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         if (sk->sk_type == SOCK_STREAM) {
 153                 struct fastopen_queue *fastopenq =
 154                         inet_csk(sk)->icsk_accept_queue.fastopenq;
 155                 kfree(fastopenq);
 156         }
 157
 158         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
 159         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
 160         WARN_ON(sk->sk_wmem_queued);
 161         WARN_ON(sk->sk_forward_alloc);
 162
 163         kfree(rcu_dereference_protected(inet->inet_opt, 1));
 164         dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
 165         dst_release(sk->sk_rx_dst);
 166         sk_refcnt_debug_dec(sk);
 167 }
 168 EXPORT_SYMBOL(inet_sock_destruct);
 169
 170 /*
 171  *      The routines beyond this point handle the behaviour of an AF_INET
 172  *      socket object. Mostly it punts to the subprotocols of IP to do
 173  *      the work.
 174  */
 175
 176 /*
 177  *      Automatically bind an unbound socket.
 178  */
 179
 180 static int inet_autobind(struct sock *sk)
 181 {
 182         struct inet_sock *inet;
 183         /* We may need to bind the socket. */
 184         lock_sock(sk);
 185         inet = inet_sk(sk);
 186         if (!inet->inet_num) {
 187                 if (sk->sk_prot->get_port(sk, 0)) {
 188                         release_sock(sk);
 189                         return -EAGAIN;
 190                 }
 191                 inet->inet_sport = htons(inet->inet_num);
 192         }
 193         release_sock(sk);
 194         return 0;
 195 }
 196
 197 /*
 198  *      Move a socket into listening state.
 199  */
 200 int inet_listen(struct socket *sock, int backlog)
 201 {
 202         struct sock *sk = sock->sk;
 203         unsigned char old_state;
 204         int err;
 205
 206         lock_sock(sk);
 207
 208         err = -EINVAL;
 209         if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
 210                 goto out;
 211
 212         old_state = sk->sk_state;
 213         if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
 214                 goto out;
 215
 216         /* Really, if the socket is already in listen state
 217          * we can only allow the backlog to be adjusted.
 218          */
 219         if (old_state != TCP_LISTEN) {
 220                 /* Check special setups for testing purpose to enable TFO w/o
 221                  * requiring TCP_FASTOPEN sockopt.
 222                  * Note that only TCP sockets (SOCK_STREAM) will reach here.
 223                  * Also fastopenq may already been allocated because this
 224                  * socket was in TCP_LISTEN state previously but was
 225                  * shutdown() (rather than close()).
 226                  */
 227                 if ((sysctl_tcp_fastopen & TFO_SERVER_ENABLE) != 0 &&
 228                     inet_csk(sk)->icsk_accept_queue.fastopenq == NULL) {
 229                         if ((sysctl_tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) != 0)
 230                                 err = fastopen_init_queue(sk, backlog);
 231                         else if ((sysctl_tcp_fastopen &
 232                                   TFO_SERVER_WO_SOCKOPT2) != 0)
 233                                 err = fastopen_init_queue(sk,
 234                                     ((uint)sysctl_tcp_fastopen) >> 16);
 235                         else
 236                                 err = 0;
 237                         if (err)
 238                                 goto out;
 239                 }
 240                 err = inet_csk_listen_start(sk, backlog);
 241                 if (err)
 242                         goto out;
 243         }
 244         sk->sk_max_ack_backlog = backlog;
 245         err = 0;
 246
 247 out:
 248         release_sock(sk);
 249         return err;
 250 }
 251 EXPORT_SYMBOL(inet_listen);
 252
 253 u32 inet_ehash_secret __read_mostly;
 254 EXPORT_SYMBOL(inet_ehash_secret);
 255
 256 /*
 257  * inet_ehash_secret must be set exactly once
 258  */
 259 void build_ehash_secret(void)
 260 {
 261         u32 rnd;
 262
 263         do {
 264                 get_random_bytes(&rnd, sizeof(rnd));
 265         } while (rnd == 0);
 266
 267         cmpxchg(&inet_ehash_secret, 0, rnd);
 268 }
 269 EXPORT_SYMBOL(build_ehash_secret);
 270
 271 static inline int inet_netns_ok(struct net *net, __u8 protocol)
 272 {
 273         const struct net_protocol *ipprot;
 274
 275         if (net_eq(net, &init_net))
 276                 return 1;
 277
 278         ipprot = rcu_dereference(inet_protos[protocol]);
 279         if (ipprot == NULL) {
 280                 /* raw IP is OK */
 281                 return 1;
 282         }
 283         return ipprot->netns_ok;
 284 }
 285
 286 /*
 287  *      Create an inet socket.
 288  */
 289
 290 static int inet_create(struct net *net, struct socket *sock, int protocol,
 291                        int kern)
 292 {
 293         struct sock *sk;
 294         struct inet_protosw *answer;
 295         struct inet_sock *inet;
 296         struct proto *answer_prot;
 297         unsigned char answer_flags;
 298         char answer_no_check;
 299         int try_loading_module = 0;
 300         int err;
 301
 302         if (unlikely(!inet_ehash_secret))
 303                 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
 304                         build_ehash_secret();
 305
 306         sock->state = SS_UNCONNECTED;
 307
 308         /* Look for the requested type/protocol pair. */
 309 lookup_protocol:
 310         err = -ESOCKTNOSUPPORT;
 311         rcu_read_lock();
 312         list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
 313
 314                 err = 0;
 315                 /* Check the non-wild match. */
 316                 if (protocol == answer->protocol) {
 317                         if (protocol != IPPROTO_IP)
 318                                 break;
 319                 } else {
 320                         /* Check for the two wild cases. */
 321                         if (IPPROTO_IP == protocol) {
 322                                 protocol = answer->protocol;
 323                                 break;
 324                         }
 325                         if (IPPROTO_IP == answer->protocol)
 326                                 break;
 327                 }
 328                 err = -EPROTONOSUPPORT;
 329         }
 330
 331         if (unlikely(err)) {
 332                 if (try_loading_module < 2) {
 333                         rcu_read_unlock();
 334                         /*
 335                          * Be more specific, e.g. net-pf-2-proto-132-type-1
 336                          * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
 337                          */
 338                         if (++try_loading_module == 1)
 339                                 request_module("net-pf-%d-proto-%d-type-%d",
 340                                                PF_INET, protocol, sock->type);
 341                         /*
 342                          * Fall back to generic, e.g. net-pf-2-proto-132
 343                          * (net-pf-PF_INET-proto-IPPROTO_SCTP)
 344                          */
 345                         else
 346                                 request_module("net-pf-%d-proto-%d",
 347                                                PF_INET, protocol);
 348                         goto lookup_protocol;
 349                 } else
 350                         goto out_rcu_unlock;
 351         }
 352
 353         err = -EPERM;
 354         if (sock->type == SOCK_RAW && !kern && !capable(CAP_NET_RAW))
 355                 goto out_rcu_unlock;
 356
 357         err = -EAFNOSUPPORT;
 358         if (!inet_netns_ok(net, protocol))
 359                 goto out_rcu_unlock;
 360
 361         sock->ops = answer->ops;
 362         answer_prot = answer->prot;
 363         answer_no_check = answer->no_check;
 364         answer_flags = answer->flags;
 365         rcu_read_unlock();
 366
 367         WARN_ON(answer_prot->slab == NULL);
 368
 369         err = -ENOBUFS;
 370         sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
 371         if (sk == NULL)
 372                 goto out;
 373
 374         err = 0;
 375         sk->sk_no_check = answer_no_check;
 376         if (INET_PROTOSW_REUSE & answer_flags)
 377                 sk->sk_reuse = SK_CAN_REUSE;
 378
 379         inet = inet_sk(sk);
 380         inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
 381
 382         inet->nodefrag = 0;
 383
 384         if (SOCK_RAW == sock->type) {
 385                 inet->inet_num = protocol;
 386                 if (IPPROTO_RAW == protocol)
 387                         inet->hdrincl = 1;
 388         }
 389
 390         if (ipv4_config.no_pmtu_disc)
 391                 inet->pmtudisc = IP_PMTUDISC_DONT;
 392         else
 393                 inet->pmtudisc = IP_PMTUDISC_WANT;
 394
 395         inet->inet_id = 0;
 396
 397         sock_init_data(sock, sk);
 398
 399         sk->sk_destruct    = inet_sock_destruct;
 400         sk->sk_protocol    = protocol;
 401         sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
 402
 403         inet->uc_ttl    = -1;
 404         inet->mc_loop   = 1;
 405         inet->mc_ttl    = 1;
 406         inet->mc_all    = 1;
 407         inet->mc_index  = 0;
 408         inet->mc_list   = NULL;
 409         inet->rcv_tos   = 0;
 410
 411         sk_refcnt_debug_inc(sk);
 412
 413         if (inet->inet_num) {
 414                 /* It assumes that any protocol which allows
 415                  * the user to assign a number at socket
 416                  * creation time automatically
 417                  * shares.
 418                  */
 419                 inet->inet_sport = htons(inet->inet_num);
 420                 /* Add to protocol hash chains. */
 421                 sk->sk_prot->hash(sk);
 422         }
 423
 424         if (sk->sk_prot->init) {
 425                 err = sk->sk_prot->init(sk);
 426                 if (err)
 427                         sk_common_release(sk);
 428         }
 429 out:
 430         return err;
 431 out_rcu_unlock:
 432         rcu_read_unlock();
 433         goto out;
 434 }
 435
 436
 437 /*
 438  *      The peer socket should always be NULL (or else). When we call this
 439  *      function we are destroying the object and from then on nobody
 440  *      should refer to it.
 441  */
 442 int inet_release(struct socket *sock)
 443 {
 444         struct sock *sk = sock->sk;
 445
 446         if (sk) {
 447                 long timeout;
 448
 449                 sock_rps_reset_flow(sk);
 450
 451                 /* Applications forget to leave groups before exiting */
 452                 ip_mc_drop_socket(sk);
 453
 454                 /* If linger is set, we don't return until the close
 455                  * is complete.  Otherwise we return immediately. The
 456                  * actually closing is done the same either way.
 457                  *
 458                  * If the close is due to the process exiting, we never
 459                  * linger..
 460                  */
 461                 timeout = 0;
 462                 if (sock_flag(sk, SOCK_LINGER) &&
 463                     !(current->flags & PF_EXITING))
 464                         timeout = sk->sk_lingertime;
 465                 sock->sk = NULL;
 466                 sk->sk_prot->close(sk, timeout);
 467         }
 468         return 0;
 469 }
 470 EXPORT_SYMBOL(inet_release);
 471
 472 /* It is off by default, see below. */
 473 int sysctl_ip_nonlocal_bind __read_mostly;
 474 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
 475
 476 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 477 {
 478         struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
 479         struct sock *sk = sock->sk;
 480         struct inet_sock *inet = inet_sk(sk);
 481         unsigned short snum;
 482         int chk_addr_ret;
 483         int err;
 484
 485         /* If the socket has its own bind function then use it. (RAW) */
 486         if (sk->sk_prot->bind) {
 487                 err = sk->sk_prot->bind(sk, uaddr, addr_len);
 488                 goto out;
 489         }
 490         err = -EINVAL;
 491         if (addr_len < sizeof(struct sockaddr_in))
 492                 goto out;
 493
 494         if (addr->sin_family != AF_INET) {
 495                 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
 496                  * only if s_addr is INADDR_ANY.
 497                  */
 498                 err = -EAFNOSUPPORT;
 499                 if (addr->sin_family != AF_UNSPEC ||
 500                     addr->sin_addr.s_addr != htonl(INADDR_ANY))
 501                         goto out;
 502         }
 503
 504         chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr);
 505
 506         /* Not specified by any standard per-se, however it breaks too
 507          * many applications when removed.  It is unfortunate since
 508          * allowing applications to make a non-local bind solves
 509          * several problems with systems using dynamic addressing.
 510          * (ie. your servers still start up even if your ISDN link
 511          *  is temporarily down)
 512          */
 513         err = -EADDRNOTAVAIL;
 514         if (!sysctl_ip_nonlocal_bind &&
 515             !(inet->freebind || inet->transparent) &&
 516             addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
 517             chk_addr_ret != RTN_LOCAL &&
 518             chk_addr_ret != RTN_MULTICAST &&
 519             chk_addr_ret != RTN_BROADCAST)
 520                 goto out;
 521
 522         snum = ntohs(addr->sin_port);
 523         err = -EACCES;
 524         if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
 525                 goto out;
 526
 527         /*      We keep a pair of addresses. rcv_saddr is the one
 528          *      used by hash lookups, and saddr is used for transmit.
 529          *
 530          *      In the BSD API these are the same except where it
 531          *      would be illegal to use them (multicast/broadcast) in
 532          *      which case the sending device address is used.
 533          */
 534         lock_sock(sk);
 535
 536         /* Check these errors (active socket, double bind). */
 537         err = -EINVAL;
 538         if (sk->sk_state != TCP_CLOSE || inet->inet_num)
 539                 goto out_release_sock;
 540
 541         inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
 542         if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
 543                 inet->inet_saddr = 0;  /* Use device */
 544
 545         /* Make sure we are allowed to bind here. */
 546         if (sk->sk_prot->get_port(sk, snum)) {
 547                 inet->inet_saddr = inet->inet_rcv_saddr = 0;
 548                 err = -EADDRINUSE;
 549                 goto out_release_sock;
 550         }
 551
 552         if (inet->inet_rcv_saddr)
 553                 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
 554         if (snum)
 555                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
 556         inet->inet_sport = htons(inet->inet_num);
 557         inet->inet_daddr = 0;
 558         inet->inet_dport = 0;
 559         sk_dst_reset(sk);
 560         err = 0;
 561 out_release_sock:
 562         release_sock(sk);
 563 out:
 564         return err;
 565 }
 566 EXPORT_SYMBOL(inet_bind);
 567
 568 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
 569                        int addr_len, int flags)
 570 {
 571         struct sock *sk = sock->sk;
 572
 573         if (addr_len < sizeof(uaddr->sa_family))
 574                 return -EINVAL;
 575         if (uaddr->sa_family == AF_UNSPEC)
 576                 return sk->sk_prot->disconnect(sk, flags);
 577
 578         if (!inet_sk(sk)->inet_num && inet_autobind(sk))
 579                 return -EAGAIN;
 580         return sk->sk_prot->connect(sk, uaddr, addr_len);
 581 }
 582 EXPORT_SYMBOL(inet_dgram_connect);
 583
 584 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
 585 {
 586         DEFINE_WAIT(wait);
 587
 588         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
 589         sk->sk_write_pending += writebias;
 590
 591         /* Basic assumption: if someone sets sk->sk_err, he _must_
 592          * change state of the socket from TCP_SYN_*.
 593          * Connect() does not allow to get error notifications
 594          * without closing the socket.
 595          */
 596         while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 597                 release_sock(sk);
 598                 timeo = schedule_timeout(timeo);
 599                 lock_sock(sk);
 600                 if (signal_pending(current) || !timeo)
 601                         break;
 602                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
 603         }
 604         finish_wait(sk_sleep(sk), &wait);
 605         sk->sk_write_pending -= writebias;
 606         return timeo;
 607 }
 608
 609 /*
 610  *      Connect to a remote host. There is regrettably still a little
 611  *      TCP 'magic' in here.
 612  */
 613 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 614                           int addr_len, int flags)
 615 {
 616         struct sock *sk = sock->sk;
 617         int err;
 618         long timeo;
 619
 620         if (addr_len < sizeof(uaddr->sa_family))
 621                 return -EINVAL;
 622
 623         if (uaddr->sa_family == AF_UNSPEC) {
 624                 err = sk->sk_prot->disconnect(sk, flags);
 625                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 626                 goto out;
 627         }
 628
 629         switch (sock->state) {
 630         default:
 631                 err = -EINVAL;
 632                 goto out;
 633         case SS_CONNECTED:
 634                 err = -EISCONN;
 635                 goto out;
 636         case SS_CONNECTING:
 637                 err = -EALREADY;
 638                 /* Fall out of switch with err, set for this state */
 639                 break;
 640         case SS_UNCONNECTED:
 641                 err = -EISCONN;
 642                 if (sk->sk_state != TCP_CLOSE)
 643                         goto out;
 644
 645                 err = sk->sk_prot->connect(sk, uaddr, addr_len);
 646                 if (err < 0)
 647                         goto out;
 648
 649                 sock->state = SS_CONNECTING;
 650
 651                 /* Just entered SS_CONNECTING state; the only
 652                  * difference is that return value in non-blocking
 653                  * case is EINPROGRESS, rather than EALREADY.
 654                  */
 655                 err = -EINPROGRESS;
 656                 break;
 657         }
 658
 659         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
 660
 661         if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 662                 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
 663                                 tcp_sk(sk)->fastopen_req &&
 664                                 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
 665
 666                 /* Error code is set above */
 667                 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
 668                         goto out;
 669
 670                 err = sock_intr_errno(timeo);
 671                 if (signal_pending(current))
 672                         goto out;
 673         }
 674
 675         /* Connection was closed by RST, timeout, ICMP error
 676          * or another process disconnected us.
 677          */
 678         if (sk->sk_state == TCP_CLOSE)
 679                 goto sock_error;
 680
 681         /* sk->sk_err may be not zero now, if RECVERR was ordered by user
 682          * and error was received after socket entered established state.
 683          * Hence, it is handled normally after connect() return successfully.
 684          */
 685
 686         sock->state = SS_CONNECTED;
 687         err = 0;
 688 out:
 689         return err;
 690
 691 sock_error:
 692         err = sock_error(sk) ? : -ECONNABORTED;
 693         sock->state = SS_UNCONNECTED;
 694         if (sk->sk_prot->disconnect(sk, flags))
 695                 sock->state = SS_DISCONNECTING;
 696         goto out;
 697 }
 698 EXPORT_SYMBOL(__inet_stream_connect);
 699
 700 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 701                         int addr_len, int flags)
 702 {
 703         int err;
 704
 705         lock_sock(sock->sk);
 706         err = __inet_stream_connect(sock, uaddr, addr_len, flags);
 707         release_sock(sock->sk);
 708         return err;
 709 }
 710 EXPORT_SYMBOL(inet_stream_connect);
 711
 712 /*
 713  *      Accept a pending connection. The TCP layer now gives BSD semantics.
 714  */
 715
 716 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
 717 {
 718         struct sock *sk1 = sock->sk;
 719         int err = -EINVAL;
 720         struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
 721
 722         if (!sk2)
 723                 goto do_err;
 724
 725         lock_sock(sk2);
 726
 727         sock_rps_record_flow(sk2);
 728         WARN_ON(!((1 << sk2->sk_state) &
 729                   (TCPF_ESTABLISHED | TCPF_SYN_RECV |
 730                   TCPF_CLOSE_WAIT | TCPF_CLOSE)));
 731
 732         sock_graft(sk2, newsock);
 733
 734         newsock->state = SS_CONNECTED;
 735         err = 0;
 736         release_sock(sk2);
 737 do_err:
 738         return err;
 739 }
 740 EXPORT_SYMBOL(inet_accept);
 741
 742
 743 /*
 744  *      This does both peername and sockname.
 745  */
 746 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
 747                         int *uaddr_len, int peer)
 748 {
 749         struct sock *sk         = sock->sk;
 750         struct inet_sock *inet  = inet_sk(sk);
 751         DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
 752
 753         sin->sin_family = AF_INET;
 754         if (peer) {
 755                 if (!inet->inet_dport ||
 756                     (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
 757                      peer == 1))
 758                         return -ENOTCONN;
 759                 sin->sin_port = inet->inet_dport;
 760                 sin->sin_addr.s_addr = inet->inet_daddr;
 761         } else {
 762                 __be32 addr = inet->inet_rcv_saddr;
 763                 if (!addr)
 764                         addr = inet->inet_saddr;
 765                 sin->sin_port = inet->inet_sport;
 766                 sin->sin_addr.s_addr = addr;
 767         }
 768         memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 769         *uaddr_len = sizeof(*sin);
 770         return 0;
 771 }
 772 EXPORT_SYMBOL(inet_getname);
 773
 774 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
 775                  size_t size)
 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         return sk->sk_prot->sendmsg(iocb, sk, msg, size);
 787 }
 788 EXPORT_SYMBOL(inet_sendmsg);
 789
 790 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
 791                       size_t size, int flags)
 792 {
 793         struct sock *sk = sock->sk;
 794
 795         sock_rps_record_flow(sk);
 796
 797         /* We may need to bind the socket. */
 798         if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
 799             inet_autobind(sk))
 800                 return -EAGAIN;
 801
 802         if (sk->sk_prot->sendpage)
 803                 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
 804         return sock_no_sendpage(sock, page, offset, size, flags);
 805 }
 806 EXPORT_SYMBOL(inet_sendpage);
 807
 808 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
 809                  size_t size, int flags)
 810 {
 811         struct sock *sk = sock->sk;
 812         int addr_len = 0;
 813         int err;
 814
 815         sock_rps_record_flow(sk);
 816
 817         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
 818                                    flags & ~MSG_DONTWAIT, &addr_len);
 819         if (err >= 0)
 820                 msg->msg_namelen = addr_len;
 821         return err;
 822 }
 823 EXPORT_SYMBOL(inet_recvmsg);
 824
 825 int inet_shutdown(struct socket *sock, int how)
 826 {
 827         struct sock *sk = sock->sk;
 828         int err = 0;
 829
 830         /* This should really check to make sure
 831          * the socket is a TCP socket. (WHY AC...)
 832          */
 833         how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
 834                        1->2 bit 2 snds.
 835                        2->3 */
 836         if ((how & ~SHUTDOWN_MASK) || !how)     /* MAXINT->0 */
 837                 return -EINVAL;
 838
 839         lock_sock(sk);
 840         if (sock->state == SS_CONNECTING) {
 841                 if ((1 << sk->sk_state) &
 842                     (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
 843                         sock->state = SS_DISCONNECTING;
 844                 else
 845                         sock->state = SS_CONNECTED;
 846         }
 847
 848         switch (sk->sk_state) {
 849         case TCP_CLOSE:
 850                 err = -ENOTCONN;
 851                 /* Hack to wake up other listeners, who can poll for
 852                    POLLHUP, even on eg. unconnected UDP sockets -- RR */
 853         default:
 854                 sk->sk_shutdown |= how;
 855                 if (sk->sk_prot->shutdown)
 856                         sk->sk_prot->shutdown(sk, how);
 857                 break;
 858
 859         /* Remaining two branches are temporary solution for missing
 860          * close() in multithreaded environment. It is _not_ a good idea,
 861          * but we have no choice until close() is repaired at VFS level.
 862          */
 863         case TCP_LISTEN:
 864                 if (!(how & RCV_SHUTDOWN))
 865                         break;
 866                 /* Fall through */
 867         case TCP_SYN_SENT:
 868                 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
 869                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 870                 break;
 871         }
 872
 873         /* Wake up anyone sleeping in poll. */
 874         sk->sk_state_change(sk);
 875         release_sock(sk);
 876         return err;
 877 }
 878 EXPORT_SYMBOL(inet_shutdown);
 879
 880 /*
 881  *      ioctl() calls you can issue on an INET socket. Most of these are
 882  *      device configuration and stuff and very rarely used. Some ioctls
 883  *      pass on to the socket itself.
 884  *
 885  *      NOTE: I like the idea of a module for the config stuff. ie ifconfig
 886  *      loads the devconfigure module does its configuring and unloads it.
 887  *      There's a good 20K of config code hanging around the kernel.
 888  */
 889
 890 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 891 {
 892         struct sock *sk = sock->sk;
 893         int err = 0;
 894         struct net *net = sock_net(sk);
 895
 896         switch (cmd) {
 897         case SIOCGSTAMP:
 898                 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
 899                 break;
 900         case SIOCGSTAMPNS:
 901                 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
 902                 break;
 903         case SIOCADDRT:
 904         case SIOCDELRT:
 905         case SIOCRTMSG:
 906                 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
 907                 break;
 908         case SIOCDARP:
 909         case SIOCGARP:
 910         case SIOCSARP:
 911                 err = arp_ioctl(net, cmd, (void __user *)arg);
 912                 break;
 913         case SIOCGIFADDR:
 914         case SIOCSIFADDR:
 915         case SIOCGIFBRDADDR:
 916         case SIOCSIFBRDADDR:
 917         case SIOCGIFNETMASK:
 918         case SIOCSIFNETMASK:
 919         case SIOCGIFDSTADDR:
 920         case SIOCSIFDSTADDR:
 921         case SIOCSIFPFLAGS:
 922         case SIOCGIFPFLAGS:
 923         case SIOCSIFFLAGS:
 924                 err = devinet_ioctl(net, cmd, (void __user *)arg);
 925                 break;
 926         default:
 927                 if (sk->sk_prot->ioctl)
 928                         err = sk->sk_prot->ioctl(sk, cmd, arg);
 929                 else
 930                         err = -ENOIOCTLCMD;
 931                 break;
 932         }
 933         return err;
 934 }
 935 EXPORT_SYMBOL(inet_ioctl);
 936
 937 #ifdef CONFIG_COMPAT
 938 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 939 {
 940         struct sock *sk = sock->sk;
 941         int err = -ENOIOCTLCMD;
 942
 943         if (sk->sk_prot->compat_ioctl)
 944                 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
 945
 946         return err;
 947 }
 948 #endif
 949
 950 const struct proto_ops inet_stream_ops = {
 951         .family            = PF_INET,
 952         .owner             = THIS_MODULE,
 953         .release           = inet_release,
 954         .bind              = inet_bind,
 955         .connect           = inet_stream_connect,
 956         .socketpair        = sock_no_socketpair,
 957         .accept            = inet_accept,
 958         .getname           = inet_getname,
 959         .poll              = tcp_poll,
 960         .ioctl             = inet_ioctl,
 961         .listen            = inet_listen,
 962         .shutdown          = inet_shutdown,
 963         .setsockopt        = sock_common_setsockopt,
 964         .getsockopt        = sock_common_getsockopt,
 965         .sendmsg           = inet_sendmsg,
 966         .recvmsg           = inet_recvmsg,
 967         .mmap              = sock_no_mmap,
 968         .sendpage          = inet_sendpage,
 969         .splice_read       = tcp_splice_read,
 970 #ifdef CONFIG_COMPAT
 971         .compat_setsockopt = compat_sock_common_setsockopt,
 972         .compat_getsockopt = compat_sock_common_getsockopt,
 973         .compat_ioctl      = inet_compat_ioctl,
 974 #endif
 975 };
 976 EXPORT_SYMBOL(inet_stream_ops);
 977
 978 const struct proto_ops inet_dgram_ops = {
 979         .family            = PF_INET,
 980         .owner             = THIS_MODULE,
 981         .release           = inet_release,
 982         .bind              = inet_bind,
 983         .connect           = inet_dgram_connect,
 984         .socketpair        = sock_no_socketpair,
 985         .accept            = sock_no_accept,
 986         .getname           = inet_getname,
 987         .poll              = udp_poll,
 988         .ioctl             = inet_ioctl,
 989         .listen            = sock_no_listen,
 990         .shutdown          = inet_shutdown,
 991         .setsockopt        = sock_common_setsockopt,
 992         .getsockopt        = sock_common_getsockopt,
 993         .sendmsg           = inet_sendmsg,
 994         .recvmsg           = inet_recvmsg,
 995         .mmap              = sock_no_mmap,
 996         .sendpage          = inet_sendpage,
 997 #ifdef CONFIG_COMPAT
 998         .compat_setsockopt = compat_sock_common_setsockopt,
 999         .compat_getsockopt = compat_sock_common_getsockopt,
1000         .compat_ioctl      = inet_compat_ioctl,
1001 #endif
1002 };
1003 EXPORT_SYMBOL(inet_dgram_ops);
1004
1005 /*
1006  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1007  * udp_poll
1008  */
1009 static const struct proto_ops inet_sockraw_ops = {
1010         .family            = PF_INET,
1011         .owner             = THIS_MODULE,
1012         .release           = inet_release,
1013         .bind              = inet_bind,
1014         .connect           = inet_dgram_connect,
1015         .socketpair        = sock_no_socketpair,
1016         .accept            = sock_no_accept,
1017         .getname           = inet_getname,
1018         .poll              = datagram_poll,
1019         .ioctl             = inet_ioctl,
1020         .listen            = sock_no_listen,
1021         .shutdown          = inet_shutdown,
1022         .setsockopt        = sock_common_setsockopt,
1023         .getsockopt        = sock_common_getsockopt,
1024         .sendmsg           = inet_sendmsg,
1025         .recvmsg           = inet_recvmsg,
1026         .mmap              = sock_no_mmap,
1027         .sendpage          = inet_sendpage,
1028 #ifdef CONFIG_COMPAT
1029         .compat_setsockopt = compat_sock_common_setsockopt,
1030         .compat_getsockopt = compat_sock_common_getsockopt,
1031         .compat_ioctl      = inet_compat_ioctl,
1032 #endif
1033 };
1034
1035 static const struct net_proto_family inet_family_ops = {
1036         .family = PF_INET,
1037         .create = inet_create,
1038         .owner  = THIS_MODULE,
1039 };
1040
1041 /* Upon startup we insert all the elements in inetsw_array[] into
1042  * the linked list inetsw.
1043  */
1044 static struct inet_protosw inetsw_array[] =
1045 {
1046         {
1047                 .type =       SOCK_STREAM,
1048                 .protocol =   IPPROTO_TCP,
1049                 .prot =       &tcp_prot,
1050                 .ops =        &inet_stream_ops,
1051                 .no_check =   0,
1052                 .flags =      INET_PROTOSW_PERMANENT |
1053                               INET_PROTOSW_ICSK,
1054         },
1055
1056         {
1057                 .type =       SOCK_DGRAM,
1058                 .protocol =   IPPROTO_UDP,
1059                 .prot =       &udp_prot,
1060                 .ops =        &inet_dgram_ops,
1061                 .no_check =   UDP_CSUM_DEFAULT,
1062                 .flags =      INET_PROTOSW_PERMANENT,
1063        },
1064
1065        {
1066                 .type =       SOCK_DGRAM,
1067                 .protocol =   IPPROTO_ICMP,
1068                 .prot =       &ping_prot,
1069                 .ops =        &inet_dgram_ops,
1070                 .no_check =   UDP_CSUM_DEFAULT,
1071                 .flags =      INET_PROTOSW_REUSE,
1072        },
1073
1074        {
1075                .type =       SOCK_RAW,
1076                .protocol =   IPPROTO_IP,        /* wild card */
1077                .prot =       &raw_prot,
1078                .ops =        &inet_sockraw_ops,
1079                .no_check =   UDP_CSUM_DEFAULT,
1080                .flags =      INET_PROTOSW_REUSE,
1081        }
1082 };
1083
1084 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1085
1086 void inet_register_protosw(struct inet_protosw *p)
1087 {
1088         struct list_head *lh;
1089         struct inet_protosw *answer;
1090         int protocol = p->protocol;
1091         struct list_head *last_perm;
1092
1093         spin_lock_bh(&inetsw_lock);
1094
1095         if (p->type >= SOCK_MAX)
1096                 goto out_illegal;
1097
1098         /* If we are trying to override a permanent protocol, bail. */
1099         answer = NULL;
1100         last_perm = &inetsw[p->type];
1101         list_for_each(lh, &inetsw[p->type]) {
1102                 answer = list_entry(lh, struct inet_protosw, list);
1103
1104                 /* Check only the non-wild match. */
1105                 if (INET_PROTOSW_PERMANENT & answer->flags) {
1106                         if (protocol == answer->protocol)
1107                                 break;
1108                         last_perm = lh;
1109                 }
1110
1111                 answer = NULL;
1112         }
1113         if (answer)
1114                 goto out_permanent;
1115
1116         /* Add the new entry after the last permanent entry if any, so that
1117          * the new entry does not override a permanent entry when matched with
1118          * a wild-card protocol. But it is allowed to override any existing
1119          * non-permanent entry.  This means that when we remove this entry, the
1120          * system automatically returns to the old behavior.
1121          */
1122         list_add_rcu(&p->list, last_perm);
1123 out:
1124         spin_unlock_bh(&inetsw_lock);
1125
1126         return;
1127
1128 out_permanent:
1129         pr_err("Attempt to override permanent protocol %d\n", protocol);
1130         goto out;
1131
1132 out_illegal:
1133         pr_err("Ignoring attempt to register invalid socket type %d\n",
1134                p->type);
1135         goto out;
1136 }
1137 EXPORT_SYMBOL(inet_register_protosw);
1138
1139 void inet_unregister_protosw(struct inet_protosw *p)
1140 {
1141         if (INET_PROTOSW_PERMANENT & p->flags) {
1142                 pr_err("Attempt to unregister permanent protocol %d\n",
1143                        p->protocol);
1144         } else {
1145                 spin_lock_bh(&inetsw_lock);
1146                 list_del_rcu(&p->list);
1147                 spin_unlock_bh(&inetsw_lock);
1148
1149                 synchronize_net();
1150         }
1151 }
1152 EXPORT_SYMBOL(inet_unregister_protosw);
1153
1154 /*
1155  *      Shall we try to damage output packets if routing dev changes?
1156  */
1157
1158 int sysctl_ip_dynaddr __read_mostly;
1159
1160 static int inet_sk_reselect_saddr(struct sock *sk)
1161 {
1162         struct inet_sock *inet = inet_sk(sk);
1163         __be32 old_saddr = inet->inet_saddr;
1164         __be32 daddr = inet->inet_daddr;
1165         struct flowi4 *fl4;
1166         struct rtable *rt;
1167         __be32 new_saddr;
1168         struct ip_options_rcu *inet_opt;
1169
1170         inet_opt = rcu_dereference_protected(inet->inet_opt,
1171                                              sock_owned_by_user(sk));
1172         if (inet_opt && inet_opt->opt.srr)
1173                 daddr = inet_opt->opt.faddr;
1174
1175         /* Query new route. */
1176         fl4 = &inet->cork.fl.u.ip4;
1177         rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1178                               sk->sk_bound_dev_if, sk->sk_protocol,
1179                               inet->inet_sport, inet->inet_dport, sk, false);
1180         if (IS_ERR(rt))
1181                 return PTR_ERR(rt);
1182
1183         sk_setup_caps(sk, &rt->dst);
1184
1185         new_saddr = fl4->saddr;
1186
1187         if (new_saddr == old_saddr)
1188                 return 0;
1189
1190         if (sysctl_ip_dynaddr > 1) {
1191                 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1192                         __func__, &old_saddr, &new_saddr);
1193         }
1194
1195         inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1196
1197         /*
1198          * XXX The only one ugly spot where we need to
1199          * XXX really change the sockets identity after
1200          * XXX it has entered the hashes. -DaveM
1201          *
1202          * Besides that, it does not check for connection
1203          * uniqueness. Wait for troubles.
1204          */
1205         __sk_prot_rehash(sk);
1206         return 0;
1207 }
1208
1209 int inet_sk_rebuild_header(struct sock *sk)
1210 {
1211         struct inet_sock *inet = inet_sk(sk);
1212         struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1213         __be32 daddr;
1214         struct ip_options_rcu *inet_opt;
1215         struct flowi4 *fl4;
1216         int err;
1217
1218         /* Route is OK, nothing to do. */
1219         if (rt)
1220                 return 0;
1221
1222         /* Reroute. */
1223         rcu_read_lock();
1224         inet_opt = rcu_dereference(inet->inet_opt);
1225         daddr = inet->inet_daddr;
1226         if (inet_opt && inet_opt->opt.srr)
1227                 daddr = inet_opt->opt.faddr;
1228         rcu_read_unlock();
1229         fl4 = &inet->cork.fl.u.ip4;
1230         rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1231                                    inet->inet_dport, inet->inet_sport,
1232                                    sk->sk_protocol, RT_CONN_FLAGS(sk),
1233                                    sk->sk_bound_dev_if);
1234         if (!IS_ERR(rt)) {
1235                 err = 0;
1236                 sk_setup_caps(sk, &rt->dst);
1237         } else {
1238                 err = PTR_ERR(rt);
1239
1240                 /* Routing failed... */
1241                 sk->sk_route_caps = 0;
1242                 /*
1243                  * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1244                  * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1245                  */
1246                 if (!sysctl_ip_dynaddr ||
1247                     sk->sk_state != TCP_SYN_SENT ||
1248                     (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1249                     (err = inet_sk_reselect_saddr(sk)) != 0)
1250                         sk->sk_err_soft = -err;
1251         }
1252
1253         return err;
1254 }
1255 EXPORT_SYMBOL(inet_sk_rebuild_header);
1256
1257 static int inet_gso_send_check(struct sk_buff *skb)
1258 {
1259         const struct net_protocol *ops;
1260         const struct iphdr *iph;
1261         int proto;
1262         int ihl;
1263         int err = -EINVAL;
1264
1265         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1266                 goto out;
1267
1268         iph = ip_hdr(skb);
1269         ihl = iph->ihl * 4;
1270         if (ihl < sizeof(*iph))
1271                 goto out;
1272
1273         if (unlikely(!pskb_may_pull(skb, ihl)))
1274                 goto out;
1275
1276         __skb_pull(skb, ihl);
1277         skb_reset_transport_header(skb);
1278         iph = ip_hdr(skb);
1279         proto = iph->protocol;
1280         err = -EPROTONOSUPPORT;
1281
1282         rcu_read_lock();
1283         ops = rcu_dereference(inet_protos[proto]);
1284         if (likely(ops && ops->gso_send_check))
1285                 err = ops->gso_send_check(skb);
1286         rcu_read_unlock();
1287
1288 out:
1289         return err;
1290 }
1291
1292 static struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1293         netdev_features_t features)
1294 {
1295         struct sk_buff *segs = ERR_PTR(-EINVAL);
1296         const struct net_protocol *ops;
1297         struct iphdr *iph;
1298         int proto;
1299         int ihl;
1300         int id;
1301         unsigned int offset = 0;
1302
1303         if (!(features & NETIF_F_V4_CSUM))
1304                 features &= ~NETIF_F_SG;
1305
1306         if (unlikely(skb_shinfo(skb)->gso_type &
1307                      ~(SKB_GSO_TCPV4 |
1308                        SKB_GSO_UDP |
1309                        SKB_GSO_DODGY |
1310                        SKB_GSO_TCP_ECN |
1311                        0)))
1312                 goto out;
1313
1314         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1315                 goto out;
1316
1317         iph = ip_hdr(skb);
1318         ihl = iph->ihl * 4;
1319         if (ihl < sizeof(*iph))
1320                 goto out;
1321
1322         if (unlikely(!pskb_may_pull(skb, ihl)))
1323                 goto out;
1324
1325         __skb_pull(skb, ihl);
1326         skb_reset_transport_header(skb);
1327         iph = ip_hdr(skb);
1328         id = ntohs(iph->id);
1329         proto = iph->protocol;
1330         segs = ERR_PTR(-EPROTONOSUPPORT);
1331
1332         rcu_read_lock();
1333         ops = rcu_dereference(inet_protos[proto]);
1334         if (likely(ops && ops->gso_segment))
1335                 segs = ops->gso_segment(skb, features);
1336         rcu_read_unlock();
1337
1338         if (!segs || IS_ERR(segs))
1339                 goto out;
1340
1341         skb = segs;
1342         do {
1343                 iph = ip_hdr(skb);
1344                 if (proto == IPPROTO_UDP) {
1345                         iph->id = htons(id);
1346                         iph->frag_off = htons(offset >> 3);
1347                         if (skb->next != NULL)
1348                                 iph->frag_off |= htons(IP_MF);
1349                         offset += (skb->len - skb->mac_len - iph->ihl * 4);
1350                 } else
1351                         iph->id = htons(id++);
1352                 iph->tot_len = htons(skb->len - skb->mac_len);
1353                 iph->check = 0;
1354                 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1355         } while ((skb = skb->next));
1356
1357 out:
1358         return segs;
1359 }
1360
1361 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1362                                          struct sk_buff *skb)
1363 {
1364         const struct net_protocol *ops;
1365         struct sk_buff **pp = NULL;
1366         struct sk_buff *p;
1367         const struct iphdr *iph;
1368         unsigned int hlen;
1369         unsigned int off;
1370         unsigned int id;
1371         int flush = 1;
1372         int proto;
1373
1374         off = skb_gro_offset(skb);
1375         hlen = off + sizeof(*iph);
1376         iph = skb_gro_header_fast(skb, off);
1377         if (skb_gro_header_hard(skb, hlen)) {
1378                 iph = skb_gro_header_slow(skb, hlen, off);
1379                 if (unlikely(!iph))
1380                         goto out;
1381         }
1382
1383         proto = iph->protocol;
1384
1385         rcu_read_lock();
1386         ops = rcu_dereference(inet_protos[proto]);
1387         if (!ops || !ops->gro_receive)
1388                 goto out_unlock;
1389
1390         if (*(u8 *)iph != 0x45)
1391                 goto out_unlock;
1392
1393         if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1394                 goto out_unlock;
1395
1396         id = ntohl(*(__be32 *)&iph->id);
1397         flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF));
1398         id >>= 16;
1399
1400         for (p = *head; p; p = p->next) {
1401                 struct iphdr *iph2;
1402
1403                 if (!NAPI_GRO_CB(p)->same_flow)
1404                         continue;
1405
1406                 iph2 = ip_hdr(p);
1407
1408                 if ((iph->protocol ^ iph2->protocol) |
1409                     ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1410                     ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1411                         NAPI_GRO_CB(p)->same_flow = 0;
1412                         continue;
1413                 }
1414
1415                 /* All fields must match except length and checksum. */
1416                 NAPI_GRO_CB(p)->flush |=
1417                         (iph->ttl ^ iph2->ttl) |
1418                         (iph->tos ^ iph2->tos) |
1419                         ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1420
1421                 NAPI_GRO_CB(p)->flush |= flush;
1422         }
1423
1424         NAPI_GRO_CB(skb)->flush |= flush;
1425         skb_gro_pull(skb, sizeof(*iph));
1426         skb_set_transport_header(skb, skb_gro_offset(skb));
1427
1428         pp = ops->gro_receive(head, skb);
1429
1430 out_unlock:
1431         rcu_read_unlock();
1432
1433 out:
1434         NAPI_GRO_CB(skb)->flush |= flush;
1435
1436         return pp;
1437 }
1438
1439 static int inet_gro_complete(struct sk_buff *skb)
1440 {
1441         __be16 newlen = htons(skb->len - skb_network_offset(skb));
1442         struct iphdr *iph = ip_hdr(skb);
1443         const struct net_protocol *ops;
1444         int proto = iph->protocol;
1445         int err = -ENOSYS;
1446
1447         csum_replace2(&iph->check, iph->tot_len, newlen);
1448         iph->tot_len = newlen;
1449
1450         rcu_read_lock();
1451         ops = rcu_dereference(inet_protos[proto]);
1452         if (WARN_ON(!ops || !ops->gro_complete))
1453                 goto out_unlock;
1454
1455         err = ops->gro_complete(skb);
1456
1457 out_unlock:
1458         rcu_read_unlock();
1459
1460         return err;
1461 }
1462
1463 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1464                          unsigned short type, unsigned char protocol,
1465                          struct net *net)
1466 {
1467         struct socket *sock;
1468         int rc = sock_create_kern(family, type, protocol, &sock);
1469
1470         if (rc == 0) {
1471                 *sk = sock->sk;
1472                 (*sk)->sk_allocation = GFP_ATOMIC;
1473                 /*
1474                  * Unhash it so that IP input processing does not even see it,
1475                  * we do not wish this socket to see incoming packets.
1476                  */
1477                 (*sk)->sk_prot->unhash(*sk);
1478
1479                 sk_change_net(*sk, net);
1480         }
1481         return rc;
1482 }
1483 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1484
1485 unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1486 {
1487         unsigned long res = 0;
1488         int i, j;
1489
1490         for_each_possible_cpu(i) {
1491                 for (j = 0; j < SNMP_ARRAY_SZ; j++)
1492                         res += *(((unsigned long *) per_cpu_ptr(mib[j], i)) + offt);
1493         }
1494         return res;
1495 }
1496 EXPORT_SYMBOL_GPL(snmp_fold_field);
1497
1498 #if BITS_PER_LONG==32
1499
1500 u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset)
1501 {
1502         u64 res = 0;
1503         int cpu;
1504
1505         for_each_possible_cpu(cpu) {
1506                 void *bhptr;
1507                 struct u64_stats_sync *syncp;
1508                 u64 v;
1509                 unsigned int start;
1510
1511                 bhptr = per_cpu_ptr(mib[0], cpu);
1512                 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1513                 do {
1514                         start = u64_stats_fetch_begin_bh(syncp);
1515                         v = *(((u64 *) bhptr) + offt);
1516                 } while (u64_stats_fetch_retry_bh(syncp, start));
1517
1518                 res += v;
1519         }
1520         return res;
1521 }
1522 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1523 #endif
1524
1525 int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align)
1526 {
1527         BUG_ON(ptr == NULL);
1528         ptr[0] = __alloc_percpu(mibsize, align);
1529         if (!ptr[0])
1530                 return -ENOMEM;
1531 #if SNMP_ARRAY_SZ == 2
1532         ptr[1] = __alloc_percpu(mibsize, align);
1533         if (!ptr[1]) {
1534                 free_percpu(ptr[0]);
1535                 ptr[0] = NULL;
1536                 return -ENOMEM;
1537         }
1538 #endif
1539         return 0;
1540 }
1541 EXPORT_SYMBOL_GPL(snmp_mib_init);
1542
1543 void snmp_mib_free(void __percpu *ptr[SNMP_ARRAY_SZ])
1544 {
1545         int i;
1546
1547         BUG_ON(ptr == NULL);
1548         for (i = 0; i < SNMP_ARRAY_SZ; i++) {
1549                 free_percpu(ptr[i]);
1550                 ptr[i] = NULL;
1551         }
1552 }
1553 EXPORT_SYMBOL_GPL(snmp_mib_free);
1554
1555 #ifdef CONFIG_IP_MULTICAST
1556 static const struct net_protocol igmp_protocol = {
1557         .handler =      igmp_rcv,
1558         .netns_ok =     1,
1559 };
1560 #endif
1561
1562 static const struct net_protocol tcp_protocol = {
1563         .early_demux    =       tcp_v4_early_demux,
1564         .handler        =       tcp_v4_rcv,
1565         .err_handler    =       tcp_v4_err,
1566         .gso_send_check =       tcp_v4_gso_send_check,
1567         .gso_segment    =       tcp_tso_segment,
1568         .gro_receive    =       tcp4_gro_receive,
1569         .gro_complete   =       tcp4_gro_complete,
1570         .no_policy      =       1,
1571         .netns_ok       =       1,
1572 };
1573
1574 static const struct net_protocol udp_protocol = {
1575         .handler =      udp_rcv,
1576         .err_handler =  udp_err,
1577         .gso_send_check = udp4_ufo_send_check,
1578         .gso_segment = udp4_ufo_fragment,
1579         .no_policy =    1,
1580         .netns_ok =     1,
1581 };
1582
1583 static const struct net_protocol icmp_protocol = {
1584         .handler =      icmp_rcv,
1585         .err_handler =  ping_err,
1586         .no_policy =    1,
1587         .netns_ok =     1,
1588 };
1589
1590 static __net_init int ipv4_mib_init_net(struct net *net)
1591 {
1592         if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1593                           sizeof(struct tcp_mib),
1594                           __alignof__(struct tcp_mib)) < 0)
1595                 goto err_tcp_mib;
1596         if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1597                           sizeof(struct ipstats_mib),
1598                           __alignof__(struct ipstats_mib)) < 0)
1599                 goto err_ip_mib;
1600         if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1601                           sizeof(struct linux_mib),
1602                           __alignof__(struct linux_mib)) < 0)
1603                 goto err_net_mib;
1604         if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1605                           sizeof(struct udp_mib),
1606                           __alignof__(struct udp_mib)) < 0)
1607                 goto err_udp_mib;
1608         if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1609                           sizeof(struct udp_mib),
1610                           __alignof__(struct udp_mib)) < 0)
1611                 goto err_udplite_mib;
1612         if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1613                           sizeof(struct icmp_mib),
1614                           __alignof__(struct icmp_mib)) < 0)
1615                 goto err_icmp_mib;
1616         net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1617                                               GFP_KERNEL);
1618         if (!net->mib.icmpmsg_statistics)
1619                 goto err_icmpmsg_mib;
1620
1621         tcp_mib_init(net);
1622         return 0;
1623
1624 err_icmpmsg_mib:
1625         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1626 err_icmp_mib:
1627         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1628 err_udplite_mib:
1629         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1630 err_udp_mib:
1631         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1632 err_net_mib:
1633         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1634 err_ip_mib:
1635         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1636 err_tcp_mib:
1637         return -ENOMEM;
1638 }
1639
1640 static __net_exit void ipv4_mib_exit_net(struct net *net)
1641 {
1642         kfree(net->mib.icmpmsg_statistics);
1643         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1644         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1645         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1646         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1647         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1648         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1649 }
1650
1651 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1652         .init = ipv4_mib_init_net,
1653         .exit = ipv4_mib_exit_net,
1654 };
1655
1656 static int __init init_ipv4_mibs(void)
1657 {
1658         return register_pernet_subsys(&ipv4_mib_ops);
1659 }
1660
1661 static int ipv4_proc_init(void);
1662
1663 /*
1664  *      IP protocol layer initialiser
1665  */
1666
1667 static struct packet_type ip_packet_type __read_mostly = {
1668         .type = cpu_to_be16(ETH_P_IP),
1669         .func = ip_rcv,
1670         .gso_send_check = inet_gso_send_check,
1671         .gso_segment = inet_gso_segment,
1672         .gro_receive = inet_gro_receive,
1673         .gro_complete = inet_gro_complete,
1674 };
1675
1676 static int __init inet_init(void)
1677 {
1678         struct sk_buff *dummy_skb;
1679         struct inet_protosw *q;
1680         struct list_head *r;
1681         int rc = -EINVAL;
1682
1683         BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1684
1685         sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL);
1686         if (!sysctl_local_reserved_ports)
1687                 goto out;
1688
1689         rc = proto_register(&tcp_prot, 1);
1690         if (rc)
1691                 goto out_free_reserved_ports;
1692
1693         rc = proto_register(&udp_prot, 1);
1694         if (rc)
1695                 goto out_unregister_tcp_proto;
1696
1697         rc = proto_register(&raw_prot, 1);
1698         if (rc)
1699                 goto out_unregister_udp_proto;
1700
1701         rc = proto_register(&ping_prot, 1);
1702         if (rc)
1703                 goto out_unregister_raw_proto;
1704
1705         /*
1706          *      Tell SOCKET that we are alive...
1707          */
1708
1709         (void)sock_register(&inet_family_ops);
1710
1711 #ifdef CONFIG_SYSCTL
1712         ip_static_sysctl_init();
1713 #endif
1714
1715         tcp_prot.sysctl_mem = init_net.ipv4.sysctl_tcp_mem;
1716
1717         /*
1718          *      Add all the base protocols.
1719          */
1720
1721         if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1722                 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1723         if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1724                 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1725         if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1726                 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1727 #ifdef CONFIG_IP_MULTICAST
1728         if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1729                 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1730 #endif
1731
1732         /* Register the socket-side information for inet_create. */
1733         for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1734                 INIT_LIST_HEAD(r);
1735
1736         for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1737                 inet_register_protosw(q);
1738
1739         /*
1740          *      Set the ARP module up
1741          */
1742
1743         arp_init();
1744
1745         /*
1746          *      Set the IP module up
1747          */
1748
1749         ip_init();
1750
1751         tcp_v4_init();
1752
1753         /* Setup TCP slab cache for open requests. */
1754         tcp_init();
1755
1756         /* Setup UDP memory threshold */
1757         udp_init();
1758
1759         /* Add UDP-Lite (RFC 3828) */
1760         udplite4_register();
1761
1762         ping_init();
1763
1764         /*
1765          *      Set the ICMP layer up
1766          */
1767
1768         if (icmp_init() < 0)
1769                 panic("Failed to create the ICMP control socket.\n");
1770
1771         /*
1772          *      Initialise the multicast router
1773          */
1774 #if defined(CONFIG_IP_MROUTE)
1775         if (ip_mr_init())
1776                 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1777 #endif
1778         /*
1779          *      Initialise per-cpu ipv4 mibs
1780          */
1781
1782         if (init_ipv4_mibs())
1783                 pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1784
1785         ipv4_proc_init();
1786
1787         ipfrag_init();
1788
1789         dev_add_pack(&ip_packet_type);
1790
1791         rc = 0;
1792 out:
1793         return rc;
1794 out_unregister_raw_proto:
1795         proto_unregister(&raw_prot);
1796 out_unregister_udp_proto:
1797         proto_unregister(&udp_prot);
1798 out_unregister_tcp_proto:
1799         proto_unregister(&tcp_prot);
1800 out_free_reserved_ports:
1801         kfree(sysctl_local_reserved_ports);
1802         goto out;
1803 }
1804
1805 fs_initcall(inet_init);
1806
1807 /* ------------------------------------------------------------------------ */
1808
1809 #ifdef CONFIG_PROC_FS
1810 static int __init ipv4_proc_init(void)
1811 {
1812         int rc = 0;
1813
1814         if (raw_proc_init())
1815                 goto out_raw;
1816         if (tcp4_proc_init())
1817                 goto out_tcp;
1818         if (udp4_proc_init())
1819                 goto out_udp;
1820         if (ping_proc_init())
1821                 goto out_ping;
1822         if (ip_misc_proc_init())
1823                 goto out_misc;
1824 out:
1825         return rc;
1826 out_misc:
1827         ping_proc_exit();
1828 out_ping:
1829         udp4_proc_exit();
1830 out_udp:
1831         tcp4_proc_exit();
1832 out_tcp:
1833         raw_proc_exit();
1834 out_raw:
1835         rc = -ENOMEM;
1836         goto out;
1837 }
1838
1839 #else /* CONFIG_PROC_FS */
1840 static int __init ipv4_proc_init(void)
1841 {
1842         return 0;
1843 }
1844 #endif /* CONFIG_PROC_FS */
1845
1846 MODULE_ALIAS_NETPROTO(PF_INET);
1847