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/gso.h>
 104 #include <net/tcp.h>
 105 #include <net/udp.h>
 106 #include <net/udplite.h>
 107 #include <net/ping.h>
 108 #include <linux/skbuff.h>
 109 #include <net/sock.h>
 110 #include <net/raw.h>
 111 #include <net/icmp.h>
 112 #include <net/inet_common.h>
 113 #include <net/ip_tunnels.h>
 114 #include <net/xfrm.h>
 115 #include <net/net_namespace.h>
 116 #include <net/secure_seq.h>
 117 #ifdef CONFIG_IP_MROUTE
 118 #include <linux/mroute.h>
 119 #endif
 120 #include <net/l3mdev.h>
 121 #include <net/compat.h>
 122
 123 #include <trace/events/sock.h>
 124
 125 /* The inetsw table contains everything that inet_create needs to
 126  * build a new socket.
 127  */
 128 static struct list_head inetsw[SOCK_MAX];
 129 static DEFINE_SPINLOCK(inetsw_lock);
 130
 131 /* New destruction routine */
 132
 133 void inet_sock_destruct(struct sock *sk)
 134 {
 135         struct inet_sock *inet = inet_sk(sk);
 136
 137         __skb_queue_purge(&sk->sk_receive_queue);
 138         __skb_queue_purge(&sk->sk_error_queue);
 139
 140         sk_mem_reclaim_final(sk);
 141
 142         if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
 143                 pr_err("Attempt to release TCP socket in state %d %p\n",
 144                        sk->sk_state, sk);
 145                 return;
 146         }
 147         if (!sock_flag(sk, SOCK_DEAD)) {
 148                 pr_err("Attempt to release alive inet socket %p\n", sk);
 149                 return;
 150         }
 151
 152         WARN_ON_ONCE(atomic_read(&sk->sk_rmem_alloc));
 153         WARN_ON_ONCE(refcount_read(&sk->sk_wmem_alloc));
 154         WARN_ON_ONCE(sk->sk_wmem_queued);
 155         WARN_ON_ONCE(sk_forward_alloc_get(sk));
 156
 157         kfree(rcu_dereference_protected(inet->inet_opt, 1));
 158         dst_release(rcu_dereference_protected(sk->sk_dst_cache, 1));
 159         dst_release(rcu_dereference_protected(sk->sk_rx_dst, 1));
 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 int __inet_listen_sk(struct sock *sk, int backlog)
 191 {
 192         unsigned char old_state = sk->sk_state;
 193         int err, tcp_fastopen;
 194
 195         if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
 196                 return -EINVAL;
 197
 198         WRITE_ONCE(sk->sk_max_ack_backlog, backlog);
 199         /* Really, if the socket is already in listen state
 200          * we can only allow the backlog to be adjusted.
 201          */
 202         if (old_state != TCP_LISTEN) {
 203                 /* Enable TFO w/o requiring TCP_FASTOPEN socket option.
 204                  * Note that only TCP sockets (SOCK_STREAM) will reach here.
 205                  * Also fastopen backlog may already been set via the option
 206                  * because the socket was in TCP_LISTEN state previously but
 207                  * was shutdown() rather than close().
 208                  */
 209                 tcp_fastopen = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fastopen);
 210                 if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
 211                     (tcp_fastopen & TFO_SERVER_ENABLE) &&
 212                     !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
 213                         fastopen_queue_tune(sk, backlog);
 214                         tcp_fastopen_init_key_once(sock_net(sk));
 215                 }
 216
 217                 err = inet_csk_listen_start(sk);
 218                 if (err)
 219                         return err;
 220
 221                 tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL);
 222         }
 223         return 0;
 224 }
 225
 226 /*
 227  *      Move a socket into listening state.
 228  */
 229 int inet_listen(struct socket *sock, int backlog)
 230 {
 231         struct sock *sk = sock->sk;
 232         int err = -EINVAL;
 233
 234         lock_sock(sk);
 235
 236         if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
 237                 goto out;
 238
 239         err = __inet_listen_sk(sk, backlog);
 240
 241 out:
 242         release_sock(sk);
 243         return err;
 244 }
 245 EXPORT_SYMBOL(inet_listen);
 246
 247 /*
 248  *      Create an inet socket.
 249  */
 250
 251 static int inet_create(struct net *net, struct socket *sock, int protocol,
 252                        int kern)
 253 {
 254         struct sock *sk;
 255         struct inet_protosw *answer;
 256         struct inet_sock *inet;
 257         struct proto *answer_prot;
 258         unsigned char answer_flags;
 259         int try_loading_module = 0;
 260         int err;
 261
 262         if (protocol < 0 || protocol >= IPPROTO_MAX)
 263                 return -EINVAL;
 264
 265         sock->state = SS_UNCONNECTED;
 266
 267         /* Look for the requested type/protocol pair. */
 268 lookup_protocol:
 269         err = -ESOCKTNOSUPPORT;
 270         rcu_read_lock();
 271         list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
 272
 273                 err = 0;
 274                 /* Check the non-wild match. */
 275                 if (protocol == answer->protocol) {
 276                         if (protocol != IPPROTO_IP)
 277                                 break;
 278                 } else {
 279                         /* Check for the two wild cases. */
 280                         if (IPPROTO_IP == protocol) {
 281                                 protocol = answer->protocol;
 282                                 break;
 283                         }
 284                         if (IPPROTO_IP == answer->protocol)
 285                                 break;
 286                 }
 287                 err = -EPROTONOSUPPORT;
 288         }
 289
 290         if (unlikely(err)) {
 291                 if (try_loading_module < 2) {
 292                         rcu_read_unlock();
 293                         /*
 294                          * Be more specific, e.g. net-pf-2-proto-132-type-1
 295                          * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
 296                          */
 297                         if (++try_loading_module == 1)
 298                                 request_module("net-pf-%d-proto-%d-type-%d",
 299                                                PF_INET, protocol, sock->type);
 300                         /*
 301                          * Fall back to generic, e.g. net-pf-2-proto-132
 302                          * (net-pf-PF_INET-proto-IPPROTO_SCTP)
 303                          */
 304                         else
 305                                 request_module("net-pf-%d-proto-%d",
 306                                                PF_INET, protocol);
 307                         goto lookup_protocol;
 308                 } else
 309                         goto out_rcu_unlock;
 310         }
 311
 312         err = -EPERM;
 313         if (sock->type == SOCK_RAW && !kern &&
 314             !ns_capable(net->user_ns, CAP_NET_RAW))
 315                 goto out_rcu_unlock;
 316
 317         sock->ops = answer->ops;
 318         answer_prot = answer->prot;
 319         answer_flags = answer->flags;
 320         rcu_read_unlock();
 321
 322         WARN_ON(!answer_prot->slab);
 323
 324         err = -ENOMEM;
 325         sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
 326         if (!sk)
 327                 goto out;
 328
 329         err = 0;
 330         if (INET_PROTOSW_REUSE & answer_flags)
 331                 sk->sk_reuse = SK_CAN_REUSE;
 332
 333         if (INET_PROTOSW_ICSK & answer_flags)
 334                 inet_init_csk_locks(sk);
 335
 336         inet = inet_sk(sk);
 337         inet_assign_bit(IS_ICSK, sk, INET_PROTOSW_ICSK & answer_flags);
 338
 339         inet_clear_bit(NODEFRAG, sk);
 340
 341         if (SOCK_RAW == sock->type) {
 342                 inet->inet_num = protocol;
 343                 if (IPPROTO_RAW == protocol)
 344                         inet_set_bit(HDRINCL, sk);
 345         }
 346
 347         if (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc))
 348                 inet->pmtudisc = IP_PMTUDISC_DONT;
 349         else
 350                 inet->pmtudisc = IP_PMTUDISC_WANT;
 351
 352         atomic_set(&inet->inet_id, 0);
 353
 354         sock_init_data(sock, sk);
 355
 356         sk->sk_destruct    = inet_sock_destruct;
 357         sk->sk_protocol    = protocol;
 358         sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
 359         sk->sk_txrehash = READ_ONCE(net->core.sysctl_txrehash);
 360
 361         inet->uc_ttl    = -1;
 362         inet_set_bit(MC_LOOP, sk);
 363         inet->mc_ttl    = 1;
 364         inet_set_bit(MC_ALL, sk);
 365         inet->mc_index  = 0;
 366         inet->mc_list   = NULL;
 367         inet->rcv_tos   = 0;
 368
 369         if (inet->inet_num) {
 370                 /* It assumes that any protocol which allows
 371                  * the user to assign a number at socket
 372                  * creation time automatically
 373                  * shares.
 374                  */
 375                 inet->inet_sport = htons(inet->inet_num);
 376                 /* Add to protocol hash chains. */
 377                 err = sk->sk_prot->hash(sk);
 378                 if (err) {
 379                         sk_common_release(sk);
 380                         goto out;
 381                 }
 382         }
 383
 384         if (sk->sk_prot->init) {
 385                 err = sk->sk_prot->init(sk);
 386                 if (err) {
 387                         sk_common_release(sk);
 388                         goto out;
 389                 }
 390         }
 391
 392         if (!kern) {
 393                 err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
 394                 if (err) {
 395                         sk_common_release(sk);
 396                         goto out;
 397                 }
 398         }
 399 out:
 400         return err;
 401 out_rcu_unlock:
 402         rcu_read_unlock();
 403         goto out;
 404 }
 405
 406
 407 /*
 408  *      The peer socket should always be NULL (or else). When we call this
 409  *      function we are destroying the object and from then on nobody
 410  *      should refer to it.
 411  */
 412 int inet_release(struct socket *sock)
 413 {
 414         struct sock *sk = sock->sk;
 415
 416         if (sk) {
 417                 long timeout;
 418
 419                 if (!sk->sk_kern_sock)
 420                         BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk);
 421
 422                 /* Applications forget to leave groups before exiting */
 423                 ip_mc_drop_socket(sk);
 424
 425                 /* If linger is set, we don't return until the close
 426                  * is complete.  Otherwise we return immediately. The
 427                  * actually closing is done the same either way.
 428                  *
 429                  * If the close is due to the process exiting, we never
 430                  * linger..
 431                  */
 432                 timeout = 0;
 433                 if (sock_flag(sk, SOCK_LINGER) &&
 434                     !(current->flags & PF_EXITING))
 435                         timeout = sk->sk_lingertime;
 436                 sk->sk_prot->close(sk, timeout);
 437                 sock->sk = NULL;
 438         }
 439         return 0;
 440 }
 441 EXPORT_SYMBOL(inet_release);
 442
 443 int inet_bind_sk(struct sock *sk, struct sockaddr *uaddr, int addr_len)
 444 {
 445         u32 flags = BIND_WITH_LOCK;
 446         int err;
 447
 448         /* If the socket has its own bind function then use it. (RAW) */
 449         if (sk->sk_prot->bind) {
 450                 return sk->sk_prot->bind(sk, uaddr, addr_len);
 451         }
 452         if (addr_len < sizeof(struct sockaddr_in))
 453                 return -EINVAL;
 454
 455         /* BPF prog is run before any checks are done so that if the prog
 456          * changes context in a wrong way it will be caught.
 457          */
 458         err = BPF_CGROUP_RUN_PROG_INET_BIND_LOCK(sk, uaddr, &addr_len,
 459                                                  CGROUP_INET4_BIND, &flags);
 460         if (err)
 461                 return err;
 462
 463         return __inet_bind(sk, uaddr, addr_len, flags);
 464 }
 465
 466 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 467 {
 468         return inet_bind_sk(sock->sk, uaddr, addr_len);
 469 }
 470 EXPORT_SYMBOL(inet_bind);
 471
 472 int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
 473                 u32 flags)
 474 {
 475         struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
 476         struct inet_sock *inet = inet_sk(sk);
 477         struct net *net = sock_net(sk);
 478         unsigned short snum;
 479         int chk_addr_ret;
 480         u32 tb_id = RT_TABLE_LOCAL;
 481         int err;
 482
 483         if (addr->sin_family != AF_INET) {
 484                 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
 485                  * only if s_addr is INADDR_ANY.
 486                  */
 487                 err = -EAFNOSUPPORT;
 488                 if (addr->sin_family != AF_UNSPEC ||
 489                     addr->sin_addr.s_addr != htonl(INADDR_ANY))
 490                         goto out;
 491         }
 492
 493         tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
 494         chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
 495
 496         /* Not specified by any standard per-se, however it breaks too
 497          * many applications when removed.  It is unfortunate since
 498          * allowing applications to make a non-local bind solves
 499          * several problems with systems using dynamic addressing.
 500          * (ie. your servers still start up even if your ISDN link
 501          *  is temporarily down)
 502          */
 503         err = -EADDRNOTAVAIL;
 504         if (!inet_addr_valid_or_nonlocal(net, inet, addr->sin_addr.s_addr,
 505                                          chk_addr_ret))
 506                 goto out;
 507
 508         snum = ntohs(addr->sin_port);
 509         err = -EACCES;
 510         if (!(flags & BIND_NO_CAP_NET_BIND_SERVICE) &&
 511             snum && inet_port_requires_bind_service(net, snum) &&
 512             !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
 513                 goto out;
 514
 515         /*      We keep a pair of addresses. rcv_saddr is the one
 516          *      used by hash lookups, and saddr is used for transmit.
 517          *
 518          *      In the BSD API these are the same except where it
 519          *      would be illegal to use them (multicast/broadcast) in
 520          *      which case the sending device address is used.
 521          */
 522         if (flags & BIND_WITH_LOCK)
 523                 lock_sock(sk);
 524
 525         /* Check these errors (active socket, double bind). */
 526         err = -EINVAL;
 527         if (sk->sk_state != TCP_CLOSE || inet->inet_num)
 528                 goto out_release_sock;
 529
 530         inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
 531         if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
 532                 inet->inet_saddr = 0;  /* Use device */
 533
 534         /* Make sure we are allowed to bind here. */
 535         if (snum || !(inet_test_bit(BIND_ADDRESS_NO_PORT, sk) ||
 536                       (flags & BIND_FORCE_ADDRESS_NO_PORT))) {
 537                 err = sk->sk_prot->get_port(sk, snum);
 538                 if (err) {
 539                         inet->inet_saddr = inet->inet_rcv_saddr = 0;
 540                         goto out_release_sock;
 541                 }
 542                 if (!(flags & BIND_FROM_BPF)) {
 543                         err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
 544                         if (err) {
 545                                 inet->inet_saddr = inet->inet_rcv_saddr = 0;
 546                                 if (sk->sk_prot->put_port)
 547                                         sk->sk_prot->put_port(sk);
 548                                 goto out_release_sock;
 549                         }
 550                 }
 551         }
 552
 553         if (inet->inet_rcv_saddr)
 554                 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
 555         if (snum)
 556                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
 557         inet->inet_sport = htons(inet->inet_num);
 558         inet->inet_daddr = 0;
 559         inet->inet_dport = 0;
 560         sk_dst_reset(sk);
 561         err = 0;
 562 out_release_sock:
 563         if (flags & BIND_WITH_LOCK)
 564                 release_sock(sk);
 565 out:
 566         return err;
 567 }
 568
 569 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
 570                        int addr_len, int flags)
 571 {
 572         struct sock *sk = sock->sk;
 573         const struct proto *prot;
 574         int err;
 575
 576         if (addr_len < sizeof(uaddr->sa_family))
 577                 return -EINVAL;
 578
 579         /* IPV6_ADDRFORM can change sk->sk_prot under us. */
 580         prot = READ_ONCE(sk->sk_prot);
 581
 582         if (uaddr->sa_family == AF_UNSPEC)
 583                 return prot->disconnect(sk, flags);
 584
 585         if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
 586                 err = prot->pre_connect(sk, uaddr, addr_len);
 587                 if (err)
 588                         return err;
 589         }
 590
 591         if (data_race(!inet_sk(sk)->inet_num) && inet_autobind(sk))
 592                 return -EAGAIN;
 593         return prot->connect(sk, uaddr, addr_len);
 594 }
 595 EXPORT_SYMBOL(inet_dgram_connect);
 596
 597 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
 598 {
 599         DEFINE_WAIT_FUNC(wait, woken_wake_function);
 600
 601         add_wait_queue(sk_sleep(sk), &wait);
 602         sk->sk_write_pending += writebias;
 603
 604         /* Basic assumption: if someone sets sk->sk_err, he _must_
 605          * change state of the socket from TCP_SYN_*.
 606          * Connect() does not allow to get error notifications
 607          * without closing the socket.
 608          */
 609         while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 610                 release_sock(sk);
 611                 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
 612                 lock_sock(sk);
 613                 if (signal_pending(current) || !timeo)
 614                         break;
 615         }
 616         remove_wait_queue(sk_sleep(sk), &wait);
 617         sk->sk_write_pending -= writebias;
 618         return timeo;
 619 }
 620
 621 /*
 622  *      Connect to a remote host. There is regrettably still a little
 623  *      TCP 'magic' in here.
 624  */
 625 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 626                           int addr_len, int flags, int is_sendmsg)
 627 {
 628         struct sock *sk = sock->sk;
 629         int err;
 630         long timeo;
 631
 632         /*
 633          * uaddr can be NULL and addr_len can be 0 if:
 634          * sk is a TCP fastopen active socket and
 635          * TCP_FASTOPEN_CONNECT sockopt is set and
 636          * we already have a valid cookie for this socket.
 637          * In this case, user can call write() after connect().
 638          * write() will invoke tcp_sendmsg_fastopen() which calls
 639          * __inet_stream_connect().
 640          */
 641         if (uaddr) {
 642                 if (addr_len < sizeof(uaddr->sa_family))
 643                         return -EINVAL;
 644
 645                 if (uaddr->sa_family == AF_UNSPEC) {
 646                         sk->sk_disconnects++;
 647                         err = sk->sk_prot->disconnect(sk, flags);
 648                         sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 649                         goto out;
 650                 }
 651         }
 652
 653         switch (sock->state) {
 654         default:
 655                 err = -EINVAL;
 656                 goto out;
 657         case SS_CONNECTED:
 658                 err = -EISCONN;
 659                 goto out;
 660         case SS_CONNECTING:
 661                 if (inet_test_bit(DEFER_CONNECT, sk))
 662                         err = is_sendmsg ? -EINPROGRESS : -EISCONN;
 663                 else
 664                         err = -EALREADY;
 665                 /* Fall out of switch with err, set for this state */
 666                 break;
 667         case SS_UNCONNECTED:
 668                 err = -EISCONN;
 669                 if (sk->sk_state != TCP_CLOSE)
 670                         goto out;
 671
 672                 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
 673                         err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
 674                         if (err)
 675                                 goto out;
 676                 }
 677
 678                 err = sk->sk_prot->connect(sk, uaddr, addr_len);
 679                 if (err < 0)
 680                         goto out;
 681
 682                 sock->state = SS_CONNECTING;
 683
 684                 if (!err && inet_test_bit(DEFER_CONNECT, sk))
 685                         goto out;
 686
 687                 /* Just entered SS_CONNECTING state; the only
 688                  * difference is that return value in non-blocking
 689                  * case is EINPROGRESS, rather than EALREADY.
 690                  */
 691                 err = -EINPROGRESS;
 692                 break;
 693         }
 694
 695         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
 696
 697         if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 698                 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
 699                                 tcp_sk(sk)->fastopen_req &&
 700                                 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
 701                 int dis = sk->sk_disconnects;
 702
 703                 /* Error code is set above */
 704                 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
 705                         goto out;
 706
 707                 err = sock_intr_errno(timeo);
 708                 if (signal_pending(current))
 709                         goto out;
 710
 711                 if (dis != sk->sk_disconnects) {
 712                         err = -EPIPE;
 713                         goto out;
 714                 }
 715         }
 716
 717         /* Connection was closed by RST, timeout, ICMP error
 718          * or another process disconnected us.
 719          */
 720         if (sk->sk_state == TCP_CLOSE)
 721                 goto sock_error;
 722
 723         /* sk->sk_err may be not zero now, if RECVERR was ordered by user
 724          * and error was received after socket entered established state.
 725          * Hence, it is handled normally after connect() return successfully.
 726          */
 727
 728         sock->state = SS_CONNECTED;
 729         err = 0;
 730 out:
 731         return err;
 732
 733 sock_error:
 734         err = sock_error(sk) ? : -ECONNABORTED;
 735         sock->state = SS_UNCONNECTED;
 736         sk->sk_disconnects++;
 737         if (sk->sk_prot->disconnect(sk, flags))
 738                 sock->state = SS_DISCONNECTING;
 739         goto out;
 740 }
 741 EXPORT_SYMBOL(__inet_stream_connect);
 742
 743 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 744                         int addr_len, int flags)
 745 {
 746         int err;
 747
 748         lock_sock(sock->sk);
 749         err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
 750         release_sock(sock->sk);
 751         return err;
 752 }
 753 EXPORT_SYMBOL(inet_stream_connect);
 754
 755 void __inet_accept(struct socket *sock, struct socket *newsock, struct sock *newsk)
 756 {
 757         sock_rps_record_flow(newsk);
 758         WARN_ON(!((1 << newsk->sk_state) &
 759                   (TCPF_ESTABLISHED | TCPF_SYN_RECV |
 760                   TCPF_CLOSE_WAIT | TCPF_CLOSE)));
 761
 762         if (test_bit(SOCK_SUPPORT_ZC, &sock->flags))
 763                 set_bit(SOCK_SUPPORT_ZC, &newsock->flags);
 764         sock_graft(newsk, newsock);
 765
 766         newsock->state = SS_CONNECTED;
 767 }
 768
 769 /*
 770  *      Accept a pending connection. The TCP layer now gives BSD semantics.
 771  */
 772
 773 int inet_accept(struct socket *sock, struct socket *newsock, int flags,
 774                 bool kern)
 775 {
 776         struct sock *sk1 = sock->sk, *sk2;
 777         int err = -EINVAL;
 778
 779         /* IPV6_ADDRFORM can change sk->sk_prot under us. */
 780         sk2 = READ_ONCE(sk1->sk_prot)->accept(sk1, flags, &err, kern);
 781         if (!sk2)
 782                 return err;
 783
 784         lock_sock(sk2);
 785         __inet_accept(sock, newsock, sk2);
 786         release_sock(sk2);
 787         return 0;
 788 }
 789 EXPORT_SYMBOL(inet_accept);
 790
 791 /*
 792  *      This does both peername and sockname.
 793  */
 794 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
 795                  int peer)
 796 {
 797         struct sock *sk         = sock->sk;
 798         struct inet_sock *inet  = inet_sk(sk);
 799         DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
 800         int sin_addr_len = sizeof(*sin);
 801
 802         sin->sin_family = AF_INET;
 803         lock_sock(sk);
 804         if (peer) {
 805                 if (!inet->inet_dport ||
 806                     (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
 807                      peer == 1)) {
 808                         release_sock(sk);
 809                         return -ENOTCONN;
 810                 }
 811                 sin->sin_port = inet->inet_dport;
 812                 sin->sin_addr.s_addr = inet->inet_daddr;
 813                 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin, &sin_addr_len,
 814                                        CGROUP_INET4_GETPEERNAME);
 815         } else {
 816                 __be32 addr = inet->inet_rcv_saddr;
 817                 if (!addr)
 818                         addr = inet->inet_saddr;
 819                 sin->sin_port = inet->inet_sport;
 820                 sin->sin_addr.s_addr = addr;
 821                 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin, &sin_addr_len,
 822                                        CGROUP_INET4_GETSOCKNAME);
 823         }
 824         release_sock(sk);
 825         memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 826         return sin_addr_len;
 827 }
 828 EXPORT_SYMBOL(inet_getname);
 829
 830 int inet_send_prepare(struct sock *sk)
 831 {
 832         sock_rps_record_flow(sk);
 833
 834         /* We may need to bind the socket. */
 835         if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind &&
 836             inet_autobind(sk))
 837                 return -EAGAIN;
 838
 839         return 0;
 840 }
 841 EXPORT_SYMBOL_GPL(inet_send_prepare);
 842
 843 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
 844 {
 845         struct sock *sk = sock->sk;
 846
 847         if (unlikely(inet_send_prepare(sk)))
 848                 return -EAGAIN;
 849
 850         return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
 851                                sk, msg, size);
 852 }
 853 EXPORT_SYMBOL(inet_sendmsg);
 854
 855 void inet_splice_eof(struct socket *sock)
 856 {
 857         const struct proto *prot;
 858         struct sock *sk = sock->sk;
 859
 860         if (unlikely(inet_send_prepare(sk)))
 861                 return;
 862
 863         /* IPV6_ADDRFORM can change sk->sk_prot under us. */
 864         prot = READ_ONCE(sk->sk_prot);
 865         if (prot->splice_eof)
 866                 prot->splice_eof(sock);
 867 }
 868 EXPORT_SYMBOL_GPL(inet_splice_eof);
 869
 870 INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
 871                                           size_t, int, int *));
 872 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
 873                  int flags)
 874 {
 875         struct sock *sk = sock->sk;
 876         int addr_len = 0;
 877         int err;
 878
 879         if (likely(!(flags & MSG_ERRQUEUE)))
 880                 sock_rps_record_flow(sk);
 881
 882         err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
 883                               sk, msg, size, flags, &addr_len);
 884         if (err >= 0)
 885                 msg->msg_namelen = addr_len;
 886         return err;
 887 }
 888 EXPORT_SYMBOL(inet_recvmsg);
 889
 890 int inet_shutdown(struct socket *sock, int how)
 891 {
 892         struct sock *sk = sock->sk;
 893         int err = 0;
 894
 895         /* This should really check to make sure
 896          * the socket is a TCP socket. (WHY AC...)
 897          */
 898         how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
 899                        1->2 bit 2 snds.
 900                        2->3 */
 901         if ((how & ~SHUTDOWN_MASK) || !how)     /* MAXINT->0 */
 902                 return -EINVAL;
 903
 904         lock_sock(sk);
 905         if (sock->state == SS_CONNECTING) {
 906                 if ((1 << sk->sk_state) &
 907                     (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
 908                         sock->state = SS_DISCONNECTING;
 909                 else
 910                         sock->state = SS_CONNECTED;
 911         }
 912
 913         switch (sk->sk_state) {
 914         case TCP_CLOSE:
 915                 err = -ENOTCONN;
 916                 /* Hack to wake up other listeners, who can poll for
 917                    EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
 918                 fallthrough;
 919         default:
 920                 WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | how);
 921                 if (sk->sk_prot->shutdown)
 922                         sk->sk_prot->shutdown(sk, how);
 923                 break;
 924
 925         /* Remaining two branches are temporary solution for missing
 926          * close() in multithreaded environment. It is _not_ a good idea,
 927          * but we have no choice until close() is repaired at VFS level.
 928          */
 929         case TCP_LISTEN:
 930                 if (!(how & RCV_SHUTDOWN))
 931                         break;
 932                 fallthrough;
 933         case TCP_SYN_SENT:
 934                 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
 935                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 936                 break;
 937         }
 938
 939         /* Wake up anyone sleeping in poll. */
 940         sk->sk_state_change(sk);
 941         release_sock(sk);
 942         return err;
 943 }
 944 EXPORT_SYMBOL(inet_shutdown);
 945
 946 /*
 947  *      ioctl() calls you can issue on an INET socket. Most of these are
 948  *      device configuration and stuff and very rarely used. Some ioctls
 949  *      pass on to the socket itself.
 950  *
 951  *      NOTE: I like the idea of a module for the config stuff. ie ifconfig
 952  *      loads the devconfigure module does its configuring and unloads it.
 953  *      There's a good 20K of config code hanging around the kernel.
 954  */
 955
 956 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 957 {
 958         struct sock *sk = sock->sk;
 959         int err = 0;
 960         struct net *net = sock_net(sk);
 961         void __user *p = (void __user *)arg;
 962         struct ifreq ifr;
 963         struct rtentry rt;
 964
 965         switch (cmd) {
 966         case SIOCADDRT:
 967         case SIOCDELRT:
 968                 if (copy_from_user(&rt, p, sizeof(struct rtentry)))
 969                         return -EFAULT;
 970                 err = ip_rt_ioctl(net, cmd, &rt);
 971                 break;
 972         case SIOCRTMSG:
 973                 err = -EINVAL;
 974                 break;
 975         case SIOCDARP:
 976         case SIOCGARP:
 977         case SIOCSARP:
 978                 err = arp_ioctl(net, cmd, (void __user *)arg);
 979                 break;
 980         case SIOCGIFADDR:
 981         case SIOCGIFBRDADDR:
 982         case SIOCGIFNETMASK:
 983         case SIOCGIFDSTADDR:
 984         case SIOCGIFPFLAGS:
 985                 if (get_user_ifreq(&ifr, NULL, p))
 986                         return -EFAULT;
 987                 err = devinet_ioctl(net, cmd, &ifr);
 988                 if (!err && put_user_ifreq(&ifr, p))
 989                         err = -EFAULT;
 990                 break;
 991
 992         case SIOCSIFADDR:
 993         case SIOCSIFBRDADDR:
 994         case SIOCSIFNETMASK:
 995         case SIOCSIFDSTADDR:
 996         case SIOCSIFPFLAGS:
 997         case SIOCSIFFLAGS:
 998                 if (get_user_ifreq(&ifr, NULL, p))
 999                         return -EFAULT;
1000                 err = devinet_ioctl(net, cmd, &ifr);
1001                 break;
1002         default:
1003                 if (sk->sk_prot->ioctl)
1004                         err = sk_ioctl(sk, cmd, (void __user *)arg);
1005                 else
1006                         err = -ENOIOCTLCMD;
1007                 break;
1008         }
1009         return err;
1010 }
1011 EXPORT_SYMBOL(inet_ioctl);
1012
1013 #ifdef CONFIG_COMPAT
1014 static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd,
1015                 struct compat_rtentry __user *ur)
1016 {
1017         compat_uptr_t rtdev;
1018         struct rtentry rt;
1019
1020         if (copy_from_user(&rt.rt_dst, &ur->rt_dst,
1021                         3 * sizeof(struct sockaddr)) ||
1022             get_user(rt.rt_flags, &ur->rt_flags) ||
1023             get_user(rt.rt_metric, &ur->rt_metric) ||
1024             get_user(rt.rt_mtu, &ur->rt_mtu) ||
1025             get_user(rt.rt_window, &ur->rt_window) ||
1026             get_user(rt.rt_irtt, &ur->rt_irtt) ||
1027             get_user(rtdev, &ur->rt_dev))
1028                 return -EFAULT;
1029
1030         rt.rt_dev = compat_ptr(rtdev);
1031         return ip_rt_ioctl(sock_net(sk), cmd, &rt);
1032 }
1033
1034 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1035 {
1036         void __user *argp = compat_ptr(arg);
1037         struct sock *sk = sock->sk;
1038
1039         switch (cmd) {
1040         case SIOCADDRT:
1041         case SIOCDELRT:
1042                 return inet_compat_routing_ioctl(sk, cmd, argp);
1043         default:
1044                 if (!sk->sk_prot->compat_ioctl)
1045                         return -ENOIOCTLCMD;
1046                 return sk->sk_prot->compat_ioctl(sk, cmd, arg);
1047         }
1048 }
1049 #endif /* CONFIG_COMPAT */
1050
1051 const struct proto_ops inet_stream_ops = {
1052         .family            = PF_INET,
1053         .owner             = THIS_MODULE,
1054         .release           = inet_release,
1055         .bind              = inet_bind,
1056         .connect           = inet_stream_connect,
1057         .socketpair        = sock_no_socketpair,
1058         .accept            = inet_accept,
1059         .getname           = inet_getname,
1060         .poll              = tcp_poll,
1061         .ioctl             = inet_ioctl,
1062         .gettstamp         = sock_gettstamp,
1063         .listen            = inet_listen,
1064         .shutdown          = inet_shutdown,
1065         .setsockopt        = sock_common_setsockopt,
1066         .getsockopt        = sock_common_getsockopt,
1067         .sendmsg           = inet_sendmsg,
1068         .recvmsg           = inet_recvmsg,
1069 #ifdef CONFIG_MMU
1070         .mmap              = tcp_mmap,
1071 #endif
1072         .splice_eof        = inet_splice_eof,
1073         .splice_read       = tcp_splice_read,
1074         .read_sock         = tcp_read_sock,
1075         .read_skb          = tcp_read_skb,
1076         .sendmsg_locked    = tcp_sendmsg_locked,
1077         .peek_len          = tcp_peek_len,
1078 #ifdef CONFIG_COMPAT
1079         .compat_ioctl      = inet_compat_ioctl,
1080 #endif
1081         .set_rcvlowat      = tcp_set_rcvlowat,
1082 };
1083 EXPORT_SYMBOL(inet_stream_ops);
1084
1085 const struct proto_ops inet_dgram_ops = {
1086         .family            = PF_INET,
1087         .owner             = THIS_MODULE,
1088         .release           = inet_release,
1089         .bind              = inet_bind,
1090         .connect           = inet_dgram_connect,
1091         .socketpair        = sock_no_socketpair,
1092         .accept            = sock_no_accept,
1093         .getname           = inet_getname,
1094         .poll              = udp_poll,
1095         .ioctl             = inet_ioctl,
1096         .gettstamp         = sock_gettstamp,
1097         .listen            = sock_no_listen,
1098         .shutdown          = inet_shutdown,
1099         .setsockopt        = sock_common_setsockopt,
1100         .getsockopt        = sock_common_getsockopt,
1101         .sendmsg           = inet_sendmsg,
1102         .read_skb          = udp_read_skb,
1103         .recvmsg           = inet_recvmsg,
1104         .mmap              = sock_no_mmap,
1105         .splice_eof        = inet_splice_eof,
1106         .set_peek_off      = sk_set_peek_off,
1107 #ifdef CONFIG_COMPAT
1108         .compat_ioctl      = inet_compat_ioctl,
1109 #endif
1110 };
1111 EXPORT_SYMBOL(inet_dgram_ops);
1112
1113 /*
1114  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1115  * udp_poll
1116  */
1117 static const struct proto_ops inet_sockraw_ops = {
1118         .family            = PF_INET,
1119         .owner             = THIS_MODULE,
1120         .release           = inet_release,
1121         .bind              = inet_bind,
1122         .connect           = inet_dgram_connect,
1123         .socketpair        = sock_no_socketpair,
1124         .accept            = sock_no_accept,
1125         .getname           = inet_getname,
1126         .poll              = datagram_poll,
1127         .ioctl             = inet_ioctl,
1128         .gettstamp         = sock_gettstamp,
1129         .listen            = sock_no_listen,
1130         .shutdown          = inet_shutdown,
1131         .setsockopt        = sock_common_setsockopt,
1132         .getsockopt        = sock_common_getsockopt,
1133         .sendmsg           = inet_sendmsg,
1134         .recvmsg           = inet_recvmsg,
1135         .mmap              = sock_no_mmap,
1136         .splice_eof        = inet_splice_eof,
1137 #ifdef CONFIG_COMPAT
1138         .compat_ioctl      = inet_compat_ioctl,
1139 #endif
1140 };
1141
1142 static const struct net_proto_family inet_family_ops = {
1143         .family = PF_INET,
1144         .create = inet_create,
1145         .owner  = THIS_MODULE,
1146 };
1147
1148 /* Upon startup we insert all the elements in inetsw_array[] into
1149  * the linked list inetsw.
1150  */
1151 static struct inet_protosw inetsw_array[] =
1152 {
1153         {
1154                 .type =       SOCK_STREAM,
1155                 .protocol =   IPPROTO_TCP,
1156                 .prot =       &tcp_prot,
1157                 .ops =        &inet_stream_ops,
1158                 .flags =      INET_PROTOSW_PERMANENT |
1159                               INET_PROTOSW_ICSK,
1160         },
1161
1162         {
1163                 .type =       SOCK_DGRAM,
1164                 .protocol =   IPPROTO_UDP,
1165                 .prot =       &udp_prot,
1166                 .ops =        &inet_dgram_ops,
1167                 .flags =      INET_PROTOSW_PERMANENT,
1168        },
1169
1170        {
1171                 .type =       SOCK_DGRAM,
1172                 .protocol =   IPPROTO_ICMP,
1173                 .prot =       &ping_prot,
1174                 .ops =        &inet_sockraw_ops,
1175                 .flags =      INET_PROTOSW_REUSE,
1176        },
1177
1178        {
1179                .type =       SOCK_RAW,
1180                .protocol =   IPPROTO_IP,        /* wild card */
1181                .prot =       &raw_prot,
1182                .ops =        &inet_sockraw_ops,
1183                .flags =      INET_PROTOSW_REUSE,
1184        }
1185 };
1186
1187 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1188
1189 void inet_register_protosw(struct inet_protosw *p)
1190 {
1191         struct list_head *lh;
1192         struct inet_protosw *answer;
1193         int protocol = p->protocol;
1194         struct list_head *last_perm;
1195
1196         spin_lock_bh(&inetsw_lock);
1197
1198         if (p->type >= SOCK_MAX)
1199                 goto out_illegal;
1200
1201         /* If we are trying to override a permanent protocol, bail. */
1202         last_perm = &inetsw[p->type];
1203         list_for_each(lh, &inetsw[p->type]) {
1204                 answer = list_entry(lh, struct inet_protosw, list);
1205                 /* Check only the non-wild match. */
1206                 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1207                         break;
1208                 if (protocol == answer->protocol)
1209                         goto out_permanent;
1210                 last_perm = lh;
1211         }
1212
1213         /* Add the new entry after the last permanent entry if any, so that
1214          * the new entry does not override a permanent entry when matched with
1215          * a wild-card protocol. But it is allowed to override any existing
1216          * non-permanent entry.  This means that when we remove this entry, the
1217          * system automatically returns to the old behavior.
1218          */
1219         list_add_rcu(&p->list, last_perm);
1220 out:
1221         spin_unlock_bh(&inetsw_lock);
1222
1223         return;
1224
1225 out_permanent:
1226         pr_err("Attempt to override permanent protocol %d\n", protocol);
1227         goto out;
1228
1229 out_illegal:
1230         pr_err("Ignoring attempt to register invalid socket type %d\n",
1231                p->type);
1232         goto out;
1233 }
1234 EXPORT_SYMBOL(inet_register_protosw);
1235
1236 void inet_unregister_protosw(struct inet_protosw *p)
1237 {
1238         if (INET_PROTOSW_PERMANENT & p->flags) {
1239                 pr_err("Attempt to unregister permanent protocol %d\n",
1240                        p->protocol);
1241         } else {
1242                 spin_lock_bh(&inetsw_lock);
1243                 list_del_rcu(&p->list);
1244                 spin_unlock_bh(&inetsw_lock);
1245
1246                 synchronize_net();
1247         }
1248 }
1249 EXPORT_SYMBOL(inet_unregister_protosw);
1250
1251 static int inet_sk_reselect_saddr(struct sock *sk)
1252 {
1253         struct inet_sock *inet = inet_sk(sk);
1254         __be32 old_saddr = inet->inet_saddr;
1255         __be32 daddr = inet->inet_daddr;
1256         struct flowi4 *fl4;
1257         struct rtable *rt;
1258         __be32 new_saddr;
1259         struct ip_options_rcu *inet_opt;
1260         int err;
1261
1262         inet_opt = rcu_dereference_protected(inet->inet_opt,
1263                                              lockdep_sock_is_held(sk));
1264         if (inet_opt && inet_opt->opt.srr)
1265                 daddr = inet_opt->opt.faddr;
1266
1267         /* Query new route. */
1268         fl4 = &inet->cork.fl.u.ip4;
1269         rt = ip_route_connect(fl4, daddr, 0, sk->sk_bound_dev_if,
1270                               sk->sk_protocol, inet->inet_sport,
1271                               inet->inet_dport, sk);
1272         if (IS_ERR(rt))
1273                 return PTR_ERR(rt);
1274
1275         new_saddr = fl4->saddr;
1276
1277         if (new_saddr == old_saddr) {
1278                 sk_setup_caps(sk, &rt->dst);
1279                 return 0;
1280         }
1281
1282         err = inet_bhash2_update_saddr(sk, &new_saddr, AF_INET);
1283         if (err) {
1284                 ip_rt_put(rt);
1285                 return err;
1286         }
1287
1288         sk_setup_caps(sk, &rt->dst);
1289
1290         if (READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) > 1) {
1291                 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1292                         __func__, &old_saddr, &new_saddr);
1293         }
1294
1295         /*
1296          * XXX The only one ugly spot where we need to
1297          * XXX really change the sockets identity after
1298          * XXX it has entered the hashes. -DaveM
1299          *
1300          * Besides that, it does not check for connection
1301          * uniqueness. Wait for troubles.
1302          */
1303         return __sk_prot_rehash(sk);
1304 }
1305
1306 int inet_sk_rebuild_header(struct sock *sk)
1307 {
1308         struct inet_sock *inet = inet_sk(sk);
1309         struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1310         __be32 daddr;
1311         struct ip_options_rcu *inet_opt;
1312         struct flowi4 *fl4;
1313         int err;
1314
1315         /* Route is OK, nothing to do. */
1316         if (rt)
1317                 return 0;
1318
1319         /* Reroute. */
1320         rcu_read_lock();
1321         inet_opt = rcu_dereference(inet->inet_opt);
1322         daddr = inet->inet_daddr;
1323         if (inet_opt && inet_opt->opt.srr)
1324                 daddr = inet_opt->opt.faddr;
1325         rcu_read_unlock();
1326         fl4 = &inet->cork.fl.u.ip4;
1327         rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1328                                    inet->inet_dport, inet->inet_sport,
1329                                    sk->sk_protocol, RT_CONN_FLAGS(sk),
1330                                    sk->sk_bound_dev_if);
1331         if (!IS_ERR(rt)) {
1332                 err = 0;
1333                 sk_setup_caps(sk, &rt->dst);
1334         } else {
1335                 err = PTR_ERR(rt);
1336
1337                 /* Routing failed... */
1338                 sk->sk_route_caps = 0;
1339                 /*
1340                  * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1341                  * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1342                  */
1343                 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) ||
1344                     sk->sk_state != TCP_SYN_SENT ||
1345                     (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1346                     (err = inet_sk_reselect_saddr(sk)) != 0)
1347                         WRITE_ONCE(sk->sk_err_soft, -err);
1348         }
1349
1350         return err;
1351 }
1352 EXPORT_SYMBOL(inet_sk_rebuild_header);
1353
1354 void inet_sk_set_state(struct sock *sk, int state)
1355 {
1356         trace_inet_sock_set_state(sk, sk->sk_state, state);
1357         sk->sk_state = state;
1358 }
1359 EXPORT_SYMBOL(inet_sk_set_state);
1360
1361 void inet_sk_state_store(struct sock *sk, int newstate)
1362 {
1363         trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1364         smp_store_release(&sk->sk_state, newstate);
1365 }
1366
1367 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1368                                  netdev_features_t features)
1369 {
1370         bool udpfrag = false, fixedid = false, gso_partial, encap;
1371         struct sk_buff *segs = ERR_PTR(-EINVAL);
1372         const struct net_offload *ops;
1373         unsigned int offset = 0;
1374         struct iphdr *iph;
1375         int proto, tot_len;
1376         int nhoff;
1377         int ihl;
1378         int id;
1379
1380         skb_reset_network_header(skb);
1381         nhoff = skb_network_header(skb) - skb_mac_header(skb);
1382         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1383                 goto out;
1384
1385         iph = ip_hdr(skb);
1386         ihl = iph->ihl * 4;
1387         if (ihl < sizeof(*iph))
1388                 goto out;
1389
1390         id = ntohs(iph->id);
1391         proto = iph->protocol;
1392
1393         /* Warning: after this point, iph might be no longer valid */
1394         if (unlikely(!pskb_may_pull(skb, ihl)))
1395                 goto out;
1396         __skb_pull(skb, ihl);
1397
1398         encap = SKB_GSO_CB(skb)->encap_level > 0;
1399         if (encap)
1400                 features &= skb->dev->hw_enc_features;
1401         SKB_GSO_CB(skb)->encap_level += ihl;
1402
1403         skb_reset_transport_header(skb);
1404
1405         segs = ERR_PTR(-EPROTONOSUPPORT);
1406
1407         if (!skb->encapsulation || encap) {
1408                 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1409                 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1410
1411                 /* fixed ID is invalid if DF bit is not set */
1412                 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1413                         goto out;
1414         }
1415
1416         ops = rcu_dereference(inet_offloads[proto]);
1417         if (likely(ops && ops->callbacks.gso_segment)) {
1418                 segs = ops->callbacks.gso_segment(skb, features);
1419                 if (!segs)
1420                         skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
1421         }
1422
1423         if (IS_ERR_OR_NULL(segs))
1424                 goto out;
1425
1426         gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1427
1428         skb = segs;
1429         do {
1430                 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1431                 if (udpfrag) {
1432                         iph->frag_off = htons(offset >> 3);
1433                         if (skb->next)
1434                                 iph->frag_off |= htons(IP_MF);
1435                         offset += skb->len - nhoff - ihl;
1436                         tot_len = skb->len - nhoff;
1437                 } else if (skb_is_gso(skb)) {
1438                         if (!fixedid) {
1439                                 iph->id = htons(id);
1440                                 id += skb_shinfo(skb)->gso_segs;
1441                         }
1442
1443                         if (gso_partial)
1444                                 tot_len = skb_shinfo(skb)->gso_size +
1445                                           SKB_GSO_CB(skb)->data_offset +
1446                                           skb->head - (unsigned char *)iph;
1447                         else
1448                                 tot_len = skb->len - nhoff;
1449                 } else {
1450                         if (!fixedid)
1451                                 iph->id = htons(id++);
1452                         tot_len = skb->len - nhoff;
1453                 }
1454                 iph->tot_len = htons(tot_len);
1455                 ip_send_check(iph);
1456                 if (encap)
1457                         skb_reset_inner_headers(skb);
1458                 skb->network_header = (u8 *)iph - skb->head;
1459                 skb_reset_mac_len(skb);
1460         } while ((skb = skb->next));
1461
1462 out:
1463         return segs;
1464 }
1465
1466 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1467                                         netdev_features_t features)
1468 {
1469         if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1470                 return ERR_PTR(-EINVAL);
1471
1472         return inet_gso_segment(skb, features);
1473 }
1474
1475 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1476 {
1477         const struct net_offload *ops;
1478         struct sk_buff *pp = NULL;
1479         const struct iphdr *iph;
1480         struct sk_buff *p;
1481         unsigned int hlen;
1482         unsigned int off;
1483         unsigned int id;
1484         int flush = 1;
1485         int proto;
1486
1487         off = skb_gro_offset(skb);
1488         hlen = off + sizeof(*iph);
1489         iph = skb_gro_header(skb, hlen, off);
1490         if (unlikely(!iph))
1491                 goto out;
1492
1493         proto = iph->protocol;
1494
1495         ops = rcu_dereference(inet_offloads[proto]);
1496         if (!ops || !ops->callbacks.gro_receive)
1497                 goto out;
1498
1499         if (*(u8 *)iph != 0x45)
1500                 goto out;
1501
1502         if (ip_is_fragment(iph))
1503                 goto out;
1504
1505         if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1506                 goto out;
1507
1508         NAPI_GRO_CB(skb)->proto = proto;
1509         id = ntohl(*(__be32 *)&iph->id);
1510         flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1511         id >>= 16;
1512
1513         list_for_each_entry(p, head, list) {
1514                 struct iphdr *iph2;
1515                 u16 flush_id;
1516
1517                 if (!NAPI_GRO_CB(p)->same_flow)
1518                         continue;
1519
1520                 iph2 = (struct iphdr *)(p->data + off);
1521                 /* The above works because, with the exception of the top
1522                  * (inner most) layer, we only aggregate pkts with the same
1523                  * hdr length so all the hdrs we'll need to verify will start
1524                  * at the same offset.
1525                  */
1526                 if ((iph->protocol ^ iph2->protocol) |
1527                     ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1528                     ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1529                         NAPI_GRO_CB(p)->same_flow = 0;
1530                         continue;
1531                 }
1532
1533                 /* All fields must match except length and checksum. */
1534                 NAPI_GRO_CB(p)->flush |=
1535                         (iph->ttl ^ iph2->ttl) |
1536                         (iph->tos ^ iph2->tos) |
1537                         ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1538
1539                 NAPI_GRO_CB(p)->flush |= flush;
1540
1541                 /* We need to store of the IP ID check to be included later
1542                  * when we can verify that this packet does in fact belong
1543                  * to a given flow.
1544                  */
1545                 flush_id = (u16)(id - ntohs(iph2->id));
1546
1547                 /* This bit of code makes it much easier for us to identify
1548                  * the cases where we are doing atomic vs non-atomic IP ID
1549                  * checks.  Specifically an atomic check can return IP ID
1550                  * values 0 - 0xFFFF, while a non-atomic check can only
1551                  * return 0 or 0xFFFF.
1552                  */
1553                 if (!NAPI_GRO_CB(p)->is_atomic ||
1554                     !(iph->frag_off & htons(IP_DF))) {
1555                         flush_id ^= NAPI_GRO_CB(p)->count;
1556                         flush_id = flush_id ? 0xFFFF : 0;
1557                 }
1558
1559                 /* If the previous IP ID value was based on an atomic
1560                  * datagram we can overwrite the value and ignore it.
1561                  */
1562                 if (NAPI_GRO_CB(skb)->is_atomic)
1563                         NAPI_GRO_CB(p)->flush_id = flush_id;
1564                 else
1565                         NAPI_GRO_CB(p)->flush_id |= flush_id;
1566         }
1567
1568         NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1569         NAPI_GRO_CB(skb)->flush |= flush;
1570         skb_set_network_header(skb, off);
1571         /* The above will be needed by the transport layer if there is one
1572          * immediately following this IP hdr.
1573          */
1574
1575         /* Note : No need to call skb_gro_postpull_rcsum() here,
1576          * as we already checked checksum over ipv4 header was 0
1577          */
1578         skb_gro_pull(skb, sizeof(*iph));
1579         skb_set_transport_header(skb, skb_gro_offset(skb));
1580
1581         pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1582                                        ops->callbacks.gro_receive, head, skb);
1583
1584 out:
1585         skb_gro_flush_final(skb, pp, flush);
1586
1587         return pp;
1588 }
1589
1590 static struct sk_buff *ipip_gro_receive(struct list_head *head,
1591                                         struct sk_buff *skb)
1592 {
1593         if (NAPI_GRO_CB(skb)->encap_mark) {
1594                 NAPI_GRO_CB(skb)->flush = 1;
1595                 return NULL;
1596         }
1597
1598         NAPI_GRO_CB(skb)->encap_mark = 1;
1599
1600         return inet_gro_receive(head, skb);
1601 }
1602
1603 #define SECONDS_PER_DAY 86400
1604
1605 /* inet_current_timestamp - Return IP network timestamp
1606  *
1607  * Return milliseconds since midnight in network byte order.
1608  */
1609 __be32 inet_current_timestamp(void)
1610 {
1611         u32 secs;
1612         u32 msecs;
1613         struct timespec64 ts;
1614
1615         ktime_get_real_ts64(&ts);
1616
1617         /* Get secs since midnight. */
1618         (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1619         /* Convert to msecs. */
1620         msecs = secs * MSEC_PER_SEC;
1621         /* Convert nsec to msec. */
1622         msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1623
1624         /* Convert to network byte order. */
1625         return htonl(msecs);
1626 }
1627 EXPORT_SYMBOL(inet_current_timestamp);
1628
1629 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1630 {
1631         unsigned int family = READ_ONCE(sk->sk_family);
1632
1633         if (family == AF_INET)
1634                 return ip_recv_error(sk, msg, len, addr_len);
1635 #if IS_ENABLED(CONFIG_IPV6)
1636         if (family == AF_INET6)
1637                 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1638 #endif
1639         return -EINVAL;
1640 }
1641 EXPORT_SYMBOL(inet_recv_error);
1642
1643 int inet_gro_complete(struct sk_buff *skb, int nhoff)
1644 {
1645         struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1646         const struct net_offload *ops;
1647         __be16 totlen = iph->tot_len;
1648         int proto = iph->protocol;
1649         int err = -ENOSYS;
1650
1651         if (skb->encapsulation) {
1652                 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1653                 skb_set_inner_network_header(skb, nhoff);
1654         }
1655
1656         iph_set_totlen(iph, skb->len - nhoff);
1657         csum_replace2(&iph->check, totlen, iph->tot_len);
1658
1659         ops = rcu_dereference(inet_offloads[proto]);
1660         if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1661                 goto out;
1662
1663         /* Only need to add sizeof(*iph) to get to the next hdr below
1664          * because any hdr with option will have been flushed in
1665          * inet_gro_receive().
1666          */
1667         err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1668                               tcp4_gro_complete, udp4_gro_complete,
1669                               skb, nhoff + sizeof(*iph));
1670
1671 out:
1672         return err;
1673 }
1674
1675 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1676 {
1677         skb->encapsulation = 1;
1678         skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1679         return inet_gro_complete(skb, nhoff);
1680 }
1681
1682 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1683                          unsigned short type, unsigned char protocol,
1684                          struct net *net)
1685 {
1686         struct socket *sock;
1687         int rc = sock_create_kern(net, family, type, protocol, &sock);
1688
1689         if (rc == 0) {
1690                 *sk = sock->sk;
1691                 (*sk)->sk_allocation = GFP_ATOMIC;
1692                 (*sk)->sk_use_task_frag = false;
1693                 /*
1694                  * Unhash it so that IP input processing does not even see it,
1695                  * we do not wish this socket to see incoming packets.
1696                  */
1697                 (*sk)->sk_prot->unhash(*sk);
1698         }
1699         return rc;
1700 }
1701 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1702
1703 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1704 {
1705         unsigned long res = 0;
1706         int i;
1707
1708         for_each_possible_cpu(i)
1709                 res += snmp_get_cpu_field(mib, i, offt);
1710         return res;
1711 }
1712 EXPORT_SYMBOL_GPL(snmp_fold_field);
1713
1714 #if BITS_PER_LONG==32
1715
1716 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1717                          size_t syncp_offset)
1718 {
1719         void *bhptr;
1720         struct u64_stats_sync *syncp;
1721         u64 v;
1722         unsigned int start;
1723
1724         bhptr = per_cpu_ptr(mib, cpu);
1725         syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1726         do {
1727                 start = u64_stats_fetch_begin(syncp);
1728                 v = *(((u64 *)bhptr) + offt);
1729         } while (u64_stats_fetch_retry(syncp, start));
1730
1731         return v;
1732 }
1733 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1734
1735 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1736 {
1737         u64 res = 0;
1738         int cpu;
1739
1740         for_each_possible_cpu(cpu) {
1741                 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1742         }
1743         return res;
1744 }
1745 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1746 #endif
1747
1748 #ifdef CONFIG_IP_MULTICAST
1749 static const struct net_protocol igmp_protocol = {
1750         .handler =      igmp_rcv,
1751 };
1752 #endif
1753
1754 static const struct net_protocol tcp_protocol = {
1755         .handler        =       tcp_v4_rcv,
1756         .err_handler    =       tcp_v4_err,
1757         .no_policy      =       1,
1758         .icmp_strict_tag_validation = 1,
1759 };
1760
1761 static const struct net_protocol udp_protocol = {
1762         .handler =      udp_rcv,
1763         .err_handler =  udp_err,
1764         .no_policy =    1,
1765 };
1766
1767 static const struct net_protocol icmp_protocol = {
1768         .handler =      icmp_rcv,
1769         .err_handler =  icmp_err,
1770         .no_policy =    1,
1771 };
1772
1773 static __net_init int ipv4_mib_init_net(struct net *net)
1774 {
1775         int i;
1776
1777         net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1778         if (!net->mib.tcp_statistics)
1779                 goto err_tcp_mib;
1780         net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1781         if (!net->mib.ip_statistics)
1782                 goto err_ip_mib;
1783
1784         for_each_possible_cpu(i) {
1785                 struct ipstats_mib *af_inet_stats;
1786                 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1787                 u64_stats_init(&af_inet_stats->syncp);
1788         }
1789
1790         net->mib.net_statistics = alloc_percpu(struct linux_mib);
1791         if (!net->mib.net_statistics)
1792                 goto err_net_mib;
1793         net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1794         if (!net->mib.udp_statistics)
1795                 goto err_udp_mib;
1796         net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1797         if (!net->mib.udplite_statistics)
1798                 goto err_udplite_mib;
1799         net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1800         if (!net->mib.icmp_statistics)
1801                 goto err_icmp_mib;
1802         net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1803                                               GFP_KERNEL);
1804         if (!net->mib.icmpmsg_statistics)
1805                 goto err_icmpmsg_mib;
1806
1807         tcp_mib_init(net);
1808         return 0;
1809
1810 err_icmpmsg_mib:
1811         free_percpu(net->mib.icmp_statistics);
1812 err_icmp_mib:
1813         free_percpu(net->mib.udplite_statistics);
1814 err_udplite_mib:
1815         free_percpu(net->mib.udp_statistics);
1816 err_udp_mib:
1817         free_percpu(net->mib.net_statistics);
1818 err_net_mib:
1819         free_percpu(net->mib.ip_statistics);
1820 err_ip_mib:
1821         free_percpu(net->mib.tcp_statistics);
1822 err_tcp_mib:
1823         return -ENOMEM;
1824 }
1825
1826 static __net_exit void ipv4_mib_exit_net(struct net *net)
1827 {
1828         kfree(net->mib.icmpmsg_statistics);
1829         free_percpu(net->mib.icmp_statistics);
1830         free_percpu(net->mib.udplite_statistics);
1831         free_percpu(net->mib.udp_statistics);
1832         free_percpu(net->mib.net_statistics);
1833         free_percpu(net->mib.ip_statistics);
1834         free_percpu(net->mib.tcp_statistics);
1835 #ifdef CONFIG_MPTCP
1836         /* allocated on demand, see mptcp_init_sock() */
1837         free_percpu(net->mib.mptcp_statistics);
1838 #endif
1839 }
1840
1841 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1842         .init = ipv4_mib_init_net,
1843         .exit = ipv4_mib_exit_net,
1844 };
1845
1846 static int __init init_ipv4_mibs(void)
1847 {
1848         return register_pernet_subsys(&ipv4_mib_ops);
1849 }
1850
1851 static __net_init int inet_init_net(struct net *net)
1852 {
1853         /*
1854          * Set defaults for local port range
1855          */
1856         seqlock_init(&net->ipv4.ip_local_ports.lock);
1857         net->ipv4.ip_local_ports.range[0] =  32768;
1858         net->ipv4.ip_local_ports.range[1] =  60999;
1859
1860         seqlock_init(&net->ipv4.ping_group_range.lock);
1861         /*
1862          * Sane defaults - nobody may create ping sockets.
1863          * Boot scripts should set this to distro-specific group.
1864          */
1865         net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1866         net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1867
1868         /* Default values for sysctl-controlled parameters.
1869          * We set them here, in case sysctl is not compiled.
1870          */
1871         net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1872         net->ipv4.sysctl_ip_fwd_update_priority = 1;
1873         net->ipv4.sysctl_ip_dynaddr = 0;
1874         net->ipv4.sysctl_ip_early_demux = 1;
1875         net->ipv4.sysctl_udp_early_demux = 1;
1876         net->ipv4.sysctl_tcp_early_demux = 1;
1877         net->ipv4.sysctl_nexthop_compat_mode = 1;
1878 #ifdef CONFIG_SYSCTL
1879         net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1880 #endif
1881
1882         /* Some igmp sysctl, whose values are always used */
1883         net->ipv4.sysctl_igmp_max_memberships = 20;
1884         net->ipv4.sysctl_igmp_max_msf = 10;
1885         /* IGMP reports for link-local multicast groups are enabled by default */
1886         net->ipv4.sysctl_igmp_llm_reports = 1;
1887         net->ipv4.sysctl_igmp_qrv = 2;
1888
1889         net->ipv4.sysctl_fib_notify_on_flag_change = 0;
1890
1891         return 0;
1892 }
1893
1894 static __net_initdata struct pernet_operations af_inet_ops = {
1895         .init = inet_init_net,
1896 };
1897
1898 static int __init init_inet_pernet_ops(void)
1899 {
1900         return register_pernet_subsys(&af_inet_ops);
1901 }
1902
1903 static int ipv4_proc_init(void);
1904
1905 /*
1906  *      IP protocol layer initialiser
1907  */
1908
1909 static struct packet_offload ip_packet_offload __read_mostly = {
1910         .type = cpu_to_be16(ETH_P_IP),
1911         .callbacks = {
1912                 .gso_segment = inet_gso_segment,
1913                 .gro_receive = inet_gro_receive,
1914                 .gro_complete = inet_gro_complete,
1915         },
1916 };
1917
1918 static const struct net_offload ipip_offload = {
1919         .callbacks = {
1920                 .gso_segment    = ipip_gso_segment,
1921                 .gro_receive    = ipip_gro_receive,
1922                 .gro_complete   = ipip_gro_complete,
1923         },
1924 };
1925
1926 static int __init ipip_offload_init(void)
1927 {
1928         return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1929 }
1930
1931 static int __init ipv4_offload_init(void)
1932 {
1933         /*
1934          * Add offloads
1935          */
1936         if (udpv4_offload_init() < 0)
1937                 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1938         if (tcpv4_offload_init() < 0)
1939                 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1940         if (ipip_offload_init() < 0)
1941                 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1942
1943         dev_add_offload(&ip_packet_offload);
1944         return 0;
1945 }
1946
1947 fs_initcall(ipv4_offload_init);
1948
1949 static struct packet_type ip_packet_type __read_mostly = {
1950         .type = cpu_to_be16(ETH_P_IP),
1951         .func = ip_rcv,
1952         .list_func = ip_list_rcv,
1953 };
1954
1955 static int __init inet_init(void)
1956 {
1957         struct inet_protosw *q;
1958         struct list_head *r;
1959         int rc;
1960
1961         sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1962
1963         raw_hashinfo_init(&raw_v4_hashinfo);
1964
1965         rc = proto_register(&tcp_prot, 1);
1966         if (rc)
1967                 goto out;
1968
1969         rc = proto_register(&udp_prot, 1);
1970         if (rc)
1971                 goto out_unregister_tcp_proto;
1972
1973         rc = proto_register(&raw_prot, 1);
1974         if (rc)
1975                 goto out_unregister_udp_proto;
1976
1977         rc = proto_register(&ping_prot, 1);
1978         if (rc)
1979                 goto out_unregister_raw_proto;
1980
1981         /*
1982          *      Tell SOCKET that we are alive...
1983          */
1984
1985         (void)sock_register(&inet_family_ops);
1986
1987 #ifdef CONFIG_SYSCTL
1988         ip_static_sysctl_init();
1989 #endif
1990
1991         /*
1992          *      Add all the base protocols.
1993          */
1994
1995         if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1996                 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1997         if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1998                 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1999         if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
2000                 pr_crit("%s: Cannot add TCP protocol\n", __func__);
2001 #ifdef CONFIG_IP_MULTICAST
2002         if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
2003                 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
2004 #endif
2005
2006         /* Register the socket-side information for inet_create. */
2007         for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
2008                 INIT_LIST_HEAD(r);
2009
2010         for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
2011                 inet_register_protosw(q);
2012
2013         /*
2014          *      Set the ARP module up
2015          */
2016
2017         arp_init();
2018
2019         /*
2020          *      Set the IP module up
2021          */
2022
2023         ip_init();
2024
2025         /* Initialise per-cpu ipv4 mibs */
2026         if (init_ipv4_mibs())
2027                 panic("%s: Cannot init ipv4 mibs\n", __func__);
2028
2029         /* Setup TCP slab cache for open requests. */
2030         tcp_init();
2031
2032         /* Setup UDP memory threshold */
2033         udp_init();
2034
2035         /* Add UDP-Lite (RFC 3828) */
2036         udplite4_register();
2037
2038         raw_init();
2039
2040         ping_init();
2041
2042         /*
2043          *      Set the ICMP layer up
2044          */
2045
2046         if (icmp_init() < 0)
2047                 panic("Failed to create the ICMP control socket.\n");
2048
2049         /*
2050          *      Initialise the multicast router
2051          */
2052 #if defined(CONFIG_IP_MROUTE)
2053         if (ip_mr_init())
2054                 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2055 #endif
2056
2057         if (init_inet_pernet_ops())
2058                 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2059
2060         ipv4_proc_init();
2061
2062         ipfrag_init();
2063
2064         dev_add_pack(&ip_packet_type);
2065
2066         ip_tunnel_core_init();
2067
2068         rc = 0;
2069 out:
2070         return rc;
2071 out_unregister_raw_proto:
2072         proto_unregister(&raw_prot);
2073 out_unregister_udp_proto:
2074         proto_unregister(&udp_prot);
2075 out_unregister_tcp_proto:
2076         proto_unregister(&tcp_prot);
2077         goto out;
2078 }
2079
2080 fs_initcall(inet_init);
2081
2082 /* ------------------------------------------------------------------------ */
2083
2084 #ifdef CONFIG_PROC_FS
2085 static int __init ipv4_proc_init(void)
2086 {
2087         int rc = 0;
2088
2089         if (raw_proc_init())
2090                 goto out_raw;
2091         if (tcp4_proc_init())
2092                 goto out_tcp;
2093         if (udp4_proc_init())
2094                 goto out_udp;
2095         if (ping_proc_init())
2096                 goto out_ping;
2097         if (ip_misc_proc_init())
2098                 goto out_misc;
2099 out:
2100         return rc;
2101 out_misc:
2102         ping_proc_exit();
2103 out_ping:
2104         udp4_proc_exit();
2105 out_udp:
2106         tcp4_proc_exit();
2107 out_tcp:
2108         raw_proc_exit();
2109 out_raw:
2110         rc = -ENOMEM;
2111         goto out;
2112 }
2113
2114 #else /* CONFIG_PROC_FS */
2115 static int __init ipv4_proc_init(void)
2116 {
2117         return 0;
2118 }
2119 #endif /* CONFIG_PROC_FS */