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         if (INET_PROTOSW_ICSK & answer_flags)
 328                 inet_init_csk_locks(sk);
 329
 330         inet = inet_sk(sk);
 331         inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
 332
 333         inet->nodefrag = 0;
 334
 335         if (SOCK_RAW == sock->type) {
 336                 inet->inet_num = protocol;
 337                 if (IPPROTO_RAW == protocol)
 338                         inet->hdrincl = 1;
 339         }
 340
 341         if (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc))
 342                 inet->pmtudisc = IP_PMTUDISC_DONT;
 343         else
 344                 inet->pmtudisc = IP_PMTUDISC_WANT;
 345
 346         atomic_set(&inet->inet_id, 0);
 347
 348         sock_init_data(sock, sk);
 349
 350         sk->sk_destruct    = inet_sock_destruct;
 351         sk->sk_protocol    = protocol;
 352         sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
 353         sk->sk_txrehash = READ_ONCE(net->core.sysctl_txrehash);
 354
 355         inet->uc_ttl    = -1;
 356         inet->mc_loop   = 1;
 357         inet->mc_ttl    = 1;
 358         inet->mc_all    = 1;
 359         inet->mc_index  = 0;
 360         inet->mc_list   = NULL;
 361         inet->rcv_tos   = 0;
 362
 363         sk_refcnt_debug_inc(sk);
 364
 365         if (inet->inet_num) {
 366                 /* It assumes that any protocol which allows
 367                  * the user to assign a number at socket
 368                  * creation time automatically
 369                  * shares.
 370                  */
 371                 inet->inet_sport = htons(inet->inet_num);
 372                 /* Add to protocol hash chains. */
 373                 err = sk->sk_prot->hash(sk);
 374                 if (err) {
 375                         sk_common_release(sk);
 376                         goto out;
 377                 }
 378         }
 379
 380         if (sk->sk_prot->init) {
 381                 err = sk->sk_prot->init(sk);
 382                 if (err) {
 383                         sk_common_release(sk);
 384                         goto out;
 385                 }
 386         }
 387
 388         if (!kern) {
 389                 err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
 390                 if (err) {
 391                         sk_common_release(sk);
 392                         goto out;
 393                 }
 394         }
 395 out:
 396         return err;
 397 out_rcu_unlock:
 398         rcu_read_unlock();
 399         goto out;
 400 }
 401
 402
 403 /*
 404  *      The peer socket should always be NULL (or else). When we call this
 405  *      function we are destroying the object and from then on nobody
 406  *      should refer to it.
 407  */
 408 int inet_release(struct socket *sock)
 409 {
 410         struct sock *sk = sock->sk;
 411
 412         if (sk) {
 413                 long timeout;
 414
 415                 if (!sk->sk_kern_sock)
 416                         BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk);
 417
 418                 /* Applications forget to leave groups before exiting */
 419                 ip_mc_drop_socket(sk);
 420
 421                 /* If linger is set, we don't return until the close
 422                  * is complete.  Otherwise we return immediately. The
 423                  * actually closing is done the same either way.
 424                  *
 425                  * If the close is due to the process exiting, we never
 426                  * linger..
 427                  */
 428                 timeout = 0;
 429                 if (sock_flag(sk, SOCK_LINGER) &&
 430                     !(current->flags & PF_EXITING))
 431                         timeout = sk->sk_lingertime;
 432                 sk->sk_prot->close(sk, timeout);
 433                 sock->sk = NULL;
 434         }
 435         return 0;
 436 }
 437 EXPORT_SYMBOL(inet_release);
 438
 439 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 440 {
 441         struct sock *sk = sock->sk;
 442         u32 flags = BIND_WITH_LOCK;
 443         int err;
 444
 445         /* If the socket has its own bind function then use it. (RAW) */
 446         if (sk->sk_prot->bind) {
 447                 return sk->sk_prot->bind(sk, uaddr, addr_len);
 448         }
 449         if (addr_len < sizeof(struct sockaddr_in))
 450                 return -EINVAL;
 451
 452         /* BPF prog is run before any checks are done so that if the prog
 453          * changes context in a wrong way it will be caught.
 454          */
 455         err = BPF_CGROUP_RUN_PROG_INET_BIND_LOCK(sk, uaddr,
 456                                                  CGROUP_INET4_BIND, &flags);
 457         if (err)
 458                 return err;
 459
 460         return __inet_bind(sk, uaddr, addr_len, flags);
 461 }
 462 EXPORT_SYMBOL(inet_bind);
 463
 464 int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
 465                 u32 flags)
 466 {
 467         struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
 468         struct inet_sock *inet = inet_sk(sk);
 469         struct net *net = sock_net(sk);
 470         unsigned short snum;
 471         int chk_addr_ret;
 472         u32 tb_id = RT_TABLE_LOCAL;
 473         int err;
 474
 475         if (addr->sin_family != AF_INET) {
 476                 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
 477                  * only if s_addr is INADDR_ANY.
 478                  */
 479                 err = -EAFNOSUPPORT;
 480                 if (addr->sin_family != AF_UNSPEC ||
 481                     addr->sin_addr.s_addr != htonl(INADDR_ANY))
 482                         goto out;
 483         }
 484
 485         tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
 486         chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
 487
 488         /* Not specified by any standard per-se, however it breaks too
 489          * many applications when removed.  It is unfortunate since
 490          * allowing applications to make a non-local bind solves
 491          * several problems with systems using dynamic addressing.
 492          * (ie. your servers still start up even if your ISDN link
 493          *  is temporarily down)
 494          */
 495         err = -EADDRNOTAVAIL;
 496         if (!inet_addr_valid_or_nonlocal(net, inet, addr->sin_addr.s_addr,
 497                                          chk_addr_ret))
 498                 goto out;
 499
 500         snum = ntohs(addr->sin_port);
 501         err = -EACCES;
 502         if (!(flags & BIND_NO_CAP_NET_BIND_SERVICE) &&
 503             snum && inet_port_requires_bind_service(net, snum) &&
 504             !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
 505                 goto out;
 506
 507         /*      We keep a pair of addresses. rcv_saddr is the one
 508          *      used by hash lookups, and saddr is used for transmit.
 509          *
 510          *      In the BSD API these are the same except where it
 511          *      would be illegal to use them (multicast/broadcast) in
 512          *      which case the sending device address is used.
 513          */
 514         if (flags & BIND_WITH_LOCK)
 515                 lock_sock(sk);
 516
 517         /* Check these errors (active socket, double bind). */
 518         err = -EINVAL;
 519         if (sk->sk_state != TCP_CLOSE || inet->inet_num)
 520                 goto out_release_sock;
 521
 522         inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
 523         if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
 524                 inet->inet_saddr = 0;  /* Use device */
 525
 526         /* Make sure we are allowed to bind here. */
 527         if (snum || !(inet->bind_address_no_port ||
 528                       (flags & BIND_FORCE_ADDRESS_NO_PORT))) {
 529                 err = sk->sk_prot->get_port(sk, snum);
 530                 if (err) {
 531                         inet->inet_saddr = inet->inet_rcv_saddr = 0;
 532                         goto out_release_sock;
 533                 }
 534                 if (!(flags & BIND_FROM_BPF)) {
 535                         err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
 536                         if (err) {
 537                                 inet->inet_saddr = inet->inet_rcv_saddr = 0;
 538                                 if (sk->sk_prot->put_port)
 539                                         sk->sk_prot->put_port(sk);
 540                                 goto out_release_sock;
 541                         }
 542                 }
 543         }
 544
 545         if (inet->inet_rcv_saddr)
 546                 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
 547         if (snum)
 548                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
 549         inet->inet_sport = htons(inet->inet_num);
 550         inet->inet_daddr = 0;
 551         inet->inet_dport = 0;
 552         sk_dst_reset(sk);
 553         err = 0;
 554 out_release_sock:
 555         if (flags & BIND_WITH_LOCK)
 556                 release_sock(sk);
 557 out:
 558         return err;
 559 }
 560
 561 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
 562                        int addr_len, int flags)
 563 {
 564         struct sock *sk = sock->sk;
 565         const struct proto *prot;
 566         int err;
 567
 568         if (addr_len < sizeof(uaddr->sa_family))
 569                 return -EINVAL;
 570
 571         /* IPV6_ADDRFORM can change sk->sk_prot under us. */
 572         prot = READ_ONCE(sk->sk_prot);
 573
 574         if (uaddr->sa_family == AF_UNSPEC)
 575                 return prot->disconnect(sk, flags);
 576
 577         if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
 578                 err = prot->pre_connect(sk, uaddr, addr_len);
 579                 if (err)
 580                         return err;
 581         }
 582
 583         if (data_race(!inet_sk(sk)->inet_num) && inet_autobind(sk))
 584                 return -EAGAIN;
 585         return prot->connect(sk, uaddr, addr_len);
 586 }
 587 EXPORT_SYMBOL(inet_dgram_connect);
 588
 589 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
 590 {
 591         DEFINE_WAIT_FUNC(wait, woken_wake_function);
 592
 593         add_wait_queue(sk_sleep(sk), &wait);
 594         sk->sk_write_pending += writebias;
 595
 596         /* Basic assumption: if someone sets sk->sk_err, he _must_
 597          * change state of the socket from TCP_SYN_*.
 598          * Connect() does not allow to get error notifications
 599          * without closing the socket.
 600          */
 601         while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 602                 release_sock(sk);
 603                 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
 604                 lock_sock(sk);
 605                 if (signal_pending(current) || !timeo)
 606                         break;
 607         }
 608         remove_wait_queue(sk_sleep(sk), &wait);
 609         sk->sk_write_pending -= writebias;
 610         return timeo;
 611 }
 612
 613 /*
 614  *      Connect to a remote host. There is regrettably still a little
 615  *      TCP 'magic' in here.
 616  */
 617 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 618                           int addr_len, int flags, int is_sendmsg)
 619 {
 620         struct sock *sk = sock->sk;
 621         int err;
 622         long timeo;
 623
 624         /*
 625          * uaddr can be NULL and addr_len can be 0 if:
 626          * sk is a TCP fastopen active socket and
 627          * TCP_FASTOPEN_CONNECT sockopt is set and
 628          * we already have a valid cookie for this socket.
 629          * In this case, user can call write() after connect().
 630          * write() will invoke tcp_sendmsg_fastopen() which calls
 631          * __inet_stream_connect().
 632          */
 633         if (uaddr) {
 634                 if (addr_len < sizeof(uaddr->sa_family))
 635                         return -EINVAL;
 636
 637                 if (uaddr->sa_family == AF_UNSPEC) {
 638                         sk->sk_disconnects++;
 639                         err = sk->sk_prot->disconnect(sk, flags);
 640                         sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 641                         goto out;
 642                 }
 643         }
 644
 645         switch (sock->state) {
 646         default:
 647                 err = -EINVAL;
 648                 goto out;
 649         case SS_CONNECTED:
 650                 err = -EISCONN;
 651                 goto out;
 652         case SS_CONNECTING:
 653                 if (inet_sk(sk)->defer_connect)
 654                         err = is_sendmsg ? -EINPROGRESS : -EISCONN;
 655                 else
 656                         err = -EALREADY;
 657                 /* Fall out of switch with err, set for this state */
 658                 break;
 659         case SS_UNCONNECTED:
 660                 err = -EISCONN;
 661                 if (sk->sk_state != TCP_CLOSE)
 662                         goto out;
 663
 664                 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
 665                         err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
 666                         if (err)
 667                                 goto out;
 668                 }
 669
 670                 err = sk->sk_prot->connect(sk, uaddr, addr_len);
 671                 if (err < 0)
 672                         goto out;
 673
 674                 sock->state = SS_CONNECTING;
 675
 676                 if (!err && inet_sk(sk)->defer_connect)
 677                         goto out;
 678
 679                 /* Just entered SS_CONNECTING state; the only
 680                  * difference is that return value in non-blocking
 681                  * case is EINPROGRESS, rather than EALREADY.
 682                  */
 683                 err = -EINPROGRESS;
 684                 break;
 685         }
 686
 687         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
 688
 689         if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 690                 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
 691                                 tcp_sk(sk)->fastopen_req &&
 692                                 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
 693                 int dis = sk->sk_disconnects;
 694
 695                 /* Error code is set above */
 696                 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
 697                         goto out;
 698
 699                 err = sock_intr_errno(timeo);
 700                 if (signal_pending(current))
 701                         goto out;
 702
 703                 if (dis != sk->sk_disconnects) {
 704                         err = -EPIPE;
 705                         goto out;
 706                 }
 707         }
 708
 709         /* Connection was closed by RST, timeout, ICMP error
 710          * or another process disconnected us.
 711          */
 712         if (sk->sk_state == TCP_CLOSE)
 713                 goto sock_error;
 714
 715         /* sk->sk_err may be not zero now, if RECVERR was ordered by user
 716          * and error was received after socket entered established state.
 717          * Hence, it is handled normally after connect() return successfully.
 718          */
 719
 720         sock->state = SS_CONNECTED;
 721         err = 0;
 722 out:
 723         return err;
 724
 725 sock_error:
 726         err = sock_error(sk) ? : -ECONNABORTED;
 727         sock->state = SS_UNCONNECTED;
 728         sk->sk_disconnects++;
 729         if (sk->sk_prot->disconnect(sk, flags))
 730                 sock->state = SS_DISCONNECTING;
 731         goto out;
 732 }
 733 EXPORT_SYMBOL(__inet_stream_connect);
 734
 735 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 736                         int addr_len, int flags)
 737 {
 738         int err;
 739
 740         lock_sock(sock->sk);
 741         err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
 742         release_sock(sock->sk);
 743         return err;
 744 }
 745 EXPORT_SYMBOL(inet_stream_connect);
 746
 747 /*
 748  *      Accept a pending connection. The TCP layer now gives BSD semantics.
 749  */
 750
 751 int inet_accept(struct socket *sock, struct socket *newsock, int flags,
 752                 bool kern)
 753 {
 754         struct sock *sk1 = sock->sk, *sk2;
 755         int err = -EINVAL;
 756
 757         /* IPV6_ADDRFORM can change sk->sk_prot under us. */
 758         sk2 = READ_ONCE(sk1->sk_prot)->accept(sk1, flags, &err, kern);
 759         if (!sk2)
 760                 goto do_err;
 761
 762         lock_sock(sk2);
 763
 764         sock_rps_record_flow(sk2);
 765         WARN_ON(!((1 << sk2->sk_state) &
 766                   (TCPF_ESTABLISHED | TCPF_SYN_RECV |
 767                   TCPF_CLOSE_WAIT | TCPF_CLOSE)));
 768
 769         if (test_bit(SOCK_SUPPORT_ZC, &sock->flags))
 770                 set_bit(SOCK_SUPPORT_ZC, &newsock->flags);
 771         sock_graft(sk2, newsock);
 772
 773         newsock->state = SS_CONNECTED;
 774         err = 0;
 775         release_sock(sk2);
 776 do_err:
 777         return err;
 778 }
 779 EXPORT_SYMBOL(inet_accept);
 780
 781 /*
 782  *      This does both peername and sockname.
 783  */
 784 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
 785                  int peer)
 786 {
 787         struct sock *sk         = sock->sk;
 788         struct inet_sock *inet  = inet_sk(sk);
 789         DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
 790
 791         sin->sin_family = AF_INET;
 792         lock_sock(sk);
 793         if (peer) {
 794                 if (!inet->inet_dport ||
 795                     (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
 796                      peer == 1)) {
 797                         release_sock(sk);
 798                         return -ENOTCONN;
 799                 }
 800                 sin->sin_port = inet->inet_dport;
 801                 sin->sin_addr.s_addr = inet->inet_daddr;
 802                 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin,
 803                                        CGROUP_INET4_GETPEERNAME);
 804         } else {
 805                 __be32 addr = inet->inet_rcv_saddr;
 806                 if (!addr)
 807                         addr = inet->inet_saddr;
 808                 sin->sin_port = inet->inet_sport;
 809                 sin->sin_addr.s_addr = addr;
 810                 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin,
 811                                        CGROUP_INET4_GETSOCKNAME);
 812         }
 813         release_sock(sk);
 814         memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 815         return sizeof(*sin);
 816 }
 817 EXPORT_SYMBOL(inet_getname);
 818
 819 int inet_send_prepare(struct sock *sk)
 820 {
 821         sock_rps_record_flow(sk);
 822
 823         /* We may need to bind the socket. */
 824         if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind &&
 825             inet_autobind(sk))
 826                 return -EAGAIN;
 827
 828         return 0;
 829 }
 830 EXPORT_SYMBOL_GPL(inet_send_prepare);
 831
 832 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
 833 {
 834         struct sock *sk = sock->sk;
 835
 836         if (unlikely(inet_send_prepare(sk)))
 837                 return -EAGAIN;
 838
 839         return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
 840                                sk, msg, size);
 841 }
 842 EXPORT_SYMBOL(inet_sendmsg);
 843
 844 void inet_splice_eof(struct socket *sock)
 845 {
 846         const struct proto *prot;
 847         struct sock *sk = sock->sk;
 848
 849         if (unlikely(inet_send_prepare(sk)))
 850                 return;
 851
 852         /* IPV6_ADDRFORM can change sk->sk_prot under us. */
 853         prot = READ_ONCE(sk->sk_prot);
 854         if (prot->splice_eof)
 855                 prot->splice_eof(sock);
 856 }
 857 EXPORT_SYMBOL_GPL(inet_splice_eof);
 858
 859 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
 860                       size_t size, int flags)
 861 {
 862         struct sock *sk = sock->sk;
 863         const struct proto *prot;
 864
 865         if (unlikely(inet_send_prepare(sk)))
 866                 return -EAGAIN;
 867
 868         /* IPV6_ADDRFORM can change sk->sk_prot under us. */
 869         prot = READ_ONCE(sk->sk_prot);
 870         if (prot->sendpage)
 871                 return prot->sendpage(sk, page, offset, size, flags);
 872         return sock_no_sendpage(sock, page, offset, size, flags);
 873 }
 874 EXPORT_SYMBOL(inet_sendpage);
 875
 876 INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
 877                                           size_t, int, int *));
 878 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
 879                  int flags)
 880 {
 881         struct sock *sk = sock->sk;
 882         int addr_len = 0;
 883         int err;
 884
 885         if (likely(!(flags & MSG_ERRQUEUE)))
 886                 sock_rps_record_flow(sk);
 887
 888         err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
 889                               sk, msg, size, flags, &addr_len);
 890         if (err >= 0)
 891                 msg->msg_namelen = addr_len;
 892         return err;
 893 }
 894 EXPORT_SYMBOL(inet_recvmsg);
 895
 896 int inet_shutdown(struct socket *sock, int how)
 897 {
 898         struct sock *sk = sock->sk;
 899         int err = 0;
 900
 901         /* This should really check to make sure
 902          * the socket is a TCP socket. (WHY AC...)
 903          */
 904         how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
 905                        1->2 bit 2 snds.
 906                        2->3 */
 907         if ((how & ~SHUTDOWN_MASK) || !how)     /* MAXINT->0 */
 908                 return -EINVAL;
 909
 910         lock_sock(sk);
 911         if (sock->state == SS_CONNECTING) {
 912                 if ((1 << sk->sk_state) &
 913                     (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
 914                         sock->state = SS_DISCONNECTING;
 915                 else
 916                         sock->state = SS_CONNECTED;
 917         }
 918
 919         switch (sk->sk_state) {
 920         case TCP_CLOSE:
 921                 err = -ENOTCONN;
 922                 /* Hack to wake up other listeners, who can poll for
 923                    EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
 924                 fallthrough;
 925         default:
 926                 WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | how);
 927                 if (sk->sk_prot->shutdown)
 928                         sk->sk_prot->shutdown(sk, how);
 929                 break;
 930
 931         /* Remaining two branches are temporary solution for missing
 932          * close() in multithreaded environment. It is _not_ a good idea,
 933          * but we have no choice until close() is repaired at VFS level.
 934          */
 935         case TCP_LISTEN:
 936                 if (!(how & RCV_SHUTDOWN))
 937                         break;
 938                 fallthrough;
 939         case TCP_SYN_SENT:
 940                 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
 941                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 942                 break;
 943         }
 944
 945         /* Wake up anyone sleeping in poll. */
 946         sk->sk_state_change(sk);
 947         release_sock(sk);
 948         return err;
 949 }
 950 EXPORT_SYMBOL(inet_shutdown);
 951
 952 /*
 953  *      ioctl() calls you can issue on an INET socket. Most of these are
 954  *      device configuration and stuff and very rarely used. Some ioctls
 955  *      pass on to the socket itself.
 956  *
 957  *      NOTE: I like the idea of a module for the config stuff. ie ifconfig
 958  *      loads the devconfigure module does its configuring and unloads it.
 959  *      There's a good 20K of config code hanging around the kernel.
 960  */
 961
 962 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 963 {
 964         struct sock *sk = sock->sk;
 965         int err = 0;
 966         struct net *net = sock_net(sk);
 967         void __user *p = (void __user *)arg;
 968         struct ifreq ifr;
 969         struct rtentry rt;
 970
 971         switch (cmd) {
 972         case SIOCADDRT:
 973         case SIOCDELRT:
 974                 if (copy_from_user(&rt, p, sizeof(struct rtentry)))
 975                         return -EFAULT;
 976                 err = ip_rt_ioctl(net, cmd, &rt);
 977                 break;
 978         case SIOCRTMSG:
 979                 err = -EINVAL;
 980                 break;
 981         case SIOCDARP:
 982         case SIOCGARP:
 983         case SIOCSARP:
 984                 err = arp_ioctl(net, cmd, (void __user *)arg);
 985                 break;
 986         case SIOCGIFADDR:
 987         case SIOCGIFBRDADDR:
 988         case SIOCGIFNETMASK:
 989         case SIOCGIFDSTADDR:
 990         case SIOCGIFPFLAGS:
 991                 if (get_user_ifreq(&ifr, NULL, p))
 992                         return -EFAULT;
 993                 err = devinet_ioctl(net, cmd, &ifr);
 994                 if (!err && put_user_ifreq(&ifr, p))
 995                         err = -EFAULT;
 996                 break;
 997
 998         case SIOCSIFADDR:
 999         case SIOCSIFBRDADDR:
1000         case SIOCSIFNETMASK:
1001         case SIOCSIFDSTADDR:
1002         case SIOCSIFPFLAGS:
1003         case SIOCSIFFLAGS:
1004                 if (get_user_ifreq(&ifr, NULL, p))
1005                         return -EFAULT;
1006                 err = devinet_ioctl(net, cmd, &ifr);
1007                 break;
1008         default:
1009                 if (sk->sk_prot->ioctl)
1010                         err = sk->sk_prot->ioctl(sk, cmd, arg);
1011                 else
1012                         err = -ENOIOCTLCMD;
1013                 break;
1014         }
1015         return err;
1016 }
1017 EXPORT_SYMBOL(inet_ioctl);
1018
1019 #ifdef CONFIG_COMPAT
1020 static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd,
1021                 struct compat_rtentry __user *ur)
1022 {
1023         compat_uptr_t rtdev;
1024         struct rtentry rt;
1025
1026         if (copy_from_user(&rt.rt_dst, &ur->rt_dst,
1027                         3 * sizeof(struct sockaddr)) ||
1028             get_user(rt.rt_flags, &ur->rt_flags) ||
1029             get_user(rt.rt_metric, &ur->rt_metric) ||
1030             get_user(rt.rt_mtu, &ur->rt_mtu) ||
1031             get_user(rt.rt_window, &ur->rt_window) ||
1032             get_user(rt.rt_irtt, &ur->rt_irtt) ||
1033             get_user(rtdev, &ur->rt_dev))
1034                 return -EFAULT;
1035
1036         rt.rt_dev = compat_ptr(rtdev);
1037         return ip_rt_ioctl(sock_net(sk), cmd, &rt);
1038 }
1039
1040 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1041 {
1042         void __user *argp = compat_ptr(arg);
1043         struct sock *sk = sock->sk;
1044
1045         switch (cmd) {
1046         case SIOCADDRT:
1047         case SIOCDELRT:
1048                 return inet_compat_routing_ioctl(sk, cmd, argp);
1049         default:
1050                 if (!sk->sk_prot->compat_ioctl)
1051                         return -ENOIOCTLCMD;
1052                 return sk->sk_prot->compat_ioctl(sk, cmd, arg);
1053         }
1054 }
1055 #endif /* CONFIG_COMPAT */
1056
1057 const struct proto_ops inet_stream_ops = {
1058         .family            = PF_INET,
1059         .owner             = THIS_MODULE,
1060         .release           = inet_release,
1061         .bind              = inet_bind,
1062         .connect           = inet_stream_connect,
1063         .socketpair        = sock_no_socketpair,
1064         .accept            = inet_accept,
1065         .getname           = inet_getname,
1066         .poll              = tcp_poll,
1067         .ioctl             = inet_ioctl,
1068         .gettstamp         = sock_gettstamp,
1069         .listen            = inet_listen,
1070         .shutdown          = inet_shutdown,
1071         .setsockopt        = sock_common_setsockopt,
1072         .getsockopt        = sock_common_getsockopt,
1073         .sendmsg           = inet_sendmsg,
1074         .recvmsg           = inet_recvmsg,
1075 #ifdef CONFIG_MMU
1076         .mmap              = tcp_mmap,
1077 #endif
1078         .splice_eof        = inet_splice_eof,
1079         .sendpage          = inet_sendpage,
1080         .splice_read       = tcp_splice_read,
1081         .read_sock         = tcp_read_sock,
1082         .read_skb          = tcp_read_skb,
1083         .sendmsg_locked    = tcp_sendmsg_locked,
1084         .sendpage_locked   = tcp_sendpage_locked,
1085         .peek_len          = tcp_peek_len,
1086 #ifdef CONFIG_COMPAT
1087         .compat_ioctl      = inet_compat_ioctl,
1088 #endif
1089         .set_rcvlowat      = tcp_set_rcvlowat,
1090 };
1091 EXPORT_SYMBOL(inet_stream_ops);
1092
1093 const struct proto_ops inet_dgram_ops = {
1094         .family            = PF_INET,
1095         .owner             = THIS_MODULE,
1096         .release           = inet_release,
1097         .bind              = inet_bind,
1098         .connect           = inet_dgram_connect,
1099         .socketpair        = sock_no_socketpair,
1100         .accept            = sock_no_accept,
1101         .getname           = inet_getname,
1102         .poll              = udp_poll,
1103         .ioctl             = inet_ioctl,
1104         .gettstamp         = sock_gettstamp,
1105         .listen            = sock_no_listen,
1106         .shutdown          = inet_shutdown,
1107         .setsockopt        = sock_common_setsockopt,
1108         .getsockopt        = sock_common_getsockopt,
1109         .sendmsg           = inet_sendmsg,
1110         .read_skb          = udp_read_skb,
1111         .recvmsg           = inet_recvmsg,
1112         .mmap              = sock_no_mmap,
1113         .splice_eof        = inet_splice_eof,
1114         .sendpage          = inet_sendpage,
1115         .set_peek_off      = sk_set_peek_off,
1116 #ifdef CONFIG_COMPAT
1117         .compat_ioctl      = inet_compat_ioctl,
1118 #endif
1119 };
1120 EXPORT_SYMBOL(inet_dgram_ops);
1121
1122 /*
1123  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1124  * udp_poll
1125  */
1126 static const struct proto_ops inet_sockraw_ops = {
1127         .family            = PF_INET,
1128         .owner             = THIS_MODULE,
1129         .release           = inet_release,
1130         .bind              = inet_bind,
1131         .connect           = inet_dgram_connect,
1132         .socketpair        = sock_no_socketpair,
1133         .accept            = sock_no_accept,
1134         .getname           = inet_getname,
1135         .poll              = datagram_poll,
1136         .ioctl             = inet_ioctl,
1137         .gettstamp         = sock_gettstamp,
1138         .listen            = sock_no_listen,
1139         .shutdown          = inet_shutdown,
1140         .setsockopt        = sock_common_setsockopt,
1141         .getsockopt        = sock_common_getsockopt,
1142         .sendmsg           = inet_sendmsg,
1143         .recvmsg           = inet_recvmsg,
1144         .mmap              = sock_no_mmap,
1145         .splice_eof        = inet_splice_eof,
1146         .sendpage          = inet_sendpage,
1147 #ifdef CONFIG_COMPAT
1148         .compat_ioctl      = inet_compat_ioctl,
1149 #endif
1150 };
1151
1152 static const struct net_proto_family inet_family_ops = {
1153         .family = PF_INET,
1154         .create = inet_create,
1155         .owner  = THIS_MODULE,
1156 };
1157
1158 /* Upon startup we insert all the elements in inetsw_array[] into
1159  * the linked list inetsw.
1160  */
1161 static struct inet_protosw inetsw_array[] =
1162 {
1163         {
1164                 .type =       SOCK_STREAM,
1165                 .protocol =   IPPROTO_TCP,
1166                 .prot =       &tcp_prot,
1167                 .ops =        &inet_stream_ops,
1168                 .flags =      INET_PROTOSW_PERMANENT |
1169                               INET_PROTOSW_ICSK,
1170         },
1171
1172         {
1173                 .type =       SOCK_DGRAM,
1174                 .protocol =   IPPROTO_UDP,
1175                 .prot =       &udp_prot,
1176                 .ops =        &inet_dgram_ops,
1177                 .flags =      INET_PROTOSW_PERMANENT,
1178        },
1179
1180        {
1181                 .type =       SOCK_DGRAM,
1182                 .protocol =   IPPROTO_ICMP,
1183                 .prot =       &ping_prot,
1184                 .ops =        &inet_sockraw_ops,
1185                 .flags =      INET_PROTOSW_REUSE,
1186        },
1187
1188        {
1189                .type =       SOCK_RAW,
1190                .protocol =   IPPROTO_IP,        /* wild card */
1191                .prot =       &raw_prot,
1192                .ops =        &inet_sockraw_ops,
1193                .flags =      INET_PROTOSW_REUSE,
1194        }
1195 };
1196
1197 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1198
1199 void inet_register_protosw(struct inet_protosw *p)
1200 {
1201         struct list_head *lh;
1202         struct inet_protosw *answer;
1203         int protocol = p->protocol;
1204         struct list_head *last_perm;
1205
1206         spin_lock_bh(&inetsw_lock);
1207
1208         if (p->type >= SOCK_MAX)
1209                 goto out_illegal;
1210
1211         /* If we are trying to override a permanent protocol, bail. */
1212         last_perm = &inetsw[p->type];
1213         list_for_each(lh, &inetsw[p->type]) {
1214                 answer = list_entry(lh, struct inet_protosw, list);
1215                 /* Check only the non-wild match. */
1216                 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1217                         break;
1218                 if (protocol == answer->protocol)
1219                         goto out_permanent;
1220                 last_perm = lh;
1221         }
1222
1223         /* Add the new entry after the last permanent entry if any, so that
1224          * the new entry does not override a permanent entry when matched with
1225          * a wild-card protocol. But it is allowed to override any existing
1226          * non-permanent entry.  This means that when we remove this entry, the
1227          * system automatically returns to the old behavior.
1228          */
1229         list_add_rcu(&p->list, last_perm);
1230 out:
1231         spin_unlock_bh(&inetsw_lock);
1232
1233         return;
1234
1235 out_permanent:
1236         pr_err("Attempt to override permanent protocol %d\n", protocol);
1237         goto out;
1238
1239 out_illegal:
1240         pr_err("Ignoring attempt to register invalid socket type %d\n",
1241                p->type);
1242         goto out;
1243 }
1244 EXPORT_SYMBOL(inet_register_protosw);
1245
1246 void inet_unregister_protosw(struct inet_protosw *p)
1247 {
1248         if (INET_PROTOSW_PERMANENT & p->flags) {
1249                 pr_err("Attempt to unregister permanent protocol %d\n",
1250                        p->protocol);
1251         } else {
1252                 spin_lock_bh(&inetsw_lock);
1253                 list_del_rcu(&p->list);
1254                 spin_unlock_bh(&inetsw_lock);
1255
1256                 synchronize_net();
1257         }
1258 }
1259 EXPORT_SYMBOL(inet_unregister_protosw);
1260
1261 static int inet_sk_reselect_saddr(struct sock *sk)
1262 {
1263         struct inet_sock *inet = inet_sk(sk);
1264         __be32 old_saddr = inet->inet_saddr;
1265         __be32 daddr = inet->inet_daddr;
1266         struct flowi4 *fl4;
1267         struct rtable *rt;
1268         __be32 new_saddr;
1269         struct ip_options_rcu *inet_opt;
1270         int err;
1271
1272         inet_opt = rcu_dereference_protected(inet->inet_opt,
1273                                              lockdep_sock_is_held(sk));
1274         if (inet_opt && inet_opt->opt.srr)
1275                 daddr = inet_opt->opt.faddr;
1276
1277         /* Query new route. */
1278         fl4 = &inet->cork.fl.u.ip4;
1279         rt = ip_route_connect(fl4, daddr, 0, sk->sk_bound_dev_if,
1280                               sk->sk_protocol, inet->inet_sport,
1281                               inet->inet_dport, sk);
1282         if (IS_ERR(rt))
1283                 return PTR_ERR(rt);
1284
1285         new_saddr = fl4->saddr;
1286
1287         if (new_saddr == old_saddr) {
1288                 sk_setup_caps(sk, &rt->dst);
1289                 return 0;
1290         }
1291
1292         err = inet_bhash2_update_saddr(sk, &new_saddr, AF_INET);
1293         if (err) {
1294                 ip_rt_put(rt);
1295                 return err;
1296         }
1297
1298         sk_setup_caps(sk, &rt->dst);
1299
1300         if (READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) > 1) {
1301                 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1302                         __func__, &old_saddr, &new_saddr);
1303         }
1304
1305         /*
1306          * XXX The only one ugly spot where we need to
1307          * XXX really change the sockets identity after
1308          * XXX it has entered the hashes. -DaveM
1309          *
1310          * Besides that, it does not check for connection
1311          * uniqueness. Wait for troubles.
1312          */
1313         return __sk_prot_rehash(sk);
1314 }
1315
1316 int inet_sk_rebuild_header(struct sock *sk)
1317 {
1318         struct inet_sock *inet = inet_sk(sk);
1319         struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1320         __be32 daddr;
1321         struct ip_options_rcu *inet_opt;
1322         struct flowi4 *fl4;
1323         int err;
1324
1325         /* Route is OK, nothing to do. */
1326         if (rt)
1327                 return 0;
1328
1329         /* Reroute. */
1330         rcu_read_lock();
1331         inet_opt = rcu_dereference(inet->inet_opt);
1332         daddr = inet->inet_daddr;
1333         if (inet_opt && inet_opt->opt.srr)
1334                 daddr = inet_opt->opt.faddr;
1335         rcu_read_unlock();
1336         fl4 = &inet->cork.fl.u.ip4;
1337         rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1338                                    inet->inet_dport, inet->inet_sport,
1339                                    sk->sk_protocol, RT_CONN_FLAGS(sk),
1340                                    sk->sk_bound_dev_if);
1341         if (!IS_ERR(rt)) {
1342                 err = 0;
1343                 sk_setup_caps(sk, &rt->dst);
1344         } else {
1345                 err = PTR_ERR(rt);
1346
1347                 /* Routing failed... */
1348                 sk->sk_route_caps = 0;
1349                 /*
1350                  * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1351                  * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1352                  */
1353                 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) ||
1354                     sk->sk_state != TCP_SYN_SENT ||
1355                     (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1356                     (err = inet_sk_reselect_saddr(sk)) != 0)
1357                         sk->sk_err_soft = -err;
1358         }
1359
1360         return err;
1361 }
1362 EXPORT_SYMBOL(inet_sk_rebuild_header);
1363
1364 void inet_sk_set_state(struct sock *sk, int state)
1365 {
1366         trace_inet_sock_set_state(sk, sk->sk_state, state);
1367         sk->sk_state = state;
1368 }
1369 EXPORT_SYMBOL(inet_sk_set_state);
1370
1371 void inet_sk_state_store(struct sock *sk, int newstate)
1372 {
1373         trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1374         smp_store_release(&sk->sk_state, newstate);
1375 }
1376
1377 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1378                                  netdev_features_t features)
1379 {
1380         bool udpfrag = false, fixedid = false, gso_partial, encap;
1381         struct sk_buff *segs = ERR_PTR(-EINVAL);
1382         const struct net_offload *ops;
1383         unsigned int offset = 0;
1384         struct iphdr *iph;
1385         int proto, tot_len;
1386         int nhoff;
1387         int ihl;
1388         int id;
1389
1390         skb_reset_network_header(skb);
1391         nhoff = skb_network_header(skb) - skb_mac_header(skb);
1392         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1393                 goto out;
1394
1395         iph = ip_hdr(skb);
1396         ihl = iph->ihl * 4;
1397         if (ihl < sizeof(*iph))
1398                 goto out;
1399
1400         id = ntohs(iph->id);
1401         proto = iph->protocol;
1402
1403         /* Warning: after this point, iph might be no longer valid */
1404         if (unlikely(!pskb_may_pull(skb, ihl)))
1405                 goto out;
1406         __skb_pull(skb, ihl);
1407
1408         encap = SKB_GSO_CB(skb)->encap_level > 0;
1409         if (encap)
1410                 features &= skb->dev->hw_enc_features;
1411         SKB_GSO_CB(skb)->encap_level += ihl;
1412
1413         skb_reset_transport_header(skb);
1414
1415         segs = ERR_PTR(-EPROTONOSUPPORT);
1416
1417         if (!skb->encapsulation || encap) {
1418                 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1419                 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1420
1421                 /* fixed ID is invalid if DF bit is not set */
1422                 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1423                         goto out;
1424         }
1425
1426         ops = rcu_dereference(inet_offloads[proto]);
1427         if (likely(ops && ops->callbacks.gso_segment)) {
1428                 segs = ops->callbacks.gso_segment(skb, features);
1429                 if (!segs)
1430                         skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
1431         }
1432
1433         if (IS_ERR_OR_NULL(segs))
1434                 goto out;
1435
1436         gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1437
1438         skb = segs;
1439         do {
1440                 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1441                 if (udpfrag) {
1442                         iph->frag_off = htons(offset >> 3);
1443                         if (skb->next)
1444                                 iph->frag_off |= htons(IP_MF);
1445                         offset += skb->len - nhoff - ihl;
1446                         tot_len = skb->len - nhoff;
1447                 } else if (skb_is_gso(skb)) {
1448                         if (!fixedid) {
1449                                 iph->id = htons(id);
1450                                 id += skb_shinfo(skb)->gso_segs;
1451                         }
1452
1453                         if (gso_partial)
1454                                 tot_len = skb_shinfo(skb)->gso_size +
1455                                           SKB_GSO_CB(skb)->data_offset +
1456                                           skb->head - (unsigned char *)iph;
1457                         else
1458                                 tot_len = skb->len - nhoff;
1459                 } else {
1460                         if (!fixedid)
1461                                 iph->id = htons(id++);
1462                         tot_len = skb->len - nhoff;
1463                 }
1464                 iph->tot_len = htons(tot_len);
1465                 ip_send_check(iph);
1466                 if (encap)
1467                         skb_reset_inner_headers(skb);
1468                 skb->network_header = (u8 *)iph - skb->head;
1469                 skb_reset_mac_len(skb);
1470         } while ((skb = skb->next));
1471
1472 out:
1473         return segs;
1474 }
1475
1476 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1477                                         netdev_features_t features)
1478 {
1479         if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1480                 return ERR_PTR(-EINVAL);
1481
1482         return inet_gso_segment(skb, features);
1483 }
1484
1485 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1486 {
1487         const struct net_offload *ops;
1488         struct sk_buff *pp = NULL;
1489         const struct iphdr *iph;
1490         struct sk_buff *p;
1491         unsigned int hlen;
1492         unsigned int off;
1493         unsigned int id;
1494         int flush = 1;
1495         int proto;
1496
1497         off = skb_gro_offset(skb);
1498         hlen = off + sizeof(*iph);
1499         iph = skb_gro_header(skb, hlen, off);
1500         if (unlikely(!iph))
1501                 goto out;
1502
1503         proto = iph->protocol;
1504
1505         ops = rcu_dereference(inet_offloads[proto]);
1506         if (!ops || !ops->callbacks.gro_receive)
1507                 goto out;
1508
1509         if (*(u8 *)iph != 0x45)
1510                 goto out;
1511
1512         if (ip_is_fragment(iph))
1513                 goto out;
1514
1515         if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1516                 goto out;
1517
1518         id = ntohl(*(__be32 *)&iph->id);
1519         flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1520         id >>= 16;
1521
1522         list_for_each_entry(p, head, list) {
1523                 struct iphdr *iph2;
1524                 u16 flush_id;
1525
1526                 if (!NAPI_GRO_CB(p)->same_flow)
1527                         continue;
1528
1529                 iph2 = (struct iphdr *)(p->data + off);
1530                 /* The above works because, with the exception of the top
1531                  * (inner most) layer, we only aggregate pkts with the same
1532                  * hdr length so all the hdrs we'll need to verify will start
1533                  * at the same offset.
1534                  */
1535                 if ((iph->protocol ^ iph2->protocol) |
1536                     ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1537                     ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1538                         NAPI_GRO_CB(p)->same_flow = 0;
1539                         continue;
1540                 }
1541
1542                 /* All fields must match except length and checksum. */
1543                 NAPI_GRO_CB(p)->flush |=
1544                         (iph->ttl ^ iph2->ttl) |
1545                         (iph->tos ^ iph2->tos) |
1546                         ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1547
1548                 NAPI_GRO_CB(p)->flush |= flush;
1549
1550                 /* We need to store of the IP ID check to be included later
1551                  * when we can verify that this packet does in fact belong
1552                  * to a given flow.
1553                  */
1554                 flush_id = (u16)(id - ntohs(iph2->id));
1555
1556                 /* This bit of code makes it much easier for us to identify
1557                  * the cases where we are doing atomic vs non-atomic IP ID
1558                  * checks.  Specifically an atomic check can return IP ID
1559                  * values 0 - 0xFFFF, while a non-atomic check can only
1560                  * return 0 or 0xFFFF.
1561                  */
1562                 if (!NAPI_GRO_CB(p)->is_atomic ||
1563                     !(iph->frag_off & htons(IP_DF))) {
1564                         flush_id ^= NAPI_GRO_CB(p)->count;
1565                         flush_id = flush_id ? 0xFFFF : 0;
1566                 }
1567
1568                 /* If the previous IP ID value was based on an atomic
1569                  * datagram we can overwrite the value and ignore it.
1570                  */
1571                 if (NAPI_GRO_CB(skb)->is_atomic)
1572                         NAPI_GRO_CB(p)->flush_id = flush_id;
1573                 else
1574                         NAPI_GRO_CB(p)->flush_id |= flush_id;
1575         }
1576
1577         NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1578         NAPI_GRO_CB(skb)->flush |= flush;
1579         skb_set_network_header(skb, off);
1580         /* The above will be needed by the transport layer if there is one
1581          * immediately following this IP hdr.
1582          */
1583
1584         /* Note : No need to call skb_gro_postpull_rcsum() here,
1585          * as we already checked checksum over ipv4 header was 0
1586          */
1587         skb_gro_pull(skb, sizeof(*iph));
1588         skb_set_transport_header(skb, skb_gro_offset(skb));
1589
1590         pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1591                                        ops->callbacks.gro_receive, head, skb);
1592
1593 out:
1594         skb_gro_flush_final(skb, pp, flush);
1595
1596         return pp;
1597 }
1598
1599 static struct sk_buff *ipip_gro_receive(struct list_head *head,
1600                                         struct sk_buff *skb)
1601 {
1602         if (NAPI_GRO_CB(skb)->encap_mark) {
1603                 NAPI_GRO_CB(skb)->flush = 1;
1604                 return NULL;
1605         }
1606
1607         NAPI_GRO_CB(skb)->encap_mark = 1;
1608
1609         return inet_gro_receive(head, skb);
1610 }
1611
1612 #define SECONDS_PER_DAY 86400
1613
1614 /* inet_current_timestamp - Return IP network timestamp
1615  *
1616  * Return milliseconds since midnight in network byte order.
1617  */
1618 __be32 inet_current_timestamp(void)
1619 {
1620         u32 secs;
1621         u32 msecs;
1622         struct timespec64 ts;
1623
1624         ktime_get_real_ts64(&ts);
1625
1626         /* Get secs since midnight. */
1627         (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1628         /* Convert to msecs. */
1629         msecs = secs * MSEC_PER_SEC;
1630         /* Convert nsec to msec. */
1631         msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1632
1633         /* Convert to network byte order. */
1634         return htonl(msecs);
1635 }
1636 EXPORT_SYMBOL(inet_current_timestamp);
1637
1638 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1639 {
1640         unsigned int family = READ_ONCE(sk->sk_family);
1641
1642         if (family == AF_INET)
1643                 return ip_recv_error(sk, msg, len, addr_len);
1644 #if IS_ENABLED(CONFIG_IPV6)
1645         if (family == AF_INET6)
1646                 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1647 #endif
1648         return -EINVAL;
1649 }
1650 EXPORT_SYMBOL(inet_recv_error);
1651
1652 int inet_gro_complete(struct sk_buff *skb, int nhoff)
1653 {
1654         __be16 newlen = htons(skb->len - nhoff);
1655         struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1656         const struct net_offload *ops;
1657         int proto = iph->protocol;
1658         int err = -ENOSYS;
1659
1660         if (skb->encapsulation) {
1661                 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1662                 skb_set_inner_network_header(skb, nhoff);
1663         }
1664
1665         csum_replace2(&iph->check, iph->tot_len, newlen);
1666         iph->tot_len = newlen;
1667
1668         ops = rcu_dereference(inet_offloads[proto]);
1669         if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1670                 goto out;
1671
1672         /* Only need to add sizeof(*iph) to get to the next hdr below
1673          * because any hdr with option will have been flushed in
1674          * inet_gro_receive().
1675          */
1676         err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1677                               tcp4_gro_complete, udp4_gro_complete,
1678                               skb, nhoff + sizeof(*iph));
1679
1680 out:
1681         return err;
1682 }
1683
1684 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1685 {
1686         skb->encapsulation = 1;
1687         skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1688         return inet_gro_complete(skb, nhoff);
1689 }
1690
1691 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1692                          unsigned short type, unsigned char protocol,
1693                          struct net *net)
1694 {
1695         struct socket *sock;
1696         int rc = sock_create_kern(net, family, type, protocol, &sock);
1697
1698         if (rc == 0) {
1699                 *sk = sock->sk;
1700                 (*sk)->sk_allocation = GFP_ATOMIC;
1701                 /*
1702                  * Unhash it so that IP input processing does not even see it,
1703                  * we do not wish this socket to see incoming packets.
1704                  */
1705                 (*sk)->sk_prot->unhash(*sk);
1706         }
1707         return rc;
1708 }
1709 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1710
1711 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1712 {
1713         unsigned long res = 0;
1714         int i;
1715
1716         for_each_possible_cpu(i)
1717                 res += snmp_get_cpu_field(mib, i, offt);
1718         return res;
1719 }
1720 EXPORT_SYMBOL_GPL(snmp_fold_field);
1721
1722 #if BITS_PER_LONG==32
1723
1724 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1725                          size_t syncp_offset)
1726 {
1727         void *bhptr;
1728         struct u64_stats_sync *syncp;
1729         u64 v;
1730         unsigned int start;
1731
1732         bhptr = per_cpu_ptr(mib, cpu);
1733         syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1734         do {
1735                 start = u64_stats_fetch_begin_irq(syncp);
1736                 v = *(((u64 *)bhptr) + offt);
1737         } while (u64_stats_fetch_retry_irq(syncp, start));
1738
1739         return v;
1740 }
1741 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1742
1743 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1744 {
1745         u64 res = 0;
1746         int cpu;
1747
1748         for_each_possible_cpu(cpu) {
1749                 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1750         }
1751         return res;
1752 }
1753 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1754 #endif
1755
1756 #ifdef CONFIG_IP_MULTICAST
1757 static const struct net_protocol igmp_protocol = {
1758         .handler =      igmp_rcv,
1759 };
1760 #endif
1761
1762 static const struct net_protocol tcp_protocol = {
1763         .handler        =       tcp_v4_rcv,
1764         .err_handler    =       tcp_v4_err,
1765         .no_policy      =       1,
1766         .icmp_strict_tag_validation = 1,
1767 };
1768
1769 static const struct net_protocol udp_protocol = {
1770         .handler =      udp_rcv,
1771         .err_handler =  udp_err,
1772         .no_policy =    1,
1773 };
1774
1775 static const struct net_protocol icmp_protocol = {
1776         .handler =      icmp_rcv,
1777         .err_handler =  icmp_err,
1778         .no_policy =    1,
1779 };
1780
1781 static __net_init int ipv4_mib_init_net(struct net *net)
1782 {
1783         int i;
1784
1785         net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1786         if (!net->mib.tcp_statistics)
1787                 goto err_tcp_mib;
1788         net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1789         if (!net->mib.ip_statistics)
1790                 goto err_ip_mib;
1791
1792         for_each_possible_cpu(i) {
1793                 struct ipstats_mib *af_inet_stats;
1794                 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1795                 u64_stats_init(&af_inet_stats->syncp);
1796         }
1797
1798         net->mib.net_statistics = alloc_percpu(struct linux_mib);
1799         if (!net->mib.net_statistics)
1800                 goto err_net_mib;
1801         net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1802         if (!net->mib.udp_statistics)
1803                 goto err_udp_mib;
1804         net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1805         if (!net->mib.udplite_statistics)
1806                 goto err_udplite_mib;
1807         net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1808         if (!net->mib.icmp_statistics)
1809                 goto err_icmp_mib;
1810         net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1811                                               GFP_KERNEL);
1812         if (!net->mib.icmpmsg_statistics)
1813                 goto err_icmpmsg_mib;
1814
1815         tcp_mib_init(net);
1816         return 0;
1817
1818 err_icmpmsg_mib:
1819         free_percpu(net->mib.icmp_statistics);
1820 err_icmp_mib:
1821         free_percpu(net->mib.udplite_statistics);
1822 err_udplite_mib:
1823         free_percpu(net->mib.udp_statistics);
1824 err_udp_mib:
1825         free_percpu(net->mib.net_statistics);
1826 err_net_mib:
1827         free_percpu(net->mib.ip_statistics);
1828 err_ip_mib:
1829         free_percpu(net->mib.tcp_statistics);
1830 err_tcp_mib:
1831         return -ENOMEM;
1832 }
1833
1834 static __net_exit void ipv4_mib_exit_net(struct net *net)
1835 {
1836         kfree(net->mib.icmpmsg_statistics);
1837         free_percpu(net->mib.icmp_statistics);
1838         free_percpu(net->mib.udplite_statistics);
1839         free_percpu(net->mib.udp_statistics);
1840         free_percpu(net->mib.net_statistics);
1841         free_percpu(net->mib.ip_statistics);
1842         free_percpu(net->mib.tcp_statistics);
1843 #ifdef CONFIG_MPTCP
1844         /* allocated on demand, see mptcp_init_sock() */
1845         free_percpu(net->mib.mptcp_statistics);
1846 #endif
1847 }
1848
1849 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1850         .init = ipv4_mib_init_net,
1851         .exit = ipv4_mib_exit_net,
1852 };
1853
1854 static int __init init_ipv4_mibs(void)
1855 {
1856         return register_pernet_subsys(&ipv4_mib_ops);
1857 }
1858
1859 static __net_init int inet_init_net(struct net *net)
1860 {
1861         /*
1862          * Set defaults for local port range
1863          */
1864         seqlock_init(&net->ipv4.ip_local_ports.lock);
1865         net->ipv4.ip_local_ports.range[0] =  32768;
1866         net->ipv4.ip_local_ports.range[1] =  60999;
1867
1868         seqlock_init(&net->ipv4.ping_group_range.lock);
1869         /*
1870          * Sane defaults - nobody may create ping sockets.
1871          * Boot scripts should set this to distro-specific group.
1872          */
1873         net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1874         net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1875
1876         /* Default values for sysctl-controlled parameters.
1877          * We set them here, in case sysctl is not compiled.
1878          */
1879         net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1880         net->ipv4.sysctl_ip_fwd_update_priority = 1;
1881         net->ipv4.sysctl_ip_dynaddr = 0;
1882         net->ipv4.sysctl_ip_early_demux = 1;
1883         net->ipv4.sysctl_udp_early_demux = 1;
1884         net->ipv4.sysctl_tcp_early_demux = 1;
1885         net->ipv4.sysctl_nexthop_compat_mode = 1;
1886 #ifdef CONFIG_SYSCTL
1887         net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1888 #endif
1889
1890         /* Some igmp sysctl, whose values are always used */
1891         net->ipv4.sysctl_igmp_max_memberships = 20;
1892         net->ipv4.sysctl_igmp_max_msf = 10;
1893         /* IGMP reports for link-local multicast groups are enabled by default */
1894         net->ipv4.sysctl_igmp_llm_reports = 1;
1895         net->ipv4.sysctl_igmp_qrv = 2;
1896
1897         net->ipv4.sysctl_fib_notify_on_flag_change = 0;
1898
1899         return 0;
1900 }
1901
1902 static __net_initdata struct pernet_operations af_inet_ops = {
1903         .init = inet_init_net,
1904 };
1905
1906 static int __init init_inet_pernet_ops(void)
1907 {
1908         return register_pernet_subsys(&af_inet_ops);
1909 }
1910
1911 static int ipv4_proc_init(void);
1912
1913 /*
1914  *      IP protocol layer initialiser
1915  */
1916
1917 static struct packet_offload ip_packet_offload __read_mostly = {
1918         .type = cpu_to_be16(ETH_P_IP),
1919         .callbacks = {
1920                 .gso_segment = inet_gso_segment,
1921                 .gro_receive = inet_gro_receive,
1922                 .gro_complete = inet_gro_complete,
1923         },
1924 };
1925
1926 static const struct net_offload ipip_offload = {
1927         .callbacks = {
1928                 .gso_segment    = ipip_gso_segment,
1929                 .gro_receive    = ipip_gro_receive,
1930                 .gro_complete   = ipip_gro_complete,
1931         },
1932 };
1933
1934 static int __init ipip_offload_init(void)
1935 {
1936         return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1937 }
1938
1939 static int __init ipv4_offload_init(void)
1940 {
1941         /*
1942          * Add offloads
1943          */
1944         if (udpv4_offload_init() < 0)
1945                 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1946         if (tcpv4_offload_init() < 0)
1947                 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1948         if (ipip_offload_init() < 0)
1949                 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1950
1951         dev_add_offload(&ip_packet_offload);
1952         return 0;
1953 }
1954
1955 fs_initcall(ipv4_offload_init);
1956
1957 static struct packet_type ip_packet_type __read_mostly = {
1958         .type = cpu_to_be16(ETH_P_IP),
1959         .func = ip_rcv,
1960         .list_func = ip_list_rcv,
1961 };
1962
1963 static int __init inet_init(void)
1964 {
1965         struct inet_protosw *q;
1966         struct list_head *r;
1967         int rc;
1968
1969         sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1970
1971         raw_hashinfo_init(&raw_v4_hashinfo);
1972
1973         rc = proto_register(&tcp_prot, 1);
1974         if (rc)
1975                 goto out;
1976
1977         rc = proto_register(&udp_prot, 1);
1978         if (rc)
1979                 goto out_unregister_tcp_proto;
1980
1981         rc = proto_register(&raw_prot, 1);
1982         if (rc)
1983                 goto out_unregister_udp_proto;
1984
1985         rc = proto_register(&ping_prot, 1);
1986         if (rc)
1987                 goto out_unregister_raw_proto;
1988
1989         /*
1990          *      Tell SOCKET that we are alive...
1991          */
1992
1993         (void)sock_register(&inet_family_ops);
1994
1995 #ifdef CONFIG_SYSCTL
1996         ip_static_sysctl_init();
1997 #endif
1998
1999         /*
2000          *      Add all the base protocols.
2001          */
2002
2003         if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
2004                 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
2005         if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
2006                 pr_crit("%s: Cannot add UDP protocol\n", __func__);
2007         if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
2008                 pr_crit("%s: Cannot add TCP protocol\n", __func__);
2009 #ifdef CONFIG_IP_MULTICAST
2010         if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
2011                 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
2012 #endif
2013
2014         /* Register the socket-side information for inet_create. */
2015         for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
2016                 INIT_LIST_HEAD(r);
2017
2018         for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
2019                 inet_register_protosw(q);
2020
2021         /*
2022          *      Set the ARP module up
2023          */
2024
2025         arp_init();
2026
2027         /*
2028          *      Set the IP module up
2029          */
2030
2031         ip_init();
2032
2033         /* Initialise per-cpu ipv4 mibs */
2034         if (init_ipv4_mibs())
2035                 panic("%s: Cannot init ipv4 mibs\n", __func__);
2036
2037         /* Setup TCP slab cache for open requests. */
2038         tcp_init();
2039
2040         /* Setup UDP memory threshold */
2041         udp_init();
2042
2043         /* Add UDP-Lite (RFC 3828) */
2044         udplite4_register();
2045
2046         raw_init();
2047
2048         ping_init();
2049
2050         /*
2051          *      Set the ICMP layer up
2052          */
2053
2054         if (icmp_init() < 0)
2055                 panic("Failed to create the ICMP control socket.\n");
2056
2057         /*
2058          *      Initialise the multicast router
2059          */
2060 #if defined(CONFIG_IP_MROUTE)
2061         if (ip_mr_init())
2062                 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2063 #endif
2064
2065         if (init_inet_pernet_ops())
2066                 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2067
2068         ipv4_proc_init();
2069
2070         ipfrag_init();
2071
2072         dev_add_pack(&ip_packet_type);
2073
2074         ip_tunnel_core_init();
2075
2076         rc = 0;
2077 out:
2078         return rc;
2079 out_unregister_raw_proto:
2080         proto_unregister(&raw_prot);
2081 out_unregister_udp_proto:
2082         proto_unregister(&udp_prot);
2083 out_unregister_tcp_proto:
2084         proto_unregister(&tcp_prot);
2085         goto out;
2086 }
2087
2088 fs_initcall(inet_init);
2089
2090 /* ------------------------------------------------------------------------ */
2091
2092 #ifdef CONFIG_PROC_FS
2093 static int __init ipv4_proc_init(void)
2094 {
2095         int rc = 0;
2096
2097         if (raw_proc_init())
2098                 goto out_raw;
2099         if (tcp4_proc_init())
2100                 goto out_tcp;
2101         if (udp4_proc_init())
2102                 goto out_udp;
2103         if (ping_proc_init())
2104                 goto out_ping;
2105         if (ip_misc_proc_init())
2106                 goto out_misc;
2107 out:
2108         return rc;
2109 out_misc:
2110         ping_proc_exit();
2111 out_ping:
2112         udp4_proc_exit();
2113 out_udp:
2114         tcp4_proc_exit();
2115 out_tcp:
2116         raw_proc_exit();
2117 out_raw:
2118         rc = -ENOMEM;
2119         goto out;
2120 }
2121
2122 #else /* CONFIG_PROC_FS */
2123 static int __init ipv4_proc_init(void)
2124 {
2125         return 0;
2126 }
2127 #endif /* CONFIG_PROC_FS */