net/ipv4/af_inet.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   4  *              operating system.  INET is implemented using the  BSD Socket
   5  *              interface as the means of communication with the user level.
   6  *
   7  *              PF_INET protocol family socket handler.
   8  *
   9  * Authors:     Ross Biro
  10  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  11  *              Florian La Roche, <flla@stud.uni-sb.de>
  12  *              Alan Cox, <A.Cox@swansea.ac.uk>
  13  *
  14  * Changes (see also sock.c)
  15  *
  16  *              piggy,
  17  *              Karl Knutson    :       Socket protocol table
  18  *              A.N.Kuznetsov   :       Socket death error in accept().
  19  *              John Richardson :       Fix non blocking error in connect()
  20  *                                      so sockets that fail to connect
  21  *                                      don't return -EINPROGRESS.
  22  *              Alan Cox        :       Asynchronous I/O support
  23  *              Alan Cox        :       Keep correct socket pointer on sock
  24  *                                      structures
  25  *                                      when accept() ed
  26  *              Alan Cox        :       Semantics of SO_LINGER aren't state
  27  *                                      moved to close when you look carefully.
  28  *                                      With this fixed and the accept bug fixed
  29  *                                      some RPC stuff seems happier.
  30  *              Niibe Yutaka    :       4.4BSD style write async I/O
  31  *              Alan Cox,
  32  *              Tony Gale       :       Fixed reuse semantics.
  33  *              Alan Cox        :       bind() shouldn't abort existing but dead
  34  *                                      sockets. Stops FTP netin:.. I hope.
  35  *              Alan Cox        :       bind() works correctly for RAW sockets.
  36  *                                      Note that FreeBSD at least was broken
  37  *                                      in this respect so be careful with
  38  *                                      compatibility tests...
  39  *              Alan Cox        :       routing cache support
  40  *              Alan Cox        :       memzero the socket structure for
  41  *                                      compactness.
  42  *              Matt Day        :       nonblock connect error handler
  43  *              Alan Cox        :       Allow large numbers of pending sockets
  44  *                                      (eg for big web sites), but only if
  45  *                                      specifically application requested.
  46  *              Alan Cox        :       New buffering throughout IP. Used
  47  *                                      dumbly.
  48  *              Alan Cox        :       New buffering now used smartly.
  49  *              Alan Cox        :       BSD rather than common sense
  50  *                                      interpretation of listen.
  51  *              Germano Caronni :       Assorted small races.
  52  *              Alan Cox        :       sendmsg/recvmsg basic support.
  53  *              Alan Cox        :       Only sendmsg/recvmsg now supported.
  54  *              Alan Cox        :       Locked down bind (see security list).
  55  *              Alan Cox        :       Loosened bind a little.
  56  *              Mike McLagan    :       ADD/DEL DLCI Ioctls
  57  *      Willy Konynenberg       :       Transparent proxying support.
  58  *              David S. Miller :       New socket lookup architecture.
  59  *                                      Some other random speedups.
  60  *              Cyrus Durgin    :       Cleaned up file for kmod hacks.
  61  *              Andi Kleen      :       Fix inet_stream_connect TCP race.
  62  */
  63
  64 #define pr_fmt(fmt) "IPv4: " fmt
  65
  66 #include <linux/err.h>
  67 #include <linux/errno.h>
  68 #include <linux/types.h>
  69 #include <linux/socket.h>
  70 #include <linux/in.h>
  71 #include <linux/kernel.h>
  72 #include <linux/kmod.h>
  73 #include <linux/sched.h>
  74 #include <linux/timer.h>
  75 #include <linux/string.h>
  76 #include <linux/sockios.h>
  77 #include <linux/net.h>
  78 #include <linux/capability.h>
  79 #include <linux/fcntl.h>
  80 #include <linux/mm.h>
  81 #include <linux/interrupt.h>
  82 #include <linux/stat.h>
  83 #include <linux/init.h>
  84 #include <linux/poll.h>
  85 #include <linux/netfilter_ipv4.h>
  86 #include <linux/random.h>
  87 #include <linux/slab.h>
  88
  89 #include <linux/uaccess.h>
  90
  91 #include <linux/inet.h>
  92 #include <linux/igmp.h>
  93 #include <linux/inetdevice.h>
  94 #include <linux/netdevice.h>
  95 #include <net/checksum.h>
  96 #include <net/ip.h>
  97 #include <net/protocol.h>
  98 #include <net/arp.h>
  99 #include <net/route.h>
 100 #include <net/ip_fib.h>
 101 #include <net/inet_connection_sock.h>
 102 #include <net/gro.h>
 103 #include <net/tcp.h>
 104 #include <net/udp.h>
 105 #include <net/udplite.h>
 106 #include <net/ping.h>
 107 #include <linux/skbuff.h>
 108 #include <net/sock.h>
 109 #include <net/raw.h>
 110 #include <net/icmp.h>
 111 #include <net/inet_common.h>
 112 #include <net/ip_tunnels.h>
 113 #include <net/xfrm.h>
 114 #include <net/net_namespace.h>
 115 #include <net/secure_seq.h>
 116 #ifdef CONFIG_IP_MROUTE
 117 #include <linux/mroute.h>
 118 #endif
 119 #include <net/l3mdev.h>
 120 #include <net/compat.h>
 121
 122 #include <trace/events/sock.h>
 123
 124 /* The inetsw table contains everything that inet_create needs to
 125  * build a new socket.
 126  */
 127 static struct list_head inetsw[SOCK_MAX];
 128 static DEFINE_SPINLOCK(inetsw_lock);
 129
 130 /* New destruction routine */
 131
 132 void inet_sock_destruct(struct sock *sk)
 133 {
 134         struct inet_sock *inet = inet_sk(sk);
 135
 136         __skb_queue_purge(&sk->sk_receive_queue);
 137         __skb_queue_purge(&sk->sk_error_queue);
 138
 139         sk_mem_reclaim_final(sk);
 140
 141         if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
 142                 pr_err("Attempt to release TCP socket in state %d %p\n",
 143                        sk->sk_state, sk);
 144                 return;
 145         }
 146         if (!sock_flag(sk, SOCK_DEAD)) {
 147                 pr_err("Attempt to release alive inet socket %p\n", sk);
 148                 return;
 149         }
 150
 151         WARN_ON_ONCE(atomic_read(&sk->sk_rmem_alloc));
 152         WARN_ON_ONCE(refcount_read(&sk->sk_wmem_alloc));
 153         WARN_ON_ONCE(sk->sk_wmem_queued);
 154         WARN_ON_ONCE(sk_forward_alloc_get(sk));
 155
 156         kfree(rcu_dereference_protected(inet->inet_opt, 1));
 157         dst_release(rcu_dereference_protected(sk->sk_dst_cache, 1));
 158         dst_release(rcu_dereference_protected(sk->sk_rx_dst, 1));
 159         sk_refcnt_debug_dec(sk);
 160 }
 161 EXPORT_SYMBOL(inet_sock_destruct);
 162
 163 /*
 164  *      The routines beyond this point handle the behaviour of an AF_INET
 165  *      socket object. Mostly it punts to the subprotocols of IP to do
 166  *      the work.
 167  */
 168
 169 /*
 170  *      Automatically bind an unbound socket.
 171  */
 172
 173 static int inet_autobind(struct sock *sk)
 174 {
 175         struct inet_sock *inet;
 176         /* We may need to bind the socket. */
 177         lock_sock(sk);
 178         inet = inet_sk(sk);
 179         if (!inet->inet_num) {
 180                 if (sk->sk_prot->get_port(sk, 0)) {
 181                         release_sock(sk);
 182                         return -EAGAIN;
 183                 }
 184                 inet->inet_sport = htons(inet->inet_num);
 185         }
 186         release_sock(sk);
 187         return 0;
 188 }
 189
 190 /*
 191  *      Move a socket into listening state.
 192  */
 193 int inet_listen(struct socket *sock, int backlog)
 194 {
 195         struct sock *sk = sock->sk;
 196         unsigned char old_state;
 197         int err, tcp_fastopen;
 198
 199         lock_sock(sk);
 200
 201         err = -EINVAL;
 202         if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
 203                 goto out;
 204
 205         old_state = sk->sk_state;
 206         if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
 207                 goto out;
 208
 209         WRITE_ONCE(sk->sk_max_ack_backlog, backlog);
 210         /* Really, if the socket is already in listen state
 211          * we can only allow the backlog to be adjusted.
 212          */
 213         if (old_state != TCP_LISTEN) {
 214                 /* Enable TFO w/o requiring TCP_FASTOPEN socket option.
 215                  * Note that only TCP sockets (SOCK_STREAM) will reach here.
 216                  * Also fastopen backlog may already been set via the option
 217                  * because the socket was in TCP_LISTEN state previously but
 218                  * was shutdown() rather than close().
 219                  */
 220                 tcp_fastopen = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fastopen);
 221                 if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
 222                     (tcp_fastopen & TFO_SERVER_ENABLE) &&
 223                     !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
 224                         fastopen_queue_tune(sk, backlog);
 225                         tcp_fastopen_init_key_once(sock_net(sk));
 226                 }
 227
 228                 err = inet_csk_listen_start(sk);
 229                 if (err)
 230                         goto out;
 231                 tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL);
 232         }
 233         err = 0;
 234
 235 out:
 236         release_sock(sk);
 237         return err;
 238 }
 239 EXPORT_SYMBOL(inet_listen);
 240
 241 /*
 242  *      Create an inet socket.
 243  */
 244
 245 static int inet_create(struct net *net, struct socket *sock, int protocol,
 246                        int kern)
 247 {
 248         struct sock *sk;
 249         struct inet_protosw *answer;
 250         struct inet_sock *inet;
 251         struct proto *answer_prot;
 252         unsigned char answer_flags;
 253         int try_loading_module = 0;
 254         int err;
 255
 256         if (protocol < 0 || protocol >= IPPROTO_MAX)
 257                 return -EINVAL;
 258
 259         sock->state = SS_UNCONNECTED;
 260
 261         /* Look for the requested type/protocol pair. */
 262 lookup_protocol:
 263         err = -ESOCKTNOSUPPORT;
 264         rcu_read_lock();
 265         list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
 266
 267                 err = 0;
 268                 /* Check the non-wild match. */
 269                 if (protocol == answer->protocol) {
 270                         if (protocol != IPPROTO_IP)
 271                                 break;
 272                 } else {
 273                         /* Check for the two wild cases. */
 274                         if (IPPROTO_IP == protocol) {
 275                                 protocol = answer->protocol;
 276                                 break;
 277                         }
 278                         if (IPPROTO_IP == answer->protocol)
 279                                 break;
 280                 }
 281                 err = -EPROTONOSUPPORT;
 282         }
 283
 284         if (unlikely(err)) {
 285                 if (try_loading_module < 2) {
 286                         rcu_read_unlock();
 287                         /*
 288                          * Be more specific, e.g. net-pf-2-proto-132-type-1
 289                          * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
 290                          */
 291                         if (++try_loading_module == 1)
 292                                 request_module("net-pf-%d-proto-%d-type-%d",
 293                                                PF_INET, protocol, sock->type);
 294                         /*
 295                          * Fall back to generic, e.g. net-pf-2-proto-132
 296                          * (net-pf-PF_INET-proto-IPPROTO_SCTP)
 297                          */
 298                         else
 299                                 request_module("net-pf-%d-proto-%d",
 300                                                PF_INET, protocol);
 301                         goto lookup_protocol;
 302                 } else
 303                         goto out_rcu_unlock;
 304         }
 305
 306         err = -EPERM;
 307         if (sock->type == SOCK_RAW && !kern &&
 308             !ns_capable(net->user_ns, CAP_NET_RAW))
 309                 goto out_rcu_unlock;
 310
 311         sock->ops = answer->ops;
 312         answer_prot = answer->prot;
 313         answer_flags = answer->flags;
 314         rcu_read_unlock();
 315
 316         WARN_ON(!answer_prot->slab);
 317
 318         err = -ENOMEM;
 319         sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
 320         if (!sk)
 321                 goto out;
 322
 323         err = 0;
 324         if (INET_PROTOSW_REUSE & answer_flags)
 325                 sk->sk_reuse = SK_CAN_REUSE;
 326
 327         inet = inet_sk(sk);
 328         inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
 329
 330         inet->nodefrag = 0;
 331
 332         if (SOCK_RAW == sock->type) {
 333                 inet->inet_num = protocol;
 334                 if (IPPROTO_RAW == protocol)
 335                         inet->hdrincl = 1;
 336         }
 337
 338         if (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc))
 339                 inet->pmtudisc = IP_PMTUDISC_DONT;
 340         else
 341                 inet->pmtudisc = IP_PMTUDISC_WANT;
 342
 343         inet->inet_id = 0;
 344
 345         sock_init_data(sock, sk);
 346
 347         sk->sk_destruct    = inet_sock_destruct;
 348         sk->sk_protocol    = protocol;
 349         sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
 350         sk->sk_txrehash = READ_ONCE(net->core.sysctl_txrehash);
 351
 352         inet->uc_ttl    = -1;
 353         inet->mc_loop   = 1;
 354         inet->mc_ttl    = 1;
 355         inet->mc_all    = 1;
 356         inet->mc_index  = 0;
 357         inet->mc_list   = NULL;
 358         inet->rcv_tos   = 0;
 359
 360         sk_refcnt_debug_inc(sk);
 361
 362         if (inet->inet_num) {
 363                 /* It assumes that any protocol which allows
 364                  * the user to assign a number at socket
 365                  * creation time automatically
 366                  * shares.
 367                  */
 368                 inet->inet_sport = htons(inet->inet_num);
 369                 /* Add to protocol hash chains. */
 370                 err = sk->sk_prot->hash(sk);
 371                 if (err) {
 372                         sk_common_release(sk);
 373                         goto out;
 374                 }
 375         }
 376
 377         if (sk->sk_prot->init) {
 378                 err = sk->sk_prot->init(sk);
 379                 if (err) {
 380                         sk_common_release(sk);
 381                         goto out;
 382                 }
 383         }
 384
 385         if (!kern) {
 386                 err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
 387                 if (err) {
 388                         sk_common_release(sk);
 389                         goto out;
 390                 }
 391         }
 392 out:
 393         return err;
 394 out_rcu_unlock:
 395         rcu_read_unlock();
 396         goto out;
 397 }
 398
 399
 400 /*
 401  *      The peer socket should always be NULL (or else). When we call this
 402  *      function we are destroying the object and from then on nobody
 403  *      should refer to it.
 404  */
 405 int inet_release(struct socket *sock)
 406 {
 407         struct sock *sk = sock->sk;
 408
 409         if (sk) {
 410                 long timeout;
 411
 412                 if (!sk->sk_kern_sock)
 413                         BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk);
 414
 415                 /* Applications forget to leave groups before exiting */
 416                 ip_mc_drop_socket(sk);
 417
 418                 /* If linger is set, we don't return until the close
 419                  * is complete.  Otherwise we return immediately. The
 420                  * actually closing is done the same either way.
 421                  *
 422                  * If the close is due to the process exiting, we never
 423                  * linger..
 424                  */
 425                 timeout = 0;
 426                 if (sock_flag(sk, SOCK_LINGER) &&
 427                     !(current->flags & PF_EXITING))
 428                         timeout = sk->sk_lingertime;
 429                 sk->sk_prot->close(sk, timeout);
 430                 sock->sk = NULL;
 431         }
 432         return 0;
 433 }
 434 EXPORT_SYMBOL(inet_release);
 435
 436 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 437 {
 438         struct sock *sk = sock->sk;
 439         u32 flags = BIND_WITH_LOCK;
 440         int err;
 441
 442         /* If the socket has its own bind function then use it. (RAW) */
 443         if (sk->sk_prot->bind) {
 444                 return sk->sk_prot->bind(sk, uaddr, addr_len);
 445         }
 446         if (addr_len < sizeof(struct sockaddr_in))
 447                 return -EINVAL;
 448
 449         /* BPF prog is run before any checks are done so that if the prog
 450          * changes context in a wrong way it will be caught.
 451          */
 452         err = BPF_CGROUP_RUN_PROG_INET_BIND_LOCK(sk, uaddr,
 453                                                  CGROUP_INET4_BIND, &flags);
 454         if (err)
 455                 return err;
 456
 457         return __inet_bind(sk, uaddr, addr_len, flags);
 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_addr_valid_or_nonlocal(net, inet, addr->sin_addr.s_addr,
 494                                          chk_addr_ret))
 495                 goto out;
 496
 497         snum = ntohs(addr->sin_port);
 498         err = -EACCES;
 499         if (!(flags & BIND_NO_CAP_NET_BIND_SERVICE) &&
 500             snum && inet_port_requires_bind_service(net, snum) &&
 501             !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
 502                 goto out;
 503
 504         /*      We keep a pair of addresses. rcv_saddr is the one
 505          *      used by hash lookups, and saddr is used for transmit.
 506          *
 507          *      In the BSD API these are the same except where it
 508          *      would be illegal to use them (multicast/broadcast) in
 509          *      which case the sending device address is used.
 510          */
 511         if (flags & BIND_WITH_LOCK)
 512                 lock_sock(sk);
 513
 514         /* Check these errors (active socket, double bind). */
 515         err = -EINVAL;
 516         if (sk->sk_state != TCP_CLOSE || inet->inet_num)
 517                 goto out_release_sock;
 518
 519         inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
 520         if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
 521                 inet->inet_saddr = 0;  /* Use device */
 522
 523         /* Make sure we are allowed to bind here. */
 524         if (snum || !(inet->bind_address_no_port ||
 525                       (flags & BIND_FORCE_ADDRESS_NO_PORT))) {
 526                 err = sk->sk_prot->get_port(sk, snum);
 527                 if (err) {
 528                         inet->inet_saddr = inet->inet_rcv_saddr = 0;
 529                         goto out_release_sock;
 530                 }
 531                 if (!(flags & BIND_FROM_BPF)) {
 532                         err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
 533                         if (err) {
 534                                 inet->inet_saddr = inet->inet_rcv_saddr = 0;
 535                                 if (sk->sk_prot->put_port)
 536                                         sk->sk_prot->put_port(sk);
 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         const struct proto *prot;
 563         int err;
 564
 565         if (addr_len < sizeof(uaddr->sa_family))
 566                 return -EINVAL;
 567
 568         /* IPV6_ADDRFORM can change sk->sk_prot under us. */
 569         prot = READ_ONCE(sk->sk_prot);
 570
 571         if (uaddr->sa_family == AF_UNSPEC)
 572                 return prot->disconnect(sk, flags);
 573
 574         if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
 575                 err = prot->pre_connect(sk, uaddr, addr_len);
 576                 if (err)
 577                         return err;
 578         }
 579
 580         if (data_race(!inet_sk(sk)->inet_num) && inet_autobind(sk))
 581                 return -EAGAIN;
 582         return prot->connect(sk, uaddr, addr_len);
 583 }
 584 EXPORT_SYMBOL(inet_dgram_connect);
 585
 586 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
 587 {
 588         DEFINE_WAIT_FUNC(wait, woken_wake_function);
 589
 590         add_wait_queue(sk_sleep(sk), &wait);
 591         sk->sk_write_pending += writebias;
 592         sk->sk_wait_pending++;
 593
 594         /* Basic assumption: if someone sets sk->sk_err, he _must_
 595          * change state of the socket from TCP_SYN_*.
 596          * Connect() does not allow to get error notifications
 597          * without closing the socket.
 598          */
 599         while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 600                 release_sock(sk);
 601                 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
 602                 lock_sock(sk);
 603                 if (signal_pending(current) || !timeo)
 604                         break;
 605         }
 606         remove_wait_queue(sk_sleep(sk), &wait);
 607         sk->sk_write_pending -= writebias;
 608         sk->sk_wait_pending--;
 609         return timeo;
 610 }
 611
 612 /*
 613  *      Connect to a remote host. There is regrettably still a little
 614  *      TCP 'magic' in here.
 615  */
 616 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 617                           int addr_len, int flags, int is_sendmsg)
 618 {
 619         struct sock *sk = sock->sk;
 620         int err;
 621         long timeo;
 622
 623         /*
 624          * uaddr can be NULL and addr_len can be 0 if:
 625          * sk is a TCP fastopen active socket and
 626          * TCP_FASTOPEN_CONNECT sockopt is set and
 627          * we already have a valid cookie for this socket.
 628          * In this case, user can call write() after connect().
 629          * write() will invoke tcp_sendmsg_fastopen() which calls
 630          * __inet_stream_connect().
 631          */
 632         if (uaddr) {
 633                 if (addr_len < sizeof(uaddr->sa_family))
 634                         return -EINVAL;
 635
 636                 if (uaddr->sa_family == AF_UNSPEC) {
 637                         err = sk->sk_prot->disconnect(sk, flags);
 638                         sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 639                         goto out;
 640                 }
 641         }
 642
 643         switch (sock->state) {
 644         default:
 645                 err = -EINVAL;
 646                 goto out;
 647         case SS_CONNECTED:
 648                 err = -EISCONN;
 649                 goto out;
 650         case SS_CONNECTING:
 651                 if (inet_sk(sk)->defer_connect)
 652                         err = is_sendmsg ? -EINPROGRESS : -EISCONN;
 653                 else
 654                         err = -EALREADY;
 655                 /* Fall out of switch with err, set for this state */
 656                 break;
 657         case SS_UNCONNECTED:
 658                 err = -EISCONN;
 659                 if (sk->sk_state != TCP_CLOSE)
 660                         goto out;
 661
 662                 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
 663                         err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
 664                         if (err)
 665                                 goto out;
 666                 }
 667
 668                 err = sk->sk_prot->connect(sk, uaddr, addr_len);
 669                 if (err < 0)
 670                         goto out;
 671
 672                 sock->state = SS_CONNECTING;
 673
 674                 if (!err && inet_sk(sk)->defer_connect)
 675                         goto out;
 676
 677                 /* Just entered SS_CONNECTING state; the only
 678                  * difference is that return value in non-blocking
 679                  * case is EINPROGRESS, rather than EALREADY.
 680                  */
 681                 err = -EINPROGRESS;
 682                 break;
 683         }
 684
 685         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
 686
 687         if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 688                 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
 689                                 tcp_sk(sk)->fastopen_req &&
 690                                 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
 691
 692                 /* Error code is set above */
 693                 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
 694                         goto out;
 695
 696                 err = sock_intr_errno(timeo);
 697                 if (signal_pending(current))
 698                         goto out;
 699         }
 700
 701         /* Connection was closed by RST, timeout, ICMP error
 702          * or another process disconnected us.
 703          */
 704         if (sk->sk_state == TCP_CLOSE)
 705                 goto sock_error;
 706
 707         /* sk->sk_err may be not zero now, if RECVERR was ordered by user
 708          * and error was received after socket entered established state.
 709          * Hence, it is handled normally after connect() return successfully.
 710          */
 711
 712         sock->state = SS_CONNECTED;
 713         err = 0;
 714 out:
 715         return err;
 716
 717 sock_error:
 718         err = sock_error(sk) ? : -ECONNABORTED;
 719         sock->state = SS_UNCONNECTED;
 720         if (sk->sk_prot->disconnect(sk, flags))
 721                 sock->state = SS_DISCONNECTING;
 722         goto out;
 723 }
 724 EXPORT_SYMBOL(__inet_stream_connect);
 725
 726 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 727                         int addr_len, int flags)
 728 {
 729         int err;
 730
 731         lock_sock(sock->sk);
 732         err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
 733         release_sock(sock->sk);
 734         return err;
 735 }
 736 EXPORT_SYMBOL(inet_stream_connect);
 737
 738 /*
 739  *      Accept a pending connection. The TCP layer now gives BSD semantics.
 740  */
 741
 742 int inet_accept(struct socket *sock, struct socket *newsock, int flags,
 743                 bool kern)
 744 {
 745         struct sock *sk1 = sock->sk, *sk2;
 746         int err = -EINVAL;
 747
 748         /* IPV6_ADDRFORM can change sk->sk_prot under us. */
 749         sk2 = READ_ONCE(sk1->sk_prot)->accept(sk1, flags, &err, kern);
 750         if (!sk2)
 751                 goto do_err;
 752
 753         lock_sock(sk2);
 754
 755         sock_rps_record_flow(sk2);
 756         WARN_ON(!((1 << sk2->sk_state) &
 757                   (TCPF_ESTABLISHED | TCPF_SYN_RECV |
 758                   TCPF_CLOSE_WAIT | TCPF_CLOSE)));
 759
 760         if (test_bit(SOCK_SUPPORT_ZC, &sock->flags))
 761                 set_bit(SOCK_SUPPORT_ZC, &newsock->flags);
 762         sock_graft(sk2, newsock);
 763
 764         newsock->state = SS_CONNECTED;
 765         err = 0;
 766         release_sock(sk2);
 767 do_err:
 768         return err;
 769 }
 770 EXPORT_SYMBOL(inet_accept);
 771
 772 /*
 773  *      This does both peername and sockname.
 774  */
 775 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
 776                  int peer)
 777 {
 778         struct sock *sk         = sock->sk;
 779         struct inet_sock *inet  = inet_sk(sk);
 780         DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
 781
 782         sin->sin_family = AF_INET;
 783         lock_sock(sk);
 784         if (peer) {
 785                 if (!inet->inet_dport ||
 786                     (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
 787                      peer == 1)) {
 788                         release_sock(sk);
 789                         return -ENOTCONN;
 790                 }
 791                 sin->sin_port = inet->inet_dport;
 792                 sin->sin_addr.s_addr = inet->inet_daddr;
 793                 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin,
 794                                        CGROUP_INET4_GETPEERNAME);
 795         } else {
 796                 __be32 addr = inet->inet_rcv_saddr;
 797                 if (!addr)
 798                         addr = inet->inet_saddr;
 799                 sin->sin_port = inet->inet_sport;
 800                 sin->sin_addr.s_addr = addr;
 801                 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin,
 802                                        CGROUP_INET4_GETSOCKNAME);
 803         }
 804         release_sock(sk);
 805         memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 806         return sizeof(*sin);
 807 }
 808 EXPORT_SYMBOL(inet_getname);
 809
 810 int inet_send_prepare(struct sock *sk)
 811 {
 812         sock_rps_record_flow(sk);
 813
 814         /* We may need to bind the socket. */
 815         if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind &&
 816             inet_autobind(sk))
 817                 return -EAGAIN;
 818
 819         return 0;
 820 }
 821 EXPORT_SYMBOL_GPL(inet_send_prepare);
 822
 823 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
 824 {
 825         struct sock *sk = sock->sk;
 826
 827         if (unlikely(inet_send_prepare(sk)))
 828                 return -EAGAIN;
 829
 830         return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
 831                                sk, msg, size);
 832 }
 833 EXPORT_SYMBOL(inet_sendmsg);
 834
 835 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
 836                       size_t size, int flags)
 837 {
 838         struct sock *sk = sock->sk;
 839         const struct proto *prot;
 840
 841         if (unlikely(inet_send_prepare(sk)))
 842                 return -EAGAIN;
 843
 844         /* IPV6_ADDRFORM can change sk->sk_prot under us. */
 845         prot = READ_ONCE(sk->sk_prot);
 846         if (prot->sendpage)
 847                 return prot->sendpage(sk, page, offset, size, flags);
 848         return sock_no_sendpage(sock, page, offset, size, flags);
 849 }
 850 EXPORT_SYMBOL(inet_sendpage);
 851
 852 INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
 853                                           size_t, int, int *));
 854 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
 855                  int flags)
 856 {
 857         struct sock *sk = sock->sk;
 858         int addr_len = 0;
 859         int err;
 860
 861         if (likely(!(flags & MSG_ERRQUEUE)))
 862                 sock_rps_record_flow(sk);
 863
 864         err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
 865                               sk, msg, size, flags, &addr_len);
 866         if (err >= 0)
 867                 msg->msg_namelen = addr_len;
 868         return err;
 869 }
 870 EXPORT_SYMBOL(inet_recvmsg);
 871
 872 int inet_shutdown(struct socket *sock, int how)
 873 {
 874         struct sock *sk = sock->sk;
 875         int err = 0;
 876
 877         /* This should really check to make sure
 878          * the socket is a TCP socket. (WHY AC...)
 879          */
 880         how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
 881                        1->2 bit 2 snds.
 882                        2->3 */
 883         if ((how & ~SHUTDOWN_MASK) || !how)     /* MAXINT->0 */
 884                 return -EINVAL;
 885
 886         lock_sock(sk);
 887         if (sock->state == SS_CONNECTING) {
 888                 if ((1 << sk->sk_state) &
 889                     (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
 890                         sock->state = SS_DISCONNECTING;
 891                 else
 892                         sock->state = SS_CONNECTED;
 893         }
 894
 895         switch (sk->sk_state) {
 896         case TCP_CLOSE:
 897                 err = -ENOTCONN;
 898                 /* Hack to wake up other listeners, who can poll for
 899                    EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
 900                 fallthrough;
 901         default:
 902                 WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | how);
 903                 if (sk->sk_prot->shutdown)
 904                         sk->sk_prot->shutdown(sk, how);
 905                 break;
 906
 907         /* Remaining two branches are temporary solution for missing
 908          * close() in multithreaded environment. It is _not_ a good idea,
 909          * but we have no choice until close() is repaired at VFS level.
 910          */
 911         case TCP_LISTEN:
 912                 if (!(how & RCV_SHUTDOWN))
 913                         break;
 914                 fallthrough;
 915         case TCP_SYN_SENT:
 916                 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
 917                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 918                 break;
 919         }
 920
 921         /* Wake up anyone sleeping in poll. */
 922         sk->sk_state_change(sk);
 923         release_sock(sk);
 924         return err;
 925 }
 926 EXPORT_SYMBOL(inet_shutdown);
 927
 928 /*
 929  *      ioctl() calls you can issue on an INET socket. Most of these are
 930  *      device configuration and stuff and very rarely used. Some ioctls
 931  *      pass on to the socket itself.
 932  *
 933  *      NOTE: I like the idea of a module for the config stuff. ie ifconfig
 934  *      loads the devconfigure module does its configuring and unloads it.
 935  *      There's a good 20K of config code hanging around the kernel.
 936  */
 937
 938 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 939 {
 940         struct sock *sk = sock->sk;
 941         int err = 0;
 942         struct net *net = sock_net(sk);
 943         void __user *p = (void __user *)arg;
 944         struct ifreq ifr;
 945         struct rtentry rt;
 946
 947         switch (cmd) {
 948         case SIOCADDRT:
 949         case SIOCDELRT:
 950                 if (copy_from_user(&rt, p, sizeof(struct rtentry)))
 951                         return -EFAULT;
 952                 err = ip_rt_ioctl(net, cmd, &rt);
 953                 break;
 954         case SIOCRTMSG:
 955                 err = -EINVAL;
 956                 break;
 957         case SIOCDARP:
 958         case SIOCGARP:
 959         case SIOCSARP:
 960                 err = arp_ioctl(net, cmd, (void __user *)arg);
 961                 break;
 962         case SIOCGIFADDR:
 963         case SIOCGIFBRDADDR:
 964         case SIOCGIFNETMASK:
 965         case SIOCGIFDSTADDR:
 966         case SIOCGIFPFLAGS:
 967                 if (get_user_ifreq(&ifr, NULL, p))
 968                         return -EFAULT;
 969                 err = devinet_ioctl(net, cmd, &ifr);
 970                 if (!err && put_user_ifreq(&ifr, p))
 971                         err = -EFAULT;
 972                 break;
 973
 974         case SIOCSIFADDR:
 975         case SIOCSIFBRDADDR:
 976         case SIOCSIFNETMASK:
 977         case SIOCSIFDSTADDR:
 978         case SIOCSIFPFLAGS:
 979         case SIOCSIFFLAGS:
 980                 if (get_user_ifreq(&ifr, NULL, p))
 981                         return -EFAULT;
 982                 err = devinet_ioctl(net, cmd, &ifr);
 983                 break;
 984         default:
 985                 if (sk->sk_prot->ioctl)
 986                         err = sk->sk_prot->ioctl(sk, cmd, arg);
 987                 else
 988                         err = -ENOIOCTLCMD;
 989                 break;
 990         }
 991         return err;
 992 }
 993 EXPORT_SYMBOL(inet_ioctl);
 994
 995 #ifdef CONFIG_COMPAT
 996 static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd,
 997                 struct compat_rtentry __user *ur)
 998 {
 999         compat_uptr_t rtdev;
1000         struct rtentry rt;
1001
1002         if (copy_from_user(&rt.rt_dst, &ur->rt_dst,
1003                         3 * sizeof(struct sockaddr)) ||
1004             get_user(rt.rt_flags, &ur->rt_flags) ||
1005             get_user(rt.rt_metric, &ur->rt_metric) ||
1006             get_user(rt.rt_mtu, &ur->rt_mtu) ||
1007             get_user(rt.rt_window, &ur->rt_window) ||
1008             get_user(rt.rt_irtt, &ur->rt_irtt) ||
1009             get_user(rtdev, &ur->rt_dev))
1010                 return -EFAULT;
1011
1012         rt.rt_dev = compat_ptr(rtdev);
1013         return ip_rt_ioctl(sock_net(sk), cmd, &rt);
1014 }
1015
1016 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1017 {
1018         void __user *argp = compat_ptr(arg);
1019         struct sock *sk = sock->sk;
1020
1021         switch (cmd) {
1022         case SIOCADDRT:
1023         case SIOCDELRT:
1024                 return inet_compat_routing_ioctl(sk, cmd, argp);
1025         default:
1026                 if (!sk->sk_prot->compat_ioctl)
1027                         return -ENOIOCTLCMD;
1028                 return sk->sk_prot->compat_ioctl(sk, cmd, arg);
1029         }
1030 }
1031 #endif /* CONFIG_COMPAT */
1032
1033 const struct proto_ops inet_stream_ops = {
1034         .family            = PF_INET,
1035         .owner             = THIS_MODULE,
1036         .release           = inet_release,
1037         .bind              = inet_bind,
1038         .connect           = inet_stream_connect,
1039         .socketpair        = sock_no_socketpair,
1040         .accept            = inet_accept,
1041         .getname           = inet_getname,
1042         .poll              = tcp_poll,
1043         .ioctl             = inet_ioctl,
1044         .gettstamp         = sock_gettstamp,
1045         .listen            = inet_listen,
1046         .shutdown          = inet_shutdown,
1047         .setsockopt        = sock_common_setsockopt,
1048         .getsockopt        = sock_common_getsockopt,
1049         .sendmsg           = inet_sendmsg,
1050         .recvmsg           = inet_recvmsg,
1051 #ifdef CONFIG_MMU
1052         .mmap              = tcp_mmap,
1053 #endif
1054         .sendpage          = inet_sendpage,
1055         .splice_read       = tcp_splice_read,
1056         .read_sock         = tcp_read_sock,
1057         .read_skb          = tcp_read_skb,
1058         .sendmsg_locked    = tcp_sendmsg_locked,
1059         .sendpage_locked   = tcp_sendpage_locked,
1060         .peek_len          = tcp_peek_len,
1061 #ifdef CONFIG_COMPAT
1062         .compat_ioctl      = inet_compat_ioctl,
1063 #endif
1064         .set_rcvlowat      = tcp_set_rcvlowat,
1065 };
1066 EXPORT_SYMBOL(inet_stream_ops);
1067
1068 const struct proto_ops inet_dgram_ops = {
1069         .family            = PF_INET,
1070         .owner             = THIS_MODULE,
1071         .release           = inet_release,
1072         .bind              = inet_bind,
1073         .connect           = inet_dgram_connect,
1074         .socketpair        = sock_no_socketpair,
1075         .accept            = sock_no_accept,
1076         .getname           = inet_getname,
1077         .poll              = udp_poll,
1078         .ioctl             = inet_ioctl,
1079         .gettstamp         = sock_gettstamp,
1080         .listen            = sock_no_listen,
1081         .shutdown          = inet_shutdown,
1082         .setsockopt        = sock_common_setsockopt,
1083         .getsockopt        = sock_common_getsockopt,
1084         .sendmsg           = inet_sendmsg,
1085         .read_skb          = udp_read_skb,
1086         .recvmsg           = inet_recvmsg,
1087         .mmap              = sock_no_mmap,
1088         .sendpage          = inet_sendpage,
1089         .set_peek_off      = sk_set_peek_off,
1090 #ifdef CONFIG_COMPAT
1091         .compat_ioctl      = inet_compat_ioctl,
1092 #endif
1093 };
1094 EXPORT_SYMBOL(inet_dgram_ops);
1095
1096 /*
1097  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1098  * udp_poll
1099  */
1100 static const struct proto_ops inet_sockraw_ops = {
1101         .family            = PF_INET,
1102         .owner             = THIS_MODULE,
1103         .release           = inet_release,
1104         .bind              = inet_bind,
1105         .connect           = inet_dgram_connect,
1106         .socketpair        = sock_no_socketpair,
1107         .accept            = sock_no_accept,
1108         .getname           = inet_getname,
1109         .poll              = datagram_poll,
1110         .ioctl             = inet_ioctl,
1111         .gettstamp         = sock_gettstamp,
1112         .listen            = sock_no_listen,
1113         .shutdown          = inet_shutdown,
1114         .setsockopt        = sock_common_setsockopt,
1115         .getsockopt        = sock_common_getsockopt,
1116         .sendmsg           = inet_sendmsg,
1117         .recvmsg           = inet_recvmsg,
1118         .mmap              = sock_no_mmap,
1119         .sendpage          = inet_sendpage,
1120 #ifdef CONFIG_COMPAT
1121         .compat_ioctl      = inet_compat_ioctl,
1122 #endif
1123 };
1124
1125 static const struct net_proto_family inet_family_ops = {
1126         .family = PF_INET,
1127         .create = inet_create,
1128         .owner  = THIS_MODULE,
1129 };
1130
1131 /* Upon startup we insert all the elements in inetsw_array[] into
1132  * the linked list inetsw.
1133  */
1134 static struct inet_protosw inetsw_array[] =
1135 {
1136         {
1137                 .type =       SOCK_STREAM,
1138                 .protocol =   IPPROTO_TCP,
1139                 .prot =       &tcp_prot,
1140                 .ops =        &inet_stream_ops,
1141                 .flags =      INET_PROTOSW_PERMANENT |
1142                               INET_PROTOSW_ICSK,
1143         },
1144
1145         {
1146                 .type =       SOCK_DGRAM,
1147                 .protocol =   IPPROTO_UDP,
1148                 .prot =       &udp_prot,
1149                 .ops =        &inet_dgram_ops,
1150                 .flags =      INET_PROTOSW_PERMANENT,
1151        },
1152
1153        {
1154                 .type =       SOCK_DGRAM,
1155                 .protocol =   IPPROTO_ICMP,
1156                 .prot =       &ping_prot,
1157                 .ops =        &inet_sockraw_ops,
1158                 .flags =      INET_PROTOSW_REUSE,
1159        },
1160
1161        {
1162                .type =       SOCK_RAW,
1163                .protocol =   IPPROTO_IP,        /* wild card */
1164                .prot =       &raw_prot,
1165                .ops =        &inet_sockraw_ops,
1166                .flags =      INET_PROTOSW_REUSE,
1167        }
1168 };
1169
1170 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1171
1172 void inet_register_protosw(struct inet_protosw *p)
1173 {
1174         struct list_head *lh;
1175         struct inet_protosw *answer;
1176         int protocol = p->protocol;
1177         struct list_head *last_perm;
1178
1179         spin_lock_bh(&inetsw_lock);
1180
1181         if (p->type >= SOCK_MAX)
1182                 goto out_illegal;
1183
1184         /* If we are trying to override a permanent protocol, bail. */
1185         last_perm = &inetsw[p->type];
1186         list_for_each(lh, &inetsw[p->type]) {
1187                 answer = list_entry(lh, struct inet_protosw, list);
1188                 /* Check only the non-wild match. */
1189                 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1190                         break;
1191                 if (protocol == answer->protocol)
1192                         goto out_permanent;
1193                 last_perm = lh;
1194         }
1195
1196         /* Add the new entry after the last permanent entry if any, so that
1197          * the new entry does not override a permanent entry when matched with
1198          * a wild-card protocol. But it is allowed to override any existing
1199          * non-permanent entry.  This means that when we remove this entry, the
1200          * system automatically returns to the old behavior.
1201          */
1202         list_add_rcu(&p->list, last_perm);
1203 out:
1204         spin_unlock_bh(&inetsw_lock);
1205
1206         return;
1207
1208 out_permanent:
1209         pr_err("Attempt to override permanent protocol %d\n", protocol);
1210         goto out;
1211
1212 out_illegal:
1213         pr_err("Ignoring attempt to register invalid socket type %d\n",
1214                p->type);
1215         goto out;
1216 }
1217 EXPORT_SYMBOL(inet_register_protosw);
1218
1219 void inet_unregister_protosw(struct inet_protosw *p)
1220 {
1221         if (INET_PROTOSW_PERMANENT & p->flags) {
1222                 pr_err("Attempt to unregister permanent protocol %d\n",
1223                        p->protocol);
1224         } else {
1225                 spin_lock_bh(&inetsw_lock);
1226                 list_del_rcu(&p->list);
1227                 spin_unlock_bh(&inetsw_lock);
1228
1229                 synchronize_net();
1230         }
1231 }
1232 EXPORT_SYMBOL(inet_unregister_protosw);
1233
1234 static int inet_sk_reselect_saddr(struct sock *sk)
1235 {
1236         struct inet_sock *inet = inet_sk(sk);
1237         __be32 old_saddr = inet->inet_saddr;
1238         __be32 daddr = inet->inet_daddr;
1239         struct flowi4 *fl4;
1240         struct rtable *rt;
1241         __be32 new_saddr;
1242         struct ip_options_rcu *inet_opt;
1243         int err;
1244
1245         inet_opt = rcu_dereference_protected(inet->inet_opt,
1246                                              lockdep_sock_is_held(sk));
1247         if (inet_opt && inet_opt->opt.srr)
1248                 daddr = inet_opt->opt.faddr;
1249
1250         /* Query new route. */
1251         fl4 = &inet->cork.fl.u.ip4;
1252         rt = ip_route_connect(fl4, daddr, 0, sk->sk_bound_dev_if,
1253                               sk->sk_protocol, inet->inet_sport,
1254                               inet->inet_dport, sk);
1255         if (IS_ERR(rt))
1256                 return PTR_ERR(rt);
1257
1258         new_saddr = fl4->saddr;
1259
1260         if (new_saddr == old_saddr) {
1261                 sk_setup_caps(sk, &rt->dst);
1262                 return 0;
1263         }
1264
1265         err = inet_bhash2_update_saddr(sk, &new_saddr, AF_INET);
1266         if (err) {
1267                 ip_rt_put(rt);
1268                 return err;
1269         }
1270
1271         sk_setup_caps(sk, &rt->dst);
1272
1273         if (READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) > 1) {
1274                 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1275                         __func__, &old_saddr, &new_saddr);
1276         }
1277
1278         /*
1279          * XXX The only one ugly spot where we need to
1280          * XXX really change the sockets identity after
1281          * XXX it has entered the hashes. -DaveM
1282          *
1283          * Besides that, it does not check for connection
1284          * uniqueness. Wait for troubles.
1285          */
1286         return __sk_prot_rehash(sk);
1287 }
1288
1289 int inet_sk_rebuild_header(struct sock *sk)
1290 {
1291         struct inet_sock *inet = inet_sk(sk);
1292         struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1293         __be32 daddr;
1294         struct ip_options_rcu *inet_opt;
1295         struct flowi4 *fl4;
1296         int err;
1297
1298         /* Route is OK, nothing to do. */
1299         if (rt)
1300                 return 0;
1301
1302         /* Reroute. */
1303         rcu_read_lock();
1304         inet_opt = rcu_dereference(inet->inet_opt);
1305         daddr = inet->inet_daddr;
1306         if (inet_opt && inet_opt->opt.srr)
1307                 daddr = inet_opt->opt.faddr;
1308         rcu_read_unlock();
1309         fl4 = &inet->cork.fl.u.ip4;
1310         rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1311                                    inet->inet_dport, inet->inet_sport,
1312                                    sk->sk_protocol, RT_CONN_FLAGS(sk),
1313                                    sk->sk_bound_dev_if);
1314         if (!IS_ERR(rt)) {
1315                 err = 0;
1316                 sk_setup_caps(sk, &rt->dst);
1317         } else {
1318                 err = PTR_ERR(rt);
1319
1320                 /* Routing failed... */
1321                 sk->sk_route_caps = 0;
1322                 /*
1323                  * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1324                  * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1325                  */
1326                 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) ||
1327                     sk->sk_state != TCP_SYN_SENT ||
1328                     (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1329                     (err = inet_sk_reselect_saddr(sk)) != 0)
1330                         sk->sk_err_soft = -err;
1331         }
1332
1333         return err;
1334 }
1335 EXPORT_SYMBOL(inet_sk_rebuild_header);
1336
1337 void inet_sk_set_state(struct sock *sk, int state)
1338 {
1339         trace_inet_sock_set_state(sk, sk->sk_state, state);
1340         sk->sk_state = state;
1341 }
1342 EXPORT_SYMBOL(inet_sk_set_state);
1343
1344 void inet_sk_state_store(struct sock *sk, int newstate)
1345 {
1346         trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1347         smp_store_release(&sk->sk_state, newstate);
1348 }
1349
1350 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1351                                  netdev_features_t features)
1352 {
1353         bool udpfrag = false, fixedid = false, gso_partial, encap;
1354         struct sk_buff *segs = ERR_PTR(-EINVAL);
1355         const struct net_offload *ops;
1356         unsigned int offset = 0;
1357         struct iphdr *iph;
1358         int proto, tot_len;
1359         int nhoff;
1360         int ihl;
1361         int id;
1362
1363         skb_reset_network_header(skb);
1364         nhoff = skb_network_header(skb) - skb_mac_header(skb);
1365         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1366                 goto out;
1367
1368         iph = ip_hdr(skb);
1369         ihl = iph->ihl * 4;
1370         if (ihl < sizeof(*iph))
1371                 goto out;
1372
1373         id = ntohs(iph->id);
1374         proto = iph->protocol;
1375
1376         /* Warning: after this point, iph might be no longer valid */
1377         if (unlikely(!pskb_may_pull(skb, ihl)))
1378                 goto out;
1379         __skb_pull(skb, ihl);
1380
1381         encap = SKB_GSO_CB(skb)->encap_level > 0;
1382         if (encap)
1383                 features &= skb->dev->hw_enc_features;
1384         SKB_GSO_CB(skb)->encap_level += ihl;
1385
1386         skb_reset_transport_header(skb);
1387
1388         segs = ERR_PTR(-EPROTONOSUPPORT);
1389
1390         if (!skb->encapsulation || encap) {
1391                 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1392                 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1393
1394                 /* fixed ID is invalid if DF bit is not set */
1395                 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1396                         goto out;
1397         }
1398
1399         ops = rcu_dereference(inet_offloads[proto]);
1400         if (likely(ops && ops->callbacks.gso_segment)) {
1401                 segs = ops->callbacks.gso_segment(skb, features);
1402                 if (!segs)
1403                         skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
1404         }
1405
1406         if (IS_ERR_OR_NULL(segs))
1407                 goto out;
1408
1409         gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1410
1411         skb = segs;
1412         do {
1413                 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1414                 if (udpfrag) {
1415                         iph->frag_off = htons(offset >> 3);
1416                         if (skb->next)
1417                                 iph->frag_off |= htons(IP_MF);
1418                         offset += skb->len - nhoff - ihl;
1419                         tot_len = skb->len - nhoff;
1420                 } else if (skb_is_gso(skb)) {
1421                         if (!fixedid) {
1422                                 iph->id = htons(id);
1423                                 id += skb_shinfo(skb)->gso_segs;
1424                         }
1425
1426                         if (gso_partial)
1427                                 tot_len = skb_shinfo(skb)->gso_size +
1428                                           SKB_GSO_CB(skb)->data_offset +
1429                                           skb->head - (unsigned char *)iph;
1430                         else
1431                                 tot_len = skb->len - nhoff;
1432                 } else {
1433                         if (!fixedid)
1434                                 iph->id = htons(id++);
1435                         tot_len = skb->len - nhoff;
1436                 }
1437                 iph->tot_len = htons(tot_len);
1438                 ip_send_check(iph);
1439                 if (encap)
1440                         skb_reset_inner_headers(skb);
1441                 skb->network_header = (u8 *)iph - skb->head;
1442                 skb_reset_mac_len(skb);
1443         } while ((skb = skb->next));
1444
1445 out:
1446         return segs;
1447 }
1448
1449 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1450                                         netdev_features_t features)
1451 {
1452         if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1453                 return ERR_PTR(-EINVAL);
1454
1455         return inet_gso_segment(skb, features);
1456 }
1457
1458 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1459 {
1460         const struct net_offload *ops;
1461         struct sk_buff *pp = NULL;
1462         const struct iphdr *iph;
1463         struct sk_buff *p;
1464         unsigned int hlen;
1465         unsigned int off;
1466         unsigned int id;
1467         int flush = 1;
1468         int proto;
1469
1470         off = skb_gro_offset(skb);
1471         hlen = off + sizeof(*iph);
1472         iph = skb_gro_header(skb, hlen, off);
1473         if (unlikely(!iph))
1474                 goto out;
1475
1476         proto = iph->protocol;
1477
1478         ops = rcu_dereference(inet_offloads[proto]);
1479         if (!ops || !ops->callbacks.gro_receive)
1480                 goto out;
1481
1482         if (*(u8 *)iph != 0x45)
1483                 goto out;
1484
1485         if (ip_is_fragment(iph))
1486                 goto out;
1487
1488         if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1489                 goto out;
1490
1491         id = ntohl(*(__be32 *)&iph->id);
1492         flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1493         id >>= 16;
1494
1495         list_for_each_entry(p, head, list) {
1496                 struct iphdr *iph2;
1497                 u16 flush_id;
1498
1499                 if (!NAPI_GRO_CB(p)->same_flow)
1500                         continue;
1501
1502                 iph2 = (struct iphdr *)(p->data + off);
1503                 /* The above works because, with the exception of the top
1504                  * (inner most) layer, we only aggregate pkts with the same
1505                  * hdr length so all the hdrs we'll need to verify will start
1506                  * at the same offset.
1507                  */
1508                 if ((iph->protocol ^ iph2->protocol) |
1509                     ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1510                     ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1511                         NAPI_GRO_CB(p)->same_flow = 0;
1512                         continue;
1513                 }
1514
1515                 /* All fields must match except length and checksum. */
1516                 NAPI_GRO_CB(p)->flush |=
1517                         (iph->ttl ^ iph2->ttl) |
1518                         (iph->tos ^ iph2->tos) |
1519                         ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1520
1521                 NAPI_GRO_CB(p)->flush |= flush;
1522
1523                 /* We need to store of the IP ID check to be included later
1524                  * when we can verify that this packet does in fact belong
1525                  * to a given flow.
1526                  */
1527                 flush_id = (u16)(id - ntohs(iph2->id));
1528
1529                 /* This bit of code makes it much easier for us to identify
1530                  * the cases where we are doing atomic vs non-atomic IP ID
1531                  * checks.  Specifically an atomic check can return IP ID
1532                  * values 0 - 0xFFFF, while a non-atomic check can only
1533                  * return 0 or 0xFFFF.
1534                  */
1535                 if (!NAPI_GRO_CB(p)->is_atomic ||
1536                     !(iph->frag_off & htons(IP_DF))) {
1537                         flush_id ^= NAPI_GRO_CB(p)->count;
1538                         flush_id = flush_id ? 0xFFFF : 0;
1539                 }
1540
1541                 /* If the previous IP ID value was based on an atomic
1542                  * datagram we can overwrite the value and ignore it.
1543                  */
1544                 if (NAPI_GRO_CB(skb)->is_atomic)
1545                         NAPI_GRO_CB(p)->flush_id = flush_id;
1546                 else
1547                         NAPI_GRO_CB(p)->flush_id |= flush_id;
1548         }
1549
1550         NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1551         NAPI_GRO_CB(skb)->flush |= flush;
1552         skb_set_network_header(skb, off);
1553         /* The above will be needed by the transport layer if there is one
1554          * immediately following this IP hdr.
1555          */
1556
1557         /* Note : No need to call skb_gro_postpull_rcsum() here,
1558          * as we already checked checksum over ipv4 header was 0
1559          */
1560         skb_gro_pull(skb, sizeof(*iph));
1561         skb_set_transport_header(skb, skb_gro_offset(skb));
1562
1563         pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1564                                        ops->callbacks.gro_receive, head, skb);
1565
1566 out:
1567         skb_gro_flush_final(skb, pp, flush);
1568
1569         return pp;
1570 }
1571
1572 static struct sk_buff *ipip_gro_receive(struct list_head *head,
1573                                         struct sk_buff *skb)
1574 {
1575         if (NAPI_GRO_CB(skb)->encap_mark) {
1576                 NAPI_GRO_CB(skb)->flush = 1;
1577                 return NULL;
1578         }
1579
1580         NAPI_GRO_CB(skb)->encap_mark = 1;
1581
1582         return inet_gro_receive(head, skb);
1583 }
1584
1585 #define SECONDS_PER_DAY 86400
1586
1587 /* inet_current_timestamp - Return IP network timestamp
1588  *
1589  * Return milliseconds since midnight in network byte order.
1590  */
1591 __be32 inet_current_timestamp(void)
1592 {
1593         u32 secs;
1594         u32 msecs;
1595         struct timespec64 ts;
1596
1597         ktime_get_real_ts64(&ts);
1598
1599         /* Get secs since midnight. */
1600         (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1601         /* Convert to msecs. */
1602         msecs = secs * MSEC_PER_SEC;
1603         /* Convert nsec to msec. */
1604         msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1605
1606         /* Convert to network byte order. */
1607         return htonl(msecs);
1608 }
1609 EXPORT_SYMBOL(inet_current_timestamp);
1610
1611 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1612 {
1613         if (sk->sk_family == AF_INET)
1614                 return ip_recv_error(sk, msg, len, addr_len);
1615 #if IS_ENABLED(CONFIG_IPV6)
1616         if (sk->sk_family == AF_INET6)
1617                 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1618 #endif
1619         return -EINVAL;
1620 }
1621
1622 int inet_gro_complete(struct sk_buff *skb, int nhoff)
1623 {
1624         __be16 newlen = htons(skb->len - nhoff);
1625         struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1626         const struct net_offload *ops;
1627         int proto = iph->protocol;
1628         int err = -ENOSYS;
1629
1630         if (skb->encapsulation) {
1631                 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1632                 skb_set_inner_network_header(skb, nhoff);
1633         }
1634
1635         csum_replace2(&iph->check, iph->tot_len, newlen);
1636         iph->tot_len = newlen;
1637
1638         ops = rcu_dereference(inet_offloads[proto]);
1639         if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1640                 goto out;
1641
1642         /* Only need to add sizeof(*iph) to get to the next hdr below
1643          * because any hdr with option will have been flushed in
1644          * inet_gro_receive().
1645          */
1646         err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1647                               tcp4_gro_complete, udp4_gro_complete,
1648                               skb, nhoff + sizeof(*iph));
1649
1650 out:
1651         return err;
1652 }
1653
1654 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1655 {
1656         skb->encapsulation = 1;
1657         skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1658         return inet_gro_complete(skb, nhoff);
1659 }
1660
1661 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1662                          unsigned short type, unsigned char protocol,
1663                          struct net *net)
1664 {
1665         struct socket *sock;
1666         int rc = sock_create_kern(net, family, type, protocol, &sock);
1667
1668         if (rc == 0) {
1669                 *sk = sock->sk;
1670                 (*sk)->sk_allocation = GFP_ATOMIC;
1671                 /*
1672                  * Unhash it so that IP input processing does not even see it,
1673                  * we do not wish this socket to see incoming packets.
1674                  */
1675                 (*sk)->sk_prot->unhash(*sk);
1676         }
1677         return rc;
1678 }
1679 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1680
1681 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1682 {
1683         unsigned long res = 0;
1684         int i;
1685
1686         for_each_possible_cpu(i)
1687                 res += snmp_get_cpu_field(mib, i, offt);
1688         return res;
1689 }
1690 EXPORT_SYMBOL_GPL(snmp_fold_field);
1691
1692 #if BITS_PER_LONG==32
1693
1694 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1695                          size_t syncp_offset)
1696 {
1697         void *bhptr;
1698         struct u64_stats_sync *syncp;
1699         u64 v;
1700         unsigned int start;
1701
1702         bhptr = per_cpu_ptr(mib, cpu);
1703         syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1704         do {
1705                 start = u64_stats_fetch_begin_irq(syncp);
1706                 v = *(((u64 *)bhptr) + offt);
1707         } while (u64_stats_fetch_retry_irq(syncp, start));
1708
1709         return v;
1710 }
1711 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1712
1713 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1714 {
1715         u64 res = 0;
1716         int cpu;
1717
1718         for_each_possible_cpu(cpu) {
1719                 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1720         }
1721         return res;
1722 }
1723 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1724 #endif
1725
1726 #ifdef CONFIG_IP_MULTICAST
1727 static const struct net_protocol igmp_protocol = {
1728         .handler =      igmp_rcv,
1729 };
1730 #endif
1731
1732 static const struct net_protocol tcp_protocol = {
1733         .handler        =       tcp_v4_rcv,
1734         .err_handler    =       tcp_v4_err,
1735         .no_policy      =       1,
1736         .icmp_strict_tag_validation = 1,
1737 };
1738
1739 static const struct net_protocol udp_protocol = {
1740         .handler =      udp_rcv,
1741         .err_handler =  udp_err,
1742         .no_policy =    1,
1743 };
1744
1745 static const struct net_protocol icmp_protocol = {
1746         .handler =      icmp_rcv,
1747         .err_handler =  icmp_err,
1748         .no_policy =    1,
1749 };
1750
1751 static __net_init int ipv4_mib_init_net(struct net *net)
1752 {
1753         int i;
1754
1755         net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1756         if (!net->mib.tcp_statistics)
1757                 goto err_tcp_mib;
1758         net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1759         if (!net->mib.ip_statistics)
1760                 goto err_ip_mib;
1761
1762         for_each_possible_cpu(i) {
1763                 struct ipstats_mib *af_inet_stats;
1764                 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1765                 u64_stats_init(&af_inet_stats->syncp);
1766         }
1767
1768         net->mib.net_statistics = alloc_percpu(struct linux_mib);
1769         if (!net->mib.net_statistics)
1770                 goto err_net_mib;
1771         net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1772         if (!net->mib.udp_statistics)
1773                 goto err_udp_mib;
1774         net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1775         if (!net->mib.udplite_statistics)
1776                 goto err_udplite_mib;
1777         net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1778         if (!net->mib.icmp_statistics)
1779                 goto err_icmp_mib;
1780         net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1781                                               GFP_KERNEL);
1782         if (!net->mib.icmpmsg_statistics)
1783                 goto err_icmpmsg_mib;
1784
1785         tcp_mib_init(net);
1786         return 0;
1787
1788 err_icmpmsg_mib:
1789         free_percpu(net->mib.icmp_statistics);
1790 err_icmp_mib:
1791         free_percpu(net->mib.udplite_statistics);
1792 err_udplite_mib:
1793         free_percpu(net->mib.udp_statistics);
1794 err_udp_mib:
1795         free_percpu(net->mib.net_statistics);
1796 err_net_mib:
1797         free_percpu(net->mib.ip_statistics);
1798 err_ip_mib:
1799         free_percpu(net->mib.tcp_statistics);
1800 err_tcp_mib:
1801         return -ENOMEM;
1802 }
1803
1804 static __net_exit void ipv4_mib_exit_net(struct net *net)
1805 {
1806         kfree(net->mib.icmpmsg_statistics);
1807         free_percpu(net->mib.icmp_statistics);
1808         free_percpu(net->mib.udplite_statistics);
1809         free_percpu(net->mib.udp_statistics);
1810         free_percpu(net->mib.net_statistics);
1811         free_percpu(net->mib.ip_statistics);
1812         free_percpu(net->mib.tcp_statistics);
1813 #ifdef CONFIG_MPTCP
1814         /* allocated on demand, see mptcp_init_sock() */
1815         free_percpu(net->mib.mptcp_statistics);
1816 #endif
1817 }
1818
1819 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1820         .init = ipv4_mib_init_net,
1821         .exit = ipv4_mib_exit_net,
1822 };
1823
1824 static int __init init_ipv4_mibs(void)
1825 {
1826         return register_pernet_subsys(&ipv4_mib_ops);
1827 }
1828
1829 static __net_init int inet_init_net(struct net *net)
1830 {
1831         /*
1832          * Set defaults for local port range
1833          */
1834         seqlock_init(&net->ipv4.ip_local_ports.lock);
1835         net->ipv4.ip_local_ports.range[0] =  32768;
1836         net->ipv4.ip_local_ports.range[1] =  60999;
1837
1838         seqlock_init(&net->ipv4.ping_group_range.lock);
1839         /*
1840          * Sane defaults - nobody may create ping sockets.
1841          * Boot scripts should set this to distro-specific group.
1842          */
1843         net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1844         net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1845
1846         /* Default values for sysctl-controlled parameters.
1847          * We set them here, in case sysctl is not compiled.
1848          */
1849         net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1850         net->ipv4.sysctl_ip_fwd_update_priority = 1;
1851         net->ipv4.sysctl_ip_dynaddr = 0;
1852         net->ipv4.sysctl_ip_early_demux = 1;
1853         net->ipv4.sysctl_udp_early_demux = 1;
1854         net->ipv4.sysctl_tcp_early_demux = 1;
1855         net->ipv4.sysctl_nexthop_compat_mode = 1;
1856 #ifdef CONFIG_SYSCTL
1857         net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1858 #endif
1859
1860         /* Some igmp sysctl, whose values are always used */
1861         net->ipv4.sysctl_igmp_max_memberships = 20;
1862         net->ipv4.sysctl_igmp_max_msf = 10;
1863         /* IGMP reports for link-local multicast groups are enabled by default */
1864         net->ipv4.sysctl_igmp_llm_reports = 1;
1865         net->ipv4.sysctl_igmp_qrv = 2;
1866
1867         net->ipv4.sysctl_fib_notify_on_flag_change = 0;
1868
1869         return 0;
1870 }
1871
1872 static __net_initdata struct pernet_operations af_inet_ops = {
1873         .init = inet_init_net,
1874 };
1875
1876 static int __init init_inet_pernet_ops(void)
1877 {
1878         return register_pernet_subsys(&af_inet_ops);
1879 }
1880
1881 static int ipv4_proc_init(void);
1882
1883 /*
1884  *      IP protocol layer initialiser
1885  */
1886
1887 static struct packet_offload ip_packet_offload __read_mostly = {
1888         .type = cpu_to_be16(ETH_P_IP),
1889         .callbacks = {
1890                 .gso_segment = inet_gso_segment,
1891                 .gro_receive = inet_gro_receive,
1892                 .gro_complete = inet_gro_complete,
1893         },
1894 };
1895
1896 static const struct net_offload ipip_offload = {
1897         .callbacks = {
1898                 .gso_segment    = ipip_gso_segment,
1899                 .gro_receive    = ipip_gro_receive,
1900                 .gro_complete   = ipip_gro_complete,
1901         },
1902 };
1903
1904 static int __init ipip_offload_init(void)
1905 {
1906         return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1907 }
1908
1909 static int __init ipv4_offload_init(void)
1910 {
1911         /*
1912          * Add offloads
1913          */
1914         if (udpv4_offload_init() < 0)
1915                 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1916         if (tcpv4_offload_init() < 0)
1917                 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1918         if (ipip_offload_init() < 0)
1919                 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1920
1921         dev_add_offload(&ip_packet_offload);
1922         return 0;
1923 }
1924
1925 fs_initcall(ipv4_offload_init);
1926
1927 static struct packet_type ip_packet_type __read_mostly = {
1928         .type = cpu_to_be16(ETH_P_IP),
1929         .func = ip_rcv,
1930         .list_func = ip_list_rcv,
1931 };
1932
1933 static int __init inet_init(void)
1934 {
1935         struct inet_protosw *q;
1936         struct list_head *r;
1937         int rc;
1938
1939         sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1940
1941         raw_hashinfo_init(&raw_v4_hashinfo);
1942
1943         rc = proto_register(&tcp_prot, 1);
1944         if (rc)
1945                 goto out;
1946
1947         rc = proto_register(&udp_prot, 1);
1948         if (rc)
1949                 goto out_unregister_tcp_proto;
1950
1951         rc = proto_register(&raw_prot, 1);
1952         if (rc)
1953                 goto out_unregister_udp_proto;
1954
1955         rc = proto_register(&ping_prot, 1);
1956         if (rc)
1957                 goto out_unregister_raw_proto;
1958
1959         /*
1960          *      Tell SOCKET that we are alive...
1961          */
1962
1963         (void)sock_register(&inet_family_ops);
1964
1965 #ifdef CONFIG_SYSCTL
1966         ip_static_sysctl_init();
1967 #endif
1968
1969         /*
1970          *      Add all the base protocols.
1971          */
1972
1973         if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1974                 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1975         if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1976                 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1977         if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1978                 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1979 #ifdef CONFIG_IP_MULTICAST
1980         if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1981                 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1982 #endif
1983
1984         /* Register the socket-side information for inet_create. */
1985         for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1986                 INIT_LIST_HEAD(r);
1987
1988         for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1989                 inet_register_protosw(q);
1990
1991         /*
1992          *      Set the ARP module up
1993          */
1994
1995         arp_init();
1996
1997         /*
1998          *      Set the IP module up
1999          */
2000
2001         ip_init();
2002
2003         /* Initialise per-cpu ipv4 mibs */
2004         if (init_ipv4_mibs())
2005                 panic("%s: Cannot init ipv4 mibs\n", __func__);
2006
2007         /* Setup TCP slab cache for open requests. */
2008         tcp_init();
2009
2010         /* Setup UDP memory threshold */
2011         udp_init();
2012
2013         /* Add UDP-Lite (RFC 3828) */
2014         udplite4_register();
2015
2016         raw_init();
2017
2018         ping_init();
2019
2020         /*
2021          *      Set the ICMP layer up
2022          */
2023
2024         if (icmp_init() < 0)
2025                 panic("Failed to create the ICMP control socket.\n");
2026
2027         /*
2028          *      Initialise the multicast router
2029          */
2030 #if defined(CONFIG_IP_MROUTE)
2031         if (ip_mr_init())
2032                 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2033 #endif
2034
2035         if (init_inet_pernet_ops())
2036                 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2037
2038         ipv4_proc_init();
2039
2040         ipfrag_init();
2041
2042         dev_add_pack(&ip_packet_type);
2043
2044         ip_tunnel_core_init();
2045
2046         rc = 0;
2047 out:
2048         return rc;
2049 out_unregister_raw_proto:
2050         proto_unregister(&raw_prot);
2051 out_unregister_udp_proto:
2052         proto_unregister(&udp_prot);
2053 out_unregister_tcp_proto:
2054         proto_unregister(&tcp_prot);
2055         goto out;
2056 }
2057
2058 fs_initcall(inet_init);
2059
2060 /* ------------------------------------------------------------------------ */
2061
2062 #ifdef CONFIG_PROC_FS
2063 static int __init ipv4_proc_init(void)
2064 {
2065         int rc = 0;
2066
2067         if (raw_proc_init())
2068                 goto out_raw;
2069         if (tcp4_proc_init())
2070                 goto out_tcp;
2071         if (udp4_proc_init())
2072                 goto out_udp;
2073         if (ping_proc_init())
2074                 goto out_ping;
2075         if (ip_misc_proc_init())
2076                 goto out_misc;
2077 out:
2078         return rc;
2079 out_misc:
2080         ping_proc_exit();
2081 out_ping:
2082         udp4_proc_exit();
2083 out_udp:
2084         tcp4_proc_exit();
2085 out_tcp:
2086         raw_proc_exit();
2087 out_raw:
2088         rc = -ENOMEM;
2089         goto out;
2090 }
2091
2092 #else /* CONFIG_PROC_FS */
2093 static int __init ipv4_proc_init(void)
2094 {
2095         return 0;
2096 }
2097 #endif /* CONFIG_PROC_FS */