net/ipv4/af_inet.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              PF_INET protocol family socket handler.
   7  *
   8  * Authors:     Ross Biro
   9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  10  *              Florian La Roche, <flla@stud.uni-sb.de>
  11  *              Alan Cox, <A.Cox@swansea.ac.uk>
  12  *
  13  * Changes (see also sock.c)
  14  *
  15  *              piggy,
  16  *              Karl Knutson    :       Socket protocol table
  17  *              A.N.Kuznetsov   :       Socket death error in accept().
  18  *              John Richardson :       Fix non blocking error in connect()
  19  *                                      so sockets that fail to connect
  20  *                                      don't return -EINPROGRESS.
  21  *              Alan Cox        :       Asynchronous I/O support
  22  *              Alan Cox        :       Keep correct socket pointer on sock
  23  *                                      structures
  24  *                                      when accept() ed
  25  *              Alan Cox        :       Semantics of SO_LINGER aren't state
  26  *                                      moved to close when you look carefully.
  27  *                                      With this fixed and the accept bug fixed
  28  *                                      some RPC stuff seems happier.
  29  *              Niibe Yutaka    :       4.4BSD style write async I/O
  30  *              Alan Cox,
  31  *              Tony Gale       :       Fixed reuse semantics.
  32  *              Alan Cox        :       bind() shouldn't abort existing but dead
  33  *                                      sockets. Stops FTP netin:.. I hope.
  34  *              Alan Cox        :       bind() works correctly for RAW sockets.
  35  *                                      Note that FreeBSD at least was broken
  36  *                                      in this respect so be careful with
  37  *                                      compatibility tests...
  38  *              Alan Cox        :       routing cache support
  39  *              Alan Cox        :       memzero the socket structure for
  40  *                                      compactness.
  41  *              Matt Day        :       nonblock connect error handler
  42  *              Alan Cox        :       Allow large numbers of pending sockets
  43  *                                      (eg for big web sites), but only if
  44  *                                      specifically application requested.
  45  *              Alan Cox        :       New buffering throughout IP. Used
  46  *                                      dumbly.
  47  *              Alan Cox        :       New buffering now used smartly.
  48  *              Alan Cox        :       BSD rather than common sense
  49  *                                      interpretation of listen.
  50  *              Germano Caronni :       Assorted small races.
  51  *              Alan Cox        :       sendmsg/recvmsg basic support.
  52  *              Alan Cox        :       Only sendmsg/recvmsg now supported.
  53  *              Alan Cox        :       Locked down bind (see security list).
  54  *              Alan Cox        :       Loosened bind a little.
  55  *              Mike McLagan    :       ADD/DEL DLCI Ioctls
  56  *      Willy Konynenberg       :       Transparent proxying support.
  57  *              David S. Miller :       New socket lookup architecture.
  58  *                                      Some other random speedups.
  59  *              Cyrus Durgin    :       Cleaned up file for kmod hacks.
  60  *              Andi Kleen      :       Fix inet_stream_connect TCP race.
  61  *
  62  *              This program is free software; you can redistribute it and/or
  63  *              modify it under the terms of the GNU General Public License
  64  *              as published by the Free Software Foundation; either version
  65  *              2 of the License, or (at your option) any later version.
  66  */
  67
  68 #define pr_fmt(fmt) "IPv4: " fmt
  69
  70 #include <linux/err.h>
  71 #include <linux/errno.h>
  72 #include <linux/types.h>
  73 #include <linux/socket.h>
  74 #include <linux/in.h>
  75 #include <linux/kernel.h>
  76 #include <linux/module.h>
  77 #include <linux/sched.h>
  78 #include <linux/timer.h>
  79 #include <linux/string.h>
  80 #include <linux/sockios.h>
  81 #include <linux/net.h>
  82 #include <linux/capability.h>
  83 #include <linux/fcntl.h>
  84 #include <linux/mm.h>
  85 #include <linux/interrupt.h>
  86 #include <linux/stat.h>
  87 #include <linux/init.h>
  88 #include <linux/poll.h>
  89 #include <linux/netfilter_ipv4.h>
  90 #include <linux/random.h>
  91 #include <linux/slab.h>
  92
  93 #include <asm/uaccess.h>
  94
  95 #include <linux/inet.h>
  96 #include <linux/igmp.h>
  97 #include <linux/inetdevice.h>
  98 #include <linux/netdevice.h>
  99 #include <net/checksum.h>
 100 #include <net/ip.h>
 101 #include <net/protocol.h>
 102 #include <net/arp.h>
 103 #include <net/route.h>
 104 #include <net/ip_fib.h>
 105 #include <net/inet_connection_sock.h>
 106 #include <net/tcp.h>
 107 #include <net/udp.h>
 108 #include <net/udplite.h>
 109 #include <net/ping.h>
 110 #include <linux/skbuff.h>
 111 #include <net/sock.h>
 112 #include <net/raw.h>
 113 #include <net/icmp.h>
 114 #include <net/inet_common.h>
 115 #include <net/xfrm.h>
 116 #include <net/net_namespace.h>
 117 #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         dst_release(sk->sk_rx_dst);
 160         sk_refcnt_debug_dec(sk);
 161 }
 162 EXPORT_SYMBOL(inet_sock_destruct);
 163
 164 /*
 165  *      The routines beyond this point handle the behaviour of an AF_INET
 166  *      socket object. Mostly it punts to the subprotocols of IP to do
 167  *      the work.
 168  */
 169
 170 /*
 171  *      Automatically bind an unbound socket.
 172  */
 173
 174 static int inet_autobind(struct sock *sk)
 175 {
 176         struct inet_sock *inet;
 177         /* We may need to bind the socket. */
 178         lock_sock(sk);
 179         inet = inet_sk(sk);
 180         if (!inet->inet_num) {
 181                 if (sk->sk_prot->get_port(sk, 0)) {
 182                         release_sock(sk);
 183                         return -EAGAIN;
 184                 }
 185                 inet->inet_sport = htons(inet->inet_num);
 186         }
 187         release_sock(sk);
 188         return 0;
 189 }
 190
 191 /*
 192  *      Move a socket into listening state.
 193  */
 194 int inet_listen(struct socket *sock, int backlog)
 195 {
 196         struct sock *sk = sock->sk;
 197         unsigned char old_state;
 198         int err;
 199
 200         lock_sock(sk);
 201
 202         err = -EINVAL;
 203         if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
 204                 goto out;
 205
 206         old_state = sk->sk_state;
 207         if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
 208                 goto out;
 209
 210         /* Really, if the socket is already in listen state
 211          * we can only allow the backlog to be adjusted.
 212          */
 213         if (old_state != TCP_LISTEN) {
 214                 /* Check special setups for testing purpose to enable TFO w/o
 215                  * requiring TCP_FASTOPEN sockopt.
 216                  * Note that only TCP sockets (SOCK_STREAM) will reach here.
 217                  * Also fastopenq may already been allocated because this
 218                  * socket was in TCP_LISTEN state previously but was
 219                  * shutdown() (rather than close()).
 220                  */
 221                 if ((sysctl_tcp_fastopen & TFO_SERVER_ENABLE) != 0 &&
 222                     inet_csk(sk)->icsk_accept_queue.fastopenq == NULL) {
 223                         if ((sysctl_tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) != 0)
 224                                 err = fastopen_init_queue(sk, backlog);
 225                         else if ((sysctl_tcp_fastopen &
 226                                   TFO_SERVER_WO_SOCKOPT2) != 0)
 227                                 err = fastopen_init_queue(sk,
 228                                     ((uint)sysctl_tcp_fastopen) >> 16);
 229                         else
 230                                 err = 0;
 231                         if (err)
 232                                 goto out;
 233                 }
 234                 err = inet_csk_listen_start(sk, backlog);
 235                 if (err)
 236                         goto out;
 237         }
 238         sk->sk_max_ack_backlog = backlog;
 239         err = 0;
 240
 241 out:
 242         release_sock(sk);
 243         return err;
 244 }
 245 EXPORT_SYMBOL(inet_listen);
 246
 247 u32 inet_ehash_secret __read_mostly;
 248 EXPORT_SYMBOL(inet_ehash_secret);
 249
 250 u32 ipv6_hash_secret __read_mostly;
 251 EXPORT_SYMBOL(ipv6_hash_secret);
 252
 253 /*
 254  * inet_ehash_secret must be set exactly once, and to a non nul value
 255  * ipv6_hash_secret must be set exactly once.
 256  */
 257 void build_ehash_secret(void)
 258 {
 259         u32 rnd;
 260
 261         do {
 262                 get_random_bytes(&rnd, sizeof(rnd));
 263         } while (rnd == 0);
 264
 265         if (cmpxchg(&inet_ehash_secret, 0, rnd) == 0)
 266                 get_random_bytes(&ipv6_hash_secret, sizeof(ipv6_hash_secret));
 267 }
 268 EXPORT_SYMBOL(build_ehash_secret);
 269
 270 /*
 271  *      Create an inet socket.
 272  */
 273
 274 static int inet_create(struct net *net, struct socket *sock, int protocol,
 275                        int kern)
 276 {
 277         struct sock *sk;
 278         struct inet_protosw *answer;
 279         struct inet_sock *inet;
 280         struct proto *answer_prot;
 281         unsigned char answer_flags;
 282         char answer_no_check;
 283         int try_loading_module = 0;
 284         int err;
 285
 286         if (unlikely(!inet_ehash_secret))
 287                 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
 288                         build_ehash_secret();
 289
 290         sock->state = SS_UNCONNECTED;
 291
 292         /* Look for the requested type/protocol pair. */
 293 lookup_protocol:
 294         err = -ESOCKTNOSUPPORT;
 295         rcu_read_lock();
 296         list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
 297
 298                 err = 0;
 299                 /* Check the non-wild match. */
 300                 if (protocol == answer->protocol) {
 301                         if (protocol != IPPROTO_IP)
 302                                 break;
 303                 } else {
 304                         /* Check for the two wild cases. */
 305                         if (IPPROTO_IP == protocol) {
 306                                 protocol = answer->protocol;
 307                                 break;
 308                         }
 309                         if (IPPROTO_IP == answer->protocol)
 310                                 break;
 311                 }
 312                 err = -EPROTONOSUPPORT;
 313         }
 314
 315         if (unlikely(err)) {
 316                 if (try_loading_module < 2) {
 317                         rcu_read_unlock();
 318                         /*
 319                          * Be more specific, e.g. net-pf-2-proto-132-type-1
 320                          * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
 321                          */
 322                         if (++try_loading_module == 1)
 323                                 request_module("net-pf-%d-proto-%d-type-%d",
 324                                                PF_INET, protocol, sock->type);
 325                         /*
 326                          * Fall back to generic, e.g. net-pf-2-proto-132
 327                          * (net-pf-PF_INET-proto-IPPROTO_SCTP)
 328                          */
 329                         else
 330                                 request_module("net-pf-%d-proto-%d",
 331                                                PF_INET, protocol);
 332                         goto lookup_protocol;
 333                 } else
 334                         goto out_rcu_unlock;
 335         }
 336
 337         err = -EPERM;
 338         if (sock->type == SOCK_RAW && !kern &&
 339             !ns_capable(net->user_ns, CAP_NET_RAW))
 340                 goto out_rcu_unlock;
 341
 342         sock->ops = answer->ops;
 343         answer_prot = answer->prot;
 344         answer_no_check = answer->no_check;
 345         answer_flags = answer->flags;
 346         rcu_read_unlock();
 347
 348         WARN_ON(answer_prot->slab == NULL);
 349
 350         err = -ENOBUFS;
 351         sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
 352         if (sk == NULL)
 353                 goto out;
 354
 355         err = 0;
 356         sk->sk_no_check = answer_no_check;
 357         if (INET_PROTOSW_REUSE & answer_flags)
 358                 sk->sk_reuse = SK_CAN_REUSE;
 359
 360         inet = inet_sk(sk);
 361         inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
 362
 363         inet->nodefrag = 0;
 364
 365         if (SOCK_RAW == sock->type) {
 366                 inet->inet_num = protocol;
 367                 if (IPPROTO_RAW == protocol)
 368                         inet->hdrincl = 1;
 369         }
 370
 371         if (ipv4_config.no_pmtu_disc)
 372                 inet->pmtudisc = IP_PMTUDISC_DONT;
 373         else
 374                 inet->pmtudisc = IP_PMTUDISC_WANT;
 375
 376         inet->inet_id = 0;
 377
 378         sock_init_data(sock, sk);
 379
 380         sk->sk_destruct    = inet_sock_destruct;
 381         sk->sk_protocol    = protocol;
 382         sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
 383
 384         inet->uc_ttl    = -1;
 385         inet->mc_loop   = 1;
 386         inet->mc_ttl    = 1;
 387         inet->mc_all    = 1;
 388         inet->mc_index  = 0;
 389         inet->mc_list   = NULL;
 390         inet->rcv_tos   = 0;
 391
 392         sk_refcnt_debug_inc(sk);
 393
 394         if (inet->inet_num) {
 395                 /* It assumes that any protocol which allows
 396                  * the user to assign a number at socket
 397                  * creation time automatically
 398                  * shares.
 399                  */
 400                 inet->inet_sport = htons(inet->inet_num);
 401                 /* Add to protocol hash chains. */
 402                 sk->sk_prot->hash(sk);
 403         }
 404
 405         if (sk->sk_prot->init) {
 406                 err = sk->sk_prot->init(sk);
 407                 if (err)
 408                         sk_common_release(sk);
 409         }
 410 out:
 411         return err;
 412 out_rcu_unlock:
 413         rcu_read_unlock();
 414         goto out;
 415 }
 416
 417
 418 /*
 419  *      The peer socket should always be NULL (or else). When we call this
 420  *      function we are destroying the object and from then on nobody
 421  *      should refer to it.
 422  */
 423 int inet_release(struct socket *sock)
 424 {
 425         struct sock *sk = sock->sk;
 426
 427         if (sk) {
 428                 long timeout;
 429
 430                 sock_rps_reset_flow(sk);
 431
 432                 /* Applications forget to leave groups before exiting */
 433                 ip_mc_drop_socket(sk);
 434
 435                 /* If linger is set, we don't return until the close
 436                  * is complete.  Otherwise we return immediately. The
 437                  * actually closing is done the same either way.
 438                  *
 439                  * If the close is due to the process exiting, we never
 440                  * linger..
 441                  */
 442                 timeout = 0;
 443                 if (sock_flag(sk, SOCK_LINGER) &&
 444                     !(current->flags & PF_EXITING))
 445                         timeout = sk->sk_lingertime;
 446                 sock->sk = NULL;
 447                 sk->sk_prot->close(sk, timeout);
 448         }
 449         return 0;
 450 }
 451 EXPORT_SYMBOL(inet_release);
 452
 453 /* It is off by default, see below. */
 454 int sysctl_ip_nonlocal_bind __read_mostly;
 455 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
 456
 457 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 458 {
 459         struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
 460         struct sock *sk = sock->sk;
 461         struct inet_sock *inet = inet_sk(sk);
 462         struct net *net = sock_net(sk);
 463         unsigned short snum;
 464         int chk_addr_ret;
 465         int err;
 466
 467         /* If the socket has its own bind function then use it. (RAW) */
 468         if (sk->sk_prot->bind) {
 469                 err = sk->sk_prot->bind(sk, uaddr, addr_len);
 470                 goto out;
 471         }
 472         err = -EINVAL;
 473         if (addr_len < sizeof(struct sockaddr_in))
 474                 goto out;
 475
 476         if (addr->sin_family != AF_INET) {
 477                 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
 478                  * only if s_addr is INADDR_ANY.
 479                  */
 480                 err = -EAFNOSUPPORT;
 481                 if (addr->sin_family != AF_UNSPEC ||
 482                     addr->sin_addr.s_addr != htonl(INADDR_ANY))
 483                         goto out;
 484         }
 485
 486         chk_addr_ret = inet_addr_type(net, addr->sin_addr.s_addr);
 487
 488         /* Not specified by any standard per-se, however it breaks too
 489          * many applications when removed.  It is unfortunate since
 490          * allowing applications to make a non-local bind solves
 491          * several problems with systems using dynamic addressing.
 492          * (ie. your servers still start up even if your ISDN link
 493          *  is temporarily down)
 494          */
 495         err = -EADDRNOTAVAIL;
 496         if (!sysctl_ip_nonlocal_bind &&
 497             !(inet->freebind || inet->transparent) &&
 498             addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
 499             chk_addr_ret != RTN_LOCAL &&
 500             chk_addr_ret != RTN_MULTICAST &&
 501             chk_addr_ret != RTN_BROADCAST)
 502                 goto out;
 503
 504         snum = ntohs(addr->sin_port);
 505         err = -EACCES;
 506         if (snum && snum < PROT_SOCK &&
 507             !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
 508                 goto out;
 509
 510         /*      We keep a pair of addresses. rcv_saddr is the one
 511          *      used by hash lookups, and saddr is used for transmit.
 512          *
 513          *      In the BSD API these are the same except where it
 514          *      would be illegal to use them (multicast/broadcast) in
 515          *      which case the sending device address is used.
 516          */
 517         lock_sock(sk);
 518
 519         /* Check these errors (active socket, double bind). */
 520         err = -EINVAL;
 521         if (sk->sk_state != TCP_CLOSE || inet->inet_num)
 522                 goto out_release_sock;
 523
 524         inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
 525         if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
 526                 inet->inet_saddr = 0;  /* Use device */
 527
 528         /* Make sure we are allowed to bind here. */
 529         if (sk->sk_prot->get_port(sk, snum)) {
 530                 inet->inet_saddr = inet->inet_rcv_saddr = 0;
 531                 err = -EADDRINUSE;
 532                 goto out_release_sock;
 533         }
 534
 535         if (inet->inet_rcv_saddr)
 536                 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
 537         if (snum)
 538                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
 539         inet->inet_sport = htons(inet->inet_num);
 540         inet->inet_daddr = 0;
 541         inet->inet_dport = 0;
 542         sk_dst_reset(sk);
 543         err = 0;
 544 out_release_sock:
 545         release_sock(sk);
 546 out:
 547         return err;
 548 }
 549 EXPORT_SYMBOL(inet_bind);
 550
 551 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
 552                        int addr_len, int flags)
 553 {
 554         struct sock *sk = sock->sk;
 555
 556         if (addr_len < sizeof(uaddr->sa_family))
 557                 return -EINVAL;
 558         if (uaddr->sa_family == AF_UNSPEC)
 559                 return sk->sk_prot->disconnect(sk, flags);
 560
 561         if (!inet_sk(sk)->inet_num && inet_autobind(sk))
 562                 return -EAGAIN;
 563         return sk->sk_prot->connect(sk, uaddr, addr_len);
 564 }
 565 EXPORT_SYMBOL(inet_dgram_connect);
 566
 567 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
 568 {
 569         DEFINE_WAIT(wait);
 570
 571         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
 572         sk->sk_write_pending += writebias;
 573
 574         /* Basic assumption: if someone sets sk->sk_err, he _must_
 575          * change state of the socket from TCP_SYN_*.
 576          * Connect() does not allow to get error notifications
 577          * without closing the socket.
 578          */
 579         while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 580                 release_sock(sk);
 581                 timeo = schedule_timeout(timeo);
 582                 lock_sock(sk);
 583                 if (signal_pending(current) || !timeo)
 584                         break;
 585                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
 586         }
 587         finish_wait(sk_sleep(sk), &wait);
 588         sk->sk_write_pending -= writebias;
 589         return timeo;
 590 }
 591
 592 /*
 593  *      Connect to a remote host. There is regrettably still a little
 594  *      TCP 'magic' in here.
 595  */
 596 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 597                           int addr_len, int flags)
 598 {
 599         struct sock *sk = sock->sk;
 600         int err;
 601         long timeo;
 602
 603         if (addr_len < sizeof(uaddr->sa_family))
 604                 return -EINVAL;
 605
 606         if (uaddr->sa_family == AF_UNSPEC) {
 607                 err = sk->sk_prot->disconnect(sk, flags);
 608                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 609                 goto out;
 610         }
 611
 612         switch (sock->state) {
 613         default:
 614                 err = -EINVAL;
 615                 goto out;
 616         case SS_CONNECTED:
 617                 err = -EISCONN;
 618                 goto out;
 619         case SS_CONNECTING:
 620                 err = -EALREADY;
 621                 /* Fall out of switch with err, set for this state */
 622                 break;
 623         case SS_UNCONNECTED:
 624                 err = -EISCONN;
 625                 if (sk->sk_state != TCP_CLOSE)
 626                         goto out;
 627
 628                 err = sk->sk_prot->connect(sk, uaddr, addr_len);
 629                 if (err < 0)
 630                         goto out;
 631
 632                 sock->state = SS_CONNECTING;
 633
 634                 /* Just entered SS_CONNECTING state; the only
 635                  * difference is that return value in non-blocking
 636                  * case is EINPROGRESS, rather than EALREADY.
 637                  */
 638                 err = -EINPROGRESS;
 639                 break;
 640         }
 641
 642         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
 643
 644         if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 645                 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
 646                                 tcp_sk(sk)->fastopen_req &&
 647                                 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
 648
 649                 /* Error code is set above */
 650                 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
 651                         goto out;
 652
 653                 err = sock_intr_errno(timeo);
 654                 if (signal_pending(current))
 655                         goto out;
 656         }
 657
 658         /* Connection was closed by RST, timeout, ICMP error
 659          * or another process disconnected us.
 660          */
 661         if (sk->sk_state == TCP_CLOSE)
 662                 goto sock_error;
 663
 664         /* sk->sk_err may be not zero now, if RECVERR was ordered by user
 665          * and error was received after socket entered established state.
 666          * Hence, it is handled normally after connect() return successfully.
 667          */
 668
 669         sock->state = SS_CONNECTED;
 670         err = 0;
 671 out:
 672         return err;
 673
 674 sock_error:
 675         err = sock_error(sk) ? : -ECONNABORTED;
 676         sock->state = SS_UNCONNECTED;
 677         if (sk->sk_prot->disconnect(sk, flags))
 678                 sock->state = SS_DISCONNECTING;
 679         goto out;
 680 }
 681 EXPORT_SYMBOL(__inet_stream_connect);
 682
 683 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 684                         int addr_len, int flags)
 685 {
 686         int err;
 687
 688         lock_sock(sock->sk);
 689         err = __inet_stream_connect(sock, uaddr, addr_len, flags);
 690         release_sock(sock->sk);
 691         return err;
 692 }
 693 EXPORT_SYMBOL(inet_stream_connect);
 694
 695 /*
 696  *      Accept a pending connection. The TCP layer now gives BSD semantics.
 697  */
 698
 699 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
 700 {
 701         struct sock *sk1 = sock->sk;
 702         int err = -EINVAL;
 703         struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
 704
 705         if (!sk2)
 706                 goto do_err;
 707
 708         lock_sock(sk2);
 709
 710         sock_rps_record_flow(sk2);
 711         WARN_ON(!((1 << sk2->sk_state) &
 712                   (TCPF_ESTABLISHED | TCPF_SYN_RECV |
 713                   TCPF_CLOSE_WAIT | TCPF_CLOSE)));
 714
 715         sock_graft(sk2, newsock);
 716
 717         newsock->state = SS_CONNECTED;
 718         err = 0;
 719         release_sock(sk2);
 720 do_err:
 721         return err;
 722 }
 723 EXPORT_SYMBOL(inet_accept);
 724
 725
 726 /*
 727  *      This does both peername and sockname.
 728  */
 729 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
 730                         int *uaddr_len, int peer)
 731 {
 732         struct sock *sk         = sock->sk;
 733         struct inet_sock *inet  = inet_sk(sk);
 734         DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
 735
 736         sin->sin_family = AF_INET;
 737         if (peer) {
 738                 if (!inet->inet_dport ||
 739                     (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
 740                      peer == 1))
 741                         return -ENOTCONN;
 742                 sin->sin_port = inet->inet_dport;
 743                 sin->sin_addr.s_addr = inet->inet_daddr;
 744         } else {
 745                 __be32 addr = inet->inet_rcv_saddr;
 746                 if (!addr)
 747                         addr = inet->inet_saddr;
 748                 sin->sin_port = inet->inet_sport;
 749                 sin->sin_addr.s_addr = addr;
 750         }
 751         memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 752         *uaddr_len = sizeof(*sin);
 753         return 0;
 754 }
 755 EXPORT_SYMBOL(inet_getname);
 756
 757 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
 758                  size_t size)
 759 {
 760         struct sock *sk = sock->sk;
 761
 762         sock_rps_record_flow(sk);
 763
 764         /* We may need to bind the socket. */
 765         if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
 766             inet_autobind(sk))
 767                 return -EAGAIN;
 768
 769         return sk->sk_prot->sendmsg(iocb, sk, msg, size);
 770 }
 771 EXPORT_SYMBOL(inet_sendmsg);
 772
 773 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
 774                       size_t size, int flags)
 775 {
 776         struct sock *sk = sock->sk;
 777
 778         sock_rps_record_flow(sk);
 779
 780         /* We may need to bind the socket. */
 781         if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
 782             inet_autobind(sk))
 783                 return -EAGAIN;
 784
 785         if (sk->sk_prot->sendpage)
 786                 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
 787         return sock_no_sendpage(sock, page, offset, size, flags);
 788 }
 789 EXPORT_SYMBOL(inet_sendpage);
 790
 791 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
 792                  size_t size, int flags)
 793 {
 794         struct sock *sk = sock->sk;
 795         int addr_len = 0;
 796         int err;
 797
 798         sock_rps_record_flow(sk);
 799
 800         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
 801                                    flags & ~MSG_DONTWAIT, &addr_len);
 802         if (err >= 0)
 803                 msg->msg_namelen = addr_len;
 804         return err;
 805 }
 806 EXPORT_SYMBOL(inet_recvmsg);
 807
 808 int inet_shutdown(struct socket *sock, int how)
 809 {
 810         struct sock *sk = sock->sk;
 811         int err = 0;
 812
 813         /* This should really check to make sure
 814          * the socket is a TCP socket. (WHY AC...)
 815          */
 816         how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
 817                        1->2 bit 2 snds.
 818                        2->3 */
 819         if ((how & ~SHUTDOWN_MASK) || !how)     /* MAXINT->0 */
 820                 return -EINVAL;
 821
 822         lock_sock(sk);
 823         if (sock->state == SS_CONNECTING) {
 824                 if ((1 << sk->sk_state) &
 825                     (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
 826                         sock->state = SS_DISCONNECTING;
 827                 else
 828                         sock->state = SS_CONNECTED;
 829         }
 830
 831         switch (sk->sk_state) {
 832         case TCP_CLOSE:
 833                 err = -ENOTCONN;
 834                 /* Hack to wake up other listeners, who can poll for
 835                    POLLHUP, even on eg. unconnected UDP sockets -- RR */
 836         default:
 837                 sk->sk_shutdown |= how;
 838                 if (sk->sk_prot->shutdown)
 839                         sk->sk_prot->shutdown(sk, how);
 840                 break;
 841
 842         /* Remaining two branches are temporary solution for missing
 843          * close() in multithreaded environment. It is _not_ a good idea,
 844          * but we have no choice until close() is repaired at VFS level.
 845          */
 846         case TCP_LISTEN:
 847                 if (!(how & RCV_SHUTDOWN))
 848                         break;
 849                 /* Fall through */
 850         case TCP_SYN_SENT:
 851                 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
 852                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 853                 break;
 854         }
 855
 856         /* Wake up anyone sleeping in poll. */
 857         sk->sk_state_change(sk);
 858         release_sock(sk);
 859         return err;
 860 }
 861 EXPORT_SYMBOL(inet_shutdown);
 862
 863 /*
 864  *      ioctl() calls you can issue on an INET socket. Most of these are
 865  *      device configuration and stuff and very rarely used. Some ioctls
 866  *      pass on to the socket itself.
 867  *
 868  *      NOTE: I like the idea of a module for the config stuff. ie ifconfig
 869  *      loads the devconfigure module does its configuring and unloads it.
 870  *      There's a good 20K of config code hanging around the kernel.
 871  */
 872
 873 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 874 {
 875         struct sock *sk = sock->sk;
 876         int err = 0;
 877         struct net *net = sock_net(sk);
 878
 879         switch (cmd) {
 880         case SIOCGSTAMP:
 881                 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
 882                 break;
 883         case SIOCGSTAMPNS:
 884                 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
 885                 break;
 886         case SIOCADDRT:
 887         case SIOCDELRT:
 888         case SIOCRTMSG:
 889                 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
 890                 break;
 891         case SIOCDARP:
 892         case SIOCGARP:
 893         case SIOCSARP:
 894                 err = arp_ioctl(net, cmd, (void __user *)arg);
 895                 break;
 896         case SIOCGIFADDR:
 897         case SIOCSIFADDR:
 898         case SIOCGIFBRDADDR:
 899         case SIOCSIFBRDADDR:
 900         case SIOCGIFNETMASK:
 901         case SIOCSIFNETMASK:
 902         case SIOCGIFDSTADDR:
 903         case SIOCSIFDSTADDR:
 904         case SIOCSIFPFLAGS:
 905         case SIOCGIFPFLAGS:
 906         case SIOCSIFFLAGS:
 907                 err = devinet_ioctl(net, cmd, (void __user *)arg);
 908                 break;
 909         default:
 910                 if (sk->sk_prot->ioctl)
 911                         err = sk->sk_prot->ioctl(sk, cmd, arg);
 912                 else
 913                         err = -ENOIOCTLCMD;
 914                 break;
 915         }
 916         return err;
 917 }
 918 EXPORT_SYMBOL(inet_ioctl);
 919
 920 #ifdef CONFIG_COMPAT
 921 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 922 {
 923         struct sock *sk = sock->sk;
 924         int err = -ENOIOCTLCMD;
 925
 926         if (sk->sk_prot->compat_ioctl)
 927                 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
 928
 929         return err;
 930 }
 931 #endif
 932
 933 const struct proto_ops inet_stream_ops = {
 934         .family            = PF_INET,
 935         .owner             = THIS_MODULE,
 936         .release           = inet_release,
 937         .bind              = inet_bind,
 938         .connect           = inet_stream_connect,
 939         .socketpair        = sock_no_socketpair,
 940         .accept            = inet_accept,
 941         .getname           = inet_getname,
 942         .poll              = tcp_poll,
 943         .ioctl             = inet_ioctl,
 944         .listen            = inet_listen,
 945         .shutdown          = inet_shutdown,
 946         .setsockopt        = sock_common_setsockopt,
 947         .getsockopt        = sock_common_getsockopt,
 948         .sendmsg           = inet_sendmsg,
 949         .recvmsg           = inet_recvmsg,
 950         .mmap              = sock_no_mmap,
 951         .sendpage          = inet_sendpage,
 952         .splice_read       = tcp_splice_read,
 953 #ifdef CONFIG_COMPAT
 954         .compat_setsockopt = compat_sock_common_setsockopt,
 955         .compat_getsockopt = compat_sock_common_getsockopt,
 956         .compat_ioctl      = inet_compat_ioctl,
 957 #endif
 958 };
 959 EXPORT_SYMBOL(inet_stream_ops);
 960
 961 const struct proto_ops inet_dgram_ops = {
 962         .family            = PF_INET,
 963         .owner             = THIS_MODULE,
 964         .release           = inet_release,
 965         .bind              = inet_bind,
 966         .connect           = inet_dgram_connect,
 967         .socketpair        = sock_no_socketpair,
 968         .accept            = sock_no_accept,
 969         .getname           = inet_getname,
 970         .poll              = udp_poll,
 971         .ioctl             = inet_ioctl,
 972         .listen            = sock_no_listen,
 973         .shutdown          = inet_shutdown,
 974         .setsockopt        = sock_common_setsockopt,
 975         .getsockopt        = sock_common_getsockopt,
 976         .sendmsg           = inet_sendmsg,
 977         .recvmsg           = inet_recvmsg,
 978         .mmap              = sock_no_mmap,
 979         .sendpage          = inet_sendpage,
 980 #ifdef CONFIG_COMPAT
 981         .compat_setsockopt = compat_sock_common_setsockopt,
 982         .compat_getsockopt = compat_sock_common_getsockopt,
 983         .compat_ioctl      = inet_compat_ioctl,
 984 #endif
 985 };
 986 EXPORT_SYMBOL(inet_dgram_ops);
 987
 988 /*
 989  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
 990  * udp_poll
 991  */
 992 static const struct proto_ops inet_sockraw_ops = {
 993         .family            = PF_INET,
 994         .owner             = THIS_MODULE,
 995         .release           = inet_release,
 996         .bind              = inet_bind,
 997         .connect           = inet_dgram_connect,
 998         .socketpair        = sock_no_socketpair,
 999         .accept            = sock_no_accept,
1000         .getname           = inet_getname,
1001         .poll              = datagram_poll,
1002         .ioctl             = inet_ioctl,
1003         .listen            = sock_no_listen,
1004         .shutdown          = inet_shutdown,
1005         .setsockopt        = sock_common_setsockopt,
1006         .getsockopt        = sock_common_getsockopt,
1007         .sendmsg           = inet_sendmsg,
1008         .recvmsg           = inet_recvmsg,
1009         .mmap              = sock_no_mmap,
1010         .sendpage          = inet_sendpage,
1011 #ifdef CONFIG_COMPAT
1012         .compat_setsockopt = compat_sock_common_setsockopt,
1013         .compat_getsockopt = compat_sock_common_getsockopt,
1014         .compat_ioctl      = inet_compat_ioctl,
1015 #endif
1016 };
1017
1018 static const struct net_proto_family inet_family_ops = {
1019         .family = PF_INET,
1020         .create = inet_create,
1021         .owner  = THIS_MODULE,
1022 };
1023
1024 /* Upon startup we insert all the elements in inetsw_array[] into
1025  * the linked list inetsw.
1026  */
1027 static struct inet_protosw inetsw_array[] =
1028 {
1029         {
1030                 .type =       SOCK_STREAM,
1031                 .protocol =   IPPROTO_TCP,
1032                 .prot =       &tcp_prot,
1033                 .ops =        &inet_stream_ops,
1034                 .no_check =   0,
1035                 .flags =      INET_PROTOSW_PERMANENT |
1036                               INET_PROTOSW_ICSK,
1037         },
1038
1039         {
1040                 .type =       SOCK_DGRAM,
1041                 .protocol =   IPPROTO_UDP,
1042                 .prot =       &udp_prot,
1043                 .ops =        &inet_dgram_ops,
1044                 .no_check =   UDP_CSUM_DEFAULT,
1045                 .flags =      INET_PROTOSW_PERMANENT,
1046        },
1047
1048        {
1049                 .type =       SOCK_DGRAM,
1050                 .protocol =   IPPROTO_ICMP,
1051                 .prot =       &ping_prot,
1052                 .ops =        &inet_dgram_ops,
1053                 .no_check =   UDP_CSUM_DEFAULT,
1054                 .flags =      INET_PROTOSW_REUSE,
1055        },
1056
1057        {
1058                .type =       SOCK_RAW,
1059                .protocol =   IPPROTO_IP,        /* wild card */
1060                .prot =       &raw_prot,
1061                .ops =        &inet_sockraw_ops,
1062                .no_check =   UDP_CSUM_DEFAULT,
1063                .flags =      INET_PROTOSW_REUSE,
1064        }
1065 };
1066
1067 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1068
1069 void inet_register_protosw(struct inet_protosw *p)
1070 {
1071         struct list_head *lh;
1072         struct inet_protosw *answer;
1073         int protocol = p->protocol;
1074         struct list_head *last_perm;
1075
1076         spin_lock_bh(&inetsw_lock);
1077
1078         if (p->type >= SOCK_MAX)
1079                 goto out_illegal;
1080
1081         /* If we are trying to override a permanent protocol, bail. */
1082         answer = NULL;
1083         last_perm = &inetsw[p->type];
1084         list_for_each(lh, &inetsw[p->type]) {
1085                 answer = list_entry(lh, struct inet_protosw, list);
1086
1087                 /* Check only the non-wild match. */
1088                 if (INET_PROTOSW_PERMANENT & answer->flags) {
1089                         if (protocol == answer->protocol)
1090                                 break;
1091                         last_perm = lh;
1092                 }
1093
1094                 answer = NULL;
1095         }
1096         if (answer)
1097                 goto out_permanent;
1098
1099         /* Add the new entry after the last permanent entry if any, so that
1100          * the new entry does not override a permanent entry when matched with
1101          * a wild-card protocol. But it is allowed to override any existing
1102          * non-permanent entry.  This means that when we remove this entry, the
1103          * system automatically returns to the old behavior.
1104          */
1105         list_add_rcu(&p->list, last_perm);
1106 out:
1107         spin_unlock_bh(&inetsw_lock);
1108
1109         return;
1110
1111 out_permanent:
1112         pr_err("Attempt to override permanent protocol %d\n", protocol);
1113         goto out;
1114
1115 out_illegal:
1116         pr_err("Ignoring attempt to register invalid socket type %d\n",
1117                p->type);
1118         goto out;
1119 }
1120 EXPORT_SYMBOL(inet_register_protosw);
1121
1122 void inet_unregister_protosw(struct inet_protosw *p)
1123 {
1124         if (INET_PROTOSW_PERMANENT & p->flags) {
1125                 pr_err("Attempt to unregister permanent protocol %d\n",
1126                        p->protocol);
1127         } else {
1128                 spin_lock_bh(&inetsw_lock);
1129                 list_del_rcu(&p->list);
1130                 spin_unlock_bh(&inetsw_lock);
1131
1132                 synchronize_net();
1133         }
1134 }
1135 EXPORT_SYMBOL(inet_unregister_protosw);
1136
1137 /*
1138  *      Shall we try to damage output packets if routing dev changes?
1139  */
1140
1141 int sysctl_ip_dynaddr __read_mostly;
1142
1143 static int inet_sk_reselect_saddr(struct sock *sk)
1144 {
1145         struct inet_sock *inet = inet_sk(sk);
1146         __be32 old_saddr = inet->inet_saddr;
1147         __be32 daddr = inet->inet_daddr;
1148         struct flowi4 *fl4;
1149         struct rtable *rt;
1150         __be32 new_saddr;
1151         struct ip_options_rcu *inet_opt;
1152
1153         inet_opt = rcu_dereference_protected(inet->inet_opt,
1154                                              sock_owned_by_user(sk));
1155         if (inet_opt && inet_opt->opt.srr)
1156                 daddr = inet_opt->opt.faddr;
1157
1158         /* Query new route. */
1159         fl4 = &inet->cork.fl.u.ip4;
1160         rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1161                               sk->sk_bound_dev_if, sk->sk_protocol,
1162                               inet->inet_sport, inet->inet_dport, sk, false);
1163         if (IS_ERR(rt))
1164                 return PTR_ERR(rt);
1165
1166         sk_setup_caps(sk, &rt->dst);
1167
1168         new_saddr = fl4->saddr;
1169
1170         if (new_saddr == old_saddr)
1171                 return 0;
1172
1173         if (sysctl_ip_dynaddr > 1) {
1174                 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1175                         __func__, &old_saddr, &new_saddr);
1176         }
1177
1178         inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1179
1180         /*
1181          * XXX The only one ugly spot where we need to
1182          * XXX really change the sockets identity after
1183          * XXX it has entered the hashes. -DaveM
1184          *
1185          * Besides that, it does not check for connection
1186          * uniqueness. Wait for troubles.
1187          */
1188         __sk_prot_rehash(sk);
1189         return 0;
1190 }
1191
1192 int inet_sk_rebuild_header(struct sock *sk)
1193 {
1194         struct inet_sock *inet = inet_sk(sk);
1195         struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1196         __be32 daddr;
1197         struct ip_options_rcu *inet_opt;
1198         struct flowi4 *fl4;
1199         int err;
1200
1201         /* Route is OK, nothing to do. */
1202         if (rt)
1203                 return 0;
1204
1205         /* Reroute. */
1206         rcu_read_lock();
1207         inet_opt = rcu_dereference(inet->inet_opt);
1208         daddr = inet->inet_daddr;
1209         if (inet_opt && inet_opt->opt.srr)
1210                 daddr = inet_opt->opt.faddr;
1211         rcu_read_unlock();
1212         fl4 = &inet->cork.fl.u.ip4;
1213         rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1214                                    inet->inet_dport, inet->inet_sport,
1215                                    sk->sk_protocol, RT_CONN_FLAGS(sk),
1216                                    sk->sk_bound_dev_if);
1217         if (!IS_ERR(rt)) {
1218                 err = 0;
1219                 sk_setup_caps(sk, &rt->dst);
1220         } else {
1221                 err = PTR_ERR(rt);
1222
1223                 /* Routing failed... */
1224                 sk->sk_route_caps = 0;
1225                 /*
1226                  * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1227                  * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1228                  */
1229                 if (!sysctl_ip_dynaddr ||
1230                     sk->sk_state != TCP_SYN_SENT ||
1231                     (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1232                     (err = inet_sk_reselect_saddr(sk)) != 0)
1233                         sk->sk_err_soft = -err;
1234         }
1235
1236         return err;
1237 }
1238 EXPORT_SYMBOL(inet_sk_rebuild_header);
1239
1240 static int inet_gso_send_check(struct sk_buff *skb)
1241 {
1242         const struct net_offload *ops;
1243         const struct iphdr *iph;
1244         int proto;
1245         int ihl;
1246         int err = -EINVAL;
1247
1248         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1249                 goto out;
1250
1251         iph = ip_hdr(skb);
1252         ihl = iph->ihl * 4;
1253         if (ihl < sizeof(*iph))
1254                 goto out;
1255
1256         if (unlikely(!pskb_may_pull(skb, ihl)))
1257                 goto out;
1258
1259         __skb_pull(skb, ihl);
1260         skb_reset_transport_header(skb);
1261         iph = ip_hdr(skb);
1262         proto = iph->protocol;
1263         err = -EPROTONOSUPPORT;
1264
1265         rcu_read_lock();
1266         ops = rcu_dereference(inet_offloads[proto]);
1267         if (likely(ops && ops->callbacks.gso_send_check))
1268                 err = ops->callbacks.gso_send_check(skb);
1269         rcu_read_unlock();
1270
1271 out:
1272         return err;
1273 }
1274
1275 static struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1276         netdev_features_t features)
1277 {
1278         struct sk_buff *segs = ERR_PTR(-EINVAL);
1279         const struct net_offload *ops;
1280         struct iphdr *iph;
1281         int proto;
1282         int ihl;
1283         int id;
1284         unsigned int offset = 0;
1285         bool tunnel;
1286
1287         if (unlikely(skb_shinfo(skb)->gso_type &
1288                      ~(SKB_GSO_TCPV4 |
1289                        SKB_GSO_UDP |
1290                        SKB_GSO_DODGY |
1291                        SKB_GSO_TCP_ECN |
1292                        SKB_GSO_GRE |
1293                        SKB_GSO_UDP_TUNNEL |
1294                        0)))
1295                 goto out;
1296
1297         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1298                 goto out;
1299
1300         iph = ip_hdr(skb);
1301         ihl = iph->ihl * 4;
1302         if (ihl < sizeof(*iph))
1303                 goto out;
1304
1305         if (unlikely(!pskb_may_pull(skb, ihl)))
1306                 goto out;
1307
1308         tunnel = !!skb->encapsulation;
1309
1310         __skb_pull(skb, ihl);
1311         skb_reset_transport_header(skb);
1312         iph = ip_hdr(skb);
1313         id = ntohs(iph->id);
1314         proto = iph->protocol;
1315         segs = ERR_PTR(-EPROTONOSUPPORT);
1316
1317         rcu_read_lock();
1318         ops = rcu_dereference(inet_offloads[proto]);
1319         if (likely(ops && ops->callbacks.gso_segment))
1320                 segs = ops->callbacks.gso_segment(skb, features);
1321         rcu_read_unlock();
1322
1323         if (IS_ERR_OR_NULL(segs))
1324                 goto out;
1325
1326         skb = segs;
1327         do {
1328                 iph = ip_hdr(skb);
1329                 if (!tunnel && proto == IPPROTO_UDP) {
1330                         iph->id = htons(id);
1331                         iph->frag_off = htons(offset >> 3);
1332                         if (skb->next != NULL)
1333                                 iph->frag_off |= htons(IP_MF);
1334                         offset += (skb->len - skb->mac_len - iph->ihl * 4);
1335                 } else  {
1336                         iph->id = htons(id++);
1337                 }
1338                 iph->tot_len = htons(skb->len - skb->mac_len);
1339                 iph->check = 0;
1340                 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1341         } while ((skb = skb->next));
1342
1343 out:
1344         return segs;
1345 }
1346
1347 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1348                                          struct sk_buff *skb)
1349 {
1350         const struct net_offload *ops;
1351         struct sk_buff **pp = NULL;
1352         struct sk_buff *p;
1353         const struct iphdr *iph;
1354         unsigned int hlen;
1355         unsigned int off;
1356         unsigned int id;
1357         int flush = 1;
1358         int proto;
1359
1360         off = skb_gro_offset(skb);
1361         hlen = off + sizeof(*iph);
1362         iph = skb_gro_header_fast(skb, off);
1363         if (skb_gro_header_hard(skb, hlen)) {
1364                 iph = skb_gro_header_slow(skb, hlen, off);
1365                 if (unlikely(!iph))
1366                         goto out;
1367         }
1368
1369         proto = iph->protocol;
1370
1371         rcu_read_lock();
1372         ops = rcu_dereference(inet_offloads[proto]);
1373         if (!ops || !ops->callbacks.gro_receive)
1374                 goto out_unlock;
1375
1376         if (*(u8 *)iph != 0x45)
1377                 goto out_unlock;
1378
1379         if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1380                 goto out_unlock;
1381
1382         id = ntohl(*(__be32 *)&iph->id);
1383         flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF));
1384         id >>= 16;
1385
1386         for (p = *head; p; p = p->next) {
1387                 struct iphdr *iph2;
1388
1389                 if (!NAPI_GRO_CB(p)->same_flow)
1390                         continue;
1391
1392                 iph2 = ip_hdr(p);
1393
1394                 if ((iph->protocol ^ iph2->protocol) |
1395                     ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1396                     ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1397                         NAPI_GRO_CB(p)->same_flow = 0;
1398                         continue;
1399                 }
1400
1401                 /* All fields must match except length and checksum. */
1402                 NAPI_GRO_CB(p)->flush |=
1403                         (iph->ttl ^ iph2->ttl) |
1404                         (iph->tos ^ iph2->tos) |
1405                         ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1406
1407                 NAPI_GRO_CB(p)->flush |= flush;
1408         }
1409
1410         NAPI_GRO_CB(skb)->flush |= flush;
1411         skb_gro_pull(skb, sizeof(*iph));
1412         skb_set_transport_header(skb, skb_gro_offset(skb));
1413
1414         pp = ops->callbacks.gro_receive(head, skb);
1415
1416 out_unlock:
1417         rcu_read_unlock();
1418
1419 out:
1420         NAPI_GRO_CB(skb)->flush |= flush;
1421
1422         return pp;
1423 }
1424
1425 static int inet_gro_complete(struct sk_buff *skb)
1426 {
1427         __be16 newlen = htons(skb->len - skb_network_offset(skb));
1428         struct iphdr *iph = ip_hdr(skb);
1429         const struct net_offload *ops;
1430         int proto = iph->protocol;
1431         int err = -ENOSYS;
1432
1433         csum_replace2(&iph->check, iph->tot_len, newlen);
1434         iph->tot_len = newlen;
1435
1436         rcu_read_lock();
1437         ops = rcu_dereference(inet_offloads[proto]);
1438         if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1439                 goto out_unlock;
1440
1441         err = ops->callbacks.gro_complete(skb);
1442
1443 out_unlock:
1444         rcu_read_unlock();
1445
1446         return err;
1447 }
1448
1449 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1450                          unsigned short type, unsigned char protocol,
1451                          struct net *net)
1452 {
1453         struct socket *sock;
1454         int rc = sock_create_kern(family, type, protocol, &sock);
1455
1456         if (rc == 0) {
1457                 *sk = sock->sk;
1458                 (*sk)->sk_allocation = GFP_ATOMIC;
1459                 /*
1460                  * Unhash it so that IP input processing does not even see it,
1461                  * we do not wish this socket to see incoming packets.
1462                  */
1463                 (*sk)->sk_prot->unhash(*sk);
1464
1465                 sk_change_net(*sk, net);
1466         }
1467         return rc;
1468 }
1469 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1470
1471 unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1472 {
1473         unsigned long res = 0;
1474         int i, j;
1475
1476         for_each_possible_cpu(i) {
1477                 for (j = 0; j < SNMP_ARRAY_SZ; j++)
1478                         res += *(((unsigned long *) per_cpu_ptr(mib[j], i)) + offt);
1479         }
1480         return res;
1481 }
1482 EXPORT_SYMBOL_GPL(snmp_fold_field);
1483
1484 #if BITS_PER_LONG==32
1485
1486 u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset)
1487 {
1488         u64 res = 0;
1489         int cpu;
1490
1491         for_each_possible_cpu(cpu) {
1492                 void *bhptr;
1493                 struct u64_stats_sync *syncp;
1494                 u64 v;
1495                 unsigned int start;
1496
1497                 bhptr = per_cpu_ptr(mib[0], cpu);
1498                 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1499                 do {
1500                         start = u64_stats_fetch_begin_bh(syncp);
1501                         v = *(((u64 *) bhptr) + offt);
1502                 } while (u64_stats_fetch_retry_bh(syncp, start));
1503
1504                 res += v;
1505         }
1506         return res;
1507 }
1508 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1509 #endif
1510
1511 int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align)
1512 {
1513         BUG_ON(ptr == NULL);
1514         ptr[0] = __alloc_percpu(mibsize, align);
1515         if (!ptr[0])
1516                 return -ENOMEM;
1517 #if SNMP_ARRAY_SZ == 2
1518         ptr[1] = __alloc_percpu(mibsize, align);
1519         if (!ptr[1]) {
1520                 free_percpu(ptr[0]);
1521                 ptr[0] = NULL;
1522                 return -ENOMEM;
1523         }
1524 #endif
1525         return 0;
1526 }
1527 EXPORT_SYMBOL_GPL(snmp_mib_init);
1528
1529 void snmp_mib_free(void __percpu *ptr[SNMP_ARRAY_SZ])
1530 {
1531         int i;
1532
1533         BUG_ON(ptr == NULL);
1534         for (i = 0; i < SNMP_ARRAY_SZ; i++) {
1535                 free_percpu(ptr[i]);
1536                 ptr[i] = NULL;
1537         }
1538 }
1539 EXPORT_SYMBOL_GPL(snmp_mib_free);
1540
1541 #ifdef CONFIG_IP_MULTICAST
1542 static const struct net_protocol igmp_protocol = {
1543         .handler =      igmp_rcv,
1544         .netns_ok =     1,
1545 };
1546 #endif
1547
1548 static const struct net_protocol tcp_protocol = {
1549         .early_demux    =       tcp_v4_early_demux,
1550         .handler        =       tcp_v4_rcv,
1551         .err_handler    =       tcp_v4_err,
1552         .no_policy      =       1,
1553         .netns_ok       =       1,
1554 };
1555
1556 static const struct net_offload tcp_offload = {
1557         .callbacks = {
1558                 .gso_send_check =       tcp_v4_gso_send_check,
1559                 .gso_segment    =       tcp_tso_segment,
1560                 .gro_receive    =       tcp4_gro_receive,
1561                 .gro_complete   =       tcp4_gro_complete,
1562         },
1563 };
1564
1565 static const struct net_protocol udp_protocol = {
1566         .handler =      udp_rcv,
1567         .err_handler =  udp_err,
1568         .no_policy =    1,
1569         .netns_ok =     1,
1570 };
1571
1572 static const struct net_offload udp_offload = {
1573         .callbacks = {
1574                 .gso_send_check = udp4_ufo_send_check,
1575                 .gso_segment = udp4_ufo_fragment,
1576         },
1577 };
1578
1579 static const struct net_protocol icmp_protocol = {
1580         .handler =      icmp_rcv,
1581         .err_handler =  icmp_err,
1582         .no_policy =    1,
1583         .netns_ok =     1,
1584 };
1585
1586 static __net_init int ipv4_mib_init_net(struct net *net)
1587 {
1588         if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1589                           sizeof(struct tcp_mib),
1590                           __alignof__(struct tcp_mib)) < 0)
1591                 goto err_tcp_mib;
1592         if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1593                           sizeof(struct ipstats_mib),
1594                           __alignof__(struct ipstats_mib)) < 0)
1595                 goto err_ip_mib;
1596         if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1597                           sizeof(struct linux_mib),
1598                           __alignof__(struct linux_mib)) < 0)
1599                 goto err_net_mib;
1600         if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1601                           sizeof(struct udp_mib),
1602                           __alignof__(struct udp_mib)) < 0)
1603                 goto err_udp_mib;
1604         if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1605                           sizeof(struct udp_mib),
1606                           __alignof__(struct udp_mib)) < 0)
1607                 goto err_udplite_mib;
1608         if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1609                           sizeof(struct icmp_mib),
1610                           __alignof__(struct icmp_mib)) < 0)
1611                 goto err_icmp_mib;
1612         net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1613                                               GFP_KERNEL);
1614         if (!net->mib.icmpmsg_statistics)
1615                 goto err_icmpmsg_mib;
1616
1617         tcp_mib_init(net);
1618         return 0;
1619
1620 err_icmpmsg_mib:
1621         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1622 err_icmp_mib:
1623         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1624 err_udplite_mib:
1625         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1626 err_udp_mib:
1627         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1628 err_net_mib:
1629         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1630 err_ip_mib:
1631         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1632 err_tcp_mib:
1633         return -ENOMEM;
1634 }
1635
1636 static __net_exit void ipv4_mib_exit_net(struct net *net)
1637 {
1638         kfree(net->mib.icmpmsg_statistics);
1639         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1640         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1641         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1642         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1643         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1644         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1645 }
1646
1647 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1648         .init = ipv4_mib_init_net,
1649         .exit = ipv4_mib_exit_net,
1650 };
1651
1652 static int __init init_ipv4_mibs(void)
1653 {
1654         return register_pernet_subsys(&ipv4_mib_ops);
1655 }
1656
1657 static int ipv4_proc_init(void);
1658
1659 /*
1660  *      IP protocol layer initialiser
1661  */
1662
1663 static struct packet_offload ip_packet_offload __read_mostly = {
1664         .type = cpu_to_be16(ETH_P_IP),
1665         .callbacks = {
1666                 .gso_send_check = inet_gso_send_check,
1667                 .gso_segment = inet_gso_segment,
1668                 .gro_receive = inet_gro_receive,
1669                 .gro_complete = inet_gro_complete,
1670         },
1671 };
1672
1673 static int __init ipv4_offload_init(void)
1674 {
1675         /*
1676          * Add offloads
1677          */
1678         if (inet_add_offload(&udp_offload, IPPROTO_UDP) < 0)
1679                 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1680         if (inet_add_offload(&tcp_offload, IPPROTO_TCP) < 0)
1681                 pr_crit("%s: Cannot add TCP protocol offlaod\n", __func__);
1682
1683         dev_add_offload(&ip_packet_offload);
1684         return 0;
1685 }
1686
1687 fs_initcall(ipv4_offload_init);
1688
1689 static struct packet_type ip_packet_type __read_mostly = {
1690         .type = cpu_to_be16(ETH_P_IP),
1691         .func = ip_rcv,
1692 };
1693
1694 static int __init inet_init(void)
1695 {
1696         struct inet_protosw *q;
1697         struct list_head *r;
1698         int rc = -EINVAL;
1699
1700         BUILD_BUG_ON(sizeof(struct inet_skb_parm) > FIELD_SIZEOF(struct sk_buff, cb));
1701
1702         sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL);
1703         if (!sysctl_local_reserved_ports)
1704                 goto out;
1705
1706         rc = proto_register(&tcp_prot, 1);
1707         if (rc)
1708                 goto out_free_reserved_ports;
1709
1710         rc = proto_register(&udp_prot, 1);
1711         if (rc)
1712                 goto out_unregister_tcp_proto;
1713
1714         rc = proto_register(&raw_prot, 1);
1715         if (rc)
1716                 goto out_unregister_udp_proto;
1717
1718         rc = proto_register(&ping_prot, 1);
1719         if (rc)
1720                 goto out_unregister_raw_proto;
1721
1722         /*
1723          *      Tell SOCKET that we are alive...
1724          */
1725
1726         (void)sock_register(&inet_family_ops);
1727
1728 #ifdef CONFIG_SYSCTL
1729         ip_static_sysctl_init();
1730 #endif
1731
1732         tcp_prot.sysctl_mem = init_net.ipv4.sysctl_tcp_mem;
1733
1734         /*
1735          *      Add all the base protocols.
1736          */
1737
1738         if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1739                 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1740         if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1741                 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1742         if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1743                 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1744 #ifdef CONFIG_IP_MULTICAST
1745         if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1746                 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1747 #endif
1748
1749         /* Register the socket-side information for inet_create. */
1750         for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1751                 INIT_LIST_HEAD(r);
1752
1753         for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1754                 inet_register_protosw(q);
1755
1756         /*
1757          *      Set the ARP module up
1758          */
1759
1760         arp_init();
1761
1762         /*
1763          *      Set the IP module up
1764          */
1765
1766         ip_init();
1767
1768         tcp_v4_init();
1769
1770         /* Setup TCP slab cache for open requests. */
1771         tcp_init();
1772
1773         /* Setup UDP memory threshold */
1774         udp_init();
1775
1776         /* Add UDP-Lite (RFC 3828) */
1777         udplite4_register();
1778
1779         ping_init();
1780
1781         /*
1782          *      Set the ICMP layer up
1783          */
1784
1785         if (icmp_init() < 0)
1786                 panic("Failed to create the ICMP control socket.\n");
1787
1788         /*
1789          *      Initialise the multicast router
1790          */
1791 #if defined(CONFIG_IP_MROUTE)
1792         if (ip_mr_init())
1793                 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1794 #endif
1795         /*
1796          *      Initialise per-cpu ipv4 mibs
1797          */
1798
1799         if (init_ipv4_mibs())
1800                 pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1801
1802         ipv4_proc_init();
1803
1804         ipfrag_init();
1805
1806         dev_add_pack(&ip_packet_type);
1807
1808         rc = 0;
1809 out:
1810         return rc;
1811 out_unregister_raw_proto:
1812         proto_unregister(&raw_prot);
1813 out_unregister_udp_proto:
1814         proto_unregister(&udp_prot);
1815 out_unregister_tcp_proto:
1816         proto_unregister(&tcp_prot);
1817 out_free_reserved_ports:
1818         kfree(sysctl_local_reserved_ports);
1819         goto out;
1820 }
1821
1822 fs_initcall(inet_init);
1823
1824 /* ------------------------------------------------------------------------ */
1825
1826 #ifdef CONFIG_PROC_FS
1827 static int __init ipv4_proc_init(void)
1828 {
1829         int rc = 0;
1830
1831         if (raw_proc_init())
1832                 goto out_raw;
1833         if (tcp4_proc_init())
1834                 goto out_tcp;
1835         if (udp4_proc_init())
1836                 goto out_udp;
1837         if (ping_proc_init())
1838                 goto out_ping;
1839         if (ip_misc_proc_init())
1840                 goto out_misc;
1841 out:
1842         return rc;
1843 out_misc:
1844         ping_proc_exit();
1845 out_ping:
1846         udp4_proc_exit();
1847 out_udp:
1848         tcp4_proc_exit();
1849 out_tcp:
1850         raw_proc_exit();
1851 out_raw:
1852         rc = -ENOMEM;
1853         goto out;
1854 }
1855
1856 #else /* CONFIG_PROC_FS */
1857 static int __init ipv4_proc_init(void)
1858 {
1859         return 0;
1860 }
1861 #endif /* CONFIG_PROC_FS */
1862
1863 MODULE_ALIAS_NETPROTO(PF_INET);
1864