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/kmod.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 #include <linux/netfilter/xt_qtaguid.h>
  93
  94 #include <linux/uaccess.h>
  95
  96 #include <linux/inet.h>
  97 #include <linux/igmp.h>
  98 #include <linux/inetdevice.h>
  99 #include <linux/netdevice.h>
 100 #include <net/checksum.h>
 101 #include <net/ip.h>
 102 #include <net/protocol.h>
 103 #include <net/arp.h>
 104 #include <net/route.h>
 105 #include <net/ip_fib.h>
 106 #include <net/inet_connection_sock.h>
 107 #include <net/tcp.h>
 108 #include <net/udp.h>
 109 #include <net/udplite.h>
 110 #include <net/ping.h>
 111 #include <linux/skbuff.h>
 112 #include <net/sock.h>
 113 #include <net/raw.h>
 114 #include <net/icmp.h>
 115 #include <net/inet_common.h>
 116 #include <net/ip_tunnels.h>
 117 #include <net/xfrm.h>
 118 #include <net/net_namespace.h>
 119 #include <net/secure_seq.h>
 120 #ifdef CONFIG_IP_MROUTE
 121 #include <linux/mroute.h>
 122 #endif
 123 #include <net/l3mdev.h>
 124
 125 #include <trace/events/sock.h>
 126
 127 #ifdef CONFIG_ANDROID_PARANOID_NETWORK
 128 #include <linux/android_aid.h>
 129
 130 static inline int current_has_network(void)
 131 {
 132         return in_egroup_p(AID_INET) || capable(CAP_NET_RAW);
 133 }
 134 #else
 135 static inline int current_has_network(void)
 136 {
 137         return 1;
 138 }
 139 #endif
 140
 141 /* The inetsw table contains everything that inet_create needs to
 142  * build a new socket.
 143  */
 144 static struct list_head inetsw[SOCK_MAX];
 145 static DEFINE_SPINLOCK(inetsw_lock);
 146
 147 /* New destruction routine */
 148
 149 void inet_sock_destruct(struct sock *sk)
 150 {
 151         struct inet_sock *inet = inet_sk(sk);
 152
 153         __skb_queue_purge(&sk->sk_receive_queue);
 154         __skb_queue_purge(&sk->sk_error_queue);
 155
 156         sk_mem_reclaim(sk);
 157
 158         if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
 159                 pr_err("Attempt to release TCP socket in state %d %p\n",
 160                        sk->sk_state, sk);
 161                 return;
 162         }
 163         if (!sock_flag(sk, SOCK_DEAD)) {
 164                 pr_err("Attempt to release alive inet socket %p\n", sk);
 165                 return;
 166         }
 167
 168         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
 169         WARN_ON(refcount_read(&sk->sk_wmem_alloc));
 170         WARN_ON(sk->sk_wmem_queued);
 171         WARN_ON(sk->sk_forward_alloc);
 172
 173         kfree(rcu_dereference_protected(inet->inet_opt, 1));
 174         dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
 175         dst_release(sk->sk_rx_dst);
 176         sk_refcnt_debug_dec(sk);
 177 }
 178 EXPORT_SYMBOL(inet_sock_destruct);
 179
 180 /*
 181  *      The routines beyond this point handle the behaviour of an AF_INET
 182  *      socket object. Mostly it punts to the subprotocols of IP to do
 183  *      the work.
 184  */
 185
 186 /*
 187  *      Automatically bind an unbound socket.
 188  */
 189
 190 static int inet_autobind(struct sock *sk)
 191 {
 192         struct inet_sock *inet;
 193         /* We may need to bind the socket. */
 194         lock_sock(sk);
 195         inet = inet_sk(sk);
 196         if (!inet->inet_num) {
 197                 if (sk->sk_prot->get_port(sk, 0)) {
 198                         release_sock(sk);
 199                         return -EAGAIN;
 200                 }
 201                 inet->inet_sport = htons(inet->inet_num);
 202         }
 203         release_sock(sk);
 204         return 0;
 205 }
 206
 207 /*
 208  *      Move a socket into listening state.
 209  */
 210 int inet_listen(struct socket *sock, int backlog)
 211 {
 212         struct sock *sk = sock->sk;
 213         unsigned char old_state;
 214         int err, tcp_fastopen;
 215
 216         lock_sock(sk);
 217
 218         err = -EINVAL;
 219         if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
 220                 goto out;
 221
 222         old_state = sk->sk_state;
 223         if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
 224                 goto out;
 225
 226         /* Really, if the socket is already in listen state
 227          * we can only allow the backlog to be adjusted.
 228          */
 229         if (old_state != TCP_LISTEN) {
 230                 /* Enable TFO w/o requiring TCP_FASTOPEN socket option.
 231                  * Note that only TCP sockets (SOCK_STREAM) will reach here.
 232                  * Also fastopen backlog may already been set via the option
 233                  * because the socket was in TCP_LISTEN state previously but
 234                  * was shutdown() rather than close().
 235                  */
 236                 tcp_fastopen = sock_net(sk)->ipv4.sysctl_tcp_fastopen;
 237                 if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
 238                     (tcp_fastopen & TFO_SERVER_ENABLE) &&
 239                     !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
 240                         fastopen_queue_tune(sk, backlog);
 241                         tcp_fastopen_init_key_once(sock_net(sk));
 242                 }
 243
 244                 err = inet_csk_listen_start(sk, backlog);
 245                 if (err)
 246                         goto out;
 247                 tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL);
 248         }
 249         sk->sk_max_ack_backlog = backlog;
 250         err = 0;
 251
 252 out:
 253         release_sock(sk);
 254         return err;
 255 }
 256 EXPORT_SYMBOL(inet_listen);
 257
 258 /*
 259  *      Create an inet socket.
 260  */
 261
 262 static int inet_create(struct net *net, struct socket *sock, int protocol,
 263                        int kern)
 264 {
 265         struct sock *sk;
 266         struct inet_protosw *answer;
 267         struct inet_sock *inet;
 268         struct proto *answer_prot;
 269         unsigned char answer_flags;
 270         int try_loading_module = 0;
 271         int err;
 272
 273         if (protocol < 0 || protocol >= IPPROTO_MAX)
 274                 return -EINVAL;
 275
 276         if (!current_has_network())
 277                 return -EACCES;
 278
 279         sock->state = SS_UNCONNECTED;
 280
 281         /* Look for the requested type/protocol pair. */
 282 lookup_protocol:
 283         err = -ESOCKTNOSUPPORT;
 284         rcu_read_lock();
 285         list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
 286
 287                 err = 0;
 288                 /* Check the non-wild match. */
 289                 if (protocol == answer->protocol) {
 290                         if (protocol != IPPROTO_IP)
 291                                 break;
 292                 } else {
 293                         /* Check for the two wild cases. */
 294                         if (IPPROTO_IP == protocol) {
 295                                 protocol = answer->protocol;
 296                                 break;
 297                         }
 298                         if (IPPROTO_IP == answer->protocol)
 299                                 break;
 300                 }
 301                 err = -EPROTONOSUPPORT;
 302         }
 303
 304         if (unlikely(err)) {
 305                 if (try_loading_module < 2) {
 306                         rcu_read_unlock();
 307                         /*
 308                          * Be more specific, e.g. net-pf-2-proto-132-type-1
 309                          * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
 310                          */
 311                         if (++try_loading_module == 1)
 312                                 request_module("net-pf-%d-proto-%d-type-%d",
 313                                                PF_INET, protocol, sock->type);
 314                         /*
 315                          * Fall back to generic, e.g. net-pf-2-proto-132
 316                          * (net-pf-PF_INET-proto-IPPROTO_SCTP)
 317                          */
 318                         else
 319                                 request_module("net-pf-%d-proto-%d",
 320                                                PF_INET, protocol);
 321                         goto lookup_protocol;
 322                 } else
 323                         goto out_rcu_unlock;
 324         }
 325
 326         err = -EPERM;
 327         if (sock->type == SOCK_RAW && !kern && !capable(CAP_NET_RAW))
 328                 goto out_rcu_unlock;
 329
 330         sock->ops = answer->ops;
 331         answer_prot = answer->prot;
 332         answer_flags = answer->flags;
 333         rcu_read_unlock();
 334
 335         WARN_ON(!answer_prot->slab);
 336
 337         err = -ENOBUFS;
 338         sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
 339         if (!sk)
 340                 goto out;
 341
 342         err = 0;
 343         if (INET_PROTOSW_REUSE & answer_flags)
 344                 sk->sk_reuse = SK_CAN_REUSE;
 345
 346         inet = inet_sk(sk);
 347         inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
 348
 349         inet->nodefrag = 0;
 350
 351         if (SOCK_RAW == sock->type) {
 352                 inet->inet_num = protocol;
 353                 if (IPPROTO_RAW == protocol)
 354                         inet->hdrincl = 1;
 355         }
 356
 357         if (net->ipv4.sysctl_ip_no_pmtu_disc)
 358                 inet->pmtudisc = IP_PMTUDISC_DONT;
 359         else
 360                 inet->pmtudisc = IP_PMTUDISC_WANT;
 361
 362         inet->inet_id = 0;
 363
 364         sock_init_data(sock, sk);
 365
 366         sk->sk_destruct    = inet_sock_destruct;
 367         sk->sk_protocol    = protocol;
 368         sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
 369
 370         inet->uc_ttl    = -1;
 371         inet->mc_loop   = 1;
 372         inet->mc_ttl    = 1;
 373         inet->mc_all    = 1;
 374         inet->mc_index  = 0;
 375         inet->mc_list   = NULL;
 376         inet->rcv_tos   = 0;
 377
 378         sk_refcnt_debug_inc(sk);
 379
 380         if (inet->inet_num) {
 381                 /* It assumes that any protocol which allows
 382                  * the user to assign a number at socket
 383                  * creation time automatically
 384                  * shares.
 385                  */
 386                 inet->inet_sport = htons(inet->inet_num);
 387                 /* Add to protocol hash chains. */
 388                 err = sk->sk_prot->hash(sk);
 389                 if (err) {
 390                         sk_common_release(sk);
 391                         goto out;
 392                 }
 393         }
 394
 395         if (sk->sk_prot->init) {
 396                 err = sk->sk_prot->init(sk);
 397                 if (err) {
 398                         sk_common_release(sk);
 399                         goto out;
 400                 }
 401         }
 402
 403         if (!kern) {
 404                 err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
 405                 if (err) {
 406                         sk_common_release(sk);
 407                         goto out;
 408                 }
 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 #ifdef CONFIG_NETFILTER_XT_MATCH_QTAGUID
 431                 qtaguid_untag(sock, true);
 432 #endif
 433                 /* Applications forget to leave groups before exiting */
 434                 ip_mc_drop_socket(sk);
 435
 436                 /* If linger is set, we don't return until the close
 437                  * is complete.  Otherwise we return immediately. The
 438                  * actually closing is done the same either way.
 439                  *
 440                  * If the close is due to the process exiting, we never
 441                  * linger..
 442                  */
 443                 timeout = 0;
 444                 if (sock_flag(sk, SOCK_LINGER) &&
 445                     !(current->flags & PF_EXITING))
 446                         timeout = sk->sk_lingertime;
 447                 sock->sk = NULL;
 448                 sk->sk_prot->close(sk, timeout);
 449         }
 450         return 0;
 451 }
 452 EXPORT_SYMBOL(inet_release);
 453
 454 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 455 {
 456         struct sock *sk = sock->sk;
 457         int err;
 458
 459         /* If the socket has its own bind function then use it. (RAW) */
 460         if (sk->sk_prot->bind) {
 461                 return sk->sk_prot->bind(sk, uaddr, addr_len);
 462         }
 463         if (addr_len < sizeof(struct sockaddr_in))
 464                 return -EINVAL;
 465
 466         /* BPF prog is run before any checks are done so that if the prog
 467          * changes context in a wrong way it will be caught.
 468          */
 469         err = BPF_CGROUP_RUN_PROG_INET4_BIND(sk, uaddr);
 470         if (err)
 471                 return err;
 472
 473         return __inet_bind(sk, uaddr, addr_len, false, true);
 474 }
 475 EXPORT_SYMBOL(inet_bind);
 476
 477 int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
 478                 bool force_bind_address_no_port, bool with_lock)
 479 {
 480         struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
 481         struct inet_sock *inet = inet_sk(sk);
 482         struct net *net = sock_net(sk);
 483         unsigned short snum;
 484         int chk_addr_ret;
 485         u32 tb_id = RT_TABLE_LOCAL;
 486         int err;
 487
 488         if (addr->sin_family != AF_INET) {
 489                 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
 490                  * only if s_addr is INADDR_ANY.
 491                  */
 492                 err = -EAFNOSUPPORT;
 493                 if (addr->sin_family != AF_UNSPEC ||
 494                     addr->sin_addr.s_addr != htonl(INADDR_ANY))
 495                         goto out;
 496         }
 497
 498         tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
 499         chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
 500
 501         /* Not specified by any standard per-se, however it breaks too
 502          * many applications when removed.  It is unfortunate since
 503          * allowing applications to make a non-local bind solves
 504          * several problems with systems using dynamic addressing.
 505          * (ie. your servers still start up even if your ISDN link
 506          *  is temporarily down)
 507          */
 508         err = -EADDRNOTAVAIL;
 509         if (!inet_can_nonlocal_bind(net, inet) &&
 510             addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
 511             chk_addr_ret != RTN_LOCAL &&
 512             chk_addr_ret != RTN_MULTICAST &&
 513             chk_addr_ret != RTN_BROADCAST)
 514                 goto out;
 515
 516         snum = ntohs(addr->sin_port);
 517         err = -EACCES;
 518         if (snum && snum < inet_prot_sock(net) &&
 519             !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
 520                 goto out;
 521
 522         /*      We keep a pair of addresses. rcv_saddr is the one
 523          *      used by hash lookups, and saddr is used for transmit.
 524          *
 525          *      In the BSD API these are the same except where it
 526          *      would be illegal to use them (multicast/broadcast) in
 527          *      which case the sending device address is used.
 528          */
 529         if (with_lock)
 530                 lock_sock(sk);
 531
 532         /* Check these errors (active socket, double bind). */
 533         err = -EINVAL;
 534         if (sk->sk_state != TCP_CLOSE || inet->inet_num)
 535                 goto out_release_sock;
 536
 537         inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
 538         if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
 539                 inet->inet_saddr = 0;  /* Use device */
 540
 541         /* Make sure we are allowed to bind here. */
 542         if (snum || !(inet->bind_address_no_port ||
 543                       force_bind_address_no_port)) {
 544                 if (sk->sk_prot->get_port(sk, snum)) {
 545                         inet->inet_saddr = inet->inet_rcv_saddr = 0;
 546                         err = -EADDRINUSE;
 547                         goto out_release_sock;
 548                 }
 549                 err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
 550                 if (err) {
 551                         inet->inet_saddr = inet->inet_rcv_saddr = 0;
 552                         goto out_release_sock;
 553                 }
 554         }
 555
 556         if (inet->inet_rcv_saddr)
 557                 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
 558         if (snum)
 559                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
 560         inet->inet_sport = htons(inet->inet_num);
 561         inet->inet_daddr = 0;
 562         inet->inet_dport = 0;
 563         sk_dst_reset(sk);
 564         err = 0;
 565 out_release_sock:
 566         if (with_lock)
 567                 release_sock(sk);
 568 out:
 569         return err;
 570 }
 571
 572 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
 573                        int addr_len, int flags)
 574 {
 575         struct sock *sk = sock->sk;
 576         int err;
 577
 578         if (addr_len < sizeof(uaddr->sa_family))
 579                 return -EINVAL;
 580         if (uaddr->sa_family == AF_UNSPEC)
 581                 return sk->sk_prot->disconnect(sk, flags);
 582
 583         if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
 584                 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
 585                 if (err)
 586                         return err;
 587         }
 588
 589         if (!inet_sk(sk)->inet_num && inet_autobind(sk))
 590                 return -EAGAIN;
 591         return sk->sk_prot->connect(sk, uaddr, addr_len);
 592 }
 593 EXPORT_SYMBOL(inet_dgram_connect);
 594
 595 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
 596 {
 597         DEFINE_WAIT_FUNC(wait, woken_wake_function);
 598
 599         add_wait_queue(sk_sleep(sk), &wait);
 600         sk->sk_write_pending += writebias;
 601
 602         /* Basic assumption: if someone sets sk->sk_err, he _must_
 603          * change state of the socket from TCP_SYN_*.
 604          * Connect() does not allow to get error notifications
 605          * without closing the socket.
 606          */
 607         while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 608                 release_sock(sk);
 609                 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
 610                 lock_sock(sk);
 611                 if (signal_pending(current) || !timeo)
 612                         break;
 613         }
 614         remove_wait_queue(sk_sleep(sk), &wait);
 615         sk->sk_write_pending -= writebias;
 616         return timeo;
 617 }
 618
 619 /*
 620  *      Connect to a remote host. There is regrettably still a little
 621  *      TCP 'magic' in here.
 622  */
 623 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 624                           int addr_len, int flags, int is_sendmsg)
 625 {
 626         struct sock *sk = sock->sk;
 627         int err;
 628         long timeo;
 629
 630         /*
 631          * uaddr can be NULL and addr_len can be 0 if:
 632          * sk is a TCP fastopen active socket and
 633          * TCP_FASTOPEN_CONNECT sockopt is set and
 634          * we already have a valid cookie for this socket.
 635          * In this case, user can call write() after connect().
 636          * write() will invoke tcp_sendmsg_fastopen() which calls
 637          * __inet_stream_connect().
 638          */
 639         if (uaddr) {
 640                 if (addr_len < sizeof(uaddr->sa_family))
 641                         return -EINVAL;
 642
 643                 if (uaddr->sa_family == AF_UNSPEC) {
 644                         err = sk->sk_prot->disconnect(sk, flags);
 645                         sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 646                         goto out;
 647                 }
 648         }
 649
 650         switch (sock->state) {
 651         default:
 652                 err = -EINVAL;
 653                 goto out;
 654         case SS_CONNECTED:
 655                 err = -EISCONN;
 656                 goto out;
 657         case SS_CONNECTING:
 658                 if (inet_sk(sk)->defer_connect)
 659                         err = is_sendmsg ? -EINPROGRESS : -EISCONN;
 660                 else
 661                         err = -EALREADY;
 662                 /* Fall out of switch with err, set for this state */
 663                 break;
 664         case SS_UNCONNECTED:
 665                 err = -EISCONN;
 666                 if (sk->sk_state != TCP_CLOSE)
 667                         goto out;
 668
 669                 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
 670                         err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
 671                         if (err)
 672                                 goto out;
 673                 }
 674
 675                 err = sk->sk_prot->connect(sk, uaddr, addr_len);
 676                 if (err < 0)
 677                         goto out;
 678
 679                 sock->state = SS_CONNECTING;
 680
 681                 if (!err && inet_sk(sk)->defer_connect)
 682                         goto out;
 683
 684                 /* Just entered SS_CONNECTING state; the only
 685                  * difference is that return value in non-blocking
 686                  * case is EINPROGRESS, rather than EALREADY.
 687                  */
 688                 err = -EINPROGRESS;
 689                 break;
 690         }
 691
 692         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
 693
 694         if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 695                 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
 696                                 tcp_sk(sk)->fastopen_req &&
 697                                 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
 698
 699                 /* Error code is set above */
 700                 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
 701                         goto out;
 702
 703                 err = sock_intr_errno(timeo);
 704                 if (signal_pending(current))
 705                         goto out;
 706         }
 707
 708         /* Connection was closed by RST, timeout, ICMP error
 709          * or another process disconnected us.
 710          */
 711         if (sk->sk_state == TCP_CLOSE)
 712                 goto sock_error;
 713
 714         /* sk->sk_err may be not zero now, if RECVERR was ordered by user
 715          * and error was received after socket entered established state.
 716          * Hence, it is handled normally after connect() return successfully.
 717          */
 718
 719         sock->state = SS_CONNECTED;
 720         err = 0;
 721 out:
 722         return err;
 723
 724 sock_error:
 725         err = sock_error(sk) ? : -ECONNABORTED;
 726         sock->state = SS_UNCONNECTED;
 727         if (sk->sk_prot->disconnect(sk, flags))
 728                 sock->state = SS_DISCONNECTING;
 729         goto out;
 730 }
 731 EXPORT_SYMBOL(__inet_stream_connect);
 732
 733 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 734                         int addr_len, int flags)
 735 {
 736         int err;
 737
 738         lock_sock(sock->sk);
 739         err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
 740         release_sock(sock->sk);
 741         return err;
 742 }
 743 EXPORT_SYMBOL(inet_stream_connect);
 744
 745 /*
 746  *      Accept a pending connection. The TCP layer now gives BSD semantics.
 747  */
 748
 749 int inet_accept(struct socket *sock, struct socket *newsock, int flags,
 750                 bool kern)
 751 {
 752         struct sock *sk1 = sock->sk;
 753         int err = -EINVAL;
 754         struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err, kern);
 755
 756         if (!sk2)
 757                 goto do_err;
 758
 759         lock_sock(sk2);
 760
 761         sock_rps_record_flow(sk2);
 762         WARN_ON(!((1 << sk2->sk_state) &
 763                   (TCPF_ESTABLISHED | TCPF_SYN_RECV |
 764                   TCPF_CLOSE_WAIT | TCPF_CLOSE)));
 765
 766         sock_graft(sk2, newsock);
 767
 768         newsock->state = SS_CONNECTED;
 769         err = 0;
 770         release_sock(sk2);
 771 do_err:
 772         return err;
 773 }
 774 EXPORT_SYMBOL(inet_accept);
 775
 776
 777 /*
 778  *      This does both peername and sockname.
 779  */
 780 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
 781                         int peer)
 782 {
 783         struct sock *sk         = sock->sk;
 784         struct inet_sock *inet  = inet_sk(sk);
 785         DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
 786
 787         sin->sin_family = AF_INET;
 788         if (peer) {
 789                 if (!inet->inet_dport ||
 790                     (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
 791                      peer == 1))
 792                         return -ENOTCONN;
 793                 sin->sin_port = inet->inet_dport;
 794                 sin->sin_addr.s_addr = inet->inet_daddr;
 795         } else {
 796                 __be32 addr = inet->inet_rcv_saddr;
 797                 if (!addr)
 798                         addr = inet->inet_saddr;
 799                 sin->sin_port = inet->inet_sport;
 800                 sin->sin_addr.s_addr = addr;
 801         }
 802         memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 803         return sizeof(*sin);
 804 }
 805 EXPORT_SYMBOL(inet_getname);
 806
 807 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
 808 {
 809         struct sock *sk = sock->sk;
 810
 811         sock_rps_record_flow(sk);
 812
 813         /* We may need to bind the socket. */
 814         if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
 815             inet_autobind(sk))
 816                 return -EAGAIN;
 817
 818         return sk->sk_prot->sendmsg(sk, msg, size);
 819 }
 820 EXPORT_SYMBOL(inet_sendmsg);
 821
 822 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
 823                       size_t size, int flags)
 824 {
 825         struct sock *sk = sock->sk;
 826
 827         sock_rps_record_flow(sk);
 828
 829         /* We may need to bind the socket. */
 830         if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
 831             inet_autobind(sk))
 832                 return -EAGAIN;
 833
 834         if (sk->sk_prot->sendpage)
 835                 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
 836         return sock_no_sendpage(sock, page, offset, size, flags);
 837 }
 838 EXPORT_SYMBOL(inet_sendpage);
 839
 840 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
 841                  int flags)
 842 {
 843         struct sock *sk = sock->sk;
 844         int addr_len = 0;
 845         int err;
 846
 847         if (likely(!(flags & MSG_ERRQUEUE)))
 848                 sock_rps_record_flow(sk);
 849
 850         err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
 851                                    flags & ~MSG_DONTWAIT, &addr_len);
 852         if (err >= 0)
 853                 msg->msg_namelen = addr_len;
 854         return err;
 855 }
 856 EXPORT_SYMBOL(inet_recvmsg);
 857
 858 int inet_shutdown(struct socket *sock, int how)
 859 {
 860         struct sock *sk = sock->sk;
 861         int err = 0;
 862
 863         /* This should really check to make sure
 864          * the socket is a TCP socket. (WHY AC...)
 865          */
 866         how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
 867                        1->2 bit 2 snds.
 868                        2->3 */
 869         if ((how & ~SHUTDOWN_MASK) || !how)     /* MAXINT->0 */
 870                 return -EINVAL;
 871
 872         lock_sock(sk);
 873         if (sock->state == SS_CONNECTING) {
 874                 if ((1 << sk->sk_state) &
 875                     (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
 876                         sock->state = SS_DISCONNECTING;
 877                 else
 878                         sock->state = SS_CONNECTED;
 879         }
 880
 881         switch (sk->sk_state) {
 882         case TCP_CLOSE:
 883                 err = -ENOTCONN;
 884                 /* Hack to wake up other listeners, who can poll for
 885                    EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
 886                 /* fall through */
 887         default:
 888                 sk->sk_shutdown |= how;
 889                 if (sk->sk_prot->shutdown)
 890                         sk->sk_prot->shutdown(sk, how);
 891                 break;
 892
 893         /* Remaining two branches are temporary solution for missing
 894          * close() in multithreaded environment. It is _not_ a good idea,
 895          * but we have no choice until close() is repaired at VFS level.
 896          */
 897         case TCP_LISTEN:
 898                 if (!(how & RCV_SHUTDOWN))
 899                         break;
 900                 /* fall through */
 901         case TCP_SYN_SENT:
 902                 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
 903                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 904                 break;
 905         }
 906
 907         /* Wake up anyone sleeping in poll. */
 908         sk->sk_state_change(sk);
 909         release_sock(sk);
 910         return err;
 911 }
 912 EXPORT_SYMBOL(inet_shutdown);
 913
 914 /*
 915  *      ioctl() calls you can issue on an INET socket. Most of these are
 916  *      device configuration and stuff and very rarely used. Some ioctls
 917  *      pass on to the socket itself.
 918  *
 919  *      NOTE: I like the idea of a module for the config stuff. ie ifconfig
 920  *      loads the devconfigure module does its configuring and unloads it.
 921  *      There's a good 20K of config code hanging around the kernel.
 922  */
 923
 924 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 925 {
 926         struct sock *sk = sock->sk;
 927         int err = 0;
 928         struct net *net = sock_net(sk);
 929         void __user *p = (void __user *)arg;
 930         struct ifreq ifr;
 931         struct rtentry rt;
 932
 933         switch (cmd) {
 934         case SIOCGSTAMP:
 935                 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
 936                 break;
 937         case SIOCGSTAMPNS:
 938                 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
 939                 break;
 940         case SIOCADDRT:
 941         case SIOCDELRT:
 942                 if (copy_from_user(&rt, p, sizeof(struct rtentry)))
 943                         return -EFAULT;
 944                 err = ip_rt_ioctl(net, cmd, &rt);
 945                 break;
 946         case SIOCRTMSG:
 947                 err = -EINVAL;
 948                 break;
 949         case SIOCDARP:
 950         case SIOCGARP:
 951         case SIOCSARP:
 952                 err = arp_ioctl(net, cmd, (void __user *)arg);
 953                 break;
 954         case SIOCGIFADDR:
 955         case SIOCGIFBRDADDR:
 956         case SIOCGIFNETMASK:
 957         case SIOCGIFDSTADDR:
 958         case SIOCGIFPFLAGS:
 959                 if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
 960                         return -EFAULT;
 961                 err = devinet_ioctl(net, cmd, &ifr);
 962                 if (!err && copy_to_user(p, &ifr, sizeof(struct ifreq)))
 963                         err = -EFAULT;
 964                 break;
 965
 966         case SIOCSIFADDR:
 967         case SIOCSIFBRDADDR:
 968         case SIOCSIFNETMASK:
 969         case SIOCSIFDSTADDR:
 970         case SIOCSIFPFLAGS:
 971         case SIOCSIFFLAGS:
 972                 if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
 973                         return -EFAULT;
 974                 err = devinet_ioctl(net, cmd, &ifr);
 975                 break;
 976         default:
 977                 if (sk->sk_prot->ioctl)
 978                         err = sk->sk_prot->ioctl(sk, cmd, arg);
 979                 else
 980                         err = -ENOIOCTLCMD;
 981                 break;
 982         }
 983         return err;
 984 }
 985 EXPORT_SYMBOL(inet_ioctl);
 986
 987 #ifdef CONFIG_COMPAT
 988 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 989 {
 990         struct sock *sk = sock->sk;
 991         int err = -ENOIOCTLCMD;
 992
 993         if (sk->sk_prot->compat_ioctl)
 994                 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
 995
 996         return err;
 997 }
 998 #endif
 999
1000 const struct proto_ops inet_stream_ops = {
1001         .family            = PF_INET,
1002         .owner             = THIS_MODULE,
1003         .release           = inet_release,
1004         .bind              = inet_bind,
1005         .connect           = inet_stream_connect,
1006         .socketpair        = sock_no_socketpair,
1007         .accept            = inet_accept,
1008         .getname           = inet_getname,
1009         .poll              = tcp_poll,
1010         .ioctl             = inet_ioctl,
1011         .listen            = inet_listen,
1012         .shutdown          = inet_shutdown,
1013         .setsockopt        = sock_common_setsockopt,
1014         .getsockopt        = sock_common_getsockopt,
1015         .sendmsg           = inet_sendmsg,
1016         .recvmsg           = inet_recvmsg,
1017 #ifdef CONFIG_MMU
1018         .mmap              = tcp_mmap,
1019 #endif
1020         .sendpage          = inet_sendpage,
1021         .splice_read       = tcp_splice_read,
1022         .read_sock         = tcp_read_sock,
1023         .sendmsg_locked    = tcp_sendmsg_locked,
1024         .sendpage_locked   = tcp_sendpage_locked,
1025         .peek_len          = tcp_peek_len,
1026 #ifdef CONFIG_COMPAT
1027         .compat_setsockopt = compat_sock_common_setsockopt,
1028         .compat_getsockopt = compat_sock_common_getsockopt,
1029         .compat_ioctl      = inet_compat_ioctl,
1030 #endif
1031         .set_rcvlowat      = tcp_set_rcvlowat,
1032 };
1033 EXPORT_SYMBOL(inet_stream_ops);
1034
1035 const struct proto_ops inet_dgram_ops = {
1036         .family            = PF_INET,
1037         .owner             = THIS_MODULE,
1038         .release           = inet_release,
1039         .bind              = inet_bind,
1040         .connect           = inet_dgram_connect,
1041         .socketpair        = sock_no_socketpair,
1042         .accept            = sock_no_accept,
1043         .getname           = inet_getname,
1044         .poll              = udp_poll,
1045         .ioctl             = inet_ioctl,
1046         .listen            = sock_no_listen,
1047         .shutdown          = inet_shutdown,
1048         .setsockopt        = sock_common_setsockopt,
1049         .getsockopt        = sock_common_getsockopt,
1050         .sendmsg           = inet_sendmsg,
1051         .recvmsg           = inet_recvmsg,
1052         .mmap              = sock_no_mmap,
1053         .sendpage          = inet_sendpage,
1054         .set_peek_off      = sk_set_peek_off,
1055 #ifdef CONFIG_COMPAT
1056         .compat_setsockopt = compat_sock_common_setsockopt,
1057         .compat_getsockopt = compat_sock_common_getsockopt,
1058         .compat_ioctl      = inet_compat_ioctl,
1059 #endif
1060 };
1061 EXPORT_SYMBOL(inet_dgram_ops);
1062
1063 /*
1064  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1065  * udp_poll
1066  */
1067 static const struct proto_ops inet_sockraw_ops = {
1068         .family            = PF_INET,
1069         .owner             = THIS_MODULE,
1070         .release           = inet_release,
1071         .bind              = inet_bind,
1072         .connect           = inet_dgram_connect,
1073         .socketpair        = sock_no_socketpair,
1074         .accept            = sock_no_accept,
1075         .getname           = inet_getname,
1076         .poll              = datagram_poll,
1077         .ioctl             = inet_ioctl,
1078         .listen            = sock_no_listen,
1079         .shutdown          = inet_shutdown,
1080         .setsockopt        = sock_common_setsockopt,
1081         .getsockopt        = sock_common_getsockopt,
1082         .sendmsg           = inet_sendmsg,
1083         .recvmsg           = inet_recvmsg,
1084         .mmap              = sock_no_mmap,
1085         .sendpage          = inet_sendpage,
1086 #ifdef CONFIG_COMPAT
1087         .compat_setsockopt = compat_sock_common_setsockopt,
1088         .compat_getsockopt = compat_sock_common_getsockopt,
1089         .compat_ioctl      = inet_compat_ioctl,
1090 #endif
1091 };
1092
1093 static const struct net_proto_family inet_family_ops = {
1094         .family = PF_INET,
1095         .create = inet_create,
1096         .owner  = THIS_MODULE,
1097 };
1098
1099 /* Upon startup we insert all the elements in inetsw_array[] into
1100  * the linked list inetsw.
1101  */
1102 static struct inet_protosw inetsw_array[] =
1103 {
1104         {
1105                 .type =       SOCK_STREAM,
1106                 .protocol =   IPPROTO_TCP,
1107                 .prot =       &tcp_prot,
1108                 .ops =        &inet_stream_ops,
1109                 .flags =      INET_PROTOSW_PERMANENT |
1110                               INET_PROTOSW_ICSK,
1111         },
1112
1113         {
1114                 .type =       SOCK_DGRAM,
1115                 .protocol =   IPPROTO_UDP,
1116                 .prot =       &udp_prot,
1117                 .ops =        &inet_dgram_ops,
1118                 .flags =      INET_PROTOSW_PERMANENT,
1119        },
1120
1121        {
1122                 .type =       SOCK_DGRAM,
1123                 .protocol =   IPPROTO_ICMP,
1124                 .prot =       &ping_prot,
1125                 .ops =        &inet_sockraw_ops,
1126                 .flags =      INET_PROTOSW_REUSE,
1127        },
1128
1129        {
1130                .type =       SOCK_RAW,
1131                .protocol =   IPPROTO_IP,        /* wild card */
1132                .prot =       &raw_prot,
1133                .ops =        &inet_sockraw_ops,
1134                .flags =      INET_PROTOSW_REUSE,
1135        }
1136 };
1137
1138 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1139
1140 void inet_register_protosw(struct inet_protosw *p)
1141 {
1142         struct list_head *lh;
1143         struct inet_protosw *answer;
1144         int protocol = p->protocol;
1145         struct list_head *last_perm;
1146
1147         spin_lock_bh(&inetsw_lock);
1148
1149         if (p->type >= SOCK_MAX)
1150                 goto out_illegal;
1151
1152         /* If we are trying to override a permanent protocol, bail. */
1153         last_perm = &inetsw[p->type];
1154         list_for_each(lh, &inetsw[p->type]) {
1155                 answer = list_entry(lh, struct inet_protosw, list);
1156                 /* Check only the non-wild match. */
1157                 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1158                         break;
1159                 if (protocol == answer->protocol)
1160                         goto out_permanent;
1161                 last_perm = lh;
1162         }
1163
1164         /* Add the new entry after the last permanent entry if any, so that
1165          * the new entry does not override a permanent entry when matched with
1166          * a wild-card protocol. But it is allowed to override any existing
1167          * non-permanent entry.  This means that when we remove this entry, the
1168          * system automatically returns to the old behavior.
1169          */
1170         list_add_rcu(&p->list, last_perm);
1171 out:
1172         spin_unlock_bh(&inetsw_lock);
1173
1174         return;
1175
1176 out_permanent:
1177         pr_err("Attempt to override permanent protocol %d\n", protocol);
1178         goto out;
1179
1180 out_illegal:
1181         pr_err("Ignoring attempt to register invalid socket type %d\n",
1182                p->type);
1183         goto out;
1184 }
1185 EXPORT_SYMBOL(inet_register_protosw);
1186
1187 void inet_unregister_protosw(struct inet_protosw *p)
1188 {
1189         if (INET_PROTOSW_PERMANENT & p->flags) {
1190                 pr_err("Attempt to unregister permanent protocol %d\n",
1191                        p->protocol);
1192         } else {
1193                 spin_lock_bh(&inetsw_lock);
1194                 list_del_rcu(&p->list);
1195                 spin_unlock_bh(&inetsw_lock);
1196
1197                 synchronize_net();
1198         }
1199 }
1200 EXPORT_SYMBOL(inet_unregister_protosw);
1201
1202 static int inet_sk_reselect_saddr(struct sock *sk)
1203 {
1204         struct inet_sock *inet = inet_sk(sk);
1205         __be32 old_saddr = inet->inet_saddr;
1206         __be32 daddr = inet->inet_daddr;
1207         struct flowi4 *fl4;
1208         struct rtable *rt;
1209         __be32 new_saddr;
1210         struct ip_options_rcu *inet_opt;
1211
1212         inet_opt = rcu_dereference_protected(inet->inet_opt,
1213                                              lockdep_sock_is_held(sk));
1214         if (inet_opt && inet_opt->opt.srr)
1215                 daddr = inet_opt->opt.faddr;
1216
1217         /* Query new route. */
1218         fl4 = &inet->cork.fl.u.ip4;
1219         rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1220                               sk->sk_bound_dev_if, sk->sk_protocol,
1221                               inet->inet_sport, inet->inet_dport, sk);
1222         if (IS_ERR(rt))
1223                 return PTR_ERR(rt);
1224
1225         sk_setup_caps(sk, &rt->dst);
1226
1227         new_saddr = fl4->saddr;
1228
1229         if (new_saddr == old_saddr)
1230                 return 0;
1231
1232         if (sock_net(sk)->ipv4.sysctl_ip_dynaddr > 1) {
1233                 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1234                         __func__, &old_saddr, &new_saddr);
1235         }
1236
1237         inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1238
1239         /*
1240          * XXX The only one ugly spot where we need to
1241          * XXX really change the sockets identity after
1242          * XXX it has entered the hashes. -DaveM
1243          *
1244          * Besides that, it does not check for connection
1245          * uniqueness. Wait for troubles.
1246          */
1247         return __sk_prot_rehash(sk);
1248 }
1249
1250 int inet_sk_rebuild_header(struct sock *sk)
1251 {
1252         struct inet_sock *inet = inet_sk(sk);
1253         struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1254         __be32 daddr;
1255         struct ip_options_rcu *inet_opt;
1256         struct flowi4 *fl4;
1257         int err;
1258
1259         /* Route is OK, nothing to do. */
1260         if (rt)
1261                 return 0;
1262
1263         /* Reroute. */
1264         rcu_read_lock();
1265         inet_opt = rcu_dereference(inet->inet_opt);
1266         daddr = inet->inet_daddr;
1267         if (inet_opt && inet_opt->opt.srr)
1268                 daddr = inet_opt->opt.faddr;
1269         rcu_read_unlock();
1270         fl4 = &inet->cork.fl.u.ip4;
1271         rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1272                                    inet->inet_dport, inet->inet_sport,
1273                                    sk->sk_protocol, RT_CONN_FLAGS(sk),
1274                                    sk->sk_bound_dev_if);
1275         if (!IS_ERR(rt)) {
1276                 err = 0;
1277                 sk_setup_caps(sk, &rt->dst);
1278         } else {
1279                 err = PTR_ERR(rt);
1280
1281                 /* Routing failed... */
1282                 sk->sk_route_caps = 0;
1283                 /*
1284                  * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1285                  * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1286                  */
1287                 if (!sock_net(sk)->ipv4.sysctl_ip_dynaddr ||
1288                     sk->sk_state != TCP_SYN_SENT ||
1289                     (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1290                     (err = inet_sk_reselect_saddr(sk)) != 0)
1291                         sk->sk_err_soft = -err;
1292         }
1293
1294         return err;
1295 }
1296 EXPORT_SYMBOL(inet_sk_rebuild_header);
1297
1298 void inet_sk_set_state(struct sock *sk, int state)
1299 {
1300         trace_inet_sock_set_state(sk, sk->sk_state, state);
1301         sk->sk_state = state;
1302 }
1303 EXPORT_SYMBOL(inet_sk_set_state);
1304
1305 void inet_sk_state_store(struct sock *sk, int newstate)
1306 {
1307         trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1308         smp_store_release(&sk->sk_state, newstate);
1309 }
1310
1311 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1312                                  netdev_features_t features)
1313 {
1314         bool udpfrag = false, fixedid = false, gso_partial, encap;
1315         struct sk_buff *segs = ERR_PTR(-EINVAL);
1316         const struct net_offload *ops;
1317         unsigned int offset = 0;
1318         struct iphdr *iph;
1319         int proto, tot_len;
1320         int nhoff;
1321         int ihl;
1322         int id;
1323
1324         skb_reset_network_header(skb);
1325         nhoff = skb_network_header(skb) - skb_mac_header(skb);
1326         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1327                 goto out;
1328
1329         iph = ip_hdr(skb);
1330         ihl = iph->ihl * 4;
1331         if (ihl < sizeof(*iph))
1332                 goto out;
1333
1334         id = ntohs(iph->id);
1335         proto = iph->protocol;
1336
1337         /* Warning: after this point, iph might be no longer valid */
1338         if (unlikely(!pskb_may_pull(skb, ihl)))
1339                 goto out;
1340         __skb_pull(skb, ihl);
1341
1342         encap = SKB_GSO_CB(skb)->encap_level > 0;
1343         if (encap)
1344                 features &= skb->dev->hw_enc_features;
1345         SKB_GSO_CB(skb)->encap_level += ihl;
1346
1347         skb_reset_transport_header(skb);
1348
1349         segs = ERR_PTR(-EPROTONOSUPPORT);
1350
1351         if (!skb->encapsulation || encap) {
1352                 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1353                 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1354
1355                 /* fixed ID is invalid if DF bit is not set */
1356                 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1357                         goto out;
1358         }
1359
1360         ops = rcu_dereference(inet_offloads[proto]);
1361         if (likely(ops && ops->callbacks.gso_segment))
1362                 segs = ops->callbacks.gso_segment(skb, features);
1363
1364         if (IS_ERR_OR_NULL(segs))
1365                 goto out;
1366
1367         gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1368
1369         skb = segs;
1370         do {
1371                 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1372                 if (udpfrag) {
1373                         iph->frag_off = htons(offset >> 3);
1374                         if (skb->next)
1375                                 iph->frag_off |= htons(IP_MF);
1376                         offset += skb->len - nhoff - ihl;
1377                         tot_len = skb->len - nhoff;
1378                 } else if (skb_is_gso(skb)) {
1379                         if (!fixedid) {
1380                                 iph->id = htons(id);
1381                                 id += skb_shinfo(skb)->gso_segs;
1382                         }
1383
1384                         if (gso_partial)
1385                                 tot_len = skb_shinfo(skb)->gso_size +
1386                                           SKB_GSO_CB(skb)->data_offset +
1387                                           skb->head - (unsigned char *)iph;
1388                         else
1389                                 tot_len = skb->len - nhoff;
1390                 } else {
1391                         if (!fixedid)
1392                                 iph->id = htons(id++);
1393                         tot_len = skb->len - nhoff;
1394                 }
1395                 iph->tot_len = htons(tot_len);
1396                 ip_send_check(iph);
1397                 if (encap)
1398                         skb_reset_inner_headers(skb);
1399                 skb->network_header = (u8 *)iph - skb->head;
1400                 skb_reset_mac_len(skb);
1401         } while ((skb = skb->next));
1402
1403 out:
1404         return segs;
1405 }
1406 EXPORT_SYMBOL(inet_gso_segment);
1407
1408 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1409 {
1410         const struct net_offload *ops;
1411         struct sk_buff *pp = NULL;
1412         const struct iphdr *iph;
1413         struct sk_buff *p;
1414         unsigned int hlen;
1415         unsigned int off;
1416         unsigned int id;
1417         int flush = 1;
1418         int proto;
1419
1420         off = skb_gro_offset(skb);
1421         hlen = off + sizeof(*iph);
1422         iph = skb_gro_header_fast(skb, off);
1423         if (skb_gro_header_hard(skb, hlen)) {
1424                 iph = skb_gro_header_slow(skb, hlen, off);
1425                 if (unlikely(!iph))
1426                         goto out;
1427         }
1428
1429         proto = iph->protocol;
1430
1431         rcu_read_lock();
1432         ops = rcu_dereference(inet_offloads[proto]);
1433         if (!ops || !ops->callbacks.gro_receive)
1434                 goto out_unlock;
1435
1436         if (*(u8 *)iph != 0x45)
1437                 goto out_unlock;
1438
1439         if (ip_is_fragment(iph))
1440                 goto out_unlock;
1441
1442         if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1443                 goto out_unlock;
1444
1445         id = ntohl(*(__be32 *)&iph->id);
1446         flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1447         id >>= 16;
1448
1449         list_for_each_entry(p, head, list) {
1450                 struct iphdr *iph2;
1451                 u16 flush_id;
1452
1453                 if (!NAPI_GRO_CB(p)->same_flow)
1454                         continue;
1455
1456                 iph2 = (struct iphdr *)(p->data + off);
1457                 /* The above works because, with the exception of the top
1458                  * (inner most) layer, we only aggregate pkts with the same
1459                  * hdr length so all the hdrs we'll need to verify will start
1460                  * at the same offset.
1461                  */
1462                 if ((iph->protocol ^ iph2->protocol) |
1463                     ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1464                     ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1465                         NAPI_GRO_CB(p)->same_flow = 0;
1466                         continue;
1467                 }
1468
1469                 /* All fields must match except length and checksum. */
1470                 NAPI_GRO_CB(p)->flush |=
1471                         (iph->ttl ^ iph2->ttl) |
1472                         (iph->tos ^ iph2->tos) |
1473                         ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1474
1475                 NAPI_GRO_CB(p)->flush |= flush;
1476
1477                 /* We need to store of the IP ID check to be included later
1478                  * when we can verify that this packet does in fact belong
1479                  * to a given flow.
1480                  */
1481                 flush_id = (u16)(id - ntohs(iph2->id));
1482
1483                 /* This bit of code makes it much easier for us to identify
1484                  * the cases where we are doing atomic vs non-atomic IP ID
1485                  * checks.  Specifically an atomic check can return IP ID
1486                  * values 0 - 0xFFFF, while a non-atomic check can only
1487                  * return 0 or 0xFFFF.
1488                  */
1489                 if (!NAPI_GRO_CB(p)->is_atomic ||
1490                     !(iph->frag_off & htons(IP_DF))) {
1491                         flush_id ^= NAPI_GRO_CB(p)->count;
1492                         flush_id = flush_id ? 0xFFFF : 0;
1493                 }
1494
1495                 /* If the previous IP ID value was based on an atomic
1496                  * datagram we can overwrite the value and ignore it.
1497                  */
1498                 if (NAPI_GRO_CB(skb)->is_atomic)
1499                         NAPI_GRO_CB(p)->flush_id = flush_id;
1500                 else
1501                         NAPI_GRO_CB(p)->flush_id |= flush_id;
1502         }
1503
1504         NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1505         NAPI_GRO_CB(skb)->flush |= flush;
1506         skb_set_network_header(skb, off);
1507         /* The above will be needed by the transport layer if there is one
1508          * immediately following this IP hdr.
1509          */
1510
1511         /* Note : No need to call skb_gro_postpull_rcsum() here,
1512          * as we already checked checksum over ipv4 header was 0
1513          */
1514         skb_gro_pull(skb, sizeof(*iph));
1515         skb_set_transport_header(skb, skb_gro_offset(skb));
1516
1517         pp = call_gro_receive(ops->callbacks.gro_receive, head, skb);
1518
1519 out_unlock:
1520         rcu_read_unlock();
1521
1522 out:
1523         skb_gro_flush_final(skb, pp, flush);
1524
1525         return pp;
1526 }
1527 EXPORT_SYMBOL(inet_gro_receive);
1528
1529 static struct sk_buff *ipip_gro_receive(struct list_head *head,
1530                                         struct sk_buff *skb)
1531 {
1532         if (NAPI_GRO_CB(skb)->encap_mark) {
1533                 NAPI_GRO_CB(skb)->flush = 1;
1534                 return NULL;
1535         }
1536
1537         NAPI_GRO_CB(skb)->encap_mark = 1;
1538
1539         return inet_gro_receive(head, skb);
1540 }
1541
1542 #define SECONDS_PER_DAY 86400
1543
1544 /* inet_current_timestamp - Return IP network timestamp
1545  *
1546  * Return milliseconds since midnight in network byte order.
1547  */
1548 __be32 inet_current_timestamp(void)
1549 {
1550         u32 secs;
1551         u32 msecs;
1552         struct timespec64 ts;
1553
1554         ktime_get_real_ts64(&ts);
1555
1556         /* Get secs since midnight. */
1557         (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1558         /* Convert to msecs. */
1559         msecs = secs * MSEC_PER_SEC;
1560         /* Convert nsec to msec. */
1561         msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1562
1563         /* Convert to network byte order. */
1564         return htonl(msecs);
1565 }
1566 EXPORT_SYMBOL(inet_current_timestamp);
1567
1568 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1569 {
1570         if (sk->sk_family == AF_INET)
1571                 return ip_recv_error(sk, msg, len, addr_len);
1572 #if IS_ENABLED(CONFIG_IPV6)
1573         if (sk->sk_family == AF_INET6)
1574                 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1575 #endif
1576         return -EINVAL;
1577 }
1578
1579 int inet_gro_complete(struct sk_buff *skb, int nhoff)
1580 {
1581         __be16 newlen = htons(skb->len - nhoff);
1582         struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1583         const struct net_offload *ops;
1584         int proto = iph->protocol;
1585         int err = -ENOSYS;
1586
1587         if (skb->encapsulation) {
1588                 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1589                 skb_set_inner_network_header(skb, nhoff);
1590         }
1591
1592         csum_replace2(&iph->check, iph->tot_len, newlen);
1593         iph->tot_len = newlen;
1594
1595         rcu_read_lock();
1596         ops = rcu_dereference(inet_offloads[proto]);
1597         if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1598                 goto out_unlock;
1599
1600         /* Only need to add sizeof(*iph) to get to the next hdr below
1601          * because any hdr with option will have been flushed in
1602          * inet_gro_receive().
1603          */
1604         err = ops->callbacks.gro_complete(skb, nhoff + sizeof(*iph));
1605
1606 out_unlock:
1607         rcu_read_unlock();
1608
1609         return err;
1610 }
1611 EXPORT_SYMBOL(inet_gro_complete);
1612
1613 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1614 {
1615         skb->encapsulation = 1;
1616         skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1617         return inet_gro_complete(skb, nhoff);
1618 }
1619
1620 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1621                          unsigned short type, unsigned char protocol,
1622                          struct net *net)
1623 {
1624         struct socket *sock;
1625         int rc = sock_create_kern(net, family, type, protocol, &sock);
1626
1627         if (rc == 0) {
1628                 *sk = sock->sk;
1629                 (*sk)->sk_allocation = GFP_ATOMIC;
1630                 /*
1631                  * Unhash it so that IP input processing does not even see it,
1632                  * we do not wish this socket to see incoming packets.
1633                  */
1634                 (*sk)->sk_prot->unhash(*sk);
1635         }
1636         return rc;
1637 }
1638 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1639
1640 u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt)
1641 {
1642         return  *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt);
1643 }
1644 EXPORT_SYMBOL_GPL(snmp_get_cpu_field);
1645
1646 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1647 {
1648         unsigned long res = 0;
1649         int i;
1650
1651         for_each_possible_cpu(i)
1652                 res += snmp_get_cpu_field(mib, i, offt);
1653         return res;
1654 }
1655 EXPORT_SYMBOL_GPL(snmp_fold_field);
1656
1657 #if BITS_PER_LONG==32
1658
1659 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1660                          size_t syncp_offset)
1661 {
1662         void *bhptr;
1663         struct u64_stats_sync *syncp;
1664         u64 v;
1665         unsigned int start;
1666
1667         bhptr = per_cpu_ptr(mib, cpu);
1668         syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1669         do {
1670                 start = u64_stats_fetch_begin_irq(syncp);
1671                 v = *(((u64 *)bhptr) + offt);
1672         } while (u64_stats_fetch_retry_irq(syncp, start));
1673
1674         return v;
1675 }
1676 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1677
1678 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1679 {
1680         u64 res = 0;
1681         int cpu;
1682
1683         for_each_possible_cpu(cpu) {
1684                 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1685         }
1686         return res;
1687 }
1688 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1689 #endif
1690
1691 #ifdef CONFIG_IP_MULTICAST
1692 static const struct net_protocol igmp_protocol = {
1693         .handler =      igmp_rcv,
1694         .netns_ok =     1,
1695 };
1696 #endif
1697
1698 /* thinking of making this const? Don't.
1699  * early_demux can change based on sysctl.
1700  */
1701 static struct net_protocol tcp_protocol = {
1702         .early_demux    =       tcp_v4_early_demux,
1703         .early_demux_handler =  tcp_v4_early_demux,
1704         .handler        =       tcp_v4_rcv,
1705         .err_handler    =       tcp_v4_err,
1706         .no_policy      =       1,
1707         .netns_ok       =       1,
1708         .icmp_strict_tag_validation = 1,
1709 };
1710
1711 /* thinking of making this const? Don't.
1712  * early_demux can change based on sysctl.
1713  */
1714 static struct net_protocol udp_protocol = {
1715         .early_demux =  udp_v4_early_demux,
1716         .early_demux_handler =  udp_v4_early_demux,
1717         .handler =      udp_rcv,
1718         .err_handler =  udp_err,
1719         .no_policy =    1,
1720         .netns_ok =     1,
1721 };
1722
1723 static const struct net_protocol icmp_protocol = {
1724         .handler =      icmp_rcv,
1725         .err_handler =  icmp_err,
1726         .no_policy =    1,
1727         .netns_ok =     1,
1728 };
1729
1730 static __net_init int ipv4_mib_init_net(struct net *net)
1731 {
1732         int i;
1733
1734         net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1735         if (!net->mib.tcp_statistics)
1736                 goto err_tcp_mib;
1737         net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1738         if (!net->mib.ip_statistics)
1739                 goto err_ip_mib;
1740
1741         for_each_possible_cpu(i) {
1742                 struct ipstats_mib *af_inet_stats;
1743                 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1744                 u64_stats_init(&af_inet_stats->syncp);
1745         }
1746
1747         net->mib.net_statistics = alloc_percpu(struct linux_mib);
1748         if (!net->mib.net_statistics)
1749                 goto err_net_mib;
1750         net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1751         if (!net->mib.udp_statistics)
1752                 goto err_udp_mib;
1753         net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1754         if (!net->mib.udplite_statistics)
1755                 goto err_udplite_mib;
1756         net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1757         if (!net->mib.icmp_statistics)
1758                 goto err_icmp_mib;
1759         net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1760                                               GFP_KERNEL);
1761         if (!net->mib.icmpmsg_statistics)
1762                 goto err_icmpmsg_mib;
1763
1764         tcp_mib_init(net);
1765         return 0;
1766
1767 err_icmpmsg_mib:
1768         free_percpu(net->mib.icmp_statistics);
1769 err_icmp_mib:
1770         free_percpu(net->mib.udplite_statistics);
1771 err_udplite_mib:
1772         free_percpu(net->mib.udp_statistics);
1773 err_udp_mib:
1774         free_percpu(net->mib.net_statistics);
1775 err_net_mib:
1776         free_percpu(net->mib.ip_statistics);
1777 err_ip_mib:
1778         free_percpu(net->mib.tcp_statistics);
1779 err_tcp_mib:
1780         return -ENOMEM;
1781 }
1782
1783 static __net_exit void ipv4_mib_exit_net(struct net *net)
1784 {
1785         kfree(net->mib.icmpmsg_statistics);
1786         free_percpu(net->mib.icmp_statistics);
1787         free_percpu(net->mib.udplite_statistics);
1788         free_percpu(net->mib.udp_statistics);
1789         free_percpu(net->mib.net_statistics);
1790         free_percpu(net->mib.ip_statistics);
1791         free_percpu(net->mib.tcp_statistics);
1792 }
1793
1794 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1795         .init = ipv4_mib_init_net,
1796         .exit = ipv4_mib_exit_net,
1797 };
1798
1799 static int __init init_ipv4_mibs(void)
1800 {
1801         return register_pernet_subsys(&ipv4_mib_ops);
1802 }
1803
1804 static __net_init int inet_init_net(struct net *net)
1805 {
1806         /*
1807          * Set defaults for local port range
1808          */
1809         seqlock_init(&net->ipv4.ip_local_ports.lock);
1810         net->ipv4.ip_local_ports.range[0] =  32768;
1811         net->ipv4.ip_local_ports.range[1] =  60999;
1812
1813         seqlock_init(&net->ipv4.ping_group_range.lock);
1814         /*
1815          * Sane defaults - nobody may create ping sockets.
1816          * Boot scripts should set this to distro-specific group.
1817          */
1818         net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1819         net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1820
1821         /* Default values for sysctl-controlled parameters.
1822          * We set them here, in case sysctl is not compiled.
1823          */
1824         net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1825         net->ipv4.sysctl_ip_fwd_update_priority = 1;
1826         net->ipv4.sysctl_ip_dynaddr = 0;
1827         net->ipv4.sysctl_ip_early_demux = 1;
1828         net->ipv4.sysctl_udp_early_demux = 1;
1829         net->ipv4.sysctl_tcp_early_demux = 1;
1830 #ifdef CONFIG_SYSCTL
1831         net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1832 #endif
1833
1834         /* Some igmp sysctl, whose values are always used */
1835         net->ipv4.sysctl_igmp_max_memberships = 20;
1836         net->ipv4.sysctl_igmp_max_msf = 10;
1837         /* IGMP reports for link-local multicast groups are enabled by default */
1838         net->ipv4.sysctl_igmp_llm_reports = 1;
1839         net->ipv4.sysctl_igmp_qrv = 2;
1840
1841         return 0;
1842 }
1843
1844 static __net_exit void inet_exit_net(struct net *net)
1845 {
1846 }
1847
1848 static __net_initdata struct pernet_operations af_inet_ops = {
1849         .init = inet_init_net,
1850         .exit = inet_exit_net,
1851 };
1852
1853 static int __init init_inet_pernet_ops(void)
1854 {
1855         return register_pernet_subsys(&af_inet_ops);
1856 }
1857
1858 static int ipv4_proc_init(void);
1859
1860 /*
1861  *      IP protocol layer initialiser
1862  */
1863
1864 static struct packet_offload ip_packet_offload __read_mostly = {
1865         .type = cpu_to_be16(ETH_P_IP),
1866         .callbacks = {
1867                 .gso_segment = inet_gso_segment,
1868                 .gro_receive = inet_gro_receive,
1869                 .gro_complete = inet_gro_complete,
1870         },
1871 };
1872
1873 static const struct net_offload ipip_offload = {
1874         .callbacks = {
1875                 .gso_segment    = inet_gso_segment,
1876                 .gro_receive    = ipip_gro_receive,
1877                 .gro_complete   = ipip_gro_complete,
1878         },
1879 };
1880
1881 static int __init ipip_offload_init(void)
1882 {
1883         return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1884 }
1885
1886 static int __init ipv4_offload_init(void)
1887 {
1888         /*
1889          * Add offloads
1890          */
1891         if (udpv4_offload_init() < 0)
1892                 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1893         if (tcpv4_offload_init() < 0)
1894                 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1895         if (ipip_offload_init() < 0)
1896                 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1897
1898         dev_add_offload(&ip_packet_offload);
1899         return 0;
1900 }
1901
1902 fs_initcall(ipv4_offload_init);
1903
1904 static struct packet_type ip_packet_type __read_mostly = {
1905         .type = cpu_to_be16(ETH_P_IP),
1906         .func = ip_rcv,
1907         .list_func = ip_list_rcv,
1908 };
1909
1910 static int __init inet_init(void)
1911 {
1912         struct inet_protosw *q;
1913         struct list_head *r;
1914         int rc = -EINVAL;
1915
1916         sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1917
1918         rc = proto_register(&tcp_prot, 1);
1919         if (rc)
1920                 goto out;
1921
1922         rc = proto_register(&udp_prot, 1);
1923         if (rc)
1924                 goto out_unregister_tcp_proto;
1925
1926         rc = proto_register(&raw_prot, 1);
1927         if (rc)
1928                 goto out_unregister_udp_proto;
1929
1930         rc = proto_register(&ping_prot, 1);
1931         if (rc)
1932                 goto out_unregister_raw_proto;
1933
1934         /*
1935          *      Tell SOCKET that we are alive...
1936          */
1937
1938         (void)sock_register(&inet_family_ops);
1939
1940 #ifdef CONFIG_SYSCTL
1941         ip_static_sysctl_init();
1942 #endif
1943
1944         /*
1945          *      Add all the base protocols.
1946          */
1947
1948         if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1949                 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1950         if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1951                 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1952         if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1953                 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1954 #ifdef CONFIG_IP_MULTICAST
1955         if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1956                 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1957 #endif
1958
1959         /* Register the socket-side information for inet_create. */
1960         for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1961                 INIT_LIST_HEAD(r);
1962
1963         for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1964                 inet_register_protosw(q);
1965
1966         /*
1967          *      Set the ARP module up
1968          */
1969
1970         arp_init();
1971
1972         /*
1973          *      Set the IP module up
1974          */
1975
1976         ip_init();
1977
1978         /* Setup TCP slab cache for open requests. */
1979         tcp_init();
1980
1981         /* Setup UDP memory threshold */
1982         udp_init();
1983
1984         /* Add UDP-Lite (RFC 3828) */
1985         udplite4_register();
1986
1987         ping_init();
1988
1989         /*
1990          *      Set the ICMP layer up
1991          */
1992
1993         if (icmp_init() < 0)
1994                 panic("Failed to create the ICMP control socket.\n");
1995
1996         /*
1997          *      Initialise the multicast router
1998          */
1999 #if defined(CONFIG_IP_MROUTE)
2000         if (ip_mr_init())
2001                 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2002 #endif
2003
2004         if (init_inet_pernet_ops())
2005                 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2006         /*
2007          *      Initialise per-cpu ipv4 mibs
2008          */
2009
2010         if (init_ipv4_mibs())
2011                 pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
2012
2013         ipv4_proc_init();
2014
2015         ipfrag_init();
2016
2017         dev_add_pack(&ip_packet_type);
2018
2019         ip_tunnel_core_init();
2020
2021         rc = 0;
2022 out:
2023         return rc;
2024 out_unregister_raw_proto:
2025         proto_unregister(&raw_prot);
2026 out_unregister_udp_proto:
2027         proto_unregister(&udp_prot);
2028 out_unregister_tcp_proto:
2029         proto_unregister(&tcp_prot);
2030         goto out;
2031 }
2032
2033 fs_initcall(inet_init);
2034
2035 /* ------------------------------------------------------------------------ */
2036
2037 #ifdef CONFIG_PROC_FS
2038 static int __init ipv4_proc_init(void)
2039 {
2040         int rc = 0;
2041
2042         if (raw_proc_init())
2043                 goto out_raw;
2044         if (tcp4_proc_init())
2045                 goto out_tcp;
2046         if (udp4_proc_init())
2047                 goto out_udp;
2048         if (ping_proc_init())
2049                 goto out_ping;
2050         if (ip_misc_proc_init())
2051                 goto out_misc;
2052 out:
2053         return rc;
2054 out_misc:
2055         ping_proc_exit();
2056 out_ping:
2057         udp4_proc_exit();
2058 out_udp:
2059         tcp4_proc_exit();
2060 out_tcp:
2061         raw_proc_exit();
2062 out_raw:
2063         rc = -ENOMEM;
2064         goto out;
2065 }
2066
2067 #else /* CONFIG_PROC_FS */
2068 static int __init ipv4_proc_init(void)
2069 {
2070         return 0;
2071 }
2072 #endif /* CONFIG_PROC_FS */