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                 goto out;
 470
 471         chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr);
 472
 473         /* Not specified by any standard per-se, however it breaks too
 474          * many applications when removed.  It is unfortunate since
 475          * allowing applications to make a non-local bind solves
 476          * several problems with systems using dynamic addressing.
 477          * (ie. your servers still start up even if your ISDN link
 478          *  is temporarily down)
 479          */
 480         err = -EADDRNOTAVAIL;
 481         if (!sysctl_ip_nonlocal_bind &&
 482             !(inet->freebind || inet->transparent) &&
 483             addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
 484             chk_addr_ret != RTN_LOCAL &&
 485             chk_addr_ret != RTN_MULTICAST &&
 486             chk_addr_ret != RTN_BROADCAST)
 487                 goto out;
 488
 489         snum = ntohs(addr->sin_port);
 490         err = -EACCES;
 491         if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
 492                 goto out;
 493
 494         /*      We keep a pair of addresses. rcv_saddr is the one
 495          *      used by hash lookups, and saddr is used for transmit.
 496          *
 497          *      In the BSD API these are the same except where it
 498          *      would be illegal to use them (multicast/broadcast) in
 499          *      which case the sending device address is used.
 500          */
 501         lock_sock(sk);
 502
 503         /* Check these errors (active socket, double bind). */
 504         err = -EINVAL;
 505         if (sk->sk_state != TCP_CLOSE || inet->inet_num)
 506                 goto out_release_sock;
 507
 508         inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
 509         if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
 510                 inet->inet_saddr = 0;  /* Use device */
 511
 512         /* Make sure we are allowed to bind here. */
 513         if (sk->sk_prot->get_port(sk, snum)) {
 514                 inet->inet_saddr = inet->inet_rcv_saddr = 0;
 515                 err = -EADDRINUSE;
 516                 goto out_release_sock;
 517         }
 518
 519         if (inet->inet_rcv_saddr)
 520                 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
 521         if (snum)
 522                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
 523         inet->inet_sport = htons(inet->inet_num);
 524         inet->inet_daddr = 0;
 525         inet->inet_dport = 0;
 526         sk_dst_reset(sk);
 527         err = 0;
 528 out_release_sock:
 529         release_sock(sk);
 530 out:
 531         return err;
 532 }
 533 EXPORT_SYMBOL(inet_bind);
 534
 535 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
 536                        int addr_len, int flags)
 537 {
 538         struct sock *sk = sock->sk;
 539
 540         if (addr_len < sizeof(uaddr->sa_family))
 541                 return -EINVAL;
 542         if (uaddr->sa_family == AF_UNSPEC)
 543                 return sk->sk_prot->disconnect(sk, flags);
 544
 545         if (!inet_sk(sk)->inet_num && inet_autobind(sk))
 546                 return -EAGAIN;
 547         return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
 548 }
 549 EXPORT_SYMBOL(inet_dgram_connect);
 550
 551 static long inet_wait_for_connect(struct sock *sk, long timeo)
 552 {
 553         DEFINE_WAIT(wait);
 554
 555         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
 556
 557         /* Basic assumption: if someone sets sk->sk_err, he _must_
 558          * change state of the socket from TCP_SYN_*.
 559          * Connect() does not allow to get error notifications
 560          * without closing the socket.
 561          */
 562         while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 563                 release_sock(sk);
 564                 timeo = schedule_timeout(timeo);
 565                 lock_sock(sk);
 566                 if (signal_pending(current) || !timeo)
 567                         break;
 568                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
 569         }
 570         finish_wait(sk_sleep(sk), &wait);
 571         return timeo;
 572 }
 573
 574 /*
 575  *      Connect to a remote host. There is regrettably still a little
 576  *      TCP 'magic' in here.
 577  */
 578 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 579                         int addr_len, int flags)
 580 {
 581         struct sock *sk = sock->sk;
 582         int err;
 583         long timeo;
 584
 585         if (addr_len < sizeof(uaddr->sa_family))
 586                 return -EINVAL;
 587
 588         lock_sock(sk);
 589
 590         if (uaddr->sa_family == AF_UNSPEC) {
 591                 err = sk->sk_prot->disconnect(sk, flags);
 592                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 593                 goto out;
 594         }
 595
 596         switch (sock->state) {
 597         default:
 598                 err = -EINVAL;
 599                 goto out;
 600         case SS_CONNECTED:
 601                 err = -EISCONN;
 602                 goto out;
 603         case SS_CONNECTING:
 604                 err = -EALREADY;
 605                 /* Fall out of switch with err, set for this state */
 606                 break;
 607         case SS_UNCONNECTED:
 608                 err = -EISCONN;
 609                 if (sk->sk_state != TCP_CLOSE)
 610                         goto out;
 611
 612                 err = sk->sk_prot->connect(sk, uaddr, addr_len);
 613                 if (err < 0)
 614                         goto out;
 615
 616                 sock->state = SS_CONNECTING;
 617
 618                 /* Just entered SS_CONNECTING state; the only
 619                  * difference is that return value in non-blocking
 620                  * case is EINPROGRESS, rather than EALREADY.
 621                  */
 622                 err = -EINPROGRESS;
 623                 break;
 624         }
 625
 626         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
 627
 628         if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 629                 /* Error code is set above */
 630                 if (!timeo || !inet_wait_for_connect(sk, timeo))
 631                         goto out;
 632
 633                 err = sock_intr_errno(timeo);
 634                 if (signal_pending(current))
 635                         goto out;
 636         }
 637
 638         /* Connection was closed by RST, timeout, ICMP error
 639          * or another process disconnected us.
 640          */
 641         if (sk->sk_state == TCP_CLOSE)
 642                 goto sock_error;
 643
 644         /* sk->sk_err may be not zero now, if RECVERR was ordered by user
 645          * and error was received after socket entered established state.
 646          * Hence, it is handled normally after connect() return successfully.
 647          */
 648
 649         sock->state = SS_CONNECTED;
 650         err = 0;
 651 out:
 652         release_sock(sk);
 653         return err;
 654
 655 sock_error:
 656         err = sock_error(sk) ? : -ECONNABORTED;
 657         sock->state = SS_UNCONNECTED;
 658         if (sk->sk_prot->disconnect(sk, flags))
 659                 sock->state = SS_DISCONNECTING;
 660         goto out;
 661 }
 662 EXPORT_SYMBOL(inet_stream_connect);
 663
 664 /*
 665  *      Accept a pending connection. The TCP layer now gives BSD semantics.
 666  */
 667
 668 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
 669 {
 670         struct sock *sk1 = sock->sk;
 671         int err = -EINVAL;
 672         struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
 673
 674         if (!sk2)
 675                 goto do_err;
 676
 677         lock_sock(sk2);
 678
 679         WARN_ON(!((1 << sk2->sk_state) &
 680                   (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE)));
 681
 682         sock_graft(sk2, newsock);
 683
 684         newsock->state = SS_CONNECTED;
 685         err = 0;
 686         release_sock(sk2);
 687 do_err:
 688         return err;
 689 }
 690 EXPORT_SYMBOL(inet_accept);
 691
 692
 693 /*
 694  *      This does both peername and sockname.
 695  */
 696 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
 697                         int *uaddr_len, int peer)
 698 {
 699         struct sock *sk         = sock->sk;
 700         struct inet_sock *inet  = inet_sk(sk);
 701         DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
 702
 703         sin->sin_family = AF_INET;
 704         if (peer) {
 705                 if (!inet->inet_dport ||
 706                     (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
 707                      peer == 1))
 708                         return -ENOTCONN;
 709                 sin->sin_port = inet->inet_dport;
 710                 sin->sin_addr.s_addr = inet->inet_daddr;
 711         } else {
 712                 __be32 addr = inet->inet_rcv_saddr;
 713                 if (!addr)
 714                         addr = inet->inet_saddr;
 715                 sin->sin_port = inet->inet_sport;
 716                 sin->sin_addr.s_addr = addr;
 717         }
 718         memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 719         *uaddr_len = sizeof(*sin);
 720         return 0;
 721 }
 722 EXPORT_SYMBOL(inet_getname);
 723
 724 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
 725                  size_t size)
 726 {
 727         struct sock *sk = sock->sk;
 728
 729         sock_rps_record_flow(sk);
 730
 731         /* We may need to bind the socket. */
 732         if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
 733             inet_autobind(sk))
 734                 return -EAGAIN;
 735
 736         return sk->sk_prot->sendmsg(iocb, sk, msg, size);
 737 }
 738 EXPORT_SYMBOL(inet_sendmsg);
 739
 740 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
 741                       size_t size, int flags)
 742 {
 743         struct sock *sk = sock->sk;
 744
 745         sock_rps_record_flow(sk);
 746
 747         /* We may need to bind the socket. */
 748         if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
 749             inet_autobind(sk))
 750                 return -EAGAIN;
 751
 752         if (sk->sk_prot->sendpage)
 753                 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
 754         return sock_no_sendpage(sock, page, offset, size, flags);
 755 }
 756 EXPORT_SYMBOL(inet_sendpage);
 757
 758 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
 759                  size_t size, int flags)
 760 {
 761         struct sock *sk = sock->sk;
 762         int addr_len = 0;
 763         int err;
 764
 765         sock_rps_record_flow(sk);
 766
 767         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
 768                                    flags & ~MSG_DONTWAIT, &addr_len);
 769         if (err >= 0)
 770                 msg->msg_namelen = addr_len;
 771         return err;
 772 }
 773 EXPORT_SYMBOL(inet_recvmsg);
 774
 775 int inet_shutdown(struct socket *sock, int how)
 776 {
 777         struct sock *sk = sock->sk;
 778         int err = 0;
 779
 780         /* This should really check to make sure
 781          * the socket is a TCP socket. (WHY AC...)
 782          */
 783         how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
 784                        1->2 bit 2 snds.
 785                        2->3 */
 786         if ((how & ~SHUTDOWN_MASK) || !how)     /* MAXINT->0 */
 787                 return -EINVAL;
 788
 789         lock_sock(sk);
 790         if (sock->state == SS_CONNECTING) {
 791                 if ((1 << sk->sk_state) &
 792                     (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
 793                         sock->state = SS_DISCONNECTING;
 794                 else
 795                         sock->state = SS_CONNECTED;
 796         }
 797
 798         switch (sk->sk_state) {
 799         case TCP_CLOSE:
 800                 err = -ENOTCONN;
 801                 /* Hack to wake up other listeners, who can poll for
 802                    POLLHUP, even on eg. unconnected UDP sockets -- RR */
 803         default:
 804                 sk->sk_shutdown |= how;
 805                 if (sk->sk_prot->shutdown)
 806                         sk->sk_prot->shutdown(sk, how);
 807                 break;
 808
 809         /* Remaining two branches are temporary solution for missing
 810          * close() in multithreaded environment. It is _not_ a good idea,
 811          * but we have no choice until close() is repaired at VFS level.
 812          */
 813         case TCP_LISTEN:
 814                 if (!(how & RCV_SHUTDOWN))
 815                         break;
 816                 /* Fall through */
 817         case TCP_SYN_SENT:
 818                 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
 819                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 820                 break;
 821         }
 822
 823         /* Wake up anyone sleeping in poll. */
 824         sk->sk_state_change(sk);
 825         release_sock(sk);
 826         return err;
 827 }
 828 EXPORT_SYMBOL(inet_shutdown);
 829
 830 /*
 831  *      ioctl() calls you can issue on an INET socket. Most of these are
 832  *      device configuration and stuff and very rarely used. Some ioctls
 833  *      pass on to the socket itself.
 834  *
 835  *      NOTE: I like the idea of a module for the config stuff. ie ifconfig
 836  *      loads the devconfigure module does its configuring and unloads it.
 837  *      There's a good 20K of config code hanging around the kernel.
 838  */
 839
 840 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 841 {
 842         struct sock *sk = sock->sk;
 843         int err = 0;
 844         struct net *net = sock_net(sk);
 845
 846         switch (cmd) {
 847         case SIOCGSTAMP:
 848                 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
 849                 break;
 850         case SIOCGSTAMPNS:
 851                 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
 852                 break;
 853         case SIOCADDRT:
 854         case SIOCDELRT:
 855         case SIOCRTMSG:
 856                 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
 857                 break;
 858         case SIOCDARP:
 859         case SIOCGARP:
 860         case SIOCSARP:
 861                 err = arp_ioctl(net, cmd, (void __user *)arg);
 862                 break;
 863         case SIOCGIFADDR:
 864         case SIOCSIFADDR:
 865         case SIOCGIFBRDADDR:
 866         case SIOCSIFBRDADDR:
 867         case SIOCGIFNETMASK:
 868         case SIOCSIFNETMASK:
 869         case SIOCGIFDSTADDR:
 870         case SIOCSIFDSTADDR:
 871         case SIOCSIFPFLAGS:
 872         case SIOCGIFPFLAGS:
 873         case SIOCSIFFLAGS:
 874                 err = devinet_ioctl(net, cmd, (void __user *)arg);
 875                 break;
 876         default:
 877                 if (sk->sk_prot->ioctl)
 878                         err = sk->sk_prot->ioctl(sk, cmd, arg);
 879                 else
 880                         err = -ENOIOCTLCMD;
 881                 break;
 882         }
 883         return err;
 884 }
 885 EXPORT_SYMBOL(inet_ioctl);
 886
 887 #ifdef CONFIG_COMPAT
 888 int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 889 {
 890         struct sock *sk = sock->sk;
 891         int err = -ENOIOCTLCMD;
 892
 893         if (sk->sk_prot->compat_ioctl)
 894                 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
 895
 896         return err;
 897 }
 898 #endif
 899
 900 const struct proto_ops inet_stream_ops = {
 901         .family            = PF_INET,
 902         .owner             = THIS_MODULE,
 903         .release           = inet_release,
 904         .bind              = inet_bind,
 905         .connect           = inet_stream_connect,
 906         .socketpair        = sock_no_socketpair,
 907         .accept            = inet_accept,
 908         .getname           = inet_getname,
 909         .poll              = tcp_poll,
 910         .ioctl             = inet_ioctl,
 911         .listen            = inet_listen,
 912         .shutdown          = inet_shutdown,
 913         .setsockopt        = sock_common_setsockopt,
 914         .getsockopt        = sock_common_getsockopt,
 915         .sendmsg           = inet_sendmsg,
 916         .recvmsg           = inet_recvmsg,
 917         .mmap              = sock_no_mmap,
 918         .sendpage          = inet_sendpage,
 919         .splice_read       = tcp_splice_read,
 920 #ifdef CONFIG_COMPAT
 921         .compat_setsockopt = compat_sock_common_setsockopt,
 922         .compat_getsockopt = compat_sock_common_getsockopt,
 923         .compat_ioctl      = inet_compat_ioctl,
 924 #endif
 925 };
 926 EXPORT_SYMBOL(inet_stream_ops);
 927
 928 const struct proto_ops inet_dgram_ops = {
 929         .family            = PF_INET,
 930         .owner             = THIS_MODULE,
 931         .release           = inet_release,
 932         .bind              = inet_bind,
 933         .connect           = inet_dgram_connect,
 934         .socketpair        = sock_no_socketpair,
 935         .accept            = sock_no_accept,
 936         .getname           = inet_getname,
 937         .poll              = udp_poll,
 938         .ioctl             = inet_ioctl,
 939         .listen            = sock_no_listen,
 940         .shutdown          = inet_shutdown,
 941         .setsockopt        = sock_common_setsockopt,
 942         .getsockopt        = sock_common_getsockopt,
 943         .sendmsg           = inet_sendmsg,
 944         .recvmsg           = inet_recvmsg,
 945         .mmap              = sock_no_mmap,
 946         .sendpage          = inet_sendpage,
 947 #ifdef CONFIG_COMPAT
 948         .compat_setsockopt = compat_sock_common_setsockopt,
 949         .compat_getsockopt = compat_sock_common_getsockopt,
 950         .compat_ioctl      = inet_compat_ioctl,
 951 #endif
 952 };
 953 EXPORT_SYMBOL(inet_dgram_ops);
 954
 955 /*
 956  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
 957  * udp_poll
 958  */
 959 static const struct proto_ops inet_sockraw_ops = {
 960         .family            = PF_INET,
 961         .owner             = THIS_MODULE,
 962         .release           = inet_release,
 963         .bind              = inet_bind,
 964         .connect           = inet_dgram_connect,
 965         .socketpair        = sock_no_socketpair,
 966         .accept            = sock_no_accept,
 967         .getname           = inet_getname,
 968         .poll              = datagram_poll,
 969         .ioctl             = inet_ioctl,
 970         .listen            = sock_no_listen,
 971         .shutdown          = inet_shutdown,
 972         .setsockopt        = sock_common_setsockopt,
 973         .getsockopt        = sock_common_getsockopt,
 974         .sendmsg           = inet_sendmsg,
 975         .recvmsg           = inet_recvmsg,
 976         .mmap              = sock_no_mmap,
 977         .sendpage          = inet_sendpage,
 978 #ifdef CONFIG_COMPAT
 979         .compat_setsockopt = compat_sock_common_setsockopt,
 980         .compat_getsockopt = compat_sock_common_getsockopt,
 981         .compat_ioctl      = inet_compat_ioctl,
 982 #endif
 983 };
 984
 985 static const struct net_proto_family inet_family_ops = {
 986         .family = PF_INET,
 987         .create = inet_create,
 988         .owner  = THIS_MODULE,
 989 };
 990
 991 /* Upon startup we insert all the elements in inetsw_array[] into
 992  * the linked list inetsw.
 993  */
 994 static struct inet_protosw inetsw_array[] =
 995 {
 996         {
 997                 .type =       SOCK_STREAM,
 998                 .protocol =   IPPROTO_TCP,
 999                 .prot =       &tcp_prot,
1000                 .ops =        &inet_stream_ops,
1001                 .no_check =   0,
1002                 .flags =      INET_PROTOSW_PERMANENT |
1003                               INET_PROTOSW_ICSK,
1004         },
1005
1006         {
1007                 .type =       SOCK_DGRAM,
1008                 .protocol =   IPPROTO_UDP,
1009                 .prot =       &udp_prot,
1010                 .ops =        &inet_dgram_ops,
1011                 .no_check =   UDP_CSUM_DEFAULT,
1012                 .flags =      INET_PROTOSW_PERMANENT,
1013        },
1014
1015        {
1016                 .type =       SOCK_DGRAM,
1017                 .protocol =   IPPROTO_ICMP,
1018                 .prot =       &ping_prot,
1019                 .ops =        &inet_dgram_ops,
1020                 .no_check =   UDP_CSUM_DEFAULT,
1021                 .flags =      INET_PROTOSW_REUSE,
1022        },
1023
1024        {
1025                .type =       SOCK_RAW,
1026                .protocol =   IPPROTO_IP,        /* wild card */
1027                .prot =       &raw_prot,
1028                .ops =        &inet_sockraw_ops,
1029                .no_check =   UDP_CSUM_DEFAULT,
1030                .flags =      INET_PROTOSW_REUSE,
1031        }
1032 };
1033
1034 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1035
1036 void inet_register_protosw(struct inet_protosw *p)
1037 {
1038         struct list_head *lh;
1039         struct inet_protosw *answer;
1040         int protocol = p->protocol;
1041         struct list_head *last_perm;
1042
1043         spin_lock_bh(&inetsw_lock);
1044
1045         if (p->type >= SOCK_MAX)
1046                 goto out_illegal;
1047
1048         /* If we are trying to override a permanent protocol, bail. */
1049         answer = NULL;
1050         last_perm = &inetsw[p->type];
1051         list_for_each(lh, &inetsw[p->type]) {
1052                 answer = list_entry(lh, struct inet_protosw, list);
1053
1054                 /* Check only the non-wild match. */
1055                 if (INET_PROTOSW_PERMANENT & answer->flags) {
1056                         if (protocol == answer->protocol)
1057                                 break;
1058                         last_perm = lh;
1059                 }
1060
1061                 answer = NULL;
1062         }
1063         if (answer)
1064                 goto out_permanent;
1065
1066         /* Add the new entry after the last permanent entry if any, so that
1067          * the new entry does not override a permanent entry when matched with
1068          * a wild-card protocol. But it is allowed to override any existing
1069          * non-permanent entry.  This means that when we remove this entry, the
1070          * system automatically returns to the old behavior.
1071          */
1072         list_add_rcu(&p->list, last_perm);
1073 out:
1074         spin_unlock_bh(&inetsw_lock);
1075
1076         return;
1077
1078 out_permanent:
1079         printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
1080                protocol);
1081         goto out;
1082
1083 out_illegal:
1084         printk(KERN_ERR
1085                "Ignoring attempt to register invalid socket type %d.\n",
1086                p->type);
1087         goto out;
1088 }
1089 EXPORT_SYMBOL(inet_register_protosw);
1090
1091 void inet_unregister_protosw(struct inet_protosw *p)
1092 {
1093         if (INET_PROTOSW_PERMANENT & p->flags) {
1094                 printk(KERN_ERR
1095                        "Attempt to unregister permanent protocol %d.\n",
1096                        p->protocol);
1097         } else {
1098                 spin_lock_bh(&inetsw_lock);
1099                 list_del_rcu(&p->list);
1100                 spin_unlock_bh(&inetsw_lock);
1101
1102                 synchronize_net();
1103         }
1104 }
1105 EXPORT_SYMBOL(inet_unregister_protosw);
1106
1107 /*
1108  *      Shall we try to damage output packets if routing dev changes?
1109  */
1110
1111 int sysctl_ip_dynaddr __read_mostly;
1112
1113 static int inet_sk_reselect_saddr(struct sock *sk)
1114 {
1115         struct inet_sock *inet = inet_sk(sk);
1116         __be32 old_saddr = inet->inet_saddr;
1117         __be32 daddr = inet->inet_daddr;
1118         struct flowi4 *fl4;
1119         struct rtable *rt;
1120         __be32 new_saddr;
1121         struct ip_options_rcu *inet_opt;
1122
1123         inet_opt = rcu_dereference_protected(inet->inet_opt,
1124                                              sock_owned_by_user(sk));
1125         if (inet_opt && inet_opt->opt.srr)
1126                 daddr = inet_opt->opt.faddr;
1127
1128         /* Query new route. */
1129         fl4 = &inet->cork.fl.u.ip4;
1130         rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1131                               sk->sk_bound_dev_if, sk->sk_protocol,
1132                               inet->inet_sport, inet->inet_dport, sk, false);
1133         if (IS_ERR(rt))
1134                 return PTR_ERR(rt);
1135
1136         sk_setup_caps(sk, &rt->dst);
1137
1138         new_saddr = fl4->saddr;
1139
1140         if (new_saddr == old_saddr)
1141                 return 0;
1142
1143         if (sysctl_ip_dynaddr > 1) {
1144                 printk(KERN_INFO "%s(): shifting inet->saddr from %pI4 to %pI4\n",
1145                        __func__, &old_saddr, &new_saddr);
1146         }
1147
1148         inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1149
1150         /*
1151          * XXX The only one ugly spot where we need to
1152          * XXX really change the sockets identity after
1153          * XXX it has entered the hashes. -DaveM
1154          *
1155          * Besides that, it does not check for connection
1156          * uniqueness. Wait for troubles.
1157          */
1158         __sk_prot_rehash(sk);
1159         return 0;
1160 }
1161
1162 int inet_sk_rebuild_header(struct sock *sk)
1163 {
1164         struct inet_sock *inet = inet_sk(sk);
1165         struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1166         __be32 daddr;
1167         struct ip_options_rcu *inet_opt;
1168         struct flowi4 *fl4;
1169         int err;
1170
1171         /* Route is OK, nothing to do. */
1172         if (rt)
1173                 return 0;
1174
1175         /* Reroute. */
1176         rcu_read_lock();
1177         inet_opt = rcu_dereference(inet->inet_opt);
1178         daddr = inet->inet_daddr;
1179         if (inet_opt && inet_opt->opt.srr)
1180                 daddr = inet_opt->opt.faddr;
1181         rcu_read_unlock();
1182         fl4 = &inet->cork.fl.u.ip4;
1183         rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1184                                    inet->inet_dport, inet->inet_sport,
1185                                    sk->sk_protocol, RT_CONN_FLAGS(sk),
1186                                    sk->sk_bound_dev_if);
1187         if (!IS_ERR(rt)) {
1188                 err = 0;
1189                 sk_setup_caps(sk, &rt->dst);
1190         } else {
1191                 err = PTR_ERR(rt);
1192
1193                 /* Routing failed... */
1194                 sk->sk_route_caps = 0;
1195                 /*
1196                  * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1197                  * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1198                  */
1199                 if (!sysctl_ip_dynaddr ||
1200                     sk->sk_state != TCP_SYN_SENT ||
1201                     (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1202                     (err = inet_sk_reselect_saddr(sk)) != 0)
1203                         sk->sk_err_soft = -err;
1204         }
1205
1206         return err;
1207 }
1208 EXPORT_SYMBOL(inet_sk_rebuild_header);
1209
1210 static int inet_gso_send_check(struct sk_buff *skb)
1211 {
1212         const struct iphdr *iph;
1213         const struct net_protocol *ops;
1214         int proto;
1215         int ihl;
1216         int err = -EINVAL;
1217
1218         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1219                 goto out;
1220
1221         iph = ip_hdr(skb);
1222         ihl = iph->ihl * 4;
1223         if (ihl < sizeof(*iph))
1224                 goto out;
1225
1226         if (unlikely(!pskb_may_pull(skb, ihl)))
1227                 goto out;
1228
1229         __skb_pull(skb, ihl);
1230         skb_reset_transport_header(skb);
1231         iph = ip_hdr(skb);
1232         proto = iph->protocol & (MAX_INET_PROTOS - 1);
1233         err = -EPROTONOSUPPORT;
1234
1235         rcu_read_lock();
1236         ops = rcu_dereference(inet_protos[proto]);
1237         if (likely(ops && ops->gso_send_check))
1238                 err = ops->gso_send_check(skb);
1239         rcu_read_unlock();
1240
1241 out:
1242         return err;
1243 }
1244
1245 static struct sk_buff *inet_gso_segment(struct sk_buff *skb, u32 features)
1246 {
1247         struct sk_buff *segs = ERR_PTR(-EINVAL);
1248         struct iphdr *iph;
1249         const struct net_protocol *ops;
1250         int proto;
1251         int ihl;
1252         int id;
1253         unsigned int offset = 0;
1254
1255         if (!(features & NETIF_F_V4_CSUM))
1256                 features &= ~NETIF_F_SG;
1257
1258         if (unlikely(skb_shinfo(skb)->gso_type &
1259                      ~(SKB_GSO_TCPV4 |
1260                        SKB_GSO_UDP |
1261                        SKB_GSO_DODGY |
1262                        SKB_GSO_TCP_ECN |
1263                        0)))
1264                 goto out;
1265
1266         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1267                 goto out;
1268
1269         iph = ip_hdr(skb);
1270         ihl = iph->ihl * 4;
1271         if (ihl < sizeof(*iph))
1272                 goto out;
1273
1274         if (unlikely(!pskb_may_pull(skb, ihl)))
1275                 goto out;
1276
1277         __skb_pull(skb, ihl);
1278         skb_reset_transport_header(skb);
1279         iph = ip_hdr(skb);
1280         id = ntohs(iph->id);
1281         proto = iph->protocol & (MAX_INET_PROTOS - 1);
1282         segs = ERR_PTR(-EPROTONOSUPPORT);
1283
1284         rcu_read_lock();
1285         ops = rcu_dereference(inet_protos[proto]);
1286         if (likely(ops && ops->gso_segment))
1287                 segs = ops->gso_segment(skb, features);
1288         rcu_read_unlock();
1289
1290         if (!segs || IS_ERR(segs))
1291                 goto out;
1292
1293         skb = segs;
1294         do {
1295                 iph = ip_hdr(skb);
1296                 if (proto == IPPROTO_UDP) {
1297                         iph->id = htons(id);
1298                         iph->frag_off = htons(offset >> 3);
1299                         if (skb->next != NULL)
1300                                 iph->frag_off |= htons(IP_MF);
1301                         offset += (skb->len - skb->mac_len - iph->ihl * 4);
1302                 } else
1303                         iph->id = htons(id++);
1304                 iph->tot_len = htons(skb->len - skb->mac_len);
1305                 iph->check = 0;
1306                 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1307         } while ((skb = skb->next));
1308
1309 out:
1310         return segs;
1311 }
1312
1313 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1314                                          struct sk_buff *skb)
1315 {
1316         const struct net_protocol *ops;
1317         struct sk_buff **pp = NULL;
1318         struct sk_buff *p;
1319         const struct iphdr *iph;
1320         unsigned int hlen;
1321         unsigned int off;
1322         unsigned int id;
1323         int flush = 1;
1324         int proto;
1325
1326         off = skb_gro_offset(skb);
1327         hlen = off + sizeof(*iph);
1328         iph = skb_gro_header_fast(skb, off);
1329         if (skb_gro_header_hard(skb, hlen)) {
1330                 iph = skb_gro_header_slow(skb, hlen, off);
1331                 if (unlikely(!iph))
1332                         goto out;
1333         }
1334
1335         proto = iph->protocol & (MAX_INET_PROTOS - 1);
1336
1337         rcu_read_lock();
1338         ops = rcu_dereference(inet_protos[proto]);
1339         if (!ops || !ops->gro_receive)
1340                 goto out_unlock;
1341
1342         if (*(u8 *)iph != 0x45)
1343                 goto out_unlock;
1344
1345         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1346                 goto out_unlock;
1347
1348         id = ntohl(*(__be32 *)&iph->id);
1349         flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF));
1350         id >>= 16;
1351
1352         for (p = *head; p; p = p->next) {
1353                 struct iphdr *iph2;
1354
1355                 if (!NAPI_GRO_CB(p)->same_flow)
1356                         continue;
1357
1358                 iph2 = ip_hdr(p);
1359
1360                 if ((iph->protocol ^ iph2->protocol) |
1361                     (iph->tos ^ iph2->tos) |
1362                     ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1363                     ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1364                         NAPI_GRO_CB(p)->same_flow = 0;
1365                         continue;
1366                 }
1367
1368                 /* All fields must match except length and checksum. */
1369                 NAPI_GRO_CB(p)->flush |=
1370                         (iph->ttl ^ iph2->ttl) |
1371                         ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1372
1373                 NAPI_GRO_CB(p)->flush |= flush;
1374         }
1375
1376         NAPI_GRO_CB(skb)->flush |= flush;
1377         skb_gro_pull(skb, sizeof(*iph));
1378         skb_set_transport_header(skb, skb_gro_offset(skb));
1379
1380         pp = ops->gro_receive(head, skb);
1381
1382 out_unlock:
1383         rcu_read_unlock();
1384
1385 out:
1386         NAPI_GRO_CB(skb)->flush |= flush;
1387
1388         return pp;
1389 }
1390
1391 static int inet_gro_complete(struct sk_buff *skb)
1392 {
1393         const struct net_protocol *ops;
1394         struct iphdr *iph = ip_hdr(skb);
1395         int proto = iph->protocol & (MAX_INET_PROTOS - 1);
1396         int err = -ENOSYS;
1397         __be16 newlen = htons(skb->len - skb_network_offset(skb));
1398
1399         csum_replace2(&iph->check, iph->tot_len, newlen);
1400         iph->tot_len = newlen;
1401
1402         rcu_read_lock();
1403         ops = rcu_dereference(inet_protos[proto]);
1404         if (WARN_ON(!ops || !ops->gro_complete))
1405                 goto out_unlock;
1406
1407         err = ops->gro_complete(skb);
1408
1409 out_unlock:
1410         rcu_read_unlock();
1411
1412         return err;
1413 }
1414
1415 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1416                          unsigned short type, unsigned char protocol,
1417                          struct net *net)
1418 {
1419         struct socket *sock;
1420         int rc = sock_create_kern(family, type, protocol, &sock);
1421
1422         if (rc == 0) {
1423                 *sk = sock->sk;
1424                 (*sk)->sk_allocation = GFP_ATOMIC;
1425                 /*
1426                  * Unhash it so that IP input processing does not even see it,
1427                  * we do not wish this socket to see incoming packets.
1428                  */
1429                 (*sk)->sk_prot->unhash(*sk);
1430
1431                 sk_change_net(*sk, net);
1432         }
1433         return rc;
1434 }
1435 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1436
1437 unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1438 {
1439         unsigned long res = 0;
1440         int i;
1441
1442         for_each_possible_cpu(i) {
1443                 res += *(((unsigned long *) per_cpu_ptr(mib[0], i)) + offt);
1444                 res += *(((unsigned long *) per_cpu_ptr(mib[1], i)) + offt);
1445         }
1446         return res;
1447 }
1448 EXPORT_SYMBOL_GPL(snmp_fold_field);
1449
1450 #if BITS_PER_LONG==32
1451
1452 u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset)
1453 {
1454         u64 res = 0;
1455         int cpu;
1456
1457         for_each_possible_cpu(cpu) {
1458                 void *bhptr, *userptr;
1459                 struct u64_stats_sync *syncp;
1460                 u64 v_bh, v_user;
1461                 unsigned int start;
1462
1463                 /* first mib used by softirq context, we must use _bh() accessors */
1464                 bhptr = per_cpu_ptr(SNMP_STAT_BHPTR(mib), cpu);
1465                 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1466                 do {
1467                         start = u64_stats_fetch_begin_bh(syncp);
1468                         v_bh = *(((u64 *) bhptr) + offt);
1469                 } while (u64_stats_fetch_retry_bh(syncp, start));
1470
1471                 /* second mib used in USER context */
1472                 userptr = per_cpu_ptr(SNMP_STAT_USRPTR(mib), cpu);
1473                 syncp = (struct u64_stats_sync *)(userptr + syncp_offset);
1474                 do {
1475                         start = u64_stats_fetch_begin(syncp);
1476                         v_user = *(((u64 *) userptr) + offt);
1477                 } while (u64_stats_fetch_retry(syncp, start));
1478
1479                 res += v_bh + v_user;
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                 goto err0;
1492         ptr[1] = __alloc_percpu(mibsize, align);
1493         if (!ptr[1])
1494                 goto err1;
1495         return 0;
1496 err1:
1497         free_percpu(ptr[0]);
1498         ptr[0] = NULL;
1499 err0:
1500         return -ENOMEM;
1501 }
1502 EXPORT_SYMBOL_GPL(snmp_mib_init);
1503
1504 void snmp_mib_free(void __percpu *ptr[2])
1505 {
1506         BUG_ON(ptr == NULL);
1507         free_percpu(ptr[0]);
1508         free_percpu(ptr[1]);
1509         ptr[0] = ptr[1] = NULL;
1510 }
1511 EXPORT_SYMBOL_GPL(snmp_mib_free);
1512
1513 #ifdef CONFIG_IP_MULTICAST
1514 static const struct net_protocol igmp_protocol = {
1515         .handler =      igmp_rcv,
1516         .netns_ok =     1,
1517 };
1518 #endif
1519
1520 static const struct net_protocol tcp_protocol = {
1521         .handler =      tcp_v4_rcv,
1522         .err_handler =  tcp_v4_err,
1523         .gso_send_check = tcp_v4_gso_send_check,
1524         .gso_segment =  tcp_tso_segment,
1525         .gro_receive =  tcp4_gro_receive,
1526         .gro_complete = tcp4_gro_complete,
1527         .no_policy =    1,
1528         .netns_ok =     1,
1529 };
1530
1531 static const struct net_protocol udp_protocol = {
1532         .handler =      udp_rcv,
1533         .err_handler =  udp_err,
1534         .gso_send_check = udp4_ufo_send_check,
1535         .gso_segment = udp4_ufo_fragment,
1536         .no_policy =    1,
1537         .netns_ok =     1,
1538 };
1539
1540 static const struct net_protocol icmp_protocol = {
1541         .handler =      icmp_rcv,
1542         .err_handler =  ping_err,
1543         .no_policy =    1,
1544         .netns_ok =     1,
1545 };
1546
1547 static __net_init int ipv4_mib_init_net(struct net *net)
1548 {
1549         if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1550                           sizeof(struct tcp_mib),
1551                           __alignof__(struct tcp_mib)) < 0)
1552                 goto err_tcp_mib;
1553         if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1554                           sizeof(struct ipstats_mib),
1555                           __alignof__(struct ipstats_mib)) < 0)
1556                 goto err_ip_mib;
1557         if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1558                           sizeof(struct linux_mib),
1559                           __alignof__(struct linux_mib)) < 0)
1560                 goto err_net_mib;
1561         if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1562                           sizeof(struct udp_mib),
1563                           __alignof__(struct udp_mib)) < 0)
1564                 goto err_udp_mib;
1565         if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1566                           sizeof(struct udp_mib),
1567                           __alignof__(struct udp_mib)) < 0)
1568                 goto err_udplite_mib;
1569         if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1570                           sizeof(struct icmp_mib),
1571                           __alignof__(struct icmp_mib)) < 0)
1572                 goto err_icmp_mib;
1573         if (snmp_mib_init((void __percpu **)net->mib.icmpmsg_statistics,
1574                           sizeof(struct icmpmsg_mib),
1575                           __alignof__(struct icmpmsg_mib)) < 0)
1576                 goto err_icmpmsg_mib;
1577
1578         tcp_mib_init(net);
1579         return 0;
1580
1581 err_icmpmsg_mib:
1582         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1583 err_icmp_mib:
1584         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1585 err_udplite_mib:
1586         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1587 err_udp_mib:
1588         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1589 err_net_mib:
1590         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1591 err_ip_mib:
1592         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1593 err_tcp_mib:
1594         return -ENOMEM;
1595 }
1596
1597 static __net_exit void ipv4_mib_exit_net(struct net *net)
1598 {
1599         snmp_mib_free((void __percpu **)net->mib.icmpmsg_statistics);
1600         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1601         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1602         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1603         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1604         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1605         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1606 }
1607
1608 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1609         .init = ipv4_mib_init_net,
1610         .exit = ipv4_mib_exit_net,
1611 };
1612
1613 static int __init init_ipv4_mibs(void)
1614 {
1615         return register_pernet_subsys(&ipv4_mib_ops);
1616 }
1617
1618 static int ipv4_proc_init(void);
1619
1620 /*
1621  *      IP protocol layer initialiser
1622  */
1623
1624 static struct packet_type ip_packet_type __read_mostly = {
1625         .type = cpu_to_be16(ETH_P_IP),
1626         .func = ip_rcv,
1627         .gso_send_check = inet_gso_send_check,
1628         .gso_segment = inet_gso_segment,
1629         .gro_receive = inet_gro_receive,
1630         .gro_complete = inet_gro_complete,
1631 };
1632
1633 static int __init inet_init(void)
1634 {
1635         struct sk_buff *dummy_skb;
1636         struct inet_protosw *q;
1637         struct list_head *r;
1638         int rc = -EINVAL;
1639
1640         BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1641
1642         sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL);
1643         if (!sysctl_local_reserved_ports)
1644                 goto out;
1645
1646         rc = proto_register(&tcp_prot, 1);
1647         if (rc)
1648                 goto out_free_reserved_ports;
1649
1650         rc = proto_register(&udp_prot, 1);
1651         if (rc)
1652                 goto out_unregister_tcp_proto;
1653
1654         rc = proto_register(&raw_prot, 1);
1655         if (rc)
1656                 goto out_unregister_udp_proto;
1657
1658         rc = proto_register(&ping_prot, 1);
1659         if (rc)
1660                 goto out_unregister_raw_proto;
1661
1662         /*
1663          *      Tell SOCKET that we are alive...
1664          */
1665
1666         (void)sock_register(&inet_family_ops);
1667
1668 #ifdef CONFIG_SYSCTL
1669         ip_static_sysctl_init();
1670 #endif
1671
1672         /*
1673          *      Add all the base protocols.
1674          */
1675
1676         if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1677                 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1678         if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1679                 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1680         if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1681                 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1682 #ifdef CONFIG_IP_MULTICAST
1683         if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1684                 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1685 #endif
1686
1687         /* Register the socket-side information for inet_create. */
1688         for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1689                 INIT_LIST_HEAD(r);
1690
1691         for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1692                 inet_register_protosw(q);
1693
1694         /*
1695          *      Set the ARP module up
1696          */
1697
1698         arp_init();
1699
1700         /*
1701          *      Set the IP module up
1702          */
1703
1704         ip_init();
1705
1706         tcp_v4_init();
1707
1708         /* Setup TCP slab cache for open requests. */
1709         tcp_init();
1710
1711         /* Setup UDP memory threshold */
1712         udp_init();
1713
1714         /* Add UDP-Lite (RFC 3828) */
1715         udplite4_register();
1716
1717         ping_init();
1718
1719         /*
1720          *      Set the ICMP layer up
1721          */
1722
1723         if (icmp_init() < 0)
1724                 panic("Failed to create the ICMP control socket.\n");
1725
1726         /*
1727          *      Initialise the multicast router
1728          */
1729 #if defined(CONFIG_IP_MROUTE)
1730         if (ip_mr_init())
1731                 printk(KERN_CRIT "inet_init: Cannot init ipv4 mroute\n");
1732 #endif
1733         /*
1734          *      Initialise per-cpu ipv4 mibs
1735          */
1736
1737         if (init_ipv4_mibs())
1738                 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n");
1739
1740         ipv4_proc_init();
1741
1742         ipfrag_init();
1743
1744         dev_add_pack(&ip_packet_type);
1745
1746         rc = 0;
1747 out:
1748         return rc;
1749 out_unregister_raw_proto:
1750         proto_unregister(&raw_prot);
1751 out_unregister_udp_proto:
1752         proto_unregister(&udp_prot);
1753 out_unregister_tcp_proto:
1754         proto_unregister(&tcp_prot);
1755 out_free_reserved_ports:
1756         kfree(sysctl_local_reserved_ports);
1757         goto out;
1758 }
1759
1760 fs_initcall(inet_init);
1761
1762 /* ------------------------------------------------------------------------ */
1763
1764 #ifdef CONFIG_PROC_FS
1765 static int __init ipv4_proc_init(void)
1766 {
1767         int rc = 0;
1768
1769         if (raw_proc_init())
1770                 goto out_raw;
1771         if (tcp4_proc_init())
1772                 goto out_tcp;
1773         if (udp4_proc_init())
1774                 goto out_udp;
1775         if (ping_proc_init())
1776                 goto out_ping;
1777         if (ip_misc_proc_init())
1778                 goto out_misc;
1779 out:
1780         return rc;
1781 out_misc:
1782         ping_proc_exit();
1783 out_ping:
1784         udp4_proc_exit();
1785 out_udp:
1786         tcp4_proc_exit();
1787 out_tcp:
1788         raw_proc_exit();
1789 out_raw:
1790         rc = -ENOMEM;
1791         goto out;
1792 }
1793
1794 #else /* CONFIG_PROC_FS */
1795 static int __init ipv4_proc_init(void)
1796 {
1797         return 0;
1798 }
1799 #endif /* CONFIG_PROC_FS */
1800
1801 MODULE_ALIAS_NETPROTO(PF_INET);
1802