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 #include <linux/err.h>
  69 #include <linux/errno.h>
  70 #include <linux/types.h>
  71 #include <linux/socket.h>
  72 #include <linux/in.h>
  73 #include <linux/kernel.h>
  74 #include <linux/module.h>
  75 #include <linux/sched.h>
  76 #include <linux/timer.h>
  77 #include <linux/string.h>
  78 #include <linux/sockios.h>
  79 #include <linux/net.h>
  80 #include <linux/capability.h>
  81 #include <linux/fcntl.h>
  82 #include <linux/mm.h>
  83 #include <linux/interrupt.h>
  84 #include <linux/stat.h>
  85 #include <linux/init.h>
  86 #include <linux/poll.h>
  87 #include <linux/netfilter_ipv4.h>
  88 #include <linux/random.h>
  89 #include <linux/slab.h>
  90
  91 #include <asm/uaccess.h>
  92 #include <asm/system.h>
  93
  94 #include <linux/inet.h>
  95 #include <linux/igmp.h>
  96 #include <linux/inetdevice.h>
  97 #include <linux/netdevice.h>
  98 #include <net/checksum.h>
  99 #include <net/ip.h>
 100 #include <net/protocol.h>
 101 #include <net/arp.h>
 102 #include <net/route.h>
 103 #include <net/ip_fib.h>
 104 #include <net/inet_connection_sock.h>
 105 #include <net/tcp.h>
 106 #include <net/udp.h>
 107 #include <net/udplite.h>
 108 #include <net/ping.h>
 109 #include <linux/skbuff.h>
 110 #include <net/sock.h>
 111 #include <net/raw.h>
 112 #include <net/icmp.h>
 113 #include <net/ipip.h>
 114 #include <net/inet_common.h>
 115 #include <net/xfrm.h>
 116 #include <net/net_namespace.h>
 117 #ifdef CONFIG_IP_MROUTE
 118 #include <linux/mroute.h>
 119 #endif
 120
 121
 122 /* The inetsw table contains everything that inet_create needs to
 123  * build a new socket.
 124  */
 125 static struct list_head inetsw[SOCK_MAX];
 126 static DEFINE_SPINLOCK(inetsw_lock);
 127
 128 struct ipv4_config ipv4_config;
 129 EXPORT_SYMBOL(ipv4_config);
 130
 131 /* New destruction routine */
 132
 133 void inet_sock_destruct(struct sock *sk)
 134 {
 135         struct inet_sock *inet = inet_sk(sk);
 136
 137         __skb_queue_purge(&sk->sk_receive_queue);
 138         __skb_queue_purge(&sk->sk_error_queue);
 139
 140         sk_mem_reclaim(sk);
 141
 142         if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
 143                 pr_err("Attempt to release TCP socket in state %d %p\n",
 144                        sk->sk_state, sk);
 145                 return;
 146         }
 147         if (!sock_flag(sk, SOCK_DEAD)) {
 148                 pr_err("Attempt to release alive inet socket %p\n", sk);
 149                 return;
 150         }
 151
 152         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
 153         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
 154         WARN_ON(sk->sk_wmem_queued);
 155         WARN_ON(sk->sk_forward_alloc);
 156
 157         kfree(rcu_dereference_protected(inet->inet_opt, 1));
 158         dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
 159         sk_refcnt_debug_dec(sk);
 160 }
 161 EXPORT_SYMBOL(inet_sock_destruct);
 162
 163 /*
 164  *      The routines beyond this point handle the behaviour of an AF_INET
 165  *      socket object. Mostly it punts to the subprotocols of IP to do
 166  *      the work.
 167  */
 168
 169 /*
 170  *      Automatically bind an unbound socket.
 171  */
 172
 173 static int inet_autobind(struct sock *sk)
 174 {
 175         struct inet_sock *inet;
 176         /* We may need to bind the socket. */
 177         lock_sock(sk);
 178         inet = inet_sk(sk);
 179         if (!inet->inet_num) {
 180                 if (sk->sk_prot->get_port(sk, 0)) {
 181                         release_sock(sk);
 182                         return -EAGAIN;
 183                 }
 184                 inet->inet_sport = htons(inet->inet_num);
 185         }
 186         release_sock(sk);
 187         return 0;
 188 }
 189
 190 /*
 191  *      Move a socket into listening state.
 192  */
 193 int inet_listen(struct socket *sock, int backlog)
 194 {
 195         struct sock *sk = sock->sk;
 196         unsigned char old_state;
 197         int err;
 198
 199         lock_sock(sk);
 200
 201         err = -EINVAL;
 202         if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
 203                 goto out;
 204
 205         old_state = sk->sk_state;
 206         if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
 207                 goto out;
 208
 209         /* Really, if the socket is already in listen state
 210          * we can only allow the backlog to be adjusted.
 211          */
 212         if (old_state != TCP_LISTEN) {
 213                 err = inet_csk_listen_start(sk, backlog);
 214                 if (err)
 215                         goto out;
 216         }
 217         sk->sk_max_ack_backlog = backlog;
 218         err = 0;
 219
 220 out:
 221         release_sock(sk);
 222         return err;
 223 }
 224 EXPORT_SYMBOL(inet_listen);
 225
 226 u32 inet_ehash_secret __read_mostly;
 227 EXPORT_SYMBOL(inet_ehash_secret);
 228
 229 /*
 230  * inet_ehash_secret must be set exactly once
 231  */
 232 void build_ehash_secret(void)
 233 {
 234         u32 rnd;
 235
 236         do {
 237                 get_random_bytes(&rnd, sizeof(rnd));
 238         } while (rnd == 0);
 239
 240         cmpxchg(&inet_ehash_secret, 0, rnd);
 241 }
 242 EXPORT_SYMBOL(build_ehash_secret);
 243
 244 static inline int inet_netns_ok(struct net *net, int protocol)
 245 {
 246         int hash;
 247         const struct net_protocol *ipprot;
 248
 249         if (net_eq(net, &init_net))
 250                 return 1;
 251
 252         hash = protocol & (MAX_INET_PROTOS - 1);
 253         ipprot = rcu_dereference(inet_protos[hash]);
 254
 255         if (ipprot == NULL)
 256                 /* raw IP is OK */
 257                 return 1;
 258         return ipprot->netns_ok;
 259 }
 260
 261 /*
 262  *      Create an inet socket.
 263  */
 264
 265 static int inet_create(struct net *net, struct socket *sock, int protocol,
 266                        int kern)
 267 {
 268         struct sock *sk;
 269         struct inet_protosw *answer;
 270         struct inet_sock *inet;
 271         struct proto *answer_prot;
 272         unsigned char answer_flags;
 273         char answer_no_check;
 274         int try_loading_module = 0;
 275         int err;
 276
 277         if (unlikely(!inet_ehash_secret))
 278                 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
 279                         build_ehash_secret();
 280
 281         sock->state = SS_UNCONNECTED;
 282
 283         /* Look for the requested type/protocol pair. */
 284 lookup_protocol:
 285         err = -ESOCKTNOSUPPORT;
 286         rcu_read_lock();
 287         list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
 288
 289                 err = 0;
 290                 /* Check the non-wild match. */
 291                 if (protocol == answer->protocol) {
 292                         if (protocol != IPPROTO_IP)
 293                                 break;
 294                 } else {
 295                         /* Check for the two wild cases. */
 296                         if (IPPROTO_IP == protocol) {
 297                                 protocol = answer->protocol;
 298                                 break;
 299                         }
 300                         if (IPPROTO_IP == answer->protocol)
 301                                 break;
 302                 }
 303                 err = -EPROTONOSUPPORT;
 304         }
 305
 306         if (unlikely(err)) {
 307                 if (try_loading_module < 2) {
 308                         rcu_read_unlock();
 309                         /*
 310                          * Be more specific, e.g. net-pf-2-proto-132-type-1
 311                          * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
 312                          */
 313                         if (++try_loading_module == 1)
 314                                 request_module("net-pf-%d-proto-%d-type-%d",
 315                                                PF_INET, protocol, sock->type);
 316                         /*
 317                          * Fall back to generic, e.g. net-pf-2-proto-132
 318                          * (net-pf-PF_INET-proto-IPPROTO_SCTP)
 319                          */
 320                         else
 321                                 request_module("net-pf-%d-proto-%d",
 322                                                PF_INET, protocol);
 323                         goto lookup_protocol;
 324                 } else
 325                         goto out_rcu_unlock;
 326         }
 327
 328         err = -EPERM;
 329         if (sock->type == SOCK_RAW && !kern && !capable(CAP_NET_RAW))
 330                 goto out_rcu_unlock;
 331
 332         err = -EAFNOSUPPORT;
 333         if (!inet_netns_ok(net, protocol))
 334                 goto out_rcu_unlock;
 335
 336         sock->ops = answer->ops;
 337         answer_prot = answer->prot;
 338         answer_no_check = answer->no_check;
 339         answer_flags = answer->flags;
 340         rcu_read_unlock();
 341
 342         WARN_ON(answer_prot->slab == NULL);
 343
 344         err = -ENOBUFS;
 345         sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
 346         if (sk == NULL)
 347                 goto out;
 348
 349         err = 0;
 350         sk->sk_no_check = answer_no_check;
 351         if (INET_PROTOSW_REUSE & answer_flags)
 352                 sk->sk_reuse = 1;
 353
 354         inet = inet_sk(sk);
 355         inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
 356
 357         inet->nodefrag = 0;
 358
 359         if (SOCK_RAW == sock->type) {
 360                 inet->inet_num = protocol;
 361                 if (IPPROTO_RAW == protocol)
 362                         inet->hdrincl = 1;
 363         }
 364
 365         if (ipv4_config.no_pmtu_disc)
 366                 inet->pmtudisc = IP_PMTUDISC_DONT;
 367         else
 368                 inet->pmtudisc = IP_PMTUDISC_WANT;
 369
 370         inet->inet_id = 0;
 371
 372         sock_init_data(sock, sk);
 373
 374         sk->sk_destruct    = inet_sock_destruct;
 375         sk->sk_protocol    = protocol;
 376         sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
 377
 378         inet->uc_ttl    = -1;
 379         inet->mc_loop   = 1;
 380         inet->mc_ttl    = 1;
 381         inet->mc_all    = 1;
 382         inet->mc_index  = 0;
 383         inet->mc_list   = NULL;
 384
 385         sk_refcnt_debug_inc(sk);
 386
 387         if (inet->inet_num) {
 388                 /* It assumes that any protocol which allows
 389                  * the user to assign a number at socket
 390                  * creation time automatically
 391                  * shares.
 392                  */
 393                 inet->inet_sport = htons(inet->inet_num);
 394                 /* Add to protocol hash chains. */
 395                 sk->sk_prot->hash(sk);
 396         }
 397
 398         if (sk->sk_prot->init) {
 399                 err = sk->sk_prot->init(sk);
 400                 if (err)
 401                         sk_common_release(sk);
 402         }
 403 out:
 404         return err;
 405 out_rcu_unlock:
 406         rcu_read_unlock();
 407         goto out;
 408 }
 409
 410
 411 /*
 412  *      The peer socket should always be NULL (or else). When we call this
 413  *      function we are destroying the object and from then on nobody
 414  *      should refer to it.
 415  */
 416 int inet_release(struct socket *sock)
 417 {
 418         struct sock *sk = sock->sk;
 419
 420         if (sk) {
 421                 long timeout;
 422
 423                 sock_rps_reset_flow(sk);
 424
 425                 /* Applications forget to leave groups before exiting */
 426                 ip_mc_drop_socket(sk);
 427
 428                 /* If linger is set, we don't return until the close
 429                  * is complete.  Otherwise we return immediately. The
 430                  * actually closing is done the same either way.
 431                  *
 432                  * If the close is due to the process exiting, we never
 433                  * linger..
 434                  */
 435                 timeout = 0;
 436                 if (sock_flag(sk, SOCK_LINGER) &&
 437                     !(current->flags & PF_EXITING))
 438                         timeout = sk->sk_lingertime;
 439                 sock->sk = NULL;
 440                 sk->sk_prot->close(sk, timeout);
 441         }
 442         return 0;
 443 }
 444 EXPORT_SYMBOL(inet_release);
 445
 446 /* It is off by default, see below. */
 447 int sysctl_ip_nonlocal_bind __read_mostly;
 448 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
 449
 450 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 451 {
 452         struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
 453         struct sock *sk = sock->sk;
 454         struct inet_sock *inet = inet_sk(sk);
 455         unsigned short snum;
 456         int chk_addr_ret;
 457         int err;
 458
 459         /* If the socket has its own bind function then use it. (RAW) */
 460         if (sk->sk_prot->bind) {
 461                 err = sk->sk_prot->bind(sk, uaddr, addr_len);
 462                 goto out;
 463         }
 464         err = -EINVAL;
 465         if (addr_len < sizeof(struct sockaddr_in))
 466                 goto out;
 467
 468         if (addr->sin_family != AF_INET) {
 469                 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
 470                  * only if s_addr is INADDR_ANY.
 471                  */
 472                 err = -EAFNOSUPPORT;
 473                 if (addr->sin_family != AF_UNSPEC ||
 474                     addr->sin_addr.s_addr != htonl(INADDR_ANY))
 475                         goto out;
 476         }
 477
 478         chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr);
 479
 480         /* Not specified by any standard per-se, however it breaks too
 481          * many applications when removed.  It is unfortunate since
 482          * allowing applications to make a non-local bind solves
 483          * several problems with systems using dynamic addressing.
 484          * (ie. your servers still start up even if your ISDN link
 485          *  is temporarily down)
 486          */
 487         err = -EADDRNOTAVAIL;
 488         if (!sysctl_ip_nonlocal_bind &&
 489             !(inet->freebind || inet->transparent) &&
 490             addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
 491             chk_addr_ret != RTN_LOCAL &&
 492             chk_addr_ret != RTN_MULTICAST &&
 493             chk_addr_ret != RTN_BROADCAST)
 494                 goto out;
 495
 496         snum = ntohs(addr->sin_port);
 497         err = -EACCES;
 498         if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
 499                 goto out;
 500
 501         /*      We keep a pair of addresses. rcv_saddr is the one
 502          *      used by hash lookups, and saddr is used for transmit.
 503          *
 504          *      In the BSD API these are the same except where it
 505          *      would be illegal to use them (multicast/broadcast) in
 506          *      which case the sending device address is used.
 507          */
 508         lock_sock(sk);
 509
 510         /* Check these errors (active socket, double bind). */
 511         err = -EINVAL;
 512         if (sk->sk_state != TCP_CLOSE || inet->inet_num)
 513                 goto out_release_sock;
 514
 515         inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
 516         if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
 517                 inet->inet_saddr = 0;  /* Use device */
 518
 519         /* Make sure we are allowed to bind here. */
 520         if (sk->sk_prot->get_port(sk, snum)) {
 521                 inet->inet_saddr = inet->inet_rcv_saddr = 0;
 522                 err = -EADDRINUSE;
 523                 goto out_release_sock;
 524         }
 525
 526         if (inet->inet_rcv_saddr)
 527                 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
 528         if (snum)
 529                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
 530         inet->inet_sport = htons(inet->inet_num);
 531         inet->inet_daddr = 0;
 532         inet->inet_dport = 0;
 533         sk_dst_reset(sk);
 534         err = 0;
 535 out_release_sock:
 536         release_sock(sk);
 537 out:
 538         return err;
 539 }
 540 EXPORT_SYMBOL(inet_bind);
 541
 542 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
 543                        int addr_len, int flags)
 544 {
 545         struct sock *sk = sock->sk;
 546
 547         if (addr_len < sizeof(uaddr->sa_family))
 548                 return -EINVAL;
 549         if (uaddr->sa_family == AF_UNSPEC)
 550                 return sk->sk_prot->disconnect(sk, flags);
 551
 552         if (!inet_sk(sk)->inet_num && inet_autobind(sk))
 553                 return -EAGAIN;
 554         return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
 555 }
 556 EXPORT_SYMBOL(inet_dgram_connect);
 557
 558 static long inet_wait_for_connect(struct sock *sk, long timeo)
 559 {
 560         DEFINE_WAIT(wait);
 561
 562         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
 563
 564         /* Basic assumption: if someone sets sk->sk_err, he _must_
 565          * change state of the socket from TCP_SYN_*.
 566          * Connect() does not allow to get error notifications
 567          * without closing the socket.
 568          */
 569         while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 570                 release_sock(sk);
 571                 timeo = schedule_timeout(timeo);
 572                 lock_sock(sk);
 573                 if (signal_pending(current) || !timeo)
 574                         break;
 575                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
 576         }
 577         finish_wait(sk_sleep(sk), &wait);
 578         return timeo;
 579 }
 580
 581 /*
 582  *      Connect to a remote host. There is regrettably still a little
 583  *      TCP 'magic' in here.
 584  */
 585 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 586                         int addr_len, int flags)
 587 {
 588         struct sock *sk = sock->sk;
 589         int err;
 590         long timeo;
 591
 592         if (addr_len < sizeof(uaddr->sa_family))
 593                 return -EINVAL;
 594
 595         lock_sock(sk);
 596
 597         if (uaddr->sa_family == AF_UNSPEC) {
 598                 err = sk->sk_prot->disconnect(sk, flags);
 599                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 600                 goto out;
 601         }
 602
 603         switch (sock->state) {
 604         default:
 605                 err = -EINVAL;
 606                 goto out;
 607         case SS_CONNECTED:
 608                 err = -EISCONN;
 609                 goto out;
 610         case SS_CONNECTING:
 611                 err = -EALREADY;
 612                 /* Fall out of switch with err, set for this state */
 613                 break;
 614         case SS_UNCONNECTED:
 615                 err = -EISCONN;
 616                 if (sk->sk_state != TCP_CLOSE)
 617                         goto out;
 618
 619                 err = sk->sk_prot->connect(sk, uaddr, addr_len);
 620                 if (err < 0)
 621                         goto out;
 622
 623                 sock->state = SS_CONNECTING;
 624
 625                 /* Just entered SS_CONNECTING state; the only
 626                  * difference is that return value in non-blocking
 627                  * case is EINPROGRESS, rather than EALREADY.
 628                  */
 629                 err = -EINPROGRESS;
 630                 break;
 631         }
 632
 633         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
 634
 635         if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 636                 /* Error code is set above */
 637                 if (!timeo || !inet_wait_for_connect(sk, timeo))
 638                         goto out;
 639
 640                 err = sock_intr_errno(timeo);
 641                 if (signal_pending(current))
 642                         goto out;
 643         }
 644
 645         /* Connection was closed by RST, timeout, ICMP error
 646          * or another process disconnected us.
 647          */
 648         if (sk->sk_state == TCP_CLOSE)
 649                 goto sock_error;
 650
 651         /* sk->sk_err may be not zero now, if RECVERR was ordered by user
 652          * and error was received after socket entered established state.
 653          * Hence, it is handled normally after connect() return successfully.
 654          */
 655
 656         sock->state = SS_CONNECTED;
 657         err = 0;
 658 out:
 659         release_sock(sk);
 660         return err;
 661
 662 sock_error:
 663         err = sock_error(sk) ? : -ECONNABORTED;
 664         sock->state = SS_UNCONNECTED;
 665         if (sk->sk_prot->disconnect(sk, flags))
 666                 sock->state = SS_DISCONNECTING;
 667         goto out;
 668 }
 669 EXPORT_SYMBOL(inet_stream_connect);
 670
 671 /*
 672  *      Accept a pending connection. The TCP layer now gives BSD semantics.
 673  */
 674
 675 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
 676 {
 677         struct sock *sk1 = sock->sk;
 678         int err = -EINVAL;
 679         struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
 680
 681         if (!sk2)
 682                 goto do_err;
 683
 684         lock_sock(sk2);
 685
 686         sock_rps_record_flow(sk2);
 687         WARN_ON(!((1 << sk2->sk_state) &
 688                   (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE)));
 689
 690         sock_graft(sk2, newsock);
 691
 692         newsock->state = SS_CONNECTED;
 693         err = 0;
 694         release_sock(sk2);
 695 do_err:
 696         return err;
 697 }
 698 EXPORT_SYMBOL(inet_accept);
 699
 700
 701 /*
 702  *      This does both peername and sockname.
 703  */
 704 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
 705                         int *uaddr_len, int peer)
 706 {
 707         struct sock *sk         = sock->sk;
 708         struct inet_sock *inet  = inet_sk(sk);
 709         DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
 710
 711         sin->sin_family = AF_INET;
 712         if (peer) {
 713                 if (!inet->inet_dport ||
 714                     (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
 715                      peer == 1))
 716                         return -ENOTCONN;
 717                 sin->sin_port = inet->inet_dport;
 718                 sin->sin_addr.s_addr = inet->inet_daddr;
 719         } else {
 720                 __be32 addr = inet->inet_rcv_saddr;
 721                 if (!addr)
 722                         addr = inet->inet_saddr;
 723                 sin->sin_port = inet->inet_sport;
 724                 sin->sin_addr.s_addr = addr;
 725         }
 726         memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 727         *uaddr_len = sizeof(*sin);
 728         return 0;
 729 }
 730 EXPORT_SYMBOL(inet_getname);
 731
 732 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
 733                  size_t size)
 734 {
 735         struct sock *sk = sock->sk;
 736
 737         sock_rps_record_flow(sk);
 738
 739         /* We may need to bind the socket. */
 740         if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
 741             inet_autobind(sk))
 742                 return -EAGAIN;
 743
 744         return sk->sk_prot->sendmsg(iocb, sk, msg, size);
 745 }
 746 EXPORT_SYMBOL(inet_sendmsg);
 747
 748 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
 749                       size_t size, int flags)
 750 {
 751         struct sock *sk = sock->sk;
 752
 753         sock_rps_record_flow(sk);
 754
 755         /* We may need to bind the socket. */
 756         if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
 757             inet_autobind(sk))
 758                 return -EAGAIN;
 759
 760         if (sk->sk_prot->sendpage)
 761                 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
 762         return sock_no_sendpage(sock, page, offset, size, flags);
 763 }
 764 EXPORT_SYMBOL(inet_sendpage);
 765
 766 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
 767                  size_t size, int flags)
 768 {
 769         struct sock *sk = sock->sk;
 770         int addr_len = 0;
 771         int err;
 772
 773         sock_rps_record_flow(sk);
 774
 775         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
 776                                    flags & ~MSG_DONTWAIT, &addr_len);
 777         if (err >= 0)
 778                 msg->msg_namelen = addr_len;
 779         return err;
 780 }
 781 EXPORT_SYMBOL(inet_recvmsg);
 782
 783 int inet_shutdown(struct socket *sock, int how)
 784 {
 785         struct sock *sk = sock->sk;
 786         int err = 0;
 787
 788         /* This should really check to make sure
 789          * the socket is a TCP socket. (WHY AC...)
 790          */
 791         how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
 792                        1->2 bit 2 snds.
 793                        2->3 */
 794         if ((how & ~SHUTDOWN_MASK) || !how)     /* MAXINT->0 */
 795                 return -EINVAL;
 796
 797         lock_sock(sk);
 798         if (sock->state == SS_CONNECTING) {
 799                 if ((1 << sk->sk_state) &
 800                     (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
 801                         sock->state = SS_DISCONNECTING;
 802                 else
 803                         sock->state = SS_CONNECTED;
 804         }
 805
 806         switch (sk->sk_state) {
 807         case TCP_CLOSE:
 808                 err = -ENOTCONN;
 809                 /* Hack to wake up other listeners, who can poll for
 810                    POLLHUP, even on eg. unconnected UDP sockets -- RR */
 811         default:
 812                 sk->sk_shutdown |= how;
 813                 if (sk->sk_prot->shutdown)
 814                         sk->sk_prot->shutdown(sk, how);
 815                 break;
 816
 817         /* Remaining two branches are temporary solution for missing
 818          * close() in multithreaded environment. It is _not_ a good idea,
 819          * but we have no choice until close() is repaired at VFS level.
 820          */
 821         case TCP_LISTEN:
 822                 if (!(how & RCV_SHUTDOWN))
 823                         break;
 824                 /* Fall through */
 825         case TCP_SYN_SENT:
 826                 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
 827                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 828                 break;
 829         }
 830
 831         /* Wake up anyone sleeping in poll. */
 832         sk->sk_state_change(sk);
 833         release_sock(sk);
 834         return err;
 835 }
 836 EXPORT_SYMBOL(inet_shutdown);
 837
 838 /*
 839  *      ioctl() calls you can issue on an INET socket. Most of these are
 840  *      device configuration and stuff and very rarely used. Some ioctls
 841  *      pass on to the socket itself.
 842  *
 843  *      NOTE: I like the idea of a module for the config stuff. ie ifconfig
 844  *      loads the devconfigure module does its configuring and unloads it.
 845  *      There's a good 20K of config code hanging around the kernel.
 846  */
 847
 848 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 849 {
 850         struct sock *sk = sock->sk;
 851         int err = 0;
 852         struct net *net = sock_net(sk);
 853
 854         switch (cmd) {
 855         case SIOCGSTAMP:
 856                 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
 857                 break;
 858         case SIOCGSTAMPNS:
 859                 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
 860                 break;
 861         case SIOCADDRT:
 862         case SIOCDELRT:
 863         case SIOCRTMSG:
 864                 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
 865                 break;
 866         case SIOCDARP:
 867         case SIOCGARP:
 868         case SIOCSARP:
 869                 err = arp_ioctl(net, cmd, (void __user *)arg);
 870                 break;
 871         case SIOCGIFADDR:
 872         case SIOCSIFADDR:
 873         case SIOCGIFBRDADDR:
 874         case SIOCSIFBRDADDR:
 875         case SIOCGIFNETMASK:
 876         case SIOCSIFNETMASK:
 877         case SIOCGIFDSTADDR:
 878         case SIOCSIFDSTADDR:
 879         case SIOCSIFPFLAGS:
 880         case SIOCGIFPFLAGS:
 881         case SIOCSIFFLAGS:
 882                 err = devinet_ioctl(net, cmd, (void __user *)arg);
 883                 break;
 884         default:
 885                 if (sk->sk_prot->ioctl)
 886                         err = sk->sk_prot->ioctl(sk, cmd, arg);
 887                 else
 888                         err = -ENOIOCTLCMD;
 889                 break;
 890         }
 891         return err;
 892 }
 893 EXPORT_SYMBOL(inet_ioctl);
 894
 895 #ifdef CONFIG_COMPAT
 896 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 897 {
 898         struct sock *sk = sock->sk;
 899         int err = -ENOIOCTLCMD;
 900
 901         if (sk->sk_prot->compat_ioctl)
 902                 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
 903
 904         return err;
 905 }
 906 #endif
 907
 908 const struct proto_ops inet_stream_ops = {
 909         .family            = PF_INET,
 910         .owner             = THIS_MODULE,
 911         .release           = inet_release,
 912         .bind              = inet_bind,
 913         .connect           = inet_stream_connect,
 914         .socketpair        = sock_no_socketpair,
 915         .accept            = inet_accept,
 916         .getname           = inet_getname,
 917         .poll              = tcp_poll,
 918         .ioctl             = inet_ioctl,
 919         .listen            = inet_listen,
 920         .shutdown          = inet_shutdown,
 921         .setsockopt        = sock_common_setsockopt,
 922         .getsockopt        = sock_common_getsockopt,
 923         .sendmsg           = inet_sendmsg,
 924         .recvmsg           = inet_recvmsg,
 925         .mmap              = sock_no_mmap,
 926         .sendpage          = inet_sendpage,
 927         .splice_read       = tcp_splice_read,
 928 #ifdef CONFIG_COMPAT
 929         .compat_setsockopt = compat_sock_common_setsockopt,
 930         .compat_getsockopt = compat_sock_common_getsockopt,
 931         .compat_ioctl      = inet_compat_ioctl,
 932 #endif
 933 };
 934 EXPORT_SYMBOL(inet_stream_ops);
 935
 936 const struct proto_ops inet_dgram_ops = {
 937         .family            = PF_INET,
 938         .owner             = THIS_MODULE,
 939         .release           = inet_release,
 940         .bind              = inet_bind,
 941         .connect           = inet_dgram_connect,
 942         .socketpair        = sock_no_socketpair,
 943         .accept            = sock_no_accept,
 944         .getname           = inet_getname,
 945         .poll              = udp_poll,
 946         .ioctl             = inet_ioctl,
 947         .listen            = sock_no_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 #ifdef CONFIG_COMPAT
 956         .compat_setsockopt = compat_sock_common_setsockopt,
 957         .compat_getsockopt = compat_sock_common_getsockopt,
 958         .compat_ioctl      = inet_compat_ioctl,
 959 #endif
 960 };
 961 EXPORT_SYMBOL(inet_dgram_ops);
 962
 963 /*
 964  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
 965  * udp_poll
 966  */
 967 static const struct proto_ops inet_sockraw_ops = {
 968         .family            = PF_INET,
 969         .owner             = THIS_MODULE,
 970         .release           = inet_release,
 971         .bind              = inet_bind,
 972         .connect           = inet_dgram_connect,
 973         .socketpair        = sock_no_socketpair,
 974         .accept            = sock_no_accept,
 975         .getname           = inet_getname,
 976         .poll              = datagram_poll,
 977         .ioctl             = inet_ioctl,
 978         .listen            = sock_no_listen,
 979         .shutdown          = inet_shutdown,
 980         .setsockopt        = sock_common_setsockopt,
 981         .getsockopt        = sock_common_getsockopt,
 982         .sendmsg           = inet_sendmsg,
 983         .recvmsg           = inet_recvmsg,
 984         .mmap              = sock_no_mmap,
 985         .sendpage          = inet_sendpage,
 986 #ifdef CONFIG_COMPAT
 987         .compat_setsockopt = compat_sock_common_setsockopt,
 988         .compat_getsockopt = compat_sock_common_getsockopt,
 989         .compat_ioctl      = inet_compat_ioctl,
 990 #endif
 991 };
 992
 993 static const struct net_proto_family inet_family_ops = {
 994         .family = PF_INET,
 995         .create = inet_create,
 996         .owner  = THIS_MODULE,
 997 };
 998
 999 /* Upon startup we insert all the elements in inetsw_array[] into
1000  * the linked list inetsw.
1001  */
1002 static struct inet_protosw inetsw_array[] =
1003 {
1004         {
1005                 .type =       SOCK_STREAM,
1006                 .protocol =   IPPROTO_TCP,
1007                 .prot =       &tcp_prot,
1008                 .ops =        &inet_stream_ops,
1009                 .no_check =   0,
1010                 .flags =      INET_PROTOSW_PERMANENT |
1011                               INET_PROTOSW_ICSK,
1012         },
1013
1014         {
1015                 .type =       SOCK_DGRAM,
1016                 .protocol =   IPPROTO_UDP,
1017                 .prot =       &udp_prot,
1018                 .ops =        &inet_dgram_ops,
1019                 .no_check =   UDP_CSUM_DEFAULT,
1020                 .flags =      INET_PROTOSW_PERMANENT,
1021        },
1022
1023        {
1024                 .type =       SOCK_DGRAM,
1025                 .protocol =   IPPROTO_ICMP,
1026                 .prot =       &ping_prot,
1027                 .ops =        &inet_dgram_ops,
1028                 .no_check =   UDP_CSUM_DEFAULT,
1029                 .flags =      INET_PROTOSW_REUSE,
1030        },
1031
1032        {
1033                .type =       SOCK_RAW,
1034                .protocol =   IPPROTO_IP,        /* wild card */
1035                .prot =       &raw_prot,
1036                .ops =        &inet_sockraw_ops,
1037                .no_check =   UDP_CSUM_DEFAULT,
1038                .flags =      INET_PROTOSW_REUSE,
1039        }
1040 };
1041
1042 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1043
1044 void inet_register_protosw(struct inet_protosw *p)
1045 {
1046         struct list_head *lh;
1047         struct inet_protosw *answer;
1048         int protocol = p->protocol;
1049         struct list_head *last_perm;
1050
1051         spin_lock_bh(&inetsw_lock);
1052
1053         if (p->type >= SOCK_MAX)
1054                 goto out_illegal;
1055
1056         /* If we are trying to override a permanent protocol, bail. */
1057         answer = NULL;
1058         last_perm = &inetsw[p->type];
1059         list_for_each(lh, &inetsw[p->type]) {
1060                 answer = list_entry(lh, struct inet_protosw, list);
1061
1062                 /* Check only the non-wild match. */
1063                 if (INET_PROTOSW_PERMANENT & answer->flags) {
1064                         if (protocol == answer->protocol)
1065                                 break;
1066                         last_perm = lh;
1067                 }
1068
1069                 answer = NULL;
1070         }
1071         if (answer)
1072                 goto out_permanent;
1073
1074         /* Add the new entry after the last permanent entry if any, so that
1075          * the new entry does not override a permanent entry when matched with
1076          * a wild-card protocol. But it is allowed to override any existing
1077          * non-permanent entry.  This means that when we remove this entry, the
1078          * system automatically returns to the old behavior.
1079          */
1080         list_add_rcu(&p->list, last_perm);
1081 out:
1082         spin_unlock_bh(&inetsw_lock);
1083
1084         return;
1085
1086 out_permanent:
1087         printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
1088                protocol);
1089         goto out;
1090
1091 out_illegal:
1092         printk(KERN_ERR
1093                "Ignoring attempt to register invalid socket type %d.\n",
1094                p->type);
1095         goto out;
1096 }
1097 EXPORT_SYMBOL(inet_register_protosw);
1098
1099 void inet_unregister_protosw(struct inet_protosw *p)
1100 {
1101         if (INET_PROTOSW_PERMANENT & p->flags) {
1102                 printk(KERN_ERR
1103                        "Attempt to unregister permanent protocol %d.\n",
1104                        p->protocol);
1105         } else {
1106                 spin_lock_bh(&inetsw_lock);
1107                 list_del_rcu(&p->list);
1108                 spin_unlock_bh(&inetsw_lock);
1109
1110                 synchronize_net();
1111         }
1112 }
1113 EXPORT_SYMBOL(inet_unregister_protosw);
1114
1115 /*
1116  *      Shall we try to damage output packets if routing dev changes?
1117  */
1118
1119 int sysctl_ip_dynaddr __read_mostly;
1120
1121 static int inet_sk_reselect_saddr(struct sock *sk)
1122 {
1123         struct inet_sock *inet = inet_sk(sk);
1124         __be32 old_saddr = inet->inet_saddr;
1125         __be32 daddr = inet->inet_daddr;
1126         struct flowi4 *fl4;
1127         struct rtable *rt;
1128         __be32 new_saddr;
1129         struct ip_options_rcu *inet_opt;
1130
1131         inet_opt = rcu_dereference_protected(inet->inet_opt,
1132                                              sock_owned_by_user(sk));
1133         if (inet_opt && inet_opt->opt.srr)
1134                 daddr = inet_opt->opt.faddr;
1135
1136         /* Query new route. */
1137         fl4 = &inet->cork.fl.u.ip4;
1138         rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1139                               sk->sk_bound_dev_if, sk->sk_protocol,
1140                               inet->inet_sport, inet->inet_dport, sk, false);
1141         if (IS_ERR(rt))
1142                 return PTR_ERR(rt);
1143
1144         sk_setup_caps(sk, &rt->dst);
1145
1146         new_saddr = fl4->saddr;
1147
1148         if (new_saddr == old_saddr)
1149                 return 0;
1150
1151         if (sysctl_ip_dynaddr > 1) {
1152                 printk(KERN_INFO "%s(): shifting inet->saddr from %pI4 to %pI4\n",
1153                        __func__, &old_saddr, &new_saddr);
1154         }
1155
1156         inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1157
1158         /*
1159          * XXX The only one ugly spot where we need to
1160          * XXX really change the sockets identity after
1161          * XXX it has entered the hashes. -DaveM
1162          *
1163          * Besides that, it does not check for connection
1164          * uniqueness. Wait for troubles.
1165          */
1166         __sk_prot_rehash(sk);
1167         return 0;
1168 }
1169
1170 int inet_sk_rebuild_header(struct sock *sk)
1171 {
1172         struct inet_sock *inet = inet_sk(sk);
1173         struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1174         __be32 daddr;
1175         struct ip_options_rcu *inet_opt;
1176         struct flowi4 *fl4;
1177         int err;
1178
1179         /* Route is OK, nothing to do. */
1180         if (rt)
1181                 return 0;
1182
1183         /* Reroute. */
1184         rcu_read_lock();
1185         inet_opt = rcu_dereference(inet->inet_opt);
1186         daddr = inet->inet_daddr;
1187         if (inet_opt && inet_opt->opt.srr)
1188                 daddr = inet_opt->opt.faddr;
1189         rcu_read_unlock();
1190         fl4 = &inet->cork.fl.u.ip4;
1191         rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1192                                    inet->inet_dport, inet->inet_sport,
1193                                    sk->sk_protocol, RT_CONN_FLAGS(sk),
1194                                    sk->sk_bound_dev_if);
1195         if (!IS_ERR(rt)) {
1196                 err = 0;
1197                 sk_setup_caps(sk, &rt->dst);
1198         } else {
1199                 err = PTR_ERR(rt);
1200
1201                 /* Routing failed... */
1202                 sk->sk_route_caps = 0;
1203                 /*
1204                  * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1205                  * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1206                  */
1207                 if (!sysctl_ip_dynaddr ||
1208                     sk->sk_state != TCP_SYN_SENT ||
1209                     (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1210                     (err = inet_sk_reselect_saddr(sk)) != 0)
1211                         sk->sk_err_soft = -err;
1212         }
1213
1214         return err;
1215 }
1216 EXPORT_SYMBOL(inet_sk_rebuild_header);
1217
1218 static int inet_gso_send_check(struct sk_buff *skb)
1219 {
1220         const struct iphdr *iph;
1221         const struct net_protocol *ops;
1222         int proto;
1223         int ihl;
1224         int err = -EINVAL;
1225
1226         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1227                 goto out;
1228
1229         iph = ip_hdr(skb);
1230         ihl = iph->ihl * 4;
1231         if (ihl < sizeof(*iph))
1232                 goto out;
1233
1234         if (unlikely(!pskb_may_pull(skb, ihl)))
1235                 goto out;
1236
1237         __skb_pull(skb, ihl);
1238         skb_reset_transport_header(skb);
1239         iph = ip_hdr(skb);
1240         proto = iph->protocol & (MAX_INET_PROTOS - 1);
1241         err = -EPROTONOSUPPORT;
1242
1243         rcu_read_lock();
1244         ops = rcu_dereference(inet_protos[proto]);
1245         if (likely(ops && ops->gso_send_check))
1246                 err = ops->gso_send_check(skb);
1247         rcu_read_unlock();
1248
1249 out:
1250         return err;
1251 }
1252
1253 static struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1254         netdev_features_t features)
1255 {
1256         struct sk_buff *segs = ERR_PTR(-EINVAL);
1257         struct iphdr *iph;
1258         const struct net_protocol *ops;
1259         int proto;
1260         int ihl;
1261         int id;
1262         unsigned int offset = 0;
1263
1264         if (!(features & NETIF_F_V4_CSUM))
1265                 features &= ~NETIF_F_SG;
1266
1267         if (unlikely(skb_shinfo(skb)->gso_type &
1268                      ~(SKB_GSO_TCPV4 |
1269                        SKB_GSO_UDP |
1270                        SKB_GSO_DODGY |
1271                        SKB_GSO_TCP_ECN |
1272                        0)))
1273                 goto out;
1274
1275         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1276                 goto out;
1277
1278         iph = ip_hdr(skb);
1279         ihl = iph->ihl * 4;
1280         if (ihl < sizeof(*iph))
1281                 goto out;
1282
1283         if (unlikely(!pskb_may_pull(skb, ihl)))
1284                 goto out;
1285
1286         __skb_pull(skb, ihl);
1287         skb_reset_transport_header(skb);
1288         iph = ip_hdr(skb);
1289         id = ntohs(iph->id);
1290         proto = iph->protocol & (MAX_INET_PROTOS - 1);
1291         segs = ERR_PTR(-EPROTONOSUPPORT);
1292
1293         rcu_read_lock();
1294         ops = rcu_dereference(inet_protos[proto]);
1295         if (likely(ops && ops->gso_segment))
1296                 segs = ops->gso_segment(skb, features);
1297         rcu_read_unlock();
1298
1299         if (!segs || IS_ERR(segs))
1300                 goto out;
1301
1302         skb = segs;
1303         do {
1304                 iph = ip_hdr(skb);
1305                 if (proto == IPPROTO_UDP) {
1306                         iph->id = htons(id);
1307                         iph->frag_off = htons(offset >> 3);
1308                         if (skb->next != NULL)
1309                                 iph->frag_off |= htons(IP_MF);
1310                         offset += (skb->len - skb->mac_len - iph->ihl * 4);
1311                 } else
1312                         iph->id = htons(id++);
1313                 iph->tot_len = htons(skb->len - skb->mac_len);
1314                 iph->check = 0;
1315                 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1316         } while ((skb = skb->next));
1317
1318 out:
1319         return segs;
1320 }
1321
1322 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1323                                          struct sk_buff *skb)
1324 {
1325         const struct net_protocol *ops;
1326         struct sk_buff **pp = NULL;
1327         struct sk_buff *p;
1328         const struct iphdr *iph;
1329         unsigned int hlen;
1330         unsigned int off;
1331         unsigned int id;
1332         int flush = 1;
1333         int proto;
1334
1335         off = skb_gro_offset(skb);
1336         hlen = off + sizeof(*iph);
1337         iph = skb_gro_header_fast(skb, off);
1338         if (skb_gro_header_hard(skb, hlen)) {
1339                 iph = skb_gro_header_slow(skb, hlen, off);
1340                 if (unlikely(!iph))
1341                         goto out;
1342         }
1343
1344         proto = iph->protocol & (MAX_INET_PROTOS - 1);
1345
1346         rcu_read_lock();
1347         ops = rcu_dereference(inet_protos[proto]);
1348         if (!ops || !ops->gro_receive)
1349                 goto out_unlock;
1350
1351         if (*(u8 *)iph != 0x45)
1352                 goto out_unlock;
1353
1354         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1355                 goto out_unlock;
1356
1357         id = ntohl(*(__be32 *)&iph->id);
1358         flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF));
1359         id >>= 16;
1360
1361         for (p = *head; p; p = p->next) {
1362                 struct iphdr *iph2;
1363
1364                 if (!NAPI_GRO_CB(p)->same_flow)
1365                         continue;
1366
1367                 iph2 = ip_hdr(p);
1368
1369                 if ((iph->protocol ^ iph2->protocol) |
1370                     (iph->tos ^ iph2->tos) |
1371                     ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1372                     ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1373                         NAPI_GRO_CB(p)->same_flow = 0;
1374                         continue;
1375                 }
1376
1377                 /* All fields must match except length and checksum. */
1378                 NAPI_GRO_CB(p)->flush |=
1379                         (iph->ttl ^ iph2->ttl) |
1380                         ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1381
1382                 NAPI_GRO_CB(p)->flush |= flush;
1383         }
1384
1385         NAPI_GRO_CB(skb)->flush |= flush;
1386         skb_gro_pull(skb, sizeof(*iph));
1387         skb_set_transport_header(skb, skb_gro_offset(skb));
1388
1389         pp = ops->gro_receive(head, skb);
1390
1391 out_unlock:
1392         rcu_read_unlock();
1393
1394 out:
1395         NAPI_GRO_CB(skb)->flush |= flush;
1396
1397         return pp;
1398 }
1399
1400 static int inet_gro_complete(struct sk_buff *skb)
1401 {
1402         const struct net_protocol *ops;
1403         struct iphdr *iph = ip_hdr(skb);
1404         int proto = iph->protocol & (MAX_INET_PROTOS - 1);
1405         int err = -ENOSYS;
1406         __be16 newlen = htons(skb->len - skb_network_offset(skb));
1407
1408         csum_replace2(&iph->check, iph->tot_len, newlen);
1409         iph->tot_len = newlen;
1410
1411         rcu_read_lock();
1412         ops = rcu_dereference(inet_protos[proto]);
1413         if (WARN_ON(!ops || !ops->gro_complete))
1414                 goto out_unlock;
1415
1416         err = ops->gro_complete(skb);
1417
1418 out_unlock:
1419         rcu_read_unlock();
1420
1421         return err;
1422 }
1423
1424 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1425                          unsigned short type, unsigned char protocol,
1426                          struct net *net)
1427 {
1428         struct socket *sock;
1429         int rc = sock_create_kern(family, type, protocol, &sock);
1430
1431         if (rc == 0) {
1432                 *sk = sock->sk;
1433                 (*sk)->sk_allocation = GFP_ATOMIC;
1434                 /*
1435                  * Unhash it so that IP input processing does not even see it,
1436                  * we do not wish this socket to see incoming packets.
1437                  */
1438                 (*sk)->sk_prot->unhash(*sk);
1439
1440                 sk_change_net(*sk, net);
1441         }
1442         return rc;
1443 }
1444 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1445
1446 unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1447 {
1448         unsigned long res = 0;
1449         int i, j;
1450
1451         for_each_possible_cpu(i) {
1452                 for (j = 0; j < SNMP_ARRAY_SZ; j++)
1453                         res += *(((unsigned long *) per_cpu_ptr(mib[j], i)) + offt);
1454         }
1455         return res;
1456 }
1457 EXPORT_SYMBOL_GPL(snmp_fold_field);
1458
1459 #if BITS_PER_LONG==32
1460
1461 u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset)
1462 {
1463         u64 res = 0;
1464         int cpu;
1465
1466         for_each_possible_cpu(cpu) {
1467                 void *bhptr;
1468                 struct u64_stats_sync *syncp;
1469                 u64 v;
1470                 unsigned int start;
1471
1472                 bhptr = per_cpu_ptr(mib[0], cpu);
1473                 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1474                 do {
1475                         start = u64_stats_fetch_begin_bh(syncp);
1476                         v = *(((u64 *) bhptr) + offt);
1477                 } while (u64_stats_fetch_retry_bh(syncp, start));
1478
1479                 res += v;
1480         }
1481         return res;
1482 }
1483 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1484 #endif
1485
1486 int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align)
1487 {
1488         BUG_ON(ptr == NULL);
1489         ptr[0] = __alloc_percpu(mibsize, align);
1490         if (!ptr[0])
1491                 return -ENOMEM;
1492 #if SNMP_ARRAY_SZ == 2
1493         ptr[1] = __alloc_percpu(mibsize, align);
1494         if (!ptr[1]) {
1495                 free_percpu(ptr[0]);
1496                 ptr[0] = NULL;
1497                 return -ENOMEM;
1498         }
1499 #endif
1500         return 0;
1501 }
1502 EXPORT_SYMBOL_GPL(snmp_mib_init);
1503
1504 void snmp_mib_free(void __percpu *ptr[SNMP_ARRAY_SZ])
1505 {
1506         int i;
1507
1508         BUG_ON(ptr == NULL);
1509         for (i = 0; i < SNMP_ARRAY_SZ; i++) {
1510                 free_percpu(ptr[i]);
1511                 ptr[i] = NULL;
1512         }
1513 }
1514 EXPORT_SYMBOL_GPL(snmp_mib_free);
1515
1516 #ifdef CONFIG_IP_MULTICAST
1517 static const struct net_protocol igmp_protocol = {
1518         .handler =      igmp_rcv,
1519         .netns_ok =     1,
1520 };
1521 #endif
1522
1523 static const struct net_protocol tcp_protocol = {
1524         .handler =      tcp_v4_rcv,
1525         .err_handler =  tcp_v4_err,
1526         .gso_send_check = tcp_v4_gso_send_check,
1527         .gso_segment =  tcp_tso_segment,
1528         .gro_receive =  tcp4_gro_receive,
1529         .gro_complete = tcp4_gro_complete,
1530         .no_policy =    1,
1531         .netns_ok =     1,
1532 };
1533
1534 static const struct net_protocol udp_protocol = {
1535         .handler =      udp_rcv,
1536         .err_handler =  udp_err,
1537         .gso_send_check = udp4_ufo_send_check,
1538         .gso_segment = udp4_ufo_fragment,
1539         .no_policy =    1,
1540         .netns_ok =     1,
1541 };
1542
1543 static const struct net_protocol icmp_protocol = {
1544         .handler =      icmp_rcv,
1545         .err_handler =  ping_err,
1546         .no_policy =    1,
1547         .netns_ok =     1,
1548 };
1549
1550 static __net_init int ipv4_mib_init_net(struct net *net)
1551 {
1552         if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1553                           sizeof(struct tcp_mib),
1554                           __alignof__(struct tcp_mib)) < 0)
1555                 goto err_tcp_mib;
1556         if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1557                           sizeof(struct ipstats_mib),
1558                           __alignof__(struct ipstats_mib)) < 0)
1559                 goto err_ip_mib;
1560         if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1561                           sizeof(struct linux_mib),
1562                           __alignof__(struct linux_mib)) < 0)
1563                 goto err_net_mib;
1564         if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1565                           sizeof(struct udp_mib),
1566                           __alignof__(struct udp_mib)) < 0)
1567                 goto err_udp_mib;
1568         if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1569                           sizeof(struct udp_mib),
1570                           __alignof__(struct udp_mib)) < 0)
1571                 goto err_udplite_mib;
1572         if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1573                           sizeof(struct icmp_mib),
1574                           __alignof__(struct icmp_mib)) < 0)
1575                 goto err_icmp_mib;
1576         net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1577                                               GFP_KERNEL);
1578         if (!net->mib.icmpmsg_statistics)
1579                 goto err_icmpmsg_mib;
1580
1581         tcp_mib_init(net);
1582         return 0;
1583
1584 err_icmpmsg_mib:
1585         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1586 err_icmp_mib:
1587         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1588 err_udplite_mib:
1589         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1590 err_udp_mib:
1591         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1592 err_net_mib:
1593         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1594 err_ip_mib:
1595         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1596 err_tcp_mib:
1597         return -ENOMEM;
1598 }
1599
1600 static __net_exit void ipv4_mib_exit_net(struct net *net)
1601 {
1602         kfree(net->mib.icmpmsg_statistics);
1603         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1604         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1605         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1606         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1607         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1608         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1609 }
1610
1611 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1612         .init = ipv4_mib_init_net,
1613         .exit = ipv4_mib_exit_net,
1614 };
1615
1616 static int __init init_ipv4_mibs(void)
1617 {
1618         return register_pernet_subsys(&ipv4_mib_ops);
1619 }
1620
1621 static int ipv4_proc_init(void);
1622
1623 /*
1624  *      IP protocol layer initialiser
1625  */
1626
1627 static struct packet_type ip_packet_type __read_mostly = {
1628         .type = cpu_to_be16(ETH_P_IP),
1629         .func = ip_rcv,
1630         .gso_send_check = inet_gso_send_check,
1631         .gso_segment = inet_gso_segment,
1632         .gro_receive = inet_gro_receive,
1633         .gro_complete = inet_gro_complete,
1634 };
1635
1636 static int __init inet_init(void)
1637 {
1638         struct sk_buff *dummy_skb;
1639         struct inet_protosw *q;
1640         struct list_head *r;
1641         int rc = -EINVAL;
1642
1643         BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1644
1645         sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL);
1646         if (!sysctl_local_reserved_ports)
1647                 goto out;
1648
1649         rc = proto_register(&tcp_prot, 1);
1650         if (rc)
1651                 goto out_free_reserved_ports;
1652
1653         rc = proto_register(&udp_prot, 1);
1654         if (rc)
1655                 goto out_unregister_tcp_proto;
1656
1657         rc = proto_register(&raw_prot, 1);
1658         if (rc)
1659                 goto out_unregister_udp_proto;
1660
1661         rc = proto_register(&ping_prot, 1);
1662         if (rc)
1663                 goto out_unregister_raw_proto;
1664
1665         /*
1666          *      Tell SOCKET that we are alive...
1667          */
1668
1669         (void)sock_register(&inet_family_ops);
1670
1671 #ifdef CONFIG_SYSCTL
1672         ip_static_sysctl_init();
1673 #endif
1674
1675         tcp_prot.sysctl_mem = init_net.ipv4.sysctl_tcp_mem;
1676
1677         /*
1678          *      Add all the base protocols.
1679          */
1680
1681         if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1682                 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1683         if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1684                 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1685         if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1686                 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1687 #ifdef CONFIG_IP_MULTICAST
1688         if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1689                 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1690 #endif
1691
1692         /* Register the socket-side information for inet_create. */
1693         for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1694                 INIT_LIST_HEAD(r);
1695
1696         for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1697                 inet_register_protosw(q);
1698
1699         /*
1700          *      Set the ARP module up
1701          */
1702
1703         arp_init();
1704
1705         /*
1706          *      Set the IP module up
1707          */
1708
1709         ip_init();
1710
1711         tcp_v4_init();
1712
1713         /* Setup TCP slab cache for open requests. */
1714         tcp_init();
1715
1716         /* Setup UDP memory threshold */
1717         udp_init();
1718
1719         /* Add UDP-Lite (RFC 3828) */
1720         udplite4_register();
1721
1722         ping_init();
1723
1724         /*
1725          *      Set the ICMP layer up
1726          */
1727
1728         if (icmp_init() < 0)
1729                 panic("Failed to create the ICMP control socket.\n");
1730
1731         /*
1732          *      Initialise the multicast router
1733          */
1734 #if defined(CONFIG_IP_MROUTE)
1735         if (ip_mr_init())
1736                 printk(KERN_CRIT "inet_init: Cannot init ipv4 mroute\n");
1737 #endif
1738         /*
1739          *      Initialise per-cpu ipv4 mibs
1740          */
1741
1742         if (init_ipv4_mibs())
1743                 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n");
1744
1745         ipv4_proc_init();
1746
1747         ipfrag_init();
1748
1749         dev_add_pack(&ip_packet_type);
1750
1751         rc = 0;
1752 out:
1753         return rc;
1754 out_unregister_raw_proto:
1755         proto_unregister(&raw_prot);
1756 out_unregister_udp_proto:
1757         proto_unregister(&udp_prot);
1758 out_unregister_tcp_proto:
1759         proto_unregister(&tcp_prot);
1760 out_free_reserved_ports:
1761         kfree(sysctl_local_reserved_ports);
1762         goto out;
1763 }
1764
1765 fs_initcall(inet_init);
1766
1767 /* ------------------------------------------------------------------------ */
1768
1769 #ifdef CONFIG_PROC_FS
1770 static int __init ipv4_proc_init(void)
1771 {
1772         int rc = 0;
1773
1774         if (raw_proc_init())
1775                 goto out_raw;
1776         if (tcp4_proc_init())
1777                 goto out_tcp;
1778         if (udp4_proc_init())
1779                 goto out_udp;
1780         if (ping_proc_init())
1781                 goto out_ping;
1782         if (ip_misc_proc_init())
1783                 goto out_misc;
1784 out:
1785         return rc;
1786 out_misc:
1787         ping_proc_exit();
1788 out_ping:
1789         udp4_proc_exit();
1790 out_udp:
1791         tcp4_proc_exit();
1792 out_tcp:
1793         raw_proc_exit();
1794 out_raw:
1795         rc = -ENOMEM;
1796         goto out;
1797 }
1798
1799 #else /* CONFIG_PROC_FS */
1800 static int __init ipv4_proc_init(void)
1801 {
1802         return 0;
1803 }
1804 #endif /* CONFIG_PROC_FS */
1805
1806 MODULE_ALIAS_NETPROTO(PF_INET);
1807