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
 601         /* Basic assumption: if someone sets sk->sk_err, he _must_
 602          * change state of the socket from TCP_SYN_*.
 603          * Connect() does not allow to get error notifications
 604          * without closing the socket.
 605          */
 606         while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 607                 release_sock(sk);
 608                 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
 609                 lock_sock(sk);
 610                 if (signal_pending(current) || !timeo)
 611                         break;
 612         }
 613         remove_wait_queue(sk_sleep(sk), &wait);
 614         sk->sk_write_pending -= writebias;
 615         return timeo;
 616 }
 617
 618 /*
 619  *      Connect to a remote host. There is regrettably still a little
 620  *      TCP 'magic' in here.
 621  */
 622 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 623                           int addr_len, int flags, int is_sendmsg)
 624 {
 625         struct sock *sk = sock->sk;
 626         int err;
 627         long timeo;
 628
 629         /*
 630          * uaddr can be NULL and addr_len can be 0 if:
 631          * sk is a TCP fastopen active socket and
 632          * TCP_FASTOPEN_CONNECT sockopt is set and
 633          * we already have a valid cookie for this socket.
 634          * In this case, user can call write() after connect().
 635          * write() will invoke tcp_sendmsg_fastopen() which calls
 636          * __inet_stream_connect().
 637          */
 638         if (uaddr) {
 639                 if (addr_len < sizeof(uaddr->sa_family))
 640                         return -EINVAL;
 641
 642                 if (uaddr->sa_family == AF_UNSPEC) {
 643                         sk->sk_disconnects++;
 644                         err = sk->sk_prot->disconnect(sk, flags);
 645                         sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 646                         goto out;
 647                 }
 648         }
 649
 650         switch (sock->state) {
 651         default:
 652                 err = -EINVAL;
 653                 goto out;
 654         case SS_CONNECTED:
 655                 err = -EISCONN;
 656                 goto out;
 657         case SS_CONNECTING:
 658                 if (inet_test_bit(DEFER_CONNECT, sk))
 659                         err = is_sendmsg ? -EINPROGRESS : -EISCONN;
 660                 else
 661                         err = -EALREADY;
 662                 /* Fall out of switch with err, set for this state */
 663                 break;
 664         case SS_UNCONNECTED:
 665                 err = -EISCONN;
 666                 if (sk->sk_state != TCP_CLOSE)
 667                         goto out;
 668
 669                 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
 670                         err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
 671                         if (err)
 672                                 goto out;
 673                 }
 674
 675                 err = sk->sk_prot->connect(sk, uaddr, addr_len);
 676                 if (err < 0)
 677                         goto out;
 678
 679                 sock->state = SS_CONNECTING;
 680
 681                 if (!err && inet_test_bit(DEFER_CONNECT, sk))
 682                         goto out;
 683
 684                 /* Just entered SS_CONNECTING state; the only
 685                  * difference is that return value in non-blocking
 686                  * case is EINPROGRESS, rather than EALREADY.
 687                  */
 688                 err = -EINPROGRESS;
 689                 break;
 690         }
 691
 692         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
 693
 694         if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 695                 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
 696                                 tcp_sk(sk)->fastopen_req &&
 697                                 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
 698                 int dis = sk->sk_disconnects;
 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                 if (dis != sk->sk_disconnects) {
 709                         err = -EPIPE;
 710                         goto out;
 711                 }
 712         }
 713
 714         /* Connection was closed by RST, timeout, ICMP error
 715          * or another process disconnected us.
 716          */
 717         if (sk->sk_state == TCP_CLOSE)
 718                 goto sock_error;
 719
 720         /* sk->sk_err may be not zero now, if RECVERR was ordered by user
 721          * and error was received after socket entered established state.
 722          * Hence, it is handled normally after connect() return successfully.
 723          */
 724
 725         sock->state = SS_CONNECTED;
 726         err = 0;
 727 out:
 728         return err;
 729
 730 sock_error:
 731         err = sock_error(sk) ? : -ECONNABORTED;
 732         sock->state = SS_UNCONNECTED;
 733         sk->sk_disconnects++;
 734         if (sk->sk_prot->disconnect(sk, flags))
 735                 sock->state = SS_DISCONNECTING;
 736         goto out;
 737 }
 738 EXPORT_SYMBOL(__inet_stream_connect);
 739
 740 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 741                         int addr_len, int flags)
 742 {
 743         int err;
 744
 745         lock_sock(sock->sk);
 746         err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
 747         release_sock(sock->sk);
 748         return err;
 749 }
 750 EXPORT_SYMBOL(inet_stream_connect);
 751
 752 void __inet_accept(struct socket *sock, struct socket *newsock, struct sock *newsk)
 753 {
 754         sock_rps_record_flow(newsk);
 755         WARN_ON(!((1 << newsk->sk_state) &
 756                   (TCPF_ESTABLISHED | TCPF_SYN_RECV |
 757                   TCPF_CLOSE_WAIT | TCPF_CLOSE)));
 758
 759         if (test_bit(SOCK_SUPPORT_ZC, &sock->flags))
 760                 set_bit(SOCK_SUPPORT_ZC, &newsock->flags);
 761         sock_graft(newsk, newsock);
 762
 763         newsock->state = SS_CONNECTED;
 764 }
 765
 766 /*
 767  *      Accept a pending connection. The TCP layer now gives BSD semantics.
 768  */
 769
 770 int inet_accept(struct socket *sock, struct socket *newsock, int flags,
 771                 bool kern)
 772 {
 773         struct sock *sk1 = sock->sk, *sk2;
 774         int err = -EINVAL;
 775
 776         /* IPV6_ADDRFORM can change sk->sk_prot under us. */
 777         sk2 = READ_ONCE(sk1->sk_prot)->accept(sk1, flags, &err, kern);
 778         if (!sk2)
 779                 return err;
 780
 781         lock_sock(sk2);
 782         __inet_accept(sock, newsock, sk2);
 783         release_sock(sk2);
 784         return 0;
 785 }
 786 EXPORT_SYMBOL(inet_accept);
 787
 788 /*
 789  *      This does both peername and sockname.
 790  */
 791 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
 792                  int peer)
 793 {
 794         struct sock *sk         = sock->sk;
 795         struct inet_sock *inet  = inet_sk(sk);
 796         DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
 797
 798         sin->sin_family = AF_INET;
 799         lock_sock(sk);
 800         if (peer) {
 801                 if (!inet->inet_dport ||
 802                     (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
 803                      peer == 1)) {
 804                         release_sock(sk);
 805                         return -ENOTCONN;
 806                 }
 807                 sin->sin_port = inet->inet_dport;
 808                 sin->sin_addr.s_addr = inet->inet_daddr;
 809                 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin,
 810                                        CGROUP_INET4_GETPEERNAME);
 811         } else {
 812                 __be32 addr = inet->inet_rcv_saddr;
 813                 if (!addr)
 814                         addr = inet->inet_saddr;
 815                 sin->sin_port = inet->inet_sport;
 816                 sin->sin_addr.s_addr = addr;
 817                 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin,
 818                                        CGROUP_INET4_GETSOCKNAME);
 819         }
 820         release_sock(sk);
 821         memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 822         return sizeof(*sin);
 823 }
 824 EXPORT_SYMBOL(inet_getname);
 825
 826 int inet_send_prepare(struct sock *sk)
 827 {
 828         sock_rps_record_flow(sk);
 829
 830         /* We may need to bind the socket. */
 831         if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind &&
 832             inet_autobind(sk))
 833                 return -EAGAIN;
 834
 835         return 0;
 836 }
 837 EXPORT_SYMBOL_GPL(inet_send_prepare);
 838
 839 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
 840 {
 841         struct sock *sk = sock->sk;
 842
 843         if (unlikely(inet_send_prepare(sk)))
 844                 return -EAGAIN;
 845
 846         return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
 847                                sk, msg, size);
 848 }
 849 EXPORT_SYMBOL(inet_sendmsg);
 850
 851 void inet_splice_eof(struct socket *sock)
 852 {
 853         const struct proto *prot;
 854         struct sock *sk = sock->sk;
 855
 856         if (unlikely(inet_send_prepare(sk)))
 857                 return;
 858
 859         /* IPV6_ADDRFORM can change sk->sk_prot under us. */
 860         prot = READ_ONCE(sk->sk_prot);
 861         if (prot->splice_eof)
 862                 prot->splice_eof(sock);
 863 }
 864 EXPORT_SYMBOL_GPL(inet_splice_eof);
 865
 866 INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
 867                                           size_t, int, int *));
 868 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
 869                  int flags)
 870 {
 871         struct sock *sk = sock->sk;
 872         int addr_len = 0;
 873         int err;
 874
 875         if (likely(!(flags & MSG_ERRQUEUE)))
 876                 sock_rps_record_flow(sk);
 877
 878         err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
 879                               sk, msg, size, flags, &addr_len);
 880         if (err >= 0)
 881                 msg->msg_namelen = addr_len;
 882         return err;
 883 }
 884 EXPORT_SYMBOL(inet_recvmsg);
 885
 886 int inet_shutdown(struct socket *sock, int how)
 887 {
 888         struct sock *sk = sock->sk;
 889         int err = 0;
 890
 891         /* This should really check to make sure
 892          * the socket is a TCP socket. (WHY AC...)
 893          */
 894         how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
 895                        1->2 bit 2 snds.
 896                        2->3 */
 897         if ((how & ~SHUTDOWN_MASK) || !how)     /* MAXINT->0 */
 898                 return -EINVAL;
 899
 900         lock_sock(sk);
 901         if (sock->state == SS_CONNECTING) {
 902                 if ((1 << sk->sk_state) &
 903                     (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
 904                         sock->state = SS_DISCONNECTING;
 905                 else
 906                         sock->state = SS_CONNECTED;
 907         }
 908
 909         switch (sk->sk_state) {
 910         case TCP_CLOSE:
 911                 err = -ENOTCONN;
 912                 /* Hack to wake up other listeners, who can poll for
 913                    EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
 914                 fallthrough;
 915         default:
 916                 WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | how);
 917                 if (sk->sk_prot->shutdown)
 918                         sk->sk_prot->shutdown(sk, how);
 919                 break;
 920
 921         /* Remaining two branches are temporary solution for missing
 922          * close() in multithreaded environment. It is _not_ a good idea,
 923          * but we have no choice until close() is repaired at VFS level.
 924          */
 925         case TCP_LISTEN:
 926                 if (!(how & RCV_SHUTDOWN))
 927                         break;
 928                 fallthrough;
 929         case TCP_SYN_SENT:
 930                 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
 931                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 932                 break;
 933         }
 934
 935         /* Wake up anyone sleeping in poll. */
 936         sk->sk_state_change(sk);
 937         release_sock(sk);
 938         return err;
 939 }
 940 EXPORT_SYMBOL(inet_shutdown);
 941
 942 /*
 943  *      ioctl() calls you can issue on an INET socket. Most of these are
 944  *      device configuration and stuff and very rarely used. Some ioctls
 945  *      pass on to the socket itself.
 946  *
 947  *      NOTE: I like the idea of a module for the config stuff. ie ifconfig
 948  *      loads the devconfigure module does its configuring and unloads it.
 949  *      There's a good 20K of config code hanging around the kernel.
 950  */
 951
 952 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 953 {
 954         struct sock *sk = sock->sk;
 955         int err = 0;
 956         struct net *net = sock_net(sk);
 957         void __user *p = (void __user *)arg;
 958         struct ifreq ifr;
 959         struct rtentry rt;
 960
 961         switch (cmd) {
 962         case SIOCADDRT:
 963         case SIOCDELRT:
 964                 if (copy_from_user(&rt, p, sizeof(struct rtentry)))
 965                         return -EFAULT;
 966                 err = ip_rt_ioctl(net, cmd, &rt);
 967                 break;
 968         case SIOCRTMSG:
 969                 err = -EINVAL;
 970                 break;
 971         case SIOCDARP:
 972         case SIOCGARP:
 973         case SIOCSARP:
 974                 err = arp_ioctl(net, cmd, (void __user *)arg);
 975                 break;
 976         case SIOCGIFADDR:
 977         case SIOCGIFBRDADDR:
 978         case SIOCGIFNETMASK:
 979         case SIOCGIFDSTADDR:
 980         case SIOCGIFPFLAGS:
 981                 if (get_user_ifreq(&ifr, NULL, p))
 982                         return -EFAULT;
 983                 err = devinet_ioctl(net, cmd, &ifr);
 984                 if (!err && put_user_ifreq(&ifr, p))
 985                         err = -EFAULT;
 986                 break;
 987
 988         case SIOCSIFADDR:
 989         case SIOCSIFBRDADDR:
 990         case SIOCSIFNETMASK:
 991         case SIOCSIFDSTADDR:
 992         case SIOCSIFPFLAGS:
 993         case SIOCSIFFLAGS:
 994                 if (get_user_ifreq(&ifr, NULL, p))
 995                         return -EFAULT;
 996                 err = devinet_ioctl(net, cmd, &ifr);
 997                 break;
 998         default:
 999                 if (sk->sk_prot->ioctl)
1000                         err = sk_ioctl(sk, cmd, (void __user *)arg);
1001                 else
1002                         err = -ENOIOCTLCMD;
1003                 break;
1004         }
1005         return err;
1006 }
1007 EXPORT_SYMBOL(inet_ioctl);
1008
1009 #ifdef CONFIG_COMPAT
1010 static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd,
1011                 struct compat_rtentry __user *ur)
1012 {
1013         compat_uptr_t rtdev;
1014         struct rtentry rt;
1015
1016         if (copy_from_user(&rt.rt_dst, &ur->rt_dst,
1017                         3 * sizeof(struct sockaddr)) ||
1018             get_user(rt.rt_flags, &ur->rt_flags) ||
1019             get_user(rt.rt_metric, &ur->rt_metric) ||
1020             get_user(rt.rt_mtu, &ur->rt_mtu) ||
1021             get_user(rt.rt_window, &ur->rt_window) ||
1022             get_user(rt.rt_irtt, &ur->rt_irtt) ||
1023             get_user(rtdev, &ur->rt_dev))
1024                 return -EFAULT;
1025
1026         rt.rt_dev = compat_ptr(rtdev);
1027         return ip_rt_ioctl(sock_net(sk), cmd, &rt);
1028 }
1029
1030 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1031 {
1032         void __user *argp = compat_ptr(arg);
1033         struct sock *sk = sock->sk;
1034
1035         switch (cmd) {
1036         case SIOCADDRT:
1037         case SIOCDELRT:
1038                 return inet_compat_routing_ioctl(sk, cmd, argp);
1039         default:
1040                 if (!sk->sk_prot->compat_ioctl)
1041                         return -ENOIOCTLCMD;
1042                 return sk->sk_prot->compat_ioctl(sk, cmd, arg);
1043         }
1044 }
1045 #endif /* CONFIG_COMPAT */
1046
1047 const struct proto_ops inet_stream_ops = {
1048         .family            = PF_INET,
1049         .owner             = THIS_MODULE,
1050         .release           = inet_release,
1051         .bind              = inet_bind,
1052         .connect           = inet_stream_connect,
1053         .socketpair        = sock_no_socketpair,
1054         .accept            = inet_accept,
1055         .getname           = inet_getname,
1056         .poll              = tcp_poll,
1057         .ioctl             = inet_ioctl,
1058         .gettstamp         = sock_gettstamp,
1059         .listen            = inet_listen,
1060         .shutdown          = inet_shutdown,
1061         .setsockopt        = sock_common_setsockopt,
1062         .getsockopt        = sock_common_getsockopt,
1063         .sendmsg           = inet_sendmsg,
1064         .recvmsg           = inet_recvmsg,
1065 #ifdef CONFIG_MMU
1066         .mmap              = tcp_mmap,
1067 #endif
1068         .splice_eof        = inet_splice_eof,
1069         .splice_read       = tcp_splice_read,
1070         .read_sock         = tcp_read_sock,
1071         .read_skb          = tcp_read_skb,
1072         .sendmsg_locked    = tcp_sendmsg_locked,
1073         .peek_len          = tcp_peek_len,
1074 #ifdef CONFIG_COMPAT
1075         .compat_ioctl      = inet_compat_ioctl,
1076 #endif
1077         .set_rcvlowat      = tcp_set_rcvlowat,
1078 };
1079 EXPORT_SYMBOL(inet_stream_ops);
1080
1081 const struct proto_ops inet_dgram_ops = {
1082         .family            = PF_INET,
1083         .owner             = THIS_MODULE,
1084         .release           = inet_release,
1085         .bind              = inet_bind,
1086         .connect           = inet_dgram_connect,
1087         .socketpair        = sock_no_socketpair,
1088         .accept            = sock_no_accept,
1089         .getname           = inet_getname,
1090         .poll              = udp_poll,
1091         .ioctl             = inet_ioctl,
1092         .gettstamp         = sock_gettstamp,
1093         .listen            = sock_no_listen,
1094         .shutdown          = inet_shutdown,
1095         .setsockopt        = sock_common_setsockopt,
1096         .getsockopt        = sock_common_getsockopt,
1097         .sendmsg           = inet_sendmsg,
1098         .read_skb          = udp_read_skb,
1099         .recvmsg           = inet_recvmsg,
1100         .mmap              = sock_no_mmap,
1101         .splice_eof        = inet_splice_eof,
1102         .set_peek_off      = sk_set_peek_off,
1103 #ifdef CONFIG_COMPAT
1104         .compat_ioctl      = inet_compat_ioctl,
1105 #endif
1106 };
1107 EXPORT_SYMBOL(inet_dgram_ops);
1108
1109 /*
1110  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1111  * udp_poll
1112  */
1113 static const struct proto_ops inet_sockraw_ops = {
1114         .family            = PF_INET,
1115         .owner             = THIS_MODULE,
1116         .release           = inet_release,
1117         .bind              = inet_bind,
1118         .connect           = inet_dgram_connect,
1119         .socketpair        = sock_no_socketpair,
1120         .accept            = sock_no_accept,
1121         .getname           = inet_getname,
1122         .poll              = datagram_poll,
1123         .ioctl             = inet_ioctl,
1124         .gettstamp         = sock_gettstamp,
1125         .listen            = sock_no_listen,
1126         .shutdown          = inet_shutdown,
1127         .setsockopt        = sock_common_setsockopt,
1128         .getsockopt        = sock_common_getsockopt,
1129         .sendmsg           = inet_sendmsg,
1130         .recvmsg           = inet_recvmsg,
1131         .mmap              = sock_no_mmap,
1132         .splice_eof        = inet_splice_eof,
1133 #ifdef CONFIG_COMPAT
1134         .compat_ioctl      = inet_compat_ioctl,
1135 #endif
1136 };
1137
1138 static const struct net_proto_family inet_family_ops = {
1139         .family = PF_INET,
1140         .create = inet_create,
1141         .owner  = THIS_MODULE,
1142 };
1143
1144 /* Upon startup we insert all the elements in inetsw_array[] into
1145  * the linked list inetsw.
1146  */
1147 static struct inet_protosw inetsw_array[] =
1148 {
1149         {
1150                 .type =       SOCK_STREAM,
1151                 .protocol =   IPPROTO_TCP,
1152                 .prot =       &tcp_prot,
1153                 .ops =        &inet_stream_ops,
1154                 .flags =      INET_PROTOSW_PERMANENT |
1155                               INET_PROTOSW_ICSK,
1156         },
1157
1158         {
1159                 .type =       SOCK_DGRAM,
1160                 .protocol =   IPPROTO_UDP,
1161                 .prot =       &udp_prot,
1162                 .ops =        &inet_dgram_ops,
1163                 .flags =      INET_PROTOSW_PERMANENT,
1164        },
1165
1166        {
1167                 .type =       SOCK_DGRAM,
1168                 .protocol =   IPPROTO_ICMP,
1169                 .prot =       &ping_prot,
1170                 .ops =        &inet_sockraw_ops,
1171                 .flags =      INET_PROTOSW_REUSE,
1172        },
1173
1174        {
1175                .type =       SOCK_RAW,
1176                .protocol =   IPPROTO_IP,        /* wild card */
1177                .prot =       &raw_prot,
1178                .ops =        &inet_sockraw_ops,
1179                .flags =      INET_PROTOSW_REUSE,
1180        }
1181 };
1182
1183 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1184
1185 void inet_register_protosw(struct inet_protosw *p)
1186 {
1187         struct list_head *lh;
1188         struct inet_protosw *answer;
1189         int protocol = p->protocol;
1190         struct list_head *last_perm;
1191
1192         spin_lock_bh(&inetsw_lock);
1193
1194         if (p->type >= SOCK_MAX)
1195                 goto out_illegal;
1196
1197         /* If we are trying to override a permanent protocol, bail. */
1198         last_perm = &inetsw[p->type];
1199         list_for_each(lh, &inetsw[p->type]) {
1200                 answer = list_entry(lh, struct inet_protosw, list);
1201                 /* Check only the non-wild match. */
1202                 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1203                         break;
1204                 if (protocol == answer->protocol)
1205                         goto out_permanent;
1206                 last_perm = lh;
1207         }
1208
1209         /* Add the new entry after the last permanent entry if any, so that
1210          * the new entry does not override a permanent entry when matched with
1211          * a wild-card protocol. But it is allowed to override any existing
1212          * non-permanent entry.  This means that when we remove this entry, the
1213          * system automatically returns to the old behavior.
1214          */
1215         list_add_rcu(&p->list, last_perm);
1216 out:
1217         spin_unlock_bh(&inetsw_lock);
1218
1219         return;
1220
1221 out_permanent:
1222         pr_err("Attempt to override permanent protocol %d\n", protocol);
1223         goto out;
1224
1225 out_illegal:
1226         pr_err("Ignoring attempt to register invalid socket type %d\n",
1227                p->type);
1228         goto out;
1229 }
1230 EXPORT_SYMBOL(inet_register_protosw);
1231
1232 void inet_unregister_protosw(struct inet_protosw *p)
1233 {
1234         if (INET_PROTOSW_PERMANENT & p->flags) {
1235                 pr_err("Attempt to unregister permanent protocol %d\n",
1236                        p->protocol);
1237         } else {
1238                 spin_lock_bh(&inetsw_lock);
1239                 list_del_rcu(&p->list);
1240                 spin_unlock_bh(&inetsw_lock);
1241
1242                 synchronize_net();
1243         }
1244 }
1245 EXPORT_SYMBOL(inet_unregister_protosw);
1246
1247 static int inet_sk_reselect_saddr(struct sock *sk)
1248 {
1249         struct inet_sock *inet = inet_sk(sk);
1250         __be32 old_saddr = inet->inet_saddr;
1251         __be32 daddr = inet->inet_daddr;
1252         struct flowi4 *fl4;
1253         struct rtable *rt;
1254         __be32 new_saddr;
1255         struct ip_options_rcu *inet_opt;
1256         int err;
1257
1258         inet_opt = rcu_dereference_protected(inet->inet_opt,
1259                                              lockdep_sock_is_held(sk));
1260         if (inet_opt && inet_opt->opt.srr)
1261                 daddr = inet_opt->opt.faddr;
1262
1263         /* Query new route. */
1264         fl4 = &inet->cork.fl.u.ip4;
1265         rt = ip_route_connect(fl4, daddr, 0, sk->sk_bound_dev_if,
1266                               sk->sk_protocol, inet->inet_sport,
1267                               inet->inet_dport, sk);
1268         if (IS_ERR(rt))
1269                 return PTR_ERR(rt);
1270
1271         new_saddr = fl4->saddr;
1272
1273         if (new_saddr == old_saddr) {
1274                 sk_setup_caps(sk, &rt->dst);
1275                 return 0;
1276         }
1277
1278         err = inet_bhash2_update_saddr(sk, &new_saddr, AF_INET);
1279         if (err) {
1280                 ip_rt_put(rt);
1281                 return err;
1282         }
1283
1284         sk_setup_caps(sk, &rt->dst);
1285
1286         if (READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) > 1) {
1287                 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1288                         __func__, &old_saddr, &new_saddr);
1289         }
1290
1291         /*
1292          * XXX The only one ugly spot where we need to
1293          * XXX really change the sockets identity after
1294          * XXX it has entered the hashes. -DaveM
1295          *
1296          * Besides that, it does not check for connection
1297          * uniqueness. Wait for troubles.
1298          */
1299         return __sk_prot_rehash(sk);
1300 }
1301
1302 int inet_sk_rebuild_header(struct sock *sk)
1303 {
1304         struct inet_sock *inet = inet_sk(sk);
1305         struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1306         __be32 daddr;
1307         struct ip_options_rcu *inet_opt;
1308         struct flowi4 *fl4;
1309         int err;
1310
1311         /* Route is OK, nothing to do. */
1312         if (rt)
1313                 return 0;
1314
1315         /* Reroute. */
1316         rcu_read_lock();
1317         inet_opt = rcu_dereference(inet->inet_opt);
1318         daddr = inet->inet_daddr;
1319         if (inet_opt && inet_opt->opt.srr)
1320                 daddr = inet_opt->opt.faddr;
1321         rcu_read_unlock();
1322         fl4 = &inet->cork.fl.u.ip4;
1323         rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1324                                    inet->inet_dport, inet->inet_sport,
1325                                    sk->sk_protocol, RT_CONN_FLAGS(sk),
1326                                    sk->sk_bound_dev_if);
1327         if (!IS_ERR(rt)) {
1328                 err = 0;
1329                 sk_setup_caps(sk, &rt->dst);
1330         } else {
1331                 err = PTR_ERR(rt);
1332
1333                 /* Routing failed... */
1334                 sk->sk_route_caps = 0;
1335                 /*
1336                  * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1337                  * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1338                  */
1339                 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) ||
1340                     sk->sk_state != TCP_SYN_SENT ||
1341                     (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1342                     (err = inet_sk_reselect_saddr(sk)) != 0)
1343                         WRITE_ONCE(sk->sk_err_soft, -err);
1344         }
1345
1346         return err;
1347 }
1348 EXPORT_SYMBOL(inet_sk_rebuild_header);
1349
1350 void inet_sk_set_state(struct sock *sk, int state)
1351 {
1352         trace_inet_sock_set_state(sk, sk->sk_state, state);
1353         sk->sk_state = state;
1354 }
1355 EXPORT_SYMBOL(inet_sk_set_state);
1356
1357 void inet_sk_state_store(struct sock *sk, int newstate)
1358 {
1359         trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1360         smp_store_release(&sk->sk_state, newstate);
1361 }
1362
1363 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1364                                  netdev_features_t features)
1365 {
1366         bool udpfrag = false, fixedid = false, gso_partial, encap;
1367         struct sk_buff *segs = ERR_PTR(-EINVAL);
1368         const struct net_offload *ops;
1369         unsigned int offset = 0;
1370         struct iphdr *iph;
1371         int proto, tot_len;
1372         int nhoff;
1373         int ihl;
1374         int id;
1375
1376         skb_reset_network_header(skb);
1377         nhoff = skb_network_header(skb) - skb_mac_header(skb);
1378         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1379                 goto out;
1380
1381         iph = ip_hdr(skb);
1382         ihl = iph->ihl * 4;
1383         if (ihl < sizeof(*iph))
1384                 goto out;
1385
1386         id = ntohs(iph->id);
1387         proto = iph->protocol;
1388
1389         /* Warning: after this point, iph might be no longer valid */
1390         if (unlikely(!pskb_may_pull(skb, ihl)))
1391                 goto out;
1392         __skb_pull(skb, ihl);
1393
1394         encap = SKB_GSO_CB(skb)->encap_level > 0;
1395         if (encap)
1396                 features &= skb->dev->hw_enc_features;
1397         SKB_GSO_CB(skb)->encap_level += ihl;
1398
1399         skb_reset_transport_header(skb);
1400
1401         segs = ERR_PTR(-EPROTONOSUPPORT);
1402
1403         if (!skb->encapsulation || encap) {
1404                 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1405                 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1406
1407                 /* fixed ID is invalid if DF bit is not set */
1408                 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1409                         goto out;
1410         }
1411
1412         ops = rcu_dereference(inet_offloads[proto]);
1413         if (likely(ops && ops->callbacks.gso_segment)) {
1414                 segs = ops->callbacks.gso_segment(skb, features);
1415                 if (!segs)
1416                         skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
1417         }
1418
1419         if (IS_ERR_OR_NULL(segs))
1420                 goto out;
1421
1422         gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1423
1424         skb = segs;
1425         do {
1426                 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1427                 if (udpfrag) {
1428                         iph->frag_off = htons(offset >> 3);
1429                         if (skb->next)
1430                                 iph->frag_off |= htons(IP_MF);
1431                         offset += skb->len - nhoff - ihl;
1432                         tot_len = skb->len - nhoff;
1433                 } else if (skb_is_gso(skb)) {
1434                         if (!fixedid) {
1435                                 iph->id = htons(id);
1436                                 id += skb_shinfo(skb)->gso_segs;
1437                         }
1438
1439                         if (gso_partial)
1440                                 tot_len = skb_shinfo(skb)->gso_size +
1441                                           SKB_GSO_CB(skb)->data_offset +
1442                                           skb->head - (unsigned char *)iph;
1443                         else
1444                                 tot_len = skb->len - nhoff;
1445                 } else {
1446                         if (!fixedid)
1447                                 iph->id = htons(id++);
1448                         tot_len = skb->len - nhoff;
1449                 }
1450                 iph->tot_len = htons(tot_len);
1451                 ip_send_check(iph);
1452                 if (encap)
1453                         skb_reset_inner_headers(skb);
1454                 skb->network_header = (u8 *)iph - skb->head;
1455                 skb_reset_mac_len(skb);
1456         } while ((skb = skb->next));
1457
1458 out:
1459         return segs;
1460 }
1461
1462 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1463                                         netdev_features_t features)
1464 {
1465         if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1466                 return ERR_PTR(-EINVAL);
1467
1468         return inet_gso_segment(skb, features);
1469 }
1470
1471 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1472 {
1473         const struct net_offload *ops;
1474         struct sk_buff *pp = NULL;
1475         const struct iphdr *iph;
1476         struct sk_buff *p;
1477         unsigned int hlen;
1478         unsigned int off;
1479         unsigned int id;
1480         int flush = 1;
1481         int proto;
1482
1483         off = skb_gro_offset(skb);
1484         hlen = off + sizeof(*iph);
1485         iph = skb_gro_header(skb, hlen, off);
1486         if (unlikely(!iph))
1487                 goto out;
1488
1489         proto = iph->protocol;
1490
1491         ops = rcu_dereference(inet_offloads[proto]);
1492         if (!ops || !ops->callbacks.gro_receive)
1493                 goto out;
1494
1495         if (*(u8 *)iph != 0x45)
1496                 goto out;
1497
1498         if (ip_is_fragment(iph))
1499                 goto out;
1500
1501         if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1502                 goto out;
1503
1504         NAPI_GRO_CB(skb)->proto = proto;
1505         id = ntohl(*(__be32 *)&iph->id);
1506         flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1507         id >>= 16;
1508
1509         list_for_each_entry(p, head, list) {
1510                 struct iphdr *iph2;
1511                 u16 flush_id;
1512
1513                 if (!NAPI_GRO_CB(p)->same_flow)
1514                         continue;
1515
1516                 iph2 = (struct iphdr *)(p->data + off);
1517                 /* The above works because, with the exception of the top
1518                  * (inner most) layer, we only aggregate pkts with the same
1519                  * hdr length so all the hdrs we'll need to verify will start
1520                  * at the same offset.
1521                  */
1522                 if ((iph->protocol ^ iph2->protocol) |
1523                     ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1524                     ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1525                         NAPI_GRO_CB(p)->same_flow = 0;
1526                         continue;
1527                 }
1528
1529                 /* All fields must match except length and checksum. */
1530                 NAPI_GRO_CB(p)->flush |=
1531                         (iph->ttl ^ iph2->ttl) |
1532                         (iph->tos ^ iph2->tos) |
1533                         ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1534
1535                 NAPI_GRO_CB(p)->flush |= flush;
1536
1537                 /* We need to store of the IP ID check to be included later
1538                  * when we can verify that this packet does in fact belong
1539                  * to a given flow.
1540                  */
1541                 flush_id = (u16)(id - ntohs(iph2->id));
1542
1543                 /* This bit of code makes it much easier for us to identify
1544                  * the cases where we are doing atomic vs non-atomic IP ID
1545                  * checks.  Specifically an atomic check can return IP ID
1546                  * values 0 - 0xFFFF, while a non-atomic check can only
1547                  * return 0 or 0xFFFF.
1548                  */
1549                 if (!NAPI_GRO_CB(p)->is_atomic ||
1550                     !(iph->frag_off & htons(IP_DF))) {
1551                         flush_id ^= NAPI_GRO_CB(p)->count;
1552                         flush_id = flush_id ? 0xFFFF : 0;
1553                 }
1554
1555                 /* If the previous IP ID value was based on an atomic
1556                  * datagram we can overwrite the value and ignore it.
1557                  */
1558                 if (NAPI_GRO_CB(skb)->is_atomic)
1559                         NAPI_GRO_CB(p)->flush_id = flush_id;
1560                 else
1561                         NAPI_GRO_CB(p)->flush_id |= flush_id;
1562         }
1563
1564         NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1565         NAPI_GRO_CB(skb)->flush |= flush;
1566         skb_set_network_header(skb, off);
1567         /* The above will be needed by the transport layer if there is one
1568          * immediately following this IP hdr.
1569          */
1570
1571         /* Note : No need to call skb_gro_postpull_rcsum() here,
1572          * as we already checked checksum over ipv4 header was 0
1573          */
1574         skb_gro_pull(skb, sizeof(*iph));
1575         skb_set_transport_header(skb, skb_gro_offset(skb));
1576
1577         pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1578                                        ops->callbacks.gro_receive, head, skb);
1579
1580 out:
1581         skb_gro_flush_final(skb, pp, flush);
1582
1583         return pp;
1584 }
1585
1586 static struct sk_buff *ipip_gro_receive(struct list_head *head,
1587                                         struct sk_buff *skb)
1588 {
1589         if (NAPI_GRO_CB(skb)->encap_mark) {
1590                 NAPI_GRO_CB(skb)->flush = 1;
1591                 return NULL;
1592         }
1593
1594         NAPI_GRO_CB(skb)->encap_mark = 1;
1595
1596         return inet_gro_receive(head, skb);
1597 }
1598
1599 #define SECONDS_PER_DAY 86400
1600
1601 /* inet_current_timestamp - Return IP network timestamp
1602  *
1603  * Return milliseconds since midnight in network byte order.
1604  */
1605 __be32 inet_current_timestamp(void)
1606 {
1607         u32 secs;
1608         u32 msecs;
1609         struct timespec64 ts;
1610
1611         ktime_get_real_ts64(&ts);
1612
1613         /* Get secs since midnight. */
1614         (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1615         /* Convert to msecs. */
1616         msecs = secs * MSEC_PER_SEC;
1617         /* Convert nsec to msec. */
1618         msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1619
1620         /* Convert to network byte order. */
1621         return htonl(msecs);
1622 }
1623 EXPORT_SYMBOL(inet_current_timestamp);
1624
1625 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1626 {
1627         if (sk->sk_family == AF_INET)
1628                 return ip_recv_error(sk, msg, len, addr_len);
1629 #if IS_ENABLED(CONFIG_IPV6)
1630         if (sk->sk_family == AF_INET6)
1631                 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1632 #endif
1633         return -EINVAL;
1634 }
1635
1636 int inet_gro_complete(struct sk_buff *skb, int nhoff)
1637 {
1638         struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1639         const struct net_offload *ops;
1640         __be16 totlen = iph->tot_len;
1641         int proto = iph->protocol;
1642         int err = -ENOSYS;
1643
1644         if (skb->encapsulation) {
1645                 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1646                 skb_set_inner_network_header(skb, nhoff);
1647         }
1648
1649         iph_set_totlen(iph, skb->len - nhoff);
1650         csum_replace2(&iph->check, totlen, iph->tot_len);
1651
1652         ops = rcu_dereference(inet_offloads[proto]);
1653         if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1654                 goto out;
1655
1656         /* Only need to add sizeof(*iph) to get to the next hdr below
1657          * because any hdr with option will have been flushed in
1658          * inet_gro_receive().
1659          */
1660         err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1661                               tcp4_gro_complete, udp4_gro_complete,
1662                               skb, nhoff + sizeof(*iph));
1663
1664 out:
1665         return err;
1666 }
1667
1668 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1669 {
1670         skb->encapsulation = 1;
1671         skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1672         return inet_gro_complete(skb, nhoff);
1673 }
1674
1675 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1676                          unsigned short type, unsigned char protocol,
1677                          struct net *net)
1678 {
1679         struct socket *sock;
1680         int rc = sock_create_kern(net, family, type, protocol, &sock);
1681
1682         if (rc == 0) {
1683                 *sk = sock->sk;
1684                 (*sk)->sk_allocation = GFP_ATOMIC;
1685                 (*sk)->sk_use_task_frag = false;
1686                 /*
1687                  * Unhash it so that IP input processing does not even see it,
1688                  * we do not wish this socket to see incoming packets.
1689                  */
1690                 (*sk)->sk_prot->unhash(*sk);
1691         }
1692         return rc;
1693 }
1694 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1695
1696 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1697 {
1698         unsigned long res = 0;
1699         int i;
1700
1701         for_each_possible_cpu(i)
1702                 res += snmp_get_cpu_field(mib, i, offt);
1703         return res;
1704 }
1705 EXPORT_SYMBOL_GPL(snmp_fold_field);
1706
1707 #if BITS_PER_LONG==32
1708
1709 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1710                          size_t syncp_offset)
1711 {
1712         void *bhptr;
1713         struct u64_stats_sync *syncp;
1714         u64 v;
1715         unsigned int start;
1716
1717         bhptr = per_cpu_ptr(mib, cpu);
1718         syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1719         do {
1720                 start = u64_stats_fetch_begin(syncp);
1721                 v = *(((u64 *)bhptr) + offt);
1722         } while (u64_stats_fetch_retry(syncp, start));
1723
1724         return v;
1725 }
1726 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1727
1728 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1729 {
1730         u64 res = 0;
1731         int cpu;
1732
1733         for_each_possible_cpu(cpu) {
1734                 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1735         }
1736         return res;
1737 }
1738 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1739 #endif
1740
1741 #ifdef CONFIG_IP_MULTICAST
1742 static const struct net_protocol igmp_protocol = {
1743         .handler =      igmp_rcv,
1744 };
1745 #endif
1746
1747 static const struct net_protocol tcp_protocol = {
1748         .handler        =       tcp_v4_rcv,
1749         .err_handler    =       tcp_v4_err,
1750         .no_policy      =       1,
1751         .icmp_strict_tag_validation = 1,
1752 };
1753
1754 static const struct net_protocol udp_protocol = {
1755         .handler =      udp_rcv,
1756         .err_handler =  udp_err,
1757         .no_policy =    1,
1758 };
1759
1760 static const struct net_protocol icmp_protocol = {
1761         .handler =      icmp_rcv,
1762         .err_handler =  icmp_err,
1763         .no_policy =    1,
1764 };
1765
1766 static __net_init int ipv4_mib_init_net(struct net *net)
1767 {
1768         int i;
1769
1770         net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1771         if (!net->mib.tcp_statistics)
1772                 goto err_tcp_mib;
1773         net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1774         if (!net->mib.ip_statistics)
1775                 goto err_ip_mib;
1776
1777         for_each_possible_cpu(i) {
1778                 struct ipstats_mib *af_inet_stats;
1779                 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1780                 u64_stats_init(&af_inet_stats->syncp);
1781         }
1782
1783         net->mib.net_statistics = alloc_percpu(struct linux_mib);
1784         if (!net->mib.net_statistics)
1785                 goto err_net_mib;
1786         net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1787         if (!net->mib.udp_statistics)
1788                 goto err_udp_mib;
1789         net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1790         if (!net->mib.udplite_statistics)
1791                 goto err_udplite_mib;
1792         net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1793         if (!net->mib.icmp_statistics)
1794                 goto err_icmp_mib;
1795         net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1796                                               GFP_KERNEL);
1797         if (!net->mib.icmpmsg_statistics)
1798                 goto err_icmpmsg_mib;
1799
1800         tcp_mib_init(net);
1801         return 0;
1802
1803 err_icmpmsg_mib:
1804         free_percpu(net->mib.icmp_statistics);
1805 err_icmp_mib:
1806         free_percpu(net->mib.udplite_statistics);
1807 err_udplite_mib:
1808         free_percpu(net->mib.udp_statistics);
1809 err_udp_mib:
1810         free_percpu(net->mib.net_statistics);
1811 err_net_mib:
1812         free_percpu(net->mib.ip_statistics);
1813 err_ip_mib:
1814         free_percpu(net->mib.tcp_statistics);
1815 err_tcp_mib:
1816         return -ENOMEM;
1817 }
1818
1819 static __net_exit void ipv4_mib_exit_net(struct net *net)
1820 {
1821         kfree(net->mib.icmpmsg_statistics);
1822         free_percpu(net->mib.icmp_statistics);
1823         free_percpu(net->mib.udplite_statistics);
1824         free_percpu(net->mib.udp_statistics);
1825         free_percpu(net->mib.net_statistics);
1826         free_percpu(net->mib.ip_statistics);
1827         free_percpu(net->mib.tcp_statistics);
1828 #ifdef CONFIG_MPTCP
1829         /* allocated on demand, see mptcp_init_sock() */
1830         free_percpu(net->mib.mptcp_statistics);
1831 #endif
1832 }
1833
1834 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1835         .init = ipv4_mib_init_net,
1836         .exit = ipv4_mib_exit_net,
1837 };
1838
1839 static int __init init_ipv4_mibs(void)
1840 {
1841         return register_pernet_subsys(&ipv4_mib_ops);
1842 }
1843
1844 static __net_init int inet_init_net(struct net *net)
1845 {
1846         /*
1847          * Set defaults for local port range
1848          */
1849         seqlock_init(&net->ipv4.ip_local_ports.lock);
1850         net->ipv4.ip_local_ports.range[0] =  32768;
1851         net->ipv4.ip_local_ports.range[1] =  60999;
1852
1853         seqlock_init(&net->ipv4.ping_group_range.lock);
1854         /*
1855          * Sane defaults - nobody may create ping sockets.
1856          * Boot scripts should set this to distro-specific group.
1857          */
1858         net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1859         net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1860
1861         /* Default values for sysctl-controlled parameters.
1862          * We set them here, in case sysctl is not compiled.
1863          */
1864         net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1865         net->ipv4.sysctl_ip_fwd_update_priority = 1;
1866         net->ipv4.sysctl_ip_dynaddr = 0;
1867         net->ipv4.sysctl_ip_early_demux = 1;
1868         net->ipv4.sysctl_udp_early_demux = 1;
1869         net->ipv4.sysctl_tcp_early_demux = 1;
1870         net->ipv4.sysctl_nexthop_compat_mode = 1;
1871 #ifdef CONFIG_SYSCTL
1872         net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1873 #endif
1874
1875         /* Some igmp sysctl, whose values are always used */
1876         net->ipv4.sysctl_igmp_max_memberships = 20;
1877         net->ipv4.sysctl_igmp_max_msf = 10;
1878         /* IGMP reports for link-local multicast groups are enabled by default */
1879         net->ipv4.sysctl_igmp_llm_reports = 1;
1880         net->ipv4.sysctl_igmp_qrv = 2;
1881
1882         net->ipv4.sysctl_fib_notify_on_flag_change = 0;
1883
1884         return 0;
1885 }
1886
1887 static __net_initdata struct pernet_operations af_inet_ops = {
1888         .init = inet_init_net,
1889 };
1890
1891 static int __init init_inet_pernet_ops(void)
1892 {
1893         return register_pernet_subsys(&af_inet_ops);
1894 }
1895
1896 static int ipv4_proc_init(void);
1897
1898 /*
1899  *      IP protocol layer initialiser
1900  */
1901
1902 static struct packet_offload ip_packet_offload __read_mostly = {
1903         .type = cpu_to_be16(ETH_P_IP),
1904         .callbacks = {
1905                 .gso_segment = inet_gso_segment,
1906                 .gro_receive = inet_gro_receive,
1907                 .gro_complete = inet_gro_complete,
1908         },
1909 };
1910
1911 static const struct net_offload ipip_offload = {
1912         .callbacks = {
1913                 .gso_segment    = ipip_gso_segment,
1914                 .gro_receive    = ipip_gro_receive,
1915                 .gro_complete   = ipip_gro_complete,
1916         },
1917 };
1918
1919 static int __init ipip_offload_init(void)
1920 {
1921         return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1922 }
1923
1924 static int __init ipv4_offload_init(void)
1925 {
1926         /*
1927          * Add offloads
1928          */
1929         if (udpv4_offload_init() < 0)
1930                 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1931         if (tcpv4_offload_init() < 0)
1932                 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1933         if (ipip_offload_init() < 0)
1934                 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1935
1936         dev_add_offload(&ip_packet_offload);
1937         return 0;
1938 }
1939
1940 fs_initcall(ipv4_offload_init);
1941
1942 static struct packet_type ip_packet_type __read_mostly = {
1943         .type = cpu_to_be16(ETH_P_IP),
1944         .func = ip_rcv,
1945         .list_func = ip_list_rcv,
1946 };
1947
1948 static int __init inet_init(void)
1949 {
1950         struct inet_protosw *q;
1951         struct list_head *r;
1952         int rc;
1953
1954         sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1955
1956         raw_hashinfo_init(&raw_v4_hashinfo);
1957
1958         rc = proto_register(&tcp_prot, 1);
1959         if (rc)
1960                 goto out;
1961
1962         rc = proto_register(&udp_prot, 1);
1963         if (rc)
1964                 goto out_unregister_tcp_proto;
1965
1966         rc = proto_register(&raw_prot, 1);
1967         if (rc)
1968                 goto out_unregister_udp_proto;
1969
1970         rc = proto_register(&ping_prot, 1);
1971         if (rc)
1972                 goto out_unregister_raw_proto;
1973
1974         /*
1975          *      Tell SOCKET that we are alive...
1976          */
1977
1978         (void)sock_register(&inet_family_ops);
1979
1980 #ifdef CONFIG_SYSCTL
1981         ip_static_sysctl_init();
1982 #endif
1983
1984         /*
1985          *      Add all the base protocols.
1986          */
1987
1988         if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1989                 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1990         if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1991                 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1992         if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1993                 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1994 #ifdef CONFIG_IP_MULTICAST
1995         if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1996                 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1997 #endif
1998
1999         /* Register the socket-side information for inet_create. */
2000         for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
2001                 INIT_LIST_HEAD(r);
2002
2003         for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
2004                 inet_register_protosw(q);
2005
2006         /*
2007          *      Set the ARP module up
2008          */
2009
2010         arp_init();
2011
2012         /*
2013          *      Set the IP module up
2014          */
2015
2016         ip_init();
2017
2018         /* Initialise per-cpu ipv4 mibs */
2019         if (init_ipv4_mibs())
2020                 panic("%s: Cannot init ipv4 mibs\n", __func__);
2021
2022         /* Setup TCP slab cache for open requests. */
2023         tcp_init();
2024
2025         /* Setup UDP memory threshold */
2026         udp_init();
2027
2028         /* Add UDP-Lite (RFC 3828) */
2029         udplite4_register();
2030
2031         raw_init();
2032
2033         ping_init();
2034
2035         /*
2036          *      Set the ICMP layer up
2037          */
2038
2039         if (icmp_init() < 0)
2040                 panic("Failed to create the ICMP control socket.\n");
2041
2042         /*
2043          *      Initialise the multicast router
2044          */
2045 #if defined(CONFIG_IP_MROUTE)
2046         if (ip_mr_init())
2047                 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2048 #endif
2049
2050         if (init_inet_pernet_ops())
2051                 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2052
2053         ipv4_proc_init();
2054
2055         ipfrag_init();
2056
2057         dev_add_pack(&ip_packet_type);
2058
2059         ip_tunnel_core_init();
2060
2061         rc = 0;
2062 out:
2063         return rc;
2064 out_unregister_raw_proto:
2065         proto_unregister(&raw_prot);
2066 out_unregister_udp_proto:
2067         proto_unregister(&udp_prot);
2068 out_unregister_tcp_proto:
2069         proto_unregister(&tcp_prot);
2070         goto out;
2071 }
2072
2073 fs_initcall(inet_init);
2074
2075 /* ------------------------------------------------------------------------ */
2076
2077 #ifdef CONFIG_PROC_FS
2078 static int __init ipv4_proc_init(void)
2079 {
2080         int rc = 0;
2081
2082         if (raw_proc_init())
2083                 goto out_raw;
2084         if (tcp4_proc_init())
2085                 goto out_tcp;
2086         if (udp4_proc_init())
2087                 goto out_udp;
2088         if (ping_proc_init())
2089                 goto out_ping;
2090         if (ip_misc_proc_init())
2091                 goto out_misc;
2092 out:
2093         return rc;
2094 out_misc:
2095         ping_proc_exit();
2096 out_ping:
2097         udp4_proc_exit();
2098 out_udp:
2099         tcp4_proc_exit();
2100 out_tcp:
2101         raw_proc_exit();
2102 out_raw:
2103         rc = -ENOMEM;
2104         goto out;
2105 }
2106
2107 #else /* CONFIG_PROC_FS */
2108 static int __init ipv4_proc_init(void)
2109 {
2110         return 0;
2111 }
2112 #endif /* CONFIG_PROC_FS */