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