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 #include <net/secure_seq.h>
 118 #ifdef CONFIG_IP_MROUTE
 119 #include <linux/mroute.h>
 120 #endif
 121
 122 #ifdef CONFIG_ANDROID_PARANOID_NETWORK
 123 #include <linux/android_aid.h>
 124 #endif
 125
 126 /* The inetsw table contains everything that inet_create needs to
 127  * build a new socket.
 128  */
 129 static struct list_head inetsw[SOCK_MAX];
 130 static DEFINE_SPINLOCK(inetsw_lock);
 131
 132 struct ipv4_config ipv4_config;
 133 EXPORT_SYMBOL(ipv4_config);
 134
 135 /* New destruction routine */
 136
 137 void inet_sock_destruct(struct sock *sk)
 138 {
 139         struct inet_sock *inet = inet_sk(sk);
 140
 141         __skb_queue_purge(&sk->sk_receive_queue);
 142         __skb_queue_purge(&sk->sk_error_queue);
 143
 144         sk_mem_reclaim(sk);
 145
 146         if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
 147                 pr_err("Attempt to release TCP socket in state %d %p\n",
 148                        sk->sk_state, sk);
 149                 return;
 150         }
 151         if (!sock_flag(sk, SOCK_DEAD)) {
 152                 pr_err("Attempt to release alive inet socket %p\n", sk);
 153                 return;
 154         }
 155
 156         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
 157         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
 158         WARN_ON(sk->sk_wmem_queued);
 159         WARN_ON(sk->sk_forward_alloc);
 160
 161         kfree(rcu_dereference_protected(inet->inet_opt, 1));
 162         dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
 163         dst_release(sk->sk_rx_dst);
 164         sk_refcnt_debug_dec(sk);
 165 }
 166 EXPORT_SYMBOL(inet_sock_destruct);
 167
 168 /*
 169  *      The routines beyond this point handle the behaviour of an AF_INET
 170  *      socket object. Mostly it punts to the subprotocols of IP to do
 171  *      the work.
 172  */
 173
 174 /*
 175  *      Automatically bind an unbound socket.
 176  */
 177
 178 static int inet_autobind(struct sock *sk)
 179 {
 180         struct inet_sock *inet;
 181         /* We may need to bind the socket. */
 182         lock_sock(sk);
 183         inet = inet_sk(sk);
 184         if (!inet->inet_num) {
 185                 if (sk->sk_prot->get_port(sk, 0)) {
 186                         release_sock(sk);
 187                         return -EAGAIN;
 188                 }
 189                 inet->inet_sport = htons(inet->inet_num);
 190         }
 191         release_sock(sk);
 192         return 0;
 193 }
 194
 195 /*
 196  *      Move a socket into listening state.
 197  */
 198 int inet_listen(struct socket *sock, int backlog)
 199 {
 200         struct sock *sk = sock->sk;
 201         unsigned char old_state;
 202         int err;
 203
 204         lock_sock(sk);
 205
 206         err = -EINVAL;
 207         if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
 208                 goto out;
 209
 210         old_state = sk->sk_state;
 211         if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
 212                 goto out;
 213
 214         /* Really, if the socket is already in listen state
 215          * we can only allow the backlog to be adjusted.
 216          */
 217         if (old_state != TCP_LISTEN) {
 218                 /* Check special setups for testing purpose to enable TFO w/o
 219                  * requiring TCP_FASTOPEN sockopt.
 220                  * Note that only TCP sockets (SOCK_STREAM) will reach here.
 221                  * Also fastopenq may already been allocated because this
 222                  * socket was in TCP_LISTEN state previously but was
 223                  * shutdown() (rather than close()).
 224                  */
 225                 if ((sysctl_tcp_fastopen & TFO_SERVER_ENABLE) != 0 &&
 226                     inet_csk(sk)->icsk_accept_queue.fastopenq == NULL) {
 227                         if ((sysctl_tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) != 0)
 228                                 err = fastopen_init_queue(sk, backlog);
 229                         else if ((sysctl_tcp_fastopen &
 230                                   TFO_SERVER_WO_SOCKOPT2) != 0)
 231                                 err = fastopen_init_queue(sk,
 232                                     ((uint)sysctl_tcp_fastopen) >> 16);
 233                         else
 234                                 err = 0;
 235                         if (err)
 236                                 goto out;
 237                 }
 238                 err = inet_csk_listen_start(sk, backlog);
 239                 if (err)
 240                         goto out;
 241         }
 242         sk->sk_max_ack_backlog = backlog;
 243         err = 0;
 244
 245 out:
 246         release_sock(sk);
 247         return err;
 248 }
 249 EXPORT_SYMBOL(inet_listen);
 250
 251 u32 inet_ehash_secret __read_mostly;
 252 EXPORT_SYMBOL(inet_ehash_secret);
 253
 254 u32 ipv6_hash_secret __read_mostly;
 255 EXPORT_SYMBOL(ipv6_hash_secret);
 256
 257 /*
 258  * inet_ehash_secret must be set exactly once, and to a non nul value
 259  * ipv6_hash_secret must be set exactly once.
 260  */
 261 void build_ehash_secret(void)
 262 {
 263         u32 rnd;
 264
 265         do {
 266                 get_random_bytes(&rnd, sizeof(rnd));
 267         } while (rnd == 0);
 268
 269         if (cmpxchg(&inet_ehash_secret, 0, rnd) == 0) {
 270                 get_random_bytes(&ipv6_hash_secret, sizeof(ipv6_hash_secret));
 271                 net_secret_init();
 272         }
 273 }
 274 EXPORT_SYMBOL(build_ehash_secret);
 275
 276 #ifdef CONFIG_ANDROID_PARANOID_NETWORK
 277 static inline int current_has_network(void)
 278 {
 279         return (!current_euid() || in_egroup_p(AID_INET) ||
 280                 in_egroup_p(AID_NET_RAW));
 281 }
 282 static inline int current_has_cap(struct net *net, int cap)
 283 {
 284         if (cap == CAP_NET_RAW && in_egroup_p(AID_NET_RAW))
 285                 return 1;
 286         return ns_capable(net->user_ns, cap);
 287 }
 288 # else
 289 static inline int current_has_network(void)
 290 {
 291         return 1;
 292 }
 293 static inline int current_has_cap(struct net *net, int cap)
 294 {
 295         return ns_capable(net->user_ns, cap);
 296 }
 297 #endif
 298
 299 /*
 300  *      Create an inet socket.
 301  */
 302
 303 static int inet_create(struct net *net, struct socket *sock, int protocol,
 304                        int kern)
 305 {
 306         struct sock *sk;
 307         struct inet_protosw *answer;
 308         struct inet_sock *inet;
 309         struct proto *answer_prot;
 310         unsigned char answer_flags;
 311         char answer_no_check;
 312         int try_loading_module = 0;
 313         int err;
 314
 315         if (!current_has_network())
 316                 return -EACCES;
 317
 318         if (unlikely(!inet_ehash_secret))
 319                 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
 320                         build_ehash_secret();
 321
 322         sock->state = SS_UNCONNECTED;
 323
 324         /* Look for the requested type/protocol pair. */
 325 lookup_protocol:
 326         err = -ESOCKTNOSUPPORT;
 327         rcu_read_lock();
 328         list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
 329
 330                 err = 0;
 331                 /* Check the non-wild match. */
 332                 if (protocol == answer->protocol) {
 333                         if (protocol != IPPROTO_IP)
 334                                 break;
 335                 } else {
 336                         /* Check for the two wild cases. */
 337                         if (IPPROTO_IP == protocol) {
 338                                 protocol = answer->protocol;
 339                                 break;
 340                         }
 341                         if (IPPROTO_IP == answer->protocol)
 342                                 break;
 343                 }
 344                 err = -EPROTONOSUPPORT;
 345         }
 346
 347         if (unlikely(err)) {
 348                 if (try_loading_module < 2) {
 349                         rcu_read_unlock();
 350                         /*
 351                          * Be more specific, e.g. net-pf-2-proto-132-type-1
 352                          * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
 353                          */
 354                         if (++try_loading_module == 1)
 355                                 request_module("net-pf-%d-proto-%d-type-%d",
 356                                                PF_INET, protocol, sock->type);
 357                         /*
 358                          * Fall back to generic, e.g. net-pf-2-proto-132
 359                          * (net-pf-PF_INET-proto-IPPROTO_SCTP)
 360                          */
 361                         else
 362                                 request_module("net-pf-%d-proto-%d",
 363                                                PF_INET, protocol);
 364                         goto lookup_protocol;
 365                 } else
 366                         goto out_rcu_unlock;
 367         }
 368
 369         err = -EPERM;
 370         if (sock->type == SOCK_RAW && !kern &&
 371             !current_has_cap(net, CAP_NET_RAW))
 372                 goto out_rcu_unlock;
 373
 374         sock->ops = answer->ops;
 375         answer_prot = answer->prot;
 376         answer_no_check = answer->no_check;
 377         answer_flags = answer->flags;
 378         rcu_read_unlock();
 379
 380         WARN_ON(answer_prot->slab == NULL);
 381
 382         err = -ENOBUFS;
 383         sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
 384         if (sk == NULL)
 385                 goto out;
 386
 387         err = 0;
 388         sk->sk_no_check = answer_no_check;
 389         if (INET_PROTOSW_REUSE & answer_flags)
 390                 sk->sk_reuse = SK_CAN_REUSE;
 391
 392         inet = inet_sk(sk);
 393         inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
 394
 395         inet->nodefrag = 0;
 396
 397         if (SOCK_RAW == sock->type) {
 398                 inet->inet_num = protocol;
 399                 if (IPPROTO_RAW == protocol)
 400                         inet->hdrincl = 1;
 401         }
 402
 403         if (ipv4_config.no_pmtu_disc)
 404                 inet->pmtudisc = IP_PMTUDISC_DONT;
 405         else
 406                 inet->pmtudisc = IP_PMTUDISC_WANT;
 407
 408         inet->inet_id = 0;
 409
 410         sock_init_data(sock, sk);
 411
 412         sk->sk_destruct    = inet_sock_destruct;
 413         sk->sk_protocol    = protocol;
 414         sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
 415
 416         inet->uc_ttl    = -1;
 417         inet->mc_loop   = 1;
 418         inet->mc_ttl    = 1;
 419         inet->mc_all    = 1;
 420         inet->mc_index  = 0;
 421         inet->mc_list   = NULL;
 422         inet->rcv_tos   = 0;
 423
 424         sk_refcnt_debug_inc(sk);
 425
 426         if (inet->inet_num) {
 427                 /* It assumes that any protocol which allows
 428                  * the user to assign a number at socket
 429                  * creation time automatically
 430                  * shares.
 431                  */
 432                 inet->inet_sport = htons(inet->inet_num);
 433                 /* Add to protocol hash chains. */
 434                 sk->sk_prot->hash(sk);
 435         }
 436
 437         if (sk->sk_prot->init) {
 438                 err = sk->sk_prot->init(sk);
 439                 if (err)
 440                         sk_common_release(sk);
 441         }
 442 out:
 443         return err;
 444 out_rcu_unlock:
 445         rcu_read_unlock();
 446         goto out;
 447 }
 448
 449
 450 /*
 451  *      The peer socket should always be NULL (or else). When we call this
 452  *      function we are destroying the object and from then on nobody
 453  *      should refer to it.
 454  */
 455 int inet_release(struct socket *sock)
 456 {
 457         struct sock *sk = sock->sk;
 458
 459         if (sk) {
 460                 long timeout;
 461
 462                 sock_rps_reset_flow(sk);
 463
 464                 /* Applications forget to leave groups before exiting */
 465                 ip_mc_drop_socket(sk);
 466
 467                 /* If linger is set, we don't return until the close
 468                  * is complete.  Otherwise we return immediately. The
 469                  * actually closing is done the same either way.
 470                  *
 471                  * If the close is due to the process exiting, we never
 472                  * linger..
 473                  */
 474                 timeout = 0;
 475                 if (sock_flag(sk, SOCK_LINGER) &&
 476                     !(current->flags & PF_EXITING))
 477                         timeout = sk->sk_lingertime;
 478                 sock->sk = NULL;
 479                 sk->sk_prot->close(sk, timeout);
 480         }
 481         return 0;
 482 }
 483 EXPORT_SYMBOL(inet_release);
 484
 485 /* It is off by default, see below. */
 486 int sysctl_ip_nonlocal_bind __read_mostly;
 487 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
 488
 489 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 490 {
 491         struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
 492         struct sock *sk = sock->sk;
 493         struct inet_sock *inet = inet_sk(sk);
 494         struct net *net = sock_net(sk);
 495         unsigned short snum;
 496         int chk_addr_ret;
 497         int err;
 498
 499         /* If the socket has its own bind function then use it. (RAW) */
 500         if (sk->sk_prot->bind) {
 501                 err = sk->sk_prot->bind(sk, uaddr, addr_len);
 502                 goto out;
 503         }
 504         err = -EINVAL;
 505         if (addr_len < sizeof(struct sockaddr_in))
 506                 goto out;
 507
 508         if (addr->sin_family != AF_INET) {
 509                 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
 510                  * only if s_addr is INADDR_ANY.
 511                  */
 512                 err = -EAFNOSUPPORT;
 513                 if (addr->sin_family != AF_UNSPEC ||
 514                     addr->sin_addr.s_addr != htonl(INADDR_ANY))
 515                         goto out;
 516         }
 517
 518         chk_addr_ret = inet_addr_type(net, addr->sin_addr.s_addr);
 519
 520         /* Not specified by any standard per-se, however it breaks too
 521          * many applications when removed.  It is unfortunate since
 522          * allowing applications to make a non-local bind solves
 523          * several problems with systems using dynamic addressing.
 524          * (ie. your servers still start up even if your ISDN link
 525          *  is temporarily down)
 526          */
 527         err = -EADDRNOTAVAIL;
 528         if (!sysctl_ip_nonlocal_bind &&
 529             !(inet->freebind || inet->transparent) &&
 530             addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
 531             chk_addr_ret != RTN_LOCAL &&
 532             chk_addr_ret != RTN_MULTICAST &&
 533             chk_addr_ret != RTN_BROADCAST)
 534                 goto out;
 535
 536         snum = ntohs(addr->sin_port);
 537         err = -EACCES;
 538         if (snum && snum < PROT_SOCK &&
 539             !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
 540                 goto out;
 541
 542         /*      We keep a pair of addresses. rcv_saddr is the one
 543          *      used by hash lookups, and saddr is used for transmit.
 544          *
 545          *      In the BSD API these are the same except where it
 546          *      would be illegal to use them (multicast/broadcast) in
 547          *      which case the sending device address is used.
 548          */
 549         lock_sock(sk);
 550
 551         /* Check these errors (active socket, double bind). */
 552         err = -EINVAL;
 553         if (sk->sk_state != TCP_CLOSE || inet->inet_num)
 554                 goto out_release_sock;
 555
 556         inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
 557         if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
 558                 inet->inet_saddr = 0;  /* Use device */
 559
 560         /* Make sure we are allowed to bind here. */
 561         if (sk->sk_prot->get_port(sk, snum)) {
 562                 inet->inet_saddr = inet->inet_rcv_saddr = 0;
 563                 err = -EADDRINUSE;
 564                 goto out_release_sock;
 565         }
 566
 567         if (inet->inet_rcv_saddr)
 568                 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
 569         if (snum)
 570                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
 571         inet->inet_sport = htons(inet->inet_num);
 572         inet->inet_daddr = 0;
 573         inet->inet_dport = 0;
 574         sk_dst_reset(sk);
 575         err = 0;
 576 out_release_sock:
 577         release_sock(sk);
 578 out:
 579         return err;
 580 }
 581 EXPORT_SYMBOL(inet_bind);
 582
 583 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
 584                        int addr_len, int flags)
 585 {
 586         struct sock *sk = sock->sk;
 587
 588         if (addr_len < sizeof(uaddr->sa_family))
 589                 return -EINVAL;
 590         if (uaddr->sa_family == AF_UNSPEC)
 591                 return sk->sk_prot->disconnect(sk, flags);
 592
 593         if (!inet_sk(sk)->inet_num && inet_autobind(sk))
 594                 return -EAGAIN;
 595         return sk->sk_prot->connect(sk, uaddr, addr_len);
 596 }
 597 EXPORT_SYMBOL(inet_dgram_connect);
 598
 599 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
 600 {
 601         DEFINE_WAIT(wait);
 602
 603         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
 604         sk->sk_write_pending += writebias;
 605
 606         /* Basic assumption: if someone sets sk->sk_err, he _must_
 607          * change state of the socket from TCP_SYN_*.
 608          * Connect() does not allow to get error notifications
 609          * without closing the socket.
 610          */
 611         while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 612                 release_sock(sk);
 613                 timeo = schedule_timeout(timeo);
 614                 lock_sock(sk);
 615                 if (signal_pending(current) || !timeo)
 616                         break;
 617                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
 618         }
 619         finish_wait(sk_sleep(sk), &wait);
 620         sk->sk_write_pending -= writebias;
 621         return timeo;
 622 }
 623
 624 /*
 625  *      Connect to a remote host. There is regrettably still a little
 626  *      TCP 'magic' in here.
 627  */
 628 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 629                           int addr_len, int flags)
 630 {
 631         struct sock *sk = sock->sk;
 632         int err;
 633         long timeo;
 634
 635         if (addr_len < sizeof(uaddr->sa_family))
 636                 return -EINVAL;
 637
 638         if (uaddr->sa_family == AF_UNSPEC) {
 639                 err = sk->sk_prot->disconnect(sk, flags);
 640                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 641                 goto out;
 642         }
 643
 644         switch (sock->state) {
 645         default:
 646                 err = -EINVAL;
 647                 goto out;
 648         case SS_CONNECTED:
 649                 err = -EISCONN;
 650                 goto out;
 651         case SS_CONNECTING:
 652                 err = -EALREADY;
 653                 /* Fall out of switch with err, set for this state */
 654                 break;
 655         case SS_UNCONNECTED:
 656                 err = -EISCONN;
 657                 if (sk->sk_state != TCP_CLOSE)
 658                         goto out;
 659
 660                 err = sk->sk_prot->connect(sk, uaddr, addr_len);
 661                 if (err < 0)
 662                         goto out;
 663
 664                 sock->state = SS_CONNECTING;
 665
 666                 /* Just entered SS_CONNECTING state; the only
 667                  * difference is that return value in non-blocking
 668                  * case is EINPROGRESS, rather than EALREADY.
 669                  */
 670                 err = -EINPROGRESS;
 671                 break;
 672         }
 673
 674         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
 675
 676         if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 677                 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
 678                                 tcp_sk(sk)->fastopen_req &&
 679                                 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
 680
 681                 /* Error code is set above */
 682                 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
 683                         goto out;
 684
 685                 err = sock_intr_errno(timeo);
 686                 if (signal_pending(current))
 687                         goto out;
 688         }
 689
 690         /* Connection was closed by RST, timeout, ICMP error
 691          * or another process disconnected us.
 692          */
 693         if (sk->sk_state == TCP_CLOSE)
 694                 goto sock_error;
 695
 696         /* sk->sk_err may be not zero now, if RECVERR was ordered by user
 697          * and error was received after socket entered established state.
 698          * Hence, it is handled normally after connect() return successfully.
 699          */
 700
 701         sock->state = SS_CONNECTED;
 702         err = 0;
 703 out:
 704         return err;
 705
 706 sock_error:
 707         err = sock_error(sk) ? : -ECONNABORTED;
 708         sock->state = SS_UNCONNECTED;
 709         if (sk->sk_prot->disconnect(sk, flags))
 710                 sock->state = SS_DISCONNECTING;
 711         goto out;
 712 }
 713 EXPORT_SYMBOL(__inet_stream_connect);
 714
 715 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 716                         int addr_len, int flags)
 717 {
 718         int err;
 719
 720         lock_sock(sock->sk);
 721         err = __inet_stream_connect(sock, uaddr, addr_len, flags);
 722         release_sock(sock->sk);
 723         return err;
 724 }
 725 EXPORT_SYMBOL(inet_stream_connect);
 726
 727 /*
 728  *      Accept a pending connection. The TCP layer now gives BSD semantics.
 729  */
 730
 731 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
 732 {
 733         struct sock *sk1 = sock->sk;
 734         int err = -EINVAL;
 735         struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
 736
 737         if (!sk2)
 738                 goto do_err;
 739
 740         lock_sock(sk2);
 741
 742         sock_rps_record_flow(sk2);
 743         WARN_ON(!((1 << sk2->sk_state) &
 744                   (TCPF_ESTABLISHED | TCPF_SYN_RECV |
 745                   TCPF_CLOSE_WAIT | TCPF_CLOSE)));
 746
 747         sock_graft(sk2, newsock);
 748
 749         newsock->state = SS_CONNECTED;
 750         err = 0;
 751         release_sock(sk2);
 752 do_err:
 753         return err;
 754 }
 755 EXPORT_SYMBOL(inet_accept);
 756
 757
 758 /*
 759  *      This does both peername and sockname.
 760  */
 761 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
 762                         int *uaddr_len, int peer)
 763 {
 764         struct sock *sk         = sock->sk;
 765         struct inet_sock *inet  = inet_sk(sk);
 766         DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
 767
 768         sin->sin_family = AF_INET;
 769         if (peer) {
 770                 if (!inet->inet_dport ||
 771                     (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
 772                      peer == 1))
 773                         return -ENOTCONN;
 774                 sin->sin_port = inet->inet_dport;
 775                 sin->sin_addr.s_addr = inet->inet_daddr;
 776         } else {
 777                 __be32 addr = inet->inet_rcv_saddr;
 778                 if (!addr)
 779                         addr = inet->inet_saddr;
 780                 sin->sin_port = inet->inet_sport;
 781                 sin->sin_addr.s_addr = addr;
 782         }
 783         memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 784         *uaddr_len = sizeof(*sin);
 785         return 0;
 786 }
 787 EXPORT_SYMBOL(inet_getname);
 788
 789 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
 790                  size_t size)
 791 {
 792         struct sock *sk = sock->sk;
 793
 794         sock_rps_record_flow(sk);
 795
 796         /* We may need to bind the socket. */
 797         if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
 798             inet_autobind(sk))
 799                 return -EAGAIN;
 800
 801         return sk->sk_prot->sendmsg(iocb, sk, msg, size);
 802 }
 803 EXPORT_SYMBOL(inet_sendmsg);
 804
 805 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
 806                       size_t size, int flags)
 807 {
 808         struct sock *sk = sock->sk;
 809
 810         sock_rps_record_flow(sk);
 811
 812         /* We may need to bind the socket. */
 813         if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
 814             inet_autobind(sk))
 815                 return -EAGAIN;
 816
 817         if (sk->sk_prot->sendpage)
 818                 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
 819         return sock_no_sendpage(sock, page, offset, size, flags);
 820 }
 821 EXPORT_SYMBOL(inet_sendpage);
 822
 823 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
 824                  size_t size, int flags)
 825 {
 826         struct sock *sk = sock->sk;
 827         int addr_len = 0;
 828         int err;
 829
 830         sock_rps_record_flow(sk);
 831
 832         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
 833                                    flags & ~MSG_DONTWAIT, &addr_len);
 834         if (err >= 0)
 835                 msg->msg_namelen = addr_len;
 836         return err;
 837 }
 838 EXPORT_SYMBOL(inet_recvmsg);
 839
 840 int inet_shutdown(struct socket *sock, int how)
 841 {
 842         struct sock *sk = sock->sk;
 843         int err = 0;
 844
 845         /* This should really check to make sure
 846          * the socket is a TCP socket. (WHY AC...)
 847          */
 848         how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
 849                        1->2 bit 2 snds.
 850                        2->3 */
 851         if ((how & ~SHUTDOWN_MASK) || !how)     /* MAXINT->0 */
 852                 return -EINVAL;
 853
 854         lock_sock(sk);
 855         if (sock->state == SS_CONNECTING) {
 856                 if ((1 << sk->sk_state) &
 857                     (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
 858                         sock->state = SS_DISCONNECTING;
 859                 else
 860                         sock->state = SS_CONNECTED;
 861         }
 862
 863         switch (sk->sk_state) {
 864         case TCP_CLOSE:
 865                 err = -ENOTCONN;
 866                 /* Hack to wake up other listeners, who can poll for
 867                    POLLHUP, even on eg. unconnected UDP sockets -- RR */
 868         default:
 869                 sk->sk_shutdown |= how;
 870                 if (sk->sk_prot->shutdown)
 871                         sk->sk_prot->shutdown(sk, how);
 872                 break;
 873
 874         /* Remaining two branches are temporary solution for missing
 875          * close() in multithreaded environment. It is _not_ a good idea,
 876          * but we have no choice until close() is repaired at VFS level.
 877          */
 878         case TCP_LISTEN:
 879                 if (!(how & RCV_SHUTDOWN))
 880                         break;
 881                 /* Fall through */
 882         case TCP_SYN_SENT:
 883                 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
 884                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 885                 break;
 886         }
 887
 888         /* Wake up anyone sleeping in poll. */
 889         sk->sk_state_change(sk);
 890         release_sock(sk);
 891         return err;
 892 }
 893 EXPORT_SYMBOL(inet_shutdown);
 894
 895 /*
 896  *      ioctl() calls you can issue on an INET socket. Most of these are
 897  *      device configuration and stuff and very rarely used. Some ioctls
 898  *      pass on to the socket itself.
 899  *
 900  *      NOTE: I like the idea of a module for the config stuff. ie ifconfig
 901  *      loads the devconfigure module does its configuring and unloads it.
 902  *      There's a good 20K of config code hanging around the kernel.
 903  */
 904
 905 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 906 {
 907         struct sock *sk = sock->sk;
 908         int err = 0;
 909         struct net *net = sock_net(sk);
 910
 911         switch (cmd) {
 912         case SIOCGSTAMP:
 913                 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
 914                 break;
 915         case SIOCGSTAMPNS:
 916                 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
 917                 break;
 918         case SIOCADDRT:
 919         case SIOCDELRT:
 920         case SIOCRTMSG:
 921                 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
 922                 break;
 923         case SIOCDARP:
 924         case SIOCGARP:
 925         case SIOCSARP:
 926                 err = arp_ioctl(net, cmd, (void __user *)arg);
 927                 break;
 928         case SIOCGIFADDR:
 929         case SIOCSIFADDR:
 930         case SIOCGIFBRDADDR:
 931         case SIOCSIFBRDADDR:
 932         case SIOCGIFNETMASK:
 933         case SIOCSIFNETMASK:
 934         case SIOCGIFDSTADDR:
 935         case SIOCSIFDSTADDR:
 936         case SIOCSIFPFLAGS:
 937         case SIOCGIFPFLAGS:
 938         case SIOCSIFFLAGS:
 939                 err = devinet_ioctl(net, cmd, (void __user *)arg);
 940                 break;
 941         default:
 942                 if (sk->sk_prot->ioctl)
 943                         err = sk->sk_prot->ioctl(sk, cmd, arg);
 944                 else
 945                         err = -ENOIOCTLCMD;
 946                 break;
 947         }
 948         return err;
 949 }
 950 EXPORT_SYMBOL(inet_ioctl);
 951
 952 #ifdef CONFIG_COMPAT
 953 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 954 {
 955         struct sock *sk = sock->sk;
 956         int err = -ENOIOCTLCMD;
 957
 958         if (sk->sk_prot->compat_ioctl)
 959                 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
 960
 961         return err;
 962 }
 963 #endif
 964
 965 const struct proto_ops inet_stream_ops = {
 966         .family            = PF_INET,
 967         .owner             = THIS_MODULE,
 968         .release           = inet_release,
 969         .bind              = inet_bind,
 970         .connect           = inet_stream_connect,
 971         .socketpair        = sock_no_socketpair,
 972         .accept            = inet_accept,
 973         .getname           = inet_getname,
 974         .poll              = tcp_poll,
 975         .ioctl             = inet_ioctl,
 976         .listen            = inet_listen,
 977         .shutdown          = inet_shutdown,
 978         .setsockopt        = sock_common_setsockopt,
 979         .getsockopt        = sock_common_getsockopt,
 980         .sendmsg           = inet_sendmsg,
 981         .recvmsg           = inet_recvmsg,
 982         .mmap              = sock_no_mmap,
 983         .sendpage          = inet_sendpage,
 984         .splice_read       = tcp_splice_read,
 985 #ifdef CONFIG_COMPAT
 986         .compat_setsockopt = compat_sock_common_setsockopt,
 987         .compat_getsockopt = compat_sock_common_getsockopt,
 988         .compat_ioctl      = inet_compat_ioctl,
 989 #endif
 990 };
 991 EXPORT_SYMBOL(inet_stream_ops);
 992
 993 const struct proto_ops inet_dgram_ops = {
 994         .family            = PF_INET,
 995         .owner             = THIS_MODULE,
 996         .release           = inet_release,
 997         .bind              = inet_bind,
 998         .connect           = inet_dgram_connect,
 999         .socketpair        = sock_no_socketpair,
1000         .accept            = sock_no_accept,
1001         .getname           = inet_getname,
1002         .poll              = udp_poll,
1003         .ioctl             = inet_ioctl,
1004         .listen            = sock_no_listen,
1005         .shutdown          = inet_shutdown,
1006         .setsockopt        = sock_common_setsockopt,
1007         .getsockopt        = sock_common_getsockopt,
1008         .sendmsg           = inet_sendmsg,
1009         .recvmsg           = inet_recvmsg,
1010         .mmap              = sock_no_mmap,
1011         .sendpage          = inet_sendpage,
1012 #ifdef CONFIG_COMPAT
1013         .compat_setsockopt = compat_sock_common_setsockopt,
1014         .compat_getsockopt = compat_sock_common_getsockopt,
1015         .compat_ioctl      = inet_compat_ioctl,
1016 #endif
1017 };
1018 EXPORT_SYMBOL(inet_dgram_ops);
1019
1020 /*
1021  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1022  * udp_poll
1023  */
1024 static const struct proto_ops inet_sockraw_ops = {
1025         .family            = PF_INET,
1026         .owner             = THIS_MODULE,
1027         .release           = inet_release,
1028         .bind              = inet_bind,
1029         .connect           = inet_dgram_connect,
1030         .socketpair        = sock_no_socketpair,
1031         .accept            = sock_no_accept,
1032         .getname           = inet_getname,
1033         .poll              = datagram_poll,
1034         .ioctl             = inet_ioctl,
1035         .listen            = sock_no_listen,
1036         .shutdown          = inet_shutdown,
1037         .setsockopt        = sock_common_setsockopt,
1038         .getsockopt        = sock_common_getsockopt,
1039         .sendmsg           = inet_sendmsg,
1040         .recvmsg           = inet_recvmsg,
1041         .mmap              = sock_no_mmap,
1042         .sendpage          = inet_sendpage,
1043 #ifdef CONFIG_COMPAT
1044         .compat_setsockopt = compat_sock_common_setsockopt,
1045         .compat_getsockopt = compat_sock_common_getsockopt,
1046         .compat_ioctl      = inet_compat_ioctl,
1047 #endif
1048 };
1049
1050 static const struct net_proto_family inet_family_ops = {
1051         .family = PF_INET,
1052         .create = inet_create,
1053         .owner  = THIS_MODULE,
1054 };
1055
1056 /* Upon startup we insert all the elements in inetsw_array[] into
1057  * the linked list inetsw.
1058  */
1059 static struct inet_protosw inetsw_array[] =
1060 {
1061         {
1062                 .type =       SOCK_STREAM,
1063                 .protocol =   IPPROTO_TCP,
1064                 .prot =       &tcp_prot,
1065                 .ops =        &inet_stream_ops,
1066                 .no_check =   0,
1067                 .flags =      INET_PROTOSW_PERMANENT |
1068                               INET_PROTOSW_ICSK,
1069         },
1070
1071         {
1072                 .type =       SOCK_DGRAM,
1073                 .protocol =   IPPROTO_UDP,
1074                 .prot =       &udp_prot,
1075                 .ops =        &inet_dgram_ops,
1076                 .no_check =   UDP_CSUM_DEFAULT,
1077                 .flags =      INET_PROTOSW_PERMANENT,
1078        },
1079
1080        {
1081                 .type =       SOCK_DGRAM,
1082                 .protocol =   IPPROTO_ICMP,
1083                 .prot =       &ping_prot,
1084                 .ops =        &inet_dgram_ops,
1085                 .no_check =   UDP_CSUM_DEFAULT,
1086                 .flags =      INET_PROTOSW_REUSE,
1087        },
1088
1089        {
1090                .type =       SOCK_RAW,
1091                .protocol =   IPPROTO_IP,        /* wild card */
1092                .prot =       &raw_prot,
1093                .ops =        &inet_sockraw_ops,
1094                .no_check =   UDP_CSUM_DEFAULT,
1095                .flags =      INET_PROTOSW_REUSE,
1096        }
1097 };
1098
1099 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1100
1101 void inet_register_protosw(struct inet_protosw *p)
1102 {
1103         struct list_head *lh;
1104         struct inet_protosw *answer;
1105         int protocol = p->protocol;
1106         struct list_head *last_perm;
1107
1108         spin_lock_bh(&inetsw_lock);
1109
1110         if (p->type >= SOCK_MAX)
1111                 goto out_illegal;
1112
1113         /* If we are trying to override a permanent protocol, bail. */
1114         answer = NULL;
1115         last_perm = &inetsw[p->type];
1116         list_for_each(lh, &inetsw[p->type]) {
1117                 answer = list_entry(lh, struct inet_protosw, list);
1118
1119                 /* Check only the non-wild match. */
1120                 if (INET_PROTOSW_PERMANENT & answer->flags) {
1121                         if (protocol == answer->protocol)
1122                                 break;
1123                         last_perm = lh;
1124                 }
1125
1126                 answer = NULL;
1127         }
1128         if (answer)
1129                 goto out_permanent;
1130
1131         /* Add the new entry after the last permanent entry if any, so that
1132          * the new entry does not override a permanent entry when matched with
1133          * a wild-card protocol. But it is allowed to override any existing
1134          * non-permanent entry.  This means that when we remove this entry, the
1135          * system automatically returns to the old behavior.
1136          */
1137         list_add_rcu(&p->list, last_perm);
1138 out:
1139         spin_unlock_bh(&inetsw_lock);
1140
1141         return;
1142
1143 out_permanent:
1144         pr_err("Attempt to override permanent protocol %d\n", protocol);
1145         goto out;
1146
1147 out_illegal:
1148         pr_err("Ignoring attempt to register invalid socket type %d\n",
1149                p->type);
1150         goto out;
1151 }
1152 EXPORT_SYMBOL(inet_register_protosw);
1153
1154 void inet_unregister_protosw(struct inet_protosw *p)
1155 {
1156         if (INET_PROTOSW_PERMANENT & p->flags) {
1157                 pr_err("Attempt to unregister permanent protocol %d\n",
1158                        p->protocol);
1159         } else {
1160                 spin_lock_bh(&inetsw_lock);
1161                 list_del_rcu(&p->list);
1162                 spin_unlock_bh(&inetsw_lock);
1163
1164                 synchronize_net();
1165         }
1166 }
1167 EXPORT_SYMBOL(inet_unregister_protosw);
1168
1169 /*
1170  *      Shall we try to damage output packets if routing dev changes?
1171  */
1172
1173 int sysctl_ip_dynaddr __read_mostly;
1174
1175 static int inet_sk_reselect_saddr(struct sock *sk)
1176 {
1177         struct inet_sock *inet = inet_sk(sk);
1178         __be32 old_saddr = inet->inet_saddr;
1179         __be32 daddr = inet->inet_daddr;
1180         struct flowi4 *fl4;
1181         struct rtable *rt;
1182         __be32 new_saddr;
1183         struct ip_options_rcu *inet_opt;
1184
1185         inet_opt = rcu_dereference_protected(inet->inet_opt,
1186                                              sock_owned_by_user(sk));
1187         if (inet_opt && inet_opt->opt.srr)
1188                 daddr = inet_opt->opt.faddr;
1189
1190         /* Query new route. */
1191         fl4 = &inet->cork.fl.u.ip4;
1192         rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1193                               sk->sk_bound_dev_if, sk->sk_protocol,
1194                               inet->inet_sport, inet->inet_dport, sk, false);
1195         if (IS_ERR(rt))
1196                 return PTR_ERR(rt);
1197
1198         sk_setup_caps(sk, &rt->dst);
1199
1200         new_saddr = fl4->saddr;
1201
1202         if (new_saddr == old_saddr)
1203                 return 0;
1204
1205         if (sysctl_ip_dynaddr > 1) {
1206                 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1207                         __func__, &old_saddr, &new_saddr);
1208         }
1209
1210         inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1211
1212         /*
1213          * XXX The only one ugly spot where we need to
1214          * XXX really change the sockets identity after
1215          * XXX it has entered the hashes. -DaveM
1216          *
1217          * Besides that, it does not check for connection
1218          * uniqueness. Wait for troubles.
1219          */
1220         __sk_prot_rehash(sk);
1221         return 0;
1222 }
1223
1224 int inet_sk_rebuild_header(struct sock *sk)
1225 {
1226         struct inet_sock *inet = inet_sk(sk);
1227         struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1228         __be32 daddr;
1229         struct ip_options_rcu *inet_opt;
1230         struct flowi4 *fl4;
1231         int err;
1232
1233         /* Route is OK, nothing to do. */
1234         if (rt)
1235                 return 0;
1236
1237         /* Reroute. */
1238         rcu_read_lock();
1239         inet_opt = rcu_dereference(inet->inet_opt);
1240         daddr = inet->inet_daddr;
1241         if (inet_opt && inet_opt->opt.srr)
1242                 daddr = inet_opt->opt.faddr;
1243         rcu_read_unlock();
1244         fl4 = &inet->cork.fl.u.ip4;
1245         rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1246                                    inet->inet_dport, inet->inet_sport,
1247                                    sk->sk_protocol, RT_CONN_FLAGS(sk),
1248                                    sk->sk_bound_dev_if);
1249         if (!IS_ERR(rt)) {
1250                 err = 0;
1251                 sk_setup_caps(sk, &rt->dst);
1252         } else {
1253                 err = PTR_ERR(rt);
1254
1255                 /* Routing failed... */
1256                 sk->sk_route_caps = 0;
1257                 /*
1258                  * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1259                  * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1260                  */
1261                 if (!sysctl_ip_dynaddr ||
1262                     sk->sk_state != TCP_SYN_SENT ||
1263                     (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1264                     (err = inet_sk_reselect_saddr(sk)) != 0)
1265                         sk->sk_err_soft = -err;
1266         }
1267
1268         return err;
1269 }
1270 EXPORT_SYMBOL(inet_sk_rebuild_header);
1271
1272 static int inet_gso_send_check(struct sk_buff *skb)
1273 {
1274         const struct net_offload *ops;
1275         const struct iphdr *iph;
1276         int proto;
1277         int ihl;
1278         int err = -EINVAL;
1279
1280         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1281                 goto out;
1282
1283         iph = ip_hdr(skb);
1284         ihl = iph->ihl * 4;
1285         if (ihl < sizeof(*iph))
1286                 goto out;
1287
1288         if (unlikely(!pskb_may_pull(skb, ihl)))
1289                 goto out;
1290
1291         __skb_pull(skb, ihl);
1292         skb_reset_transport_header(skb);
1293         iph = ip_hdr(skb);
1294         proto = iph->protocol;
1295         err = -EPROTONOSUPPORT;
1296
1297         rcu_read_lock();
1298         ops = rcu_dereference(inet_offloads[proto]);
1299         if (likely(ops && ops->callbacks.gso_send_check))
1300                 err = ops->callbacks.gso_send_check(skb);
1301         rcu_read_unlock();
1302
1303 out:
1304         return err;
1305 }
1306
1307 static struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1308         netdev_features_t features)
1309 {
1310         struct sk_buff *segs = ERR_PTR(-EINVAL);
1311         const struct net_offload *ops;
1312         struct iphdr *iph;
1313         int proto;
1314         int ihl;
1315         int id;
1316         unsigned int offset = 0;
1317         bool tunnel;
1318
1319         if (unlikely(skb_shinfo(skb)->gso_type &
1320                      ~(SKB_GSO_TCPV4 |
1321                        SKB_GSO_UDP |
1322                        SKB_GSO_DODGY |
1323                        SKB_GSO_TCP_ECN |
1324                        SKB_GSO_GRE |
1325                        SKB_GSO_TCPV6 |
1326                        SKB_GSO_UDP_TUNNEL |
1327                        0)))
1328                 goto out;
1329
1330         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1331                 goto out;
1332
1333         iph = ip_hdr(skb);
1334         ihl = iph->ihl * 4;
1335         if (ihl < sizeof(*iph))
1336                 goto out;
1337
1338         if (unlikely(!pskb_may_pull(skb, ihl)))
1339                 goto out;
1340
1341         tunnel = !!skb->encapsulation;
1342
1343         __skb_pull(skb, ihl);
1344         skb_reset_transport_header(skb);
1345         iph = ip_hdr(skb);
1346         id = ntohs(iph->id);
1347         proto = iph->protocol;
1348         segs = ERR_PTR(-EPROTONOSUPPORT);
1349
1350         rcu_read_lock();
1351         ops = rcu_dereference(inet_offloads[proto]);
1352         if (likely(ops && ops->callbacks.gso_segment))
1353                 segs = ops->callbacks.gso_segment(skb, features);
1354         rcu_read_unlock();
1355
1356         if (IS_ERR_OR_NULL(segs))
1357                 goto out;
1358
1359         skb = segs;
1360         do {
1361                 iph = ip_hdr(skb);
1362                 if (!tunnel && proto == IPPROTO_UDP) {
1363                         iph->id = htons(id);
1364                         iph->frag_off = htons(offset >> 3);
1365                         if (skb->next != NULL)
1366                                 iph->frag_off |= htons(IP_MF);
1367                         offset += (skb->len - skb->mac_len - iph->ihl * 4);
1368                 } else  {
1369                         iph->id = htons(id++);
1370                 }
1371                 iph->tot_len = htons(skb->len - skb->mac_len);
1372                 iph->check = 0;
1373                 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1374         } while ((skb = skb->next));
1375
1376 out:
1377         return segs;
1378 }
1379
1380 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1381                                          struct sk_buff *skb)
1382 {
1383         const struct net_offload *ops;
1384         struct sk_buff **pp = NULL;
1385         struct sk_buff *p;
1386         const struct iphdr *iph;
1387         unsigned int hlen;
1388         unsigned int off;
1389         unsigned int id;
1390         int flush = 1;
1391         int proto;
1392
1393         off = skb_gro_offset(skb);
1394         hlen = off + sizeof(*iph);
1395         iph = skb_gro_header_fast(skb, off);
1396         if (skb_gro_header_hard(skb, hlen)) {
1397                 iph = skb_gro_header_slow(skb, hlen, off);
1398                 if (unlikely(!iph))
1399                         goto out;
1400         }
1401
1402         proto = iph->protocol;
1403
1404         rcu_read_lock();
1405         ops = rcu_dereference(inet_offloads[proto]);
1406         if (!ops || !ops->callbacks.gro_receive)
1407                 goto out_unlock;
1408
1409         if (*(u8 *)iph != 0x45)
1410                 goto out_unlock;
1411
1412         if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1413                 goto out_unlock;
1414
1415         id = ntohl(*(__be32 *)&iph->id);
1416         flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF));
1417         id >>= 16;
1418
1419         for (p = *head; p; p = p->next) {
1420                 struct iphdr *iph2;
1421
1422                 if (!NAPI_GRO_CB(p)->same_flow)
1423                         continue;
1424
1425                 iph2 = ip_hdr(p);
1426
1427                 if ((iph->protocol ^ iph2->protocol) |
1428                     ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1429                     ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1430                         NAPI_GRO_CB(p)->same_flow = 0;
1431                         continue;
1432                 }
1433
1434                 /* All fields must match except length and checksum. */
1435                 NAPI_GRO_CB(p)->flush |=
1436                         (iph->ttl ^ iph2->ttl) |
1437                         (iph->tos ^ iph2->tos) |
1438                         ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1439
1440                 NAPI_GRO_CB(p)->flush |= flush;
1441         }
1442
1443         NAPI_GRO_CB(skb)->flush |= flush;
1444         skb_gro_pull(skb, sizeof(*iph));
1445         skb_set_transport_header(skb, skb_gro_offset(skb));
1446
1447         pp = ops->callbacks.gro_receive(head, skb);
1448
1449 out_unlock:
1450         rcu_read_unlock();
1451
1452 out:
1453         NAPI_GRO_CB(skb)->flush |= flush;
1454
1455         return pp;
1456 }
1457
1458 static int inet_gro_complete(struct sk_buff *skb)
1459 {
1460         __be16 newlen = htons(skb->len - skb_network_offset(skb));
1461         struct iphdr *iph = ip_hdr(skb);
1462         const struct net_offload *ops;
1463         int proto = iph->protocol;
1464         int err = -ENOSYS;
1465
1466         csum_replace2(&iph->check, iph->tot_len, newlen);
1467         iph->tot_len = newlen;
1468
1469         rcu_read_lock();
1470         ops = rcu_dereference(inet_offloads[proto]);
1471         if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1472                 goto out_unlock;
1473
1474         err = ops->callbacks.gro_complete(skb);
1475
1476 out_unlock:
1477         rcu_read_unlock();
1478
1479         return err;
1480 }
1481
1482 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1483                          unsigned short type, unsigned char protocol,
1484                          struct net *net)
1485 {
1486         struct socket *sock;
1487         int rc = sock_create_kern(family, type, protocol, &sock);
1488
1489         if (rc == 0) {
1490                 *sk = sock->sk;
1491                 (*sk)->sk_allocation = GFP_ATOMIC;
1492                 /*
1493                  * Unhash it so that IP input processing does not even see it,
1494                  * we do not wish this socket to see incoming packets.
1495                  */
1496                 (*sk)->sk_prot->unhash(*sk);
1497
1498                 sk_change_net(*sk, net);
1499         }
1500         return rc;
1501 }
1502 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1503
1504 unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1505 {
1506         unsigned long res = 0;
1507         int i, j;
1508
1509         for_each_possible_cpu(i) {
1510                 for (j = 0; j < SNMP_ARRAY_SZ; j++)
1511                         res += *(((unsigned long *) per_cpu_ptr(mib[j], i)) + offt);
1512         }
1513         return res;
1514 }
1515 EXPORT_SYMBOL_GPL(snmp_fold_field);
1516
1517 #if BITS_PER_LONG==32
1518
1519 u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset)
1520 {
1521         u64 res = 0;
1522         int cpu;
1523
1524         for_each_possible_cpu(cpu) {
1525                 void *bhptr;
1526                 struct u64_stats_sync *syncp;
1527                 u64 v;
1528                 unsigned int start;
1529
1530                 bhptr = per_cpu_ptr(mib[0], cpu);
1531                 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1532                 do {
1533                         start = u64_stats_fetch_begin_bh(syncp);
1534                         v = *(((u64 *) bhptr) + offt);
1535                 } while (u64_stats_fetch_retry_bh(syncp, start));
1536
1537                 res += v;
1538         }
1539         return res;
1540 }
1541 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1542 #endif
1543
1544 int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align)
1545 {
1546         BUG_ON(ptr == NULL);
1547         ptr[0] = __alloc_percpu(mibsize, align);
1548         if (!ptr[0])
1549                 return -ENOMEM;
1550 #if SNMP_ARRAY_SZ == 2
1551         ptr[1] = __alloc_percpu(mibsize, align);
1552         if (!ptr[1]) {
1553                 free_percpu(ptr[0]);
1554                 ptr[0] = NULL;
1555                 return -ENOMEM;
1556         }
1557 #endif
1558         return 0;
1559 }
1560 EXPORT_SYMBOL_GPL(snmp_mib_init);
1561
1562 void snmp_mib_free(void __percpu *ptr[SNMP_ARRAY_SZ])
1563 {
1564         int i;
1565
1566         BUG_ON(ptr == NULL);
1567         for (i = 0; i < SNMP_ARRAY_SZ; i++) {
1568                 free_percpu(ptr[i]);
1569                 ptr[i] = NULL;
1570         }
1571 }
1572 EXPORT_SYMBOL_GPL(snmp_mib_free);
1573
1574 #ifdef CONFIG_IP_MULTICAST
1575 static const struct net_protocol igmp_protocol = {
1576         .handler =      igmp_rcv,
1577         .netns_ok =     1,
1578 };
1579 #endif
1580
1581 static const struct net_protocol tcp_protocol = {
1582         .early_demux    =       tcp_v4_early_demux,
1583         .handler        =       tcp_v4_rcv,
1584         .err_handler    =       tcp_v4_err,
1585         .no_policy      =       1,
1586         .netns_ok       =       1,
1587 };
1588
1589 static const struct net_offload tcp_offload = {
1590         .callbacks = {
1591                 .gso_send_check =       tcp_v4_gso_send_check,
1592                 .gso_segment    =       tcp_tso_segment,
1593                 .gro_receive    =       tcp4_gro_receive,
1594                 .gro_complete   =       tcp4_gro_complete,
1595         },
1596 };
1597
1598 static const struct net_protocol udp_protocol = {
1599         .handler =      udp_rcv,
1600         .err_handler =  udp_err,
1601         .no_policy =    1,
1602         .netns_ok =     1,
1603 };
1604
1605 static const struct net_offload udp_offload = {
1606         .callbacks = {
1607                 .gso_send_check = udp4_ufo_send_check,
1608                 .gso_segment = udp4_ufo_fragment,
1609         },
1610 };
1611
1612 static const struct net_protocol icmp_protocol = {
1613         .handler =      icmp_rcv,
1614         .err_handler =  icmp_err,
1615         .no_policy =    1,
1616         .netns_ok =     1,
1617 };
1618
1619 static __net_init int ipv4_mib_init_net(struct net *net)
1620 {
1621         if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1622                           sizeof(struct tcp_mib),
1623                           __alignof__(struct tcp_mib)) < 0)
1624                 goto err_tcp_mib;
1625         if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1626                           sizeof(struct ipstats_mib),
1627                           __alignof__(struct ipstats_mib)) < 0)
1628                 goto err_ip_mib;
1629         if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1630                           sizeof(struct linux_mib),
1631                           __alignof__(struct linux_mib)) < 0)
1632                 goto err_net_mib;
1633         if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1634                           sizeof(struct udp_mib),
1635                           __alignof__(struct udp_mib)) < 0)
1636                 goto err_udp_mib;
1637         if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1638                           sizeof(struct udp_mib),
1639                           __alignof__(struct udp_mib)) < 0)
1640                 goto err_udplite_mib;
1641         if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1642                           sizeof(struct icmp_mib),
1643                           __alignof__(struct icmp_mib)) < 0)
1644                 goto err_icmp_mib;
1645         net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1646                                               GFP_KERNEL);
1647         if (!net->mib.icmpmsg_statistics)
1648                 goto err_icmpmsg_mib;
1649
1650         tcp_mib_init(net);
1651         return 0;
1652
1653 err_icmpmsg_mib:
1654         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1655 err_icmp_mib:
1656         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1657 err_udplite_mib:
1658         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1659 err_udp_mib:
1660         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1661 err_net_mib:
1662         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1663 err_ip_mib:
1664         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1665 err_tcp_mib:
1666         return -ENOMEM;
1667 }
1668
1669 static __net_exit void ipv4_mib_exit_net(struct net *net)
1670 {
1671         kfree(net->mib.icmpmsg_statistics);
1672         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1673         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1674         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1675         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1676         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1677         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1678 }
1679
1680 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1681         .init = ipv4_mib_init_net,
1682         .exit = ipv4_mib_exit_net,
1683 };
1684
1685 static int __init init_ipv4_mibs(void)
1686 {
1687         return register_pernet_subsys(&ipv4_mib_ops);
1688 }
1689
1690 static int ipv4_proc_init(void);
1691
1692 /*
1693  *      IP protocol layer initialiser
1694  */
1695
1696 static struct packet_offload ip_packet_offload __read_mostly = {
1697         .type = cpu_to_be16(ETH_P_IP),
1698         .callbacks = {
1699                 .gso_send_check = inet_gso_send_check,
1700                 .gso_segment = inet_gso_segment,
1701                 .gro_receive = inet_gro_receive,
1702                 .gro_complete = inet_gro_complete,
1703         },
1704 };
1705
1706 static int __init ipv4_offload_init(void)
1707 {
1708         /*
1709          * Add offloads
1710          */
1711         if (inet_add_offload(&udp_offload, IPPROTO_UDP) < 0)
1712                 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1713         if (inet_add_offload(&tcp_offload, IPPROTO_TCP) < 0)
1714                 pr_crit("%s: Cannot add TCP protocol offlaod\n", __func__);
1715
1716         dev_add_offload(&ip_packet_offload);
1717         return 0;
1718 }
1719
1720 fs_initcall(ipv4_offload_init);
1721
1722 static struct packet_type ip_packet_type __read_mostly = {
1723         .type = cpu_to_be16(ETH_P_IP),
1724         .func = ip_rcv,
1725 };
1726
1727 static int __init inet_init(void)
1728 {
1729         struct inet_protosw *q;
1730         struct list_head *r;
1731         int rc = -EINVAL;
1732
1733         BUILD_BUG_ON(sizeof(struct inet_skb_parm) > FIELD_SIZEOF(struct sk_buff, cb));
1734
1735         sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL);
1736         if (!sysctl_local_reserved_ports)
1737                 goto out;
1738
1739         rc = proto_register(&tcp_prot, 1);
1740         if (rc)
1741                 goto out_free_reserved_ports;
1742
1743         rc = proto_register(&udp_prot, 1);
1744         if (rc)
1745                 goto out_unregister_tcp_proto;
1746
1747         rc = proto_register(&raw_prot, 1);
1748         if (rc)
1749                 goto out_unregister_udp_proto;
1750
1751         rc = proto_register(&ping_prot, 1);
1752         if (rc)
1753                 goto out_unregister_raw_proto;
1754
1755         /*
1756          *      Tell SOCKET that we are alive...
1757          */
1758
1759         (void)sock_register(&inet_family_ops);
1760
1761 #ifdef CONFIG_SYSCTL
1762         ip_static_sysctl_init();
1763 #endif
1764
1765         tcp_prot.sysctl_mem = init_net.ipv4.sysctl_tcp_mem;
1766
1767         /*
1768          *      Add all the base protocols.
1769          */
1770
1771         if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1772                 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1773         if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1774                 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1775         if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1776                 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1777 #ifdef CONFIG_IP_MULTICAST
1778         if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1779                 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1780 #endif
1781
1782         /* Register the socket-side information for inet_create. */
1783         for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1784                 INIT_LIST_HEAD(r);
1785
1786         for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1787                 inet_register_protosw(q);
1788
1789         /*
1790          *      Set the ARP module up
1791          */
1792
1793         arp_init();
1794
1795         /*
1796          *      Set the IP module up
1797          */
1798
1799         ip_init();
1800
1801         tcp_v4_init();
1802
1803         /* Setup TCP slab cache for open requests. */
1804         tcp_init();
1805
1806         /* Setup UDP memory threshold */
1807         udp_init();
1808
1809         /* Add UDP-Lite (RFC 3828) */
1810         udplite4_register();
1811
1812         ping_init();
1813
1814         /*
1815          *      Set the ICMP layer up
1816          */
1817
1818         if (icmp_init() < 0)
1819                 panic("Failed to create the ICMP control socket.\n");
1820
1821         /*
1822          *      Initialise the multicast router
1823          */
1824 #if defined(CONFIG_IP_MROUTE)
1825         if (ip_mr_init())
1826                 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1827 #endif
1828         /*
1829          *      Initialise per-cpu ipv4 mibs
1830          */
1831
1832         if (init_ipv4_mibs())
1833                 pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1834
1835         ipv4_proc_init();
1836
1837         ipfrag_init();
1838
1839         dev_add_pack(&ip_packet_type);
1840
1841         rc = 0;
1842 out:
1843         return rc;
1844 out_unregister_raw_proto:
1845         proto_unregister(&raw_prot);
1846 out_unregister_udp_proto:
1847         proto_unregister(&udp_prot);
1848 out_unregister_tcp_proto:
1849         proto_unregister(&tcp_prot);
1850 out_free_reserved_ports:
1851         kfree(sysctl_local_reserved_ports);
1852         goto out;
1853 }
1854
1855 fs_initcall(inet_init);
1856
1857 /* ------------------------------------------------------------------------ */
1858
1859 #ifdef CONFIG_PROC_FS
1860 static int __init ipv4_proc_init(void)
1861 {
1862         int rc = 0;
1863
1864         if (raw_proc_init())
1865                 goto out_raw;
1866         if (tcp4_proc_init())
1867                 goto out_tcp;
1868         if (udp4_proc_init())
1869                 goto out_udp;
1870         if (ping_proc_init())
1871                 goto out_ping;
1872         if (ip_misc_proc_init())
1873                 goto out_misc;
1874 out:
1875         return rc;
1876 out_misc:
1877         ping_proc_exit();
1878 out_ping:
1879         udp4_proc_exit();
1880 out_udp:
1881         tcp4_proc_exit();
1882 out_tcp:
1883         raw_proc_exit();
1884 out_raw:
1885         rc = -ENOMEM;
1886         goto out;
1887 }
1888
1889 #else /* CONFIG_PROC_FS */
1890 static int __init ipv4_proc_init(void)
1891 {
1892         return 0;
1893 }
1894 #endif /* CONFIG_PROC_FS */
1895
1896 MODULE_ALIAS_NETPROTO(PF_INET);
1897