net/ipv4/af_inet.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   4  *              operating system.  INET is implemented using the  BSD Socket
   5  *              interface as the means of communication with the user level.
   6  *
   7  *              PF_INET protocol family socket handler.
   8  *
   9  * Authors:     Ross Biro
  10  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  11  *              Florian La Roche, <flla@stud.uni-sb.de>
  12  *              Alan Cox, <A.Cox@swansea.ac.uk>
  13  *
  14  * Changes (see also sock.c)
  15  *
  16  *              piggy,
  17  *              Karl Knutson    :       Socket protocol table
  18  *              A.N.Kuznetsov   :       Socket death error in accept().
  19  *              John Richardson :       Fix non blocking error in connect()
  20  *                                      so sockets that fail to connect
  21  *                                      don't return -EINPROGRESS.
  22  *              Alan Cox        :       Asynchronous I/O support
  23  *              Alan Cox        :       Keep correct socket pointer on sock
  24  *                                      structures
  25  *                                      when accept() ed
  26  *              Alan Cox        :       Semantics of SO_LINGER aren't state
  27  *                                      moved to close when you look carefully.
  28  *                                      With this fixed and the accept bug fixed
  29  *                                      some RPC stuff seems happier.
  30  *              Niibe Yutaka    :       4.4BSD style write async I/O
  31  *              Alan Cox,
  32  *              Tony Gale       :       Fixed reuse semantics.
  33  *              Alan Cox        :       bind() shouldn't abort existing but dead
  34  *                                      sockets. Stops FTP netin:.. I hope.
  35  *              Alan Cox        :       bind() works correctly for RAW sockets.
  36  *                                      Note that FreeBSD at least was broken
  37  *                                      in this respect so be careful with
  38  *                                      compatibility tests...
  39  *              Alan Cox        :       routing cache support
  40  *              Alan Cox        :       memzero the socket structure for
  41  *                                      compactness.
  42  *              Matt Day        :       nonblock connect error handler
  43  *              Alan Cox        :       Allow large numbers of pending sockets
  44  *                                      (eg for big web sites), but only if
  45  *                                      specifically application requested.
  46  *              Alan Cox        :       New buffering throughout IP. Used
  47  *                                      dumbly.
  48  *              Alan Cox        :       New buffering now used smartly.
  49  *              Alan Cox        :       BSD rather than common sense
  50  *                                      interpretation of listen.
  51  *              Germano Caronni :       Assorted small races.
  52  *              Alan Cox        :       sendmsg/recvmsg basic support.
  53  *              Alan Cox        :       Only sendmsg/recvmsg now supported.
  54  *              Alan Cox        :       Locked down bind (see security list).
  55  *              Alan Cox        :       Loosened bind a little.
  56  *              Mike McLagan    :       ADD/DEL DLCI Ioctls
  57  *      Willy Konynenberg       :       Transparent proxying support.
  58  *              David S. Miller :       New socket lookup architecture.
  59  *                                      Some other random speedups.
  60  *              Cyrus Durgin    :       Cleaned up file for kmod hacks.
  61  *              Andi Kleen      :       Fix inet_stream_connect TCP race.
  62  */
  63
  64 #define pr_fmt(fmt) "IPv4: " fmt
  65
  66 #include <linux/err.h>
  67 #include <linux/errno.h>
  68 #include <linux/types.h>
  69 #include <linux/socket.h>
  70 #include <linux/in.h>
  71 #include <linux/kernel.h>
  72 #include <linux/kmod.h>
  73 #include <linux/sched.h>
  74 #include <linux/timer.h>
  75 #include <linux/string.h>
  76 #include <linux/sockios.h>
  77 #include <linux/net.h>
  78 #include <linux/capability.h>
  79 #include <linux/fcntl.h>
  80 #include <linux/mm.h>
  81 #include <linux/interrupt.h>
  82 #include <linux/stat.h>
  83 #include <linux/init.h>
  84 #include <linux/poll.h>
  85 #include <linux/netfilter_ipv4.h>
  86 #include <linux/random.h>
  87 #include <linux/slab.h>
  88
  89 #include <linux/uaccess.h>
  90
  91 #include <linux/inet.h>
  92 #include <linux/igmp.h>
  93 #include <linux/inetdevice.h>
  94 #include <linux/netdevice.h>
  95 #include <net/checksum.h>
  96 #include <net/ip.h>
  97 #include <net/protocol.h>
  98 #include <net/arp.h>
  99 #include <net/route.h>
 100 #include <net/ip_fib.h>
 101 #include <net/inet_connection_sock.h>
 102 #include <net/tcp.h>
 103 #include <net/udp.h>
 104 #include <net/udplite.h>
 105 #include <net/ping.h>
 106 #include <linux/skbuff.h>
 107 #include <net/sock.h>
 108 #include <net/raw.h>
 109 #include <net/icmp.h>
 110 #include <net/inet_common.h>
 111 #include <net/ip_tunnels.h>
 112 #include <net/xfrm.h>
 113 #include <net/net_namespace.h>
 114 #include <net/secure_seq.h>
 115 #ifdef CONFIG_IP_MROUTE
 116 #include <linux/mroute.h>
 117 #endif
 118 #include <net/l3mdev.h>
 119 #include <net/compat.h>
 120
 121 #include <trace/events/sock.h>
 122
 123 /* The inetsw table contains everything that inet_create needs to
 124  * build a new socket.
 125  */
 126 static struct list_head inetsw[SOCK_MAX];
 127 static DEFINE_SPINLOCK(inetsw_lock);
 128
 129 /* New destruction routine */
 130
 131 void inet_sock_destruct(struct sock *sk)
 132 {
 133         struct inet_sock *inet = inet_sk(sk);
 134
 135         __skb_queue_purge(&sk->sk_receive_queue);
 136         __skb_queue_purge(&sk->sk_error_queue);
 137
 138         sk_mem_reclaim_final(sk);
 139
 140         if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
 141                 pr_err("Attempt to release TCP socket in state %d %p\n",
 142                        sk->sk_state, sk);
 143                 return;
 144         }
 145         if (!sock_flag(sk, SOCK_DEAD)) {
 146                 pr_err("Attempt to release alive inet socket %p\n", sk);
 147                 return;
 148         }
 149
 150         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
 151         WARN_ON(refcount_read(&sk->sk_wmem_alloc));
 152         WARN_ON(sk->sk_wmem_queued);
 153         WARN_ON(sk_forward_alloc_get(sk));
 154
 155         kfree(rcu_dereference_protected(inet->inet_opt, 1));
 156         dst_release(rcu_dereference_protected(sk->sk_dst_cache, 1));
 157         dst_release(sk->sk_rx_dst);
 158         sk_refcnt_debug_dec(sk);
 159 }
 160 EXPORT_SYMBOL(inet_sock_destruct);
 161
 162 /*
 163  *      The routines beyond this point handle the behaviour of an AF_INET
 164  *      socket object. Mostly it punts to the subprotocols of IP to do
 165  *      the work.
 166  */
 167
 168 /*
 169  *      Automatically bind an unbound socket.
 170  */
 171
 172 static int inet_autobind(struct sock *sk)
 173 {
 174         struct inet_sock *inet;
 175         /* We may need to bind the socket. */
 176         lock_sock(sk);
 177         inet = inet_sk(sk);
 178         if (!inet->inet_num) {
 179                 if (sk->sk_prot->get_port(sk, 0)) {
 180                         release_sock(sk);
 181                         return -EAGAIN;
 182                 }
 183                 inet->inet_sport = htons(inet->inet_num);
 184         }
 185         release_sock(sk);
 186         return 0;
 187 }
 188
 189 /*
 190  *      Move a socket into listening state.
 191  */
 192 int inet_listen(struct socket *sock, int backlog)
 193 {
 194         struct sock *sk = sock->sk;
 195         unsigned char old_state;
 196         int err, tcp_fastopen;
 197
 198         lock_sock(sk);
 199
 200         err = -EINVAL;
 201         if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
 202                 goto out;
 203
 204         old_state = sk->sk_state;
 205         if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
 206                 goto out;
 207
 208         WRITE_ONCE(sk->sk_max_ack_backlog, backlog);
 209         /* Really, if the socket is already in listen state
 210          * we can only allow the backlog to be adjusted.
 211          */
 212         if (old_state != TCP_LISTEN) {
 213                 /* Enable TFO w/o requiring TCP_FASTOPEN socket option.
 214                  * Note that only TCP sockets (SOCK_STREAM) will reach here.
 215                  * Also fastopen backlog may already been set via the option
 216                  * because the socket was in TCP_LISTEN state previously but
 217                  * was shutdown() rather than close().
 218                  */
 219                 tcp_fastopen = sock_net(sk)->ipv4.sysctl_tcp_fastopen;
 220                 if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
 221                     (tcp_fastopen & TFO_SERVER_ENABLE) &&
 222                     !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
 223                         fastopen_queue_tune(sk, backlog);
 224                         tcp_fastopen_init_key_once(sock_net(sk));
 225                 }
 226
 227                 err = inet_csk_listen_start(sk, backlog);
 228                 if (err)
 229                         goto out;
 230                 tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL);
 231         }
 232         err = 0;
 233
 234 out:
 235         release_sock(sk);
 236         return err;
 237 }
 238 EXPORT_SYMBOL(inet_listen);
 239
 240 /*
 241  *      Create an inet socket.
 242  */
 243
 244 static int inet_create(struct net *net, struct socket *sock, int protocol,
 245                        int kern)
 246 {
 247         struct sock *sk;
 248         struct inet_protosw *answer;
 249         struct inet_sock *inet;
 250         struct proto *answer_prot;
 251         unsigned char answer_flags;
 252         int try_loading_module = 0;
 253         int err;
 254
 255         if (protocol < 0 || protocol >= IPPROTO_MAX)
 256                 return -EINVAL;
 257
 258         sock->state = SS_UNCONNECTED;
 259
 260         /* Look for the requested type/protocol pair. */
 261 lookup_protocol:
 262         err = -ESOCKTNOSUPPORT;
 263         rcu_read_lock();
 264         list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
 265
 266                 err = 0;
 267                 /* Check the non-wild match. */
 268                 if (protocol == answer->protocol) {
 269                         if (protocol != IPPROTO_IP)
 270                                 break;
 271                 } else {
 272                         /* Check for the two wild cases. */
 273                         if (IPPROTO_IP == protocol) {
 274                                 protocol = answer->protocol;
 275                                 break;
 276                         }
 277                         if (IPPROTO_IP == answer->protocol)
 278                                 break;
 279                 }
 280                 err = -EPROTONOSUPPORT;
 281         }
 282
 283         if (unlikely(err)) {
 284                 if (try_loading_module < 2) {
 285                         rcu_read_unlock();
 286                         /*
 287                          * Be more specific, e.g. net-pf-2-proto-132-type-1
 288                          * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
 289                          */
 290                         if (++try_loading_module == 1)
 291                                 request_module("net-pf-%d-proto-%d-type-%d",
 292                                                PF_INET, protocol, sock->type);
 293                         /*
 294                          * Fall back to generic, e.g. net-pf-2-proto-132
 295                          * (net-pf-PF_INET-proto-IPPROTO_SCTP)
 296                          */
 297                         else
 298                                 request_module("net-pf-%d-proto-%d",
 299                                                PF_INET, protocol);
 300                         goto lookup_protocol;
 301                 } else
 302                         goto out_rcu_unlock;
 303         }
 304
 305         err = -EPERM;
 306         if (sock->type == SOCK_RAW && !kern &&
 307             !ns_capable(net->user_ns, CAP_NET_RAW))
 308                 goto out_rcu_unlock;
 309
 310         sock->ops = answer->ops;
 311         answer_prot = answer->prot;
 312         answer_flags = answer->flags;
 313         rcu_read_unlock();
 314
 315         WARN_ON(!answer_prot->slab);
 316
 317         err = -ENOMEM;
 318         sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
 319         if (!sk)
 320                 goto out;
 321
 322         err = 0;
 323         if (INET_PROTOSW_REUSE & answer_flags)
 324                 sk->sk_reuse = SK_CAN_REUSE;
 325
 326         inet = inet_sk(sk);
 327         inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
 328
 329         inet->nodefrag = 0;
 330
 331         if (SOCK_RAW == sock->type) {
 332                 inet->inet_num = protocol;
 333                 if (IPPROTO_RAW == protocol)
 334                         inet->hdrincl = 1;
 335         }
 336
 337         if (net->ipv4.sysctl_ip_no_pmtu_disc)
 338                 inet->pmtudisc = IP_PMTUDISC_DONT;
 339         else
 340                 inet->pmtudisc = IP_PMTUDISC_WANT;
 341
 342         inet->inet_id = 0;
 343
 344         sock_init_data(sock, sk);
 345
 346         sk->sk_destruct    = inet_sock_destruct;
 347         sk->sk_protocol    = protocol;
 348         sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
 349
 350         inet->uc_ttl    = -1;
 351         inet->mc_loop   = 1;
 352         inet->mc_ttl    = 1;
 353         inet->mc_all    = 1;
 354         inet->mc_index  = 0;
 355         inet->mc_list   = NULL;
 356         inet->rcv_tos   = 0;
 357
 358         sk_refcnt_debug_inc(sk);
 359
 360         if (inet->inet_num) {
 361                 /* It assumes that any protocol which allows
 362                  * the user to assign a number at socket
 363                  * creation time automatically
 364                  * shares.
 365                  */
 366                 inet->inet_sport = htons(inet->inet_num);
 367                 /* Add to protocol hash chains. */
 368                 err = sk->sk_prot->hash(sk);
 369                 if (err) {
 370                         sk_common_release(sk);
 371                         goto out;
 372                 }
 373         }
 374
 375         if (sk->sk_prot->init) {
 376                 err = sk->sk_prot->init(sk);
 377                 if (err) {
 378                         sk_common_release(sk);
 379                         goto out;
 380                 }
 381         }
 382
 383         if (!kern) {
 384                 err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
 385                 if (err) {
 386                         sk_common_release(sk);
 387                         goto out;
 388                 }
 389         }
 390 out:
 391         return err;
 392 out_rcu_unlock:
 393         rcu_read_unlock();
 394         goto out;
 395 }
 396
 397
 398 /*
 399  *      The peer socket should always be NULL (or else). When we call this
 400  *      function we are destroying the object and from then on nobody
 401  *      should refer to it.
 402  */
 403 int inet_release(struct socket *sock)
 404 {
 405         struct sock *sk = sock->sk;
 406
 407         if (sk) {
 408                 long timeout;
 409
 410                 if (!sk->sk_kern_sock)
 411                         BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk);
 412
 413                 /* Applications forget to leave groups before exiting */
 414                 ip_mc_drop_socket(sk);
 415
 416                 /* If linger is set, we don't return until the close
 417                  * is complete.  Otherwise we return immediately. The
 418                  * actually closing is done the same either way.
 419                  *
 420                  * If the close is due to the process exiting, we never
 421                  * linger..
 422                  */
 423                 timeout = 0;
 424                 if (sock_flag(sk, SOCK_LINGER) &&
 425                     !(current->flags & PF_EXITING))
 426                         timeout = sk->sk_lingertime;
 427                 sk->sk_prot->close(sk, timeout);
 428                 sock->sk = NULL;
 429         }
 430         return 0;
 431 }
 432 EXPORT_SYMBOL(inet_release);
 433
 434 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 435 {
 436         struct sock *sk = sock->sk;
 437         u32 flags = BIND_WITH_LOCK;
 438         int err;
 439
 440         /* If the socket has its own bind function then use it. (RAW) */
 441         if (sk->sk_prot->bind) {
 442                 return sk->sk_prot->bind(sk, uaddr, addr_len);
 443         }
 444         if (addr_len < sizeof(struct sockaddr_in))
 445                 return -EINVAL;
 446
 447         /* BPF prog is run before any checks are done so that if the prog
 448          * changes context in a wrong way it will be caught.
 449          */
 450         err = BPF_CGROUP_RUN_PROG_INET_BIND_LOCK(sk, uaddr,
 451                                                  CGROUP_INET4_BIND, &flags);
 452         if (err)
 453                 return err;
 454
 455         return __inet_bind(sk, uaddr, addr_len, flags);
 456 }
 457 EXPORT_SYMBOL(inet_bind);
 458
 459 int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
 460                 u32 flags)
 461 {
 462         struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
 463         struct inet_sock *inet = inet_sk(sk);
 464         struct net *net = sock_net(sk);
 465         unsigned short snum;
 466         int chk_addr_ret;
 467         u32 tb_id = RT_TABLE_LOCAL;
 468         int err;
 469
 470         if (addr->sin_family != AF_INET) {
 471                 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
 472                  * only if s_addr is INADDR_ANY.
 473                  */
 474                 err = -EAFNOSUPPORT;
 475                 if (addr->sin_family != AF_UNSPEC ||
 476                     addr->sin_addr.s_addr != htonl(INADDR_ANY))
 477                         goto out;
 478         }
 479
 480         tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
 481         chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
 482
 483         /* Not specified by any standard per-se, however it breaks too
 484          * many applications when removed.  It is unfortunate since
 485          * allowing applications to make a non-local bind solves
 486          * several problems with systems using dynamic addressing.
 487          * (ie. your servers still start up even if your ISDN link
 488          *  is temporarily down)
 489          */
 490         err = -EADDRNOTAVAIL;
 491         if (!inet_can_nonlocal_bind(net, inet) &&
 492             addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
 493             chk_addr_ret != RTN_LOCAL &&
 494             chk_addr_ret != RTN_MULTICAST &&
 495             chk_addr_ret != RTN_BROADCAST)
 496                 goto out;
 497
 498         snum = ntohs(addr->sin_port);
 499         err = -EACCES;
 500         if (!(flags & BIND_NO_CAP_NET_BIND_SERVICE) &&
 501             snum && inet_port_requires_bind_service(net, snum) &&
 502             !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
 503                 goto out;
 504
 505         /*      We keep a pair of addresses. rcv_saddr is the one
 506          *      used by hash lookups, and saddr is used for transmit.
 507          *
 508          *      In the BSD API these are the same except where it
 509          *      would be illegal to use them (multicast/broadcast) in
 510          *      which case the sending device address is used.
 511          */
 512         if (flags & BIND_WITH_LOCK)
 513                 lock_sock(sk);
 514
 515         /* Check these errors (active socket, double bind). */
 516         err = -EINVAL;
 517         if (sk->sk_state != TCP_CLOSE || inet->inet_num)
 518                 goto out_release_sock;
 519
 520         inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
 521         if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
 522                 inet->inet_saddr = 0;  /* Use device */
 523
 524         /* Make sure we are allowed to bind here. */
 525         if (snum || !(inet->bind_address_no_port ||
 526                       (flags & BIND_FORCE_ADDRESS_NO_PORT))) {
 527                 if (sk->sk_prot->get_port(sk, snum)) {
 528                         inet->inet_saddr = inet->inet_rcv_saddr = 0;
 529                         err = -EADDRINUSE;
 530                         goto out_release_sock;
 531                 }
 532                 if (!(flags & BIND_FROM_BPF)) {
 533                         err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
 534                         if (err) {
 535                                 inet->inet_saddr = inet->inet_rcv_saddr = 0;
 536                                 goto out_release_sock;
 537                         }
 538                 }
 539         }
 540
 541         if (inet->inet_rcv_saddr)
 542                 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
 543         if (snum)
 544                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
 545         inet->inet_sport = htons(inet->inet_num);
 546         inet->inet_daddr = 0;
 547         inet->inet_dport = 0;
 548         sk_dst_reset(sk);
 549         err = 0;
 550 out_release_sock:
 551         if (flags & BIND_WITH_LOCK)
 552                 release_sock(sk);
 553 out:
 554         return err;
 555 }
 556
 557 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
 558                        int addr_len, int flags)
 559 {
 560         struct sock *sk = sock->sk;
 561         int err;
 562
 563         if (addr_len < sizeof(uaddr->sa_family))
 564                 return -EINVAL;
 565         if (uaddr->sa_family == AF_UNSPEC)
 566                 return sk->sk_prot->disconnect(sk, flags);
 567
 568         if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
 569                 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
 570                 if (err)
 571                         return err;
 572         }
 573
 574         if (data_race(!inet_sk(sk)->inet_num) && inet_autobind(sk))
 575                 return -EAGAIN;
 576         return sk->sk_prot->connect(sk, uaddr, addr_len);
 577 }
 578 EXPORT_SYMBOL(inet_dgram_connect);
 579
 580 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
 581 {
 582         DEFINE_WAIT_FUNC(wait, woken_wake_function);
 583
 584         add_wait_queue(sk_sleep(sk), &wait);
 585         sk->sk_write_pending += writebias;
 586
 587         /* Basic assumption: if someone sets sk->sk_err, he _must_
 588          * change state of the socket from TCP_SYN_*.
 589          * Connect() does not allow to get error notifications
 590          * without closing the socket.
 591          */
 592         while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 593                 release_sock(sk);
 594                 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
 595                 lock_sock(sk);
 596                 if (signal_pending(current) || !timeo)
 597                         break;
 598         }
 599         remove_wait_queue(sk_sleep(sk), &wait);
 600         sk->sk_write_pending -= writebias;
 601         return timeo;
 602 }
 603
 604 /*
 605  *      Connect to a remote host. There is regrettably still a little
 606  *      TCP 'magic' in here.
 607  */
 608 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 609                           int addr_len, int flags, int is_sendmsg)
 610 {
 611         struct sock *sk = sock->sk;
 612         int err;
 613         long timeo;
 614
 615         /*
 616          * uaddr can be NULL and addr_len can be 0 if:
 617          * sk is a TCP fastopen active socket and
 618          * TCP_FASTOPEN_CONNECT sockopt is set and
 619          * we already have a valid cookie for this socket.
 620          * In this case, user can call write() after connect().
 621          * write() will invoke tcp_sendmsg_fastopen() which calls
 622          * __inet_stream_connect().
 623          */
 624         if (uaddr) {
 625                 if (addr_len < sizeof(uaddr->sa_family))
 626                         return -EINVAL;
 627
 628                 if (uaddr->sa_family == AF_UNSPEC) {
 629                         err = sk->sk_prot->disconnect(sk, flags);
 630                         sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 631                         goto out;
 632                 }
 633         }
 634
 635         switch (sock->state) {
 636         default:
 637                 err = -EINVAL;
 638                 goto out;
 639         case SS_CONNECTED:
 640                 err = -EISCONN;
 641                 goto out;
 642         case SS_CONNECTING:
 643                 if (inet_sk(sk)->defer_connect)
 644                         err = is_sendmsg ? -EINPROGRESS : -EISCONN;
 645                 else
 646                         err = -EALREADY;
 647                 /* Fall out of switch with err, set for this state */
 648                 break;
 649         case SS_UNCONNECTED:
 650                 err = -EISCONN;
 651                 if (sk->sk_state != TCP_CLOSE)
 652                         goto out;
 653
 654                 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
 655                         err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
 656                         if (err)
 657                                 goto out;
 658                 }
 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                 if (!err && inet_sk(sk)->defer_connect)
 667                         goto out;
 668
 669                 /* Just entered SS_CONNECTING state; the only
 670                  * difference is that return value in non-blocking
 671                  * case is EINPROGRESS, rather than EALREADY.
 672                  */
 673                 err = -EINPROGRESS;
 674                 break;
 675         }
 676
 677         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
 678
 679         if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 680                 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
 681                                 tcp_sk(sk)->fastopen_req &&
 682                                 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
 683
 684                 /* Error code is set above */
 685                 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
 686                         goto out;
 687
 688                 err = sock_intr_errno(timeo);
 689                 if (signal_pending(current))
 690                         goto out;
 691         }
 692
 693         /* Connection was closed by RST, timeout, ICMP error
 694          * or another process disconnected us.
 695          */
 696         if (sk->sk_state == TCP_CLOSE)
 697                 goto sock_error;
 698
 699         /* sk->sk_err may be not zero now, if RECVERR was ordered by user
 700          * and error was received after socket entered established state.
 701          * Hence, it is handled normally after connect() return successfully.
 702          */
 703
 704         sock->state = SS_CONNECTED;
 705         err = 0;
 706 out:
 707         return err;
 708
 709 sock_error:
 710         err = sock_error(sk) ? : -ECONNABORTED;
 711         sock->state = SS_UNCONNECTED;
 712         if (sk->sk_prot->disconnect(sk, flags))
 713                 sock->state = SS_DISCONNECTING;
 714         goto out;
 715 }
 716 EXPORT_SYMBOL(__inet_stream_connect);
 717
 718 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 719                         int addr_len, int flags)
 720 {
 721         int err;
 722
 723         lock_sock(sock->sk);
 724         err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
 725         release_sock(sock->sk);
 726         return err;
 727 }
 728 EXPORT_SYMBOL(inet_stream_connect);
 729
 730 /*
 731  *      Accept a pending connection. The TCP layer now gives BSD semantics.
 732  */
 733
 734 int inet_accept(struct socket *sock, struct socket *newsock, int flags,
 735                 bool kern)
 736 {
 737         struct sock *sk1 = sock->sk;
 738         int err = -EINVAL;
 739         struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err, kern);
 740
 741         if (!sk2)
 742                 goto do_err;
 743
 744         lock_sock(sk2);
 745
 746         sock_rps_record_flow(sk2);
 747         WARN_ON(!((1 << sk2->sk_state) &
 748                   (TCPF_ESTABLISHED | TCPF_SYN_RECV |
 749                   TCPF_CLOSE_WAIT | TCPF_CLOSE)));
 750
 751         sock_graft(sk2, newsock);
 752
 753         newsock->state = SS_CONNECTED;
 754         err = 0;
 755         release_sock(sk2);
 756 do_err:
 757         return err;
 758 }
 759 EXPORT_SYMBOL(inet_accept);
 760
 761 /*
 762  *      This does both peername and sockname.
 763  */
 764 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
 765                  int peer)
 766 {
 767         struct sock *sk         = sock->sk;
 768         struct inet_sock *inet  = inet_sk(sk);
 769         DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
 770
 771         sin->sin_family = AF_INET;
 772         lock_sock(sk);
 773         if (peer) {
 774                 if (!inet->inet_dport ||
 775                     (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
 776                      peer == 1)) {
 777                         release_sock(sk);
 778                         return -ENOTCONN;
 779                 }
 780                 sin->sin_port = inet->inet_dport;
 781                 sin->sin_addr.s_addr = inet->inet_daddr;
 782                 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin,
 783                                        CGROUP_INET4_GETPEERNAME);
 784         } else {
 785                 __be32 addr = inet->inet_rcv_saddr;
 786                 if (!addr)
 787                         addr = inet->inet_saddr;
 788                 sin->sin_port = inet->inet_sport;
 789                 sin->sin_addr.s_addr = addr;
 790                 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin,
 791                                        CGROUP_INET4_GETSOCKNAME);
 792         }
 793         release_sock(sk);
 794         memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 795         return sizeof(*sin);
 796 }
 797 EXPORT_SYMBOL(inet_getname);
 798
 799 int inet_send_prepare(struct sock *sk)
 800 {
 801         sock_rps_record_flow(sk);
 802
 803         /* We may need to bind the socket. */
 804         if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind &&
 805             inet_autobind(sk))
 806                 return -EAGAIN;
 807
 808         return 0;
 809 }
 810 EXPORT_SYMBOL_GPL(inet_send_prepare);
 811
 812 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
 813 {
 814         struct sock *sk = sock->sk;
 815
 816         if (unlikely(inet_send_prepare(sk)))
 817                 return -EAGAIN;
 818
 819         return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
 820                                sk, msg, size);
 821 }
 822 EXPORT_SYMBOL(inet_sendmsg);
 823
 824 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
 825                       size_t size, int flags)
 826 {
 827         struct sock *sk = sock->sk;
 828
 829         if (unlikely(inet_send_prepare(sk)))
 830                 return -EAGAIN;
 831
 832         if (sk->sk_prot->sendpage)
 833                 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
 834         return sock_no_sendpage(sock, page, offset, size, flags);
 835 }
 836 EXPORT_SYMBOL(inet_sendpage);
 837
 838 INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
 839                                           size_t, int, int, int *));
 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 = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
 851                               sk, msg, size, flags & MSG_DONTWAIT,
 852                               flags & ~MSG_DONTWAIT, &addr_len);
 853         if (err >= 0)
 854                 msg->msg_namelen = addr_len;
 855         return err;
 856 }
 857 EXPORT_SYMBOL(inet_recvmsg);
 858
 859 int inet_shutdown(struct socket *sock, int how)
 860 {
 861         struct sock *sk = sock->sk;
 862         int err = 0;
 863
 864         /* This should really check to make sure
 865          * the socket is a TCP socket. (WHY AC...)
 866          */
 867         how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
 868                        1->2 bit 2 snds.
 869                        2->3 */
 870         if ((how & ~SHUTDOWN_MASK) || !how)     /* MAXINT->0 */
 871                 return -EINVAL;
 872
 873         lock_sock(sk);
 874         if (sock->state == SS_CONNECTING) {
 875                 if ((1 << sk->sk_state) &
 876                     (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
 877                         sock->state = SS_DISCONNECTING;
 878                 else
 879                         sock->state = SS_CONNECTED;
 880         }
 881
 882         switch (sk->sk_state) {
 883         case TCP_CLOSE:
 884                 err = -ENOTCONN;
 885                 /* Hack to wake up other listeners, who can poll for
 886                    EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
 887                 fallthrough;
 888         default:
 889                 sk->sk_shutdown |= how;
 890                 if (sk->sk_prot->shutdown)
 891                         sk->sk_prot->shutdown(sk, how);
 892                 break;
 893
 894         /* Remaining two branches are temporary solution for missing
 895          * close() in multithreaded environment. It is _not_ a good idea,
 896          * but we have no choice until close() is repaired at VFS level.
 897          */
 898         case TCP_LISTEN:
 899                 if (!(how & RCV_SHUTDOWN))
 900                         break;
 901                 fallthrough;
 902         case TCP_SYN_SENT:
 903                 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
 904                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 905                 break;
 906         }
 907
 908         /* Wake up anyone sleeping in poll. */
 909         sk->sk_state_change(sk);
 910         release_sock(sk);
 911         return err;
 912 }
 913 EXPORT_SYMBOL(inet_shutdown);
 914
 915 /*
 916  *      ioctl() calls you can issue on an INET socket. Most of these are
 917  *      device configuration and stuff and very rarely used. Some ioctls
 918  *      pass on to the socket itself.
 919  *
 920  *      NOTE: I like the idea of a module for the config stuff. ie ifconfig
 921  *      loads the devconfigure module does its configuring and unloads it.
 922  *      There's a good 20K of config code hanging around the kernel.
 923  */
 924
 925 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 926 {
 927         struct sock *sk = sock->sk;
 928         int err = 0;
 929         struct net *net = sock_net(sk);
 930         void __user *p = (void __user *)arg;
 931         struct ifreq ifr;
 932         struct rtentry rt;
 933
 934         switch (cmd) {
 935         case SIOCADDRT:
 936         case SIOCDELRT:
 937                 if (copy_from_user(&rt, p, sizeof(struct rtentry)))
 938                         return -EFAULT;
 939                 err = ip_rt_ioctl(net, cmd, &rt);
 940                 break;
 941         case SIOCRTMSG:
 942                 err = -EINVAL;
 943                 break;
 944         case SIOCDARP:
 945         case SIOCGARP:
 946         case SIOCSARP:
 947                 err = arp_ioctl(net, cmd, (void __user *)arg);
 948                 break;
 949         case SIOCGIFADDR:
 950         case SIOCGIFBRDADDR:
 951         case SIOCGIFNETMASK:
 952         case SIOCGIFDSTADDR:
 953         case SIOCGIFPFLAGS:
 954                 if (get_user_ifreq(&ifr, NULL, p))
 955                         return -EFAULT;
 956                 err = devinet_ioctl(net, cmd, &ifr);
 957                 if (!err && put_user_ifreq(&ifr, p))
 958                         err = -EFAULT;
 959                 break;
 960
 961         case SIOCSIFADDR:
 962         case SIOCSIFBRDADDR:
 963         case SIOCSIFNETMASK:
 964         case SIOCSIFDSTADDR:
 965         case SIOCSIFPFLAGS:
 966         case SIOCSIFFLAGS:
 967                 if (get_user_ifreq(&ifr, NULL, p))
 968                         return -EFAULT;
 969                 err = devinet_ioctl(net, cmd, &ifr);
 970                 break;
 971         default:
 972                 if (sk->sk_prot->ioctl)
 973                         err = sk->sk_prot->ioctl(sk, cmd, arg);
 974                 else
 975                         err = -ENOIOCTLCMD;
 976                 break;
 977         }
 978         return err;
 979 }
 980 EXPORT_SYMBOL(inet_ioctl);
 981
 982 #ifdef CONFIG_COMPAT
 983 static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd,
 984                 struct compat_rtentry __user *ur)
 985 {
 986         compat_uptr_t rtdev;
 987         struct rtentry rt;
 988
 989         if (copy_from_user(&rt.rt_dst, &ur->rt_dst,
 990                         3 * sizeof(struct sockaddr)) ||
 991             get_user(rt.rt_flags, &ur->rt_flags) ||
 992             get_user(rt.rt_metric, &ur->rt_metric) ||
 993             get_user(rt.rt_mtu, &ur->rt_mtu) ||
 994             get_user(rt.rt_window, &ur->rt_window) ||
 995             get_user(rt.rt_irtt, &ur->rt_irtt) ||
 996             get_user(rtdev, &ur->rt_dev))
 997                 return -EFAULT;
 998
 999         rt.rt_dev = compat_ptr(rtdev);
1000         return ip_rt_ioctl(sock_net(sk), cmd, &rt);
1001 }
1002
1003 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1004 {
1005         void __user *argp = compat_ptr(arg);
1006         struct sock *sk = sock->sk;
1007
1008         switch (cmd) {
1009         case SIOCADDRT:
1010         case SIOCDELRT:
1011                 return inet_compat_routing_ioctl(sk, cmd, argp);
1012         default:
1013                 if (!sk->sk_prot->compat_ioctl)
1014                         return -ENOIOCTLCMD;
1015                 return sk->sk_prot->compat_ioctl(sk, cmd, arg);
1016         }
1017 }
1018 #endif /* CONFIG_COMPAT */
1019
1020 const struct proto_ops inet_stream_ops = {
1021         .family            = PF_INET,
1022         .owner             = THIS_MODULE,
1023         .release           = inet_release,
1024         .bind              = inet_bind,
1025         .connect           = inet_stream_connect,
1026         .socketpair        = sock_no_socketpair,
1027         .accept            = inet_accept,
1028         .getname           = inet_getname,
1029         .poll              = tcp_poll,
1030         .ioctl             = inet_ioctl,
1031         .gettstamp         = sock_gettstamp,
1032         .listen            = inet_listen,
1033         .shutdown          = inet_shutdown,
1034         .setsockopt        = sock_common_setsockopt,
1035         .getsockopt        = sock_common_getsockopt,
1036         .sendmsg           = inet_sendmsg,
1037         .recvmsg           = inet_recvmsg,
1038 #ifdef CONFIG_MMU
1039         .mmap              = tcp_mmap,
1040 #endif
1041         .sendpage          = inet_sendpage,
1042         .splice_read       = tcp_splice_read,
1043         .read_sock         = tcp_read_sock,
1044         .sendmsg_locked    = tcp_sendmsg_locked,
1045         .sendpage_locked   = tcp_sendpage_locked,
1046         .peek_len          = tcp_peek_len,
1047 #ifdef CONFIG_COMPAT
1048         .compat_ioctl      = inet_compat_ioctl,
1049 #endif
1050         .set_rcvlowat      = tcp_set_rcvlowat,
1051 };
1052 EXPORT_SYMBOL(inet_stream_ops);
1053
1054 const struct proto_ops inet_dgram_ops = {
1055         .family            = PF_INET,
1056         .owner             = THIS_MODULE,
1057         .release           = inet_release,
1058         .bind              = inet_bind,
1059         .connect           = inet_dgram_connect,
1060         .socketpair        = sock_no_socketpair,
1061         .accept            = sock_no_accept,
1062         .getname           = inet_getname,
1063         .poll              = udp_poll,
1064         .ioctl             = inet_ioctl,
1065         .gettstamp         = sock_gettstamp,
1066         .listen            = sock_no_listen,
1067         .shutdown          = inet_shutdown,
1068         .setsockopt        = sock_common_setsockopt,
1069         .getsockopt        = sock_common_getsockopt,
1070         .sendmsg           = inet_sendmsg,
1071         .read_sock         = udp_read_sock,
1072         .recvmsg           = inet_recvmsg,
1073         .mmap              = sock_no_mmap,
1074         .sendpage          = inet_sendpage,
1075         .set_peek_off      = sk_set_peek_off,
1076 #ifdef CONFIG_COMPAT
1077         .compat_ioctl      = inet_compat_ioctl,
1078 #endif
1079 };
1080 EXPORT_SYMBOL(inet_dgram_ops);
1081
1082 /*
1083  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1084  * udp_poll
1085  */
1086 static const struct proto_ops inet_sockraw_ops = {
1087         .family            = PF_INET,
1088         .owner             = THIS_MODULE,
1089         .release           = inet_release,
1090         .bind              = inet_bind,
1091         .connect           = inet_dgram_connect,
1092         .socketpair        = sock_no_socketpair,
1093         .accept            = sock_no_accept,
1094         .getname           = inet_getname,
1095         .poll              = datagram_poll,
1096         .ioctl             = inet_ioctl,
1097         .gettstamp         = sock_gettstamp,
1098         .listen            = sock_no_listen,
1099         .shutdown          = inet_shutdown,
1100         .setsockopt        = sock_common_setsockopt,
1101         .getsockopt        = sock_common_getsockopt,
1102         .sendmsg           = inet_sendmsg,
1103         .recvmsg           = inet_recvmsg,
1104         .mmap              = sock_no_mmap,
1105         .sendpage          = inet_sendpage,
1106 #ifdef CONFIG_COMPAT
1107         .compat_ioctl      = inet_compat_ioctl,
1108 #endif
1109 };
1110
1111 static const struct net_proto_family inet_family_ops = {
1112         .family = PF_INET,
1113         .create = inet_create,
1114         .owner  = THIS_MODULE,
1115 };
1116
1117 /* Upon startup we insert all the elements in inetsw_array[] into
1118  * the linked list inetsw.
1119  */
1120 static struct inet_protosw inetsw_array[] =
1121 {
1122         {
1123                 .type =       SOCK_STREAM,
1124                 .protocol =   IPPROTO_TCP,
1125                 .prot =       &tcp_prot,
1126                 .ops =        &inet_stream_ops,
1127                 .flags =      INET_PROTOSW_PERMANENT |
1128                               INET_PROTOSW_ICSK,
1129         },
1130
1131         {
1132                 .type =       SOCK_DGRAM,
1133                 .protocol =   IPPROTO_UDP,
1134                 .prot =       &udp_prot,
1135                 .ops =        &inet_dgram_ops,
1136                 .flags =      INET_PROTOSW_PERMANENT,
1137        },
1138
1139        {
1140                 .type =       SOCK_DGRAM,
1141                 .protocol =   IPPROTO_ICMP,
1142                 .prot =       &ping_prot,
1143                 .ops =        &inet_sockraw_ops,
1144                 .flags =      INET_PROTOSW_REUSE,
1145        },
1146
1147        {
1148                .type =       SOCK_RAW,
1149                .protocol =   IPPROTO_IP,        /* wild card */
1150                .prot =       &raw_prot,
1151                .ops =        &inet_sockraw_ops,
1152                .flags =      INET_PROTOSW_REUSE,
1153        }
1154 };
1155
1156 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1157
1158 void inet_register_protosw(struct inet_protosw *p)
1159 {
1160         struct list_head *lh;
1161         struct inet_protosw *answer;
1162         int protocol = p->protocol;
1163         struct list_head *last_perm;
1164
1165         spin_lock_bh(&inetsw_lock);
1166
1167         if (p->type >= SOCK_MAX)
1168                 goto out_illegal;
1169
1170         /* If we are trying to override a permanent protocol, bail. */
1171         last_perm = &inetsw[p->type];
1172         list_for_each(lh, &inetsw[p->type]) {
1173                 answer = list_entry(lh, struct inet_protosw, list);
1174                 /* Check only the non-wild match. */
1175                 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1176                         break;
1177                 if (protocol == answer->protocol)
1178                         goto out_permanent;
1179                 last_perm = lh;
1180         }
1181
1182         /* Add the new entry after the last permanent entry if any, so that
1183          * the new entry does not override a permanent entry when matched with
1184          * a wild-card protocol. But it is allowed to override any existing
1185          * non-permanent entry.  This means that when we remove this entry, the
1186          * system automatically returns to the old behavior.
1187          */
1188         list_add_rcu(&p->list, last_perm);
1189 out:
1190         spin_unlock_bh(&inetsw_lock);
1191
1192         return;
1193
1194 out_permanent:
1195         pr_err("Attempt to override permanent protocol %d\n", protocol);
1196         goto out;
1197
1198 out_illegal:
1199         pr_err("Ignoring attempt to register invalid socket type %d\n",
1200                p->type);
1201         goto out;
1202 }
1203 EXPORT_SYMBOL(inet_register_protosw);
1204
1205 void inet_unregister_protosw(struct inet_protosw *p)
1206 {
1207         if (INET_PROTOSW_PERMANENT & p->flags) {
1208                 pr_err("Attempt to unregister permanent protocol %d\n",
1209                        p->protocol);
1210         } else {
1211                 spin_lock_bh(&inetsw_lock);
1212                 list_del_rcu(&p->list);
1213                 spin_unlock_bh(&inetsw_lock);
1214
1215                 synchronize_net();
1216         }
1217 }
1218 EXPORT_SYMBOL(inet_unregister_protosw);
1219
1220 static int inet_sk_reselect_saddr(struct sock *sk)
1221 {
1222         struct inet_sock *inet = inet_sk(sk);
1223         __be32 old_saddr = inet->inet_saddr;
1224         __be32 daddr = inet->inet_daddr;
1225         struct flowi4 *fl4;
1226         struct rtable *rt;
1227         __be32 new_saddr;
1228         struct ip_options_rcu *inet_opt;
1229
1230         inet_opt = rcu_dereference_protected(inet->inet_opt,
1231                                              lockdep_sock_is_held(sk));
1232         if (inet_opt && inet_opt->opt.srr)
1233                 daddr = inet_opt->opt.faddr;
1234
1235         /* Query new route. */
1236         fl4 = &inet->cork.fl.u.ip4;
1237         rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1238                               sk->sk_bound_dev_if, sk->sk_protocol,
1239                               inet->inet_sport, inet->inet_dport, sk);
1240         if (IS_ERR(rt))
1241                 return PTR_ERR(rt);
1242
1243         sk_setup_caps(sk, &rt->dst);
1244
1245         new_saddr = fl4->saddr;
1246
1247         if (new_saddr == old_saddr)
1248                 return 0;
1249
1250         if (sock_net(sk)->ipv4.sysctl_ip_dynaddr > 1) {
1251                 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1252                         __func__, &old_saddr, &new_saddr);
1253         }
1254
1255         inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1256
1257         /*
1258          * XXX The only one ugly spot where we need to
1259          * XXX really change the sockets identity after
1260          * XXX it has entered the hashes. -DaveM
1261          *
1262          * Besides that, it does not check for connection
1263          * uniqueness. Wait for troubles.
1264          */
1265         return __sk_prot_rehash(sk);
1266 }
1267
1268 int inet_sk_rebuild_header(struct sock *sk)
1269 {
1270         struct inet_sock *inet = inet_sk(sk);
1271         struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1272         __be32 daddr;
1273         struct ip_options_rcu *inet_opt;
1274         struct flowi4 *fl4;
1275         int err;
1276
1277         /* Route is OK, nothing to do. */
1278         if (rt)
1279                 return 0;
1280
1281         /* Reroute. */
1282         rcu_read_lock();
1283         inet_opt = rcu_dereference(inet->inet_opt);
1284         daddr = inet->inet_daddr;
1285         if (inet_opt && inet_opt->opt.srr)
1286                 daddr = inet_opt->opt.faddr;
1287         rcu_read_unlock();
1288         fl4 = &inet->cork.fl.u.ip4;
1289         rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1290                                    inet->inet_dport, inet->inet_sport,
1291                                    sk->sk_protocol, RT_CONN_FLAGS(sk),
1292                                    sk->sk_bound_dev_if);
1293         if (!IS_ERR(rt)) {
1294                 err = 0;
1295                 sk_setup_caps(sk, &rt->dst);
1296         } else {
1297                 err = PTR_ERR(rt);
1298
1299                 /* Routing failed... */
1300                 sk->sk_route_caps = 0;
1301                 /*
1302                  * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1303                  * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1304                  */
1305                 if (!sock_net(sk)->ipv4.sysctl_ip_dynaddr ||
1306                     sk->sk_state != TCP_SYN_SENT ||
1307                     (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1308                     (err = inet_sk_reselect_saddr(sk)) != 0)
1309                         sk->sk_err_soft = -err;
1310         }
1311
1312         return err;
1313 }
1314 EXPORT_SYMBOL(inet_sk_rebuild_header);
1315
1316 void inet_sk_set_state(struct sock *sk, int state)
1317 {
1318         trace_inet_sock_set_state(sk, sk->sk_state, state);
1319         sk->sk_state = state;
1320 }
1321 EXPORT_SYMBOL(inet_sk_set_state);
1322
1323 void inet_sk_state_store(struct sock *sk, int newstate)
1324 {
1325         trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1326         smp_store_release(&sk->sk_state, newstate);
1327 }
1328
1329 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1330                                  netdev_features_t features)
1331 {
1332         bool udpfrag = false, fixedid = false, gso_partial, encap;
1333         struct sk_buff *segs = ERR_PTR(-EINVAL);
1334         const struct net_offload *ops;
1335         unsigned int offset = 0;
1336         struct iphdr *iph;
1337         int proto, tot_len;
1338         int nhoff;
1339         int ihl;
1340         int id;
1341
1342         skb_reset_network_header(skb);
1343         nhoff = skb_network_header(skb) - skb_mac_header(skb);
1344         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1345                 goto out;
1346
1347         iph = ip_hdr(skb);
1348         ihl = iph->ihl * 4;
1349         if (ihl < sizeof(*iph))
1350                 goto out;
1351
1352         id = ntohs(iph->id);
1353         proto = iph->protocol;
1354
1355         /* Warning: after this point, iph might be no longer valid */
1356         if (unlikely(!pskb_may_pull(skb, ihl)))
1357                 goto out;
1358         __skb_pull(skb, ihl);
1359
1360         encap = SKB_GSO_CB(skb)->encap_level > 0;
1361         if (encap)
1362                 features &= skb->dev->hw_enc_features;
1363         SKB_GSO_CB(skb)->encap_level += ihl;
1364
1365         skb_reset_transport_header(skb);
1366
1367         segs = ERR_PTR(-EPROTONOSUPPORT);
1368
1369         if (!skb->encapsulation || encap) {
1370                 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1371                 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1372
1373                 /* fixed ID is invalid if DF bit is not set */
1374                 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1375                         goto out;
1376         }
1377
1378         ops = rcu_dereference(inet_offloads[proto]);
1379         if (likely(ops && ops->callbacks.gso_segment))
1380                 segs = ops->callbacks.gso_segment(skb, features);
1381
1382         if (IS_ERR_OR_NULL(segs))
1383                 goto out;
1384
1385         gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1386
1387         skb = segs;
1388         do {
1389                 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1390                 if (udpfrag) {
1391                         iph->frag_off = htons(offset >> 3);
1392                         if (skb->next)
1393                                 iph->frag_off |= htons(IP_MF);
1394                         offset += skb->len - nhoff - ihl;
1395                         tot_len = skb->len - nhoff;
1396                 } else if (skb_is_gso(skb)) {
1397                         if (!fixedid) {
1398                                 iph->id = htons(id);
1399                                 id += skb_shinfo(skb)->gso_segs;
1400                         }
1401
1402                         if (gso_partial)
1403                                 tot_len = skb_shinfo(skb)->gso_size +
1404                                           SKB_GSO_CB(skb)->data_offset +
1405                                           skb->head - (unsigned char *)iph;
1406                         else
1407                                 tot_len = skb->len - nhoff;
1408                 } else {
1409                         if (!fixedid)
1410                                 iph->id = htons(id++);
1411                         tot_len = skb->len - nhoff;
1412                 }
1413                 iph->tot_len = htons(tot_len);
1414                 ip_send_check(iph);
1415                 if (encap)
1416                         skb_reset_inner_headers(skb);
1417                 skb->network_header = (u8 *)iph - skb->head;
1418                 skb_reset_mac_len(skb);
1419         } while ((skb = skb->next));
1420
1421 out:
1422         return segs;
1423 }
1424
1425 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1426                                         netdev_features_t features)
1427 {
1428         if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1429                 return ERR_PTR(-EINVAL);
1430
1431         return inet_gso_segment(skb, features);
1432 }
1433
1434 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1435 {
1436         const struct net_offload *ops;
1437         struct sk_buff *pp = NULL;
1438         const struct iphdr *iph;
1439         struct sk_buff *p;
1440         unsigned int hlen;
1441         unsigned int off;
1442         unsigned int id;
1443         int flush = 1;
1444         int proto;
1445
1446         off = skb_gro_offset(skb);
1447         hlen = off + sizeof(*iph);
1448         iph = skb_gro_header_fast(skb, off);
1449         if (skb_gro_header_hard(skb, hlen)) {
1450                 iph = skb_gro_header_slow(skb, hlen, off);
1451                 if (unlikely(!iph))
1452                         goto out;
1453         }
1454
1455         proto = iph->protocol;
1456
1457         rcu_read_lock();
1458         ops = rcu_dereference(inet_offloads[proto]);
1459         if (!ops || !ops->callbacks.gro_receive)
1460                 goto out_unlock;
1461
1462         if (*(u8 *)iph != 0x45)
1463                 goto out_unlock;
1464
1465         if (ip_is_fragment(iph))
1466                 goto out_unlock;
1467
1468         if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1469                 goto out_unlock;
1470
1471         id = ntohl(*(__be32 *)&iph->id);
1472         flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1473         id >>= 16;
1474
1475         list_for_each_entry(p, head, list) {
1476                 struct iphdr *iph2;
1477                 u16 flush_id;
1478
1479                 if (!NAPI_GRO_CB(p)->same_flow)
1480                         continue;
1481
1482                 iph2 = (struct iphdr *)(p->data + off);
1483                 /* The above works because, with the exception of the top
1484                  * (inner most) layer, we only aggregate pkts with the same
1485                  * hdr length so all the hdrs we'll need to verify will start
1486                  * at the same offset.
1487                  */
1488                 if ((iph->protocol ^ iph2->protocol) |
1489                     ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1490                     ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1491                         NAPI_GRO_CB(p)->same_flow = 0;
1492                         continue;
1493                 }
1494
1495                 /* All fields must match except length and checksum. */
1496                 NAPI_GRO_CB(p)->flush |=
1497                         (iph->ttl ^ iph2->ttl) |
1498                         (iph->tos ^ iph2->tos) |
1499                         ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1500
1501                 NAPI_GRO_CB(p)->flush |= flush;
1502
1503                 /* We need to store of the IP ID check to be included later
1504                  * when we can verify that this packet does in fact belong
1505                  * to a given flow.
1506                  */
1507                 flush_id = (u16)(id - ntohs(iph2->id));
1508
1509                 /* This bit of code makes it much easier for us to identify
1510                  * the cases where we are doing atomic vs non-atomic IP ID
1511                  * checks.  Specifically an atomic check can return IP ID
1512                  * values 0 - 0xFFFF, while a non-atomic check can only
1513                  * return 0 or 0xFFFF.
1514                  */
1515                 if (!NAPI_GRO_CB(p)->is_atomic ||
1516                     !(iph->frag_off & htons(IP_DF))) {
1517                         flush_id ^= NAPI_GRO_CB(p)->count;
1518                         flush_id = flush_id ? 0xFFFF : 0;
1519                 }
1520
1521                 /* If the previous IP ID value was based on an atomic
1522                  * datagram we can overwrite the value and ignore it.
1523                  */
1524                 if (NAPI_GRO_CB(skb)->is_atomic)
1525                         NAPI_GRO_CB(p)->flush_id = flush_id;
1526                 else
1527                         NAPI_GRO_CB(p)->flush_id |= flush_id;
1528         }
1529
1530         NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1531         NAPI_GRO_CB(skb)->flush |= flush;
1532         skb_set_network_header(skb, off);
1533         /* The above will be needed by the transport layer if there is one
1534          * immediately following this IP hdr.
1535          */
1536
1537         /* Note : No need to call skb_gro_postpull_rcsum() here,
1538          * as we already checked checksum over ipv4 header was 0
1539          */
1540         skb_gro_pull(skb, sizeof(*iph));
1541         skb_set_transport_header(skb, skb_gro_offset(skb));
1542
1543         pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1544                                        ops->callbacks.gro_receive, head, skb);
1545
1546 out_unlock:
1547         rcu_read_unlock();
1548
1549 out:
1550         skb_gro_flush_final(skb, pp, flush);
1551
1552         return pp;
1553 }
1554
1555 static struct sk_buff *ipip_gro_receive(struct list_head *head,
1556                                         struct sk_buff *skb)
1557 {
1558         if (NAPI_GRO_CB(skb)->encap_mark) {
1559                 NAPI_GRO_CB(skb)->flush = 1;
1560                 return NULL;
1561         }
1562
1563         NAPI_GRO_CB(skb)->encap_mark = 1;
1564
1565         return inet_gro_receive(head, skb);
1566 }
1567
1568 #define SECONDS_PER_DAY 86400
1569
1570 /* inet_current_timestamp - Return IP network timestamp
1571  *
1572  * Return milliseconds since midnight in network byte order.
1573  */
1574 __be32 inet_current_timestamp(void)
1575 {
1576         u32 secs;
1577         u32 msecs;
1578         struct timespec64 ts;
1579
1580         ktime_get_real_ts64(&ts);
1581
1582         /* Get secs since midnight. */
1583         (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1584         /* Convert to msecs. */
1585         msecs = secs * MSEC_PER_SEC;
1586         /* Convert nsec to msec. */
1587         msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1588
1589         /* Convert to network byte order. */
1590         return htonl(msecs);
1591 }
1592 EXPORT_SYMBOL(inet_current_timestamp);
1593
1594 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1595 {
1596         if (sk->sk_family == AF_INET)
1597                 return ip_recv_error(sk, msg, len, addr_len);
1598 #if IS_ENABLED(CONFIG_IPV6)
1599         if (sk->sk_family == AF_INET6)
1600                 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1601 #endif
1602         return -EINVAL;
1603 }
1604
1605 int inet_gro_complete(struct sk_buff *skb, int nhoff)
1606 {
1607         __be16 newlen = htons(skb->len - nhoff);
1608         struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1609         const struct net_offload *ops;
1610         int proto = iph->protocol;
1611         int err = -ENOSYS;
1612
1613         if (skb->encapsulation) {
1614                 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1615                 skb_set_inner_network_header(skb, nhoff);
1616         }
1617
1618         csum_replace2(&iph->check, iph->tot_len, newlen);
1619         iph->tot_len = newlen;
1620
1621         rcu_read_lock();
1622         ops = rcu_dereference(inet_offloads[proto]);
1623         if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1624                 goto out_unlock;
1625
1626         /* Only need to add sizeof(*iph) to get to the next hdr below
1627          * because any hdr with option will have been flushed in
1628          * inet_gro_receive().
1629          */
1630         err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1631                               tcp4_gro_complete, udp4_gro_complete,
1632                               skb, nhoff + sizeof(*iph));
1633
1634 out_unlock:
1635         rcu_read_unlock();
1636
1637         return err;
1638 }
1639
1640 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1641 {
1642         skb->encapsulation = 1;
1643         skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1644         return inet_gro_complete(skb, nhoff);
1645 }
1646
1647 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1648                          unsigned short type, unsigned char protocol,
1649                          struct net *net)
1650 {
1651         struct socket *sock;
1652         int rc = sock_create_kern(net, family, type, protocol, &sock);
1653
1654         if (rc == 0) {
1655                 *sk = sock->sk;
1656                 (*sk)->sk_allocation = GFP_ATOMIC;
1657                 /*
1658                  * Unhash it so that IP input processing does not even see it,
1659                  * we do not wish this socket to see incoming packets.
1660                  */
1661                 (*sk)->sk_prot->unhash(*sk);
1662         }
1663         return rc;
1664 }
1665 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1666
1667 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1668 {
1669         unsigned long res = 0;
1670         int i;
1671
1672         for_each_possible_cpu(i)
1673                 res += snmp_get_cpu_field(mib, i, offt);
1674         return res;
1675 }
1676 EXPORT_SYMBOL_GPL(snmp_fold_field);
1677
1678 #if BITS_PER_LONG==32
1679
1680 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1681                          size_t syncp_offset)
1682 {
1683         void *bhptr;
1684         struct u64_stats_sync *syncp;
1685         u64 v;
1686         unsigned int start;
1687
1688         bhptr = per_cpu_ptr(mib, cpu);
1689         syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1690         do {
1691                 start = u64_stats_fetch_begin_irq(syncp);
1692                 v = *(((u64 *)bhptr) + offt);
1693         } while (u64_stats_fetch_retry_irq(syncp, start));
1694
1695         return v;
1696 }
1697 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1698
1699 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1700 {
1701         u64 res = 0;
1702         int cpu;
1703
1704         for_each_possible_cpu(cpu) {
1705                 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1706         }
1707         return res;
1708 }
1709 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1710 #endif
1711
1712 #ifdef CONFIG_IP_MULTICAST
1713 static const struct net_protocol igmp_protocol = {
1714         .handler =      igmp_rcv,
1715 };
1716 #endif
1717
1718 /* thinking of making this const? Don't.
1719  * early_demux can change based on sysctl.
1720  */
1721 static struct net_protocol tcp_protocol = {
1722         .early_demux    =       tcp_v4_early_demux,
1723         .early_demux_handler =  tcp_v4_early_demux,
1724         .handler        =       tcp_v4_rcv,
1725         .err_handler    =       tcp_v4_err,
1726         .no_policy      =       1,
1727         .icmp_strict_tag_validation = 1,
1728 };
1729
1730 /* thinking of making this const? Don't.
1731  * early_demux can change based on sysctl.
1732  */
1733 static struct net_protocol udp_protocol = {
1734         .early_demux =  udp_v4_early_demux,
1735         .early_demux_handler =  udp_v4_early_demux,
1736         .handler =      udp_rcv,
1737         .err_handler =  udp_err,
1738         .no_policy =    1,
1739 };
1740
1741 static const struct net_protocol icmp_protocol = {
1742         .handler =      icmp_rcv,
1743         .err_handler =  icmp_err,
1744         .no_policy =    1,
1745 };
1746
1747 static __net_init int ipv4_mib_init_net(struct net *net)
1748 {
1749         int i;
1750
1751         net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1752         if (!net->mib.tcp_statistics)
1753                 goto err_tcp_mib;
1754         net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1755         if (!net->mib.ip_statistics)
1756                 goto err_ip_mib;
1757
1758         for_each_possible_cpu(i) {
1759                 struct ipstats_mib *af_inet_stats;
1760                 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1761                 u64_stats_init(&af_inet_stats->syncp);
1762         }
1763
1764         net->mib.net_statistics = alloc_percpu(struct linux_mib);
1765         if (!net->mib.net_statistics)
1766                 goto err_net_mib;
1767         net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1768         if (!net->mib.udp_statistics)
1769                 goto err_udp_mib;
1770         net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1771         if (!net->mib.udplite_statistics)
1772                 goto err_udplite_mib;
1773         net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1774         if (!net->mib.icmp_statistics)
1775                 goto err_icmp_mib;
1776         net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1777                                               GFP_KERNEL);
1778         if (!net->mib.icmpmsg_statistics)
1779                 goto err_icmpmsg_mib;
1780
1781         tcp_mib_init(net);
1782         return 0;
1783
1784 err_icmpmsg_mib:
1785         free_percpu(net->mib.icmp_statistics);
1786 err_icmp_mib:
1787         free_percpu(net->mib.udplite_statistics);
1788 err_udplite_mib:
1789         free_percpu(net->mib.udp_statistics);
1790 err_udp_mib:
1791         free_percpu(net->mib.net_statistics);
1792 err_net_mib:
1793         free_percpu(net->mib.ip_statistics);
1794 err_ip_mib:
1795         free_percpu(net->mib.tcp_statistics);
1796 err_tcp_mib:
1797         return -ENOMEM;
1798 }
1799
1800 static __net_exit void ipv4_mib_exit_net(struct net *net)
1801 {
1802         kfree(net->mib.icmpmsg_statistics);
1803         free_percpu(net->mib.icmp_statistics);
1804         free_percpu(net->mib.udplite_statistics);
1805         free_percpu(net->mib.udp_statistics);
1806         free_percpu(net->mib.net_statistics);
1807         free_percpu(net->mib.ip_statistics);
1808         free_percpu(net->mib.tcp_statistics);
1809 #ifdef CONFIG_MPTCP
1810         /* allocated on demand, see mptcp_init_sock() */
1811         free_percpu(net->mib.mptcp_statistics);
1812 #endif
1813 }
1814
1815 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1816         .init = ipv4_mib_init_net,
1817         .exit = ipv4_mib_exit_net,
1818 };
1819
1820 static int __init init_ipv4_mibs(void)
1821 {
1822         return register_pernet_subsys(&ipv4_mib_ops);
1823 }
1824
1825 static __net_init int inet_init_net(struct net *net)
1826 {
1827         /*
1828          * Set defaults for local port range
1829          */
1830         seqlock_init(&net->ipv4.ip_local_ports.lock);
1831         net->ipv4.ip_local_ports.range[0] =  32768;
1832         net->ipv4.ip_local_ports.range[1] =  60999;
1833
1834         seqlock_init(&net->ipv4.ping_group_range.lock);
1835         /*
1836          * Sane defaults - nobody may create ping sockets.
1837          * Boot scripts should set this to distro-specific group.
1838          */
1839         net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1840         net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1841
1842         /* Default values for sysctl-controlled parameters.
1843          * We set them here, in case sysctl is not compiled.
1844          */
1845         net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1846         net->ipv4.sysctl_ip_fwd_update_priority = 1;
1847         net->ipv4.sysctl_ip_dynaddr = 0;
1848         net->ipv4.sysctl_ip_early_demux = 1;
1849         net->ipv4.sysctl_udp_early_demux = 1;
1850         net->ipv4.sysctl_tcp_early_demux = 1;
1851         net->ipv4.sysctl_nexthop_compat_mode = 1;
1852 #ifdef CONFIG_SYSCTL
1853         net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1854 #endif
1855
1856         /* Some igmp sysctl, whose values are always used */
1857         net->ipv4.sysctl_igmp_max_memberships = 20;
1858         net->ipv4.sysctl_igmp_max_msf = 10;
1859         /* IGMP reports for link-local multicast groups are enabled by default */
1860         net->ipv4.sysctl_igmp_llm_reports = 1;
1861         net->ipv4.sysctl_igmp_qrv = 2;
1862
1863         net->ipv4.sysctl_fib_notify_on_flag_change = 0;
1864
1865         return 0;
1866 }
1867
1868 static __net_initdata struct pernet_operations af_inet_ops = {
1869         .init = inet_init_net,
1870 };
1871
1872 static int __init init_inet_pernet_ops(void)
1873 {
1874         return register_pernet_subsys(&af_inet_ops);
1875 }
1876
1877 static int ipv4_proc_init(void);
1878
1879 /*
1880  *      IP protocol layer initialiser
1881  */
1882
1883 static struct packet_offload ip_packet_offload __read_mostly = {
1884         .type = cpu_to_be16(ETH_P_IP),
1885         .callbacks = {
1886                 .gso_segment = inet_gso_segment,
1887                 .gro_receive = inet_gro_receive,
1888                 .gro_complete = inet_gro_complete,
1889         },
1890 };
1891
1892 static const struct net_offload ipip_offload = {
1893         .callbacks = {
1894                 .gso_segment    = ipip_gso_segment,
1895                 .gro_receive    = ipip_gro_receive,
1896                 .gro_complete   = ipip_gro_complete,
1897         },
1898 };
1899
1900 static int __init ipip_offload_init(void)
1901 {
1902         return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1903 }
1904
1905 static int __init ipv4_offload_init(void)
1906 {
1907         /*
1908          * Add offloads
1909          */
1910         if (udpv4_offload_init() < 0)
1911                 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1912         if (tcpv4_offload_init() < 0)
1913                 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1914         if (ipip_offload_init() < 0)
1915                 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1916
1917         dev_add_offload(&ip_packet_offload);
1918         return 0;
1919 }
1920
1921 fs_initcall(ipv4_offload_init);
1922
1923 static struct packet_type ip_packet_type __read_mostly = {
1924         .type = cpu_to_be16(ETH_P_IP),
1925         .func = ip_rcv,
1926         .list_func = ip_list_rcv,
1927 };
1928
1929 static int __init inet_init(void)
1930 {
1931         struct inet_protosw *q;
1932         struct list_head *r;
1933         int rc;
1934
1935         sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1936
1937         rc = proto_register(&tcp_prot, 1);
1938         if (rc)
1939                 goto out;
1940
1941         rc = proto_register(&udp_prot, 1);
1942         if (rc)
1943                 goto out_unregister_tcp_proto;
1944
1945         rc = proto_register(&raw_prot, 1);
1946         if (rc)
1947                 goto out_unregister_udp_proto;
1948
1949         rc = proto_register(&ping_prot, 1);
1950         if (rc)
1951                 goto out_unregister_raw_proto;
1952
1953         /*
1954          *      Tell SOCKET that we are alive...
1955          */
1956
1957         (void)sock_register(&inet_family_ops);
1958
1959 #ifdef CONFIG_SYSCTL
1960         ip_static_sysctl_init();
1961 #endif
1962
1963         /*
1964          *      Add all the base protocols.
1965          */
1966
1967         if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1968                 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1969         if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1970                 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1971         if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1972                 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1973 #ifdef CONFIG_IP_MULTICAST
1974         if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1975                 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1976 #endif
1977
1978         /* Register the socket-side information for inet_create. */
1979         for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1980                 INIT_LIST_HEAD(r);
1981
1982         for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1983                 inet_register_protosw(q);
1984
1985         /*
1986          *      Set the ARP module up
1987          */
1988
1989         arp_init();
1990
1991         /*
1992          *      Set the IP module up
1993          */
1994
1995         ip_init();
1996
1997         /* Setup TCP slab cache for open requests. */
1998         tcp_init();
1999
2000         /* Setup UDP memory threshold */
2001         udp_init();
2002
2003         /* Add UDP-Lite (RFC 3828) */
2004         udplite4_register();
2005
2006         raw_init();
2007
2008         ping_init();
2009
2010         /*
2011          *      Set the ICMP layer up
2012          */
2013
2014         if (icmp_init() < 0)
2015                 panic("Failed to create the ICMP control socket.\n");
2016
2017         /*
2018          *      Initialise the multicast router
2019          */
2020 #if defined(CONFIG_IP_MROUTE)
2021         if (ip_mr_init())
2022                 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2023 #endif
2024
2025         if (init_inet_pernet_ops())
2026                 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2027         /*
2028          *      Initialise per-cpu ipv4 mibs
2029          */
2030
2031         if (init_ipv4_mibs())
2032                 pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
2033
2034         ipv4_proc_init();
2035
2036         ipfrag_init();
2037
2038         dev_add_pack(&ip_packet_type);
2039
2040         ip_tunnel_core_init();
2041
2042         rc = 0;
2043 out:
2044         return rc;
2045 out_unregister_raw_proto:
2046         proto_unregister(&raw_prot);
2047 out_unregister_udp_proto:
2048         proto_unregister(&udp_prot);
2049 out_unregister_tcp_proto:
2050         proto_unregister(&tcp_prot);
2051         goto out;
2052 }
2053
2054 fs_initcall(inet_init);
2055
2056 /* ------------------------------------------------------------------------ */
2057
2058 #ifdef CONFIG_PROC_FS
2059 static int __init ipv4_proc_init(void)
2060 {
2061         int rc = 0;
2062
2063         if (raw_proc_init())
2064                 goto out_raw;
2065         if (tcp4_proc_init())
2066                 goto out_tcp;
2067         if (udp4_proc_init())
2068                 goto out_udp;
2069         if (ping_proc_init())
2070                 goto out_ping;
2071         if (ip_misc_proc_init())
2072                 goto out_misc;
2073 out:
2074         return rc;
2075 out_misc:
2076         ping_proc_exit();
2077 out_ping:
2078         udp4_proc_exit();
2079 out_udp:
2080         tcp4_proc_exit();
2081 out_tcp:
2082         raw_proc_exit();
2083 out_raw:
2084         rc = -ENOMEM;
2085         goto out;
2086 }
2087
2088 #else /* CONFIG_PROC_FS */
2089 static int __init ipv4_proc_init(void)
2090 {
2091         return 0;
2092 }
2093 #endif /* CONFIG_PROC_FS */