SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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
  90 #include <asm/uaccess.h>
  91 #include <asm/system.h>
  92
  93 #include <linux/inet.h>
  94 #include <linux/igmp.h>
  95 #include <linux/inetdevice.h>
  96 #include <linux/netdevice.h>
  97 #include <net/checksum.h>
  98 #include <net/ip.h>
  99 #include <net/protocol.h>
 100 #include <net/arp.h>
 101 #include <net/route.h>
 102 #include <net/ip_fib.h>
 103 #include <net/inet_connection_sock.h>
 104 #include <net/tcp.h>
 105 #include <net/udp.h>
 106 #include <net/udplite.h>
 107 #include <linux/skbuff.h>
 108 #include <net/sock.h>
 109 #include <net/raw.h>
 110 #include <net/icmp.h>
 111 #include <net/ipip.h>
 112 #include <net/inet_common.h>
 113 #include <net/xfrm.h>
 114 #include <net/net_namespace.h>
 115 #ifdef CONFIG_IP_MROUTE
 116 #include <linux/mroute.h>
 117 #endif
 118
 119
 120 /* The inetsw table contains everything that inet_create needs to
 121  * build a new socket.
 122  */
 123 static struct list_head inetsw[SOCK_MAX];
 124 static DEFINE_SPINLOCK(inetsw_lock);
 125
 126 struct ipv4_config ipv4_config;
 127 EXPORT_SYMBOL(ipv4_config);
 128
 129 /* New destruction routine */
 130
 131 void inet_sock_destruct(struct sock *sk)
 132 {
 133         struct inet_sock *inet = inet_sk(sk);
 134
 135         __skb_queue_purge(&sk->sk_receive_queue);
 136         __skb_queue_purge(&sk->sk_error_queue);
 137
 138         sk_mem_reclaim(sk);
 139
 140         if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
 141                 pr_err("Attempt to release TCP socket in state %d %p\n",
 142                        sk->sk_state, sk);
 143                 return;
 144         }
 145         if (!sock_flag(sk, SOCK_DEAD)) {
 146                 pr_err("Attempt to release alive inet socket %p\n", sk);
 147                 return;
 148         }
 149
 150         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
 151         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
 152         WARN_ON(sk->sk_wmem_queued);
 153         WARN_ON(sk->sk_forward_alloc);
 154
 155         kfree(inet->opt);
 156         dst_release(sk->sk_dst_cache);
 157         sk_refcnt_debug_dec(sk);
 158 }
 159 EXPORT_SYMBOL(inet_sock_destruct);
 160
 161 /*
 162  *      The routines beyond this point handle the behaviour of an AF_INET
 163  *      socket object. Mostly it punts to the subprotocols of IP to do
 164  *      the work.
 165  */
 166
 167 /*
 168  *      Automatically bind an unbound socket.
 169  */
 170
 171 static int inet_autobind(struct sock *sk)
 172 {
 173         struct inet_sock *inet;
 174         /* We may need to bind the socket. */
 175         lock_sock(sk);
 176         inet = inet_sk(sk);
 177         if (!inet->inet_num) {
 178                 if (sk->sk_prot->get_port(sk, 0)) {
 179                         release_sock(sk);
 180                         return -EAGAIN;
 181                 }
 182                 inet->inet_sport = htons(inet->inet_num);
 183         }
 184         release_sock(sk);
 185         return 0;
 186 }
 187
 188 /*
 189  *      Move a socket into listening state.
 190  */
 191 int inet_listen(struct socket *sock, int backlog)
 192 {
 193         struct sock *sk = sock->sk;
 194         unsigned char old_state;
 195         int err;
 196
 197         lock_sock(sk);
 198
 199         err = -EINVAL;
 200         if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
 201                 goto out;
 202
 203         old_state = sk->sk_state;
 204         if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
 205                 goto out;
 206
 207         /* Really, if the socket is already in listen state
 208          * we can only allow the backlog to be adjusted.
 209          */
 210         if (old_state != TCP_LISTEN) {
 211                 err = inet_csk_listen_start(sk, backlog);
 212                 if (err)
 213                         goto out;
 214         }
 215         sk->sk_max_ack_backlog = backlog;
 216         err = 0;
 217
 218 out:
 219         release_sock(sk);
 220         return err;
 221 }
 222 EXPORT_SYMBOL(inet_listen);
 223
 224 u32 inet_ehash_secret __read_mostly;
 225 EXPORT_SYMBOL(inet_ehash_secret);
 226
 227 /*
 228  * inet_ehash_secret must be set exactly once
 229  * Instead of using a dedicated spinlock, we (ab)use inetsw_lock
 230  */
 231 void build_ehash_secret(void)
 232 {
 233         u32 rnd;
 234         do {
 235                 get_random_bytes(&rnd, sizeof(rnd));
 236         } while (rnd == 0);
 237         spin_lock_bh(&inetsw_lock);
 238         if (!inet_ehash_secret)
 239                 inet_ehash_secret = rnd;
 240         spin_unlock_bh(&inetsw_lock);
 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         if (SOCK_RAW == sock->type) {
 358                 inet->inet_num = protocol;
 359                 if (IPPROTO_RAW == protocol)
 360                         inet->hdrincl = 1;
 361         }
 362
 363         if (ipv4_config.no_pmtu_disc)
 364                 inet->pmtudisc = IP_PMTUDISC_DONT;
 365         else
 366                 inet->pmtudisc = IP_PMTUDISC_WANT;
 367
 368         inet->inet_id = 0;
 369
 370         sock_init_data(sock, sk);
 371
 372         sk->sk_destruct    = inet_sock_destruct;
 373         sk->sk_protocol    = protocol;
 374         sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
 375
 376         inet->uc_ttl    = -1;
 377         inet->mc_loop   = 1;
 378         inet->mc_ttl    = 1;
 379         inet->mc_all    = 1;
 380         inet->mc_index  = 0;
 381         inet->mc_list   = NULL;
 382
 383         sk_refcnt_debug_inc(sk);
 384
 385         if (inet->inet_num) {
 386                 /* It assumes that any protocol which allows
 387                  * the user to assign a number at socket
 388                  * creation time automatically
 389                  * shares.
 390                  */
 391                 inet->inet_sport = htons(inet->inet_num);
 392                 /* Add to protocol hash chains. */
 393                 sk->sk_prot->hash(sk);
 394         }
 395
 396         if (sk->sk_prot->init) {
 397                 err = sk->sk_prot->init(sk);
 398                 if (err)
 399                         sk_common_release(sk);
 400         }
 401 out:
 402         return err;
 403 out_rcu_unlock:
 404         rcu_read_unlock();
 405         goto out;
 406 }
 407
 408
 409 /*
 410  *      The peer socket should always be NULL (or else). When we call this
 411  *      function we are destroying the object and from then on nobody
 412  *      should refer to it.
 413  */
 414 int inet_release(struct socket *sock)
 415 {
 416         struct sock *sk = sock->sk;
 417
 418         if (sk) {
 419                 long timeout;
 420
 421                 /* Applications forget to leave groups before exiting */
 422                 ip_mc_drop_socket(sk);
 423
 424                 /* If linger is set, we don't return until the close
 425                  * is complete.  Otherwise we return immediately. The
 426                  * actually closing is done the same either way.
 427                  *
 428                  * If the close is due to the process exiting, we never
 429                  * linger..
 430                  */
 431                 timeout = 0;
 432                 if (sock_flag(sk, SOCK_LINGER) &&
 433                     !(current->flags & PF_EXITING))
 434                         timeout = sk->sk_lingertime;
 435                 sock->sk = NULL;
 436                 sk->sk_prot->close(sk, timeout);
 437         }
 438         return 0;
 439 }
 440 EXPORT_SYMBOL(inet_release);
 441
 442 /* It is off by default, see below. */
 443 int sysctl_ip_nonlocal_bind __read_mostly;
 444 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
 445
 446 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 447 {
 448         struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
 449         struct sock *sk = sock->sk;
 450         struct inet_sock *inet = inet_sk(sk);
 451         unsigned short snum;
 452         int chk_addr_ret;
 453         int err;
 454
 455         /* If the socket has its own bind function then use it. (RAW) */
 456         if (sk->sk_prot->bind) {
 457                 err = sk->sk_prot->bind(sk, uaddr, addr_len);
 458                 goto out;
 459         }
 460         err = -EINVAL;
 461         if (addr_len < sizeof(struct sockaddr_in))
 462                 goto out;
 463
 464         chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr);
 465
 466         /* Not specified by any standard per-se, however it breaks too
 467          * many applications when removed.  It is unfortunate since
 468          * allowing applications to make a non-local bind solves
 469          * several problems with systems using dynamic addressing.
 470          * (ie. your servers still start up even if your ISDN link
 471          *  is temporarily down)
 472          */
 473         err = -EADDRNOTAVAIL;
 474         if (!sysctl_ip_nonlocal_bind &&
 475             !(inet->freebind || inet->transparent) &&
 476             addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
 477             chk_addr_ret != RTN_LOCAL &&
 478             chk_addr_ret != RTN_MULTICAST &&
 479             chk_addr_ret != RTN_BROADCAST)
 480                 goto out;
 481
 482         snum = ntohs(addr->sin_port);
 483         err = -EACCES;
 484         if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
 485                 goto out;
 486
 487         /*      We keep a pair of addresses. rcv_saddr is the one
 488          *      used by hash lookups, and saddr is used for transmit.
 489          *
 490          *      In the BSD API these are the same except where it
 491          *      would be illegal to use them (multicast/broadcast) in
 492          *      which case the sending device address is used.
 493          */
 494         lock_sock(sk);
 495
 496         /* Check these errors (active socket, double bind). */
 497         err = -EINVAL;
 498         if (sk->sk_state != TCP_CLOSE || inet->inet_num)
 499                 goto out_release_sock;
 500
 501         inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
 502         if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
 503                 inet->inet_saddr = 0;  /* Use device */
 504
 505         /* Make sure we are allowed to bind here. */
 506         if (sk->sk_prot->get_port(sk, snum)) {
 507                 inet->inet_saddr = inet->inet_rcv_saddr = 0;
 508                 err = -EADDRINUSE;
 509                 goto out_release_sock;
 510         }
 511
 512         if (inet->inet_rcv_saddr)
 513                 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
 514         if (snum)
 515                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
 516         inet->inet_sport = htons(inet->inet_num);
 517         inet->inet_daddr = 0;
 518         inet->inet_dport = 0;
 519         sk_dst_reset(sk);
 520         err = 0;
 521 out_release_sock:
 522         release_sock(sk);
 523 out:
 524         return err;
 525 }
 526 EXPORT_SYMBOL(inet_bind);
 527
 528 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
 529                        int addr_len, int flags)
 530 {
 531         struct sock *sk = sock->sk;
 532
 533         if (addr_len < sizeof(uaddr->sa_family))
 534                 return -EINVAL;
 535         if (uaddr->sa_family == AF_UNSPEC)
 536                 return sk->sk_prot->disconnect(sk, flags);
 537
 538         if (!inet_sk(sk)->inet_num && inet_autobind(sk))
 539                 return -EAGAIN;
 540         return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
 541 }
 542 EXPORT_SYMBOL(inet_dgram_connect);
 543
 544 static long inet_wait_for_connect(struct sock *sk, long timeo)
 545 {
 546         DEFINE_WAIT(wait);
 547
 548         prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 549
 550         /* Basic assumption: if someone sets sk->sk_err, he _must_
 551          * change state of the socket from TCP_SYN_*.
 552          * Connect() does not allow to get error notifications
 553          * without closing the socket.
 554          */
 555         while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 556                 release_sock(sk);
 557                 timeo = schedule_timeout(timeo);
 558                 lock_sock(sk);
 559                 if (signal_pending(current) || !timeo)
 560                         break;
 561                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 562         }
 563         finish_wait(sk->sk_sleep, &wait);
 564         return timeo;
 565 }
 566
 567 /*
 568  *      Connect to a remote host. There is regrettably still a little
 569  *      TCP 'magic' in here.
 570  */
 571 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 572                         int addr_len, int flags)
 573 {
 574         struct sock *sk = sock->sk;
 575         int err;
 576         long timeo;
 577
 578         if (addr_len < sizeof(uaddr->sa_family))
 579                 return -EINVAL;
 580
 581         lock_sock(sk);
 582
 583         if (uaddr->sa_family == AF_UNSPEC) {
 584                 err = sk->sk_prot->disconnect(sk, flags);
 585                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 586                 goto out;
 587         }
 588
 589         switch (sock->state) {
 590         default:
 591                 err = -EINVAL;
 592                 goto out;
 593         case SS_CONNECTED:
 594                 err = -EISCONN;
 595                 goto out;
 596         case SS_CONNECTING:
 597                 err = -EALREADY;
 598                 /* Fall out of switch with err, set for this state */
 599                 break;
 600         case SS_UNCONNECTED:
 601                 err = -EISCONN;
 602                 if (sk->sk_state != TCP_CLOSE)
 603                         goto out;
 604
 605                 err = sk->sk_prot->connect(sk, uaddr, addr_len);
 606                 if (err < 0)
 607                         goto out;
 608
 609                 sock->state = SS_CONNECTING;
 610
 611                 /* Just entered SS_CONNECTING state; the only
 612                  * difference is that return value in non-blocking
 613                  * case is EINPROGRESS, rather than EALREADY.
 614                  */
 615                 err = -EINPROGRESS;
 616                 break;
 617         }
 618
 619         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
 620
 621         if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 622                 /* Error code is set above */
 623                 if (!timeo || !inet_wait_for_connect(sk, timeo))
 624                         goto out;
 625
 626                 err = sock_intr_errno(timeo);
 627                 if (signal_pending(current))
 628                         goto out;
 629         }
 630
 631         /* Connection was closed by RST, timeout, ICMP error
 632          * or another process disconnected us.
 633          */
 634         if (sk->sk_state == TCP_CLOSE)
 635                 goto sock_error;
 636
 637         /* sk->sk_err may be not zero now, if RECVERR was ordered by user
 638          * and error was received after socket entered established state.
 639          * Hence, it is handled normally after connect() return successfully.
 640          */
 641
 642         sock->state = SS_CONNECTED;
 643         err = 0;
 644 out:
 645         release_sock(sk);
 646         return err;
 647
 648 sock_error:
 649         err = sock_error(sk) ? : -ECONNABORTED;
 650         sock->state = SS_UNCONNECTED;
 651         if (sk->sk_prot->disconnect(sk, flags))
 652                 sock->state = SS_DISCONNECTING;
 653         goto out;
 654 }
 655 EXPORT_SYMBOL(inet_stream_connect);
 656
 657 /*
 658  *      Accept a pending connection. The TCP layer now gives BSD semantics.
 659  */
 660
 661 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
 662 {
 663         struct sock *sk1 = sock->sk;
 664         int err = -EINVAL;
 665         struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
 666
 667         if (!sk2)
 668                 goto do_err;
 669
 670         lock_sock(sk2);
 671
 672         WARN_ON(!((1 << sk2->sk_state) &
 673                   (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE)));
 674
 675         sock_graft(sk2, newsock);
 676
 677         newsock->state = SS_CONNECTED;
 678         err = 0;
 679         release_sock(sk2);
 680 do_err:
 681         return err;
 682 }
 683 EXPORT_SYMBOL(inet_accept);
 684
 685
 686 /*
 687  *      This does both peername and sockname.
 688  */
 689 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
 690                         int *uaddr_len, int peer)
 691 {
 692         struct sock *sk         = sock->sk;
 693         struct inet_sock *inet  = inet_sk(sk);
 694         DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
 695
 696         sin->sin_family = AF_INET;
 697         if (peer) {
 698                 if (!inet->inet_dport ||
 699                     (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
 700                      peer == 1))
 701                         return -ENOTCONN;
 702                 sin->sin_port = inet->inet_dport;
 703                 sin->sin_addr.s_addr = inet->inet_daddr;
 704         } else {
 705                 __be32 addr = inet->inet_rcv_saddr;
 706                 if (!addr)
 707                         addr = inet->inet_saddr;
 708                 sin->sin_port = inet->inet_sport;
 709                 sin->sin_addr.s_addr = addr;
 710         }
 711         memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 712         *uaddr_len = sizeof(*sin);
 713         return 0;
 714 }
 715 EXPORT_SYMBOL(inet_getname);
 716
 717 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
 718                  size_t size)
 719 {
 720         struct sock *sk = sock->sk;
 721
 722         /* We may need to bind the socket. */
 723         if (!inet_sk(sk)->inet_num && inet_autobind(sk))
 724                 return -EAGAIN;
 725
 726         return sk->sk_prot->sendmsg(iocb, sk, msg, size);
 727 }
 728 EXPORT_SYMBOL(inet_sendmsg);
 729
 730
 731 static ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
 732                              size_t size, int flags)
 733 {
 734         struct sock *sk = sock->sk;
 735
 736         /* We may need to bind the socket. */
 737         if (!inet_sk(sk)->inet_num && inet_autobind(sk))
 738                 return -EAGAIN;
 739
 740         if (sk->sk_prot->sendpage)
 741                 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
 742         return sock_no_sendpage(sock, page, offset, size, flags);
 743 }
 744
 745
 746 int inet_shutdown(struct socket *sock, int how)
 747 {
 748         struct sock *sk = sock->sk;
 749         int err = 0;
 750
 751         /* This should really check to make sure
 752          * the socket is a TCP socket. (WHY AC...)
 753          */
 754         how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
 755                        1->2 bit 2 snds.
 756                        2->3 */
 757         if ((how & ~SHUTDOWN_MASK) || !how)     /* MAXINT->0 */
 758                 return -EINVAL;
 759
 760         lock_sock(sk);
 761         if (sock->state == SS_CONNECTING) {
 762                 if ((1 << sk->sk_state) &
 763                     (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
 764                         sock->state = SS_DISCONNECTING;
 765                 else
 766                         sock->state = SS_CONNECTED;
 767         }
 768
 769         switch (sk->sk_state) {
 770         case TCP_CLOSE:
 771                 err = -ENOTCONN;
 772                 /* Hack to wake up other listeners, who can poll for
 773                    POLLHUP, even on eg. unconnected UDP sockets -- RR */
 774         default:
 775                 sk->sk_shutdown |= how;
 776                 if (sk->sk_prot->shutdown)
 777                         sk->sk_prot->shutdown(sk, how);
 778                 break;
 779
 780         /* Remaining two branches are temporary solution for missing
 781          * close() in multithreaded environment. It is _not_ a good idea,
 782          * but we have no choice until close() is repaired at VFS level.
 783          */
 784         case TCP_LISTEN:
 785                 if (!(how & RCV_SHUTDOWN))
 786                         break;
 787                 /* Fall through */
 788         case TCP_SYN_SENT:
 789                 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
 790                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 791                 break;
 792         }
 793
 794         /* Wake up anyone sleeping in poll. */
 795         sk->sk_state_change(sk);
 796         release_sock(sk);
 797         return err;
 798 }
 799 EXPORT_SYMBOL(inet_shutdown);
 800
 801 /*
 802  *      ioctl() calls you can issue on an INET socket. Most of these are
 803  *      device configuration and stuff and very rarely used. Some ioctls
 804  *      pass on to the socket itself.
 805  *
 806  *      NOTE: I like the idea of a module for the config stuff. ie ifconfig
 807  *      loads the devconfigure module does its configuring and unloads it.
 808  *      There's a good 20K of config code hanging around the kernel.
 809  */
 810
 811 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 812 {
 813         struct sock *sk = sock->sk;
 814         int err = 0;
 815         struct net *net = sock_net(sk);
 816
 817         switch (cmd) {
 818         case SIOCGSTAMP:
 819                 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
 820                 break;
 821         case SIOCGSTAMPNS:
 822                 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
 823                 break;
 824         case SIOCADDRT:
 825         case SIOCDELRT:
 826         case SIOCRTMSG:
 827                 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
 828                 break;
 829         case SIOCDARP:
 830         case SIOCGARP:
 831         case SIOCSARP:
 832                 err = arp_ioctl(net, cmd, (void __user *)arg);
 833                 break;
 834         case SIOCGIFADDR:
 835         case SIOCSIFADDR:
 836         case SIOCGIFBRDADDR:
 837         case SIOCSIFBRDADDR:
 838         case SIOCGIFNETMASK:
 839         case SIOCSIFNETMASK:
 840         case SIOCGIFDSTADDR:
 841         case SIOCSIFDSTADDR:
 842         case SIOCSIFPFLAGS:
 843         case SIOCGIFPFLAGS:
 844         case SIOCSIFFLAGS:
 845                 err = devinet_ioctl(net, cmd, (void __user *)arg);
 846                 break;
 847         default:
 848                 if (sk->sk_prot->ioctl)
 849                         err = sk->sk_prot->ioctl(sk, cmd, arg);
 850                 else
 851                         err = -ENOIOCTLCMD;
 852                 break;
 853         }
 854         return err;
 855 }
 856 EXPORT_SYMBOL(inet_ioctl);
 857
 858 const struct proto_ops inet_stream_ops = {
 859         .family            = PF_INET,
 860         .owner             = THIS_MODULE,
 861         .release           = inet_release,
 862         .bind              = inet_bind,
 863         .connect           = inet_stream_connect,
 864         .socketpair        = sock_no_socketpair,
 865         .accept            = inet_accept,
 866         .getname           = inet_getname,
 867         .poll              = tcp_poll,
 868         .ioctl             = inet_ioctl,
 869         .listen            = inet_listen,
 870         .shutdown          = inet_shutdown,
 871         .setsockopt        = sock_common_setsockopt,
 872         .getsockopt        = sock_common_getsockopt,
 873         .sendmsg           = tcp_sendmsg,
 874         .recvmsg           = sock_common_recvmsg,
 875         .mmap              = sock_no_mmap,
 876         .sendpage          = tcp_sendpage,
 877         .splice_read       = tcp_splice_read,
 878 #ifdef CONFIG_COMPAT
 879         .compat_setsockopt = compat_sock_common_setsockopt,
 880         .compat_getsockopt = compat_sock_common_getsockopt,
 881 #endif
 882 };
 883 EXPORT_SYMBOL(inet_stream_ops);
 884
 885 const struct proto_ops inet_dgram_ops = {
 886         .family            = PF_INET,
 887         .owner             = THIS_MODULE,
 888         .release           = inet_release,
 889         .bind              = inet_bind,
 890         .connect           = inet_dgram_connect,
 891         .socketpair        = sock_no_socketpair,
 892         .accept            = sock_no_accept,
 893         .getname           = inet_getname,
 894         .poll              = udp_poll,
 895         .ioctl             = inet_ioctl,
 896         .listen            = sock_no_listen,
 897         .shutdown          = inet_shutdown,
 898         .setsockopt        = sock_common_setsockopt,
 899         .getsockopt        = sock_common_getsockopt,
 900         .sendmsg           = inet_sendmsg,
 901         .recvmsg           = sock_common_recvmsg,
 902         .mmap              = sock_no_mmap,
 903         .sendpage          = inet_sendpage,
 904 #ifdef CONFIG_COMPAT
 905         .compat_setsockopt = compat_sock_common_setsockopt,
 906         .compat_getsockopt = compat_sock_common_getsockopt,
 907 #endif
 908 };
 909 EXPORT_SYMBOL(inet_dgram_ops);
 910
 911 /*
 912  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
 913  * udp_poll
 914  */
 915 static const struct proto_ops inet_sockraw_ops = {
 916         .family            = PF_INET,
 917         .owner             = THIS_MODULE,
 918         .release           = inet_release,
 919         .bind              = inet_bind,
 920         .connect           = inet_dgram_connect,
 921         .socketpair        = sock_no_socketpair,
 922         .accept            = sock_no_accept,
 923         .getname           = inet_getname,
 924         .poll              = datagram_poll,
 925         .ioctl             = inet_ioctl,
 926         .listen            = sock_no_listen,
 927         .shutdown          = inet_shutdown,
 928         .setsockopt        = sock_common_setsockopt,
 929         .getsockopt        = sock_common_getsockopt,
 930         .sendmsg           = inet_sendmsg,
 931         .recvmsg           = sock_common_recvmsg,
 932         .mmap              = sock_no_mmap,
 933         .sendpage          = inet_sendpage,
 934 #ifdef CONFIG_COMPAT
 935         .compat_setsockopt = compat_sock_common_setsockopt,
 936         .compat_getsockopt = compat_sock_common_getsockopt,
 937 #endif
 938 };
 939
 940 static const struct net_proto_family inet_family_ops = {
 941         .family = PF_INET,
 942         .create = inet_create,
 943         .owner  = THIS_MODULE,
 944 };
 945
 946 /* Upon startup we insert all the elements in inetsw_array[] into
 947  * the linked list inetsw.
 948  */
 949 static struct inet_protosw inetsw_array[] =
 950 {
 951         {
 952                 .type =       SOCK_STREAM,
 953                 .protocol =   IPPROTO_TCP,
 954                 .prot =       &tcp_prot,
 955                 .ops =        &inet_stream_ops,
 956                 .no_check =   0,
 957                 .flags =      INET_PROTOSW_PERMANENT |
 958                               INET_PROTOSW_ICSK,
 959         },
 960
 961         {
 962                 .type =       SOCK_DGRAM,
 963                 .protocol =   IPPROTO_UDP,
 964                 .prot =       &udp_prot,
 965                 .ops =        &inet_dgram_ops,
 966                 .no_check =   UDP_CSUM_DEFAULT,
 967                 .flags =      INET_PROTOSW_PERMANENT,
 968        },
 969
 970
 971        {
 972                .type =       SOCK_RAW,
 973                .protocol =   IPPROTO_IP,        /* wild card */
 974                .prot =       &raw_prot,
 975                .ops =        &inet_sockraw_ops,
 976                .no_check =   UDP_CSUM_DEFAULT,
 977                .flags =      INET_PROTOSW_REUSE,
 978        }
 979 };
 980
 981 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
 982
 983 void inet_register_protosw(struct inet_protosw *p)
 984 {
 985         struct list_head *lh;
 986         struct inet_protosw *answer;
 987         int protocol = p->protocol;
 988         struct list_head *last_perm;
 989
 990         spin_lock_bh(&inetsw_lock);
 991
 992         if (p->type >= SOCK_MAX)
 993                 goto out_illegal;
 994
 995         /* If we are trying to override a permanent protocol, bail. */
 996         answer = NULL;
 997         last_perm = &inetsw[p->type];
 998         list_for_each(lh, &inetsw[p->type]) {
 999                 answer = list_entry(lh, struct inet_protosw, list);
1000
1001                 /* Check only the non-wild match. */
1002                 if (INET_PROTOSW_PERMANENT & answer->flags) {
1003                         if (protocol == answer->protocol)
1004                                 break;
1005                         last_perm = lh;
1006                 }
1007
1008                 answer = NULL;
1009         }
1010         if (answer)
1011                 goto out_permanent;
1012
1013         /* Add the new entry after the last permanent entry if any, so that
1014          * the new entry does not override a permanent entry when matched with
1015          * a wild-card protocol. But it is allowed to override any existing
1016          * non-permanent entry.  This means that when we remove this entry, the
1017          * system automatically returns to the old behavior.
1018          */
1019         list_add_rcu(&p->list, last_perm);
1020 out:
1021         spin_unlock_bh(&inetsw_lock);
1022
1023         return;
1024
1025 out_permanent:
1026         printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
1027                protocol);
1028         goto out;
1029
1030 out_illegal:
1031         printk(KERN_ERR
1032                "Ignoring attempt to register invalid socket type %d.\n",
1033                p->type);
1034         goto out;
1035 }
1036 EXPORT_SYMBOL(inet_register_protosw);
1037
1038 void inet_unregister_protosw(struct inet_protosw *p)
1039 {
1040         if (INET_PROTOSW_PERMANENT & p->flags) {
1041                 printk(KERN_ERR
1042                        "Attempt to unregister permanent protocol %d.\n",
1043                        p->protocol);
1044         } else {
1045                 spin_lock_bh(&inetsw_lock);
1046                 list_del_rcu(&p->list);
1047                 spin_unlock_bh(&inetsw_lock);
1048
1049                 synchronize_net();
1050         }
1051 }
1052 EXPORT_SYMBOL(inet_unregister_protosw);
1053
1054 /*
1055  *      Shall we try to damage output packets if routing dev changes?
1056  */
1057
1058 int sysctl_ip_dynaddr __read_mostly;
1059
1060 static int inet_sk_reselect_saddr(struct sock *sk)
1061 {
1062         struct inet_sock *inet = inet_sk(sk);
1063         int err;
1064         struct rtable *rt;
1065         __be32 old_saddr = inet->inet_saddr;
1066         __be32 new_saddr;
1067         __be32 daddr = inet->inet_daddr;
1068
1069         if (inet->opt && inet->opt->srr)
1070                 daddr = inet->opt->faddr;
1071
1072         /* Query new route. */
1073         err = ip_route_connect(&rt, daddr, 0,
1074                                RT_CONN_FLAGS(sk),
1075                                sk->sk_bound_dev_if,
1076                                sk->sk_protocol,
1077                                inet->inet_sport, inet->inet_dport, sk, 0);
1078         if (err)
1079                 return err;
1080
1081         sk_setup_caps(sk, &rt->u.dst);
1082
1083         new_saddr = rt->rt_src;
1084
1085         if (new_saddr == old_saddr)
1086                 return 0;
1087
1088         if (sysctl_ip_dynaddr > 1) {
1089                 printk(KERN_INFO "%s(): shifting inet->saddr from %pI4 to %pI4\n",
1090                        __func__, &old_saddr, &new_saddr);
1091         }
1092
1093         inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1094
1095         /*
1096          * XXX The only one ugly spot where we need to
1097          * XXX really change the sockets identity after
1098          * XXX it has entered the hashes. -DaveM
1099          *
1100          * Besides that, it does not check for connection
1101          * uniqueness. Wait for troubles.
1102          */
1103         __sk_prot_rehash(sk);
1104         return 0;
1105 }
1106
1107 int inet_sk_rebuild_header(struct sock *sk)
1108 {
1109         struct inet_sock *inet = inet_sk(sk);
1110         struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1111         __be32 daddr;
1112         int err;
1113
1114         /* Route is OK, nothing to do. */
1115         if (rt)
1116                 return 0;
1117
1118         /* Reroute. */
1119         daddr = inet->inet_daddr;
1120         if (inet->opt && inet->opt->srr)
1121                 daddr = inet->opt->faddr;
1122 {
1123         struct flowi fl = {
1124                 .oif = sk->sk_bound_dev_if,
1125                 .mark = sk->sk_mark,
1126                 .nl_u = {
1127                         .ip4_u = {
1128                                 .daddr  = daddr,
1129                                 .saddr  = inet->inet_saddr,
1130                                 .tos    = RT_CONN_FLAGS(sk),
1131                         },
1132                 },
1133                 .proto = sk->sk_protocol,
1134                 .flags = inet_sk_flowi_flags(sk),
1135                 .uli_u = {
1136                         .ports = {
1137                                 .sport = inet->inet_sport,
1138                                 .dport = inet->inet_dport,
1139                         },
1140                 },
1141         };
1142
1143         security_sk_classify_flow(sk, &fl);
1144         err = ip_route_output_flow(sock_net(sk), &rt, &fl, sk, 0);
1145 }
1146         if (!err)
1147                 sk_setup_caps(sk, &rt->u.dst);
1148         else {
1149                 /* Routing failed... */
1150                 sk->sk_route_caps = 0;
1151                 /*
1152                  * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1153                  * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1154                  */
1155                 if (!sysctl_ip_dynaddr ||
1156                     sk->sk_state != TCP_SYN_SENT ||
1157                     (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1158                     (err = inet_sk_reselect_saddr(sk)) != 0)
1159                         sk->sk_err_soft = -err;
1160         }
1161
1162         return err;
1163 }
1164 EXPORT_SYMBOL(inet_sk_rebuild_header);
1165
1166 static int inet_gso_send_check(struct sk_buff *skb)
1167 {
1168         struct iphdr *iph;
1169         const struct net_protocol *ops;
1170         int proto;
1171         int ihl;
1172         int err = -EINVAL;
1173
1174         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1175                 goto out;
1176
1177         iph = ip_hdr(skb);
1178         ihl = iph->ihl * 4;
1179         if (ihl < sizeof(*iph))
1180                 goto out;
1181
1182         if (unlikely(!pskb_may_pull(skb, ihl)))
1183                 goto out;
1184
1185         __skb_pull(skb, ihl);
1186         skb_reset_transport_header(skb);
1187         iph = ip_hdr(skb);
1188         proto = iph->protocol & (MAX_INET_PROTOS - 1);
1189         err = -EPROTONOSUPPORT;
1190
1191         rcu_read_lock();
1192         ops = rcu_dereference(inet_protos[proto]);
1193         if (likely(ops && ops->gso_send_check))
1194                 err = ops->gso_send_check(skb);
1195         rcu_read_unlock();
1196
1197 out:
1198         return err;
1199 }
1200
1201 static struct sk_buff *inet_gso_segment(struct sk_buff *skb, int features)
1202 {
1203         struct sk_buff *segs = ERR_PTR(-EINVAL);
1204         struct iphdr *iph;
1205         const struct net_protocol *ops;
1206         int proto;
1207         int ihl;
1208         int id;
1209         unsigned int offset = 0;
1210
1211         if (!(features & NETIF_F_V4_CSUM))
1212                 features &= ~NETIF_F_SG;
1213
1214         if (unlikely(skb_shinfo(skb)->gso_type &
1215                      ~(SKB_GSO_TCPV4 |
1216                        SKB_GSO_UDP |
1217                        SKB_GSO_DODGY |
1218                        SKB_GSO_TCP_ECN |
1219                        0)))
1220                 goto out;
1221
1222         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1223                 goto out;
1224
1225         iph = ip_hdr(skb);
1226         ihl = iph->ihl * 4;
1227         if (ihl < sizeof(*iph))
1228                 goto out;
1229
1230         if (unlikely(!pskb_may_pull(skb, ihl)))
1231                 goto out;
1232
1233         __skb_pull(skb, ihl);
1234         skb_reset_transport_header(skb);
1235         iph = ip_hdr(skb);
1236         id = ntohs(iph->id);
1237         proto = iph->protocol & (MAX_INET_PROTOS - 1);
1238         segs = ERR_PTR(-EPROTONOSUPPORT);
1239
1240         rcu_read_lock();
1241         ops = rcu_dereference(inet_protos[proto]);
1242         if (likely(ops && ops->gso_segment))
1243                 segs = ops->gso_segment(skb, features);
1244         rcu_read_unlock();
1245
1246         if (!segs || IS_ERR(segs))
1247                 goto out;
1248
1249         skb = segs;
1250         do {
1251                 iph = ip_hdr(skb);
1252                 if (proto == IPPROTO_UDP) {
1253                         iph->id = htons(id);
1254                         iph->frag_off = htons(offset >> 3);
1255                         if (skb->next != NULL)
1256                                 iph->frag_off |= htons(IP_MF);
1257                         offset += (skb->len - skb->mac_len - iph->ihl * 4);
1258                 } else
1259                         iph->id = htons(id++);
1260                 iph->tot_len = htons(skb->len - skb->mac_len);
1261                 iph->check = 0;
1262                 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1263         } while ((skb = skb->next));
1264
1265 out:
1266         return segs;
1267 }
1268
1269 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1270                                          struct sk_buff *skb)
1271 {
1272         const struct net_protocol *ops;
1273         struct sk_buff **pp = NULL;
1274         struct sk_buff *p;
1275         struct iphdr *iph;
1276         unsigned int hlen;
1277         unsigned int off;
1278         unsigned int id;
1279         int flush = 1;
1280         int proto;
1281
1282         off = skb_gro_offset(skb);
1283         hlen = off + sizeof(*iph);
1284         iph = skb_gro_header_fast(skb, off);
1285         if (skb_gro_header_hard(skb, hlen)) {
1286                 iph = skb_gro_header_slow(skb, hlen, off);
1287                 if (unlikely(!iph))
1288                         goto out;
1289         }
1290
1291         proto = iph->protocol & (MAX_INET_PROTOS - 1);
1292
1293         rcu_read_lock();
1294         ops = rcu_dereference(inet_protos[proto]);
1295         if (!ops || !ops->gro_receive)
1296                 goto out_unlock;
1297
1298         if (*(u8 *)iph != 0x45)
1299                 goto out_unlock;
1300
1301         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1302                 goto out_unlock;
1303
1304         id = ntohl(*(u32 *)&iph->id);
1305         flush = (u16)((ntohl(*(u32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF));
1306         id >>= 16;
1307
1308         for (p = *head; p; p = p->next) {
1309                 struct iphdr *iph2;
1310
1311                 if (!NAPI_GRO_CB(p)->same_flow)
1312                         continue;
1313
1314                 iph2 = ip_hdr(p);
1315
1316                 if ((iph->protocol ^ iph2->protocol) |
1317                     (iph->tos ^ iph2->tos) |
1318                     (iph->saddr ^ iph2->saddr) |
1319                     (iph->daddr ^ iph2->daddr)) {
1320                         NAPI_GRO_CB(p)->same_flow = 0;
1321                         continue;
1322                 }
1323
1324                 /* All fields must match except length and checksum. */
1325                 NAPI_GRO_CB(p)->flush |=
1326                         (iph->ttl ^ iph2->ttl) |
1327                         ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1328
1329                 NAPI_GRO_CB(p)->flush |= flush;
1330         }
1331
1332         NAPI_GRO_CB(skb)->flush |= flush;
1333         skb_gro_pull(skb, sizeof(*iph));
1334         skb_set_transport_header(skb, skb_gro_offset(skb));
1335
1336         pp = ops->gro_receive(head, skb);
1337
1338 out_unlock:
1339         rcu_read_unlock();
1340
1341 out:
1342         NAPI_GRO_CB(skb)->flush |= flush;
1343
1344         return pp;
1345 }
1346
1347 static int inet_gro_complete(struct sk_buff *skb)
1348 {
1349         const struct net_protocol *ops;
1350         struct iphdr *iph = ip_hdr(skb);
1351         int proto = iph->protocol & (MAX_INET_PROTOS - 1);
1352         int err = -ENOSYS;
1353         __be16 newlen = htons(skb->len - skb_network_offset(skb));
1354
1355         csum_replace2(&iph->check, iph->tot_len, newlen);
1356         iph->tot_len = newlen;
1357
1358         rcu_read_lock();
1359         ops = rcu_dereference(inet_protos[proto]);
1360         if (WARN_ON(!ops || !ops->gro_complete))
1361                 goto out_unlock;
1362
1363         err = ops->gro_complete(skb);
1364
1365 out_unlock:
1366         rcu_read_unlock();
1367
1368         return err;
1369 }
1370
1371 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1372                          unsigned short type, unsigned char protocol,
1373                          struct net *net)
1374 {
1375         struct socket *sock;
1376         int rc = sock_create_kern(family, type, protocol, &sock);
1377
1378         if (rc == 0) {
1379                 *sk = sock->sk;
1380                 (*sk)->sk_allocation = GFP_ATOMIC;
1381                 /*
1382                  * Unhash it so that IP input processing does not even see it,
1383                  * we do not wish this socket to see incoming packets.
1384                  */
1385                 (*sk)->sk_prot->unhash(*sk);
1386
1387                 sk_change_net(*sk, net);
1388         }
1389         return rc;
1390 }
1391 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1392
1393 unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1394 {
1395         unsigned long res = 0;
1396         int i;
1397
1398         for_each_possible_cpu(i) {
1399                 res += *(((unsigned long *) per_cpu_ptr(mib[0], i)) + offt);
1400                 res += *(((unsigned long *) per_cpu_ptr(mib[1], i)) + offt);
1401         }
1402         return res;
1403 }
1404 EXPORT_SYMBOL_GPL(snmp_fold_field);
1405
1406 int snmp_mib_init(void __percpu *ptr[2], size_t mibsize)
1407 {
1408         BUG_ON(ptr == NULL);
1409         ptr[0] = __alloc_percpu(mibsize, __alignof__(unsigned long long));
1410         if (!ptr[0])
1411                 goto err0;
1412         ptr[1] = __alloc_percpu(mibsize, __alignof__(unsigned long long));
1413         if (!ptr[1])
1414                 goto err1;
1415         return 0;
1416 err1:
1417         free_percpu(ptr[0]);
1418         ptr[0] = NULL;
1419 err0:
1420         return -ENOMEM;
1421 }
1422 EXPORT_SYMBOL_GPL(snmp_mib_init);
1423
1424 void snmp_mib_free(void __percpu *ptr[2])
1425 {
1426         BUG_ON(ptr == NULL);
1427         free_percpu(ptr[0]);
1428         free_percpu(ptr[1]);
1429         ptr[0] = ptr[1] = NULL;
1430 }
1431 EXPORT_SYMBOL_GPL(snmp_mib_free);
1432
1433 #ifdef CONFIG_IP_MULTICAST
1434 static const struct net_protocol igmp_protocol = {
1435         .handler =      igmp_rcv,
1436         .netns_ok =     1,
1437 };
1438 #endif
1439
1440 static const struct net_protocol tcp_protocol = {
1441         .handler =      tcp_v4_rcv,
1442         .err_handler =  tcp_v4_err,
1443         .gso_send_check = tcp_v4_gso_send_check,
1444         .gso_segment =  tcp_tso_segment,
1445         .gro_receive =  tcp4_gro_receive,
1446         .gro_complete = tcp4_gro_complete,
1447         .no_policy =    1,
1448         .netns_ok =     1,
1449 };
1450
1451 static const struct net_protocol udp_protocol = {
1452         .handler =      udp_rcv,
1453         .err_handler =  udp_err,
1454         .gso_send_check = udp4_ufo_send_check,
1455         .gso_segment = udp4_ufo_fragment,
1456         .no_policy =    1,
1457         .netns_ok =     1,
1458 };
1459
1460 static const struct net_protocol icmp_protocol = {
1461         .handler =      icmp_rcv,
1462         .no_policy =    1,
1463         .netns_ok =     1,
1464 };
1465
1466 static __net_init int ipv4_mib_init_net(struct net *net)
1467 {
1468         if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1469                           sizeof(struct tcp_mib)) < 0)
1470                 goto err_tcp_mib;
1471         if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1472                           sizeof(struct ipstats_mib)) < 0)
1473                 goto err_ip_mib;
1474         if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1475                           sizeof(struct linux_mib)) < 0)
1476                 goto err_net_mib;
1477         if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1478                           sizeof(struct udp_mib)) < 0)
1479                 goto err_udp_mib;
1480         if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1481                           sizeof(struct udp_mib)) < 0)
1482                 goto err_udplite_mib;
1483         if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1484                           sizeof(struct icmp_mib)) < 0)
1485                 goto err_icmp_mib;
1486         if (snmp_mib_init((void __percpu **)net->mib.icmpmsg_statistics,
1487                           sizeof(struct icmpmsg_mib)) < 0)
1488                 goto err_icmpmsg_mib;
1489
1490         tcp_mib_init(net);
1491         return 0;
1492
1493 err_icmpmsg_mib:
1494         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1495 err_icmp_mib:
1496         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1497 err_udplite_mib:
1498         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1499 err_udp_mib:
1500         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1501 err_net_mib:
1502         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1503 err_ip_mib:
1504         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1505 err_tcp_mib:
1506         return -ENOMEM;
1507 }
1508
1509 static __net_exit void ipv4_mib_exit_net(struct net *net)
1510 {
1511         snmp_mib_free((void __percpu **)net->mib.icmpmsg_statistics);
1512         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1513         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1514         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1515         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1516         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1517         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1518 }
1519
1520 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1521         .init = ipv4_mib_init_net,
1522         .exit = ipv4_mib_exit_net,
1523 };
1524
1525 static int __init init_ipv4_mibs(void)
1526 {
1527         return register_pernet_subsys(&ipv4_mib_ops);
1528 }
1529
1530 static int ipv4_proc_init(void);
1531
1532 /*
1533  *      IP protocol layer initialiser
1534  */
1535
1536 static struct packet_type ip_packet_type __read_mostly = {
1537         .type = cpu_to_be16(ETH_P_IP),
1538         .func = ip_rcv,
1539         .gso_send_check = inet_gso_send_check,
1540         .gso_segment = inet_gso_segment,
1541         .gro_receive = inet_gro_receive,
1542         .gro_complete = inet_gro_complete,
1543 };
1544
1545 static int __init inet_init(void)
1546 {
1547         struct sk_buff *dummy_skb;
1548         struct inet_protosw *q;
1549         struct list_head *r;
1550         int rc = -EINVAL;
1551
1552         BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1553
1554         rc = proto_register(&tcp_prot, 1);
1555         if (rc)
1556                 goto out;
1557
1558         rc = proto_register(&udp_prot, 1);
1559         if (rc)
1560                 goto out_unregister_tcp_proto;
1561
1562         rc = proto_register(&raw_prot, 1);
1563         if (rc)
1564                 goto out_unregister_udp_proto;
1565
1566         /*
1567          *      Tell SOCKET that we are alive...
1568          */
1569
1570         (void)sock_register(&inet_family_ops);
1571
1572 #ifdef CONFIG_SYSCTL
1573         ip_static_sysctl_init();
1574 #endif
1575
1576         /*
1577          *      Add all the base protocols.
1578          */
1579
1580         if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1581                 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1582         if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1583                 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1584         if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1585                 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1586 #ifdef CONFIG_IP_MULTICAST
1587         if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1588                 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1589 #endif
1590
1591         /* Register the socket-side information for inet_create. */
1592         for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1593                 INIT_LIST_HEAD(r);
1594
1595         for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1596                 inet_register_protosw(q);
1597
1598         /*
1599          *      Set the ARP module up
1600          */
1601
1602         arp_init();
1603
1604         /*
1605          *      Set the IP module up
1606          */
1607
1608         ip_init();
1609
1610         tcp_v4_init();
1611
1612         /* Setup TCP slab cache for open requests. */
1613         tcp_init();
1614
1615         /* Setup UDP memory threshold */
1616         udp_init();
1617
1618         /* Add UDP-Lite (RFC 3828) */
1619         udplite4_register();
1620
1621         /*
1622          *      Set the ICMP layer up
1623          */
1624
1625         if (icmp_init() < 0)
1626                 panic("Failed to create the ICMP control socket.\n");
1627
1628         /*
1629          *      Initialise the multicast router
1630          */
1631 #if defined(CONFIG_IP_MROUTE)
1632         if (ip_mr_init())
1633                 printk(KERN_CRIT "inet_init: Cannot init ipv4 mroute\n");
1634 #endif
1635         /*
1636          *      Initialise per-cpu ipv4 mibs
1637          */
1638
1639         if (init_ipv4_mibs())
1640                 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n");
1641
1642         ipv4_proc_init();
1643
1644         ipfrag_init();
1645
1646         dev_add_pack(&ip_packet_type);
1647
1648         rc = 0;
1649 out:
1650         return rc;
1651 out_unregister_udp_proto:
1652         proto_unregister(&udp_prot);
1653 out_unregister_tcp_proto:
1654         proto_unregister(&tcp_prot);
1655         goto out;
1656 }
1657
1658 fs_initcall(inet_init);
1659
1660 /* ------------------------------------------------------------------------ */
1661
1662 #ifdef CONFIG_PROC_FS
1663 static int __init ipv4_proc_init(void)
1664 {
1665         int rc = 0;
1666
1667         if (raw_proc_init())
1668                 goto out_raw;
1669         if (tcp4_proc_init())
1670                 goto out_tcp;
1671         if (udp4_proc_init())
1672                 goto out_udp;
1673         if (ip_misc_proc_init())
1674                 goto out_misc;
1675 out:
1676         return rc;
1677 out_misc:
1678         udp4_proc_exit();
1679 out_udp:
1680         tcp4_proc_exit();
1681 out_tcp:
1682         raw_proc_exit();
1683 out_raw:
1684         rc = -ENOMEM;
1685         goto out;
1686 }
1687
1688 #else /* CONFIG_PROC_FS */
1689 static int __init ipv4_proc_init(void)
1690 {
1691         return 0;
1692 }
1693 #endif /* CONFIG_PROC_FS */
1694
1695 MODULE_ALIAS_NETPROTO(PF_INET);
1696