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_ONCE(atomic_read(&sk->sk_rmem_alloc));
 152         WARN_ON_ONCE(refcount_read(&sk->sk_wmem_alloc));
 153         WARN_ON_ONCE(sk->sk_wmem_queued);
 154         WARN_ON_ONCE(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 = READ_ONCE(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 (READ_ONCE(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         const struct proto *prot;
 562         int err;
 563
 564         if (addr_len < sizeof(uaddr->sa_family))
 565                 return -EINVAL;
 566
 567         /* IPV6_ADDRFORM can change sk->sk_prot under us. */
 568         prot = READ_ONCE(sk->sk_prot);
 569
 570         if (uaddr->sa_family == AF_UNSPEC)
 571                 return prot->disconnect(sk, flags);
 572
 573         if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
 574                 err = prot->pre_connect(sk, uaddr, addr_len);
 575                 if (err)
 576                         return err;
 577         }
 578
 579         if (data_race(!inet_sk(sk)->inet_num) && inet_autobind(sk))
 580                 return -EAGAIN;
 581         return prot->connect(sk, uaddr, addr_len);
 582 }
 583 EXPORT_SYMBOL(inet_dgram_connect);
 584
 585 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
 586 {
 587         DEFINE_WAIT_FUNC(wait, woken_wake_function);
 588
 589         add_wait_queue(sk_sleep(sk), &wait);
 590         sk->sk_write_pending += writebias;
 591
 592         /* Basic assumption: if someone sets sk->sk_err, he _must_
 593          * change state of the socket from TCP_SYN_*.
 594          * Connect() does not allow to get error notifications
 595          * without closing the socket.
 596          */
 597         while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 598                 release_sock(sk);
 599                 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
 600                 lock_sock(sk);
 601                 if (signal_pending(current) || !timeo)
 602                         break;
 603         }
 604         remove_wait_queue(sk_sleep(sk), &wait);
 605         sk->sk_write_pending -= writebias;
 606         return timeo;
 607 }
 608
 609 /*
 610  *      Connect to a remote host. There is regrettably still a little
 611  *      TCP 'magic' in here.
 612  */
 613 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 614                           int addr_len, int flags, int is_sendmsg)
 615 {
 616         struct sock *sk = sock->sk;
 617         int err;
 618         long timeo;
 619
 620         /*
 621          * uaddr can be NULL and addr_len can be 0 if:
 622          * sk is a TCP fastopen active socket and
 623          * TCP_FASTOPEN_CONNECT sockopt is set and
 624          * we already have a valid cookie for this socket.
 625          * In this case, user can call write() after connect().
 626          * write() will invoke tcp_sendmsg_fastopen() which calls
 627          * __inet_stream_connect().
 628          */
 629         if (uaddr) {
 630                 if (addr_len < sizeof(uaddr->sa_family))
 631                         return -EINVAL;
 632
 633                 if (uaddr->sa_family == AF_UNSPEC) {
 634                         err = sk->sk_prot->disconnect(sk, flags);
 635                         sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 636                         goto out;
 637                 }
 638         }
 639
 640         switch (sock->state) {
 641         default:
 642                 err = -EINVAL;
 643                 goto out;
 644         case SS_CONNECTED:
 645                 err = -EISCONN;
 646                 goto out;
 647         case SS_CONNECTING:
 648                 if (inet_sk(sk)->defer_connect)
 649                         err = is_sendmsg ? -EINPROGRESS : -EISCONN;
 650                 else
 651                         err = -EALREADY;
 652                 /* Fall out of switch with err, set for this state */
 653                 break;
 654         case SS_UNCONNECTED:
 655                 err = -EISCONN;
 656                 if (sk->sk_state != TCP_CLOSE)
 657                         goto out;
 658
 659                 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
 660                         err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
 661                         if (err)
 662                                 goto out;
 663                 }
 664
 665                 err = sk->sk_prot->connect(sk, uaddr, addr_len);
 666                 if (err < 0)
 667                         goto out;
 668
 669                 sock->state = SS_CONNECTING;
 670
 671                 if (!err && inet_sk(sk)->defer_connect)
 672                         goto out;
 673
 674                 /* Just entered SS_CONNECTING state; the only
 675                  * difference is that return value in non-blocking
 676                  * case is EINPROGRESS, rather than EALREADY.
 677                  */
 678                 err = -EINPROGRESS;
 679                 break;
 680         }
 681
 682         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
 683
 684         if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 685                 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
 686                                 tcp_sk(sk)->fastopen_req &&
 687                                 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
 688
 689                 /* Error code is set above */
 690                 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
 691                         goto out;
 692
 693                 err = sock_intr_errno(timeo);
 694                 if (signal_pending(current))
 695                         goto out;
 696         }
 697
 698         /* Connection was closed by RST, timeout, ICMP error
 699          * or another process disconnected us.
 700          */
 701         if (sk->sk_state == TCP_CLOSE)
 702                 goto sock_error;
 703
 704         /* sk->sk_err may be not zero now, if RECVERR was ordered by user
 705          * and error was received after socket entered established state.
 706          * Hence, it is handled normally after connect() return successfully.
 707          */
 708
 709         sock->state = SS_CONNECTED;
 710         err = 0;
 711 out:
 712         return err;
 713
 714 sock_error:
 715         err = sock_error(sk) ? : -ECONNABORTED;
 716         sock->state = SS_UNCONNECTED;
 717         if (sk->sk_prot->disconnect(sk, flags))
 718                 sock->state = SS_DISCONNECTING;
 719         goto out;
 720 }
 721 EXPORT_SYMBOL(__inet_stream_connect);
 722
 723 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 724                         int addr_len, int flags)
 725 {
 726         int err;
 727
 728         lock_sock(sock->sk);
 729         err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
 730         release_sock(sock->sk);
 731         return err;
 732 }
 733 EXPORT_SYMBOL(inet_stream_connect);
 734
 735 /*
 736  *      Accept a pending connection. The TCP layer now gives BSD semantics.
 737  */
 738
 739 int inet_accept(struct socket *sock, struct socket *newsock, int flags,
 740                 bool kern)
 741 {
 742         struct sock *sk1 = sock->sk, *sk2;
 743         int err = -EINVAL;
 744
 745         /* IPV6_ADDRFORM can change sk->sk_prot under us. */
 746         sk2 = READ_ONCE(sk1->sk_prot)->accept(sk1, flags, &err, kern);
 747         if (!sk2)
 748                 goto do_err;
 749
 750         lock_sock(sk2);
 751
 752         sock_rps_record_flow(sk2);
 753         WARN_ON(!((1 << sk2->sk_state) &
 754                   (TCPF_ESTABLISHED | TCPF_SYN_RECV |
 755                   TCPF_CLOSE_WAIT | TCPF_CLOSE)));
 756
 757         sock_graft(sk2, newsock);
 758
 759         newsock->state = SS_CONNECTED;
 760         err = 0;
 761         release_sock(sk2);
 762 do_err:
 763         return err;
 764 }
 765 EXPORT_SYMBOL(inet_accept);
 766
 767 /*
 768  *      This does both peername and sockname.
 769  */
 770 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
 771                  int peer)
 772 {
 773         struct sock *sk         = sock->sk;
 774         struct inet_sock *inet  = inet_sk(sk);
 775         DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
 776
 777         sin->sin_family = AF_INET;
 778         lock_sock(sk);
 779         if (peer) {
 780                 if (!inet->inet_dport ||
 781                     (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
 782                      peer == 1)) {
 783                         release_sock(sk);
 784                         return -ENOTCONN;
 785                 }
 786                 sin->sin_port = inet->inet_dport;
 787                 sin->sin_addr.s_addr = inet->inet_daddr;
 788                 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin,
 789                                        CGROUP_INET4_GETPEERNAME);
 790         } else {
 791                 __be32 addr = inet->inet_rcv_saddr;
 792                 if (!addr)
 793                         addr = inet->inet_saddr;
 794                 sin->sin_port = inet->inet_sport;
 795                 sin->sin_addr.s_addr = addr;
 796                 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin,
 797                                        CGROUP_INET4_GETSOCKNAME);
 798         }
 799         release_sock(sk);
 800         memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 801         return sizeof(*sin);
 802 }
 803 EXPORT_SYMBOL(inet_getname);
 804
 805 int inet_send_prepare(struct sock *sk)
 806 {
 807         sock_rps_record_flow(sk);
 808
 809         /* We may need to bind the socket. */
 810         if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind &&
 811             inet_autobind(sk))
 812                 return -EAGAIN;
 813
 814         return 0;
 815 }
 816 EXPORT_SYMBOL_GPL(inet_send_prepare);
 817
 818 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
 819 {
 820         struct sock *sk = sock->sk;
 821
 822         if (unlikely(inet_send_prepare(sk)))
 823                 return -EAGAIN;
 824
 825         return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
 826                                sk, msg, size);
 827 }
 828 EXPORT_SYMBOL(inet_sendmsg);
 829
 830 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
 831                       size_t size, int flags)
 832 {
 833         struct sock *sk = sock->sk;
 834         const struct proto *prot;
 835
 836         if (unlikely(inet_send_prepare(sk)))
 837                 return -EAGAIN;
 838
 839         /* IPV6_ADDRFORM can change sk->sk_prot under us. */
 840         prot = READ_ONCE(sk->sk_prot);
 841         if (prot->sendpage)
 842                 return prot->sendpage(sk, page, offset, size, flags);
 843         return sock_no_sendpage(sock, page, offset, size, flags);
 844 }
 845 EXPORT_SYMBOL(inet_sendpage);
 846
 847 INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
 848                                           size_t, int, int *));
 849 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
 850                  int flags)
 851 {
 852         struct sock *sk = sock->sk;
 853         int addr_len = 0;
 854         int err;
 855
 856         if (likely(!(flags & MSG_ERRQUEUE)))
 857                 sock_rps_record_flow(sk);
 858
 859         err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
 860                               sk, msg, size, flags, &addr_len);
 861         if (err >= 0)
 862                 msg->msg_namelen = addr_len;
 863         return err;
 864 }
 865 EXPORT_SYMBOL(inet_recvmsg);
 866
 867 int inet_shutdown(struct socket *sock, int how)
 868 {
 869         struct sock *sk = sock->sk;
 870         int err = 0;
 871
 872         /* This should really check to make sure
 873          * the socket is a TCP socket. (WHY AC...)
 874          */
 875         how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
 876                        1->2 bit 2 snds.
 877                        2->3 */
 878         if ((how & ~SHUTDOWN_MASK) || !how)     /* MAXINT->0 */
 879                 return -EINVAL;
 880
 881         lock_sock(sk);
 882         if (sock->state == SS_CONNECTING) {
 883                 if ((1 << sk->sk_state) &
 884                     (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
 885                         sock->state = SS_DISCONNECTING;
 886                 else
 887                         sock->state = SS_CONNECTED;
 888         }
 889
 890         switch (sk->sk_state) {
 891         case TCP_CLOSE:
 892                 err = -ENOTCONN;
 893                 /* Hack to wake up other listeners, who can poll for
 894                    EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
 895                 fallthrough;
 896         default:
 897                 sk->sk_shutdown |= how;
 898                 if (sk->sk_prot->shutdown)
 899                         sk->sk_prot->shutdown(sk, how);
 900                 break;
 901
 902         /* Remaining two branches are temporary solution for missing
 903          * close() in multithreaded environment. It is _not_ a good idea,
 904          * but we have no choice until close() is repaired at VFS level.
 905          */
 906         case TCP_LISTEN:
 907                 if (!(how & RCV_SHUTDOWN))
 908                         break;
 909                 fallthrough;
 910         case TCP_SYN_SENT:
 911                 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
 912                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 913                 break;
 914         }
 915
 916         /* Wake up anyone sleeping in poll. */
 917         sk->sk_state_change(sk);
 918         release_sock(sk);
 919         return err;
 920 }
 921 EXPORT_SYMBOL(inet_shutdown);
 922
 923 /*
 924  *      ioctl() calls you can issue on an INET socket. Most of these are
 925  *      device configuration and stuff and very rarely used. Some ioctls
 926  *      pass on to the socket itself.
 927  *
 928  *      NOTE: I like the idea of a module for the config stuff. ie ifconfig
 929  *      loads the devconfigure module does its configuring and unloads it.
 930  *      There's a good 20K of config code hanging around the kernel.
 931  */
 932
 933 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 934 {
 935         struct sock *sk = sock->sk;
 936         int err = 0;
 937         struct net *net = sock_net(sk);
 938         void __user *p = (void __user *)arg;
 939         struct ifreq ifr;
 940         struct rtentry rt;
 941
 942         switch (cmd) {
 943         case SIOCADDRT:
 944         case SIOCDELRT:
 945                 if (copy_from_user(&rt, p, sizeof(struct rtentry)))
 946                         return -EFAULT;
 947                 err = ip_rt_ioctl(net, cmd, &rt);
 948                 break;
 949         case SIOCRTMSG:
 950                 err = -EINVAL;
 951                 break;
 952         case SIOCDARP:
 953         case SIOCGARP:
 954         case SIOCSARP:
 955                 err = arp_ioctl(net, cmd, (void __user *)arg);
 956                 break;
 957         case SIOCGIFADDR:
 958         case SIOCGIFBRDADDR:
 959         case SIOCGIFNETMASK:
 960         case SIOCGIFDSTADDR:
 961         case SIOCGIFPFLAGS:
 962                 if (get_user_ifreq(&ifr, NULL, p))
 963                         return -EFAULT;
 964                 err = devinet_ioctl(net, cmd, &ifr);
 965                 if (!err && put_user_ifreq(&ifr, p))
 966                         err = -EFAULT;
 967                 break;
 968
 969         case SIOCSIFADDR:
 970         case SIOCSIFBRDADDR:
 971         case SIOCSIFNETMASK:
 972         case SIOCSIFDSTADDR:
 973         case SIOCSIFPFLAGS:
 974         case SIOCSIFFLAGS:
 975                 if (get_user_ifreq(&ifr, NULL, p))
 976                         return -EFAULT;
 977                 err = devinet_ioctl(net, cmd, &ifr);
 978                 break;
 979         default:
 980                 if (sk->sk_prot->ioctl)
 981                         err = sk->sk_prot->ioctl(sk, cmd, arg);
 982                 else
 983                         err = -ENOIOCTLCMD;
 984                 break;
 985         }
 986         return err;
 987 }
 988 EXPORT_SYMBOL(inet_ioctl);
 989
 990 #ifdef CONFIG_COMPAT
 991 static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd,
 992                 struct compat_rtentry __user *ur)
 993 {
 994         compat_uptr_t rtdev;
 995         struct rtentry rt;
 996
 997         if (copy_from_user(&rt.rt_dst, &ur->rt_dst,
 998                         3 * sizeof(struct sockaddr)) ||
 999             get_user(rt.rt_flags, &ur->rt_flags) ||
1000             get_user(rt.rt_metric, &ur->rt_metric) ||
1001             get_user(rt.rt_mtu, &ur->rt_mtu) ||
1002             get_user(rt.rt_window, &ur->rt_window) ||
1003             get_user(rt.rt_irtt, &ur->rt_irtt) ||
1004             get_user(rtdev, &ur->rt_dev))
1005                 return -EFAULT;
1006
1007         rt.rt_dev = compat_ptr(rtdev);
1008         return ip_rt_ioctl(sock_net(sk), cmd, &rt);
1009 }
1010
1011 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1012 {
1013         void __user *argp = compat_ptr(arg);
1014         struct sock *sk = sock->sk;
1015
1016         switch (cmd) {
1017         case SIOCADDRT:
1018         case SIOCDELRT:
1019                 return inet_compat_routing_ioctl(sk, cmd, argp);
1020         default:
1021                 if (!sk->sk_prot->compat_ioctl)
1022                         return -ENOIOCTLCMD;
1023                 return sk->sk_prot->compat_ioctl(sk, cmd, arg);
1024         }
1025 }
1026 #endif /* CONFIG_COMPAT */
1027
1028 const struct proto_ops inet_stream_ops = {
1029         .family            = PF_INET,
1030         .owner             = THIS_MODULE,
1031         .release           = inet_release,
1032         .bind              = inet_bind,
1033         .connect           = inet_stream_connect,
1034         .socketpair        = sock_no_socketpair,
1035         .accept            = inet_accept,
1036         .getname           = inet_getname,
1037         .poll              = tcp_poll,
1038         .ioctl             = inet_ioctl,
1039         .gettstamp         = sock_gettstamp,
1040         .listen            = inet_listen,
1041         .shutdown          = inet_shutdown,
1042         .setsockopt        = sock_common_setsockopt,
1043         .getsockopt        = sock_common_getsockopt,
1044         .sendmsg           = inet_sendmsg,
1045         .recvmsg           = inet_recvmsg,
1046 #ifdef CONFIG_MMU
1047         .mmap              = tcp_mmap,
1048 #endif
1049         .sendpage          = inet_sendpage,
1050         .splice_read       = tcp_splice_read,
1051         .read_sock         = tcp_read_sock,
1052         .read_skb          = tcp_read_skb,
1053         .sendmsg_locked    = tcp_sendmsg_locked,
1054         .sendpage_locked   = tcp_sendpage_locked,
1055         .peek_len          = tcp_peek_len,
1056 #ifdef CONFIG_COMPAT
1057         .compat_ioctl      = inet_compat_ioctl,
1058 #endif
1059         .set_rcvlowat      = tcp_set_rcvlowat,
1060 };
1061 EXPORT_SYMBOL(inet_stream_ops);
1062
1063 const struct proto_ops inet_dgram_ops = {
1064         .family            = PF_INET,
1065         .owner             = THIS_MODULE,
1066         .release           = inet_release,
1067         .bind              = inet_bind,
1068         .connect           = inet_dgram_connect,
1069         .socketpair        = sock_no_socketpair,
1070         .accept            = sock_no_accept,
1071         .getname           = inet_getname,
1072         .poll              = udp_poll,
1073         .ioctl             = inet_ioctl,
1074         .gettstamp         = sock_gettstamp,
1075         .listen            = sock_no_listen,
1076         .shutdown          = inet_shutdown,
1077         .setsockopt        = sock_common_setsockopt,
1078         .getsockopt        = sock_common_getsockopt,
1079         .sendmsg           = inet_sendmsg,
1080         .read_skb          = udp_read_skb,
1081         .recvmsg           = inet_recvmsg,
1082         .mmap              = sock_no_mmap,
1083         .sendpage          = inet_sendpage,
1084         .set_peek_off      = sk_set_peek_off,
1085 #ifdef CONFIG_COMPAT
1086         .compat_ioctl      = inet_compat_ioctl,
1087 #endif
1088 };
1089 EXPORT_SYMBOL(inet_dgram_ops);
1090
1091 /*
1092  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1093  * udp_poll
1094  */
1095 static const struct proto_ops inet_sockraw_ops = {
1096         .family            = PF_INET,
1097         .owner             = THIS_MODULE,
1098         .release           = inet_release,
1099         .bind              = inet_bind,
1100         .connect           = inet_dgram_connect,
1101         .socketpair        = sock_no_socketpair,
1102         .accept            = sock_no_accept,
1103         .getname           = inet_getname,
1104         .poll              = datagram_poll,
1105         .ioctl             = inet_ioctl,
1106         .gettstamp         = sock_gettstamp,
1107         .listen            = sock_no_listen,
1108         .shutdown          = inet_shutdown,
1109         .setsockopt        = sock_common_setsockopt,
1110         .getsockopt        = sock_common_getsockopt,
1111         .sendmsg           = inet_sendmsg,
1112         .recvmsg           = inet_recvmsg,
1113         .mmap              = sock_no_mmap,
1114         .sendpage          = inet_sendpage,
1115 #ifdef CONFIG_COMPAT
1116         .compat_ioctl      = inet_compat_ioctl,
1117 #endif
1118 };
1119
1120 static const struct net_proto_family inet_family_ops = {
1121         .family = PF_INET,
1122         .create = inet_create,
1123         .owner  = THIS_MODULE,
1124 };
1125
1126 /* Upon startup we insert all the elements in inetsw_array[] into
1127  * the linked list inetsw.
1128  */
1129 static struct inet_protosw inetsw_array[] =
1130 {
1131         {
1132                 .type =       SOCK_STREAM,
1133                 .protocol =   IPPROTO_TCP,
1134                 .prot =       &tcp_prot,
1135                 .ops =        &inet_stream_ops,
1136                 .flags =      INET_PROTOSW_PERMANENT |
1137                               INET_PROTOSW_ICSK,
1138         },
1139
1140         {
1141                 .type =       SOCK_DGRAM,
1142                 .protocol =   IPPROTO_UDP,
1143                 .prot =       &udp_prot,
1144                 .ops =        &inet_dgram_ops,
1145                 .flags =      INET_PROTOSW_PERMANENT,
1146        },
1147
1148        {
1149                 .type =       SOCK_DGRAM,
1150                 .protocol =   IPPROTO_ICMP,
1151                 .prot =       &ping_prot,
1152                 .ops =        &inet_sockraw_ops,
1153                 .flags =      INET_PROTOSW_REUSE,
1154        },
1155
1156        {
1157                .type =       SOCK_RAW,
1158                .protocol =   IPPROTO_IP,        /* wild card */
1159                .prot =       &raw_prot,
1160                .ops =        &inet_sockraw_ops,
1161                .flags =      INET_PROTOSW_REUSE,
1162        }
1163 };
1164
1165 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1166
1167 void inet_register_protosw(struct inet_protosw *p)
1168 {
1169         struct list_head *lh;
1170         struct inet_protosw *answer;
1171         int protocol = p->protocol;
1172         struct list_head *last_perm;
1173
1174         spin_lock_bh(&inetsw_lock);
1175
1176         if (p->type >= SOCK_MAX)
1177                 goto out_illegal;
1178
1179         /* If we are trying to override a permanent protocol, bail. */
1180         last_perm = &inetsw[p->type];
1181         list_for_each(lh, &inetsw[p->type]) {
1182                 answer = list_entry(lh, struct inet_protosw, list);
1183                 /* Check only the non-wild match. */
1184                 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1185                         break;
1186                 if (protocol == answer->protocol)
1187                         goto out_permanent;
1188                 last_perm = lh;
1189         }
1190
1191         /* Add the new entry after the last permanent entry if any, so that
1192          * the new entry does not override a permanent entry when matched with
1193          * a wild-card protocol. But it is allowed to override any existing
1194          * non-permanent entry.  This means that when we remove this entry, the
1195          * system automatically returns to the old behavior.
1196          */
1197         list_add_rcu(&p->list, last_perm);
1198 out:
1199         spin_unlock_bh(&inetsw_lock);
1200
1201         return;
1202
1203 out_permanent:
1204         pr_err("Attempt to override permanent protocol %d\n", protocol);
1205         goto out;
1206
1207 out_illegal:
1208         pr_err("Ignoring attempt to register invalid socket type %d\n",
1209                p->type);
1210         goto out;
1211 }
1212 EXPORT_SYMBOL(inet_register_protosw);
1213
1214 void inet_unregister_protosw(struct inet_protosw *p)
1215 {
1216         if (INET_PROTOSW_PERMANENT & p->flags) {
1217                 pr_err("Attempt to unregister permanent protocol %d\n",
1218                        p->protocol);
1219         } else {
1220                 spin_lock_bh(&inetsw_lock);
1221                 list_del_rcu(&p->list);
1222                 spin_unlock_bh(&inetsw_lock);
1223
1224                 synchronize_net();
1225         }
1226 }
1227 EXPORT_SYMBOL(inet_unregister_protosw);
1228
1229 static int inet_sk_reselect_saddr(struct sock *sk)
1230 {
1231         struct inet_bind_hashbucket *prev_addr_hashbucket;
1232         struct inet_sock *inet = inet_sk(sk);
1233         __be32 old_saddr = inet->inet_saddr;
1234         __be32 daddr = inet->inet_daddr;
1235         struct flowi4 *fl4;
1236         struct rtable *rt;
1237         __be32 new_saddr;
1238         struct ip_options_rcu *inet_opt;
1239         int err;
1240
1241         inet_opt = rcu_dereference_protected(inet->inet_opt,
1242                                              lockdep_sock_is_held(sk));
1243         if (inet_opt && inet_opt->opt.srr)
1244                 daddr = inet_opt->opt.faddr;
1245
1246         /* Query new route. */
1247         fl4 = &inet->cork.fl.u.ip4;
1248         rt = ip_route_connect(fl4, daddr, 0, sk->sk_bound_dev_if,
1249                               sk->sk_protocol, inet->inet_sport,
1250                               inet->inet_dport, sk);
1251         if (IS_ERR(rt))
1252                 return PTR_ERR(rt);
1253
1254         new_saddr = fl4->saddr;
1255
1256         if (new_saddr == old_saddr) {
1257                 sk_setup_caps(sk, &rt->dst);
1258                 return 0;
1259         }
1260
1261         prev_addr_hashbucket =
1262                 inet_bhashfn_portaddr(tcp_or_dccp_get_hashinfo(sk), sk,
1263                                       sock_net(sk), inet->inet_num);
1264
1265         inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1266
1267         err = inet_bhash2_update_saddr(prev_addr_hashbucket, sk);
1268         if (err) {
1269                 inet->inet_saddr = old_saddr;
1270                 inet->inet_rcv_saddr = old_saddr;
1271                 ip_rt_put(rt);
1272                 return err;
1273         }
1274
1275         sk_setup_caps(sk, &rt->dst);
1276
1277         if (READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) > 1) {
1278                 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1279                         __func__, &old_saddr, &new_saddr);
1280         }
1281
1282         /*
1283          * XXX The only one ugly spot where we need to
1284          * XXX really change the sockets identity after
1285          * XXX it has entered the hashes. -DaveM
1286          *
1287          * Besides that, it does not check for connection
1288          * uniqueness. Wait for troubles.
1289          */
1290         return __sk_prot_rehash(sk);
1291 }
1292
1293 int inet_sk_rebuild_header(struct sock *sk)
1294 {
1295         struct inet_sock *inet = inet_sk(sk);
1296         struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1297         __be32 daddr;
1298         struct ip_options_rcu *inet_opt;
1299         struct flowi4 *fl4;
1300         int err;
1301
1302         /* Route is OK, nothing to do. */
1303         if (rt)
1304                 return 0;
1305
1306         /* Reroute. */
1307         rcu_read_lock();
1308         inet_opt = rcu_dereference(inet->inet_opt);
1309         daddr = inet->inet_daddr;
1310         if (inet_opt && inet_opt->opt.srr)
1311                 daddr = inet_opt->opt.faddr;
1312         rcu_read_unlock();
1313         fl4 = &inet->cork.fl.u.ip4;
1314         rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1315                                    inet->inet_dport, inet->inet_sport,
1316                                    sk->sk_protocol, RT_CONN_FLAGS(sk),
1317                                    sk->sk_bound_dev_if);
1318         if (!IS_ERR(rt)) {
1319                 err = 0;
1320                 sk_setup_caps(sk, &rt->dst);
1321         } else {
1322                 err = PTR_ERR(rt);
1323
1324                 /* Routing failed... */
1325                 sk->sk_route_caps = 0;
1326                 /*
1327                  * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1328                  * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1329                  */
1330                 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) ||
1331                     sk->sk_state != TCP_SYN_SENT ||
1332                     (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1333                     (err = inet_sk_reselect_saddr(sk)) != 0)
1334                         sk->sk_err_soft = -err;
1335         }
1336
1337         return err;
1338 }
1339 EXPORT_SYMBOL(inet_sk_rebuild_header);
1340
1341 void inet_sk_set_state(struct sock *sk, int state)
1342 {
1343         trace_inet_sock_set_state(sk, sk->sk_state, state);
1344         sk->sk_state = state;
1345 }
1346 EXPORT_SYMBOL(inet_sk_set_state);
1347
1348 void inet_sk_state_store(struct sock *sk, int newstate)
1349 {
1350         trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1351         smp_store_release(&sk->sk_state, newstate);
1352 }
1353
1354 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1355                                  netdev_features_t features)
1356 {
1357         bool udpfrag = false, fixedid = false, gso_partial, encap;
1358         struct sk_buff *segs = ERR_PTR(-EINVAL);
1359         const struct net_offload *ops;
1360         unsigned int offset = 0;
1361         struct iphdr *iph;
1362         int proto, tot_len;
1363         int nhoff;
1364         int ihl;
1365         int id;
1366
1367         skb_reset_network_header(skb);
1368         nhoff = skb_network_header(skb) - skb_mac_header(skb);
1369         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1370                 goto out;
1371
1372         iph = ip_hdr(skb);
1373         ihl = iph->ihl * 4;
1374         if (ihl < sizeof(*iph))
1375                 goto out;
1376
1377         id = ntohs(iph->id);
1378         proto = iph->protocol;
1379
1380         /* Warning: after this point, iph might be no longer valid */
1381         if (unlikely(!pskb_may_pull(skb, ihl)))
1382                 goto out;
1383         __skb_pull(skb, ihl);
1384
1385         encap = SKB_GSO_CB(skb)->encap_level > 0;
1386         if (encap)
1387                 features &= skb->dev->hw_enc_features;
1388         SKB_GSO_CB(skb)->encap_level += ihl;
1389
1390         skb_reset_transport_header(skb);
1391
1392         segs = ERR_PTR(-EPROTONOSUPPORT);
1393
1394         if (!skb->encapsulation || encap) {
1395                 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1396                 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1397
1398                 /* fixed ID is invalid if DF bit is not set */
1399                 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1400                         goto out;
1401         }
1402
1403         ops = rcu_dereference(inet_offloads[proto]);
1404         if (likely(ops && ops->callbacks.gso_segment)) {
1405                 segs = ops->callbacks.gso_segment(skb, features);
1406                 if (!segs)
1407                         skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
1408         }
1409
1410         if (IS_ERR_OR_NULL(segs))
1411                 goto out;
1412
1413         gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1414
1415         skb = segs;
1416         do {
1417                 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1418                 if (udpfrag) {
1419                         iph->frag_off = htons(offset >> 3);
1420                         if (skb->next)
1421                                 iph->frag_off |= htons(IP_MF);
1422                         offset += skb->len - nhoff - ihl;
1423                         tot_len = skb->len - nhoff;
1424                 } else if (skb_is_gso(skb)) {
1425                         if (!fixedid) {
1426                                 iph->id = htons(id);
1427                                 id += skb_shinfo(skb)->gso_segs;
1428                         }
1429
1430                         if (gso_partial)
1431                                 tot_len = skb_shinfo(skb)->gso_size +
1432                                           SKB_GSO_CB(skb)->data_offset +
1433                                           skb->head - (unsigned char *)iph;
1434                         else
1435                                 tot_len = skb->len - nhoff;
1436                 } else {
1437                         if (!fixedid)
1438                                 iph->id = htons(id++);
1439                         tot_len = skb->len - nhoff;
1440                 }
1441                 iph->tot_len = htons(tot_len);
1442                 ip_send_check(iph);
1443                 if (encap)
1444                         skb_reset_inner_headers(skb);
1445                 skb->network_header = (u8 *)iph - skb->head;
1446                 skb_reset_mac_len(skb);
1447         } while ((skb = skb->next));
1448
1449 out:
1450         return segs;
1451 }
1452
1453 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1454                                         netdev_features_t features)
1455 {
1456         if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1457                 return ERR_PTR(-EINVAL);
1458
1459         return inet_gso_segment(skb, features);
1460 }
1461
1462 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1463 {
1464         const struct net_offload *ops;
1465         struct sk_buff *pp = NULL;
1466         const struct iphdr *iph;
1467         struct sk_buff *p;
1468         unsigned int hlen;
1469         unsigned int off;
1470         unsigned int id;
1471         int flush = 1;
1472         int proto;
1473
1474         off = skb_gro_offset(skb);
1475         hlen = off + sizeof(*iph);
1476         iph = skb_gro_header(skb, hlen, off);
1477         if (unlikely(!iph))
1478                 goto out;
1479
1480         proto = iph->protocol;
1481
1482         ops = rcu_dereference(inet_offloads[proto]);
1483         if (!ops || !ops->callbacks.gro_receive)
1484                 goto out;
1485
1486         if (*(u8 *)iph != 0x45)
1487                 goto out;
1488
1489         if (ip_is_fragment(iph))
1490                 goto out;
1491
1492         if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1493                 goto out;
1494
1495         id = ntohl(*(__be32 *)&iph->id);
1496         flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1497         id >>= 16;
1498
1499         list_for_each_entry(p, head, list) {
1500                 struct iphdr *iph2;
1501                 u16 flush_id;
1502
1503                 if (!NAPI_GRO_CB(p)->same_flow)
1504                         continue;
1505
1506                 iph2 = (struct iphdr *)(p->data + off);
1507                 /* The above works because, with the exception of the top
1508                  * (inner most) layer, we only aggregate pkts with the same
1509                  * hdr length so all the hdrs we'll need to verify will start
1510                  * at the same offset.
1511                  */
1512                 if ((iph->protocol ^ iph2->protocol) |
1513                     ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1514                     ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1515                         NAPI_GRO_CB(p)->same_flow = 0;
1516                         continue;
1517                 }
1518
1519                 /* All fields must match except length and checksum. */
1520                 NAPI_GRO_CB(p)->flush |=
1521                         (iph->ttl ^ iph2->ttl) |
1522                         (iph->tos ^ iph2->tos) |
1523                         ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1524
1525                 NAPI_GRO_CB(p)->flush |= flush;
1526
1527                 /* We need to store of the IP ID check to be included later
1528                  * when we can verify that this packet does in fact belong
1529                  * to a given flow.
1530                  */
1531                 flush_id = (u16)(id - ntohs(iph2->id));
1532
1533                 /* This bit of code makes it much easier for us to identify
1534                  * the cases where we are doing atomic vs non-atomic IP ID
1535                  * checks.  Specifically an atomic check can return IP ID
1536                  * values 0 - 0xFFFF, while a non-atomic check can only
1537                  * return 0 or 0xFFFF.
1538                  */
1539                 if (!NAPI_GRO_CB(p)->is_atomic ||
1540                     !(iph->frag_off & htons(IP_DF))) {
1541                         flush_id ^= NAPI_GRO_CB(p)->count;
1542                         flush_id = flush_id ? 0xFFFF : 0;
1543                 }
1544
1545                 /* If the previous IP ID value was based on an atomic
1546                  * datagram we can overwrite the value and ignore it.
1547                  */
1548                 if (NAPI_GRO_CB(skb)->is_atomic)
1549                         NAPI_GRO_CB(p)->flush_id = flush_id;
1550                 else
1551                         NAPI_GRO_CB(p)->flush_id |= flush_id;
1552         }
1553
1554         NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1555         NAPI_GRO_CB(skb)->flush |= flush;
1556         skb_set_network_header(skb, off);
1557         /* The above will be needed by the transport layer if there is one
1558          * immediately following this IP hdr.
1559          */
1560
1561         /* Note : No need to call skb_gro_postpull_rcsum() here,
1562          * as we already checked checksum over ipv4 header was 0
1563          */
1564         skb_gro_pull(skb, sizeof(*iph));
1565         skb_set_transport_header(skb, skb_gro_offset(skb));
1566
1567         pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1568                                        ops->callbacks.gro_receive, head, skb);
1569
1570 out:
1571         skb_gro_flush_final(skb, pp, flush);
1572
1573         return pp;
1574 }
1575
1576 static struct sk_buff *ipip_gro_receive(struct list_head *head,
1577                                         struct sk_buff *skb)
1578 {
1579         if (NAPI_GRO_CB(skb)->encap_mark) {
1580                 NAPI_GRO_CB(skb)->flush = 1;
1581                 return NULL;
1582         }
1583
1584         NAPI_GRO_CB(skb)->encap_mark = 1;
1585
1586         return inet_gro_receive(head, skb);
1587 }
1588
1589 #define SECONDS_PER_DAY 86400
1590
1591 /* inet_current_timestamp - Return IP network timestamp
1592  *
1593  * Return milliseconds since midnight in network byte order.
1594  */
1595 __be32 inet_current_timestamp(void)
1596 {
1597         u32 secs;
1598         u32 msecs;
1599         struct timespec64 ts;
1600
1601         ktime_get_real_ts64(&ts);
1602
1603         /* Get secs since midnight. */
1604         (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1605         /* Convert to msecs. */
1606         msecs = secs * MSEC_PER_SEC;
1607         /* Convert nsec to msec. */
1608         msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1609
1610         /* Convert to network byte order. */
1611         return htonl(msecs);
1612 }
1613 EXPORT_SYMBOL(inet_current_timestamp);
1614
1615 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1616 {
1617         if (sk->sk_family == AF_INET)
1618                 return ip_recv_error(sk, msg, len, addr_len);
1619 #if IS_ENABLED(CONFIG_IPV6)
1620         if (sk->sk_family == AF_INET6)
1621                 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1622 #endif
1623         return -EINVAL;
1624 }
1625
1626 int inet_gro_complete(struct sk_buff *skb, int nhoff)
1627 {
1628         __be16 newlen = htons(skb->len - nhoff);
1629         struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1630         const struct net_offload *ops;
1631         int proto = iph->protocol;
1632         int err = -ENOSYS;
1633
1634         if (skb->encapsulation) {
1635                 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1636                 skb_set_inner_network_header(skb, nhoff);
1637         }
1638
1639         csum_replace2(&iph->check, iph->tot_len, newlen);
1640         iph->tot_len = newlen;
1641
1642         ops = rcu_dereference(inet_offloads[proto]);
1643         if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1644                 goto out;
1645
1646         /* Only need to add sizeof(*iph) to get to the next hdr below
1647          * because any hdr with option will have been flushed in
1648          * inet_gro_receive().
1649          */
1650         err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1651                               tcp4_gro_complete, udp4_gro_complete,
1652                               skb, nhoff + sizeof(*iph));
1653
1654 out:
1655         return err;
1656 }
1657
1658 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1659 {
1660         skb->encapsulation = 1;
1661         skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1662         return inet_gro_complete(skb, nhoff);
1663 }
1664
1665 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1666                          unsigned short type, unsigned char protocol,
1667                          struct net *net)
1668 {
1669         struct socket *sock;
1670         int rc = sock_create_kern(net, family, type, protocol, &sock);
1671
1672         if (rc == 0) {
1673                 *sk = sock->sk;
1674                 (*sk)->sk_allocation = GFP_ATOMIC;
1675                 /*
1676                  * Unhash it so that IP input processing does not even see it,
1677                  * we do not wish this socket to see incoming packets.
1678                  */
1679                 (*sk)->sk_prot->unhash(*sk);
1680         }
1681         return rc;
1682 }
1683 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1684
1685 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1686 {
1687         unsigned long res = 0;
1688         int i;
1689
1690         for_each_possible_cpu(i)
1691                 res += snmp_get_cpu_field(mib, i, offt);
1692         return res;
1693 }
1694 EXPORT_SYMBOL_GPL(snmp_fold_field);
1695
1696 #if BITS_PER_LONG==32
1697
1698 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1699                          size_t syncp_offset)
1700 {
1701         void *bhptr;
1702         struct u64_stats_sync *syncp;
1703         u64 v;
1704         unsigned int start;
1705
1706         bhptr = per_cpu_ptr(mib, cpu);
1707         syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1708         do {
1709                 start = u64_stats_fetch_begin_irq(syncp);
1710                 v = *(((u64 *)bhptr) + offt);
1711         } while (u64_stats_fetch_retry_irq(syncp, start));
1712
1713         return v;
1714 }
1715 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1716
1717 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1718 {
1719         u64 res = 0;
1720         int cpu;
1721
1722         for_each_possible_cpu(cpu) {
1723                 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1724         }
1725         return res;
1726 }
1727 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1728 #endif
1729
1730 #ifdef CONFIG_IP_MULTICAST
1731 static const struct net_protocol igmp_protocol = {
1732         .handler =      igmp_rcv,
1733 };
1734 #endif
1735
1736 static const struct net_protocol tcp_protocol = {
1737         .handler        =       tcp_v4_rcv,
1738         .err_handler    =       tcp_v4_err,
1739         .no_policy      =       1,
1740         .icmp_strict_tag_validation = 1,
1741 };
1742
1743 static const struct net_protocol udp_protocol = {
1744         .handler =      udp_rcv,
1745         .err_handler =  udp_err,
1746         .no_policy =    1,
1747 };
1748
1749 static const struct net_protocol icmp_protocol = {
1750         .handler =      icmp_rcv,
1751         .err_handler =  icmp_err,
1752         .no_policy =    1,
1753 };
1754
1755 static __net_init int ipv4_mib_init_net(struct net *net)
1756 {
1757         int i;
1758
1759         net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1760         if (!net->mib.tcp_statistics)
1761                 goto err_tcp_mib;
1762         net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1763         if (!net->mib.ip_statistics)
1764                 goto err_ip_mib;
1765
1766         for_each_possible_cpu(i) {
1767                 struct ipstats_mib *af_inet_stats;
1768                 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1769                 u64_stats_init(&af_inet_stats->syncp);
1770         }
1771
1772         net->mib.net_statistics = alloc_percpu(struct linux_mib);
1773         if (!net->mib.net_statistics)
1774                 goto err_net_mib;
1775         net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1776         if (!net->mib.udp_statistics)
1777                 goto err_udp_mib;
1778         net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1779         if (!net->mib.udplite_statistics)
1780                 goto err_udplite_mib;
1781         net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1782         if (!net->mib.icmp_statistics)
1783                 goto err_icmp_mib;
1784         net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1785                                               GFP_KERNEL);
1786         if (!net->mib.icmpmsg_statistics)
1787                 goto err_icmpmsg_mib;
1788
1789         tcp_mib_init(net);
1790         return 0;
1791
1792 err_icmpmsg_mib:
1793         free_percpu(net->mib.icmp_statistics);
1794 err_icmp_mib:
1795         free_percpu(net->mib.udplite_statistics);
1796 err_udplite_mib:
1797         free_percpu(net->mib.udp_statistics);
1798 err_udp_mib:
1799         free_percpu(net->mib.net_statistics);
1800 err_net_mib:
1801         free_percpu(net->mib.ip_statistics);
1802 err_ip_mib:
1803         free_percpu(net->mib.tcp_statistics);
1804 err_tcp_mib:
1805         return -ENOMEM;
1806 }
1807
1808 static __net_exit void ipv4_mib_exit_net(struct net *net)
1809 {
1810         kfree(net->mib.icmpmsg_statistics);
1811         free_percpu(net->mib.icmp_statistics);
1812         free_percpu(net->mib.udplite_statistics);
1813         free_percpu(net->mib.udp_statistics);
1814         free_percpu(net->mib.net_statistics);
1815         free_percpu(net->mib.ip_statistics);
1816         free_percpu(net->mib.tcp_statistics);
1817 #ifdef CONFIG_MPTCP
1818         /* allocated on demand, see mptcp_init_sock() */
1819         free_percpu(net->mib.mptcp_statistics);
1820 #endif
1821 }
1822
1823 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1824         .init = ipv4_mib_init_net,
1825         .exit = ipv4_mib_exit_net,
1826 };
1827
1828 static int __init init_ipv4_mibs(void)
1829 {
1830         return register_pernet_subsys(&ipv4_mib_ops);
1831 }
1832
1833 static __net_init int inet_init_net(struct net *net)
1834 {
1835         /*
1836          * Set defaults for local port range
1837          */
1838         seqlock_init(&net->ipv4.ip_local_ports.lock);
1839         net->ipv4.ip_local_ports.range[0] =  32768;
1840         net->ipv4.ip_local_ports.range[1] =  60999;
1841
1842         seqlock_init(&net->ipv4.ping_group_range.lock);
1843         /*
1844          * Sane defaults - nobody may create ping sockets.
1845          * Boot scripts should set this to distro-specific group.
1846          */
1847         net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1848         net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1849
1850         /* Default values for sysctl-controlled parameters.
1851          * We set them here, in case sysctl is not compiled.
1852          */
1853         net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1854         net->ipv4.sysctl_ip_fwd_update_priority = 1;
1855         net->ipv4.sysctl_ip_dynaddr = 0;
1856         net->ipv4.sysctl_ip_early_demux = 1;
1857         net->ipv4.sysctl_udp_early_demux = 1;
1858         net->ipv4.sysctl_tcp_early_demux = 1;
1859         net->ipv4.sysctl_nexthop_compat_mode = 1;
1860 #ifdef CONFIG_SYSCTL
1861         net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1862 #endif
1863
1864         /* Some igmp sysctl, whose values are always used */
1865         net->ipv4.sysctl_igmp_max_memberships = 20;
1866         net->ipv4.sysctl_igmp_max_msf = 10;
1867         /* IGMP reports for link-local multicast groups are enabled by default */
1868         net->ipv4.sysctl_igmp_llm_reports = 1;
1869         net->ipv4.sysctl_igmp_qrv = 2;
1870
1871         net->ipv4.sysctl_fib_notify_on_flag_change = 0;
1872
1873         return 0;
1874 }
1875
1876 static __net_initdata struct pernet_operations af_inet_ops = {
1877         .init = inet_init_net,
1878 };
1879
1880 static int __init init_inet_pernet_ops(void)
1881 {
1882         return register_pernet_subsys(&af_inet_ops);
1883 }
1884
1885 static int ipv4_proc_init(void);
1886
1887 /*
1888  *      IP protocol layer initialiser
1889  */
1890
1891 static struct packet_offload ip_packet_offload __read_mostly = {
1892         .type = cpu_to_be16(ETH_P_IP),
1893         .callbacks = {
1894                 .gso_segment = inet_gso_segment,
1895                 .gro_receive = inet_gro_receive,
1896                 .gro_complete = inet_gro_complete,
1897         },
1898 };
1899
1900 static const struct net_offload ipip_offload = {
1901         .callbacks = {
1902                 .gso_segment    = ipip_gso_segment,
1903                 .gro_receive    = ipip_gro_receive,
1904                 .gro_complete   = ipip_gro_complete,
1905         },
1906 };
1907
1908 static int __init ipip_offload_init(void)
1909 {
1910         return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1911 }
1912
1913 static int __init ipv4_offload_init(void)
1914 {
1915         /*
1916          * Add offloads
1917          */
1918         if (udpv4_offload_init() < 0)
1919                 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1920         if (tcpv4_offload_init() < 0)
1921                 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1922         if (ipip_offload_init() < 0)
1923                 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1924
1925         dev_add_offload(&ip_packet_offload);
1926         return 0;
1927 }
1928
1929 fs_initcall(ipv4_offload_init);
1930
1931 static struct packet_type ip_packet_type __read_mostly = {
1932         .type = cpu_to_be16(ETH_P_IP),
1933         .func = ip_rcv,
1934         .list_func = ip_list_rcv,
1935 };
1936
1937 static int __init inet_init(void)
1938 {
1939         struct inet_protosw *q;
1940         struct list_head *r;
1941         int rc;
1942
1943         sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1944
1945         raw_hashinfo_init(&raw_v4_hashinfo);
1946
1947         rc = proto_register(&tcp_prot, 1);
1948         if (rc)
1949                 goto out;
1950
1951         rc = proto_register(&udp_prot, 1);
1952         if (rc)
1953                 goto out_unregister_tcp_proto;
1954
1955         rc = proto_register(&raw_prot, 1);
1956         if (rc)
1957                 goto out_unregister_udp_proto;
1958
1959         rc = proto_register(&ping_prot, 1);
1960         if (rc)
1961                 goto out_unregister_raw_proto;
1962
1963         /*
1964          *      Tell SOCKET that we are alive...
1965          */
1966
1967         (void)sock_register(&inet_family_ops);
1968
1969 #ifdef CONFIG_SYSCTL
1970         ip_static_sysctl_init();
1971 #endif
1972
1973         /*
1974          *      Add all the base protocols.
1975          */
1976
1977         if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1978                 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1979         if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1980                 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1981         if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1982                 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1983 #ifdef CONFIG_IP_MULTICAST
1984         if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1985                 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1986 #endif
1987
1988         /* Register the socket-side information for inet_create. */
1989         for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1990                 INIT_LIST_HEAD(r);
1991
1992         for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1993                 inet_register_protosw(q);
1994
1995         /*
1996          *      Set the ARP module up
1997          */
1998
1999         arp_init();
2000
2001         /*
2002          *      Set the IP module up
2003          */
2004
2005         ip_init();
2006
2007         /* Initialise per-cpu ipv4 mibs */
2008         if (init_ipv4_mibs())
2009                 panic("%s: Cannot init ipv4 mibs\n", __func__);
2010
2011         /* Setup TCP slab cache for open requests. */
2012         tcp_init();
2013
2014         /* Setup UDP memory threshold */
2015         udp_init();
2016
2017         /* Add UDP-Lite (RFC 3828) */
2018         udplite4_register();
2019
2020         raw_init();
2021
2022         ping_init();
2023
2024         /*
2025          *      Set the ICMP layer up
2026          */
2027
2028         if (icmp_init() < 0)
2029                 panic("Failed to create the ICMP control socket.\n");
2030
2031         /*
2032          *      Initialise the multicast router
2033          */
2034 #if defined(CONFIG_IP_MROUTE)
2035         if (ip_mr_init())
2036                 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2037 #endif
2038
2039         if (init_inet_pernet_ops())
2040                 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2041
2042         ipv4_proc_init();
2043
2044         ipfrag_init();
2045
2046         dev_add_pack(&ip_packet_type);
2047
2048         ip_tunnel_core_init();
2049
2050         rc = 0;
2051 out:
2052         return rc;
2053 out_unregister_raw_proto:
2054         proto_unregister(&raw_prot);
2055 out_unregister_udp_proto:
2056         proto_unregister(&udp_prot);
2057 out_unregister_tcp_proto:
2058         proto_unregister(&tcp_prot);
2059         goto out;
2060 }
2061
2062 fs_initcall(inet_init);
2063
2064 /* ------------------------------------------------------------------------ */
2065
2066 #ifdef CONFIG_PROC_FS
2067 static int __init ipv4_proc_init(void)
2068 {
2069         int rc = 0;
2070
2071         if (raw_proc_init())
2072                 goto out_raw;
2073         if (tcp4_proc_init())
2074                 goto out_tcp;
2075         if (udp4_proc_init())
2076                 goto out_udp;
2077         if (ping_proc_init())
2078                 goto out_ping;
2079         if (ip_misc_proc_init())
2080                 goto out_misc;
2081 out:
2082         return rc;
2083 out_misc:
2084         ping_proc_exit();
2085 out_ping:
2086         udp4_proc_exit();
2087 out_udp:
2088         tcp4_proc_exit();
2089 out_tcp:
2090         raw_proc_exit();
2091 out_raw:
2092         rc = -ENOMEM;
2093         goto out;
2094 }
2095
2096 #else /* CONFIG_PROC_FS */
2097 static int __init ipv4_proc_init(void)
2098 {
2099         return 0;
2100 }
2101 #endif /* CONFIG_PROC_FS */