af_inet.c

   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * INET		An implementation of the TCP/IP protocol suite for the LINUX
   4 *		operating system.  INET is implemented using the  BSD Socket
   5 *		interface as the means of communication with the user level.
   6 *
   7 *		PF_INET protocol family socket handler.
   8 *
   9 * Authors:	Ross Biro
  10 *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  11 *		Florian La Roche, <flla@stud.uni-sb.de>
  12 *		Alan Cox, <A.Cox@swansea.ac.uk>
  13 *
  14 * Changes (see also sock.c)
  15 *
  16 *		piggy,
  17 *		Karl Knutson	:	Socket protocol table
  18 *		A.N.Kuznetsov	:	Socket death error in accept().
  19 *		John Richardson :	Fix non blocking error in connect()
  20 *					so sockets that fail to connect
  21 *					don't return -EINPROGRESS.
  22 *		Alan Cox	:	Asynchronous I/O support
  23 *		Alan Cox	:	Keep correct socket pointer on sock
  24 *					structures
  25 *					when accept() ed
  26 *		Alan Cox	:	Semantics of SO_LINGER aren't state
  27 *					moved to close when you look carefully.
  28 *					With this fixed and the accept bug fixed
  29 *					some RPC stuff seems happier.
  30 *		Niibe Yutaka	:	4.4BSD style write async I/O
  31 *		Alan Cox,
  32 *		Tony Gale 	:	Fixed reuse semantics.
  33 *		Alan Cox	:	bind() shouldn't abort existing but dead
  34 *					sockets. Stops FTP netin:.. I hope.
  35 *		Alan Cox	:	bind() works correctly for RAW sockets.
  36 *					Note that FreeBSD at least was broken
  37 *					in this respect so be careful with
  38 *					compatibility tests...
  39 *		Alan Cox	:	routing cache support
  40 *		Alan Cox	:	memzero the socket structure for
  41 *					compactness.
  42 *		Matt Day	:	nonblock connect error handler
  43 *		Alan Cox	:	Allow large numbers of pending sockets
  44 *					(eg for big web sites), but only if
  45 *					specifically application requested.
  46 *		Alan Cox	:	New buffering throughout IP. Used
  47 *					dumbly.
  48 *		Alan Cox	:	New buffering now used smartly.
  49 *		Alan Cox	:	BSD rather than common sense
  50 *					interpretation of listen.
  51 *		Germano Caronni	:	Assorted small races.
  52 *		Alan Cox	:	sendmsg/recvmsg basic support.
  53 *		Alan Cox	:	Only sendmsg/recvmsg now supported.
  54 *		Alan Cox	:	Locked down bind (see security list).
  55 *		Alan Cox	:	Loosened bind a little.
  56 *		Mike McLagan	:	ADD/DEL DLCI Ioctls
  57 *	Willy Konynenberg	:	Transparent proxying support.
  58 *		David S. Miller	:	New socket lookup architecture.
  59 *					Some other random speedups.
  60 *		Cyrus Durgin	:	Cleaned up file for kmod hacks.
  61 *		Andi Kleen	:	Fix inet_stream_connect TCP race.
  62 */
  63
  64#define pr_fmt(fmt) "IPv4: " fmt
  65
  66#include <linux/err.h>
  67#include <linux/errno.h>
  68#include <linux/types.h>
  69#include <linux/socket.h>
  70#include <linux/in.h>
  71#include <linux/kernel.h>
  72#include <linux/kmod.h>
  73#include <linux/sched.h>
  74#include <linux/timer.h>
  75#include <linux/string.h>
  76#include <linux/sockios.h>
  77#include <linux/net.h>
  78#include <linux/capability.h>
  79#include <linux/fcntl.h>
  80#include <linux/mm.h>
  81#include <linux/interrupt.h>
  82#include <linux/stat.h>
  83#include <linux/init.h>
  84#include <linux/poll.h>
  85#include <linux/netfilter_ipv4.h>
  86#include <linux/random.h>
  87#include <linux/slab.h>
  88
  89#include <linux/uaccess.h>
  90
  91#include <linux/inet.h>
  92#include <linux/igmp.h>
  93#include <linux/inetdevice.h>
  94#include <linux/netdevice.h>
  95#include <net/checksum.h>
  96#include <net/ip.h>
  97#include <net/protocol.h>
  98#include <net/arp.h>
  99#include <net/route.h>
 100#include <net/ip_fib.h>
 101#include <net/inet_connection_sock.h>
 102#include <net/gro.h>
 103#include <net/gso.h>
 104#include <net/tcp.h>
 105#include <net/psp.h>
 106#include <net/udp.h>
 107#include <net/udplite.h>
 108#include <net/ping.h>
 109#include <linux/skbuff.h>
 110#include <net/sock.h>
 111#include <net/raw.h>
 112#include <net/icmp.h>
 113#include <net/inet_common.h>
 114#include <net/ip_tunnels.h>
 115#include <net/xfrm.h>
 116#include <net/net_namespace.h>
 117#include <net/secure_seq.h>
 118#ifdef CONFIG_IP_MROUTE
 119#include <linux/mroute.h>
 120#endif
 121#include <net/l3mdev.h>
 122#include <net/compat.h>
 123#include <net/rps.h>
 124
 125#include <trace/events/sock.h>
 126
 127/* The inetsw table contains everything that inet_create needs to
 128 * build a new socket.
 129 */
 130static struct list_head inetsw[SOCK_MAX];
 131static DEFINE_SPINLOCK(inetsw_lock);
 132
 133/* New destruction routine */
 134
 135void inet_sock_destruct(struct sock *sk)
 136{
 137	struct inet_sock *inet = inet_sk(sk);
 138
 139	__skb_queue_purge(&sk->sk_receive_queue);
 140	__skb_queue_purge(&sk->sk_error_queue);
 141
 142	sk_mem_reclaim_final(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_ONCE(atomic_read(&sk->sk_rmem_alloc));
 155	WARN_ON_ONCE(refcount_read(&sk->sk_wmem_alloc));
 156	WARN_ON_ONCE(sk->sk_wmem_queued);
 157	WARN_ON_ONCE(sk->sk_forward_alloc);
 158
 159	kfree(rcu_dereference_protected(inet->inet_opt, 1));
 160	dst_release(rcu_dereference_protected(sk->sk_dst_cache, 1));
 161	dst_release(rcu_dereference_protected(sk->sk_rx_dst, 1));
 162	psp_sk_assoc_free(sk);
 163}
 164EXPORT_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
 176static 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
 193int __inet_listen_sk(struct sock *sk, int backlog)
 194{
 195	unsigned char old_state = sk->sk_state;
 196	int err, tcp_fastopen;
 197
 198	if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
 199		return -EINVAL;
 200
 201	WRITE_ONCE(sk->sk_max_ack_backlog, backlog);
 202	/* Really, if the socket is already in listen state
 203	 * we can only allow the backlog to be adjusted.
 204	 */
 205	if (old_state != TCP_LISTEN) {
 206		/* Enable TFO w/o requiring TCP_FASTOPEN socket option.
 207		 * Note that only TCP sockets (SOCK_STREAM) will reach here.
 208		 * Also fastopen backlog may already been set via the option
 209		 * because the socket was in TCP_LISTEN state previously but
 210		 * was shutdown() rather than close().
 211		 */
 212		tcp_fastopen = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fastopen);
 213		if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
 214		    (tcp_fastopen & TFO_SERVER_ENABLE) &&
 215		    !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
 216			fastopen_queue_tune(sk, backlog);
 217			tcp_fastopen_init_key_once(sock_net(sk));
 218		}
 219
 220		err = inet_csk_listen_start(sk);
 221		if (err)
 222			return err;
 223
 224		tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL);
 225	}
 226	return 0;
 227}
 228
 229/*
 230 *	Move a socket into listening state.
 231 */
 232int inet_listen(struct socket *sock, int backlog)
 233{
 234	struct sock *sk = sock->sk;
 235	int err = -EINVAL;
 236
 237	lock_sock(sk);
 238
 239	if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
 240		goto out;
 241
 242	err = __inet_listen_sk(sk, backlog);
 243
 244out:
 245	release_sock(sk);
 246	return err;
 247}
 248EXPORT_SYMBOL(inet_listen);
 249
 250/*
 251 *	Create an inet socket.
 252 */
 253
 254static int inet_create(struct net *net, struct socket *sock, int protocol,
 255		       int kern)
 256{
 257	struct sock *sk;
 258	struct inet_protosw *answer;
 259	struct inet_sock *inet;
 260	struct proto *answer_prot;
 261	unsigned char answer_flags;
 262	int try_loading_module = 0;
 263	int err;
 264
 265	if (protocol < 0 || protocol >= IPPROTO_MAX)
 266		return -EINVAL;
 267
 268	sock->state = SS_UNCONNECTED;
 269
 270	/* Look for the requested type/protocol pair. */
 271lookup_protocol:
 272	err = -ESOCKTNOSUPPORT;
 273	rcu_read_lock();
 274	list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
 275
 276		err = 0;
 277		/* Check the non-wild match. */
 278		if (protocol == answer->protocol) {
 279			if (protocol != IPPROTO_IP)
 280				break;
 281		} else {
 282			/* Check for the two wild cases. */
 283			if (IPPROTO_IP == protocol) {
 284				protocol = answer->protocol;
 285				break;
 286			}
 287			if (IPPROTO_IP == answer->protocol)
 288				break;
 289		}
 290		err = -EPROTONOSUPPORT;
 291	}
 292
 293	if (unlikely(err)) {
 294		if (try_loading_module < 2) {
 295			rcu_read_unlock();
 296			/*
 297			 * Be more specific, e.g. net-pf-2-proto-132-type-1
 298			 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
 299			 */
 300			if (++try_loading_module == 1)
 301				request_module("net-pf-%d-proto-%d-type-%d",
 302					       PF_INET, protocol, sock->type);
 303			/*
 304			 * Fall back to generic, e.g. net-pf-2-proto-132
 305			 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
 306			 */
 307			else
 308				request_module("net-pf-%d-proto-%d",
 309					       PF_INET, protocol);
 310			goto lookup_protocol;
 311		} else
 312			goto out_rcu_unlock;
 313	}
 314
 315	err = -EPERM;
 316	if (sock->type == SOCK_RAW && !kern &&
 317	    !ns_capable(net->user_ns, CAP_NET_RAW))
 318		goto out_rcu_unlock;
 319
 320	sock->ops = answer->ops;
 321	answer_prot = answer->prot;
 322	answer_flags = answer->flags;
 323	rcu_read_unlock();
 324
 325	WARN_ON(!answer_prot->slab);
 326
 327	err = -ENOMEM;
 328	sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
 329	if (!sk)
 330		goto out;
 331
 332	err = 0;
 333	if (INET_PROTOSW_REUSE & answer_flags)
 334		sk->sk_reuse = SK_CAN_REUSE;
 335
 336	if (INET_PROTOSW_ICSK & answer_flags)
 337		inet_init_csk_locks(sk);
 338
 339	inet = inet_sk(sk);
 340	inet_assign_bit(IS_ICSK, sk, INET_PROTOSW_ICSK & answer_flags);
 341
 342	inet_clear_bit(NODEFRAG, sk);
 343
 344	if (SOCK_RAW == sock->type) {
 345		inet->inet_num = protocol;
 346		if (IPPROTO_RAW == protocol)
 347			inet_set_bit(HDRINCL, sk);
 348	}
 349
 350	if (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc))
 351		inet->pmtudisc = IP_PMTUDISC_DONT;
 352	else
 353		inet->pmtudisc = IP_PMTUDISC_WANT;
 354
 355	atomic_set(&inet->inet_id, 0);
 356
 357	sock_init_data(sock, sk);
 358
 359	sk->sk_destruct	   = inet_sock_destruct;
 360	sk->sk_protocol	   = protocol;
 361	sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
 362	sk->sk_txrehash = READ_ONCE(net->core.sysctl_txrehash);
 363
 364	inet->uc_ttl	= -1;
 365	inet_set_bit(MC_LOOP, sk);
 366	inet->mc_ttl	= 1;
 367	inet_set_bit(MC_ALL, sk);
 368	inet->mc_index	= 0;
 369	inet->mc_list	= NULL;
 370	inet->rcv_tos	= 0;
 371
 372	if (inet->inet_num) {
 373		/* It assumes that any protocol which allows
 374		 * the user to assign a number at socket
 375		 * creation time automatically
 376		 * shares.
 377		 */
 378		inet->inet_sport = htons(inet->inet_num);
 379		/* Add to protocol hash chains. */
 380		err = sk->sk_prot->hash(sk);
 381		if (err)
 382			goto out_sk_release;
 383	}
 384
 385	if (sk->sk_prot->init) {
 386		err = sk->sk_prot->init(sk);
 387		if (err)
 388			goto out_sk_release;
 389	}
 390
 391	if (!kern) {
 392		err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
 393		if (err)
 394			goto out_sk_release;
 395	}
 396out:
 397	return err;
 398out_rcu_unlock:
 399	rcu_read_unlock();
 400	goto out;
 401out_sk_release:
 402	sk_common_release(sk);
 403	sock->sk = NULL;
 404	goto out;
 405}
 406
 407
 408/*
 409 *	The peer socket should always be NULL (or else). When we call this
 410 *	function we are destroying the object and from then on nobody
 411 *	should refer to it.
 412 */
 413int inet_release(struct socket *sock)
 414{
 415	struct sock *sk = sock->sk;
 416
 417	if (sk) {
 418		long timeout;
 419
 420		if (!sk->sk_kern_sock)
 421			BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk);
 422
 423		/* Applications forget to leave groups before exiting */
 424		ip_mc_drop_socket(sk);
 425
 426		/* If linger is set, we don't return until the close
 427		 * is complete.  Otherwise we return immediately. The
 428		 * actually closing is done the same either way.
 429		 *
 430		 * If the close is due to the process exiting, we never
 431		 * linger..
 432		 */
 433		timeout = 0;
 434		if (sock_flag(sk, SOCK_LINGER) &&
 435		    !(current->flags & PF_EXITING))
 436			timeout = sk->sk_lingertime;
 437		sk->sk_prot->close(sk, timeout);
 438		sock->sk = NULL;
 439	}
 440	return 0;
 441}
 442EXPORT_SYMBOL(inet_release);
 443
 444int inet_bind_sk(struct sock *sk, struct sockaddr_unsized *uaddr, int addr_len)
 445{
 446	u32 flags = BIND_WITH_LOCK;
 447	int err;
 448
 449	/* If the socket has its own bind function then use it. (RAW) */
 450	if (sk->sk_prot->bind) {
 451		return sk->sk_prot->bind(sk, uaddr, addr_len);
 452	}
 453	if (addr_len < sizeof(struct sockaddr_in))
 454		return -EINVAL;
 455
 456	/* BPF prog is run before any checks are done so that if the prog
 457	 * changes context in a wrong way it will be caught.
 458	 */
 459	err = BPF_CGROUP_RUN_PROG_INET_BIND_LOCK(sk, uaddr, &addr_len,
 460						 CGROUP_INET4_BIND, &flags);
 461	if (err)
 462		return err;
 463
 464	return __inet_bind(sk, uaddr, addr_len, flags);
 465}
 466
 467int inet_bind(struct socket *sock, struct sockaddr_unsized *uaddr, int addr_len)
 468{
 469	return inet_bind_sk(sock->sk, uaddr, addr_len);
 470}
 471EXPORT_SYMBOL(inet_bind);
 472
 473int __inet_bind(struct sock *sk, struct sockaddr_unsized *uaddr, int addr_len,
 474		u32 flags)
 475{
 476	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
 477	struct inet_sock *inet = inet_sk(sk);
 478	struct net *net = sock_net(sk);
 479	unsigned short snum;
 480	int chk_addr_ret;
 481	u32 tb_id = RT_TABLE_LOCAL;
 482	int err;
 483
 484	if (addr->sin_family != AF_INET) {
 485		/* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
 486		 * only if s_addr is INADDR_ANY.
 487		 */
 488		err = -EAFNOSUPPORT;
 489		if (addr->sin_family != AF_UNSPEC ||
 490		    addr->sin_addr.s_addr != htonl(INADDR_ANY))
 491			goto out;
 492	}
 493
 494	tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
 495	chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
 496
 497	/* Not specified by any standard per-se, however it breaks too
 498	 * many applications when removed.  It is unfortunate since
 499	 * allowing applications to make a non-local bind solves
 500	 * several problems with systems using dynamic addressing.
 501	 * (ie. your servers still start up even if your ISDN link
 502	 *  is temporarily down)
 503	 */
 504	err = -EADDRNOTAVAIL;
 505	if (!inet_addr_valid_or_nonlocal(net, inet, addr->sin_addr.s_addr,
 506	                                 chk_addr_ret))
 507		goto out;
 508
 509	snum = ntohs(addr->sin_port);
 510	err = -EACCES;
 511	if (!(flags & BIND_NO_CAP_NET_BIND_SERVICE) &&
 512	    snum && inet_port_requires_bind_service(net, snum) &&
 513	    !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
 514		goto out;
 515
 516	/*      We keep a pair of addresses. rcv_saddr is the one
 517	 *      used by hash lookups, and saddr is used for transmit.
 518	 *
 519	 *      In the BSD API these are the same except where it
 520	 *      would be illegal to use them (multicast/broadcast) in
 521	 *      which case the sending device address is used.
 522	 */
 523	if (flags & BIND_WITH_LOCK)
 524		lock_sock(sk);
 525
 526	/* Check these errors (active socket, double bind). */
 527	err = -EINVAL;
 528	if (sk->sk_state != TCP_CLOSE || inet->inet_num)
 529		goto out_release_sock;
 530
 531	inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
 532	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
 533		inet->inet_saddr = 0;  /* Use device */
 534
 535	/* Make sure we are allowed to bind here. */
 536	if (snum || !(inet_test_bit(BIND_ADDRESS_NO_PORT, sk) ||
 537		      (flags & BIND_FORCE_ADDRESS_NO_PORT))) {
 538		err = sk->sk_prot->get_port(sk, snum);
 539		if (err) {
 540			inet->inet_saddr = inet->inet_rcv_saddr = 0;
 541			goto out_release_sock;
 542		}
 543		if (!(flags & BIND_FROM_BPF)) {
 544			err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
 545			if (err) {
 546				inet->inet_saddr = inet->inet_rcv_saddr = 0;
 547				if (sk->sk_prot->put_port)
 548					sk->sk_prot->put_port(sk);
 549				goto out_release_sock;
 550			}
 551		}
 552	}
 553
 554	if (inet->inet_rcv_saddr)
 555		sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
 556	if (snum)
 557		sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
 558	inet->inet_sport = htons(inet->inet_num);
 559	inet->inet_daddr = 0;
 560	inet->inet_dport = 0;
 561	sk_dst_reset(sk);
 562	err = 0;
 563out_release_sock:
 564	if (flags & BIND_WITH_LOCK)
 565		release_sock(sk);
 566out:
 567	return err;
 568}
 569
 570int inet_dgram_connect(struct socket *sock, struct sockaddr_unsized *uaddr,
 571		       int addr_len, int flags)
 572{
 573	struct sock *sk = sock->sk;
 574	const struct proto *prot;
 575	int err;
 576
 577	if (addr_len < sizeof(uaddr->sa_family))
 578		return -EINVAL;
 579
 580	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
 581	prot = READ_ONCE(sk->sk_prot);
 582
 583	if (uaddr->sa_family == AF_UNSPEC)
 584		return prot->disconnect(sk, flags);
 585
 586	if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
 587		err = prot->pre_connect(sk, uaddr, addr_len);
 588		if (err)
 589			return err;
 590	}
 591
 592	if (data_race(!inet_sk(sk)->inet_num) && inet_autobind(sk))
 593		return -EAGAIN;
 594	return prot->connect(sk, uaddr, addr_len);
 595}
 596EXPORT_SYMBOL(inet_dgram_connect);
 597
 598static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
 599{
 600	DEFINE_WAIT_FUNC(wait, woken_wake_function);
 601
 602	add_wait_queue(sk_sleep(sk), &wait);
 603	sk->sk_write_pending += writebias;
 604
 605	/* Basic assumption: if someone sets sk->sk_err, he _must_
 606	 * change state of the socket from TCP_SYN_*.
 607	 * Connect() does not allow to get error notifications
 608	 * without closing the socket.
 609	 */
 610	while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 611		release_sock(sk);
 612		timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
 613		lock_sock(sk);
 614		if (signal_pending(current) || !timeo)
 615			break;
 616	}
 617	remove_wait_queue(sk_sleep(sk), &wait);
 618	sk->sk_write_pending -= writebias;
 619	return timeo;
 620}
 621
 622/*
 623 *	Connect to a remote host. There is regrettably still a little
 624 *	TCP 'magic' in here.
 625 */
 626int __inet_stream_connect(struct socket *sock, struct sockaddr_unsized *uaddr,
 627			  int addr_len, int flags, int is_sendmsg)
 628{
 629	struct sock *sk = sock->sk;
 630	int err;
 631	long timeo;
 632
 633	/*
 634	 * uaddr can be NULL and addr_len can be 0 if:
 635	 * sk is a TCP fastopen active socket and
 636	 * TCP_FASTOPEN_CONNECT sockopt is set and
 637	 * we already have a valid cookie for this socket.
 638	 * In this case, user can call write() after connect().
 639	 * write() will invoke tcp_sendmsg_fastopen() which calls
 640	 * __inet_stream_connect().
 641	 */
 642	if (uaddr) {
 643		if (addr_len < sizeof(uaddr->sa_family))
 644			return -EINVAL;
 645
 646		if (uaddr->sa_family == AF_UNSPEC) {
 647			sk->sk_disconnects++;
 648			err = sk->sk_prot->disconnect(sk, flags);
 649			sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 650			goto out;
 651		}
 652	}
 653
 654	switch (sock->state) {
 655	default:
 656		err = -EINVAL;
 657		goto out;
 658	case SS_CONNECTED:
 659		err = -EISCONN;
 660		goto out;
 661	case SS_CONNECTING:
 662		if (inet_test_bit(DEFER_CONNECT, sk))
 663			err = is_sendmsg ? -EINPROGRESS : -EISCONN;
 664		else
 665			err = -EALREADY;
 666		/* Fall out of switch with err, set for this state */
 667		break;
 668	case SS_UNCONNECTED:
 669		err = -EISCONN;
 670		if (sk->sk_state != TCP_CLOSE)
 671			goto out;
 672
 673		if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
 674			err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
 675			if (err)
 676				goto out;
 677		}
 678
 679		err = sk->sk_prot->connect(sk, uaddr, addr_len);
 680		if (err < 0)
 681			goto out;
 682
 683		sock->state = SS_CONNECTING;
 684
 685		if (!err && inet_test_bit(DEFER_CONNECT, sk))
 686			goto out;
 687
 688		/* Just entered SS_CONNECTING state; the only
 689		 * difference is that return value in non-blocking
 690		 * case is EINPROGRESS, rather than EALREADY.
 691		 */
 692		err = -EINPROGRESS;
 693		break;
 694	}
 695
 696	timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
 697
 698	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 699		int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
 700				tcp_sk(sk)->fastopen_req &&
 701				tcp_sk(sk)->fastopen_req->data ? 1 : 0;
 702		int dis = sk->sk_disconnects;
 703
 704		/* Error code is set above */
 705		if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
 706			goto out;
 707
 708		err = sock_intr_errno(timeo);
 709		if (signal_pending(current))
 710			goto out;
 711
 712		if (dis != sk->sk_disconnects) {
 713			err = -EPIPE;
 714			goto out;
 715		}
 716	}
 717
 718	/* Connection was closed by RST, timeout, ICMP error
 719	 * or another process disconnected us.
 720	 */
 721	if (sk->sk_state == TCP_CLOSE)
 722		goto sock_error;
 723
 724	/* sk->sk_err may be not zero now, if RECVERR was ordered by user
 725	 * and error was received after socket entered established state.
 726	 * Hence, it is handled normally after connect() return successfully.
 727	 */
 728
 729	sock->state = SS_CONNECTED;
 730	err = 0;
 731out:
 732	return err;
 733
 734sock_error:
 735	err = sock_error(sk) ? : -ECONNABORTED;
 736	sock->state = SS_UNCONNECTED;
 737	sk->sk_disconnects++;
 738	if (sk->sk_prot->disconnect(sk, flags))
 739		sock->state = SS_DISCONNECTING;
 740	goto out;
 741}
 742EXPORT_SYMBOL(__inet_stream_connect);
 743
 744int inet_stream_connect(struct socket *sock, struct sockaddr_unsized *uaddr,
 745			int addr_len, int flags)
 746{
 747	int err;
 748
 749	lock_sock(sock->sk);
 750	err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
 751	release_sock(sock->sk);
 752	return err;
 753}
 754EXPORT_SYMBOL(inet_stream_connect);
 755
 756void __inet_accept(struct socket *sock, struct socket *newsock, struct sock *newsk)
 757{
 758	if (mem_cgroup_sockets_enabled) {
 759		mem_cgroup_sk_alloc(newsk);
 760		__sk_charge(newsk, GFP_KERNEL);
 761	}
 762
 763	sock_rps_record_flow(newsk);
 764	WARN_ON(!((1 << newsk->sk_state) &
 765		  (TCPF_ESTABLISHED | TCPF_SYN_RECV |
 766		   TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2 |
 767		   TCPF_CLOSING | TCPF_CLOSE_WAIT |
 768		   TCPF_CLOSE)));
 769
 770	if (test_bit(SOCK_SUPPORT_ZC, &sock->flags))
 771		set_bit(SOCK_SUPPORT_ZC, &newsock->flags);
 772	sock_graft(newsk, newsock);
 773
 774	newsock->state = SS_CONNECTED;
 775}
 776EXPORT_SYMBOL_GPL(__inet_accept);
 777
 778/*
 779 *	Accept a pending connection. The TCP layer now gives BSD semantics.
 780 */
 781
 782int inet_accept(struct socket *sock, struct socket *newsock,
 783		struct proto_accept_arg *arg)
 784{
 785	struct sock *sk1 = sock->sk, *sk2;
 786
 787	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
 788	arg->err = -EINVAL;
 789	sk2 = READ_ONCE(sk1->sk_prot)->accept(sk1, arg);
 790	if (!sk2)
 791		return arg->err;
 792
 793	lock_sock(sk2);
 794	__inet_accept(sock, newsock, sk2);
 795	release_sock(sk2);
 796	return 0;
 797}
 798EXPORT_SYMBOL(inet_accept);
 799
 800/*
 801 *	This does both peername and sockname.
 802 */
 803int inet_getname(struct socket *sock, struct sockaddr *uaddr,
 804		 int peer)
 805{
 806	struct sock *sk		= sock->sk;
 807	struct inet_sock *inet	= inet_sk(sk);
 808	DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
 809	int sin_addr_len = sizeof(*sin);
 810
 811	sin->sin_family = AF_INET;
 812	lock_sock(sk);
 813	if (peer) {
 814		if (!inet->inet_dport ||
 815		    (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
 816		     peer == 1)) {
 817			release_sock(sk);
 818			return -ENOTCONN;
 819		}
 820		sin->sin_port = inet->inet_dport;
 821		sin->sin_addr.s_addr = inet->inet_daddr;
 822		BPF_CGROUP_RUN_SA_PROG(sk, sin, &sin_addr_len,
 823				       CGROUP_INET4_GETPEERNAME);
 824	} else {
 825		__be32 addr = inet->inet_rcv_saddr;
 826		if (!addr)
 827			addr = inet->inet_saddr;
 828		sin->sin_port = inet->inet_sport;
 829		sin->sin_addr.s_addr = addr;
 830		BPF_CGROUP_RUN_SA_PROG(sk, sin, &sin_addr_len,
 831				       CGROUP_INET4_GETSOCKNAME);
 832	}
 833	release_sock(sk);
 834	memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 835	return sin_addr_len;
 836}
 837EXPORT_SYMBOL(inet_getname);
 838
 839int inet_send_prepare(struct sock *sk)
 840{
 841	sock_rps_record_flow(sk);
 842
 843	/* We may need to bind the socket. */
 844	if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind &&
 845	    inet_autobind(sk))
 846		return -EAGAIN;
 847
 848	return 0;
 849}
 850EXPORT_SYMBOL_GPL(inet_send_prepare);
 851
 852int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
 853{
 854	struct sock *sk = sock->sk;
 855
 856	if (unlikely(inet_send_prepare(sk)))
 857		return -EAGAIN;
 858
 859	return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
 860			       sk, msg, size);
 861}
 862EXPORT_SYMBOL(inet_sendmsg);
 863
 864void inet_splice_eof(struct socket *sock)
 865{
 866	const struct proto *prot;
 867	struct sock *sk = sock->sk;
 868
 869	if (unlikely(inet_send_prepare(sk)))
 870		return;
 871
 872	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
 873	prot = READ_ONCE(sk->sk_prot);
 874	if (prot->splice_eof)
 875		prot->splice_eof(sock);
 876}
 877EXPORT_SYMBOL_GPL(inet_splice_eof);
 878
 879INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
 880					  size_t, int, int *));
 881int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
 882		 int flags)
 883{
 884	struct sock *sk = sock->sk;
 885	int addr_len = 0;
 886	int err;
 887
 888	if (likely(!(flags & MSG_ERRQUEUE)))
 889		sock_rps_record_flow(sk);
 890
 891	err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
 892			      sk, msg, size, flags, &addr_len);
 893	if (err >= 0)
 894		msg->msg_namelen = addr_len;
 895	return err;
 896}
 897EXPORT_SYMBOL(inet_recvmsg);
 898
 899int inet_shutdown(struct socket *sock, int how)
 900{
 901	struct sock *sk = sock->sk;
 902	int err = 0;
 903
 904	/* This should really check to make sure
 905	 * the socket is a TCP socket. (WHY AC...)
 906	 */
 907	how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
 908		       1->2 bit 2 snds.
 909		       2->3 */
 910	if ((how & ~SHUTDOWN_MASK) || !how)	/* MAXINT->0 */
 911		return -EINVAL;
 912
 913	lock_sock(sk);
 914	if (sock->state == SS_CONNECTING) {
 915		if ((1 << sk->sk_state) &
 916		    (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
 917			sock->state = SS_DISCONNECTING;
 918		else
 919			sock->state = SS_CONNECTED;
 920	}
 921
 922	switch (sk->sk_state) {
 923	case TCP_CLOSE:
 924		err = -ENOTCONN;
 925		/* Hack to wake up other listeners, who can poll for
 926		   EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
 927		fallthrough;
 928	default:
 929		WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | how);
 930		if (sk->sk_prot->shutdown)
 931			sk->sk_prot->shutdown(sk, how);
 932		break;
 933
 934	/* Remaining two branches are temporary solution for missing
 935	 * close() in multithreaded environment. It is _not_ a good idea,
 936	 * but we have no choice until close() is repaired at VFS level.
 937	 */
 938	case TCP_LISTEN:
 939		if (!(how & RCV_SHUTDOWN))
 940			break;
 941		fallthrough;
 942	case TCP_SYN_SENT:
 943		err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
 944		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 945		break;
 946	}
 947
 948	/* Wake up anyone sleeping in poll. */
 949	sk->sk_state_change(sk);
 950	release_sock(sk);
 951	return err;
 952}
 953EXPORT_SYMBOL(inet_shutdown);
 954
 955/*
 956 *	ioctl() calls you can issue on an INET socket. Most of these are
 957 *	device configuration and stuff and very rarely used. Some ioctls
 958 *	pass on to the socket itself.
 959 *
 960 *	NOTE: I like the idea of a module for the config stuff. ie ifconfig
 961 *	loads the devconfigure module does its configuring and unloads it.
 962 *	There's a good 20K of config code hanging around the kernel.
 963 */
 964
 965int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 966{
 967	struct sock *sk = sock->sk;
 968	int err = 0;
 969	struct net *net = sock_net(sk);
 970	void __user *p = (void __user *)arg;
 971	struct ifreq ifr;
 972	struct rtentry rt;
 973
 974	switch (cmd) {
 975	case SIOCADDRT:
 976	case SIOCDELRT:
 977		if (copy_from_user(&rt, p, sizeof(struct rtentry)))
 978			return -EFAULT;
 979		err = ip_rt_ioctl(net, cmd, &rt);
 980		break;
 981	case SIOCRTMSG:
 982		err = -EINVAL;
 983		break;
 984	case SIOCDARP:
 985	case SIOCGARP:
 986	case SIOCSARP:
 987		err = arp_ioctl(net, cmd, (void __user *)arg);
 988		break;
 989	case SIOCGIFADDR:
 990	case SIOCGIFBRDADDR:
 991	case SIOCGIFNETMASK:
 992	case SIOCGIFDSTADDR:
 993	case SIOCGIFPFLAGS:
 994		if (get_user_ifreq(&ifr, NULL, p))
 995			return -EFAULT;
 996		err = devinet_ioctl(net, cmd, &ifr);
 997		if (!err && put_user_ifreq(&ifr, p))
 998			err = -EFAULT;
 999		break;
1000
1001	case SIOCSIFADDR:
1002	case SIOCSIFBRDADDR:
1003	case SIOCSIFNETMASK:
1004	case SIOCSIFDSTADDR:
1005	case SIOCSIFPFLAGS:
1006	case SIOCSIFFLAGS:
1007		if (get_user_ifreq(&ifr, NULL, p))
1008			return -EFAULT;
1009		err = devinet_ioctl(net, cmd, &ifr);
1010		break;
1011	default:
1012		if (sk->sk_prot->ioctl)
1013			err = sk_ioctl(sk, cmd, (void __user *)arg);
1014		else
1015			err = -ENOIOCTLCMD;
1016		break;
1017	}
1018	return err;
1019}
1020EXPORT_SYMBOL(inet_ioctl);
1021
1022#ifdef CONFIG_COMPAT
1023static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd,
1024		struct compat_rtentry __user *ur)
1025{
1026	compat_uptr_t rtdev;
1027	struct rtentry rt;
1028
1029	if (copy_from_user(&rt.rt_dst, &ur->rt_dst,
1030			3 * sizeof(struct sockaddr)) ||
1031	    get_user(rt.rt_flags, &ur->rt_flags) ||
1032	    get_user(rt.rt_metric, &ur->rt_metric) ||
1033	    get_user(rt.rt_mtu, &ur->rt_mtu) ||
1034	    get_user(rt.rt_window, &ur->rt_window) ||
1035	    get_user(rt.rt_irtt, &ur->rt_irtt) ||
1036	    get_user(rtdev, &ur->rt_dev))
1037		return -EFAULT;
1038
1039	rt.rt_dev = compat_ptr(rtdev);
1040	return ip_rt_ioctl(sock_net(sk), cmd, &rt);
1041}
1042
1043static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1044{
1045	void __user *argp = compat_ptr(arg);
1046	struct sock *sk = sock->sk;
1047
1048	switch (cmd) {
1049	case SIOCADDRT:
1050	case SIOCDELRT:
1051		return inet_compat_routing_ioctl(sk, cmd, argp);
1052	default:
1053		if (!sk->sk_prot->compat_ioctl)
1054			return -ENOIOCTLCMD;
1055		return sk->sk_prot->compat_ioctl(sk, cmd, arg);
1056	}
1057}
1058#endif /* CONFIG_COMPAT */
1059
1060const struct proto_ops inet_stream_ops = {
1061	.family		   = PF_INET,
1062	.owner		   = THIS_MODULE,
1063	.release	   = inet_release,
1064	.bind		   = inet_bind,
1065	.connect	   = inet_stream_connect,
1066	.socketpair	   = sock_no_socketpair,
1067	.accept		   = inet_accept,
1068	.getname	   = inet_getname,
1069	.poll		   = tcp_poll,
1070	.ioctl		   = inet_ioctl,
1071	.gettstamp	   = sock_gettstamp,
1072	.listen		   = inet_listen,
1073	.shutdown	   = inet_shutdown,
1074	.setsockopt	   = sock_common_setsockopt,
1075	.getsockopt	   = sock_common_getsockopt,
1076	.sendmsg	   = inet_sendmsg,
1077	.recvmsg	   = inet_recvmsg,
1078#ifdef CONFIG_MMU
1079	.mmap		   = tcp_mmap,
1080#endif
1081	.splice_eof	   = inet_splice_eof,
1082	.splice_read	   = tcp_splice_read,
1083	.set_peek_off      = sk_set_peek_off,
1084	.read_sock	   = tcp_read_sock,
1085	.read_skb	   = tcp_read_skb,
1086	.sendmsg_locked    = tcp_sendmsg_locked,
1087	.peek_len	   = tcp_peek_len,
1088#ifdef CONFIG_COMPAT
1089	.compat_ioctl	   = inet_compat_ioctl,
1090#endif
1091	.set_rcvlowat	   = tcp_set_rcvlowat,
1092};
1093EXPORT_SYMBOL(inet_stream_ops);
1094
1095const struct proto_ops inet_dgram_ops = {
1096	.family		   = PF_INET,
1097	.owner		   = THIS_MODULE,
1098	.release	   = inet_release,
1099	.bind		   = inet_bind,
1100	.connect	   = inet_dgram_connect,
1101	.socketpair	   = sock_no_socketpair,
1102	.accept		   = sock_no_accept,
1103	.getname	   = inet_getname,
1104	.poll		   = udp_poll,
1105	.ioctl		   = inet_ioctl,
1106	.gettstamp	   = sock_gettstamp,
1107	.listen		   = sock_no_listen,
1108	.shutdown	   = inet_shutdown,
1109	.setsockopt	   = sock_common_setsockopt,
1110	.getsockopt	   = sock_common_getsockopt,
1111	.sendmsg	   = inet_sendmsg,
1112	.read_skb	   = udp_read_skb,
1113	.recvmsg	   = inet_recvmsg,
1114	.mmap		   = sock_no_mmap,
1115	.splice_eof	   = inet_splice_eof,
1116	.set_peek_off	   = udp_set_peek_off,
1117#ifdef CONFIG_COMPAT
1118	.compat_ioctl	   = inet_compat_ioctl,
1119#endif
1120};
1121EXPORT_SYMBOL(inet_dgram_ops);
1122
1123/*
1124 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1125 * udp_poll
1126 */
1127static const struct proto_ops inet_sockraw_ops = {
1128	.family		   = PF_INET,
1129	.owner		   = THIS_MODULE,
1130	.release	   = inet_release,
1131	.bind		   = inet_bind,
1132	.connect	   = inet_dgram_connect,
1133	.socketpair	   = sock_no_socketpair,
1134	.accept		   = sock_no_accept,
1135	.getname	   = inet_getname,
1136	.poll		   = datagram_poll,
1137	.ioctl		   = inet_ioctl,
1138	.gettstamp	   = sock_gettstamp,
1139	.listen		   = sock_no_listen,
1140	.shutdown	   = inet_shutdown,
1141	.setsockopt	   = sock_common_setsockopt,
1142	.getsockopt	   = sock_common_getsockopt,
1143	.sendmsg	   = inet_sendmsg,
1144	.recvmsg	   = inet_recvmsg,
1145	.mmap		   = sock_no_mmap,
1146	.splice_eof	   = inet_splice_eof,
1147#ifdef CONFIG_COMPAT
1148	.compat_ioctl	   = inet_compat_ioctl,
1149#endif
1150};
1151
1152static const struct net_proto_family inet_family_ops = {
1153	.family = PF_INET,
1154	.create = inet_create,
1155	.owner	= THIS_MODULE,
1156};
1157
1158/* Upon startup we insert all the elements in inetsw_array[] into
1159 * the linked list inetsw.
1160 */
1161static struct inet_protosw inetsw_array[] =
1162{
1163	{
1164		.type =       SOCK_STREAM,
1165		.protocol =   IPPROTO_TCP,
1166		.prot =       &tcp_prot,
1167		.ops =        &inet_stream_ops,
1168		.flags =      INET_PROTOSW_PERMANENT |
1169			      INET_PROTOSW_ICSK,
1170	},
1171
1172	{
1173		.type =       SOCK_DGRAM,
1174		.protocol =   IPPROTO_UDP,
1175		.prot =       &udp_prot,
1176		.ops =        &inet_dgram_ops,
1177		.flags =      INET_PROTOSW_PERMANENT,
1178       },
1179
1180       {
1181		.type =       SOCK_DGRAM,
1182		.protocol =   IPPROTO_ICMP,
1183		.prot =       &ping_prot,
1184		.ops =        &inet_sockraw_ops,
1185		.flags =      INET_PROTOSW_REUSE,
1186       },
1187
1188       {
1189	       .type =       SOCK_RAW,
1190	       .protocol =   IPPROTO_IP,	/* wild card */
1191	       .prot =       &raw_prot,
1192	       .ops =        &inet_sockraw_ops,
1193	       .flags =      INET_PROTOSW_REUSE,
1194       }
1195};
1196
1197#define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1198
1199void inet_register_protosw(struct inet_protosw *p)
1200{
1201	struct list_head *lh;
1202	struct inet_protosw *answer;
1203	int protocol = p->protocol;
1204	struct list_head *last_perm;
1205
1206	spin_lock_bh(&inetsw_lock);
1207
1208	if (p->type >= SOCK_MAX)
1209		goto out_illegal;
1210
1211	/* If we are trying to override a permanent protocol, bail. */
1212	last_perm = &inetsw[p->type];
1213	list_for_each(lh, &inetsw[p->type]) {
1214		answer = list_entry(lh, struct inet_protosw, list);
1215		/* Check only the non-wild match. */
1216		if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1217			break;
1218		if (protocol == answer->protocol)
1219			goto out_permanent;
1220		last_perm = lh;
1221	}
1222
1223	/* Add the new entry after the last permanent entry if any, so that
1224	 * the new entry does not override a permanent entry when matched with
1225	 * a wild-card protocol. But it is allowed to override any existing
1226	 * non-permanent entry.  This means that when we remove this entry, the
1227	 * system automatically returns to the old behavior.
1228	 */
1229	list_add_rcu(&p->list, last_perm);
1230out:
1231	spin_unlock_bh(&inetsw_lock);
1232
1233	return;
1234
1235out_permanent:
1236	pr_err("Attempt to override permanent protocol %d\n", protocol);
1237	goto out;
1238
1239out_illegal:
1240	pr_err("Ignoring attempt to register invalid socket type %d\n",
1241	       p->type);
1242	goto out;
1243}
1244EXPORT_SYMBOL(inet_register_protosw);
1245
1246void inet_unregister_protosw(struct inet_protosw *p)
1247{
1248	if (INET_PROTOSW_PERMANENT & p->flags) {
1249		pr_err("Attempt to unregister permanent protocol %d\n",
1250		       p->protocol);
1251	} else {
1252		spin_lock_bh(&inetsw_lock);
1253		list_del_rcu(&p->list);
1254		spin_unlock_bh(&inetsw_lock);
1255
1256		synchronize_net();
1257	}
1258}
1259EXPORT_SYMBOL(inet_unregister_protosw);
1260
1261static int inet_sk_reselect_saddr(struct sock *sk)
1262{
1263	struct inet_sock *inet = inet_sk(sk);
1264	__be32 old_saddr = inet->inet_saddr;
1265	__be32 daddr = inet->inet_daddr;
1266	struct flowi4 *fl4;
1267	struct rtable *rt;
1268	__be32 new_saddr;
1269	struct ip_options_rcu *inet_opt;
1270	int err;
1271
1272	inet_opt = rcu_dereference_protected(inet->inet_opt,
1273					     lockdep_sock_is_held(sk));
1274	if (inet_opt && inet_opt->opt.srr)
1275		daddr = inet_opt->opt.faddr;
1276
1277	/* Query new route. */
1278	fl4 = &inet->cork.fl.u.ip4;
1279	rt = ip_route_connect(fl4, daddr, 0, sk->sk_bound_dev_if,
1280			      sk->sk_protocol, inet->inet_sport,
1281			      inet->inet_dport, sk);
1282	if (IS_ERR(rt))
1283		return PTR_ERR(rt);
1284
1285	new_saddr = fl4->saddr;
1286
1287	if (new_saddr == old_saddr) {
1288		sk_setup_caps(sk, &rt->dst);
1289		return 0;
1290	}
1291
1292	err = inet_bhash2_update_saddr(sk, &new_saddr, AF_INET);
1293	if (err) {
1294		ip_rt_put(rt);
1295		return err;
1296	}
1297
1298	sk_setup_caps(sk, &rt->dst);
1299
1300	if (READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) > 1) {
1301		pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1302			__func__, &old_saddr, &new_saddr);
1303	}
1304
1305	/*
1306	 * XXX The only one ugly spot where we need to
1307	 * XXX really change the sockets identity after
1308	 * XXX it has entered the hashes. -DaveM
1309	 *
1310	 * Besides that, it does not check for connection
1311	 * uniqueness. Wait for troubles.
1312	 */
1313	return __sk_prot_rehash(sk);
1314}
1315
1316int inet_sk_rebuild_header(struct sock *sk)
1317{
1318	struct rtable *rt = dst_rtable(__sk_dst_check(sk, 0));
1319	struct inet_sock *inet = inet_sk(sk);
1320	struct flowi4 *fl4;
1321	int err;
1322
1323	/* Route is OK, nothing to do. */
1324	if (rt)
1325		return 0;
1326
1327	/* Reroute. */
1328	fl4 = &inet->cork.fl.u.ip4;
1329	inet_sk_init_flowi4(inet, fl4);
1330	rt = ip_route_output_flow(sock_net(sk), fl4, sk);
1331	if (!IS_ERR(rt)) {
1332		err = 0;
1333		sk_setup_caps(sk, &rt->dst);
1334	} else {
1335		err = PTR_ERR(rt);
1336
1337		/* Routing failed... */
1338		sk->sk_route_caps = 0;
1339
1340		if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) ||
1341		    sk->sk_state != TCP_SYN_SENT ||
1342		    (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1343		    (err = inet_sk_reselect_saddr(sk)) != 0)
1344			WRITE_ONCE(sk->sk_err_soft, -err);
1345	}
1346
1347	return err;
1348}
1349EXPORT_SYMBOL(inet_sk_rebuild_header);
1350
1351void inet_sk_set_state(struct sock *sk, int state)
1352{
1353	trace_inet_sock_set_state(sk, sk->sk_state, state);
1354	sk->sk_state = state;
1355}
1356EXPORT_SYMBOL(inet_sk_set_state);
1357
1358void inet_sk_state_store(struct sock *sk, int newstate)
1359{
1360	trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1361	smp_store_release(&sk->sk_state, newstate);
1362}
1363
1364struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1365				 netdev_features_t features)
1366{
1367	bool udpfrag = false, fixedid = false, gso_partial, encap;
1368	struct sk_buff *segs = ERR_PTR(-EINVAL);
1369	const struct net_offload *ops;
1370	unsigned int offset = 0;
1371	struct iphdr *iph;
1372	int proto, tot_len;
1373	int nhoff;
1374	int ihl;
1375	int id;
1376
1377	skb_reset_network_header(skb);
1378	nhoff = skb_network_header(skb) - skb_mac_header(skb);
1379	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1380		goto out;
1381
1382	iph = ip_hdr(skb);
1383	ihl = iph->ihl * 4;
1384	if (ihl < sizeof(*iph))
1385		goto out;
1386
1387	id = ntohs(iph->id);
1388	proto = iph->protocol;
1389
1390	/* Warning: after this point, iph might be no longer valid */
1391	if (unlikely(!pskb_may_pull(skb, ihl)))
1392		goto out;
1393	__skb_pull(skb, ihl);
1394
1395	encap = SKB_GSO_CB(skb)->encap_level > 0;
1396	if (encap)
1397		features &= skb->dev->hw_enc_features;
1398	SKB_GSO_CB(skb)->encap_level += ihl;
1399
1400	skb_reset_transport_header(skb);
1401
1402	segs = ERR_PTR(-EPROTONOSUPPORT);
1403
1404	fixedid = !!(skb_shinfo(skb)->gso_type & (SKB_GSO_TCP_FIXEDID << encap));
1405
1406	if (!skb->encapsulation || encap)
1407		udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1408
1409	ops = rcu_dereference(inet_offloads[proto]);
1410	if (likely(ops && ops->callbacks.gso_segment)) {
1411		segs = ops->callbacks.gso_segment(skb, features);
1412		if (!segs)
1413			skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
1414	}
1415
1416	if (IS_ERR_OR_NULL(segs))
1417		goto out;
1418
1419	gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1420
1421	skb = segs;
1422	do {
1423		iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1424		if (udpfrag) {
1425			iph->frag_off = htons(offset >> 3);
1426			if (skb->next)
1427				iph->frag_off |= htons(IP_MF);
1428			offset += skb->len - nhoff - ihl;
1429			tot_len = skb->len - nhoff;
1430		} else if (skb_is_gso(skb)) {
1431			if (!fixedid) {
1432				iph->id = htons(id);
1433				id += skb_shinfo(skb)->gso_segs;
1434			}
1435
1436			if (gso_partial)
1437				tot_len = skb_shinfo(skb)->gso_size +
1438					  SKB_GSO_CB(skb)->data_offset +
1439					  skb->head - (unsigned char *)iph;
1440			else
1441				tot_len = skb->len - nhoff;
1442		} else {
1443			if (!fixedid)
1444				iph->id = htons(id++);
1445			tot_len = skb->len - nhoff;
1446		}
1447		iph->tot_len = htons(tot_len);
1448		ip_send_check(iph);
1449		if (encap)
1450			skb_reset_inner_headers(skb);
1451		skb->network_header = (u8 *)iph - skb->head;
1452		skb_reset_mac_len(skb);
1453	} while ((skb = skb->next));
1454
1455out:
1456	return segs;
1457}
1458
1459static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1460					netdev_features_t features)
1461{
1462	if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1463		return ERR_PTR(-EINVAL);
1464
1465	return inet_gso_segment(skb, features);
1466}
1467
1468struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1469{
1470	const struct net_offload *ops;
1471	struct sk_buff *pp = NULL;
1472	const struct iphdr *iph;
1473	struct sk_buff *p;
1474	unsigned int hlen;
1475	unsigned int off;
1476	int flush = 1;
1477	int proto;
1478
1479	off = skb_gro_offset(skb);
1480	hlen = off + sizeof(*iph);
1481	iph = skb_gro_header(skb, hlen, off);
1482	if (unlikely(!iph))
1483		goto out;
1484
1485	proto = iph->protocol;
1486
1487	ops = rcu_dereference(inet_offloads[proto]);
1488	if (!ops || !ops->callbacks.gro_receive)
1489		goto out;
1490
1491	if (*(u8 *)iph != 0x45)
1492		goto out;
1493
1494	if (ip_is_fragment(iph))
1495		goto out;
1496
1497	if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1498		goto out;
1499
1500	NAPI_GRO_CB(skb)->proto = proto;
1501	flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (ntohl(*(__be32 *)&iph->id) & ~IP_DF));
1502
1503	list_for_each_entry(p, head, list) {
1504		struct iphdr *iph2;
1505
1506		if (!NAPI_GRO_CB(p)->same_flow)
1507			continue;
1508
1509		iph2 = (struct iphdr *)(p->data + off);
1510		/* The above works because, with the exception of the top
1511		 * (inner most) layer, we only aggregate pkts with the same
1512		 * hdr length so all the hdrs we'll need to verify will start
1513		 * at the same offset.
1514		 */
1515		if ((iph->protocol ^ iph2->protocol) |
1516		    ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1517		    ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1518			NAPI_GRO_CB(p)->same_flow = 0;
1519			continue;
1520		}
1521	}
1522
1523	NAPI_GRO_CB(skb)->flush |= flush;
1524	NAPI_GRO_CB(skb)->network_offsets[NAPI_GRO_CB(skb)->encap_mark] = off;
1525
1526	/* Note : No need to call skb_gro_postpull_rcsum() here,
1527	 * as we already checked checksum over ipv4 header was 0
1528	 */
1529	skb_gro_pull(skb, sizeof(*iph));
1530	skb_set_transport_header(skb, skb_gro_offset(skb));
1531
1532	pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1533				       ops->callbacks.gro_receive, head, skb);
1534
1535out:
1536	skb_gro_flush_final(skb, pp, flush);
1537
1538	return pp;
1539}
1540
1541static struct sk_buff *ipip_gro_receive(struct list_head *head,
1542					struct sk_buff *skb)
1543{
1544	if (NAPI_GRO_CB(skb)->encap_mark) {
1545		NAPI_GRO_CB(skb)->flush = 1;
1546		return NULL;
1547	}
1548
1549	NAPI_GRO_CB(skb)->encap_mark = 1;
1550
1551	return inet_gro_receive(head, skb);
1552}
1553
1554#define SECONDS_PER_DAY	86400
1555
1556/* inet_current_timestamp - Return IP network timestamp
1557 *
1558 * Return milliseconds since midnight in network byte order.
1559 */
1560__be32 inet_current_timestamp(void)
1561{
1562	u32 secs;
1563	u32 msecs;
1564	struct timespec64 ts;
1565
1566	ktime_get_real_ts64(&ts);
1567
1568	/* Get secs since midnight. */
1569	(void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1570	/* Convert to msecs. */
1571	msecs = secs * MSEC_PER_SEC;
1572	/* Convert nsec to msec. */
1573	msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1574
1575	/* Convert to network byte order. */
1576	return htonl(msecs);
1577}
1578EXPORT_SYMBOL(inet_current_timestamp);
1579
1580int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1581{
1582	unsigned int family = READ_ONCE(sk->sk_family);
1583
1584	if (family == AF_INET)
1585		return ip_recv_error(sk, msg, len, addr_len);
1586#if IS_ENABLED(CONFIG_IPV6)
1587	if (family == AF_INET6)
1588		return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1589#endif
1590	return -EINVAL;
1591}
1592EXPORT_SYMBOL(inet_recv_error);
1593
1594int inet_gro_complete(struct sk_buff *skb, int nhoff)
1595{
1596	struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1597	const struct net_offload *ops;
1598	__be16 totlen = iph->tot_len;
1599	int proto = iph->protocol;
1600	int err = -ENOSYS;
1601
1602	if (skb->encapsulation) {
1603		skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1604		skb_set_inner_network_header(skb, nhoff);
1605	}
1606
1607	iph_set_totlen(iph, skb->len - nhoff);
1608	csum_replace2(&iph->check, totlen, iph->tot_len);
1609
1610	ops = rcu_dereference(inet_offloads[proto]);
1611	if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1612		goto out;
1613
1614	/* Only need to add sizeof(*iph) to get to the next hdr below
1615	 * because any hdr with option will have been flushed in
1616	 * inet_gro_receive().
1617	 */
1618	err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1619			      tcp4_gro_complete, udp4_gro_complete,
1620			      skb, nhoff + sizeof(*iph));
1621
1622out:
1623	return err;
1624}
1625
1626static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1627{
1628	skb->encapsulation = 1;
1629	skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1630	return inet_gro_complete(skb, nhoff);
1631}
1632
1633int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1634			 unsigned short type, unsigned char protocol,
1635			 struct net *net)
1636{
1637	struct socket *sock;
1638	int rc = sock_create_kern(net, family, type, protocol, &sock);
1639
1640	if (rc == 0) {
1641		*sk = sock->sk;
1642		(*sk)->sk_allocation = GFP_ATOMIC;
1643		(*sk)->sk_use_task_frag = false;
1644		/*
1645		 * Unhash it so that IP input processing does not even see it,
1646		 * we do not wish this socket to see incoming packets.
1647		 */
1648		(*sk)->sk_prot->unhash(*sk);
1649	}
1650	return rc;
1651}
1652EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1653
1654unsigned long snmp_fold_field(void __percpu *mib, int offt)
1655{
1656	unsigned long res = 0;
1657	int i;
1658
1659	for_each_possible_cpu(i)
1660		res += snmp_get_cpu_field(mib, i, offt);
1661	return res;
1662}
1663EXPORT_SYMBOL_GPL(snmp_fold_field);
1664
1665#if BITS_PER_LONG==32
1666
1667u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1668			 size_t syncp_offset)
1669{
1670	void *bhptr;
1671	struct u64_stats_sync *syncp;
1672	u64 v;
1673	unsigned int start;
1674
1675	bhptr = per_cpu_ptr(mib, cpu);
1676	syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1677	do {
1678		start = u64_stats_fetch_begin(syncp);
1679		v = *(((u64 *)bhptr) + offt);
1680	} while (u64_stats_fetch_retry(syncp, start));
1681
1682	return v;
1683}
1684EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1685
1686u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1687{
1688	u64 res = 0;
1689	int cpu;
1690
1691	for_each_possible_cpu(cpu) {
1692		res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1693	}
1694	return res;
1695}
1696EXPORT_SYMBOL_GPL(snmp_fold_field64);
1697#endif
1698
1699#ifdef CONFIG_IP_MULTICAST
1700static const struct net_protocol igmp_protocol = {
1701	.handler =	igmp_rcv,
1702};
1703#endif
1704
1705static const struct net_protocol icmp_protocol = {
1706	.handler =	icmp_rcv,
1707	.err_handler =	icmp_err,
1708	.no_policy =	1,
1709};
1710
1711static __net_init int ipv4_mib_init_net(struct net *net)
1712{
1713	int i;
1714
1715	net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1716	if (!net->mib.tcp_statistics)
1717		goto err_tcp_mib;
1718	net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1719	if (!net->mib.ip_statistics)
1720		goto err_ip_mib;
1721
1722	for_each_possible_cpu(i) {
1723		struct ipstats_mib *af_inet_stats;
1724		af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1725		u64_stats_init(&af_inet_stats->syncp);
1726	}
1727
1728	net->mib.net_statistics = alloc_percpu(struct linux_mib);
1729	if (!net->mib.net_statistics)
1730		goto err_net_mib;
1731	net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1732	if (!net->mib.udp_statistics)
1733		goto err_udp_mib;
1734	net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1735	if (!net->mib.udplite_statistics)
1736		goto err_udplite_mib;
1737	net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1738	if (!net->mib.icmp_statistics)
1739		goto err_icmp_mib;
1740	net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1741					      GFP_KERNEL);
1742	if (!net->mib.icmpmsg_statistics)
1743		goto err_icmpmsg_mib;
1744
1745	tcp_mib_init(net);
1746	return 0;
1747
1748err_icmpmsg_mib:
1749	free_percpu(net->mib.icmp_statistics);
1750err_icmp_mib:
1751	free_percpu(net->mib.udplite_statistics);
1752err_udplite_mib:
1753	free_percpu(net->mib.udp_statistics);
1754err_udp_mib:
1755	free_percpu(net->mib.net_statistics);
1756err_net_mib:
1757	free_percpu(net->mib.ip_statistics);
1758err_ip_mib:
1759	free_percpu(net->mib.tcp_statistics);
1760err_tcp_mib:
1761	return -ENOMEM;
1762}
1763
1764static __net_exit void ipv4_mib_exit_net(struct net *net)
1765{
1766	kfree(net->mib.icmpmsg_statistics);
1767	free_percpu(net->mib.icmp_statistics);
1768	free_percpu(net->mib.udplite_statistics);
1769	free_percpu(net->mib.udp_statistics);
1770	free_percpu(net->mib.net_statistics);
1771	free_percpu(net->mib.ip_statistics);
1772	free_percpu(net->mib.tcp_statistics);
1773#ifdef CONFIG_MPTCP
1774	/* allocated on demand, see mptcp_init_sock() */
1775	free_percpu(net->mib.mptcp_statistics);
1776#endif
1777}
1778
1779static __net_initdata struct pernet_operations ipv4_mib_ops = {
1780	.init = ipv4_mib_init_net,
1781	.exit = ipv4_mib_exit_net,
1782};
1783
1784static int __init init_ipv4_mibs(void)
1785{
1786	return register_pernet_subsys(&ipv4_mib_ops);
1787}
1788
1789static __net_init int inet_init_net(struct net *net)
1790{
1791	/*
1792	 * Set defaults for local port range
1793	 */
1794	net->ipv4.ip_local_ports.range = 60999u << 16 | 32768u;
1795
1796	seqlock_init(&net->ipv4.ping_group_range.lock);
1797	/*
1798	 * Sane defaults - nobody may create ping sockets.
1799	 * Boot scripts should set this to distro-specific group.
1800	 */
1801	net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1802	net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1803
1804	/* Default values for sysctl-controlled parameters.
1805	 * We set them here, in case sysctl is not compiled.
1806	 */
1807	net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1808	net->ipv4.sysctl_ip_fwd_update_priority = 1;
1809	net->ipv4.sysctl_ip_dynaddr = 0;
1810	net->ipv4.sysctl_ip_early_demux = 1;
1811	net->ipv4.sysctl_udp_early_demux = 1;
1812	net->ipv4.sysctl_tcp_early_demux = 1;
1813	net->ipv4.sysctl_nexthop_compat_mode = 1;
1814#ifdef CONFIG_SYSCTL
1815	net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1816#endif
1817
1818	/* Some igmp sysctl, whose values are always used */
1819	net->ipv4.sysctl_igmp_max_memberships = 20;
1820	net->ipv4.sysctl_igmp_max_msf = 10;
1821	/* IGMP reports for link-local multicast groups are enabled by default */
1822	net->ipv4.sysctl_igmp_llm_reports = 1;
1823	net->ipv4.sysctl_igmp_qrv = 2;
1824
1825	net->ipv4.sysctl_fib_notify_on_flag_change = 0;
1826
1827	return 0;
1828}
1829
1830static __net_initdata struct pernet_operations af_inet_ops = {
1831	.init = inet_init_net,
1832};
1833
1834static int __init init_inet_pernet_ops(void)
1835{
1836	return register_pernet_subsys(&af_inet_ops);
1837}
1838
1839static int ipv4_proc_init(void);
1840
1841/*
1842 *	IP protocol layer initialiser
1843 */
1844
1845
1846static const struct net_offload ipip_offload = {
1847	.callbacks = {
1848		.gso_segment	= ipip_gso_segment,
1849		.gro_receive	= ipip_gro_receive,
1850		.gro_complete	= ipip_gro_complete,
1851	},
1852};
1853
1854static int __init ipip_offload_init(void)
1855{
1856	return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1857}
1858
1859static int __init ipv4_offload_init(void)
1860{
1861	/*
1862	 * Add offloads
1863	 */
1864	if (udpv4_offload_init() < 0)
1865		pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1866	if (tcpv4_offload_init() < 0)
1867		pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1868	if (ipip_offload_init() < 0)
1869		pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1870
1871	net_hotdata.ip_packet_offload = (struct packet_offload) {
1872		.type = cpu_to_be16(ETH_P_IP),
1873		.callbacks = {
1874			.gso_segment = inet_gso_segment,
1875			.gro_receive = inet_gro_receive,
1876			.gro_complete = inet_gro_complete,
1877		},
1878	};
1879	dev_add_offload(&net_hotdata.ip_packet_offload);
1880	return 0;
1881}
1882
1883fs_initcall(ipv4_offload_init);
1884
1885static struct packet_type ip_packet_type __read_mostly = {
1886	.type = cpu_to_be16(ETH_P_IP),
1887	.func = ip_rcv,
1888	.list_func = ip_list_rcv,
1889};
1890
1891static int __init inet_init(void)
1892{
1893	struct inet_protosw *q;
1894	struct list_head *r;
1895	int rc;
1896
1897	sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1898
1899	raw_hashinfo_init(&raw_v4_hashinfo);
1900
1901	rc = proto_register(&tcp_prot, 1);
1902	if (rc)
1903		goto out;
1904
1905	rc = proto_register(&udp_prot, 1);
1906	if (rc)
1907		goto out_unregister_tcp_proto;
1908
1909	rc = proto_register(&raw_prot, 1);
1910	if (rc)
1911		goto out_unregister_udp_proto;
1912
1913	rc = proto_register(&ping_prot, 1);
1914	if (rc)
1915		goto out_unregister_raw_proto;
1916
1917	/*
1918	 *	Tell SOCKET that we are alive...
1919	 */
1920
1921	(void)sock_register(&inet_family_ops);
1922
1923#ifdef CONFIG_SYSCTL
1924	ip_static_sysctl_init();
1925#endif
1926
1927	/*
1928	 *	Add all the base protocols.
1929	 */
1930
1931	if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1932		pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1933
1934	net_hotdata.udp_protocol = (struct net_protocol) {
1935		.handler =	udp_rcv,
1936		.err_handler =	udp_err,
1937		.no_policy =	1,
1938	};
1939	if (inet_add_protocol(&net_hotdata.udp_protocol, IPPROTO_UDP) < 0)
1940		pr_crit("%s: Cannot add UDP protocol\n", __func__);
1941
1942	net_hotdata.tcp_protocol = (struct net_protocol) {
1943		.handler	=	tcp_v4_rcv,
1944		.err_handler	=	tcp_v4_err,
1945		.no_policy	=	1,
1946		.icmp_strict_tag_validation = 1,
1947	};
1948	if (inet_add_protocol(&net_hotdata.tcp_protocol, IPPROTO_TCP) < 0)
1949		pr_crit("%s: Cannot add TCP protocol\n", __func__);
1950#ifdef CONFIG_IP_MULTICAST
1951	if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1952		pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1953#endif
1954
1955	/* Register the socket-side information for inet_create. */
1956	for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1957		INIT_LIST_HEAD(r);
1958
1959	for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1960		inet_register_protosw(q);
1961
1962	/*
1963	 *	Set the ARP module up
1964	 */
1965
1966	arp_init();
1967
1968	/*
1969	 *	Set the IP module up
1970	 */
1971
1972	ip_init();
1973
1974	/* Initialise per-cpu ipv4 mibs */
1975	if (init_ipv4_mibs())
1976		panic("%s: Cannot init ipv4 mibs\n", __func__);
1977
1978	/* Setup TCP slab cache for open requests. */
1979	tcp_init();
1980
1981	/* Setup UDP memory threshold */
1982	udp_init();
1983
1984	/* Add UDP-Lite (RFC 3828) */
1985	udplite4_register();
1986
1987	raw_init();
1988
1989	ping_init();
1990
1991	/*
1992	 *	Set the ICMP layer up
1993	 */
1994
1995	if (icmp_init() < 0)
1996		panic("Failed to create the ICMP control socket.\n");
1997
1998	/*
1999	 *	Initialise the multicast router
2000	 */
2001#if defined(CONFIG_IP_MROUTE)
2002	if (ip_mr_init())
2003		pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2004#endif
2005
2006	if (init_inet_pernet_ops())
2007		pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2008
2009	ipv4_proc_init();
2010
2011	ipfrag_init();
2012
2013	dev_add_pack(&ip_packet_type);
2014
2015	ip_tunnel_core_init();
2016
2017	rc = 0;
2018out:
2019	return rc;
2020out_unregister_raw_proto:
2021	proto_unregister(&raw_prot);
2022out_unregister_udp_proto:
2023	proto_unregister(&udp_prot);
2024out_unregister_tcp_proto:
2025	proto_unregister(&tcp_prot);
2026	goto out;
2027}
2028
2029fs_initcall(inet_init);
2030
2031/* ------------------------------------------------------------------------ */
2032
2033#ifdef CONFIG_PROC_FS
2034static int __init ipv4_proc_init(void)
2035{
2036	int rc = 0;
2037
2038	if (raw_proc_init())
2039		goto out_raw;
2040	if (tcp4_proc_init())
2041		goto out_tcp;
2042	if (udp4_proc_init())
2043		goto out_udp;
2044	if (ping_proc_init())
2045		goto out_ping;
2046	if (ip_misc_proc_init())
2047		goto out_misc;
2048out:
2049	return rc;
2050out_misc:
2051	ping_proc_exit();
2052out_ping:
2053	udp4_proc_exit();
2054out_udp:
2055	tcp4_proc_exit();
2056out_tcp:
2057	raw_proc_exit();
2058out_raw:
2059	rc = -ENOMEM;
2060	goto out;
2061}
2062
2063#else /* CONFIG_PROC_FS */
2064static int __init ipv4_proc_init(void)
2065{
2066	return 0;
2067}
2068#endif /* CONFIG_PROC_FS */