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