
 377                 inet->pmtudisc = IP_PMTUDISC_WANT;
 378
 379         inet->inet_id = 0;
 380
 381         sock_init_data(sock, sk);
 382
 383         sk->sk_destruct    = inet_sock_destruct;
 384         sk->sk_protocol    = protocol;
 385         sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
 386
 387         inet->uc_ttl    = -1;
 388         inet->mc_loop   = 1;
 389         inet->mc_ttl    = 1;
 390         inet->mc_all    = 1;
 391         inet->mc_index  = 0;
 392         inet->mc_list   = NULL;
 393         inet->rcv_tos   = 0;
 394
 395         sk_refcnt_debug_inc(sk);
 396
 397         if (inet->inet_num) {
 398                 /* It assumes that any protocol which allows
 399                  * the user to assign a number at socket
 400                  * creation time automatically
 401                  * shares.
 402                  */
 403                 inet->inet_sport = htons(inet->inet_num);
 404                 /* Add to protocol hash chains. */
 405                 sk->sk_prot->hash(sk);
 406         }
 407
 408         if (sk->sk_prot->init) {
 409                 err = sk->sk_prot->init(sk);
 410                 if (err)
 411                         sk_common_release(sk);
 412         }
 413 out:
 414         return err;
 415 out_rcu_unlock:
 416         rcu_read_unlock();
 417         goto out;
 418 }
 419
 420
 421 /*
 422  *      The peer socket should always be NULL (or else). When we call this
 423  *      function we are destroying the object and from then on nobody
 424  *      should refer to it.
 425  */
 426 int inet_release(struct socket *sock)
 427 {
 428         struct sock *sk = sock->sk;
 429
 430         if (sk) {
 431                 long timeout;
 432
 433                 sock_rps_reset_flow(sk);
 434
 435                 /* Applications forget to leave groups before exiting */
 436                 ip_mc_drop_socket(sk);
 437
 438                 /* If linger is set, we don't return until the close
 439                  * is complete.  Otherwise we return immediately. The
 440                  * actually closing is done the same either way.
 441                  *
 442                  * If the close is due to the process exiting, we never
 443                  * linger..
 444                  */
 445                 timeout = 0;
 446                 if (sock_flag(sk, SOCK_LINGER) &&
 447                     !(current->flags & PF_EXITING))
 448                         timeout = sk->sk_lingertime;
 449                 sock->sk = NULL;
 450                 sk->sk_prot->close(sk, timeout);
 451         }
 452         return 0;
 453 }
 454 EXPORT_SYMBOL(inet_release);
 455
 456 /* It is off by default, see below. */
 457 int sysctl_ip_nonlocal_bind __read_mostly;
 458 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
 459
 460 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 461 {
 462         struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
 463         struct sock *sk = sock->sk;
 464         struct inet_sock *inet = inet_sk(sk);
 465         struct net *net = sock_net(sk);
 466         unsigned short snum;
 467         int chk_addr_ret;
 468         int err;
 469
 470         /* If the socket has its own bind function then use it. (RAW) */
 471         if (sk->sk_prot->bind) {
 472                 err = sk->sk_prot->bind(sk, uaddr, addr_len);
 473                 goto out;
 474         }
 475         err = -EINVAL;
 476         if (addr_len < sizeof(struct sockaddr_in))
 477                 goto out;
 478
 479         if (addr->sin_family != AF_INET) {
 480                 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
 481                  * only if s_addr is INADDR_ANY.
 482                  */
 483                 err = -EAFNOSUPPORT;
 484                 if (addr->sin_family != AF_UNSPEC ||
 485                     addr->sin_addr.s_addr != htonl(INADDR_ANY))
 486                         goto out;
 487         }
 488
 489         chk_addr_ret = inet_addr_type(net, addr->sin_addr.s_addr);
 490
 491         /* Not specified by any standard per-se, however it breaks too
 492          * many applications when removed.  It is unfortunate since
 493          * allowing applications to make a non-local bind solves
 494          * several problems with systems using dynamic addressing.
 495          * (ie. your servers still start up even if your ISDN link
 496          *  is temporarily down)
 497          */
 498         err = -EADDRNOTAVAIL;
 499         if (!sysctl_ip_nonlocal_bind &&
 500             !(inet->freebind || inet->transparent) &&
 501             addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
 502             chk_addr_ret != RTN_LOCAL &&
 503             chk_addr_ret != RTN_MULTICAST &&
 504             chk_addr_ret != RTN_BROADCAST)
 505                 goto out;
 506
 507         snum = ntohs(addr->sin_port);
 508         err = -EACCES;
 509         if (snum && snum < PROT_SOCK &&
 510             !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
 511                 goto out;
 512
 513         /*      We keep a pair of addresses. rcv_saddr is the one
 514          *      used by hash lookups, and saddr is used for transmit.
 515          *
 516          *      In the BSD API these are the same except where it
 517          *      would be illegal to use them (multicast/broadcast) in
 518          *      which case the sending device address is used.
 519          */
 520         lock_sock(sk);
 521
 522         /* Check these errors (active socket, double bind). */
 523         err = -EINVAL;
 524         if (sk->sk_state != TCP_CLOSE || inet->inet_num)
 525                 goto out_release_sock;
 526
 527         inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
 528         if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
 529                 inet->inet_saddr = 0;  /* Use device */
 530
 531         /* Make sure we are allowed to bind here. */
 532         if (sk->sk_prot->get_port(sk, snum)) {
 533                 inet->inet_saddr = inet->inet_rcv_saddr = 0;
 534                 err = -EADDRINUSE;
 535                 goto out_release_sock;
 536         }
 537
 538         if (inet->inet_rcv_saddr)
 539                 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
 540         if (snum)
 541                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
 542         inet->inet_sport = htons(inet->inet_num);
 543         inet->inet_daddr = 0;
 544         inet->inet_dport = 0;
 545         sk_dst_reset(sk);
 546         err = 0;
 547 out_release_sock:
 548         release_sock(sk);
 549 out:
 550         return err;
 551 }
 552 EXPORT_SYMBOL(inet_bind);
 553
 554 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
 555                        int addr_len, int flags)
 556 {
 557         struct sock *sk = sock->sk;
 558
 559         if (addr_len < sizeof(uaddr->sa_family))
 560                 return -EINVAL;
 561         if (uaddr->sa_family == AF_UNSPEC)
 562                 return sk->sk_prot->disconnect(sk, flags);
 563
 564         if (!inet_sk(sk)->inet_num && inet_autobind(sk))
 565                 return -EAGAIN;
 566         return sk->sk_prot->connect(sk, uaddr, addr_len);
 567 }
 568 EXPORT_SYMBOL(inet_dgram_connect);
 569
 570 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
 571 {
 572         DEFINE_WAIT(wait);
 573
 574         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
 575         sk->sk_write_pending += writebias;
 576
 577         /* Basic assumption: if someone sets sk->sk_err, he _must_
 578          * change state of the socket from TCP_SYN_*.
 579          * Connect() does not allow to get error notifications
 580          * without closing the socket.
 581          */
 582         while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 583                 release_sock(sk);
 584                 timeo = schedule_timeout(timeo);
 585                 lock_sock(sk);
 586                 if (signal_pending(current) || !timeo)
 587                         break;
 588                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
 589         }
 590         finish_wait(sk_sleep(sk), &wait);
 591         sk->sk_write_pending -= writebias;
 592         return timeo;
 593 }
 594
 595 /*
 596  *      Connect to a remote host. There is regrettably still a little
 597  *      TCP 'magic' in here.
 598  */
 599 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 600                           int addr_len, int flags)
 601 {
 602         struct sock *sk = sock->sk;
 603         int err;
 604         long timeo;
 605
 606         if (addr_len < sizeof(uaddr->sa_family))
 607                 return -EINVAL;
 608
 609         if (uaddr->sa_family == AF_UNSPEC) {
 610                 err = sk->sk_prot->disconnect(sk, flags);
 611                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 612                 goto out;
 613         }
 614
 615         switch (sock->state) {
 616         default:
 617                 err = -EINVAL;
 618                 goto out;
 619         case SS_CONNECTED:
 620                 err = -EISCONN;
 621                 goto out;
 622         case SS_CONNECTING:
 623                 err = -EALREADY;
 624                 /* Fall out of switch with err, set for this state */
 625                 break;
 626         case SS_UNCONNECTED:
 627                 err = -EISCONN;
 628                 if (sk->sk_state != TCP_CLOSE)
 629                         goto out;
 630
 631                 err = sk->sk_prot->connect(sk, uaddr, addr_len);
 632                 if (err < 0)
 633                         goto out;
 634
 635                 sock->state = SS_CONNECTING;
 636
 637                 /* Just entered SS_CONNECTING state; the only
 638                  * difference is that return value in non-blocking
 639                  * case is EINPROGRESS, rather than EALREADY.
 640                  */
 641                 err = -EINPROGRESS;
 642                 break;
 643         }
 644
 645         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
 646
 647         if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 648                 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
 649                                 tcp_sk(sk)->fastopen_req &&
 650                                 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
 651
 652                 /* Error code is set above */
 653                 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
 654                         goto out;
 655
 656                 err = sock_intr_errno(timeo);
 657                 if (signal_pending(current))
 658                         goto out;
 659         }
 660
 661         /* Connection was closed by RST, timeout, ICMP error
 662          * or another process disconnected us.
 663          */
 664         if (sk->sk_state == TCP_CLOSE)
 665                 goto sock_error;
 666
 667         /* sk->sk_err may be not zero now, if RECVERR was ordered by user
 668          * and error was received after socket entered established state.
 669          * Hence, it is handled normally after connect() return successfully.
 670          */
 671
 672         sock->state = SS_CONNECTED;
 673         err = 0;
 674 out:
 675         return err;
 676
 677 sock_error:
 678         err = sock_error(sk) ? : -ECONNABORTED;
 679         sock->state = SS_UNCONNECTED;
 680         if (sk->sk_prot->disconnect(sk, flags))
 681                 sock->state = SS_DISCONNECTING;
 682         goto out;
 683 }
 684 EXPORT_SYMBOL(__inet_stream_connect);
 685
 686 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 687                         int addr_len, int flags)
 688 {
 689         int err;
 690
 691         lock_sock(sock->sk);
 692         err = __inet_stream_connect(sock, uaddr, addr_len, flags);
 693         release_sock(sock->sk);
 694         return err;
 695 }
 696 EXPORT_SYMBOL(inet_stream_connect);
 697
 698 /*
 699  *      Accept a pending connection. The TCP layer now gives BSD semantics.
 700  */
 701
 702 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
 703 {
 704         struct sock *sk1 = sock->sk;
 705         int err = -EINVAL;
 706         struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
 707
 708         if (!sk2)
 709                 goto do_err;
 710
 711         lock_sock(sk2);
 712
 713         sock_rps_record_flow(sk2);
 714         WARN_ON(!((1 << sk2->sk_state) &
 715                   (TCPF_ESTABLISHED | TCPF_SYN_RECV |
 716                   TCPF_CLOSE_WAIT | TCPF_CLOSE)));
 717
 718         sock_graft(sk2, newsock);
 719
 720         newsock->state = SS_CONNECTED;
 721         err = 0;
 722         release_sock(sk2);
 723 do_err:
 724         return err;
 725 }
 726 EXPORT_SYMBOL(inet_accept);
 727
 728
 729 /*
 730  *      This does both peername and sockname.
 731  */
 732 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
 733                         int *uaddr_len, int peer)
 734 {
 735         struct sock *sk         = sock->sk;
 736         struct inet_sock *inet  = inet_sk(sk);
 737         DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
 738
 739         sin->sin_family = AF_INET;
 740         if (peer) {
 741                 if (!inet->inet_dport ||
 742                     (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
 743                      peer == 1))
 744                         return -ENOTCONN;
 745                 sin->sin_port = inet->inet_dport;
 746                 sin->sin_addr.s_addr = inet->inet_daddr;
 747         } else {
 748                 __be32 addr = inet->inet_rcv_saddr;
 749                 if (!addr)
 750                         addr = inet->inet_saddr;
 751                 sin->sin_port = inet->inet_sport;
 752                 sin->sin_addr.s_addr = addr;
 753         }
 754         memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 755         *uaddr_len = sizeof(*sin);
 756         return 0;
 757 }
 758 EXPORT_SYMBOL(inet_getname);
 759
 760 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
 761                  size_t size)
 762 {
 763         struct sock *sk = sock->sk;
 764
 765         sock_rps_record_flow(sk);
 766
 767         /* We may need to bind the socket. */
 768         if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
 769             inet_autobind(sk))
 770                 return -EAGAIN;
 771
 772         return sk->sk_prot->sendmsg(iocb, sk, msg, size);
 773 }
 774 EXPORT_SYMBOL(inet_sendmsg);
 775
 776 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
 777                       size_t size, int flags)
 778 {
 779         struct sock *sk = sock->sk;
 780
 781         sock_rps_record_flow(sk);
 782
 783         /* We may need to bind the socket. */
 784         if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
 785             inet_autobind(sk))
 786                 return -EAGAIN;
 787
 788         if (sk->sk_prot->sendpage)
 789                 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
 790         return sock_no_sendpage(sock, page, offset, size, flags);
 791 }
 792 EXPORT_SYMBOL(inet_sendpage);
 793
 794 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
 795                  size_t size, int flags)
 796 {
 797         struct sock *sk = sock->sk;
 798         int addr_len = 0;
 799         int err;
 800
 801         sock_rps_record_flow(sk);
 802
 803         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
 804                                    flags & ~MSG_DONTWAIT, &addr_len);
 805         if (err >= 0)
 806                 msg->msg_namelen = addr_len;
 807         return err;
 808 }
 809 EXPORT_SYMBOL(inet_recvmsg);
 810
 811 int inet_shutdown(struct socket *sock, int how)
 812 {
 813         struct sock *sk = sock->sk;
 814         int err = 0;
 815
 816         /* This should really check to make sure
 817          * the socket is a TCP socket. (WHY AC...)
 818          */
 819         how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
 820                        1->2 bit 2 snds.
 821                        2->3 */
 822         if ((how & ~SHUTDOWN_MASK) || !how)     /* MAXINT->0 */
 823                 return -EINVAL;
 824
 825         lock_sock(sk);
 826         if (sock->state == SS_CONNECTING) {
 827                 if ((1 << sk->sk_state) &
 828                     (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
 829                         sock->state = SS_DISCONNECTING;
 830                 else
 831                         sock->state = SS_CONNECTED;
 832         }
 833
 834         switch (sk->sk_state) {
 835         case TCP_CLOSE:
 836                 err = -ENOTCONN;
 837                 /* Hack to wake up other listeners, who can poll for
 838                    POLLHUP, even on eg. unconnected UDP sockets -- RR */
 839         default:
 840                 sk->sk_shutdown |= how;
 841                 if (sk->sk_prot->shutdown)
 842                         sk->sk_prot->shutdown(sk, how);
 843                 break;
 844
 845         /* Remaining two branches are temporary solution for missing
 846          * close() in multithreaded environment. It is _not_ a good idea,
 847          * but we have no choice until close() is repaired at VFS level.
 848          */
 849         case TCP_LISTEN:
 850                 if (!(how & RCV_SHUTDOWN))
 851                         break;
 852                 /* Fall through */
 853         case TCP_SYN_SENT:
 854                 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
 855                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 856                 break;
 857         }
 858
 859         /* Wake up anyone sleeping in poll. */
 860         sk->sk_state_change(sk);
 861         release_sock(sk);
 862         return err;
 863 }
 864 EXPORT_SYMBOL(inet_shutdown);
 865
 866 /*
 867  *      ioctl() calls you can issue on an INET socket. Most of these are
 868  *      device configuration and stuff and very rarely used. Some ioctls
 869  *      pass on to the socket itself.
 870  *
 871  *      NOTE: I like the idea of a module for the config stuff. ie ifconfig
 872  *      loads the devconfigure module does its configuring and unloads it.
 873  *      There's a good 20K of config code hanging around the kernel.
 874  */
 875
 876 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 877 {
 878         struct sock *sk = sock->sk;
 879         int err = 0;
 880         struct net *net = sock_net(sk);
 881
 882         switch (cmd) {
 883         case SIOCGSTAMP:
 884                 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
 885                 break;
 886         case SIOCGSTAMPNS:
 887                 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
 888                 break;
 889         case SIOCADDRT:
 890         case SIOCDELRT:
 891         case SIOCRTMSG:
 892                 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
 893                 break;
 894         case SIOCDARP:
 895         case SIOCGARP:
 896         case SIOCSARP:
 897                 err = arp_ioctl(net, cmd, (void __user *)arg);
 898                 break;
 899         case SIOCGIFADDR:
 900         case SIOCSIFADDR:
 901         case SIOCGIFBRDADDR:
 902         case SIOCSIFBRDADDR:
 903         case SIOCGIFNETMASK:
 904         case SIOCSIFNETMASK:
 905         case SIOCGIFDSTADDR:
 906         case SIOCSIFDSTADDR:
 907         case SIOCSIFPFLAGS:
 908         case SIOCGIFPFLAGS:
 909         case SIOCSIFFLAGS:
 910                 err = devinet_ioctl(net, cmd, (void __user *)arg);
 911                 break;
 912         default:
 913                 if (sk->sk_prot->ioctl)
 914                         err = sk->sk_prot->ioctl(sk, cmd, arg);
 915                 else
 916                         err = -ENOIOCTLCMD;
 917                 break;
 918         }
 919         return err;
 920 }
 921 EXPORT_SYMBOL(inet_ioctl);
 922
 923 #ifdef CONFIG_COMPAT
 924 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 925 {
 926         struct sock *sk = sock->sk;
 927         int err = -ENOIOCTLCMD;
 928
 929         if (sk->sk_prot->compat_ioctl)
 930                 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
 931
 932         return err;
 933 }
 934 #endif
 935
 936 const struct proto_ops inet_stream_ops = {
 937         .family            = PF_INET,
 938         .owner             = THIS_MODULE,
 939         .release           = inet_release,
 940         .bind              = inet_bind,
 941         .connect           = inet_stream_connect,
 942         .socketpair        = sock_no_socketpair,
 943         .accept            = inet_accept,
 944         .getname           = inet_getname,
 945         .poll              = tcp_poll,
 946         .ioctl             = inet_ioctl,
 947         .listen            = inet_listen,
 948         .shutdown          = inet_shutdown,
 949         .setsockopt        = sock_common_setsockopt,
 950         .getsockopt        = sock_common_getsockopt,
 951         .sendmsg           = inet_sendmsg,
 952         .recvmsg           = inet_recvmsg,
 953         .mmap              = sock_no_mmap,
 954         .sendpage          = inet_sendpage,
 955         .splice_read       = tcp_splice_read,
 956 #ifdef CONFIG_COMPAT
 957         .compat_setsockopt = compat_sock_common_setsockopt,
 958         .compat_getsockopt = compat_sock_common_getsockopt,
 959         .compat_ioctl      = inet_compat_ioctl,
 960 #endif
 961 };
 962 EXPORT_SYMBOL(inet_stream_ops);
 963
 964 const struct proto_ops inet_dgram_ops = {
 965         .family            = PF_INET,
 966         .owner             = THIS_MODULE,
 967         .release           = inet_release,
 968         .bind              = inet_bind,
 969         .connect           = inet_dgram_connect,
 970         .socketpair        = sock_no_socketpair,
 971         .accept            = sock_no_accept,
 972         .getname           = inet_getname,
 973         .poll              = udp_poll,
 974         .ioctl             = inet_ioctl,
 975         .listen            = sock_no_listen,
 976         .shutdown          = inet_shutdown,
 977         .setsockopt        = sock_common_setsockopt,
 978         .getsockopt        = sock_common_getsockopt,
 979         .sendmsg           = inet_sendmsg,
 980         .recvmsg           = inet_recvmsg,
 981         .mmap              = sock_no_mmap,
 982         .sendpage          = inet_sendpage,
 983 #ifdef CONFIG_COMPAT
 984         .compat_setsockopt = compat_sock_common_setsockopt,
 985         .compat_getsockopt = compat_sock_common_getsockopt,
 986         .compat_ioctl      = inet_compat_ioctl,
 987 #endif
 988 };
 989 EXPORT_SYMBOL(inet_dgram_ops);
 990
 991 /*
 992  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
 993  * udp_poll
 994  */
 995 static const struct proto_ops inet_sockraw_ops = {
 996         .family            = PF_INET,
 997         .owner             = THIS_MODULE,
 998         .release           = inet_release,
 999         .bind              = inet_bind,
1000         .connect           = inet_dgram_connect,
1001         .socketpair        = sock_no_socketpair,
1002         .accept            = sock_no_accept,
1003         .getname           = inet_getname,
1004         .poll              = datagram_poll,
1005         .ioctl             = inet_ioctl,
1006         .listen            = sock_no_listen,
1007         .shutdown          = inet_shutdown,
1008         .setsockopt        = sock_common_setsockopt,
1009         .getsockopt        = sock_common_getsockopt,
1010         .sendmsg           = inet_sendmsg,
1011         .recvmsg           = inet_recvmsg,
1012         .mmap              = sock_no_mmap,
1013         .sendpage          = inet_sendpage,
1014 #ifdef CONFIG_COMPAT
1015         .compat_setsockopt = compat_sock_common_setsockopt,
1016         .compat_getsockopt = compat_sock_common_getsockopt,
1017         .compat_ioctl      = inet_compat_ioctl,
1018 #endif
1019 };
1020
1021 static const struct net_proto_family inet_family_ops = {
1022         .family = PF_INET,
1023         .create = inet_create,
1024         .owner  = THIS_MODULE,
1025 };
1026
1027 /* Upon startup we insert all the elements in inetsw_array[] into
1028  * the linked list inetsw.
1029  */
1030 static struct inet_protosw inetsw_array[] =
1031 {
1032         {
1033                 .type =       SOCK_STREAM,
1034                 .protocol =   IPPROTO_TCP,
1035                 .prot =       &tcp_prot,
1036                 .ops =        &inet_stream_ops,
1037                 .no_check =   0,
1038                 .flags =      INET_PROTOSW_PERMANENT |
1039                               INET_PROTOSW_ICSK,
1040         },
1041
1042         {
1043                 .type =       SOCK_DGRAM,
1044                 .protocol =   IPPROTO_UDP,
1045                 .prot =       &udp_prot,
1046                 .ops =        &inet_dgram_ops,
1047                 .no_check =   UDP_CSUM_DEFAULT,
1048                 .flags =      INET_PROTOSW_PERMANENT,
1049        },
1050
1051        {
1052                 .type =       SOCK_DGRAM,
1053                 .protocol =   IPPROTO_ICMP,
1054                 .prot =       &ping_prot,
1055                 .ops =        &inet_dgram_ops,
1056                 .no_check =   UDP_CSUM_DEFAULT,
1057                 .flags =      INET_PROTOSW_REUSE,
1058        },
1059
1060        {
1061                .type =       SOCK_RAW,
1062                .protocol =   IPPROTO_IP,        /* wild card */
1063                .prot =       &raw_prot,
1064                .ops =        &inet_sockraw_ops,
1065                .no_check =   UDP_CSUM_DEFAULT,
1066                .flags =      INET_PROTOSW_REUSE,
1067        }
1068 };
1069
1070 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1071
1072 void inet_register_protosw(struct inet_protosw *p)
1073 {
1074         struct list_head *lh;
1075         struct inet_protosw *answer;
1076         int protocol = p->protocol;
1077         struct list_head *last_perm;
1078
1079         spin_lock_bh(&inetsw_lock);
1080
1081         if (p->type >= SOCK_MAX)
1082                 goto out_illegal;
1083
1084         /* If we are trying to override a permanent protocol, bail. */
1085         answer = NULL;
1086         last_perm = &inetsw[p->type];
1087         list_for_each(lh, &inetsw[p->type]) {
1088                 answer = list_entry(lh, struct inet_protosw, list);
1089
1090                 /* Check only the non-wild match. */
1091                 if (INET_PROTOSW_PERMANENT & answer->flags) {
1092                         if (protocol == answer->protocol)
1093                                 break;
1094                         last_perm = lh;
1095                 }
1096
1097                 answer = NULL;
1098         }
1099         if (answer)
1100                 goto out_permanent;
1101
1102         /* Add the new entry after the last permanent entry if any, so that
1103          * the new entry does not override a permanent entry when matched with
1104          * a wild-card protocol. But it is allowed to override any existing
1105          * non-permanent entry.  This means that when we remove this entry, the
1106          * system automatically returns to the old behavior.
1107          */
1108         list_add_rcu(&p->list, last_perm);
1109 out:
1110         spin_unlock_bh(&inetsw_lock);
1111
1112         return;
1113
1114 out_permanent:
1115         pr_err("Attempt to override permanent protocol %d\n", protocol);
1116         goto out;
1117
1118 out_illegal:
1119         pr_err("Ignoring attempt to register invalid socket type %d\n",
1120                p->type);
1121         goto out;
1122 }
1123 EXPORT_SYMBOL(inet_register_protosw);
1124
1125 void inet_unregister_protosw(struct inet_protosw *p)
1126 {
1127         if (INET_PROTOSW_PERMANENT & p->flags) {
1128                 pr_err("Attempt to unregister permanent protocol %d\n",
1129                        p->protocol);
1130         } else {
1131                 spin_lock_bh(&inetsw_lock);
1132                 list_del_rcu(&p->list);
1133                 spin_unlock_bh(&inetsw_lock);
1134
1135                 synchronize_net();
1136         }
1137 }
1138 EXPORT_SYMBOL(inet_unregister_protosw);
1139
1140 /*
1141  *      Shall we try to damage output packets if routing dev changes?
1142  */
1143
1144 int sysctl_ip_dynaddr __read_mostly;
1145
1146 static int inet_sk_reselect_saddr(struct sock *sk)
1147 {
1148         struct inet_sock *inet = inet_sk(sk);
1149         __be32 old_saddr = inet->inet_saddr;
1150         __be32 daddr = inet->inet_daddr;
1151         struct flowi4 *fl4;
1152         struct rtable *rt;
1153         __be32 new_saddr;
1154         struct ip_options_rcu *inet_opt;
1155
1156         inet_opt = rcu_dereference_protected(inet->inet_opt,
1157                                              sock_owned_by_user(sk));
1158         if (inet_opt && inet_opt->opt.srr)
1159                 daddr = inet_opt->opt.faddr;
1160
1161         /* Query new route. */
1162         fl4 = &inet->cork.fl.u.ip4;
1163         rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1164                               sk->sk_bound_dev_if, sk->sk_protocol,
1165                               inet->inet_sport, inet->inet_dport, sk, false);
1166         if (IS_ERR(rt))
1167                 return PTR_ERR(rt);
1168
1169         sk_setup_caps(sk, &rt->dst);
1170
1171         new_saddr = fl4->saddr;
1172
1173         if (new_saddr == old_saddr)
1174                 return 0;
1175
1176         if (sysctl_ip_dynaddr > 1) {
1177                 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1178                         __func__, &old_saddr, &new_saddr);
1179         }
1180
1181         inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1182
1183         /*
1184          * XXX The only one ugly spot where we need to
1185          * XXX really change the sockets identity after
1186          * XXX it has entered the hashes. -DaveM
1187          *
1188          * Besides that, it does not check for connection
1189          * uniqueness. Wait for troubles.
1190          */
1191         __sk_prot_rehash(sk);
1192         return 0;
1193 }
1194
1195 int inet_sk_rebuild_header(struct sock *sk)
1196 {
1197         struct inet_sock *inet = inet_sk(sk);
1198         struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1199         __be32 daddr;
1200         struct ip_options_rcu *inet_opt;
1201         struct flowi4 *fl4;
1202         int err;
1203
1204         /* Route is OK, nothing to do. */
1205         if (rt)
1206                 return 0;
1207
1208         /* Reroute. */
1209         rcu_read_lock();
1210         inet_opt = rcu_dereference(inet->inet_opt);
1211         daddr = inet->inet_daddr;
1212         if (inet_opt && inet_opt->opt.srr)
1213                 daddr = inet_opt->opt.faddr;
1214         rcu_read_unlock();
1215         fl4 = &inet->cork.fl.u.ip4;
1216         rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1217                                    inet->inet_dport, inet->inet_sport,
1218                                    sk->sk_protocol, RT_CONN_FLAGS(sk),
1219                                    sk->sk_bound_dev_if);
1220         if (!IS_ERR(rt)) {
1221                 err = 0;
1222                 sk_setup_caps(sk, &rt->dst);
1223         } else {
1224                 err = PTR_ERR(rt);
1225
1226                 /* Routing failed... */
1227                 sk->sk_route_caps = 0;
1228                 /*
1229                  * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1230                  * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1231                  */
1232                 if (!sysctl_ip_dynaddr ||
1233                     sk->sk_state != TCP_SYN_SENT ||
1234                     (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1235                     (err = inet_sk_reselect_saddr(sk)) != 0)
1236                         sk->sk_err_soft = -err;
1237         }
1238
1239         return err;
1240 }
1241 EXPORT_SYMBOL(inet_sk_rebuild_header);
1242
1243 static int inet_gso_send_check(struct sk_buff *skb)
1244 {
1245         const struct net_offload *ops;
1246         const struct iphdr *iph;
1247         int proto;
1248         int ihl;
1249         int err = -EINVAL;
1250
1251         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1252                 goto out;
1253
1254         iph = ip_hdr(skb);
1255         ihl = iph->ihl * 4;
1256         if (ihl < sizeof(*iph))
1257                 goto out;
1258
1259         if (unlikely(!pskb_may_pull(skb, ihl)))
1260                 goto out;
1261
1262         __skb_pull(skb, ihl);
1263         skb_reset_transport_header(skb);
1264         iph = ip_hdr(skb);
1265         proto = iph->protocol;
1266         err = -EPROTONOSUPPORT;
1267
1268         rcu_read_lock();
1269         ops = rcu_dereference(inet_offloads[proto]);
1270         if (likely(ops && ops->callbacks.gso_send_check))
1271                 err = ops->callbacks.gso_send_check(skb);
1272         rcu_read_unlock();
1273
1274 out:
1275         return err;
1276 }
1277
1278 static struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1279         netdev_features_t features)
1280 {
1281         struct sk_buff *segs = ERR_PTR(-EINVAL);
1282         const struct net_offload *ops;
1283         struct iphdr *iph;
1284         int proto;
1285         int ihl;
1286         int id;
1287         unsigned int offset = 0;
1288         bool tunnel;
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                        SKB_GSO_TCPV6 |
1297                        SKB_GSO_UDP_TUNNEL |
1298                        SKB_GSO_MPLS |
1299                        0)))
1300                 goto out;
1301
1302         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1303                 goto out;
1304
1305         iph = ip_hdr(skb);
1306         ihl = iph->ihl * 4;
1307         if (ihl < sizeof(*iph))
1308                 goto out;
1309
1310         if (unlikely(!pskb_may_pull(skb, ihl)))
1311                 goto out;
1312
1313         tunnel = !!skb->encapsulation;
1314
1315         __skb_pull(skb, ihl);
1316         skb_reset_transport_header(skb);
1317         iph = ip_hdr(skb);
1318         id = ntohs(iph->id);
1319         proto = iph->protocol;
1320         segs = ERR_PTR(-EPROTONOSUPPORT);
1321
1322         rcu_read_lock();
1323         ops = rcu_dereference(inet_offloads[proto]);
1324         if (likely(ops && ops->callbacks.gso_segment))
1325                 segs = ops->callbacks.gso_segment(skb, features);
1326         rcu_read_unlock();
1327
1328         if (IS_ERR_OR_NULL(segs))
1329                 goto out;
1330
1331         skb = segs;
1332         do {
1333                 iph = ip_hdr(skb);
1334                 if (!tunnel && proto == IPPROTO_UDP) {
1335                         iph->id = htons(id);
1336                         iph->frag_off = htons(offset >> 3);
1337                         if (skb->next != NULL)
1338                                 iph->frag_off |= htons(IP_MF);
1339                         offset += (skb->len - skb->mac_len - iph->ihl * 4);
1340                 } else  {
1341                         iph->id = htons(id++);
1342                 }
1343                 iph->tot_len = htons(skb->len - skb->mac_len);
1344                 iph->check = 0;
1345                 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1346         } while ((skb = skb->next));
1347
1348 out:
1349         return segs;
1350 }
1351
1352 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1353                                          struct sk_buff *skb)
1354 {
1355         const struct net_offload *ops;
1356         struct sk_buff **pp = NULL;
1357         struct sk_buff *p;
1358         const struct iphdr *iph;
1359         unsigned int hlen;
1360         unsigned int off;
1361         unsigned int id;
1362         int flush = 1;
1363         int proto;
1364
1365         off = skb_gro_offset(skb);
1366         hlen = off + sizeof(*iph);
1367         iph = skb_gro_header_fast(skb, off);
1368         if (skb_gro_header_hard(skb, hlen)) {
1369                 iph = skb_gro_header_slow(skb, hlen, off);
1370                 if (unlikely(!iph))
1371                         goto out;
1372         }
1373
1374         proto = iph->protocol;
1375
1376         rcu_read_lock();
1377         ops = rcu_dereference(inet_offloads[proto]);
1378         if (!ops || !ops->callbacks.gro_receive)
1379                 goto out_unlock;
1380
1381         if (*(u8 *)iph != 0x45)
1382                 goto out_unlock;
1383
1384         if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1385                 goto out_unlock;
1386
1387         id = ntohl(*(__be32 *)&iph->id);
1388         flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1389         id >>= 16;
1390
1391         for (p = *head; p; p = p->next) {
1392                 struct iphdr *iph2;
1393
1394                 if (!NAPI_GRO_CB(p)->same_flow)
1395                         continue;
1396
1397                 iph2 = ip_hdr(p);
1398
1399                 if ((iph->protocol ^ iph2->protocol) |
1400                     ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1401                     ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1402                         NAPI_GRO_CB(p)->same_flow = 0;
1403                         continue;
1404                 }
1405
1406                 /* All fields must match except length and checksum. */
1407                 NAPI_GRO_CB(p)->flush |=
1408                         (iph->ttl ^ iph2->ttl) |
1409                         (iph->tos ^ iph2->tos) |
1410                         (__force int)((iph->frag_off ^ iph2->frag_off) & htons(IP_DF)) |
1411                         ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1412
1413                 NAPI_GRO_CB(p)->flush |= flush;
1414         }
1415
1416         NAPI_GRO_CB(skb)->flush |= flush;
1417         skb_gro_pull(skb, sizeof(*iph));
1418         skb_set_transport_header(skb, skb_gro_offset(skb));
1419
1420         pp = ops->callbacks.gro_receive(head, skb);
1421
1422 out_unlock:
1423         rcu_read_unlock();
1424
1425 out:
1426         NAPI_GRO_CB(skb)->flush |= flush;
1427
1428         return pp;
1429 }
1430
1431 static int inet_gro_complete(struct sk_buff *skb)
1432 {
1433         __be16 newlen = htons(skb->len - skb_network_offset(skb));
1434         struct iphdr *iph = ip_hdr(skb);
1435         const struct net_offload *ops;
1436         int proto = iph->protocol;
1437         int err = -ENOSYS;
1438
1439         csum_replace2(&iph->check, iph->tot_len, newlen);
1440         iph->tot_len = newlen;
1441
1442         rcu_read_lock();
1443         ops = rcu_dereference(inet_offloads[proto]);
1444         if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1445                 goto out_unlock;
1446
1447         err = ops->callbacks.gro_complete(skb);
1448
1449 out_unlock:
1450         rcu_read_unlock();
1451
1452         return err;
1453 }
1454
1455 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1456                          unsigned short type, unsigned char protocol,
1457                          struct net *net)
1458 {
1459         struct socket *sock;
1460         int rc = sock_create_kern(family, type, protocol, &sock);
1461
1462         if (rc == 0) {
1463                 *sk = sock->sk;
1464                 (*sk)->sk_allocation = GFP_ATOMIC;
1465                 /*
1466                  * Unhash it so that IP input processing does not even see it,
1467                  * we do not wish this socket to see incoming packets.
1468                  */
1469                 (*sk)->sk_prot->unhash(*sk);
1470
1471                 sk_change_net(*sk, net);
1472         }
1473         return rc;
1474 }
1475 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1476
1477 unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1478 {
1479         unsigned long res = 0;
1480         int i, j;
1481
1482         for_each_possible_cpu(i) {
1483                 for (j = 0; j < SNMP_ARRAY_SZ; j++)
1484                         res += *(((unsigned long *) per_cpu_ptr(mib[j], i)) + offt);
1485         }
1486         return res;
1487 }
1488 EXPORT_SYMBOL_GPL(snmp_fold_field);
1489
1490 #if BITS_PER_LONG==32
1491
1492 u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset)
1493 {
1494         u64 res = 0;
1495         int cpu;
1496
1497         for_each_possible_cpu(cpu) {
1498                 void *bhptr;
1499                 struct u64_stats_sync *syncp;
1500                 u64 v;
1501                 unsigned int start;
1502
1503                 bhptr = per_cpu_ptr(mib[0], cpu);
1504                 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1505                 do {
1506                         start = u64_stats_fetch_begin_bh(syncp);
1507                         v = *(((u64 *) bhptr) + offt);
1508                 } while (u64_stats_fetch_retry_bh(syncp, start));
1509
1510                 res += v;
1511         }
1512         return res;
1513 }
1514 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1515 #endif
1516
1517 int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align)
1518 {
1519         BUG_ON(ptr == NULL);
1520         ptr[0] = __alloc_percpu(mibsize, align);
1521         if (!ptr[0])
1522                 return -ENOMEM;
1523 #if SNMP_ARRAY_SZ == 2
1524         ptr[1] = __alloc_percpu(mibsize, align);
1525         if (!ptr[1]) {
1526                 free_percpu(ptr[0]);
1527                 ptr[0] = NULL;
1528                 return -ENOMEM;
1529         }
1530 #endif
1531         return 0;
1532 }
1533 EXPORT_SYMBOL_GPL(snmp_mib_init);
1534
1535 void snmp_mib_free(void __percpu *ptr[SNMP_ARRAY_SZ])
1536 {
1537         int i;
1538
1539         BUG_ON(ptr == NULL);
1540         for (i = 0; i < SNMP_ARRAY_SZ; i++) {
1541                 free_percpu(ptr[i]);
1542                 ptr[i] = NULL;
1543         }
1544 }
1545 EXPORT_SYMBOL_GPL(snmp_mib_free);
1546
1547 #ifdef CONFIG_IP_MULTICAST
1548 static const struct net_protocol igmp_protocol = {
1549         .handler =      igmp_rcv,
1550         .netns_ok =     1,
1551 };
1552 #endif
1553
1554 static const struct net_protocol tcp_protocol = {
1555         .early_demux    =       tcp_v4_early_demux,
1556         .handler        =       tcp_v4_rcv,
1557         .err_handler    =       tcp_v4_err,
1558         .no_policy      =       1,
1559         .netns_ok       =       1,
1560 };
1561
1562 static const struct net_protocol udp_protocol = {
1563         .handler =      udp_rcv,
1564         .err_handler =  udp_err,
1565         .no_policy =    1,
1566         .netns_ok =     1,
1567 };
1568
1569 static const struct net_protocol icmp_protocol = {
1570         .handler =      icmp_rcv,
1571         .err_handler =  icmp_err,
1572         .no_policy =    1,
1573         .netns_ok =     1,
1574 };
1575
1576 static __net_init int ipv4_mib_init_net(struct net *net)
1577 {
1578         if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1579                           sizeof(struct tcp_mib),
1580                           __alignof__(struct tcp_mib)) < 0)
1581                 goto err_tcp_mib;
1582         if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1583                           sizeof(struct ipstats_mib),
1584                           __alignof__(struct ipstats_mib)) < 0)
1585                 goto err_ip_mib;
1586         if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1587                           sizeof(struct linux_mib),
1588                           __alignof__(struct linux_mib)) < 0)
1589                 goto err_net_mib;
1590         if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1591                           sizeof(struct udp_mib),
1592                           __alignof__(struct udp_mib)) < 0)
1593                 goto err_udp_mib;
1594         if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1595                           sizeof(struct udp_mib),
1596                           __alignof__(struct udp_mib)) < 0)
1597                 goto err_udplite_mib;
1598         if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1599                           sizeof(struct icmp_mib),
1600                           __alignof__(struct icmp_mib)) < 0)
1601                 goto err_icmp_mib;
1602         net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1603                                               GFP_KERNEL);
1604         if (!net->mib.icmpmsg_statistics)
1605                 goto err_icmpmsg_mib;
1606
1607         tcp_mib_init(net);
1608         return 0;
1609
1610 err_icmpmsg_mib:
1611         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1612 err_icmp_mib:
1613         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1614 err_udplite_mib:
1615         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1616 err_udp_mib:
1617         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1618 err_net_mib:
1619         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1620 err_ip_mib:
1621         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1622 err_tcp_mib:
1623         return -ENOMEM;
1624 }
1625
1626 static __net_exit void ipv4_mib_exit_net(struct net *net)
1627 {
1628         kfree(net->mib.icmpmsg_statistics);
1629         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1630         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1631         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1632         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1633         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1634         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1635 }
1636
1637 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1638         .init = ipv4_mib_init_net,
1639         .exit = ipv4_mib_exit_net,
1640 };
1641
1642 static int __init init_ipv4_mibs(void)
1643 {
1644         return register_pernet_subsys(&ipv4_mib_ops);
1645 }
1646
1647 static int ipv4_proc_init(void);
1648
1649 /*
1650  *      IP protocol layer initialiser
1651  */
1652
1653 static struct packet_offload ip_packet_offload __read_mostly = {
1654         .type = cpu_to_be16(ETH_P_IP),
1655         .callbacks = {
1656                 .gso_send_check = inet_gso_send_check,
1657                 .gso_segment = inet_gso_segment,
1658                 .gro_receive = inet_gro_receive,
1659                 .gro_complete = inet_gro_complete,
1660         },
1661 };
1662
1663 static int __init ipv4_offload_init(void)
1664 {
1665         /*
1666          * Add offloads
1667          */
1668         if (udpv4_offload_init() < 0)
1669                 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1670         if (tcpv4_offload_init() < 0)
1671                 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1672
1673         dev_add_offload(&ip_packet_offload);
1674         return 0;
1675 }
1676
1677 fs_initcall(ipv4_offload_init);
1678
1679 static struct packet_type ip_packet_type __read_mostly = {
1680         .type = cpu_to_be16(ETH_P_IP),
1681         .func = ip_rcv,
1682 };
1683
1684 static int __init inet_init(void)
1685 {
1686         struct inet_protosw *q;
1687         struct list_head *r;
1688         int rc = -EINVAL;
1689
1690         BUILD_BUG_ON(sizeof(struct inet_skb_parm) > FIELD_SIZEOF(struct sk_buff, cb));
1691
1692         sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL);
1693         if (!sysctl_local_reserved_ports)
1694                 goto out;
1695
1696         rc = proto_register(&tcp_prot, 1);
1697         if (rc)
1698                 goto out_free_reserved_ports;
1699
1700         rc = proto_register(&udp_prot, 1);
1701         if (rc)
1702                 goto out_unregister_tcp_proto;
1703
1704         rc = proto_register(&raw_prot, 1);
1705         if (rc)
1706                 goto out_unregister_udp_proto;
1707
1708         rc = proto_register(&ping_prot, 1);
1709         if (rc)
1710                 goto out_unregister_raw_proto;
1711
1712         /*
1713          *      Tell SOCKET that we are alive...
1714          */
1715
1716         (void)sock_register(&inet_family_ops);
1717
1718 #ifdef CONFIG_SYSCTL
1719         ip_static_sysctl_init();
1720 #endif
1721
1722         tcp_prot.sysctl_mem = init_net.ipv4.sysctl_tcp_mem;
1723
1724         /*
1725          *      Add all the base protocols.
1726          */
1727
1728         if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1729                 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1730         if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1731                 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1732         if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1733                 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1734 #ifdef CONFIG_IP_MULTICAST
1735         if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1736                 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1737 #endif
1738
1739         /* Register the socket-side information for inet_create. */
1740         for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1741                 INIT_LIST_HEAD(r);
1742
1743         for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1744                 inet_register_protosw(q);
1745
1746         /*
1747          *      Set the ARP module up
1748          */
1749
1750         arp_init();
1751
1752         /*
1753          *      Set the IP module up
1754          */
1755
1756         ip_init();
1757
1758         tcp_v4_init();
1759
1760         /* Setup TCP slab cache for open requests. */
1761         tcp_init();
1762
1763         /* Setup UDP memory threshold */
1764         udp_init();
1765
1766         /* Add UDP-Lite (RFC 3828) */
1767         udplite4_register();
1768
1769         ping_init();
1770
1771         /*
1772          *      Set the ICMP layer up
1773          */
1774
1775         if (icmp_init() < 0)
1776                 panic("Failed to create the ICMP control socket.\n");
1777
1778         /*
1779          *      Initialise the multicast router
1780          */
1781 #if defined(CONFIG_IP_MROUTE)
1782         if (ip_mr_init())
1783                 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1784 #endif
1785         /*
1786          *      Initialise per-cpu ipv4 mibs
1787          */
1788
1789         if (init_ipv4_mibs())
1790                 pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1791
1792         ipv4_proc_init();
1793
1794         ipfrag_init();
1795
1796         dev_add_pack(&ip_packet_type);
1797
1798         rc = 0;
1799 out:
1800         return rc;
1801 out_unregister_raw_proto:
1802         proto_unregister(&raw_prot);
1803 out_unregister_udp_proto:
1804         proto_unregister(&udp_prot);
1805 out_unregister_tcp_proto:
1806         proto_unregister(&tcp_prot);
1807 out_free_reserved_ports:
1808         kfree(sysctl_local_reserved_ports);
1809         goto out;
1810 }
1811
1812 fs_initcall(inet_init);
1813
1814 /* ------------------------------------------------------------------------ */
1815
1816 #ifdef CONFIG_PROC_FS
1817 static int __init ipv4_proc_init(void)
1818 {
1819         int rc = 0;
1820
1821         if (raw_proc_init())
1822                 goto out_raw;
1823         if (tcp4_proc_init())
1824                 goto out_tcp;
1825         if (udp4_proc_init())
1826                 goto out_udp;
1827         if (ping_proc_init())
1828                 goto out_ping;
1829         if (ip_misc_proc_init())
1830                 goto out_misc;
1831 out:
1832         return rc;
1833 out_misc:
1834         ping_proc_exit();
1835 out_ping:
1836         udp4_proc_exit();
1837 out_udp:
1838         tcp4_proc_exit();
1839 out_tcp:
1840         raw_proc_exit();
1841 out_raw:
1842         rc = -ENOMEM;
1843         goto out;
1844 }
1845
1846 #else /* CONFIG_PROC_FS */
1847 static int __init ipv4_proc_init(void)
1848 {
1849         return 0;
1850 }
1851 #endif /* CONFIG_PROC_FS */
1852
1853 MODULE_ALIAS_NETPROTO(PF_INET);
1854