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                 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         int err;
 442
 443         /* If the socket has its own bind function then use it. (RAW) */
 444         if (sk->sk_prot->bind) {
 445                 return sk->sk_prot->bind(sk, uaddr, addr_len);
 446         }
 447         if (addr_len < sizeof(struct sockaddr_in))
 448                 return -EINVAL;
 449
 450         /* BPF prog is run before any checks are done so that if the prog
 451          * changes context in a wrong way it will be caught.
 452          */
 453         err = BPF_CGROUP_RUN_PROG_INET4_BIND(sk, uaddr);
 454         if (err)
 455                 return err;
 456
 457         return __inet_bind(sk, uaddr, addr_len, BIND_WITH_LOCK);
 458 }
 459 EXPORT_SYMBOL(inet_bind);
 460
 461 int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
 462                 u32 flags)
 463 {
 464         struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
 465         struct inet_sock *inet = inet_sk(sk);
 466         struct net *net = sock_net(sk);
 467         unsigned short snum;
 468         int chk_addr_ret;
 469         u32 tb_id = RT_TABLE_LOCAL;
 470         int err;
 471
 472         if (addr->sin_family != AF_INET) {
 473                 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
 474                  * only if s_addr is INADDR_ANY.
 475                  */
 476                 err = -EAFNOSUPPORT;
 477                 if (addr->sin_family != AF_UNSPEC ||
 478                     addr->sin_addr.s_addr != htonl(INADDR_ANY))
 479                         goto out;
 480         }
 481
 482         tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
 483         chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
 484
 485         /* Not specified by any standard per-se, however it breaks too
 486          * many applications when removed.  It is unfortunate since
 487          * allowing applications to make a non-local bind solves
 488          * several problems with systems using dynamic addressing.
 489          * (ie. your servers still start up even if your ISDN link
 490          *  is temporarily down)
 491          */
 492         err = -EADDRNOTAVAIL;
 493         if (!inet_can_nonlocal_bind(net, inet) &&
 494             addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
 495             chk_addr_ret != RTN_LOCAL &&
 496             chk_addr_ret != RTN_MULTICAST &&
 497             chk_addr_ret != RTN_BROADCAST)
 498                 goto out;
 499
 500         snum = ntohs(addr->sin_port);
 501         err = -EACCES;
 502         if (snum && inet_port_requires_bind_service(net, snum) &&
 503             !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
 504                 goto out;
 505
 506         /*      We keep a pair of addresses. rcv_saddr is the one
 507          *      used by hash lookups, and saddr is used for transmit.
 508          *
 509          *      In the BSD API these are the same except where it
 510          *      would be illegal to use them (multicast/broadcast) in
 511          *      which case the sending device address is used.
 512          */
 513         if (flags & BIND_WITH_LOCK)
 514                 lock_sock(sk);
 515
 516         /* Check these errors (active socket, double bind). */
 517         err = -EINVAL;
 518         if (sk->sk_state != TCP_CLOSE || inet->inet_num)
 519                 goto out_release_sock;
 520
 521         inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
 522         if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
 523                 inet->inet_saddr = 0;  /* Use device */
 524
 525         /* Make sure we are allowed to bind here. */
 526         if (snum || !(inet->bind_address_no_port ||
 527                       (flags & BIND_FORCE_ADDRESS_NO_PORT))) {
 528                 if (sk->sk_prot->get_port(sk, snum)) {
 529                         inet->inet_saddr = inet->inet_rcv_saddr = 0;
 530                         err = -EADDRINUSE;
 531                         goto out_release_sock;
 532                 }
 533                 if (!(flags & BIND_FROM_BPF)) {
 534                         err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
 535                         if (err) {
 536                                 inet->inet_saddr = inet->inet_rcv_saddr = 0;
 537                                 goto out_release_sock;
 538                         }
 539                 }
 540         }
 541
 542         if (inet->inet_rcv_saddr)
 543                 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
 544         if (snum)
 545                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
 546         inet->inet_sport = htons(inet->inet_num);
 547         inet->inet_daddr = 0;
 548         inet->inet_dport = 0;
 549         sk_dst_reset(sk);
 550         err = 0;
 551 out_release_sock:
 552         if (flags & BIND_WITH_LOCK)
 553                 release_sock(sk);
 554 out:
 555         return err;
 556 }
 557
 558 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
 559                        int addr_len, int flags)
 560 {
 561         struct sock *sk = sock->sk;
 562         int err;
 563
 564         if (addr_len < sizeof(uaddr->sa_family))
 565                 return -EINVAL;
 566         if (uaddr->sa_family == AF_UNSPEC)
 567                 return sk->sk_prot->disconnect(sk, flags);
 568
 569         if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
 570                 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
 571                 if (err)
 572                         return err;
 573         }
 574
 575         if (!inet_sk(sk)->inet_num && inet_autobind(sk))
 576                 return -EAGAIN;
 577         return sk->sk_prot->connect(sk, uaddr, addr_len);
 578 }
 579 EXPORT_SYMBOL(inet_dgram_connect);
 580
 581 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
 582 {
 583         DEFINE_WAIT_FUNC(wait, woken_wake_function);
 584
 585         add_wait_queue(sk_sleep(sk), &wait);
 586         sk->sk_write_pending += writebias;
 587
 588         /* Basic assumption: if someone sets sk->sk_err, he _must_
 589          * change state of the socket from TCP_SYN_*.
 590          * Connect() does not allow to get error notifications
 591          * without closing the socket.
 592          */
 593         while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 594                 release_sock(sk);
 595                 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
 596                 lock_sock(sk);
 597                 if (signal_pending(current) || !timeo)
 598                         break;
 599         }
 600         remove_wait_queue(sk_sleep(sk), &wait);
 601         sk->sk_write_pending -= writebias;
 602         return timeo;
 603 }
 604
 605 /*
 606  *      Connect to a remote host. There is regrettably still a little
 607  *      TCP 'magic' in here.
 608  */
 609 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 610                           int addr_len, int flags, int is_sendmsg)
 611 {
 612         struct sock *sk = sock->sk;
 613         int err;
 614         long timeo;
 615
 616         /*
 617          * uaddr can be NULL and addr_len can be 0 if:
 618          * sk is a TCP fastopen active socket and
 619          * TCP_FASTOPEN_CONNECT sockopt is set and
 620          * we already have a valid cookie for this socket.
 621          * In this case, user can call write() after connect().
 622          * write() will invoke tcp_sendmsg_fastopen() which calls
 623          * __inet_stream_connect().
 624          */
 625         if (uaddr) {
 626                 if (addr_len < sizeof(uaddr->sa_family))
 627                         return -EINVAL;
 628
 629                 if (uaddr->sa_family == AF_UNSPEC) {
 630                         err = sk->sk_prot->disconnect(sk, flags);
 631                         sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 632                         goto out;
 633                 }
 634         }
 635
 636         switch (sock->state) {
 637         default:
 638                 err = -EINVAL;
 639                 goto out;
 640         case SS_CONNECTED:
 641                 err = -EISCONN;
 642                 goto out;
 643         case SS_CONNECTING:
 644                 if (inet_sk(sk)->defer_connect)
 645                         err = is_sendmsg ? -EINPROGRESS : -EISCONN;
 646                 else
 647                         err = -EALREADY;
 648                 /* Fall out of switch with err, set for this state */
 649                 break;
 650         case SS_UNCONNECTED:
 651                 err = -EISCONN;
 652                 if (sk->sk_state != TCP_CLOSE)
 653                         goto out;
 654
 655                 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
 656                         err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
 657                         if (err)
 658                                 goto out;
 659                 }
 660
 661                 err = sk->sk_prot->connect(sk, uaddr, addr_len);
 662                 if (err < 0)
 663                         goto out;
 664
 665                 sock->state = SS_CONNECTING;
 666
 667                 if (!err && inet_sk(sk)->defer_connect)
 668                         goto out;
 669
 670                 /* Just entered SS_CONNECTING state; the only
 671                  * difference is that return value in non-blocking
 672                  * case is EINPROGRESS, rather than EALREADY.
 673                  */
 674                 err = -EINPROGRESS;
 675                 break;
 676         }
 677
 678         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
 679
 680         if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 681                 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
 682                                 tcp_sk(sk)->fastopen_req &&
 683                                 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
 684
 685                 /* Error code is set above */
 686                 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
 687                         goto out;
 688
 689                 err = sock_intr_errno(timeo);
 690                 if (signal_pending(current))
 691                         goto out;
 692         }
 693
 694         /* Connection was closed by RST, timeout, ICMP error
 695          * or another process disconnected us.
 696          */
 697         if (sk->sk_state == TCP_CLOSE)
 698                 goto sock_error;
 699
 700         /* sk->sk_err may be not zero now, if RECVERR was ordered by user
 701          * and error was received after socket entered established state.
 702          * Hence, it is handled normally after connect() return successfully.
 703          */
 704
 705         sock->state = SS_CONNECTED;
 706         err = 0;
 707 out:
 708         return err;
 709
 710 sock_error:
 711         err = sock_error(sk) ? : -ECONNABORTED;
 712         sock->state = SS_UNCONNECTED;
 713         if (sk->sk_prot->disconnect(sk, flags))
 714                 sock->state = SS_DISCONNECTING;
 715         goto out;
 716 }
 717 EXPORT_SYMBOL(__inet_stream_connect);
 718
 719 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 720                         int addr_len, int flags)
 721 {
 722         int err;
 723
 724         lock_sock(sock->sk);
 725         err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
 726         release_sock(sock->sk);
 727         return err;
 728 }
 729 EXPORT_SYMBOL(inet_stream_connect);
 730
 731 /*
 732  *      Accept a pending connection. The TCP layer now gives BSD semantics.
 733  */
 734
 735 int inet_accept(struct socket *sock, struct socket *newsock, int flags,
 736                 bool kern)
 737 {
 738         struct sock *sk1 = sock->sk;
 739         int err = -EINVAL;
 740         struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err, kern);
 741
 742         if (!sk2)
 743                 goto do_err;
 744
 745         lock_sock(sk2);
 746
 747         sock_rps_record_flow(sk2);
 748         WARN_ON(!((1 << sk2->sk_state) &
 749                   (TCPF_ESTABLISHED | TCPF_SYN_RECV |
 750                   TCPF_CLOSE_WAIT | TCPF_CLOSE)));
 751
 752         sock_graft(sk2, newsock);
 753
 754         newsock->state = SS_CONNECTED;
 755         err = 0;
 756         release_sock(sk2);
 757 do_err:
 758         return err;
 759 }
 760 EXPORT_SYMBOL(inet_accept);
 761
 762 /*
 763  *      This does both peername and sockname.
 764  */
 765 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
 766                  int peer)
 767 {
 768         struct sock *sk         = sock->sk;
 769         struct inet_sock *inet  = inet_sk(sk);
 770         DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
 771
 772         sin->sin_family = AF_INET;
 773         if (peer) {
 774                 if (!inet->inet_dport ||
 775                     (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
 776                      peer == 1))
 777                         return -ENOTCONN;
 778                 sin->sin_port = inet->inet_dport;
 779                 sin->sin_addr.s_addr = inet->inet_daddr;
 780         } else {
 781                 __be32 addr = inet->inet_rcv_saddr;
 782                 if (!addr)
 783                         addr = inet->inet_saddr;
 784                 sin->sin_port = inet->inet_sport;
 785                 sin->sin_addr.s_addr = addr;
 786         }
 787         if (cgroup_bpf_enabled)
 788                 BPF_CGROUP_RUN_SA_PROG_LOCK(sk, (struct sockaddr *)sin,
 789                                             peer ? BPF_CGROUP_INET4_GETPEERNAME :
 790                                                    BPF_CGROUP_INET4_GETSOCKNAME,
 791                                             NULL);
 792         memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 793         return sizeof(*sin);
 794 }
 795 EXPORT_SYMBOL(inet_getname);
 796
 797 int inet_send_prepare(struct sock *sk)
 798 {
 799         sock_rps_record_flow(sk);
 800
 801         /* We may need to bind the socket. */
 802         if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
 803             inet_autobind(sk))
 804                 return -EAGAIN;
 805
 806         return 0;
 807 }
 808 EXPORT_SYMBOL_GPL(inet_send_prepare);
 809
 810 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
 811 {
 812         struct sock *sk = sock->sk;
 813
 814         if (unlikely(inet_send_prepare(sk)))
 815                 return -EAGAIN;
 816
 817         return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
 818                                sk, msg, size);
 819 }
 820 EXPORT_SYMBOL(inet_sendmsg);
 821
 822 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
 823                       size_t size, int flags)
 824 {
 825         struct sock *sk = sock->sk;
 826
 827         if (unlikely(inet_send_prepare(sk)))
 828                 return -EAGAIN;
 829
 830         if (sk->sk_prot->sendpage)
 831                 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
 832         return sock_no_sendpage(sock, page, offset, size, flags);
 833 }
 834 EXPORT_SYMBOL(inet_sendpage);
 835
 836 INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
 837                                           size_t, int, int, int *));
 838 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
 839                  int flags)
 840 {
 841         struct sock *sk = sock->sk;
 842         int addr_len = 0;
 843         int err;
 844
 845         if (likely(!(flags & MSG_ERRQUEUE)))
 846                 sock_rps_record_flow(sk);
 847
 848         err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
 849                               sk, msg, size, flags & MSG_DONTWAIT,
 850                               flags & ~MSG_DONTWAIT, &addr_len);
 851         if (err >= 0)
 852                 msg->msg_namelen = addr_len;
 853         return err;
 854 }
 855 EXPORT_SYMBOL(inet_recvmsg);
 856
 857 int inet_shutdown(struct socket *sock, int how)
 858 {
 859         struct sock *sk = sock->sk;
 860         int err = 0;
 861
 862         /* This should really check to make sure
 863          * the socket is a TCP socket. (WHY AC...)
 864          */
 865         how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
 866                        1->2 bit 2 snds.
 867                        2->3 */
 868         if ((how & ~SHUTDOWN_MASK) || !how)     /* MAXINT->0 */
 869                 return -EINVAL;
 870
 871         lock_sock(sk);
 872         if (sock->state == SS_CONNECTING) {
 873                 if ((1 << sk->sk_state) &
 874                     (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
 875                         sock->state = SS_DISCONNECTING;
 876                 else
 877                         sock->state = SS_CONNECTED;
 878         }
 879
 880         switch (sk->sk_state) {
 881         case TCP_CLOSE:
 882                 err = -ENOTCONN;
 883                 /* Hack to wake up other listeners, who can poll for
 884                    EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
 885                 fallthrough;
 886         default:
 887                 sk->sk_shutdown |= how;
 888                 if (sk->sk_prot->shutdown)
 889                         sk->sk_prot->shutdown(sk, how);
 890                 break;
 891
 892         /* Remaining two branches are temporary solution for missing
 893          * close() in multithreaded environment. It is _not_ a good idea,
 894          * but we have no choice until close() is repaired at VFS level.
 895          */
 896         case TCP_LISTEN:
 897                 if (!(how & RCV_SHUTDOWN))
 898                         break;
 899                 fallthrough;
 900         case TCP_SYN_SENT:
 901                 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
 902                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 903                 break;
 904         }
 905
 906         /* Wake up anyone sleeping in poll. */
 907         sk->sk_state_change(sk);
 908         release_sock(sk);
 909         return err;
 910 }
 911 EXPORT_SYMBOL(inet_shutdown);
 912
 913 /*
 914  *      ioctl() calls you can issue on an INET socket. Most of these are
 915  *      device configuration and stuff and very rarely used. Some ioctls
 916  *      pass on to the socket itself.
 917  *
 918  *      NOTE: I like the idea of a module for the config stuff. ie ifconfig
 919  *      loads the devconfigure module does its configuring and unloads it.
 920  *      There's a good 20K of config code hanging around the kernel.
 921  */
 922
 923 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 924 {
 925         struct sock *sk = sock->sk;
 926         int err = 0;
 927         struct net *net = sock_net(sk);
 928         void __user *p = (void __user *)arg;
 929         struct ifreq ifr;
 930         struct rtentry rt;
 931
 932         switch (cmd) {
 933         case SIOCADDRT:
 934         case SIOCDELRT:
 935                 if (copy_from_user(&rt, p, sizeof(struct rtentry)))
 936                         return -EFAULT;
 937                 err = ip_rt_ioctl(net, cmd, &rt);
 938                 break;
 939         case SIOCRTMSG:
 940                 err = -EINVAL;
 941                 break;
 942         case SIOCDARP:
 943         case SIOCGARP:
 944         case SIOCSARP:
 945                 err = arp_ioctl(net, cmd, (void __user *)arg);
 946                 break;
 947         case SIOCGIFADDR:
 948         case SIOCGIFBRDADDR:
 949         case SIOCGIFNETMASK:
 950         case SIOCGIFDSTADDR:
 951         case SIOCGIFPFLAGS:
 952                 if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
 953                         return -EFAULT;
 954                 err = devinet_ioctl(net, cmd, &ifr);
 955                 if (!err && copy_to_user(p, &ifr, sizeof(struct ifreq)))
 956                         err = -EFAULT;
 957                 break;
 958
 959         case SIOCSIFADDR:
 960         case SIOCSIFBRDADDR:
 961         case SIOCSIFNETMASK:
 962         case SIOCSIFDSTADDR:
 963         case SIOCSIFPFLAGS:
 964         case SIOCSIFFLAGS:
 965                 if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
 966                         return -EFAULT;
 967                 err = devinet_ioctl(net, cmd, &ifr);
 968                 break;
 969         default:
 970                 if (sk->sk_prot->ioctl)
 971                         err = sk->sk_prot->ioctl(sk, cmd, arg);
 972                 else
 973                         err = -ENOIOCTLCMD;
 974                 break;
 975         }
 976         return err;
 977 }
 978 EXPORT_SYMBOL(inet_ioctl);
 979
 980 #ifdef CONFIG_COMPAT
 981 static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd,
 982                 struct compat_rtentry __user *ur)
 983 {
 984         compat_uptr_t rtdev;
 985         struct rtentry rt;
 986
 987         if (copy_from_user(&rt.rt_dst, &ur->rt_dst,
 988                         3 * sizeof(struct sockaddr)) ||
 989             get_user(rt.rt_flags, &ur->rt_flags) ||
 990             get_user(rt.rt_metric, &ur->rt_metric) ||
 991             get_user(rt.rt_mtu, &ur->rt_mtu) ||
 992             get_user(rt.rt_window, &ur->rt_window) ||
 993             get_user(rt.rt_irtt, &ur->rt_irtt) ||
 994             get_user(rtdev, &ur->rt_dev))
 995                 return -EFAULT;
 996
 997         rt.rt_dev = compat_ptr(rtdev);
 998         return ip_rt_ioctl(sock_net(sk), cmd, &rt);
 999 }
1000
1001 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1002 {
1003         void __user *argp = compat_ptr(arg);
1004         struct sock *sk = sock->sk;
1005
1006         switch (cmd) {
1007         case SIOCADDRT:
1008         case SIOCDELRT:
1009                 return inet_compat_routing_ioctl(sk, cmd, argp);
1010         default:
1011                 if (!sk->sk_prot->compat_ioctl)
1012                         return -ENOIOCTLCMD;
1013                 return sk->sk_prot->compat_ioctl(sk, cmd, arg);
1014         }
1015 }
1016 #endif /* CONFIG_COMPAT */
1017
1018 const struct proto_ops inet_stream_ops = {
1019         .family            = PF_INET,
1020         .owner             = THIS_MODULE,
1021         .release           = inet_release,
1022         .bind              = inet_bind,
1023         .connect           = inet_stream_connect,
1024         .socketpair        = sock_no_socketpair,
1025         .accept            = inet_accept,
1026         .getname           = inet_getname,
1027         .poll              = tcp_poll,
1028         .ioctl             = inet_ioctl,
1029         .gettstamp         = sock_gettstamp,
1030         .listen            = inet_listen,
1031         .shutdown          = inet_shutdown,
1032         .setsockopt        = sock_common_setsockopt,
1033         .getsockopt        = sock_common_getsockopt,
1034         .sendmsg           = inet_sendmsg,
1035         .recvmsg           = inet_recvmsg,
1036 #ifdef CONFIG_MMU
1037         .mmap              = tcp_mmap,
1038 #endif
1039         .sendpage          = inet_sendpage,
1040         .splice_read       = tcp_splice_read,
1041         .read_sock         = tcp_read_sock,
1042         .sendmsg_locked    = tcp_sendmsg_locked,
1043         .sendpage_locked   = tcp_sendpage_locked,
1044         .peek_len          = tcp_peek_len,
1045 #ifdef CONFIG_COMPAT
1046         .compat_ioctl      = inet_compat_ioctl,
1047 #endif
1048         .set_rcvlowat      = tcp_set_rcvlowat,
1049 };
1050 EXPORT_SYMBOL(inet_stream_ops);
1051
1052 const struct proto_ops inet_dgram_ops = {
1053         .family            = PF_INET,
1054         .owner             = THIS_MODULE,
1055         .release           = inet_release,
1056         .bind              = inet_bind,
1057         .connect           = inet_dgram_connect,
1058         .socketpair        = sock_no_socketpair,
1059         .accept            = sock_no_accept,
1060         .getname           = inet_getname,
1061         .poll              = udp_poll,
1062         .ioctl             = inet_ioctl,
1063         .gettstamp         = sock_gettstamp,
1064         .listen            = sock_no_listen,
1065         .shutdown          = inet_shutdown,
1066         .setsockopt        = sock_common_setsockopt,
1067         .getsockopt        = sock_common_getsockopt,
1068         .sendmsg           = inet_sendmsg,
1069         .recvmsg           = inet_recvmsg,
1070         .mmap              = sock_no_mmap,
1071         .sendpage          = inet_sendpage,
1072         .set_peek_off      = sk_set_peek_off,
1073 #ifdef CONFIG_COMPAT
1074         .compat_ioctl      = inet_compat_ioctl,
1075 #endif
1076 };
1077 EXPORT_SYMBOL(inet_dgram_ops);
1078
1079 /*
1080  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1081  * udp_poll
1082  */
1083 static const struct proto_ops inet_sockraw_ops = {
1084         .family            = PF_INET,
1085         .owner             = THIS_MODULE,
1086         .release           = inet_release,
1087         .bind              = inet_bind,
1088         .connect           = inet_dgram_connect,
1089         .socketpair        = sock_no_socketpair,
1090         .accept            = sock_no_accept,
1091         .getname           = inet_getname,
1092         .poll              = datagram_poll,
1093         .ioctl             = inet_ioctl,
1094         .gettstamp         = sock_gettstamp,
1095         .listen            = sock_no_listen,
1096         .shutdown          = inet_shutdown,
1097         .setsockopt        = sock_common_setsockopt,
1098         .getsockopt        = sock_common_getsockopt,
1099         .sendmsg           = inet_sendmsg,
1100         .recvmsg           = inet_recvmsg,
1101         .mmap              = sock_no_mmap,
1102         .sendpage          = inet_sendpage,
1103 #ifdef CONFIG_COMPAT
1104         .compat_ioctl      = inet_compat_ioctl,
1105 #endif
1106 };
1107
1108 static const struct net_proto_family inet_family_ops = {
1109         .family = PF_INET,
1110         .create = inet_create,
1111         .owner  = THIS_MODULE,
1112 };
1113
1114 /* Upon startup we insert all the elements in inetsw_array[] into
1115  * the linked list inetsw.
1116  */
1117 static struct inet_protosw inetsw_array[] =
1118 {
1119         {
1120                 .type =       SOCK_STREAM,
1121                 .protocol =   IPPROTO_TCP,
1122                 .prot =       &tcp_prot,
1123                 .ops =        &inet_stream_ops,
1124                 .flags =      INET_PROTOSW_PERMANENT |
1125                               INET_PROTOSW_ICSK,
1126         },
1127
1128         {
1129                 .type =       SOCK_DGRAM,
1130                 .protocol =   IPPROTO_UDP,
1131                 .prot =       &udp_prot,
1132                 .ops =        &inet_dgram_ops,
1133                 .flags =      INET_PROTOSW_PERMANENT,
1134        },
1135
1136        {
1137                 .type =       SOCK_DGRAM,
1138                 .protocol =   IPPROTO_ICMP,
1139                 .prot =       &ping_prot,
1140                 .ops =        &inet_sockraw_ops,
1141                 .flags =      INET_PROTOSW_REUSE,
1142        },
1143
1144        {
1145                .type =       SOCK_RAW,
1146                .protocol =   IPPROTO_IP,        /* wild card */
1147                .prot =       &raw_prot,
1148                .ops =        &inet_sockraw_ops,
1149                .flags =      INET_PROTOSW_REUSE,
1150        }
1151 };
1152
1153 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1154
1155 void inet_register_protosw(struct inet_protosw *p)
1156 {
1157         struct list_head *lh;
1158         struct inet_protosw *answer;
1159         int protocol = p->protocol;
1160         struct list_head *last_perm;
1161
1162         spin_lock_bh(&inetsw_lock);
1163
1164         if (p->type >= SOCK_MAX)
1165                 goto out_illegal;
1166
1167         /* If we are trying to override a permanent protocol, bail. */
1168         last_perm = &inetsw[p->type];
1169         list_for_each(lh, &inetsw[p->type]) {
1170                 answer = list_entry(lh, struct inet_protosw, list);
1171                 /* Check only the non-wild match. */
1172                 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1173                         break;
1174                 if (protocol == answer->protocol)
1175                         goto out_permanent;
1176                 last_perm = lh;
1177         }
1178
1179         /* Add the new entry after the last permanent entry if any, so that
1180          * the new entry does not override a permanent entry when matched with
1181          * a wild-card protocol. But it is allowed to override any existing
1182          * non-permanent entry.  This means that when we remove this entry, the
1183          * system automatically returns to the old behavior.
1184          */
1185         list_add_rcu(&p->list, last_perm);
1186 out:
1187         spin_unlock_bh(&inetsw_lock);
1188
1189         return;
1190
1191 out_permanent:
1192         pr_err("Attempt to override permanent protocol %d\n", protocol);
1193         goto out;
1194
1195 out_illegal:
1196         pr_err("Ignoring attempt to register invalid socket type %d\n",
1197                p->type);
1198         goto out;
1199 }
1200 EXPORT_SYMBOL(inet_register_protosw);
1201
1202 void inet_unregister_protosw(struct inet_protosw *p)
1203 {
1204         if (INET_PROTOSW_PERMANENT & p->flags) {
1205                 pr_err("Attempt to unregister permanent protocol %d\n",
1206                        p->protocol);
1207         } else {
1208                 spin_lock_bh(&inetsw_lock);
1209                 list_del_rcu(&p->list);
1210                 spin_unlock_bh(&inetsw_lock);
1211
1212                 synchronize_net();
1213         }
1214 }
1215 EXPORT_SYMBOL(inet_unregister_protosw);
1216
1217 static int inet_sk_reselect_saddr(struct sock *sk)
1218 {
1219         struct inet_sock *inet = inet_sk(sk);
1220         __be32 old_saddr = inet->inet_saddr;
1221         __be32 daddr = inet->inet_daddr;
1222         struct flowi4 *fl4;
1223         struct rtable *rt;
1224         __be32 new_saddr;
1225         struct ip_options_rcu *inet_opt;
1226
1227         inet_opt = rcu_dereference_protected(inet->inet_opt,
1228                                              lockdep_sock_is_held(sk));
1229         if (inet_opt && inet_opt->opt.srr)
1230                 daddr = inet_opt->opt.faddr;
1231
1232         /* Query new route. */
1233         fl4 = &inet->cork.fl.u.ip4;
1234         rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1235                               sk->sk_bound_dev_if, sk->sk_protocol,
1236                               inet->inet_sport, inet->inet_dport, sk);
1237         if (IS_ERR(rt))
1238                 return PTR_ERR(rt);
1239
1240         sk_setup_caps(sk, &rt->dst);
1241
1242         new_saddr = fl4->saddr;
1243
1244         if (new_saddr == old_saddr)
1245                 return 0;
1246
1247         if (sock_net(sk)->ipv4.sysctl_ip_dynaddr > 1) {
1248                 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1249                         __func__, &old_saddr, &new_saddr);
1250         }
1251
1252         inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1253
1254         /*
1255          * XXX The only one ugly spot where we need to
1256          * XXX really change the sockets identity after
1257          * XXX it has entered the hashes. -DaveM
1258          *
1259          * Besides that, it does not check for connection
1260          * uniqueness. Wait for troubles.
1261          */
1262         return __sk_prot_rehash(sk);
1263 }
1264
1265 int inet_sk_rebuild_header(struct sock *sk)
1266 {
1267         struct inet_sock *inet = inet_sk(sk);
1268         struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1269         __be32 daddr;
1270         struct ip_options_rcu *inet_opt;
1271         struct flowi4 *fl4;
1272         int err;
1273
1274         /* Route is OK, nothing to do. */
1275         if (rt)
1276                 return 0;
1277
1278         /* Reroute. */
1279         rcu_read_lock();
1280         inet_opt = rcu_dereference(inet->inet_opt);
1281         daddr = inet->inet_daddr;
1282         if (inet_opt && inet_opt->opt.srr)
1283                 daddr = inet_opt->opt.faddr;
1284         rcu_read_unlock();
1285         fl4 = &inet->cork.fl.u.ip4;
1286         rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1287                                    inet->inet_dport, inet->inet_sport,
1288                                    sk->sk_protocol, RT_CONN_FLAGS(sk),
1289                                    sk->sk_bound_dev_if);
1290         if (!IS_ERR(rt)) {
1291                 err = 0;
1292                 sk_setup_caps(sk, &rt->dst);
1293         } else {
1294                 err = PTR_ERR(rt);
1295
1296                 /* Routing failed... */
1297                 sk->sk_route_caps = 0;
1298                 /*
1299                  * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1300                  * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1301                  */
1302                 if (!sock_net(sk)->ipv4.sysctl_ip_dynaddr ||
1303                     sk->sk_state != TCP_SYN_SENT ||
1304                     (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1305                     (err = inet_sk_reselect_saddr(sk)) != 0)
1306                         sk->sk_err_soft = -err;
1307         }
1308
1309         return err;
1310 }
1311 EXPORT_SYMBOL(inet_sk_rebuild_header);
1312
1313 void inet_sk_set_state(struct sock *sk, int state)
1314 {
1315         trace_inet_sock_set_state(sk, sk->sk_state, state);
1316         sk->sk_state = state;
1317 }
1318 EXPORT_SYMBOL(inet_sk_set_state);
1319
1320 void inet_sk_state_store(struct sock *sk, int newstate)
1321 {
1322         trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1323         smp_store_release(&sk->sk_state, newstate);
1324 }
1325
1326 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1327                                  netdev_features_t features)
1328 {
1329         bool udpfrag = false, fixedid = false, gso_partial, encap;
1330         struct sk_buff *segs = ERR_PTR(-EINVAL);
1331         const struct net_offload *ops;
1332         unsigned int offset = 0;
1333         struct iphdr *iph;
1334         int proto, tot_len;
1335         int nhoff;
1336         int ihl;
1337         int id;
1338
1339         skb_reset_network_header(skb);
1340         nhoff = skb_network_header(skb) - skb_mac_header(skb);
1341         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1342                 goto out;
1343
1344         iph = ip_hdr(skb);
1345         ihl = iph->ihl * 4;
1346         if (ihl < sizeof(*iph))
1347                 goto out;
1348
1349         id = ntohs(iph->id);
1350         proto = iph->protocol;
1351
1352         /* Warning: after this point, iph might be no longer valid */
1353         if (unlikely(!pskb_may_pull(skb, ihl)))
1354                 goto out;
1355         __skb_pull(skb, ihl);
1356
1357         encap = SKB_GSO_CB(skb)->encap_level > 0;
1358         if (encap)
1359                 features &= skb->dev->hw_enc_features;
1360         SKB_GSO_CB(skb)->encap_level += ihl;
1361
1362         skb_reset_transport_header(skb);
1363
1364         segs = ERR_PTR(-EPROTONOSUPPORT);
1365
1366         if (!skb->encapsulation || encap) {
1367                 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1368                 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1369
1370                 /* fixed ID is invalid if DF bit is not set */
1371                 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1372                         goto out;
1373         }
1374
1375         ops = rcu_dereference(inet_offloads[proto]);
1376         if (likely(ops && ops->callbacks.gso_segment))
1377                 segs = ops->callbacks.gso_segment(skb, features);
1378
1379         if (IS_ERR_OR_NULL(segs))
1380                 goto out;
1381
1382         gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1383
1384         skb = segs;
1385         do {
1386                 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1387                 if (udpfrag) {
1388                         iph->frag_off = htons(offset >> 3);
1389                         if (skb->next)
1390                                 iph->frag_off |= htons(IP_MF);
1391                         offset += skb->len - nhoff - ihl;
1392                         tot_len = skb->len - nhoff;
1393                 } else if (skb_is_gso(skb)) {
1394                         if (!fixedid) {
1395                                 iph->id = htons(id);
1396                                 id += skb_shinfo(skb)->gso_segs;
1397                         }
1398
1399                         if (gso_partial)
1400                                 tot_len = skb_shinfo(skb)->gso_size +
1401                                           SKB_GSO_CB(skb)->data_offset +
1402                                           skb->head - (unsigned char *)iph;
1403                         else
1404                                 tot_len = skb->len - nhoff;
1405                 } else {
1406                         if (!fixedid)
1407                                 iph->id = htons(id++);
1408                         tot_len = skb->len - nhoff;
1409                 }
1410                 iph->tot_len = htons(tot_len);
1411                 ip_send_check(iph);
1412                 if (encap)
1413                         skb_reset_inner_headers(skb);
1414                 skb->network_header = (u8 *)iph - skb->head;
1415                 skb_reset_mac_len(skb);
1416         } while ((skb = skb->next));
1417
1418 out:
1419         return segs;
1420 }
1421 EXPORT_SYMBOL(inet_gso_segment);
1422
1423 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1424                                         netdev_features_t features)
1425 {
1426         if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1427                 return ERR_PTR(-EINVAL);
1428
1429         return inet_gso_segment(skb, features);
1430 }
1431
1432 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1433 {
1434         const struct net_offload *ops;
1435         struct sk_buff *pp = NULL;
1436         const struct iphdr *iph;
1437         struct sk_buff *p;
1438         unsigned int hlen;
1439         unsigned int off;
1440         unsigned int id;
1441         int flush = 1;
1442         int proto;
1443
1444         off = skb_gro_offset(skb);
1445         hlen = off + sizeof(*iph);
1446         iph = skb_gro_header_fast(skb, off);
1447         if (skb_gro_header_hard(skb, hlen)) {
1448                 iph = skb_gro_header_slow(skb, hlen, off);
1449                 if (unlikely(!iph))
1450                         goto out;
1451         }
1452
1453         proto = iph->protocol;
1454
1455         rcu_read_lock();
1456         ops = rcu_dereference(inet_offloads[proto]);
1457         if (!ops || !ops->callbacks.gro_receive)
1458                 goto out_unlock;
1459
1460         if (*(u8 *)iph != 0x45)
1461                 goto out_unlock;
1462
1463         if (ip_is_fragment(iph))
1464                 goto out_unlock;
1465
1466         if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1467                 goto out_unlock;
1468
1469         id = ntohl(*(__be32 *)&iph->id);
1470         flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1471         id >>= 16;
1472
1473         list_for_each_entry(p, head, list) {
1474                 struct iphdr *iph2;
1475                 u16 flush_id;
1476
1477                 if (!NAPI_GRO_CB(p)->same_flow)
1478                         continue;
1479
1480                 iph2 = (struct iphdr *)(p->data + off);
1481                 /* The above works because, with the exception of the top
1482                  * (inner most) layer, we only aggregate pkts with the same
1483                  * hdr length so all the hdrs we'll need to verify will start
1484                  * at the same offset.
1485                  */
1486                 if ((iph->protocol ^ iph2->protocol) |
1487                     ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1488                     ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1489                         NAPI_GRO_CB(p)->same_flow = 0;
1490                         continue;
1491                 }
1492
1493                 /* All fields must match except length and checksum. */
1494                 NAPI_GRO_CB(p)->flush |=
1495                         (iph->ttl ^ iph2->ttl) |
1496                         (iph->tos ^ iph2->tos) |
1497                         ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1498
1499                 NAPI_GRO_CB(p)->flush |= flush;
1500
1501                 /* We need to store of the IP ID check to be included later
1502                  * when we can verify that this packet does in fact belong
1503                  * to a given flow.
1504                  */
1505                 flush_id = (u16)(id - ntohs(iph2->id));
1506
1507                 /* This bit of code makes it much easier for us to identify
1508                  * the cases where we are doing atomic vs non-atomic IP ID
1509                  * checks.  Specifically an atomic check can return IP ID
1510                  * values 0 - 0xFFFF, while a non-atomic check can only
1511                  * return 0 or 0xFFFF.
1512                  */
1513                 if (!NAPI_GRO_CB(p)->is_atomic ||
1514                     !(iph->frag_off & htons(IP_DF))) {
1515                         flush_id ^= NAPI_GRO_CB(p)->count;
1516                         flush_id = flush_id ? 0xFFFF : 0;
1517                 }
1518
1519                 /* If the previous IP ID value was based on an atomic
1520                  * datagram we can overwrite the value and ignore it.
1521                  */
1522                 if (NAPI_GRO_CB(skb)->is_atomic)
1523                         NAPI_GRO_CB(p)->flush_id = flush_id;
1524                 else
1525                         NAPI_GRO_CB(p)->flush_id |= flush_id;
1526         }
1527
1528         NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1529         NAPI_GRO_CB(skb)->flush |= flush;
1530         skb_set_network_header(skb, off);
1531         /* The above will be needed by the transport layer if there is one
1532          * immediately following this IP hdr.
1533          */
1534
1535         /* Note : No need to call skb_gro_postpull_rcsum() here,
1536          * as we already checked checksum over ipv4 header was 0
1537          */
1538         skb_gro_pull(skb, sizeof(*iph));
1539         skb_set_transport_header(skb, skb_gro_offset(skb));
1540
1541         pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1542                                        ops->callbacks.gro_receive, head, skb);
1543
1544 out_unlock:
1545         rcu_read_unlock();
1546
1547 out:
1548         skb_gro_flush_final(skb, pp, flush);
1549
1550         return pp;
1551 }
1552 EXPORT_SYMBOL(inet_gro_receive);
1553
1554 static struct sk_buff *ipip_gro_receive(struct list_head *head,
1555                                         struct sk_buff *skb)
1556 {
1557         if (NAPI_GRO_CB(skb)->encap_mark) {
1558                 NAPI_GRO_CB(skb)->flush = 1;
1559                 return NULL;
1560         }
1561
1562         NAPI_GRO_CB(skb)->encap_mark = 1;
1563
1564         return inet_gro_receive(head, skb);
1565 }
1566
1567 #define SECONDS_PER_DAY 86400
1568
1569 /* inet_current_timestamp - Return IP network timestamp
1570  *
1571  * Return milliseconds since midnight in network byte order.
1572  */
1573 __be32 inet_current_timestamp(void)
1574 {
1575         u32 secs;
1576         u32 msecs;
1577         struct timespec64 ts;
1578
1579         ktime_get_real_ts64(&ts);
1580
1581         /* Get secs since midnight. */
1582         (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1583         /* Convert to msecs. */
1584         msecs = secs * MSEC_PER_SEC;
1585         /* Convert nsec to msec. */
1586         msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1587
1588         /* Convert to network byte order. */
1589         return htonl(msecs);
1590 }
1591 EXPORT_SYMBOL(inet_current_timestamp);
1592
1593 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1594 {
1595         if (sk->sk_family == AF_INET)
1596                 return ip_recv_error(sk, msg, len, addr_len);
1597 #if IS_ENABLED(CONFIG_IPV6)
1598         if (sk->sk_family == AF_INET6)
1599                 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1600 #endif
1601         return -EINVAL;
1602 }
1603
1604 int inet_gro_complete(struct sk_buff *skb, int nhoff)
1605 {
1606         __be16 newlen = htons(skb->len - nhoff);
1607         struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1608         const struct net_offload *ops;
1609         int proto = iph->protocol;
1610         int err = -ENOSYS;
1611
1612         if (skb->encapsulation) {
1613                 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1614                 skb_set_inner_network_header(skb, nhoff);
1615         }
1616
1617         csum_replace2(&iph->check, iph->tot_len, newlen);
1618         iph->tot_len = newlen;
1619
1620         rcu_read_lock();
1621         ops = rcu_dereference(inet_offloads[proto]);
1622         if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1623                 goto out_unlock;
1624
1625         /* Only need to add sizeof(*iph) to get to the next hdr below
1626          * because any hdr with option will have been flushed in
1627          * inet_gro_receive().
1628          */
1629         err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1630                               tcp4_gro_complete, udp4_gro_complete,
1631                               skb, nhoff + sizeof(*iph));
1632
1633 out_unlock:
1634         rcu_read_unlock();
1635
1636         return err;
1637 }
1638 EXPORT_SYMBOL(inet_gro_complete);
1639
1640 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1641 {
1642         skb->encapsulation = 1;
1643         skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1644         return inet_gro_complete(skb, nhoff);
1645 }
1646
1647 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1648                          unsigned short type, unsigned char protocol,
1649                          struct net *net)
1650 {
1651         struct socket *sock;
1652         int rc = sock_create_kern(net, family, type, protocol, &sock);
1653
1654         if (rc == 0) {
1655                 *sk = sock->sk;
1656                 (*sk)->sk_allocation = GFP_ATOMIC;
1657                 /*
1658                  * Unhash it so that IP input processing does not even see it,
1659                  * we do not wish this socket to see incoming packets.
1660                  */
1661                 (*sk)->sk_prot->unhash(*sk);
1662         }
1663         return rc;
1664 }
1665 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1666
1667 u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt)
1668 {
1669         return  *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt);
1670 }
1671 EXPORT_SYMBOL_GPL(snmp_get_cpu_field);
1672
1673 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1674 {
1675         unsigned long res = 0;
1676         int i;
1677
1678         for_each_possible_cpu(i)
1679                 res += snmp_get_cpu_field(mib, i, offt);
1680         return res;
1681 }
1682 EXPORT_SYMBOL_GPL(snmp_fold_field);
1683
1684 #if BITS_PER_LONG==32
1685
1686 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1687                          size_t syncp_offset)
1688 {
1689         void *bhptr;
1690         struct u64_stats_sync *syncp;
1691         u64 v;
1692         unsigned int start;
1693
1694         bhptr = per_cpu_ptr(mib, cpu);
1695         syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1696         do {
1697                 start = u64_stats_fetch_begin_irq(syncp);
1698                 v = *(((u64 *)bhptr) + offt);
1699         } while (u64_stats_fetch_retry_irq(syncp, start));
1700
1701         return v;
1702 }
1703 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1704
1705 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1706 {
1707         u64 res = 0;
1708         int cpu;
1709
1710         for_each_possible_cpu(cpu) {
1711                 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1712         }
1713         return res;
1714 }
1715 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1716 #endif
1717
1718 #ifdef CONFIG_IP_MULTICAST
1719 static const struct net_protocol igmp_protocol = {
1720         .handler =      igmp_rcv,
1721         .netns_ok =     1,
1722 };
1723 #endif
1724
1725 /* thinking of making this const? Don't.
1726  * early_demux can change based on sysctl.
1727  */
1728 static struct net_protocol tcp_protocol = {
1729         .early_demux    =       tcp_v4_early_demux,
1730         .early_demux_handler =  tcp_v4_early_demux,
1731         .handler        =       tcp_v4_rcv,
1732         .err_handler    =       tcp_v4_err,
1733         .no_policy      =       1,
1734         .netns_ok       =       1,
1735         .icmp_strict_tag_validation = 1,
1736 };
1737
1738 /* thinking of making this const? Don't.
1739  * early_demux can change based on sysctl.
1740  */
1741 static struct net_protocol udp_protocol = {
1742         .early_demux =  udp_v4_early_demux,
1743         .early_demux_handler =  udp_v4_early_demux,
1744         .handler =      udp_rcv,
1745         .err_handler =  udp_err,
1746         .no_policy =    1,
1747         .netns_ok =     1,
1748 };
1749
1750 static const struct net_protocol icmp_protocol = {
1751         .handler =      icmp_rcv,
1752         .err_handler =  icmp_err,
1753         .no_policy =    1,
1754         .netns_ok =     1,
1755 };
1756
1757 static __net_init int ipv4_mib_init_net(struct net *net)
1758 {
1759         int i;
1760
1761         net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1762         if (!net->mib.tcp_statistics)
1763                 goto err_tcp_mib;
1764         net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1765         if (!net->mib.ip_statistics)
1766                 goto err_ip_mib;
1767
1768         for_each_possible_cpu(i) {
1769                 struct ipstats_mib *af_inet_stats;
1770                 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1771                 u64_stats_init(&af_inet_stats->syncp);
1772         }
1773
1774         net->mib.net_statistics = alloc_percpu(struct linux_mib);
1775         if (!net->mib.net_statistics)
1776                 goto err_net_mib;
1777         net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1778         if (!net->mib.udp_statistics)
1779                 goto err_udp_mib;
1780         net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1781         if (!net->mib.udplite_statistics)
1782                 goto err_udplite_mib;
1783         net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1784         if (!net->mib.icmp_statistics)
1785                 goto err_icmp_mib;
1786         net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1787                                               GFP_KERNEL);
1788         if (!net->mib.icmpmsg_statistics)
1789                 goto err_icmpmsg_mib;
1790
1791         tcp_mib_init(net);
1792         return 0;
1793
1794 err_icmpmsg_mib:
1795         free_percpu(net->mib.icmp_statistics);
1796 err_icmp_mib:
1797         free_percpu(net->mib.udplite_statistics);
1798 err_udplite_mib:
1799         free_percpu(net->mib.udp_statistics);
1800 err_udp_mib:
1801         free_percpu(net->mib.net_statistics);
1802 err_net_mib:
1803         free_percpu(net->mib.ip_statistics);
1804 err_ip_mib:
1805         free_percpu(net->mib.tcp_statistics);
1806 err_tcp_mib:
1807         return -ENOMEM;
1808 }
1809
1810 static __net_exit void ipv4_mib_exit_net(struct net *net)
1811 {
1812         kfree(net->mib.icmpmsg_statistics);
1813         free_percpu(net->mib.icmp_statistics);
1814         free_percpu(net->mib.udplite_statistics);
1815         free_percpu(net->mib.udp_statistics);
1816         free_percpu(net->mib.net_statistics);
1817         free_percpu(net->mib.ip_statistics);
1818         free_percpu(net->mib.tcp_statistics);
1819 #ifdef CONFIG_MPTCP
1820         /* allocated on demand, see mptcp_init_sock() */
1821         free_percpu(net->mib.mptcp_statistics);
1822 #endif
1823 }
1824
1825 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1826         .init = ipv4_mib_init_net,
1827         .exit = ipv4_mib_exit_net,
1828 };
1829
1830 static int __init init_ipv4_mibs(void)
1831 {
1832         return register_pernet_subsys(&ipv4_mib_ops);
1833 }
1834
1835 static __net_init int inet_init_net(struct net *net)
1836 {
1837         /*
1838          * Set defaults for local port range
1839          */
1840         seqlock_init(&net->ipv4.ip_local_ports.lock);
1841         net->ipv4.ip_local_ports.range[0] =  32768;
1842         net->ipv4.ip_local_ports.range[1] =  60999;
1843
1844         seqlock_init(&net->ipv4.ping_group_range.lock);
1845         /*
1846          * Sane defaults - nobody may create ping sockets.
1847          * Boot scripts should set this to distro-specific group.
1848          */
1849         net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1850         net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1851
1852         /* Default values for sysctl-controlled parameters.
1853          * We set them here, in case sysctl is not compiled.
1854          */
1855         net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1856         net->ipv4.sysctl_ip_fwd_update_priority = 1;
1857         net->ipv4.sysctl_ip_dynaddr = 0;
1858         net->ipv4.sysctl_ip_early_demux = 1;
1859         net->ipv4.sysctl_udp_early_demux = 1;
1860         net->ipv4.sysctl_tcp_early_demux = 1;
1861         net->ipv4.sysctl_nexthop_compat_mode = 1;
1862 #ifdef CONFIG_SYSCTL
1863         net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1864 #endif
1865
1866         /* Some igmp sysctl, whose values are always used */
1867         net->ipv4.sysctl_igmp_max_memberships = 20;
1868         net->ipv4.sysctl_igmp_max_msf = 10;
1869         /* IGMP reports for link-local multicast groups are enabled by default */
1870         net->ipv4.sysctl_igmp_llm_reports = 1;
1871         net->ipv4.sysctl_igmp_qrv = 2;
1872
1873         return 0;
1874 }
1875
1876 static __net_initdata struct pernet_operations af_inet_ops = {
1877         .init = inet_init_net,
1878 };
1879
1880 static int __init init_inet_pernet_ops(void)
1881 {
1882         return register_pernet_subsys(&af_inet_ops);
1883 }
1884
1885 static int ipv4_proc_init(void);
1886
1887 /*
1888  *      IP protocol layer initialiser
1889  */
1890
1891 static struct packet_offload ip_packet_offload __read_mostly = {
1892         .type = cpu_to_be16(ETH_P_IP),
1893         .callbacks = {
1894                 .gso_segment = inet_gso_segment,
1895                 .gro_receive = inet_gro_receive,
1896                 .gro_complete = inet_gro_complete,
1897         },
1898 };
1899
1900 static const struct net_offload ipip_offload = {
1901         .callbacks = {
1902                 .gso_segment    = ipip_gso_segment,
1903                 .gro_receive    = ipip_gro_receive,
1904                 .gro_complete   = ipip_gro_complete,
1905         },
1906 };
1907
1908 static int __init ipip_offload_init(void)
1909 {
1910         return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1911 }
1912
1913 static int __init ipv4_offload_init(void)
1914 {
1915         /*
1916          * Add offloads
1917          */
1918         if (udpv4_offload_init() < 0)
1919                 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1920         if (tcpv4_offload_init() < 0)
1921                 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1922         if (ipip_offload_init() < 0)
1923                 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1924
1925         dev_add_offload(&ip_packet_offload);
1926         return 0;
1927 }
1928
1929 fs_initcall(ipv4_offload_init);
1930
1931 static struct packet_type ip_packet_type __read_mostly = {
1932         .type = cpu_to_be16(ETH_P_IP),
1933         .func = ip_rcv,
1934         .list_func = ip_list_rcv,
1935 };
1936
1937 static int __init inet_init(void)
1938 {
1939         struct inet_protosw *q;
1940         struct list_head *r;
1941         int rc;
1942
1943         sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1944
1945         rc = proto_register(&tcp_prot, 1);
1946         if (rc)
1947                 goto out;
1948
1949         rc = proto_register(&udp_prot, 1);
1950         if (rc)
1951                 goto out_unregister_tcp_proto;
1952
1953         rc = proto_register(&raw_prot, 1);
1954         if (rc)
1955                 goto out_unregister_udp_proto;
1956
1957         rc = proto_register(&ping_prot, 1);
1958         if (rc)
1959                 goto out_unregister_raw_proto;
1960
1961         /*
1962          *      Tell SOCKET that we are alive...
1963          */
1964
1965         (void)sock_register(&inet_family_ops);
1966
1967 #ifdef CONFIG_SYSCTL
1968         ip_static_sysctl_init();
1969 #endif
1970
1971         /*
1972          *      Add all the base protocols.
1973          */
1974
1975         if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1976                 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1977         if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1978                 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1979         if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1980                 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1981 #ifdef CONFIG_IP_MULTICAST
1982         if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1983                 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1984 #endif
1985
1986         /* Register the socket-side information for inet_create. */
1987         for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1988                 INIT_LIST_HEAD(r);
1989
1990         for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1991                 inet_register_protosw(q);
1992
1993         /*
1994          *      Set the ARP module up
1995          */
1996
1997         arp_init();
1998
1999         /*
2000          *      Set the IP module up
2001          */
2002
2003         ip_init();
2004
2005         /* Setup TCP slab cache for open requests. */
2006         tcp_init();
2007
2008         /* Setup UDP memory threshold */
2009         udp_init();
2010
2011         /* Add UDP-Lite (RFC 3828) */
2012         udplite4_register();
2013
2014         raw_init();
2015
2016         ping_init();
2017
2018         /*
2019          *      Set the ICMP layer up
2020          */
2021
2022         if (icmp_init() < 0)
2023                 panic("Failed to create the ICMP control socket.\n");
2024
2025         /*
2026          *      Initialise the multicast router
2027          */
2028 #if defined(CONFIG_IP_MROUTE)
2029         if (ip_mr_init())
2030                 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2031 #endif
2032
2033         if (init_inet_pernet_ops())
2034                 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2035         /*
2036          *      Initialise per-cpu ipv4 mibs
2037          */
2038
2039         if (init_ipv4_mibs())
2040                 pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
2041
2042         ipv4_proc_init();
2043
2044         ipfrag_init();
2045
2046         dev_add_pack(&ip_packet_type);
2047
2048         ip_tunnel_core_init();
2049
2050         rc = 0;
2051 out:
2052         return rc;
2053 out_unregister_raw_proto:
2054         proto_unregister(&raw_prot);
2055 out_unregister_udp_proto:
2056         proto_unregister(&udp_prot);
2057 out_unregister_tcp_proto:
2058         proto_unregister(&tcp_prot);
2059         goto out;
2060 }
2061
2062 fs_initcall(inet_init);
2063
2064 /* ------------------------------------------------------------------------ */
2065
2066 #ifdef CONFIG_PROC_FS
2067 static int __init ipv4_proc_init(void)
2068 {
2069         int rc = 0;
2070
2071         if (raw_proc_init())
2072                 goto out_raw;
2073         if (tcp4_proc_init())
2074                 goto out_tcp;
2075         if (udp4_proc_init())
2076                 goto out_udp;
2077         if (ping_proc_init())
2078                 goto out_ping;
2079         if (ip_misc_proc_init())
2080                 goto out_misc;
2081 out:
2082         return rc;
2083 out_misc:
2084         ping_proc_exit();
2085 out_ping:
2086         udp4_proc_exit();
2087 out_udp:
2088         tcp4_proc_exit();
2089 out_tcp:
2090         raw_proc_exit();
2091 out_raw:
2092         rc = -ENOMEM;
2093         goto out;
2094 }
2095
2096 #else /* CONFIG_PROC_FS */
2097 static int __init ipv4_proc_init(void)
2098 {
2099         return 0;
2100 }
2101 #endif /* CONFIG_PROC_FS */