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/gro.h>
 103 #include <net/tcp.h>
 104 #include <net/udp.h>
 105 #include <net/udplite.h>
 106 #include <net/ping.h>
 107 #include <linux/skbuff.h>
 108 #include <net/sock.h>
 109 #include <net/raw.h>
 110 #include <net/icmp.h>
 111 #include <net/inet_common.h>
 112 #include <net/ip_tunnels.h>
 113 #include <net/xfrm.h>
 114 #include <net/net_namespace.h>
 115 #include <net/secure_seq.h>
 116 #ifdef CONFIG_IP_MROUTE
 117 #include <linux/mroute.h>
 118 #endif
 119 #include <net/l3mdev.h>
 120 #include <net/compat.h>
 121
 122 #include <trace/events/sock.h>
 123
 124 /* The inetsw table contains everything that inet_create needs to
 125  * build a new socket.
 126  */
 127 static struct list_head inetsw[SOCK_MAX];
 128 static DEFINE_SPINLOCK(inetsw_lock);
 129
 130 /* New destruction routine */
 131
 132 void inet_sock_destruct(struct sock *sk)
 133 {
 134         struct inet_sock *inet = inet_sk(sk);
 135
 136         __skb_queue_purge(&sk->sk_receive_queue);
 137         __skb_queue_purge(&sk->sk_error_queue);
 138
 139         sk_mem_reclaim_final(sk);
 140
 141         if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
 142                 pr_err("Attempt to release TCP socket in state %d %p\n",
 143                        sk->sk_state, sk);
 144                 return;
 145         }
 146         if (!sock_flag(sk, SOCK_DEAD)) {
 147                 pr_err("Attempt to release alive inet socket %p\n", sk);
 148                 return;
 149         }
 150
 151         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
 152         WARN_ON(refcount_read(&sk->sk_wmem_alloc));
 153         WARN_ON(sk->sk_wmem_queued);
 154         WARN_ON(sk_forward_alloc_get(sk));
 155
 156         kfree(rcu_dereference_protected(inet->inet_opt, 1));
 157         dst_release(rcu_dereference_protected(sk->sk_dst_cache, 1));
 158         dst_release(rcu_dereference_protected(sk->sk_rx_dst, 1));
 159         sk_refcnt_debug_dec(sk);
 160 }
 161 EXPORT_SYMBOL(inet_sock_destruct);
 162
 163 /*
 164  *      The routines beyond this point handle the behaviour of an AF_INET
 165  *      socket object. Mostly it punts to the subprotocols of IP to do
 166  *      the work.
 167  */
 168
 169 /*
 170  *      Automatically bind an unbound socket.
 171  */
 172
 173 static int inet_autobind(struct sock *sk)
 174 {
 175         struct inet_sock *inet;
 176         /* We may need to bind the socket. */
 177         lock_sock(sk);
 178         inet = inet_sk(sk);
 179         if (!inet->inet_num) {
 180                 if (sk->sk_prot->get_port(sk, 0)) {
 181                         release_sock(sk);
 182                         return -EAGAIN;
 183                 }
 184                 inet->inet_sport = htons(inet->inet_num);
 185         }
 186         release_sock(sk);
 187         return 0;
 188 }
 189
 190 /*
 191  *      Move a socket into listening state.
 192  */
 193 int inet_listen(struct socket *sock, int backlog)
 194 {
 195         struct sock *sk = sock->sk;
 196         unsigned char old_state;
 197         int err, tcp_fastopen;
 198
 199         lock_sock(sk);
 200
 201         err = -EINVAL;
 202         if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
 203                 goto out;
 204
 205         old_state = sk->sk_state;
 206         if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
 207                 goto out;
 208
 209         WRITE_ONCE(sk->sk_max_ack_backlog, backlog);
 210         /* Really, if the socket is already in listen state
 211          * we can only allow the backlog to be adjusted.
 212          */
 213         if (old_state != TCP_LISTEN) {
 214                 /* Enable TFO w/o requiring TCP_FASTOPEN socket option.
 215                  * Note that only TCP sockets (SOCK_STREAM) will reach here.
 216                  * Also fastopen backlog may already been set via the option
 217                  * because the socket was in TCP_LISTEN state previously but
 218                  * was shutdown() rather than close().
 219                  */
 220                 tcp_fastopen = sock_net(sk)->ipv4.sysctl_tcp_fastopen;
 221                 if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
 222                     (tcp_fastopen & TFO_SERVER_ENABLE) &&
 223                     !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
 224                         fastopen_queue_tune(sk, backlog);
 225                         tcp_fastopen_init_key_once(sock_net(sk));
 226                 }
 227
 228                 err = inet_csk_listen_start(sk);
 229                 if (err)
 230                         goto out;
 231                 tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL);
 232         }
 233         err = 0;
 234
 235 out:
 236         release_sock(sk);
 237         return err;
 238 }
 239 EXPORT_SYMBOL(inet_listen);
 240
 241 /*
 242  *      Create an inet socket.
 243  */
 244
 245 static int inet_create(struct net *net, struct socket *sock, int protocol,
 246                        int kern)
 247 {
 248         struct sock *sk;
 249         struct inet_protosw *answer;
 250         struct inet_sock *inet;
 251         struct proto *answer_prot;
 252         unsigned char answer_flags;
 253         int try_loading_module = 0;
 254         int err;
 255
 256         if (protocol < 0 || protocol >= IPPROTO_MAX)
 257                 return -EINVAL;
 258
 259         sock->state = SS_UNCONNECTED;
 260
 261         /* Look for the requested type/protocol pair. */
 262 lookup_protocol:
 263         err = -ESOCKTNOSUPPORT;
 264         rcu_read_lock();
 265         list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
 266
 267                 err = 0;
 268                 /* Check the non-wild match. */
 269                 if (protocol == answer->protocol) {
 270                         if (protocol != IPPROTO_IP)
 271                                 break;
 272                 } else {
 273                         /* Check for the two wild cases. */
 274                         if (IPPROTO_IP == protocol) {
 275                                 protocol = answer->protocol;
 276                                 break;
 277                         }
 278                         if (IPPROTO_IP == answer->protocol)
 279                                 break;
 280                 }
 281                 err = -EPROTONOSUPPORT;
 282         }
 283
 284         if (unlikely(err)) {
 285                 if (try_loading_module < 2) {
 286                         rcu_read_unlock();
 287                         /*
 288                          * Be more specific, e.g. net-pf-2-proto-132-type-1
 289                          * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
 290                          */
 291                         if (++try_loading_module == 1)
 292                                 request_module("net-pf-%d-proto-%d-type-%d",
 293                                                PF_INET, protocol, sock->type);
 294                         /*
 295                          * Fall back to generic, e.g. net-pf-2-proto-132
 296                          * (net-pf-PF_INET-proto-IPPROTO_SCTP)
 297                          */
 298                         else
 299                                 request_module("net-pf-%d-proto-%d",
 300                                                PF_INET, protocol);
 301                         goto lookup_protocol;
 302                 } else
 303                         goto out_rcu_unlock;
 304         }
 305
 306         err = -EPERM;
 307         if (sock->type == SOCK_RAW && !kern &&
 308             !ns_capable(net->user_ns, CAP_NET_RAW))
 309                 goto out_rcu_unlock;
 310
 311         sock->ops = answer->ops;
 312         answer_prot = answer->prot;
 313         answer_flags = answer->flags;
 314         rcu_read_unlock();
 315
 316         WARN_ON(!answer_prot->slab);
 317
 318         err = -ENOMEM;
 319         sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
 320         if (!sk)
 321                 goto out;
 322
 323         err = 0;
 324         if (INET_PROTOSW_REUSE & answer_flags)
 325                 sk->sk_reuse = SK_CAN_REUSE;
 326
 327         inet = inet_sk(sk);
 328         inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
 329
 330         inet->nodefrag = 0;
 331
 332         if (SOCK_RAW == sock->type) {
 333                 inet->inet_num = protocol;
 334                 if (IPPROTO_RAW == protocol)
 335                         inet->hdrincl = 1;
 336         }
 337
 338         if (net->ipv4.sysctl_ip_no_pmtu_disc)
 339                 inet->pmtudisc = IP_PMTUDISC_DONT;
 340         else
 341                 inet->pmtudisc = IP_PMTUDISC_WANT;
 342
 343         inet->inet_id = 0;
 344
 345         sock_init_data(sock, sk);
 346
 347         sk->sk_destruct    = inet_sock_destruct;
 348         sk->sk_protocol    = protocol;
 349         sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
 350
 351         inet->uc_ttl    = -1;
 352         inet->mc_loop   = 1;
 353         inet->mc_ttl    = 1;
 354         inet->mc_all    = 1;
 355         inet->mc_index  = 0;
 356         inet->mc_list   = NULL;
 357         inet->rcv_tos   = 0;
 358
 359         sk_refcnt_debug_inc(sk);
 360
 361         if (inet->inet_num) {
 362                 /* It assumes that any protocol which allows
 363                  * the user to assign a number at socket
 364                  * creation time automatically
 365                  * shares.
 366                  */
 367                 inet->inet_sport = htons(inet->inet_num);
 368                 /* Add to protocol hash chains. */
 369                 err = sk->sk_prot->hash(sk);
 370                 if (err) {
 371                         sk_common_release(sk);
 372                         goto out;
 373                 }
 374         }
 375
 376         if (sk->sk_prot->init) {
 377                 err = sk->sk_prot->init(sk);
 378                 if (err) {
 379                         sk_common_release(sk);
 380                         goto out;
 381                 }
 382         }
 383
 384         if (!kern) {
 385                 err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
 386                 if (err) {
 387                         sk_common_release(sk);
 388                         goto out;
 389                 }
 390         }
 391 out:
 392         return err;
 393 out_rcu_unlock:
 394         rcu_read_unlock();
 395         goto out;
 396 }
 397
 398
 399 /*
 400  *      The peer socket should always be NULL (or else). When we call this
 401  *      function we are destroying the object and from then on nobody
 402  *      should refer to it.
 403  */
 404 int inet_release(struct socket *sock)
 405 {
 406         struct sock *sk = sock->sk;
 407
 408         if (sk) {
 409                 long timeout;
 410
 411                 if (!sk->sk_kern_sock)
 412                         BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk);
 413
 414                 /* Applications forget to leave groups before exiting */
 415                 ip_mc_drop_socket(sk);
 416
 417                 /* If linger is set, we don't return until the close
 418                  * is complete.  Otherwise we return immediately. The
 419                  * actually closing is done the same either way.
 420                  *
 421                  * If the close is due to the process exiting, we never
 422                  * linger..
 423                  */
 424                 timeout = 0;
 425                 if (sock_flag(sk, SOCK_LINGER) &&
 426                     !(current->flags & PF_EXITING))
 427                         timeout = sk->sk_lingertime;
 428                 sk->sk_prot->close(sk, timeout);
 429                 sock->sk = NULL;
 430         }
 431         return 0;
 432 }
 433 EXPORT_SYMBOL(inet_release);
 434
 435 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 436 {
 437         struct sock *sk = sock->sk;
 438         u32 flags = BIND_WITH_LOCK;
 439         int err;
 440
 441         /* If the socket has its own bind function then use it. (RAW) */
 442         if (sk->sk_prot->bind) {
 443                 return sk->sk_prot->bind(sk, uaddr, addr_len);
 444         }
 445         if (addr_len < sizeof(struct sockaddr_in))
 446                 return -EINVAL;
 447
 448         /* BPF prog is run before any checks are done so that if the prog
 449          * changes context in a wrong way it will be caught.
 450          */
 451         err = BPF_CGROUP_RUN_PROG_INET_BIND_LOCK(sk, uaddr,
 452                                                  CGROUP_INET4_BIND, &flags);
 453         if (err)
 454                 return err;
 455
 456         return __inet_bind(sk, uaddr, addr_len, flags);
 457 }
 458 EXPORT_SYMBOL(inet_bind);
 459
 460 int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
 461                 u32 flags)
 462 {
 463         struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
 464         struct inet_sock *inet = inet_sk(sk);
 465         struct net *net = sock_net(sk);
 466         unsigned short snum;
 467         int chk_addr_ret;
 468         u32 tb_id = RT_TABLE_LOCAL;
 469         int err;
 470
 471         if (addr->sin_family != AF_INET) {
 472                 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
 473                  * only if s_addr is INADDR_ANY.
 474                  */
 475                 err = -EAFNOSUPPORT;
 476                 if (addr->sin_family != AF_UNSPEC ||
 477                     addr->sin_addr.s_addr != htonl(INADDR_ANY))
 478                         goto out;
 479         }
 480
 481         tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
 482         chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
 483
 484         /* Not specified by any standard per-se, however it breaks too
 485          * many applications when removed.  It is unfortunate since
 486          * allowing applications to make a non-local bind solves
 487          * several problems with systems using dynamic addressing.
 488          * (ie. your servers still start up even if your ISDN link
 489          *  is temporarily down)
 490          */
 491         err = -EADDRNOTAVAIL;
 492         if (!inet_addr_valid_or_nonlocal(net, inet, addr->sin_addr.s_addr,
 493                                          chk_addr_ret))
 494                 goto out;
 495
 496         snum = ntohs(addr->sin_port);
 497         err = -EACCES;
 498         if (!(flags & BIND_NO_CAP_NET_BIND_SERVICE) &&
 499             snum && inet_port_requires_bind_service(net, snum) &&
 500             !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
 501                 goto out;
 502
 503         /*      We keep a pair of addresses. rcv_saddr is the one
 504          *      used by hash lookups, and saddr is used for transmit.
 505          *
 506          *      In the BSD API these are the same except where it
 507          *      would be illegal to use them (multicast/broadcast) in
 508          *      which case the sending device address is used.
 509          */
 510         if (flags & BIND_WITH_LOCK)
 511                 lock_sock(sk);
 512
 513         /* Check these errors (active socket, double bind). */
 514         err = -EINVAL;
 515         if (sk->sk_state != TCP_CLOSE || inet->inet_num)
 516                 goto out_release_sock;
 517
 518         inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
 519         if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
 520                 inet->inet_saddr = 0;  /* Use device */
 521
 522         /* Make sure we are allowed to bind here. */
 523         if (snum || !(inet->bind_address_no_port ||
 524                       (flags & BIND_FORCE_ADDRESS_NO_PORT))) {
 525                 if (sk->sk_prot->get_port(sk, snum)) {
 526                         inet->inet_saddr = inet->inet_rcv_saddr = 0;
 527                         err = -EADDRINUSE;
 528                         goto out_release_sock;
 529                 }
 530                 if (!(flags & BIND_FROM_BPF)) {
 531                         err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
 532                         if (err) {
 533                                 inet->inet_saddr = inet->inet_rcv_saddr = 0;
 534                                 if (sk->sk_prot->put_port)
 535                                         sk->sk_prot->put_port(sk);
 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 *));
 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, &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                 fallthrough;
 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                 fallthrough;
 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 SIOCADDRT:
 935         case SIOCDELRT:
 936                 if (copy_from_user(&rt, p, sizeof(struct rtentry)))
 937                         return -EFAULT;
 938                 err = ip_rt_ioctl(net, cmd, &rt);
 939                 break;
 940         case SIOCRTMSG:
 941                 err = -EINVAL;
 942                 break;
 943         case SIOCDARP:
 944         case SIOCGARP:
 945         case SIOCSARP:
 946                 err = arp_ioctl(net, cmd, (void __user *)arg);
 947                 break;
 948         case SIOCGIFADDR:
 949         case SIOCGIFBRDADDR:
 950         case SIOCGIFNETMASK:
 951         case SIOCGIFDSTADDR:
 952         case SIOCGIFPFLAGS:
 953                 if (get_user_ifreq(&ifr, NULL, p))
 954                         return -EFAULT;
 955                 err = devinet_ioctl(net, cmd, &ifr);
 956                 if (!err && put_user_ifreq(&ifr, p))
 957                         err = -EFAULT;
 958                 break;
 959
 960         case SIOCSIFADDR:
 961         case SIOCSIFBRDADDR:
 962         case SIOCSIFNETMASK:
 963         case SIOCSIFDSTADDR:
 964         case SIOCSIFPFLAGS:
 965         case SIOCSIFFLAGS:
 966                 if (get_user_ifreq(&ifr, NULL, p))
 967                         return -EFAULT;
 968                 err = devinet_ioctl(net, cmd, &ifr);
 969                 break;
 970         default:
 971                 if (sk->sk_prot->ioctl)
 972                         err = sk->sk_prot->ioctl(sk, cmd, arg);
 973                 else
 974                         err = -ENOIOCTLCMD;
 975                 break;
 976         }
 977         return err;
 978 }
 979 EXPORT_SYMBOL(inet_ioctl);
 980
 981 #ifdef CONFIG_COMPAT
 982 static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd,
 983                 struct compat_rtentry __user *ur)
 984 {
 985         compat_uptr_t rtdev;
 986         struct rtentry rt;
 987
 988         if (copy_from_user(&rt.rt_dst, &ur->rt_dst,
 989                         3 * sizeof(struct sockaddr)) ||
 990             get_user(rt.rt_flags, &ur->rt_flags) ||
 991             get_user(rt.rt_metric, &ur->rt_metric) ||
 992             get_user(rt.rt_mtu, &ur->rt_mtu) ||
 993             get_user(rt.rt_window, &ur->rt_window) ||
 994             get_user(rt.rt_irtt, &ur->rt_irtt) ||
 995             get_user(rtdev, &ur->rt_dev))
 996                 return -EFAULT;
 997
 998         rt.rt_dev = compat_ptr(rtdev);
 999         return ip_rt_ioctl(sock_net(sk), cmd, &rt);
1000 }
1001
1002 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1003 {
1004         void __user *argp = compat_ptr(arg);
1005         struct sock *sk = sock->sk;
1006
1007         switch (cmd) {
1008         case SIOCADDRT:
1009         case SIOCDELRT:
1010                 return inet_compat_routing_ioctl(sk, cmd, argp);
1011         default:
1012                 if (!sk->sk_prot->compat_ioctl)
1013                         return -ENOIOCTLCMD;
1014                 return sk->sk_prot->compat_ioctl(sk, cmd, arg);
1015         }
1016 }
1017 #endif /* CONFIG_COMPAT */
1018
1019 const struct proto_ops inet_stream_ops = {
1020         .family            = PF_INET,
1021         .owner             = THIS_MODULE,
1022         .release           = inet_release,
1023         .bind              = inet_bind,
1024         .connect           = inet_stream_connect,
1025         .socketpair        = sock_no_socketpair,
1026         .accept            = inet_accept,
1027         .getname           = inet_getname,
1028         .poll              = tcp_poll,
1029         .ioctl             = inet_ioctl,
1030         .gettstamp         = sock_gettstamp,
1031         .listen            = inet_listen,
1032         .shutdown          = inet_shutdown,
1033         .setsockopt        = sock_common_setsockopt,
1034         .getsockopt        = sock_common_getsockopt,
1035         .sendmsg           = inet_sendmsg,
1036         .recvmsg           = inet_recvmsg,
1037 #ifdef CONFIG_MMU
1038         .mmap              = tcp_mmap,
1039 #endif
1040         .sendpage          = inet_sendpage,
1041         .splice_read       = tcp_splice_read,
1042         .read_sock         = tcp_read_sock,
1043         .sendmsg_locked    = tcp_sendmsg_locked,
1044         .sendpage_locked   = tcp_sendpage_locked,
1045         .peek_len          = tcp_peek_len,
1046 #ifdef CONFIG_COMPAT
1047         .compat_ioctl      = inet_compat_ioctl,
1048 #endif
1049         .set_rcvlowat      = tcp_set_rcvlowat,
1050 };
1051 EXPORT_SYMBOL(inet_stream_ops);
1052
1053 const struct proto_ops inet_dgram_ops = {
1054         .family            = PF_INET,
1055         .owner             = THIS_MODULE,
1056         .release           = inet_release,
1057         .bind              = inet_bind,
1058         .connect           = inet_dgram_connect,
1059         .socketpair        = sock_no_socketpair,
1060         .accept            = sock_no_accept,
1061         .getname           = inet_getname,
1062         .poll              = udp_poll,
1063         .ioctl             = inet_ioctl,
1064         .gettstamp         = sock_gettstamp,
1065         .listen            = sock_no_listen,
1066         .shutdown          = inet_shutdown,
1067         .setsockopt        = sock_common_setsockopt,
1068         .getsockopt        = sock_common_getsockopt,
1069         .sendmsg           = inet_sendmsg,
1070         .read_sock         = udp_read_sock,
1071         .recvmsg           = inet_recvmsg,
1072         .mmap              = sock_no_mmap,
1073         .sendpage          = inet_sendpage,
1074         .set_peek_off      = sk_set_peek_off,
1075 #ifdef CONFIG_COMPAT
1076         .compat_ioctl      = inet_compat_ioctl,
1077 #endif
1078 };
1079 EXPORT_SYMBOL(inet_dgram_ops);
1080
1081 /*
1082  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1083  * udp_poll
1084  */
1085 static const struct proto_ops inet_sockraw_ops = {
1086         .family            = PF_INET,
1087         .owner             = THIS_MODULE,
1088         .release           = inet_release,
1089         .bind              = inet_bind,
1090         .connect           = inet_dgram_connect,
1091         .socketpair        = sock_no_socketpair,
1092         .accept            = sock_no_accept,
1093         .getname           = inet_getname,
1094         .poll              = datagram_poll,
1095         .ioctl             = inet_ioctl,
1096         .gettstamp         = sock_gettstamp,
1097         .listen            = sock_no_listen,
1098         .shutdown          = inet_shutdown,
1099         .setsockopt        = sock_common_setsockopt,
1100         .getsockopt        = sock_common_getsockopt,
1101         .sendmsg           = inet_sendmsg,
1102         .recvmsg           = inet_recvmsg,
1103         .mmap              = sock_no_mmap,
1104         .sendpage          = inet_sendpage,
1105 #ifdef CONFIG_COMPAT
1106         .compat_ioctl      = inet_compat_ioctl,
1107 #endif
1108 };
1109
1110 static const struct net_proto_family inet_family_ops = {
1111         .family = PF_INET,
1112         .create = inet_create,
1113         .owner  = THIS_MODULE,
1114 };
1115
1116 /* Upon startup we insert all the elements in inetsw_array[] into
1117  * the linked list inetsw.
1118  */
1119 static struct inet_protosw inetsw_array[] =
1120 {
1121         {
1122                 .type =       SOCK_STREAM,
1123                 .protocol =   IPPROTO_TCP,
1124                 .prot =       &tcp_prot,
1125                 .ops =        &inet_stream_ops,
1126                 .flags =      INET_PROTOSW_PERMANENT |
1127                               INET_PROTOSW_ICSK,
1128         },
1129
1130         {
1131                 .type =       SOCK_DGRAM,
1132                 .protocol =   IPPROTO_UDP,
1133                 .prot =       &udp_prot,
1134                 .ops =        &inet_dgram_ops,
1135                 .flags =      INET_PROTOSW_PERMANENT,
1136        },
1137
1138        {
1139                 .type =       SOCK_DGRAM,
1140                 .protocol =   IPPROTO_ICMP,
1141                 .prot =       &ping_prot,
1142                 .ops =        &inet_sockraw_ops,
1143                 .flags =      INET_PROTOSW_REUSE,
1144        },
1145
1146        {
1147                .type =       SOCK_RAW,
1148                .protocol =   IPPROTO_IP,        /* wild card */
1149                .prot =       &raw_prot,
1150                .ops =        &inet_sockraw_ops,
1151                .flags =      INET_PROTOSW_REUSE,
1152        }
1153 };
1154
1155 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1156
1157 void inet_register_protosw(struct inet_protosw *p)
1158 {
1159         struct list_head *lh;
1160         struct inet_protosw *answer;
1161         int protocol = p->protocol;
1162         struct list_head *last_perm;
1163
1164         spin_lock_bh(&inetsw_lock);
1165
1166         if (p->type >= SOCK_MAX)
1167                 goto out_illegal;
1168
1169         /* If we are trying to override a permanent protocol, bail. */
1170         last_perm = &inetsw[p->type];
1171         list_for_each(lh, &inetsw[p->type]) {
1172                 answer = list_entry(lh, struct inet_protosw, list);
1173                 /* Check only the non-wild match. */
1174                 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1175                         break;
1176                 if (protocol == answer->protocol)
1177                         goto out_permanent;
1178                 last_perm = lh;
1179         }
1180
1181         /* Add the new entry after the last permanent entry if any, so that
1182          * the new entry does not override a permanent entry when matched with
1183          * a wild-card protocol. But it is allowed to override any existing
1184          * non-permanent entry.  This means that when we remove this entry, the
1185          * system automatically returns to the old behavior.
1186          */
1187         list_add_rcu(&p->list, last_perm);
1188 out:
1189         spin_unlock_bh(&inetsw_lock);
1190
1191         return;
1192
1193 out_permanent:
1194         pr_err("Attempt to override permanent protocol %d\n", protocol);
1195         goto out;
1196
1197 out_illegal:
1198         pr_err("Ignoring attempt to register invalid socket type %d\n",
1199                p->type);
1200         goto out;
1201 }
1202 EXPORT_SYMBOL(inet_register_protosw);
1203
1204 void inet_unregister_protosw(struct inet_protosw *p)
1205 {
1206         if (INET_PROTOSW_PERMANENT & p->flags) {
1207                 pr_err("Attempt to unregister permanent protocol %d\n",
1208                        p->protocol);
1209         } else {
1210                 spin_lock_bh(&inetsw_lock);
1211                 list_del_rcu(&p->list);
1212                 spin_unlock_bh(&inetsw_lock);
1213
1214                 synchronize_net();
1215         }
1216 }
1217 EXPORT_SYMBOL(inet_unregister_protosw);
1218
1219 static int inet_sk_reselect_saddr(struct sock *sk)
1220 {
1221         struct inet_sock *inet = inet_sk(sk);
1222         __be32 old_saddr = inet->inet_saddr;
1223         __be32 daddr = inet->inet_daddr;
1224         struct flowi4 *fl4;
1225         struct rtable *rt;
1226         __be32 new_saddr;
1227         struct ip_options_rcu *inet_opt;
1228
1229         inet_opt = rcu_dereference_protected(inet->inet_opt,
1230                                              lockdep_sock_is_held(sk));
1231         if (inet_opt && inet_opt->opt.srr)
1232                 daddr = inet_opt->opt.faddr;
1233
1234         /* Query new route. */
1235         fl4 = &inet->cork.fl.u.ip4;
1236         rt = ip_route_connect(fl4, daddr, 0, sk->sk_bound_dev_if,
1237                               sk->sk_protocol, inet->inet_sport,
1238                               inet->inet_dport, sk);
1239         if (IS_ERR(rt))
1240                 return PTR_ERR(rt);
1241
1242         sk_setup_caps(sk, &rt->dst);
1243
1244         new_saddr = fl4->saddr;
1245
1246         if (new_saddr == old_saddr)
1247                 return 0;
1248
1249         if (sock_net(sk)->ipv4.sysctl_ip_dynaddr > 1) {
1250                 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1251                         __func__, &old_saddr, &new_saddr);
1252         }
1253
1254         inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1255
1256         /*
1257          * XXX The only one ugly spot where we need to
1258          * XXX really change the sockets identity after
1259          * XXX it has entered the hashes. -DaveM
1260          *
1261          * Besides that, it does not check for connection
1262          * uniqueness. Wait for troubles.
1263          */
1264         return __sk_prot_rehash(sk);
1265 }
1266
1267 int inet_sk_rebuild_header(struct sock *sk)
1268 {
1269         struct inet_sock *inet = inet_sk(sk);
1270         struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1271         __be32 daddr;
1272         struct ip_options_rcu *inet_opt;
1273         struct flowi4 *fl4;
1274         int err;
1275
1276         /* Route is OK, nothing to do. */
1277         if (rt)
1278                 return 0;
1279
1280         /* Reroute. */
1281         rcu_read_lock();
1282         inet_opt = rcu_dereference(inet->inet_opt);
1283         daddr = inet->inet_daddr;
1284         if (inet_opt && inet_opt->opt.srr)
1285                 daddr = inet_opt->opt.faddr;
1286         rcu_read_unlock();
1287         fl4 = &inet->cork.fl.u.ip4;
1288         rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1289                                    inet->inet_dport, inet->inet_sport,
1290                                    sk->sk_protocol, RT_CONN_FLAGS(sk),
1291                                    sk->sk_bound_dev_if);
1292         if (!IS_ERR(rt)) {
1293                 err = 0;
1294                 sk_setup_caps(sk, &rt->dst);
1295         } else {
1296                 err = PTR_ERR(rt);
1297
1298                 /* Routing failed... */
1299                 sk->sk_route_caps = 0;
1300                 /*
1301                  * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1302                  * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1303                  */
1304                 if (!sock_net(sk)->ipv4.sysctl_ip_dynaddr ||
1305                     sk->sk_state != TCP_SYN_SENT ||
1306                     (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1307                     (err = inet_sk_reselect_saddr(sk)) != 0)
1308                         sk->sk_err_soft = -err;
1309         }
1310
1311         return err;
1312 }
1313 EXPORT_SYMBOL(inet_sk_rebuild_header);
1314
1315 void inet_sk_set_state(struct sock *sk, int state)
1316 {
1317         trace_inet_sock_set_state(sk, sk->sk_state, state);
1318         sk->sk_state = state;
1319 }
1320 EXPORT_SYMBOL(inet_sk_set_state);
1321
1322 void inet_sk_state_store(struct sock *sk, int newstate)
1323 {
1324         trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1325         smp_store_release(&sk->sk_state, newstate);
1326 }
1327
1328 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1329                                  netdev_features_t features)
1330 {
1331         bool udpfrag = false, fixedid = false, gso_partial, encap;
1332         struct sk_buff *segs = ERR_PTR(-EINVAL);
1333         const struct net_offload *ops;
1334         unsigned int offset = 0;
1335         struct iphdr *iph;
1336         int proto, tot_len;
1337         int nhoff;
1338         int ihl;
1339         int id;
1340
1341         skb_reset_network_header(skb);
1342         nhoff = skb_network_header(skb) - skb_mac_header(skb);
1343         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1344                 goto out;
1345
1346         iph = ip_hdr(skb);
1347         ihl = iph->ihl * 4;
1348         if (ihl < sizeof(*iph))
1349                 goto out;
1350
1351         id = ntohs(iph->id);
1352         proto = iph->protocol;
1353
1354         /* Warning: after this point, iph might be no longer valid */
1355         if (unlikely(!pskb_may_pull(skb, ihl)))
1356                 goto out;
1357         __skb_pull(skb, ihl);
1358
1359         encap = SKB_GSO_CB(skb)->encap_level > 0;
1360         if (encap)
1361                 features &= skb->dev->hw_enc_features;
1362         SKB_GSO_CB(skb)->encap_level += ihl;
1363
1364         skb_reset_transport_header(skb);
1365
1366         segs = ERR_PTR(-EPROTONOSUPPORT);
1367
1368         if (!skb->encapsulation || encap) {
1369                 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1370                 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1371
1372                 /* fixed ID is invalid if DF bit is not set */
1373                 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1374                         goto out;
1375         }
1376
1377         ops = rcu_dereference(inet_offloads[proto]);
1378         if (likely(ops && ops->callbacks.gso_segment)) {
1379                 segs = ops->callbacks.gso_segment(skb, features);
1380                 if (!segs)
1381                         skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
1382         }
1383
1384         if (IS_ERR_OR_NULL(segs))
1385                 goto out;
1386
1387         gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1388
1389         skb = segs;
1390         do {
1391                 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1392                 if (udpfrag) {
1393                         iph->frag_off = htons(offset >> 3);
1394                         if (skb->next)
1395                                 iph->frag_off |= htons(IP_MF);
1396                         offset += skb->len - nhoff - ihl;
1397                         tot_len = skb->len - nhoff;
1398                 } else if (skb_is_gso(skb)) {
1399                         if (!fixedid) {
1400                                 iph->id = htons(id);
1401                                 id += skb_shinfo(skb)->gso_segs;
1402                         }
1403
1404                         if (gso_partial)
1405                                 tot_len = skb_shinfo(skb)->gso_size +
1406                                           SKB_GSO_CB(skb)->data_offset +
1407                                           skb->head - (unsigned char *)iph;
1408                         else
1409                                 tot_len = skb->len - nhoff;
1410                 } else {
1411                         if (!fixedid)
1412                                 iph->id = htons(id++);
1413                         tot_len = skb->len - nhoff;
1414                 }
1415                 iph->tot_len = htons(tot_len);
1416                 ip_send_check(iph);
1417                 if (encap)
1418                         skb_reset_inner_headers(skb);
1419                 skb->network_header = (u8 *)iph - skb->head;
1420                 skb_reset_mac_len(skb);
1421         } while ((skb = skb->next));
1422
1423 out:
1424         return segs;
1425 }
1426
1427 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1428                                         netdev_features_t features)
1429 {
1430         if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1431                 return ERR_PTR(-EINVAL);
1432
1433         return inet_gso_segment(skb, features);
1434 }
1435
1436 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1437 {
1438         const struct net_offload *ops;
1439         struct sk_buff *pp = NULL;
1440         const struct iphdr *iph;
1441         struct sk_buff *p;
1442         unsigned int hlen;
1443         unsigned int off;
1444         unsigned int id;
1445         int flush = 1;
1446         int proto;
1447
1448         off = skb_gro_offset(skb);
1449         hlen = off + sizeof(*iph);
1450         iph = skb_gro_header_fast(skb, off);
1451         if (skb_gro_header_hard(skb, hlen)) {
1452                 iph = skb_gro_header_slow(skb, hlen, off);
1453                 if (unlikely(!iph))
1454                         goto out;
1455         }
1456
1457         proto = iph->protocol;
1458
1459         ops = rcu_dereference(inet_offloads[proto]);
1460         if (!ops || !ops->callbacks.gro_receive)
1461                 goto out;
1462
1463         if (*(u8 *)iph != 0x45)
1464                 goto out;
1465
1466         if (ip_is_fragment(iph))
1467                 goto out;
1468
1469         if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1470                 goto out;
1471
1472         id = ntohl(*(__be32 *)&iph->id);
1473         flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1474         id >>= 16;
1475
1476         list_for_each_entry(p, head, list) {
1477                 struct iphdr *iph2;
1478                 u16 flush_id;
1479
1480                 if (!NAPI_GRO_CB(p)->same_flow)
1481                         continue;
1482
1483                 iph2 = (struct iphdr *)(p->data + off);
1484                 /* The above works because, with the exception of the top
1485                  * (inner most) layer, we only aggregate pkts with the same
1486                  * hdr length so all the hdrs we'll need to verify will start
1487                  * at the same offset.
1488                  */
1489                 if ((iph->protocol ^ iph2->protocol) |
1490                     ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1491                     ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1492                         NAPI_GRO_CB(p)->same_flow = 0;
1493                         continue;
1494                 }
1495
1496                 /* All fields must match except length and checksum. */
1497                 NAPI_GRO_CB(p)->flush |=
1498                         (iph->ttl ^ iph2->ttl) |
1499                         (iph->tos ^ iph2->tos) |
1500                         ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1501
1502                 NAPI_GRO_CB(p)->flush |= flush;
1503
1504                 /* We need to store of the IP ID check to be included later
1505                  * when we can verify that this packet does in fact belong
1506                  * to a given flow.
1507                  */
1508                 flush_id = (u16)(id - ntohs(iph2->id));
1509
1510                 /* This bit of code makes it much easier for us to identify
1511                  * the cases where we are doing atomic vs non-atomic IP ID
1512                  * checks.  Specifically an atomic check can return IP ID
1513                  * values 0 - 0xFFFF, while a non-atomic check can only
1514                  * return 0 or 0xFFFF.
1515                  */
1516                 if (!NAPI_GRO_CB(p)->is_atomic ||
1517                     !(iph->frag_off & htons(IP_DF))) {
1518                         flush_id ^= NAPI_GRO_CB(p)->count;
1519                         flush_id = flush_id ? 0xFFFF : 0;
1520                 }
1521
1522                 /* If the previous IP ID value was based on an atomic
1523                  * datagram we can overwrite the value and ignore it.
1524                  */
1525                 if (NAPI_GRO_CB(skb)->is_atomic)
1526                         NAPI_GRO_CB(p)->flush_id = flush_id;
1527                 else
1528                         NAPI_GRO_CB(p)->flush_id |= flush_id;
1529         }
1530
1531         NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1532         NAPI_GRO_CB(skb)->flush |= flush;
1533         skb_set_network_header(skb, off);
1534         /* The above will be needed by the transport layer if there is one
1535          * immediately following this IP hdr.
1536          */
1537
1538         /* Note : No need to call skb_gro_postpull_rcsum() here,
1539          * as we already checked checksum over ipv4 header was 0
1540          */
1541         skb_gro_pull(skb, sizeof(*iph));
1542         skb_set_transport_header(skb, skb_gro_offset(skb));
1543
1544         pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1545                                        ops->callbacks.gro_receive, head, skb);
1546
1547 out:
1548         skb_gro_flush_final(skb, pp, flush);
1549
1550         return pp;
1551 }
1552
1553 static struct sk_buff *ipip_gro_receive(struct list_head *head,
1554                                         struct sk_buff *skb)
1555 {
1556         if (NAPI_GRO_CB(skb)->encap_mark) {
1557                 NAPI_GRO_CB(skb)->flush = 1;
1558                 return NULL;
1559         }
1560
1561         NAPI_GRO_CB(skb)->encap_mark = 1;
1562
1563         return inet_gro_receive(head, skb);
1564 }
1565
1566 #define SECONDS_PER_DAY 86400
1567
1568 /* inet_current_timestamp - Return IP network timestamp
1569  *
1570  * Return milliseconds since midnight in network byte order.
1571  */
1572 __be32 inet_current_timestamp(void)
1573 {
1574         u32 secs;
1575         u32 msecs;
1576         struct timespec64 ts;
1577
1578         ktime_get_real_ts64(&ts);
1579
1580         /* Get secs since midnight. */
1581         (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1582         /* Convert to msecs. */
1583         msecs = secs * MSEC_PER_SEC;
1584         /* Convert nsec to msec. */
1585         msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1586
1587         /* Convert to network byte order. */
1588         return htonl(msecs);
1589 }
1590 EXPORT_SYMBOL(inet_current_timestamp);
1591
1592 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1593 {
1594         if (sk->sk_family == AF_INET)
1595                 return ip_recv_error(sk, msg, len, addr_len);
1596 #if IS_ENABLED(CONFIG_IPV6)
1597         if (sk->sk_family == AF_INET6)
1598                 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1599 #endif
1600         return -EINVAL;
1601 }
1602
1603 int inet_gro_complete(struct sk_buff *skb, int nhoff)
1604 {
1605         __be16 newlen = htons(skb->len - nhoff);
1606         struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1607         const struct net_offload *ops;
1608         int proto = iph->protocol;
1609         int err = -ENOSYS;
1610
1611         if (skb->encapsulation) {
1612                 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1613                 skb_set_inner_network_header(skb, nhoff);
1614         }
1615
1616         csum_replace2(&iph->check, iph->tot_len, newlen);
1617         iph->tot_len = newlen;
1618
1619         ops = rcu_dereference(inet_offloads[proto]);
1620         if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1621                 goto out;
1622
1623         /* Only need to add sizeof(*iph) to get to the next hdr below
1624          * because any hdr with option will have been flushed in
1625          * inet_gro_receive().
1626          */
1627         err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1628                               tcp4_gro_complete, udp4_gro_complete,
1629                               skb, nhoff + sizeof(*iph));
1630
1631 out:
1632         return err;
1633 }
1634
1635 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1636 {
1637         skb->encapsulation = 1;
1638         skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1639         return inet_gro_complete(skb, nhoff);
1640 }
1641
1642 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1643                          unsigned short type, unsigned char protocol,
1644                          struct net *net)
1645 {
1646         struct socket *sock;
1647         int rc = sock_create_kern(net, family, type, protocol, &sock);
1648
1649         if (rc == 0) {
1650                 *sk = sock->sk;
1651                 (*sk)->sk_allocation = GFP_ATOMIC;
1652                 /*
1653                  * Unhash it so that IP input processing does not even see it,
1654                  * we do not wish this socket to see incoming packets.
1655                  */
1656                 (*sk)->sk_prot->unhash(*sk);
1657         }
1658         return rc;
1659 }
1660 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1661
1662 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1663 {
1664         unsigned long res = 0;
1665         int i;
1666
1667         for_each_possible_cpu(i)
1668                 res += snmp_get_cpu_field(mib, i, offt);
1669         return res;
1670 }
1671 EXPORT_SYMBOL_GPL(snmp_fold_field);
1672
1673 #if BITS_PER_LONG==32
1674
1675 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1676                          size_t syncp_offset)
1677 {
1678         void *bhptr;
1679         struct u64_stats_sync *syncp;
1680         u64 v;
1681         unsigned int start;
1682
1683         bhptr = per_cpu_ptr(mib, cpu);
1684         syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1685         do {
1686                 start = u64_stats_fetch_begin_irq(syncp);
1687                 v = *(((u64 *)bhptr) + offt);
1688         } while (u64_stats_fetch_retry_irq(syncp, start));
1689
1690         return v;
1691 }
1692 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1693
1694 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1695 {
1696         u64 res = 0;
1697         int cpu;
1698
1699         for_each_possible_cpu(cpu) {
1700                 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1701         }
1702         return res;
1703 }
1704 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1705 #endif
1706
1707 #ifdef CONFIG_IP_MULTICAST
1708 static const struct net_protocol igmp_protocol = {
1709         .handler =      igmp_rcv,
1710 };
1711 #endif
1712
1713 /* thinking of making this const? Don't.
1714  * early_demux can change based on sysctl.
1715  */
1716 static struct net_protocol tcp_protocol = {
1717         .early_demux    =       tcp_v4_early_demux,
1718         .early_demux_handler =  tcp_v4_early_demux,
1719         .handler        =       tcp_v4_rcv,
1720         .err_handler    =       tcp_v4_err,
1721         .no_policy      =       1,
1722         .icmp_strict_tag_validation = 1,
1723 };
1724
1725 /* thinking of making this const? Don't.
1726  * early_demux can change based on sysctl.
1727  */
1728 static struct net_protocol udp_protocol = {
1729         .early_demux =  udp_v4_early_demux,
1730         .early_demux_handler =  udp_v4_early_demux,
1731         .handler =      udp_rcv,
1732         .err_handler =  udp_err,
1733         .no_policy =    1,
1734 };
1735
1736 static const struct net_protocol icmp_protocol = {
1737         .handler =      icmp_rcv,
1738         .err_handler =  icmp_err,
1739         .no_policy =    1,
1740 };
1741
1742 static __net_init int ipv4_mib_init_net(struct net *net)
1743 {
1744         int i;
1745
1746         net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1747         if (!net->mib.tcp_statistics)
1748                 goto err_tcp_mib;
1749         net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1750         if (!net->mib.ip_statistics)
1751                 goto err_ip_mib;
1752
1753         for_each_possible_cpu(i) {
1754                 struct ipstats_mib *af_inet_stats;
1755                 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1756                 u64_stats_init(&af_inet_stats->syncp);
1757         }
1758
1759         net->mib.net_statistics = alloc_percpu(struct linux_mib);
1760         if (!net->mib.net_statistics)
1761                 goto err_net_mib;
1762         net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1763         if (!net->mib.udp_statistics)
1764                 goto err_udp_mib;
1765         net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1766         if (!net->mib.udplite_statistics)
1767                 goto err_udplite_mib;
1768         net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1769         if (!net->mib.icmp_statistics)
1770                 goto err_icmp_mib;
1771         net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1772                                               GFP_KERNEL);
1773         if (!net->mib.icmpmsg_statistics)
1774                 goto err_icmpmsg_mib;
1775
1776         tcp_mib_init(net);
1777         return 0;
1778
1779 err_icmpmsg_mib:
1780         free_percpu(net->mib.icmp_statistics);
1781 err_icmp_mib:
1782         free_percpu(net->mib.udplite_statistics);
1783 err_udplite_mib:
1784         free_percpu(net->mib.udp_statistics);
1785 err_udp_mib:
1786         free_percpu(net->mib.net_statistics);
1787 err_net_mib:
1788         free_percpu(net->mib.ip_statistics);
1789 err_ip_mib:
1790         free_percpu(net->mib.tcp_statistics);
1791 err_tcp_mib:
1792         return -ENOMEM;
1793 }
1794
1795 static __net_exit void ipv4_mib_exit_net(struct net *net)
1796 {
1797         kfree(net->mib.icmpmsg_statistics);
1798         free_percpu(net->mib.icmp_statistics);
1799         free_percpu(net->mib.udplite_statistics);
1800         free_percpu(net->mib.udp_statistics);
1801         free_percpu(net->mib.net_statistics);
1802         free_percpu(net->mib.ip_statistics);
1803         free_percpu(net->mib.tcp_statistics);
1804 #ifdef CONFIG_MPTCP
1805         /* allocated on demand, see mptcp_init_sock() */
1806         free_percpu(net->mib.mptcp_statistics);
1807 #endif
1808 }
1809
1810 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1811         .init = ipv4_mib_init_net,
1812         .exit = ipv4_mib_exit_net,
1813 };
1814
1815 static int __init init_ipv4_mibs(void)
1816 {
1817         return register_pernet_subsys(&ipv4_mib_ops);
1818 }
1819
1820 static __net_init int inet_init_net(struct net *net)
1821 {
1822         /*
1823          * Set defaults for local port range
1824          */
1825         seqlock_init(&net->ipv4.ip_local_ports.lock);
1826         net->ipv4.ip_local_ports.range[0] =  32768;
1827         net->ipv4.ip_local_ports.range[1] =  60999;
1828
1829         seqlock_init(&net->ipv4.ping_group_range.lock);
1830         /*
1831          * Sane defaults - nobody may create ping sockets.
1832          * Boot scripts should set this to distro-specific group.
1833          */
1834         net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1835         net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1836
1837         /* Default values for sysctl-controlled parameters.
1838          * We set them here, in case sysctl is not compiled.
1839          */
1840         net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1841         net->ipv4.sysctl_ip_fwd_update_priority = 1;
1842         net->ipv4.sysctl_ip_dynaddr = 0;
1843         net->ipv4.sysctl_ip_early_demux = 1;
1844         net->ipv4.sysctl_udp_early_demux = 1;
1845         net->ipv4.sysctl_tcp_early_demux = 1;
1846         net->ipv4.sysctl_nexthop_compat_mode = 1;
1847 #ifdef CONFIG_SYSCTL
1848         net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1849 #endif
1850
1851         /* Some igmp sysctl, whose values are always used */
1852         net->ipv4.sysctl_igmp_max_memberships = 20;
1853         net->ipv4.sysctl_igmp_max_msf = 10;
1854         /* IGMP reports for link-local multicast groups are enabled by default */
1855         net->ipv4.sysctl_igmp_llm_reports = 1;
1856         net->ipv4.sysctl_igmp_qrv = 2;
1857
1858         net->ipv4.sysctl_fib_notify_on_flag_change = 0;
1859
1860         return 0;
1861 }
1862
1863 static __net_initdata struct pernet_operations af_inet_ops = {
1864         .init = inet_init_net,
1865 };
1866
1867 static int __init init_inet_pernet_ops(void)
1868 {
1869         return register_pernet_subsys(&af_inet_ops);
1870 }
1871
1872 static int ipv4_proc_init(void);
1873
1874 /*
1875  *      IP protocol layer initialiser
1876  */
1877
1878 static struct packet_offload ip_packet_offload __read_mostly = {
1879         .type = cpu_to_be16(ETH_P_IP),
1880         .callbacks = {
1881                 .gso_segment = inet_gso_segment,
1882                 .gro_receive = inet_gro_receive,
1883                 .gro_complete = inet_gro_complete,
1884         },
1885 };
1886
1887 static const struct net_offload ipip_offload = {
1888         .callbacks = {
1889                 .gso_segment    = ipip_gso_segment,
1890                 .gro_receive    = ipip_gro_receive,
1891                 .gro_complete   = ipip_gro_complete,
1892         },
1893 };
1894
1895 static int __init ipip_offload_init(void)
1896 {
1897         return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1898 }
1899
1900 static int __init ipv4_offload_init(void)
1901 {
1902         /*
1903          * Add offloads
1904          */
1905         if (udpv4_offload_init() < 0)
1906                 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1907         if (tcpv4_offload_init() < 0)
1908                 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1909         if (ipip_offload_init() < 0)
1910                 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1911
1912         dev_add_offload(&ip_packet_offload);
1913         return 0;
1914 }
1915
1916 fs_initcall(ipv4_offload_init);
1917
1918 static struct packet_type ip_packet_type __read_mostly = {
1919         .type = cpu_to_be16(ETH_P_IP),
1920         .func = ip_rcv,
1921         .list_func = ip_list_rcv,
1922 };
1923
1924 static int __init inet_init(void)
1925 {
1926         struct inet_protosw *q;
1927         struct list_head *r;
1928         int rc;
1929
1930         sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1931
1932         rc = proto_register(&tcp_prot, 1);
1933         if (rc)
1934                 goto out;
1935
1936         rc = proto_register(&udp_prot, 1);
1937         if (rc)
1938                 goto out_unregister_tcp_proto;
1939
1940         rc = proto_register(&raw_prot, 1);
1941         if (rc)
1942                 goto out_unregister_udp_proto;
1943
1944         rc = proto_register(&ping_prot, 1);
1945         if (rc)
1946                 goto out_unregister_raw_proto;
1947
1948         /*
1949          *      Tell SOCKET that we are alive...
1950          */
1951
1952         (void)sock_register(&inet_family_ops);
1953
1954 #ifdef CONFIG_SYSCTL
1955         ip_static_sysctl_init();
1956 #endif
1957
1958         /*
1959          *      Add all the base protocols.
1960          */
1961
1962         if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1963                 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1964         if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1965                 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1966         if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1967                 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1968 #ifdef CONFIG_IP_MULTICAST
1969         if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1970                 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1971 #endif
1972
1973         /* Register the socket-side information for inet_create. */
1974         for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1975                 INIT_LIST_HEAD(r);
1976
1977         for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1978                 inet_register_protosw(q);
1979
1980         /*
1981          *      Set the ARP module up
1982          */
1983
1984         arp_init();
1985
1986         /*
1987          *      Set the IP module up
1988          */
1989
1990         ip_init();
1991
1992         /* Initialise per-cpu ipv4 mibs */
1993         if (init_ipv4_mibs())
1994                 panic("%s: Cannot init ipv4 mibs\n", __func__);
1995
1996         /* Setup TCP slab cache for open requests. */
1997         tcp_init();
1998
1999         /* Setup UDP memory threshold */
2000         udp_init();
2001
2002         /* Add UDP-Lite (RFC 3828) */
2003         udplite4_register();
2004
2005         raw_init();
2006
2007         ping_init();
2008
2009         /*
2010          *      Set the ICMP layer up
2011          */
2012
2013         if (icmp_init() < 0)
2014                 panic("Failed to create the ICMP control socket.\n");
2015
2016         /*
2017          *      Initialise the multicast router
2018          */
2019 #if defined(CONFIG_IP_MROUTE)
2020         if (ip_mr_init())
2021                 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2022 #endif
2023
2024         if (init_inet_pernet_ops())
2025                 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2026
2027         ipv4_proc_init();
2028
2029         ipfrag_init();
2030
2031         dev_add_pack(&ip_packet_type);
2032
2033         ip_tunnel_core_init();
2034
2035         rc = 0;
2036 out:
2037         return rc;
2038 out_unregister_raw_proto:
2039         proto_unregister(&raw_prot);
2040 out_unregister_udp_proto:
2041         proto_unregister(&udp_prot);
2042 out_unregister_tcp_proto:
2043         proto_unregister(&tcp_prot);
2044         goto out;
2045 }
2046
2047 fs_initcall(inet_init);
2048
2049 /* ------------------------------------------------------------------------ */
2050
2051 #ifdef CONFIG_PROC_FS
2052 static int __init ipv4_proc_init(void)
2053 {
2054         int rc = 0;
2055
2056         if (raw_proc_init())
2057                 goto out_raw;
2058         if (tcp4_proc_init())
2059                 goto out_tcp;
2060         if (udp4_proc_init())
2061                 goto out_udp;
2062         if (ping_proc_init())
2063                 goto out_ping;
2064         if (ip_misc_proc_init())
2065                 goto out_misc;
2066 out:
2067         return rc;
2068 out_misc:
2069         ping_proc_exit();
2070 out_ping:
2071         udp4_proc_exit();
2072 out_udp:
2073         tcp4_proc_exit();
2074 out_tcp:
2075         raw_proc_exit();
2076 out_raw:
2077         rc = -ENOMEM;
2078         goto out;
2079 }
2080
2081 #else /* CONFIG_PROC_FS */
2082 static int __init ipv4_proc_init(void)
2083 {
2084         return 0;
2085 }
2086 #endif /* CONFIG_PROC_FS */