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/tcp.h>
 103 #include <net/udp.h>
 104 #include <net/udplite.h>
 105 #include <net/ping.h>
 106 #include <linux/skbuff.h>
 107 #include <net/sock.h>
 108 #include <net/raw.h>
 109 #include <net/icmp.h>
 110 #include <net/inet_common.h>
 111 #include <net/ip_tunnels.h>
 112 #include <net/xfrm.h>
 113 #include <net/net_namespace.h>
 114 #include <net/secure_seq.h>
 115 #ifdef CONFIG_IP_MROUTE
 116 #include <linux/mroute.h>
 117 #endif
 118 #include <net/l3mdev.h>
 119 #include <net/compat.h>
 120
 121 #include <trace/events/sock.h>
 122
 123 /* The inetsw table contains everything that inet_create needs to
 124  * build a new socket.
 125  */
 126 static struct list_head inetsw[SOCK_MAX];
 127 static DEFINE_SPINLOCK(inetsw_lock);
 128
 129 /* New destruction routine */
 130
 131 void inet_sock_destruct(struct sock *sk)
 132 {
 133         struct inet_sock *inet = inet_sk(sk);
 134
 135         __skb_queue_purge(&sk->sk_receive_queue);
 136         if (sk->sk_rx_skb_cache) {
 137                 __kfree_skb(sk->sk_rx_skb_cache);
 138                 sk->sk_rx_skb_cache = NULL;
 139         }
 140         __skb_queue_purge(&sk->sk_error_queue);
 141
 142         sk_mem_reclaim(sk);
 143
 144         if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
 145                 pr_err("Attempt to release TCP socket in state %d %p\n",
 146                        sk->sk_state, sk);
 147                 return;
 148         }
 149         if (!sock_flag(sk, SOCK_DEAD)) {
 150                 pr_err("Attempt to release alive inet socket %p\n", sk);
 151                 return;
 152         }
 153
 154         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
 155         WARN_ON(refcount_read(&sk->sk_wmem_alloc));
 156         WARN_ON(sk->sk_wmem_queued);
 157         WARN_ON(sk->sk_forward_alloc);
 158
 159         kfree(rcu_dereference_protected(inet->inet_opt, 1));
 160         dst_release(rcu_dereference_protected(sk->sk_dst_cache, 1));
 161         dst_release(sk->sk_rx_dst);
 162         sk_refcnt_debug_dec(sk);
 163 }
 164 EXPORT_SYMBOL(inet_sock_destruct);
 165
 166 /*
 167  *      The routines beyond this point handle the behaviour of an AF_INET
 168  *      socket object. Mostly it punts to the subprotocols of IP to do
 169  *      the work.
 170  */
 171
 172 /*
 173  *      Automatically bind an unbound socket.
 174  */
 175
 176 static int inet_autobind(struct sock *sk)
 177 {
 178         struct inet_sock *inet;
 179         /* We may need to bind the socket. */
 180         lock_sock(sk);
 181         inet = inet_sk(sk);
 182         if (!inet->inet_num) {
 183                 if (sk->sk_prot->get_port(sk, 0)) {
 184                         release_sock(sk);
 185                         return -EAGAIN;
 186                 }
 187                 inet->inet_sport = htons(inet->inet_num);
 188         }
 189         release_sock(sk);
 190         return 0;
 191 }
 192
 193 /*
 194  *      Move a socket into listening state.
 195  */
 196 int inet_listen(struct socket *sock, int backlog)
 197 {
 198         struct sock *sk = sock->sk;
 199         unsigned char old_state;
 200         int err, tcp_fastopen;
 201
 202         lock_sock(sk);
 203
 204         err = -EINVAL;
 205         if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
 206                 goto out;
 207
 208         old_state = sk->sk_state;
 209         if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
 210                 goto out;
 211
 212         WRITE_ONCE(sk->sk_max_ack_backlog, backlog);
 213         /* Really, if the socket is already in listen state
 214          * we can only allow the backlog to be adjusted.
 215          */
 216         if (old_state != TCP_LISTEN) {
 217                 /* Enable TFO w/o requiring TCP_FASTOPEN socket option.
 218                  * Note that only TCP sockets (SOCK_STREAM) will reach here.
 219                  * Also fastopen backlog may already been set via the option
 220                  * because the socket was in TCP_LISTEN state previously but
 221                  * was shutdown() rather than close().
 222                  */
 223                 tcp_fastopen = sock_net(sk)->ipv4.sysctl_tcp_fastopen;
 224                 if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
 225                     (tcp_fastopen & TFO_SERVER_ENABLE) &&
 226                     !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
 227                         fastopen_queue_tune(sk, backlog);
 228                         tcp_fastopen_init_key_once(sock_net(sk));
 229                 }
 230
 231                 err = inet_csk_listen_start(sk, backlog);
 232                 if (err)
 233                         goto out;
 234                 tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL);
 235         }
 236         err = 0;
 237
 238 out:
 239         release_sock(sk);
 240         return err;
 241 }
 242 EXPORT_SYMBOL(inet_listen);
 243
 244 /*
 245  *      Create an inet socket.
 246  */
 247
 248 static int inet_create(struct net *net, struct socket *sock, int protocol,
 249                        int kern)
 250 {
 251         struct sock *sk;
 252         struct inet_protosw *answer;
 253         struct inet_sock *inet;
 254         struct proto *answer_prot;
 255         unsigned char answer_flags;
 256         int try_loading_module = 0;
 257         int err;
 258
 259         if (protocol < 0 || protocol >= IPPROTO_MAX)
 260                 return -EINVAL;
 261
 262         sock->state = SS_UNCONNECTED;
 263
 264         /* Look for the requested type/protocol pair. */
 265 lookup_protocol:
 266         err = -ESOCKTNOSUPPORT;
 267         rcu_read_lock();
 268         list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
 269
 270                 err = 0;
 271                 /* Check the non-wild match. */
 272                 if (protocol == answer->protocol) {
 273                         if (protocol != IPPROTO_IP)
 274                                 break;
 275                 } else {
 276                         /* Check for the two wild cases. */
 277                         if (IPPROTO_IP == protocol) {
 278                                 protocol = answer->protocol;
 279                                 break;
 280                         }
 281                         if (IPPROTO_IP == answer->protocol)
 282                                 break;
 283                 }
 284                 err = -EPROTONOSUPPORT;
 285         }
 286
 287         if (unlikely(err)) {
 288                 if (try_loading_module < 2) {
 289                         rcu_read_unlock();
 290                         /*
 291                          * Be more specific, e.g. net-pf-2-proto-132-type-1
 292                          * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
 293                          */
 294                         if (++try_loading_module == 1)
 295                                 request_module("net-pf-%d-proto-%d-type-%d",
 296                                                PF_INET, protocol, sock->type);
 297                         /*
 298                          * Fall back to generic, e.g. net-pf-2-proto-132
 299                          * (net-pf-PF_INET-proto-IPPROTO_SCTP)
 300                          */
 301                         else
 302                                 request_module("net-pf-%d-proto-%d",
 303                                                PF_INET, protocol);
 304                         goto lookup_protocol;
 305                 } else
 306                         goto out_rcu_unlock;
 307         }
 308
 309         err = -EPERM;
 310         if (sock->type == SOCK_RAW && !kern &&
 311             !ns_capable(net->user_ns, CAP_NET_RAW))
 312                 goto out_rcu_unlock;
 313
 314         sock->ops = answer->ops;
 315         answer_prot = answer->prot;
 316         answer_flags = answer->flags;
 317         rcu_read_unlock();
 318
 319         WARN_ON(!answer_prot->slab);
 320
 321         err = -ENOBUFS;
 322         sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
 323         if (!sk)
 324                 goto out;
 325
 326         err = 0;
 327         if (INET_PROTOSW_REUSE & answer_flags)
 328                 sk->sk_reuse = SK_CAN_REUSE;
 329
 330         inet = inet_sk(sk);
 331         inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
 332
 333         inet->nodefrag = 0;
 334
 335         if (SOCK_RAW == sock->type) {
 336                 inet->inet_num = protocol;
 337                 if (IPPROTO_RAW == protocol)
 338                         inet->hdrincl = 1;
 339         }
 340
 341         if (net->ipv4.sysctl_ip_no_pmtu_disc)
 342                 inet->pmtudisc = IP_PMTUDISC_DONT;
 343         else
 344                 inet->pmtudisc = IP_PMTUDISC_WANT;
 345
 346         inet->inet_id = 0;
 347
 348         sock_init_data(sock, sk);
 349
 350         sk->sk_destruct    = inet_sock_destruct;
 351         sk->sk_protocol    = protocol;
 352         sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
 353
 354         inet->uc_ttl    = -1;
 355         inet->mc_loop   = 1;
 356         inet->mc_ttl    = 1;
 357         inet->mc_all    = 1;
 358         inet->mc_index  = 0;
 359         inet->mc_list   = NULL;
 360         inet->rcv_tos   = 0;
 361
 362         sk_refcnt_debug_inc(sk);
 363
 364         if (inet->inet_num) {
 365                 /* It assumes that any protocol which allows
 366                  * the user to assign a number at socket
 367                  * creation time automatically
 368                  * shares.
 369                  */
 370                 inet->inet_sport = htons(inet->inet_num);
 371                 /* Add to protocol hash chains. */
 372                 err = sk->sk_prot->hash(sk);
 373                 if (err) {
 374                         sk_common_release(sk);
 375                         goto out;
 376                 }
 377         }
 378
 379         if (sk->sk_prot->init) {
 380                 err = sk->sk_prot->init(sk);
 381                 if (err) {
 382                         sk_common_release(sk);
 383                         goto out;
 384                 }
 385         }
 386
 387         if (!kern) {
 388                 err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
 389                 if (err) {
 390                         sk_common_release(sk);
 391                         goto out;
 392                 }
 393         }
 394 out:
 395         return err;
 396 out_rcu_unlock:
 397         rcu_read_unlock();
 398         goto out;
 399 }
 400
 401
 402 /*
 403  *      The peer socket should always be NULL (or else). When we call this
 404  *      function we are destroying the object and from then on nobody
 405  *      should refer to it.
 406  */
 407 int inet_release(struct socket *sock)
 408 {
 409         struct sock *sk = sock->sk;
 410
 411         if (sk) {
 412                 long timeout;
 413
 414                 /* Applications forget to leave groups before exiting */
 415                 ip_mc_drop_socket(sk);
 416
 417                 /* If linger is set, we don't return until the close
 418                  * is complete.  Otherwise we return immediately. The
 419                  * actually closing is done the same either way.
 420                  *
 421                  * If the close is due to the process exiting, we never
 422                  * linger..
 423                  */
 424                 timeout = 0;
 425                 if (sock_flag(sk, SOCK_LINGER) &&
 426                     !(current->flags & PF_EXITING))
 427                         timeout = sk->sk_lingertime;
 428                 sk->sk_prot->close(sk, timeout);
 429                 sock->sk = NULL;
 430         }
 431         return 0;
 432 }
 433 EXPORT_SYMBOL(inet_release);
 434
 435 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 436 {
 437         struct sock *sk = sock->sk;
 438         int err;
 439
 440         /* If the socket has its own bind function then use it. (RAW) */
 441         if (sk->sk_prot->bind) {
 442                 return sk->sk_prot->bind(sk, uaddr, addr_len);
 443         }
 444         if (addr_len < sizeof(struct sockaddr_in))
 445                 return -EINVAL;
 446
 447         /* BPF prog is run before any checks are done so that if the prog
 448          * changes context in a wrong way it will be caught.
 449          */
 450         err = BPF_CGROUP_RUN_PROG_INET4_BIND(sk, uaddr);
 451         if (err)
 452                 return err;
 453
 454         return __inet_bind(sk, uaddr, addr_len, BIND_WITH_LOCK);
 455 }
 456 EXPORT_SYMBOL(inet_bind);
 457
 458 int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
 459                 u32 flags)
 460 {
 461         struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
 462         struct inet_sock *inet = inet_sk(sk);
 463         struct net *net = sock_net(sk);
 464         unsigned short snum;
 465         int chk_addr_ret;
 466         u32 tb_id = RT_TABLE_LOCAL;
 467         int err;
 468
 469         if (addr->sin_family != AF_INET) {
 470                 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
 471                  * only if s_addr is INADDR_ANY.
 472                  */
 473                 err = -EAFNOSUPPORT;
 474                 if (addr->sin_family != AF_UNSPEC ||
 475                     addr->sin_addr.s_addr != htonl(INADDR_ANY))
 476                         goto out;
 477         }
 478
 479         tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
 480         chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
 481
 482         /* Not specified by any standard per-se, however it breaks too
 483          * many applications when removed.  It is unfortunate since
 484          * allowing applications to make a non-local bind solves
 485          * several problems with systems using dynamic addressing.
 486          * (ie. your servers still start up even if your ISDN link
 487          *  is temporarily down)
 488          */
 489         err = -EADDRNOTAVAIL;
 490         if (!inet_can_nonlocal_bind(net, inet) &&
 491             addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
 492             chk_addr_ret != RTN_LOCAL &&
 493             chk_addr_ret != RTN_MULTICAST &&
 494             chk_addr_ret != RTN_BROADCAST)
 495                 goto out;
 496
 497         snum = ntohs(addr->sin_port);
 498         err = -EACCES;
 499         if (snum && inet_port_requires_bind_service(net, snum) &&
 500             !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
 501                 goto out;
 502
 503         /*      We keep a pair of addresses. rcv_saddr is the one
 504          *      used by hash lookups, and saddr is used for transmit.
 505          *
 506          *      In the BSD API these are the same except where it
 507          *      would be illegal to use them (multicast/broadcast) in
 508          *      which case the sending device address is used.
 509          */
 510         if (flags & BIND_WITH_LOCK)
 511                 lock_sock(sk);
 512
 513         /* Check these errors (active socket, double bind). */
 514         err = -EINVAL;
 515         if (sk->sk_state != TCP_CLOSE || inet->inet_num)
 516                 goto out_release_sock;
 517
 518         inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
 519         if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
 520                 inet->inet_saddr = 0;  /* Use device */
 521
 522         /* Make sure we are allowed to bind here. */
 523         if (snum || !(inet->bind_address_no_port ||
 524                       (flags & BIND_FORCE_ADDRESS_NO_PORT))) {
 525                 if (sk->sk_prot->get_port(sk, snum)) {
 526                         inet->inet_saddr = inet->inet_rcv_saddr = 0;
 527                         err = -EADDRINUSE;
 528                         goto out_release_sock;
 529                 }
 530                 if (!(flags & BIND_FROM_BPF)) {
 531                         err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
 532                         if (err) {
 533                                 inet->inet_saddr = inet->inet_rcv_saddr = 0;
 534                                 goto out_release_sock;
 535                         }
 536                 }
 537         }
 538
 539         if (inet->inet_rcv_saddr)
 540                 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
 541         if (snum)
 542                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
 543         inet->inet_sport = htons(inet->inet_num);
 544         inet->inet_daddr = 0;
 545         inet->inet_dport = 0;
 546         sk_dst_reset(sk);
 547         err = 0;
 548 out_release_sock:
 549         if (flags & BIND_WITH_LOCK)
 550                 release_sock(sk);
 551 out:
 552         return err;
 553 }
 554
 555 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
 556                        int addr_len, int flags)
 557 {
 558         struct sock *sk = sock->sk;
 559         int err;
 560
 561         if (addr_len < sizeof(uaddr->sa_family))
 562                 return -EINVAL;
 563         if (uaddr->sa_family == AF_UNSPEC)
 564                 return sk->sk_prot->disconnect(sk, flags);
 565
 566         if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
 567                 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
 568                 if (err)
 569                         return err;
 570         }
 571
 572         if (!inet_sk(sk)->inet_num && inet_autobind(sk))
 573                 return -EAGAIN;
 574         return sk->sk_prot->connect(sk, uaddr, addr_len);
 575 }
 576 EXPORT_SYMBOL(inet_dgram_connect);
 577
 578 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
 579 {
 580         DEFINE_WAIT_FUNC(wait, woken_wake_function);
 581
 582         add_wait_queue(sk_sleep(sk), &wait);
 583         sk->sk_write_pending += writebias;
 584
 585         /* Basic assumption: if someone sets sk->sk_err, he _must_
 586          * change state of the socket from TCP_SYN_*.
 587          * Connect() does not allow to get error notifications
 588          * without closing the socket.
 589          */
 590         while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 591                 release_sock(sk);
 592                 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
 593                 lock_sock(sk);
 594                 if (signal_pending(current) || !timeo)
 595                         break;
 596         }
 597         remove_wait_queue(sk_sleep(sk), &wait);
 598         sk->sk_write_pending -= writebias;
 599         return timeo;
 600 }
 601
 602 /*
 603  *      Connect to a remote host. There is regrettably still a little
 604  *      TCP 'magic' in here.
 605  */
 606 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 607                           int addr_len, int flags, int is_sendmsg)
 608 {
 609         struct sock *sk = sock->sk;
 610         int err;
 611         long timeo;
 612
 613         /*
 614          * uaddr can be NULL and addr_len can be 0 if:
 615          * sk is a TCP fastopen active socket and
 616          * TCP_FASTOPEN_CONNECT sockopt is set and
 617          * we already have a valid cookie for this socket.
 618          * In this case, user can call write() after connect().
 619          * write() will invoke tcp_sendmsg_fastopen() which calls
 620          * __inet_stream_connect().
 621          */
 622         if (uaddr) {
 623                 if (addr_len < sizeof(uaddr->sa_family))
 624                         return -EINVAL;
 625
 626                 if (uaddr->sa_family == AF_UNSPEC) {
 627                         err = sk->sk_prot->disconnect(sk, flags);
 628                         sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 629                         goto out;
 630                 }
 631         }
 632
 633         switch (sock->state) {
 634         default:
 635                 err = -EINVAL;
 636                 goto out;
 637         case SS_CONNECTED:
 638                 err = -EISCONN;
 639                 goto out;
 640         case SS_CONNECTING:
 641                 if (inet_sk(sk)->defer_connect)
 642                         err = is_sendmsg ? -EINPROGRESS : -EISCONN;
 643                 else
 644                         err = -EALREADY;
 645                 /* Fall out of switch with err, set for this state */
 646                 break;
 647         case SS_UNCONNECTED:
 648                 err = -EISCONN;
 649                 if (sk->sk_state != TCP_CLOSE)
 650                         goto out;
 651
 652                 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
 653                         err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
 654                         if (err)
 655                                 goto out;
 656                 }
 657
 658                 err = sk->sk_prot->connect(sk, uaddr, addr_len);
 659                 if (err < 0)
 660                         goto out;
 661
 662                 sock->state = SS_CONNECTING;
 663
 664                 if (!err && inet_sk(sk)->defer_connect)
 665                         goto out;
 666
 667                 /* Just entered SS_CONNECTING state; the only
 668                  * difference is that return value in non-blocking
 669                  * case is EINPROGRESS, rather than EALREADY.
 670                  */
 671                 err = -EINPROGRESS;
 672                 break;
 673         }
 674
 675         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
 676
 677         if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 678                 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
 679                                 tcp_sk(sk)->fastopen_req &&
 680                                 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
 681
 682                 /* Error code is set above */
 683                 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
 684                         goto out;
 685
 686                 err = sock_intr_errno(timeo);
 687                 if (signal_pending(current))
 688                         goto out;
 689         }
 690
 691         /* Connection was closed by RST, timeout, ICMP error
 692          * or another process disconnected us.
 693          */
 694         if (sk->sk_state == TCP_CLOSE)
 695                 goto sock_error;
 696
 697         /* sk->sk_err may be not zero now, if RECVERR was ordered by user
 698          * and error was received after socket entered established state.
 699          * Hence, it is handled normally after connect() return successfully.
 700          */
 701
 702         sock->state = SS_CONNECTED;
 703         err = 0;
 704 out:
 705         return err;
 706
 707 sock_error:
 708         err = sock_error(sk) ? : -ECONNABORTED;
 709         sock->state = SS_UNCONNECTED;
 710         if (sk->sk_prot->disconnect(sk, flags))
 711                 sock->state = SS_DISCONNECTING;
 712         goto out;
 713 }
 714 EXPORT_SYMBOL(__inet_stream_connect);
 715
 716 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 717                         int addr_len, int flags)
 718 {
 719         int err;
 720
 721         lock_sock(sock->sk);
 722         err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
 723         release_sock(sock->sk);
 724         return err;
 725 }
 726 EXPORT_SYMBOL(inet_stream_connect);
 727
 728 /*
 729  *      Accept a pending connection. The TCP layer now gives BSD semantics.
 730  */
 731
 732 int inet_accept(struct socket *sock, struct socket *newsock, int flags,
 733                 bool kern)
 734 {
 735         struct sock *sk1 = sock->sk;
 736         int err = -EINVAL;
 737         struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err, kern);
 738
 739         if (!sk2)
 740                 goto do_err;
 741
 742         lock_sock(sk2);
 743
 744         sock_rps_record_flow(sk2);
 745         WARN_ON(!((1 << sk2->sk_state) &
 746                   (TCPF_ESTABLISHED | TCPF_SYN_RECV |
 747                   TCPF_CLOSE_WAIT | TCPF_CLOSE)));
 748
 749         sock_graft(sk2, newsock);
 750
 751         newsock->state = SS_CONNECTED;
 752         err = 0;
 753         release_sock(sk2);
 754 do_err:
 755         return err;
 756 }
 757 EXPORT_SYMBOL(inet_accept);
 758
 759 /*
 760  *      This does both peername and sockname.
 761  */
 762 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
 763                  int peer)
 764 {
 765         struct sock *sk         = sock->sk;
 766         struct inet_sock *inet  = inet_sk(sk);
 767         DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
 768
 769         sin->sin_family = AF_INET;
 770         if (peer) {
 771                 if (!inet->inet_dport ||
 772                     (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
 773                      peer == 1))
 774                         return -ENOTCONN;
 775                 sin->sin_port = inet->inet_dport;
 776                 sin->sin_addr.s_addr = inet->inet_daddr;
 777         } else {
 778                 __be32 addr = inet->inet_rcv_saddr;
 779                 if (!addr)
 780                         addr = inet->inet_saddr;
 781                 sin->sin_port = inet->inet_sport;
 782                 sin->sin_addr.s_addr = addr;
 783         }
 784         if (cgroup_bpf_enabled)
 785                 BPF_CGROUP_RUN_SA_PROG_LOCK(sk, (struct sockaddr *)sin,
 786                                             peer ? BPF_CGROUP_INET4_GETPEERNAME :
 787                                                    BPF_CGROUP_INET4_GETSOCKNAME,
 788                                             NULL);
 789         memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 790         return sizeof(*sin);
 791 }
 792 EXPORT_SYMBOL(inet_getname);
 793
 794 int inet_send_prepare(struct sock *sk)
 795 {
 796         sock_rps_record_flow(sk);
 797
 798         /* We may need to bind the socket. */
 799         if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
 800             inet_autobind(sk))
 801                 return -EAGAIN;
 802
 803         return 0;
 804 }
 805 EXPORT_SYMBOL_GPL(inet_send_prepare);
 806
 807 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
 808 {
 809         struct sock *sk = sock->sk;
 810
 811         if (unlikely(inet_send_prepare(sk)))
 812                 return -EAGAIN;
 813
 814         return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
 815                                sk, msg, size);
 816 }
 817 EXPORT_SYMBOL(inet_sendmsg);
 818
 819 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
 820                       size_t size, int flags)
 821 {
 822         struct sock *sk = sock->sk;
 823
 824         if (unlikely(inet_send_prepare(sk)))
 825                 return -EAGAIN;
 826
 827         if (sk->sk_prot->sendpage)
 828                 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
 829         return sock_no_sendpage(sock, page, offset, size, flags);
 830 }
 831 EXPORT_SYMBOL(inet_sendpage);
 832
 833 INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
 834                                           size_t, int, int, int *));
 835 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
 836                  int flags)
 837 {
 838         struct sock *sk = sock->sk;
 839         int addr_len = 0;
 840         int err;
 841
 842         if (likely(!(flags & MSG_ERRQUEUE)))
 843                 sock_rps_record_flow(sk);
 844
 845         err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
 846                               sk, msg, size, flags & MSG_DONTWAIT,
 847                               flags & ~MSG_DONTWAIT, &addr_len);
 848         if (err >= 0)
 849                 msg->msg_namelen = addr_len;
 850         return err;
 851 }
 852 EXPORT_SYMBOL(inet_recvmsg);
 853
 854 int inet_shutdown(struct socket *sock, int how)
 855 {
 856         struct sock *sk = sock->sk;
 857         int err = 0;
 858
 859         /* This should really check to make sure
 860          * the socket is a TCP socket. (WHY AC...)
 861          */
 862         how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
 863                        1->2 bit 2 snds.
 864                        2->3 */
 865         if ((how & ~SHUTDOWN_MASK) || !how)     /* MAXINT->0 */
 866                 return -EINVAL;
 867
 868         lock_sock(sk);
 869         if (sock->state == SS_CONNECTING) {
 870                 if ((1 << sk->sk_state) &
 871                     (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
 872                         sock->state = SS_DISCONNECTING;
 873                 else
 874                         sock->state = SS_CONNECTED;
 875         }
 876
 877         switch (sk->sk_state) {
 878         case TCP_CLOSE:
 879                 err = -ENOTCONN;
 880                 /* Hack to wake up other listeners, who can poll for
 881                    EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
 882                 fallthrough;
 883         default:
 884                 sk->sk_shutdown |= how;
 885                 if (sk->sk_prot->shutdown)
 886                         sk->sk_prot->shutdown(sk, how);
 887                 break;
 888
 889         /* Remaining two branches are temporary solution for missing
 890          * close() in multithreaded environment. It is _not_ a good idea,
 891          * but we have no choice until close() is repaired at VFS level.
 892          */
 893         case TCP_LISTEN:
 894                 if (!(how & RCV_SHUTDOWN))
 895                         break;
 896                 fallthrough;
 897         case TCP_SYN_SENT:
 898                 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
 899                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 900                 break;
 901         }
 902
 903         /* Wake up anyone sleeping in poll. */
 904         sk->sk_state_change(sk);
 905         release_sock(sk);
 906         return err;
 907 }
 908 EXPORT_SYMBOL(inet_shutdown);
 909
 910 /*
 911  *      ioctl() calls you can issue on an INET socket. Most of these are
 912  *      device configuration and stuff and very rarely used. Some ioctls
 913  *      pass on to the socket itself.
 914  *
 915  *      NOTE: I like the idea of a module for the config stuff. ie ifconfig
 916  *      loads the devconfigure module does its configuring and unloads it.
 917  *      There's a good 20K of config code hanging around the kernel.
 918  */
 919
 920 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 921 {
 922         struct sock *sk = sock->sk;
 923         int err = 0;
 924         struct net *net = sock_net(sk);
 925         void __user *p = (void __user *)arg;
 926         struct ifreq ifr;
 927         struct rtentry rt;
 928
 929         switch (cmd) {
 930         case SIOCADDRT:
 931         case SIOCDELRT:
 932                 if (copy_from_user(&rt, p, sizeof(struct rtentry)))
 933                         return -EFAULT;
 934                 err = ip_rt_ioctl(net, cmd, &rt);
 935                 break;
 936         case SIOCRTMSG:
 937                 err = -EINVAL;
 938                 break;
 939         case SIOCDARP:
 940         case SIOCGARP:
 941         case SIOCSARP:
 942                 err = arp_ioctl(net, cmd, (void __user *)arg);
 943                 break;
 944         case SIOCGIFADDR:
 945         case SIOCGIFBRDADDR:
 946         case SIOCGIFNETMASK:
 947         case SIOCGIFDSTADDR:
 948         case SIOCGIFPFLAGS:
 949                 if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
 950                         return -EFAULT;
 951                 err = devinet_ioctl(net, cmd, &ifr);
 952                 if (!err && copy_to_user(p, &ifr, sizeof(struct ifreq)))
 953                         err = -EFAULT;
 954                 break;
 955
 956         case SIOCSIFADDR:
 957         case SIOCSIFBRDADDR:
 958         case SIOCSIFNETMASK:
 959         case SIOCSIFDSTADDR:
 960         case SIOCSIFPFLAGS:
 961         case SIOCSIFFLAGS:
 962                 if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
 963                         return -EFAULT;
 964                 err = devinet_ioctl(net, cmd, &ifr);
 965                 break;
 966         default:
 967                 if (sk->sk_prot->ioctl)
 968                         err = sk->sk_prot->ioctl(sk, cmd, arg);
 969                 else
 970                         err = -ENOIOCTLCMD;
 971                 break;
 972         }
 973         return err;
 974 }
 975 EXPORT_SYMBOL(inet_ioctl);
 976
 977 #ifdef CONFIG_COMPAT
 978 static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd,
 979                 struct compat_rtentry __user *ur)
 980 {
 981         compat_uptr_t rtdev;
 982         struct rtentry rt;
 983
 984         if (copy_from_user(&rt.rt_dst, &ur->rt_dst,
 985                         3 * sizeof(struct sockaddr)) ||
 986             get_user(rt.rt_flags, &ur->rt_flags) ||
 987             get_user(rt.rt_metric, &ur->rt_metric) ||
 988             get_user(rt.rt_mtu, &ur->rt_mtu) ||
 989             get_user(rt.rt_window, &ur->rt_window) ||
 990             get_user(rt.rt_irtt, &ur->rt_irtt) ||
 991             get_user(rtdev, &ur->rt_dev))
 992                 return -EFAULT;
 993
 994         rt.rt_dev = compat_ptr(rtdev);
 995         return ip_rt_ioctl(sock_net(sk), cmd, &rt);
 996 }
 997
 998 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 999 {
1000         void __user *argp = compat_ptr(arg);
1001         struct sock *sk = sock->sk;
1002
1003         switch (cmd) {
1004         case SIOCADDRT:
1005         case SIOCDELRT:
1006                 return inet_compat_routing_ioctl(sk, cmd, argp);
1007         default:
1008                 if (!sk->sk_prot->compat_ioctl)
1009                         return -ENOIOCTLCMD;
1010                 return sk->sk_prot->compat_ioctl(sk, cmd, arg);
1011         }
1012 }
1013 #endif /* CONFIG_COMPAT */
1014
1015 const struct proto_ops inet_stream_ops = {
1016         .family            = PF_INET,
1017         .owner             = THIS_MODULE,
1018         .release           = inet_release,
1019         .bind              = inet_bind,
1020         .connect           = inet_stream_connect,
1021         .socketpair        = sock_no_socketpair,
1022         .accept            = inet_accept,
1023         .getname           = inet_getname,
1024         .poll              = tcp_poll,
1025         .ioctl             = inet_ioctl,
1026         .gettstamp         = sock_gettstamp,
1027         .listen            = inet_listen,
1028         .shutdown          = inet_shutdown,
1029         .setsockopt        = sock_common_setsockopt,
1030         .getsockopt        = sock_common_getsockopt,
1031         .sendmsg           = inet_sendmsg,
1032         .recvmsg           = inet_recvmsg,
1033 #ifdef CONFIG_MMU
1034         .mmap              = tcp_mmap,
1035 #endif
1036         .sendpage          = inet_sendpage,
1037         .splice_read       = tcp_splice_read,
1038         .read_sock         = tcp_read_sock,
1039         .sendmsg_locked    = tcp_sendmsg_locked,
1040         .sendpage_locked   = tcp_sendpage_locked,
1041         .peek_len          = tcp_peek_len,
1042 #ifdef CONFIG_COMPAT
1043         .compat_setsockopt = compat_sock_common_setsockopt,
1044         .compat_getsockopt = compat_sock_common_getsockopt,
1045         .compat_ioctl      = inet_compat_ioctl,
1046 #endif
1047         .set_rcvlowat      = tcp_set_rcvlowat,
1048 };
1049 EXPORT_SYMBOL(inet_stream_ops);
1050
1051 const struct proto_ops inet_dgram_ops = {
1052         .family            = PF_INET,
1053         .owner             = THIS_MODULE,
1054         .release           = inet_release,
1055         .bind              = inet_bind,
1056         .connect           = inet_dgram_connect,
1057         .socketpair        = sock_no_socketpair,
1058         .accept            = sock_no_accept,
1059         .getname           = inet_getname,
1060         .poll              = udp_poll,
1061         .ioctl             = inet_ioctl,
1062         .gettstamp         = sock_gettstamp,
1063         .listen            = sock_no_listen,
1064         .shutdown          = inet_shutdown,
1065         .setsockopt        = sock_common_setsockopt,
1066         .getsockopt        = sock_common_getsockopt,
1067         .sendmsg           = inet_sendmsg,
1068         .recvmsg           = inet_recvmsg,
1069         .mmap              = sock_no_mmap,
1070         .sendpage          = inet_sendpage,
1071         .set_peek_off      = sk_set_peek_off,
1072 #ifdef CONFIG_COMPAT
1073         .compat_setsockopt = compat_sock_common_setsockopt,
1074         .compat_getsockopt = compat_sock_common_getsockopt,
1075         .compat_ioctl      = inet_compat_ioctl,
1076 #endif
1077 };
1078 EXPORT_SYMBOL(inet_dgram_ops);
1079
1080 /*
1081  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1082  * udp_poll
1083  */
1084 static const struct proto_ops inet_sockraw_ops = {
1085         .family            = PF_INET,
1086         .owner             = THIS_MODULE,
1087         .release           = inet_release,
1088         .bind              = inet_bind,
1089         .connect           = inet_dgram_connect,
1090         .socketpair        = sock_no_socketpair,
1091         .accept            = sock_no_accept,
1092         .getname           = inet_getname,
1093         .poll              = datagram_poll,
1094         .ioctl             = inet_ioctl,
1095         .gettstamp         = sock_gettstamp,
1096         .listen            = sock_no_listen,
1097         .shutdown          = inet_shutdown,
1098         .setsockopt        = sock_common_setsockopt,
1099         .getsockopt        = sock_common_getsockopt,
1100         .sendmsg           = inet_sendmsg,
1101         .recvmsg           = inet_recvmsg,
1102         .mmap              = sock_no_mmap,
1103         .sendpage          = inet_sendpage,
1104 #ifdef CONFIG_COMPAT
1105         .compat_setsockopt = compat_sock_common_setsockopt,
1106         .compat_getsockopt = compat_sock_common_getsockopt,
1107         .compat_ioctl      = inet_compat_ioctl,
1108 #endif
1109 };
1110
1111 static const struct net_proto_family inet_family_ops = {
1112         .family = PF_INET,
1113         .create = inet_create,
1114         .owner  = THIS_MODULE,
1115 };
1116
1117 /* Upon startup we insert all the elements in inetsw_array[] into
1118  * the linked list inetsw.
1119  */
1120 static struct inet_protosw inetsw_array[] =
1121 {
1122         {
1123                 .type =       SOCK_STREAM,
1124                 .protocol =   IPPROTO_TCP,
1125                 .prot =       &tcp_prot,
1126                 .ops =        &inet_stream_ops,
1127                 .flags =      INET_PROTOSW_PERMANENT |
1128                               INET_PROTOSW_ICSK,
1129         },
1130
1131         {
1132                 .type =       SOCK_DGRAM,
1133                 .protocol =   IPPROTO_UDP,
1134                 .prot =       &udp_prot,
1135                 .ops =        &inet_dgram_ops,
1136                 .flags =      INET_PROTOSW_PERMANENT,
1137        },
1138
1139        {
1140                 .type =       SOCK_DGRAM,
1141                 .protocol =   IPPROTO_ICMP,
1142                 .prot =       &ping_prot,
1143                 .ops =        &inet_sockraw_ops,
1144                 .flags =      INET_PROTOSW_REUSE,
1145        },
1146
1147        {
1148                .type =       SOCK_RAW,
1149                .protocol =   IPPROTO_IP,        /* wild card */
1150                .prot =       &raw_prot,
1151                .ops =        &inet_sockraw_ops,
1152                .flags =      INET_PROTOSW_REUSE,
1153        }
1154 };
1155
1156 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1157
1158 void inet_register_protosw(struct inet_protosw *p)
1159 {
1160         struct list_head *lh;
1161         struct inet_protosw *answer;
1162         int protocol = p->protocol;
1163         struct list_head *last_perm;
1164
1165         spin_lock_bh(&inetsw_lock);
1166
1167         if (p->type >= SOCK_MAX)
1168                 goto out_illegal;
1169
1170         /* If we are trying to override a permanent protocol, bail. */
1171         last_perm = &inetsw[p->type];
1172         list_for_each(lh, &inetsw[p->type]) {
1173                 answer = list_entry(lh, struct inet_protosw, list);
1174                 /* Check only the non-wild match. */
1175                 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1176                         break;
1177                 if (protocol == answer->protocol)
1178                         goto out_permanent;
1179                 last_perm = lh;
1180         }
1181
1182         /* Add the new entry after the last permanent entry if any, so that
1183          * the new entry does not override a permanent entry when matched with
1184          * a wild-card protocol. But it is allowed to override any existing
1185          * non-permanent entry.  This means that when we remove this entry, the
1186          * system automatically returns to the old behavior.
1187          */
1188         list_add_rcu(&p->list, last_perm);
1189 out:
1190         spin_unlock_bh(&inetsw_lock);
1191
1192         return;
1193
1194 out_permanent:
1195         pr_err("Attempt to override permanent protocol %d\n", protocol);
1196         goto out;
1197
1198 out_illegal:
1199         pr_err("Ignoring attempt to register invalid socket type %d\n",
1200                p->type);
1201         goto out;
1202 }
1203 EXPORT_SYMBOL(inet_register_protosw);
1204
1205 void inet_unregister_protosw(struct inet_protosw *p)
1206 {
1207         if (INET_PROTOSW_PERMANENT & p->flags) {
1208                 pr_err("Attempt to unregister permanent protocol %d\n",
1209                        p->protocol);
1210         } else {
1211                 spin_lock_bh(&inetsw_lock);
1212                 list_del_rcu(&p->list);
1213                 spin_unlock_bh(&inetsw_lock);
1214
1215                 synchronize_net();
1216         }
1217 }
1218 EXPORT_SYMBOL(inet_unregister_protosw);
1219
1220 static int inet_sk_reselect_saddr(struct sock *sk)
1221 {
1222         struct inet_sock *inet = inet_sk(sk);
1223         __be32 old_saddr = inet->inet_saddr;
1224         __be32 daddr = inet->inet_daddr;
1225         struct flowi4 *fl4;
1226         struct rtable *rt;
1227         __be32 new_saddr;
1228         struct ip_options_rcu *inet_opt;
1229
1230         inet_opt = rcu_dereference_protected(inet->inet_opt,
1231                                              lockdep_sock_is_held(sk));
1232         if (inet_opt && inet_opt->opt.srr)
1233                 daddr = inet_opt->opt.faddr;
1234
1235         /* Query new route. */
1236         fl4 = &inet->cork.fl.u.ip4;
1237         rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1238                               sk->sk_bound_dev_if, sk->sk_protocol,
1239                               inet->inet_sport, inet->inet_dport, sk);
1240         if (IS_ERR(rt))
1241                 return PTR_ERR(rt);
1242
1243         sk_setup_caps(sk, &rt->dst);
1244
1245         new_saddr = fl4->saddr;
1246
1247         if (new_saddr == old_saddr)
1248                 return 0;
1249
1250         if (sock_net(sk)->ipv4.sysctl_ip_dynaddr > 1) {
1251                 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1252                         __func__, &old_saddr, &new_saddr);
1253         }
1254
1255         inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1256
1257         /*
1258          * XXX The only one ugly spot where we need to
1259          * XXX really change the sockets identity after
1260          * XXX it has entered the hashes. -DaveM
1261          *
1262          * Besides that, it does not check for connection
1263          * uniqueness. Wait for troubles.
1264          */
1265         return __sk_prot_rehash(sk);
1266 }
1267
1268 int inet_sk_rebuild_header(struct sock *sk)
1269 {
1270         struct inet_sock *inet = inet_sk(sk);
1271         struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1272         __be32 daddr;
1273         struct ip_options_rcu *inet_opt;
1274         struct flowi4 *fl4;
1275         int err;
1276
1277         /* Route is OK, nothing to do. */
1278         if (rt)
1279                 return 0;
1280
1281         /* Reroute. */
1282         rcu_read_lock();
1283         inet_opt = rcu_dereference(inet->inet_opt);
1284         daddr = inet->inet_daddr;
1285         if (inet_opt && inet_opt->opt.srr)
1286                 daddr = inet_opt->opt.faddr;
1287         rcu_read_unlock();
1288         fl4 = &inet->cork.fl.u.ip4;
1289         rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1290                                    inet->inet_dport, inet->inet_sport,
1291                                    sk->sk_protocol, RT_CONN_FLAGS(sk),
1292                                    sk->sk_bound_dev_if);
1293         if (!IS_ERR(rt)) {
1294                 err = 0;
1295                 sk_setup_caps(sk, &rt->dst);
1296         } else {
1297                 err = PTR_ERR(rt);
1298
1299                 /* Routing failed... */
1300                 sk->sk_route_caps = 0;
1301                 /*
1302                  * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1303                  * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1304                  */
1305                 if (!sock_net(sk)->ipv4.sysctl_ip_dynaddr ||
1306                     sk->sk_state != TCP_SYN_SENT ||
1307                     (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1308                     (err = inet_sk_reselect_saddr(sk)) != 0)
1309                         sk->sk_err_soft = -err;
1310         }
1311
1312         return err;
1313 }
1314 EXPORT_SYMBOL(inet_sk_rebuild_header);
1315
1316 void inet_sk_set_state(struct sock *sk, int state)
1317 {
1318         trace_inet_sock_set_state(sk, sk->sk_state, state);
1319         sk->sk_state = state;
1320 }
1321 EXPORT_SYMBOL(inet_sk_set_state);
1322
1323 void inet_sk_state_store(struct sock *sk, int newstate)
1324 {
1325         trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1326         smp_store_release(&sk->sk_state, newstate);
1327 }
1328
1329 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1330                                  netdev_features_t features)
1331 {
1332         bool udpfrag = false, fixedid = false, gso_partial, encap;
1333         struct sk_buff *segs = ERR_PTR(-EINVAL);
1334         const struct net_offload *ops;
1335         unsigned int offset = 0;
1336         struct iphdr *iph;
1337         int proto, tot_len;
1338         int nhoff;
1339         int ihl;
1340         int id;
1341
1342         skb_reset_network_header(skb);
1343         nhoff = skb_network_header(skb) - skb_mac_header(skb);
1344         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1345                 goto out;
1346
1347         iph = ip_hdr(skb);
1348         ihl = iph->ihl * 4;
1349         if (ihl < sizeof(*iph))
1350                 goto out;
1351
1352         id = ntohs(iph->id);
1353         proto = iph->protocol;
1354
1355         /* Warning: after this point, iph might be no longer valid */
1356         if (unlikely(!pskb_may_pull(skb, ihl)))
1357                 goto out;
1358         __skb_pull(skb, ihl);
1359
1360         encap = SKB_GSO_CB(skb)->encap_level > 0;
1361         if (encap)
1362                 features &= skb->dev->hw_enc_features;
1363         SKB_GSO_CB(skb)->encap_level += ihl;
1364
1365         skb_reset_transport_header(skb);
1366
1367         segs = ERR_PTR(-EPROTONOSUPPORT);
1368
1369         if (!skb->encapsulation || encap) {
1370                 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1371                 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1372
1373                 /* fixed ID is invalid if DF bit is not set */
1374                 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1375                         goto out;
1376         }
1377
1378         ops = rcu_dereference(inet_offloads[proto]);
1379         if (likely(ops && ops->callbacks.gso_segment))
1380                 segs = ops->callbacks.gso_segment(skb, features);
1381
1382         if (IS_ERR_OR_NULL(segs))
1383                 goto out;
1384
1385         gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1386
1387         skb = segs;
1388         do {
1389                 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1390                 if (udpfrag) {
1391                         iph->frag_off = htons(offset >> 3);
1392                         if (skb->next)
1393                                 iph->frag_off |= htons(IP_MF);
1394                         offset += skb->len - nhoff - ihl;
1395                         tot_len = skb->len - nhoff;
1396                 } else if (skb_is_gso(skb)) {
1397                         if (!fixedid) {
1398                                 iph->id = htons(id);
1399                                 id += skb_shinfo(skb)->gso_segs;
1400                         }
1401
1402                         if (gso_partial)
1403                                 tot_len = skb_shinfo(skb)->gso_size +
1404                                           SKB_GSO_CB(skb)->data_offset +
1405                                           skb->head - (unsigned char *)iph;
1406                         else
1407                                 tot_len = skb->len - nhoff;
1408                 } else {
1409                         if (!fixedid)
1410                                 iph->id = htons(id++);
1411                         tot_len = skb->len - nhoff;
1412                 }
1413                 iph->tot_len = htons(tot_len);
1414                 ip_send_check(iph);
1415                 if (encap)
1416                         skb_reset_inner_headers(skb);
1417                 skb->network_header = (u8 *)iph - skb->head;
1418                 skb_reset_mac_len(skb);
1419         } while ((skb = skb->next));
1420
1421 out:
1422         return segs;
1423 }
1424 EXPORT_SYMBOL(inet_gso_segment);
1425
1426 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1427                                         netdev_features_t features)
1428 {
1429         if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1430                 return ERR_PTR(-EINVAL);
1431
1432         return inet_gso_segment(skb, features);
1433 }
1434
1435 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *,
1436                                                            struct sk_buff *));
1437 INDIRECT_CALLABLE_DECLARE(struct sk_buff *udp4_gro_receive(struct list_head *,
1438                                                            struct sk_buff *));
1439 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1440 {
1441         const struct net_offload *ops;
1442         struct sk_buff *pp = NULL;
1443         const struct iphdr *iph;
1444         struct sk_buff *p;
1445         unsigned int hlen;
1446         unsigned int off;
1447         unsigned int id;
1448         int flush = 1;
1449         int proto;
1450
1451         off = skb_gro_offset(skb);
1452         hlen = off + sizeof(*iph);
1453         iph = skb_gro_header_fast(skb, off);
1454         if (skb_gro_header_hard(skb, hlen)) {
1455                 iph = skb_gro_header_slow(skb, hlen, off);
1456                 if (unlikely(!iph))
1457                         goto out;
1458         }
1459
1460         proto = iph->protocol;
1461
1462         rcu_read_lock();
1463         ops = rcu_dereference(inet_offloads[proto]);
1464         if (!ops || !ops->callbacks.gro_receive)
1465                 goto out_unlock;
1466
1467         if (*(u8 *)iph != 0x45)
1468                 goto out_unlock;
1469
1470         if (ip_is_fragment(iph))
1471                 goto out_unlock;
1472
1473         if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1474                 goto out_unlock;
1475
1476         id = ntohl(*(__be32 *)&iph->id);
1477         flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1478         id >>= 16;
1479
1480         list_for_each_entry(p, head, list) {
1481                 struct iphdr *iph2;
1482                 u16 flush_id;
1483
1484                 if (!NAPI_GRO_CB(p)->same_flow)
1485                         continue;
1486
1487                 iph2 = (struct iphdr *)(p->data + off);
1488                 /* The above works because, with the exception of the top
1489                  * (inner most) layer, we only aggregate pkts with the same
1490                  * hdr length so all the hdrs we'll need to verify will start
1491                  * at the same offset.
1492                  */
1493                 if ((iph->protocol ^ iph2->protocol) |
1494                     ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1495                     ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1496                         NAPI_GRO_CB(p)->same_flow = 0;
1497                         continue;
1498                 }
1499
1500                 /* All fields must match except length and checksum. */
1501                 NAPI_GRO_CB(p)->flush |=
1502                         (iph->ttl ^ iph2->ttl) |
1503                         (iph->tos ^ iph2->tos) |
1504                         ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1505
1506                 NAPI_GRO_CB(p)->flush |= flush;
1507
1508                 /* We need to store of the IP ID check to be included later
1509                  * when we can verify that this packet does in fact belong
1510                  * to a given flow.
1511                  */
1512                 flush_id = (u16)(id - ntohs(iph2->id));
1513
1514                 /* This bit of code makes it much easier for us to identify
1515                  * the cases where we are doing atomic vs non-atomic IP ID
1516                  * checks.  Specifically an atomic check can return IP ID
1517                  * values 0 - 0xFFFF, while a non-atomic check can only
1518                  * return 0 or 0xFFFF.
1519                  */
1520                 if (!NAPI_GRO_CB(p)->is_atomic ||
1521                     !(iph->frag_off & htons(IP_DF))) {
1522                         flush_id ^= NAPI_GRO_CB(p)->count;
1523                         flush_id = flush_id ? 0xFFFF : 0;
1524                 }
1525
1526                 /* If the previous IP ID value was based on an atomic
1527                  * datagram we can overwrite the value and ignore it.
1528                  */
1529                 if (NAPI_GRO_CB(skb)->is_atomic)
1530                         NAPI_GRO_CB(p)->flush_id = flush_id;
1531                 else
1532                         NAPI_GRO_CB(p)->flush_id |= flush_id;
1533         }
1534
1535         NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1536         NAPI_GRO_CB(skb)->flush |= flush;
1537         skb_set_network_header(skb, off);
1538         /* The above will be needed by the transport layer if there is one
1539          * immediately following this IP hdr.
1540          */
1541
1542         /* Note : No need to call skb_gro_postpull_rcsum() here,
1543          * as we already checked checksum over ipv4 header was 0
1544          */
1545         skb_gro_pull(skb, sizeof(*iph));
1546         skb_set_transport_header(skb, skb_gro_offset(skb));
1547
1548         pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1549                                        ops->callbacks.gro_receive, head, skb);
1550
1551 out_unlock:
1552         rcu_read_unlock();
1553
1554 out:
1555         skb_gro_flush_final(skb, pp, flush);
1556
1557         return pp;
1558 }
1559 EXPORT_SYMBOL(inet_gro_receive);
1560
1561 static struct sk_buff *ipip_gro_receive(struct list_head *head,
1562                                         struct sk_buff *skb)
1563 {
1564         if (NAPI_GRO_CB(skb)->encap_mark) {
1565                 NAPI_GRO_CB(skb)->flush = 1;
1566                 return NULL;
1567         }
1568
1569         NAPI_GRO_CB(skb)->encap_mark = 1;
1570
1571         return inet_gro_receive(head, skb);
1572 }
1573
1574 #define SECONDS_PER_DAY 86400
1575
1576 /* inet_current_timestamp - Return IP network timestamp
1577  *
1578  * Return milliseconds since midnight in network byte order.
1579  */
1580 __be32 inet_current_timestamp(void)
1581 {
1582         u32 secs;
1583         u32 msecs;
1584         struct timespec64 ts;
1585
1586         ktime_get_real_ts64(&ts);
1587
1588         /* Get secs since midnight. */
1589         (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1590         /* Convert to msecs. */
1591         msecs = secs * MSEC_PER_SEC;
1592         /* Convert nsec to msec. */
1593         msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1594
1595         /* Convert to network byte order. */
1596         return htonl(msecs);
1597 }
1598 EXPORT_SYMBOL(inet_current_timestamp);
1599
1600 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1601 {
1602         if (sk->sk_family == AF_INET)
1603                 return ip_recv_error(sk, msg, len, addr_len);
1604 #if IS_ENABLED(CONFIG_IPV6)
1605         if (sk->sk_family == AF_INET6)
1606                 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1607 #endif
1608         return -EINVAL;
1609 }
1610
1611 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *, int));
1612 INDIRECT_CALLABLE_DECLARE(int udp4_gro_complete(struct sk_buff *, int));
1613 int inet_gro_complete(struct sk_buff *skb, int nhoff)
1614 {
1615         __be16 newlen = htons(skb->len - nhoff);
1616         struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1617         const struct net_offload *ops;
1618         int proto = iph->protocol;
1619         int err = -ENOSYS;
1620
1621         if (skb->encapsulation) {
1622                 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1623                 skb_set_inner_network_header(skb, nhoff);
1624         }
1625
1626         csum_replace2(&iph->check, iph->tot_len, newlen);
1627         iph->tot_len = newlen;
1628
1629         rcu_read_lock();
1630         ops = rcu_dereference(inet_offloads[proto]);
1631         if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1632                 goto out_unlock;
1633
1634         /* Only need to add sizeof(*iph) to get to the next hdr below
1635          * because any hdr with option will have been flushed in
1636          * inet_gro_receive().
1637          */
1638         err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1639                               tcp4_gro_complete, udp4_gro_complete,
1640                               skb, nhoff + sizeof(*iph));
1641
1642 out_unlock:
1643         rcu_read_unlock();
1644
1645         return err;
1646 }
1647 EXPORT_SYMBOL(inet_gro_complete);
1648
1649 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1650 {
1651         skb->encapsulation = 1;
1652         skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1653         return inet_gro_complete(skb, nhoff);
1654 }
1655
1656 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1657                          unsigned short type, unsigned char protocol,
1658                          struct net *net)
1659 {
1660         struct socket *sock;
1661         int rc = sock_create_kern(net, family, type, protocol, &sock);
1662
1663         if (rc == 0) {
1664                 *sk = sock->sk;
1665                 (*sk)->sk_allocation = GFP_ATOMIC;
1666                 /*
1667                  * Unhash it so that IP input processing does not even see it,
1668                  * we do not wish this socket to see incoming packets.
1669                  */
1670                 (*sk)->sk_prot->unhash(*sk);
1671         }
1672         return rc;
1673 }
1674 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1675
1676 u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt)
1677 {
1678         return  *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt);
1679 }
1680 EXPORT_SYMBOL_GPL(snmp_get_cpu_field);
1681
1682 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1683 {
1684         unsigned long res = 0;
1685         int i;
1686
1687         for_each_possible_cpu(i)
1688                 res += snmp_get_cpu_field(mib, i, offt);
1689         return res;
1690 }
1691 EXPORT_SYMBOL_GPL(snmp_fold_field);
1692
1693 #if BITS_PER_LONG==32
1694
1695 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1696                          size_t syncp_offset)
1697 {
1698         void *bhptr;
1699         struct u64_stats_sync *syncp;
1700         u64 v;
1701         unsigned int start;
1702
1703         bhptr = per_cpu_ptr(mib, cpu);
1704         syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1705         do {
1706                 start = u64_stats_fetch_begin_irq(syncp);
1707                 v = *(((u64 *)bhptr) + offt);
1708         } while (u64_stats_fetch_retry_irq(syncp, start));
1709
1710         return v;
1711 }
1712 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1713
1714 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1715 {
1716         u64 res = 0;
1717         int cpu;
1718
1719         for_each_possible_cpu(cpu) {
1720                 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1721         }
1722         return res;
1723 }
1724 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1725 #endif
1726
1727 #ifdef CONFIG_IP_MULTICAST
1728 static const struct net_protocol igmp_protocol = {
1729         .handler =      igmp_rcv,
1730         .netns_ok =     1,
1731 };
1732 #endif
1733
1734 /* thinking of making this const? Don't.
1735  * early_demux can change based on sysctl.
1736  */
1737 static struct net_protocol tcp_protocol = {
1738         .early_demux    =       tcp_v4_early_demux,
1739         .early_demux_handler =  tcp_v4_early_demux,
1740         .handler        =       tcp_v4_rcv,
1741         .err_handler    =       tcp_v4_err,
1742         .no_policy      =       1,
1743         .netns_ok       =       1,
1744         .icmp_strict_tag_validation = 1,
1745 };
1746
1747 /* thinking of making this const? Don't.
1748  * early_demux can change based on sysctl.
1749  */
1750 static struct net_protocol udp_protocol = {
1751         .early_demux =  udp_v4_early_demux,
1752         .early_demux_handler =  udp_v4_early_demux,
1753         .handler =      udp_rcv,
1754         .err_handler =  udp_err,
1755         .no_policy =    1,
1756         .netns_ok =     1,
1757 };
1758
1759 static const struct net_protocol icmp_protocol = {
1760         .handler =      icmp_rcv,
1761         .err_handler =  icmp_err,
1762         .no_policy =    1,
1763         .netns_ok =     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         return 0;
1883 }
1884
1885 static __net_initdata struct pernet_operations af_inet_ops = {
1886         .init = inet_init_net,
1887 };
1888
1889 static int __init init_inet_pernet_ops(void)
1890 {
1891         return register_pernet_subsys(&af_inet_ops);
1892 }
1893
1894 static int ipv4_proc_init(void);
1895
1896 /*
1897  *      IP protocol layer initialiser
1898  */
1899
1900 static struct packet_offload ip_packet_offload __read_mostly = {
1901         .type = cpu_to_be16(ETH_P_IP),
1902         .callbacks = {
1903                 .gso_segment = inet_gso_segment,
1904                 .gro_receive = inet_gro_receive,
1905                 .gro_complete = inet_gro_complete,
1906         },
1907 };
1908
1909 static const struct net_offload ipip_offload = {
1910         .callbacks = {
1911                 .gso_segment    = ipip_gso_segment,
1912                 .gro_receive    = ipip_gro_receive,
1913                 .gro_complete   = ipip_gro_complete,
1914         },
1915 };
1916
1917 static int __init ipip_offload_init(void)
1918 {
1919         return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1920 }
1921
1922 static int __init ipv4_offload_init(void)
1923 {
1924         /*
1925          * Add offloads
1926          */
1927         if (udpv4_offload_init() < 0)
1928                 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1929         if (tcpv4_offload_init() < 0)
1930                 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1931         if (ipip_offload_init() < 0)
1932                 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1933
1934         dev_add_offload(&ip_packet_offload);
1935         return 0;
1936 }
1937
1938 fs_initcall(ipv4_offload_init);
1939
1940 static struct packet_type ip_packet_type __read_mostly = {
1941         .type = cpu_to_be16(ETH_P_IP),
1942         .func = ip_rcv,
1943         .list_func = ip_list_rcv,
1944 };
1945
1946 static int __init inet_init(void)
1947 {
1948         struct inet_protosw *q;
1949         struct list_head *r;
1950         int rc;
1951
1952         sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1953
1954         rc = proto_register(&tcp_prot, 1);
1955         if (rc)
1956                 goto out;
1957
1958         rc = proto_register(&udp_prot, 1);
1959         if (rc)
1960                 goto out_unregister_tcp_proto;
1961
1962         rc = proto_register(&raw_prot, 1);
1963         if (rc)
1964                 goto out_unregister_udp_proto;
1965
1966         rc = proto_register(&ping_prot, 1);
1967         if (rc)
1968                 goto out_unregister_raw_proto;
1969
1970         /*
1971          *      Tell SOCKET that we are alive...
1972          */
1973
1974         (void)sock_register(&inet_family_ops);
1975
1976 #ifdef CONFIG_SYSCTL
1977         ip_static_sysctl_init();
1978 #endif
1979
1980         /*
1981          *      Add all the base protocols.
1982          */
1983
1984         if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1985                 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1986         if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1987                 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1988         if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1989                 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1990 #ifdef CONFIG_IP_MULTICAST
1991         if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1992                 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1993 #endif
1994
1995         /* Register the socket-side information for inet_create. */
1996         for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1997                 INIT_LIST_HEAD(r);
1998
1999         for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
2000                 inet_register_protosw(q);
2001
2002         /*
2003          *      Set the ARP module up
2004          */
2005
2006         arp_init();
2007
2008         /*
2009          *      Set the IP module up
2010          */
2011
2012         ip_init();
2013
2014         /* Setup TCP slab cache for open requests. */
2015         tcp_init();
2016
2017         /* Setup UDP memory threshold */
2018         udp_init();
2019
2020         /* Add UDP-Lite (RFC 3828) */
2021         udplite4_register();
2022
2023         raw_init();
2024
2025         ping_init();
2026
2027         /*
2028          *      Set the ICMP layer up
2029          */
2030
2031         if (icmp_init() < 0)
2032                 panic("Failed to create the ICMP control socket.\n");
2033
2034         /*
2035          *      Initialise the multicast router
2036          */
2037 #if defined(CONFIG_IP_MROUTE)
2038         if (ip_mr_init())
2039                 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2040 #endif
2041
2042         if (init_inet_pernet_ops())
2043                 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2044         /*
2045          *      Initialise per-cpu ipv4 mibs
2046          */
2047
2048         if (init_ipv4_mibs())
2049                 pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
2050
2051         ipv4_proc_init();
2052
2053         ipfrag_init();
2054
2055         dev_add_pack(&ip_packet_type);
2056
2057         ip_tunnel_core_init();
2058
2059         rc = 0;
2060 out:
2061         return rc;
2062 out_unregister_raw_proto:
2063         proto_unregister(&raw_prot);
2064 out_unregister_udp_proto:
2065         proto_unregister(&udp_prot);
2066 out_unregister_tcp_proto:
2067         proto_unregister(&tcp_prot);
2068         goto out;
2069 }
2070
2071 fs_initcall(inet_init);
2072
2073 /* ------------------------------------------------------------------------ */
2074
2075 #ifdef CONFIG_PROC_FS
2076 static int __init ipv4_proc_init(void)
2077 {
2078         int rc = 0;
2079
2080         if (raw_proc_init())
2081                 goto out_raw;
2082         if (tcp4_proc_init())
2083                 goto out_tcp;
2084         if (udp4_proc_init())
2085                 goto out_udp;
2086         if (ping_proc_init())
2087                 goto out_ping;
2088         if (ip_misc_proc_init())
2089                 goto out_misc;
2090 out:
2091         return rc;
2092 out_misc:
2093         ping_proc_exit();
2094 out_ping:
2095         udp4_proc_exit();
2096 out_udp:
2097         tcp4_proc_exit();
2098 out_tcp:
2099         raw_proc_exit();
2100 out_raw:
2101         rc = -ENOMEM;
2102         goto out;
2103 }
2104
2105 #else /* CONFIG_PROC_FS */
2106 static int __init ipv4_proc_init(void)
2107 {
2108         return 0;
2109 }
2110 #endif /* CONFIG_PROC_FS */