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/udp.h>
 106#include <net/udplite.h>
 107#include <net/ping.h>
 108#include <linux/skbuff.h>
 109#include <net/sock.h>
 110#include <net/raw.h>
 111#include <net/icmp.h>
 112#include <net/inet_common.h>
 113#include <net/ip_tunnels.h>
 114#include <net/xfrm.h>
 115#include <net/net_namespace.h>
 116#include <net/secure_seq.h>
 117#ifdef CONFIG_IP_MROUTE
 118#include <linux/mroute.h>
 119#endif
 120#include <net/l3mdev.h>
 121#include <net/compat.h>
 122#include <net/rps.h>
 123
 124#include <trace/events/sock.h>
 125
 126/* The inetsw table contains everything that inet_create needs to
 127 * build a new socket.
 128 */
 129static struct list_head inetsw[SOCK_MAX];
 130static DEFINE_SPINLOCK(inetsw_lock);
 131
 132/* New destruction routine */
 133
 134void inet_sock_destruct(struct sock *sk)
 135{
 136	struct inet_sock *inet = inet_sk(sk);
 137
 138	__skb_queue_purge(&sk->sk_receive_queue);
 139	__skb_queue_purge(&sk->sk_error_queue);
 140
 141	sk_mem_reclaim_final(sk);
 142
 143	if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
 144		pr_err("Attempt to release TCP socket in state %d %p\n",
 145		       sk->sk_state, sk);
 146		return;
 147	}
 148	if (!sock_flag(sk, SOCK_DEAD)) {
 149		pr_err("Attempt to release alive inet socket %p\n", sk);
 150		return;
 151	}
 152
 153	WARN_ON_ONCE(atomic_read(&sk->sk_rmem_alloc));
 154	WARN_ON_ONCE(refcount_read(&sk->sk_wmem_alloc));
 155	WARN_ON_ONCE(sk->sk_wmem_queued);
 156	WARN_ON_ONCE(sk_forward_alloc_get(sk));
 157
 158	kfree(rcu_dereference_protected(inet->inet_opt, 1));
 159	dst_release(rcu_dereference_protected(sk->sk_dst_cache, 1));
 160	dst_release(rcu_dereference_protected(sk->sk_rx_dst, 1));
 161}
 162EXPORT_SYMBOL(inet_sock_destruct);
 163
 164/*
 165 *	The routines beyond this point handle the behaviour of an AF_INET
 166 *	socket object. Mostly it punts to the subprotocols of IP to do
 167 *	the work.
 168 */
 169
 170/*
 171 *	Automatically bind an unbound socket.
 172 */
 173
 174static int inet_autobind(struct sock *sk)
 175{
 176	struct inet_sock *inet;
 177	/* We may need to bind the socket. */
 178	lock_sock(sk);
 179	inet = inet_sk(sk);
 180	if (!inet->inet_num) {
 181		if (sk->sk_prot->get_port(sk, 0)) {
 182			release_sock(sk);
 183			return -EAGAIN;
 184		}
 185		inet->inet_sport = htons(inet->inet_num);
 186	}
 187	release_sock(sk);
 188	return 0;
 189}
 190
 191int __inet_listen_sk(struct sock *sk, int backlog)
 192{
 193	unsigned char old_state = sk->sk_state;
 194	int err, tcp_fastopen;
 195
 196	if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
 197		return -EINVAL;
 198
 199	WRITE_ONCE(sk->sk_max_ack_backlog, backlog);
 200	/* Really, if the socket is already in listen state
 201	 * we can only allow the backlog to be adjusted.
 202	 */
 203	if (old_state != TCP_LISTEN) {
 204		/* Enable TFO w/o requiring TCP_FASTOPEN socket option.
 205		 * Note that only TCP sockets (SOCK_STREAM) will reach here.
 206		 * Also fastopen backlog may already been set via the option
 207		 * because the socket was in TCP_LISTEN state previously but
 208		 * was shutdown() rather than close().
 209		 */
 210		tcp_fastopen = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fastopen);
 211		if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
 212		    (tcp_fastopen & TFO_SERVER_ENABLE) &&
 213		    !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
 214			fastopen_queue_tune(sk, backlog);
 215			tcp_fastopen_init_key_once(sock_net(sk));
 216		}
 217
 218		err = inet_csk_listen_start(sk);
 219		if (err)
 220			return err;
 221
 222		tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL);
 223	}
 224	return 0;
 225}
 226
 227/*
 228 *	Move a socket into listening state.
 229 */
 230int inet_listen(struct socket *sock, int backlog)
 231{
 232	struct sock *sk = sock->sk;
 233	int err = -EINVAL;
 234
 235	lock_sock(sk);
 236
 237	if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
 238		goto out;
 239
 240	err = __inet_listen_sk(sk, backlog);
 241
 242out:
 243	release_sock(sk);
 244	return err;
 245}
 246EXPORT_SYMBOL(inet_listen);
 247
 248/*
 249 *	Create an inet socket.
 250 */
 251
 252static int inet_create(struct net *net, struct socket *sock, int protocol,
 253		       int kern)
 254{
 255	struct sock *sk;
 256	struct inet_protosw *answer;
 257	struct inet_sock *inet;
 258	struct proto *answer_prot;
 259	unsigned char answer_flags;
 260	int try_loading_module = 0;
 261	int err;
 262
 263	if (protocol < 0 || protocol >= IPPROTO_MAX)
 264		return -EINVAL;
 265
 266	sock->state = SS_UNCONNECTED;
 267
 268	/* Look for the requested type/protocol pair. */
 269lookup_protocol:
 270	err = -ESOCKTNOSUPPORT;
 271	rcu_read_lock();
 272	list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
 273
 274		err = 0;
 275		/* Check the non-wild match. */
 276		if (protocol == answer->protocol) {
 277			if (protocol != IPPROTO_IP)
 278				break;
 279		} else {
 280			/* Check for the two wild cases. */
 281			if (IPPROTO_IP == protocol) {
 282				protocol = answer->protocol;
 283				break;
 284			}
 285			if (IPPROTO_IP == answer->protocol)
 286				break;
 287		}
 288		err = -EPROTONOSUPPORT;
 289	}
 290
 291	if (unlikely(err)) {
 292		if (try_loading_module < 2) {
 293			rcu_read_unlock();
 294			/*
 295			 * Be more specific, e.g. net-pf-2-proto-132-type-1
 296			 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
 297			 */
 298			if (++try_loading_module == 1)
 299				request_module("net-pf-%d-proto-%d-type-%d",
 300					       PF_INET, protocol, sock->type);
 301			/*
 302			 * Fall back to generic, e.g. net-pf-2-proto-132
 303			 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
 304			 */
 305			else
 306				request_module("net-pf-%d-proto-%d",
 307					       PF_INET, protocol);
 308			goto lookup_protocol;
 309		} else
 310			goto out_rcu_unlock;
 311	}
 312
 313	err = -EPERM;
 314	if (sock->type == SOCK_RAW && !kern &&
 315	    !ns_capable(net->user_ns, CAP_NET_RAW))
 316		goto out_rcu_unlock;
 317
 318	sock->ops = answer->ops;
 319	answer_prot = answer->prot;
 320	answer_flags = answer->flags;
 321	rcu_read_unlock();
 322
 323	WARN_ON(!answer_prot->slab);
 324
 325	err = -ENOMEM;
 326	sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
 327	if (!sk)
 328		goto out;
 329
 330	err = 0;
 331	if (INET_PROTOSW_REUSE & answer_flags)
 332		sk->sk_reuse = SK_CAN_REUSE;
 333
 334	if (INET_PROTOSW_ICSK & answer_flags)
 335		inet_init_csk_locks(sk);
 336
 337	inet = inet_sk(sk);
 338	inet_assign_bit(IS_ICSK, sk, INET_PROTOSW_ICSK & answer_flags);
 339
 340	inet_clear_bit(NODEFRAG, sk);
 341
 342	if (SOCK_RAW == sock->type) {
 343		inet->inet_num = protocol;
 344		if (IPPROTO_RAW == protocol)
 345			inet_set_bit(HDRINCL, sk);
 346	}
 347
 348	if (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc))
 349		inet->pmtudisc = IP_PMTUDISC_DONT;
 350	else
 351		inet->pmtudisc = IP_PMTUDISC_WANT;
 352
 353	atomic_set(&inet->inet_id, 0);
 354
 355	sock_init_data(sock, sk);
 356
 357	sk->sk_destruct	   = inet_sock_destruct;
 358	sk->sk_protocol	   = protocol;
 359	sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
 360	sk->sk_txrehash = READ_ONCE(net->core.sysctl_txrehash);
 361
 362	inet->uc_ttl	= -1;
 363	inet_set_bit(MC_LOOP, sk);
 364	inet->mc_ttl	= 1;
 365	inet_set_bit(MC_ALL, sk);
 366	inet->mc_index	= 0;
 367	inet->mc_list	= NULL;
 368	inet->rcv_tos	= 0;
 369
 370	if (inet->inet_num) {
 371		/* It assumes that any protocol which allows
 372		 * the user to assign a number at socket
 373		 * creation time automatically
 374		 * shares.
 375		 */
 376		inet->inet_sport = htons(inet->inet_num);
 377		/* Add to protocol hash chains. */
 378		err = sk->sk_prot->hash(sk);
 379		if (err)
 380			goto out_sk_release;
 
 
 381	}
 382
 383	if (sk->sk_prot->init) {
 384		err = sk->sk_prot->init(sk);
 385		if (err)
 386			goto out_sk_release;
 
 
 387	}
 388
 389	if (!kern) {
 390		err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
 391		if (err)
 392			goto out_sk_release;
 
 
 393	}
 394out:
 395	return err;
 396out_rcu_unlock:
 397	rcu_read_unlock();
 398	goto out;
 399out_sk_release:
 400	sk_common_release(sk);
 401	sock->sk = NULL;
 402	goto out;
 403}
 404
 405
 406/*
 407 *	The peer socket should always be NULL (or else). When we call this
 408 *	function we are destroying the object and from then on nobody
 409 *	should refer to it.
 410 */
 411int inet_release(struct socket *sock)
 412{
 413	struct sock *sk = sock->sk;
 414
 415	if (sk) {
 416		long timeout;
 417
 418		if (!sk->sk_kern_sock)
 419			BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk);
 420
 421		/* Applications forget to leave groups before exiting */
 422		ip_mc_drop_socket(sk);
 423
 424		/* If linger is set, we don't return until the close
 425		 * is complete.  Otherwise we return immediately. The
 426		 * actually closing is done the same either way.
 427		 *
 428		 * If the close is due to the process exiting, we never
 429		 * linger..
 430		 */
 431		timeout = 0;
 432		if (sock_flag(sk, SOCK_LINGER) &&
 433		    !(current->flags & PF_EXITING))
 434			timeout = sk->sk_lingertime;
 435		sk->sk_prot->close(sk, timeout);
 436		sock->sk = NULL;
 437	}
 438	return 0;
 439}
 440EXPORT_SYMBOL(inet_release);
 441
 442int inet_bind_sk(struct sock *sk, struct sockaddr *uaddr, int addr_len)
 443{
 444	u32 flags = BIND_WITH_LOCK;
 445	int err;
 446
 447	/* If the socket has its own bind function then use it. (RAW) */
 448	if (sk->sk_prot->bind) {
 449		return sk->sk_prot->bind(sk, uaddr, addr_len);
 450	}
 451	if (addr_len < sizeof(struct sockaddr_in))
 452		return -EINVAL;
 453
 454	/* BPF prog is run before any checks are done so that if the prog
 455	 * changes context in a wrong way it will be caught.
 456	 */
 457	err = BPF_CGROUP_RUN_PROG_INET_BIND_LOCK(sk, uaddr, &addr_len,
 458						 CGROUP_INET4_BIND, &flags);
 459	if (err)
 460		return err;
 461
 462	return __inet_bind(sk, uaddr, addr_len, flags);
 463}
 464
 465int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 466{
 467	return inet_bind_sk(sock->sk, uaddr, addr_len);
 468}
 469EXPORT_SYMBOL(inet_bind);
 470
 471int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
 472		u32 flags)
 473{
 474	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
 475	struct inet_sock *inet = inet_sk(sk);
 476	struct net *net = sock_net(sk);
 477	unsigned short snum;
 478	int chk_addr_ret;
 479	u32 tb_id = RT_TABLE_LOCAL;
 480	int err;
 481
 482	if (addr->sin_family != AF_INET) {
 483		/* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
 484		 * only if s_addr is INADDR_ANY.
 485		 */
 486		err = -EAFNOSUPPORT;
 487		if (addr->sin_family != AF_UNSPEC ||
 488		    addr->sin_addr.s_addr != htonl(INADDR_ANY))
 489			goto out;
 490	}
 491
 492	tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
 493	chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
 494
 495	/* Not specified by any standard per-se, however it breaks too
 496	 * many applications when removed.  It is unfortunate since
 497	 * allowing applications to make a non-local bind solves
 498	 * several problems with systems using dynamic addressing.
 499	 * (ie. your servers still start up even if your ISDN link
 500	 *  is temporarily down)
 501	 */
 502	err = -EADDRNOTAVAIL;
 503	if (!inet_addr_valid_or_nonlocal(net, inet, addr->sin_addr.s_addr,
 504	                                 chk_addr_ret))
 505		goto out;
 506
 507	snum = ntohs(addr->sin_port);
 508	err = -EACCES;
 509	if (!(flags & BIND_NO_CAP_NET_BIND_SERVICE) &&
 510	    snum && inet_port_requires_bind_service(net, snum) &&
 511	    !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
 512		goto out;
 513
 514	/*      We keep a pair of addresses. rcv_saddr is the one
 515	 *      used by hash lookups, and saddr is used for transmit.
 516	 *
 517	 *      In the BSD API these are the same except where it
 518	 *      would be illegal to use them (multicast/broadcast) in
 519	 *      which case the sending device address is used.
 520	 */
 521	if (flags & BIND_WITH_LOCK)
 522		lock_sock(sk);
 523
 524	/* Check these errors (active socket, double bind). */
 525	err = -EINVAL;
 526	if (sk->sk_state != TCP_CLOSE || inet->inet_num)
 527		goto out_release_sock;
 528
 529	inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
 530	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
 531		inet->inet_saddr = 0;  /* Use device */
 532
 533	/* Make sure we are allowed to bind here. */
 534	if (snum || !(inet_test_bit(BIND_ADDRESS_NO_PORT, sk) ||
 535		      (flags & BIND_FORCE_ADDRESS_NO_PORT))) {
 536		err = sk->sk_prot->get_port(sk, snum);
 537		if (err) {
 538			inet->inet_saddr = inet->inet_rcv_saddr = 0;
 539			goto out_release_sock;
 540		}
 541		if (!(flags & BIND_FROM_BPF)) {
 542			err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
 543			if (err) {
 544				inet->inet_saddr = inet->inet_rcv_saddr = 0;
 545				if (sk->sk_prot->put_port)
 546					sk->sk_prot->put_port(sk);
 547				goto out_release_sock;
 548			}
 549		}
 550	}
 551
 552	if (inet->inet_rcv_saddr)
 553		sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
 554	if (snum)
 555		sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
 556	inet->inet_sport = htons(inet->inet_num);
 557	inet->inet_daddr = 0;
 558	inet->inet_dport = 0;
 559	sk_dst_reset(sk);
 560	err = 0;
 561out_release_sock:
 562	if (flags & BIND_WITH_LOCK)
 563		release_sock(sk);
 564out:
 565	return err;
 566}
 567
 568int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
 569		       int addr_len, int flags)
 570{
 571	struct sock *sk = sock->sk;
 572	const struct proto *prot;
 573	int err;
 574
 575	if (addr_len < sizeof(uaddr->sa_family))
 576		return -EINVAL;
 577
 578	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
 579	prot = READ_ONCE(sk->sk_prot);
 580
 581	if (uaddr->sa_family == AF_UNSPEC)
 582		return prot->disconnect(sk, flags);
 583
 584	if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
 585		err = prot->pre_connect(sk, uaddr, addr_len);
 586		if (err)
 587			return err;
 588	}
 589
 590	if (data_race(!inet_sk(sk)->inet_num) && inet_autobind(sk))
 591		return -EAGAIN;
 592	return prot->connect(sk, uaddr, addr_len);
 593}
 594EXPORT_SYMBOL(inet_dgram_connect);
 595
 596static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
 597{
 598	DEFINE_WAIT_FUNC(wait, woken_wake_function);
 599
 600	add_wait_queue(sk_sleep(sk), &wait);
 601	sk->sk_write_pending += writebias;
 602
 603	/* Basic assumption: if someone sets sk->sk_err, he _must_
 604	 * change state of the socket from TCP_SYN_*.
 605	 * Connect() does not allow to get error notifications
 606	 * without closing the socket.
 607	 */
 608	while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 609		release_sock(sk);
 610		timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
 611		lock_sock(sk);
 612		if (signal_pending(current) || !timeo)
 613			break;
 614	}
 615	remove_wait_queue(sk_sleep(sk), &wait);
 616	sk->sk_write_pending -= writebias;
 617	return timeo;
 618}
 619
 620/*
 621 *	Connect to a remote host. There is regrettably still a little
 622 *	TCP 'magic' in here.
 623 */
 624int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 625			  int addr_len, int flags, int is_sendmsg)
 626{
 627	struct sock *sk = sock->sk;
 628	int err;
 629	long timeo;
 630
 631	/*
 632	 * uaddr can be NULL and addr_len can be 0 if:
 633	 * sk is a TCP fastopen active socket and
 634	 * TCP_FASTOPEN_CONNECT sockopt is set and
 635	 * we already have a valid cookie for this socket.
 636	 * In this case, user can call write() after connect().
 637	 * write() will invoke tcp_sendmsg_fastopen() which calls
 638	 * __inet_stream_connect().
 639	 */
 640	if (uaddr) {
 641		if (addr_len < sizeof(uaddr->sa_family))
 642			return -EINVAL;
 643
 644		if (uaddr->sa_family == AF_UNSPEC) {
 645			sk->sk_disconnects++;
 646			err = sk->sk_prot->disconnect(sk, flags);
 647			sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 648			goto out;
 649		}
 650	}
 651
 652	switch (sock->state) {
 653	default:
 654		err = -EINVAL;
 655		goto out;
 656	case SS_CONNECTED:
 657		err = -EISCONN;
 658		goto out;
 659	case SS_CONNECTING:
 660		if (inet_test_bit(DEFER_CONNECT, sk))
 661			err = is_sendmsg ? -EINPROGRESS : -EISCONN;
 662		else
 663			err = -EALREADY;
 664		/* Fall out of switch with err, set for this state */
 665		break;
 666	case SS_UNCONNECTED:
 667		err = -EISCONN;
 668		if (sk->sk_state != TCP_CLOSE)
 669			goto out;
 670
 671		if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
 672			err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
 673			if (err)
 674				goto out;
 675		}
 676
 677		err = sk->sk_prot->connect(sk, uaddr, addr_len);
 678		if (err < 0)
 679			goto out;
 680
 681		sock->state = SS_CONNECTING;
 682
 683		if (!err && inet_test_bit(DEFER_CONNECT, sk))
 684			goto out;
 685
 686		/* Just entered SS_CONNECTING state; the only
 687		 * difference is that return value in non-blocking
 688		 * case is EINPROGRESS, rather than EALREADY.
 689		 */
 690		err = -EINPROGRESS;
 691		break;
 692	}
 693
 694	timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
 695
 696	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 697		int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
 698				tcp_sk(sk)->fastopen_req &&
 699				tcp_sk(sk)->fastopen_req->data ? 1 : 0;
 700		int dis = sk->sk_disconnects;
 701
 702		/* Error code is set above */
 703		if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
 704			goto out;
 705
 706		err = sock_intr_errno(timeo);
 707		if (signal_pending(current))
 708			goto out;
 709
 710		if (dis != sk->sk_disconnects) {
 711			err = -EPIPE;
 712			goto out;
 713		}
 714	}
 715
 716	/* Connection was closed by RST, timeout, ICMP error
 717	 * or another process disconnected us.
 718	 */
 719	if (sk->sk_state == TCP_CLOSE)
 720		goto sock_error;
 721
 722	/* sk->sk_err may be not zero now, if RECVERR was ordered by user
 723	 * and error was received after socket entered established state.
 724	 * Hence, it is handled normally after connect() return successfully.
 725	 */
 726
 727	sock->state = SS_CONNECTED;
 728	err = 0;
 729out:
 730	return err;
 731
 732sock_error:
 733	err = sock_error(sk) ? : -ECONNABORTED;
 734	sock->state = SS_UNCONNECTED;
 735	sk->sk_disconnects++;
 736	if (sk->sk_prot->disconnect(sk, flags))
 737		sock->state = SS_DISCONNECTING;
 738	goto out;
 739}
 740EXPORT_SYMBOL(__inet_stream_connect);
 741
 742int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 743			int addr_len, int flags)
 744{
 745	int err;
 746
 747	lock_sock(sock->sk);
 748	err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
 749	release_sock(sock->sk);
 750	return err;
 751}
 752EXPORT_SYMBOL(inet_stream_connect);
 753
 754void __inet_accept(struct socket *sock, struct socket *newsock, struct sock *newsk)
 755{
 756	sock_rps_record_flow(newsk);
 757	WARN_ON(!((1 << newsk->sk_state) &
 758		  (TCPF_ESTABLISHED | TCPF_SYN_RECV |
 759		   TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2 |
 760		   TCPF_CLOSING | TCPF_CLOSE_WAIT |
 761		   TCPF_CLOSE)));
 762
 763	if (test_bit(SOCK_SUPPORT_ZC, &sock->flags))
 764		set_bit(SOCK_SUPPORT_ZC, &newsock->flags);
 765	sock_graft(newsk, newsock);
 766
 767	newsock->state = SS_CONNECTED;
 768}
 769
 770/*
 771 *	Accept a pending connection. The TCP layer now gives BSD semantics.
 772 */
 773
 774int inet_accept(struct socket *sock, struct socket *newsock,
 775		struct proto_accept_arg *arg)
 776{
 777	struct sock *sk1 = sock->sk, *sk2;
 
 778
 779	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
 780	arg->err = -EINVAL;
 781	sk2 = READ_ONCE(sk1->sk_prot)->accept(sk1, arg);
 782	if (!sk2)
 783		return arg->err;
 784
 785	lock_sock(sk2);
 786	__inet_accept(sock, newsock, sk2);
 787	release_sock(sk2);
 788	return 0;
 789}
 790EXPORT_SYMBOL(inet_accept);
 791
 792/*
 793 *	This does both peername and sockname.
 794 */
 795int inet_getname(struct socket *sock, struct sockaddr *uaddr,
 796		 int peer)
 797{
 798	struct sock *sk		= sock->sk;
 799	struct inet_sock *inet	= inet_sk(sk);
 800	DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
 801	int sin_addr_len = sizeof(*sin);
 802
 803	sin->sin_family = AF_INET;
 804	lock_sock(sk);
 805	if (peer) {
 806		if (!inet->inet_dport ||
 807		    (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
 808		     peer == 1)) {
 809			release_sock(sk);
 810			return -ENOTCONN;
 811		}
 812		sin->sin_port = inet->inet_dport;
 813		sin->sin_addr.s_addr = inet->inet_daddr;
 814		BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin, &sin_addr_len,
 815				       CGROUP_INET4_GETPEERNAME);
 816	} else {
 817		__be32 addr = inet->inet_rcv_saddr;
 818		if (!addr)
 819			addr = inet->inet_saddr;
 820		sin->sin_port = inet->inet_sport;
 821		sin->sin_addr.s_addr = addr;
 822		BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin, &sin_addr_len,
 823				       CGROUP_INET4_GETSOCKNAME);
 824	}
 825	release_sock(sk);
 826	memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 827	return sin_addr_len;
 828}
 829EXPORT_SYMBOL(inet_getname);
 830
 831int inet_send_prepare(struct sock *sk)
 832{
 833	sock_rps_record_flow(sk);
 834
 835	/* We may need to bind the socket. */
 836	if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind &&
 837	    inet_autobind(sk))
 838		return -EAGAIN;
 839
 840	return 0;
 841}
 842EXPORT_SYMBOL_GPL(inet_send_prepare);
 843
 844int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
 845{
 846	struct sock *sk = sock->sk;
 847
 848	if (unlikely(inet_send_prepare(sk)))
 849		return -EAGAIN;
 850
 851	return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
 852			       sk, msg, size);
 853}
 854EXPORT_SYMBOL(inet_sendmsg);
 855
 856void inet_splice_eof(struct socket *sock)
 857{
 858	const struct proto *prot;
 859	struct sock *sk = sock->sk;
 860
 861	if (unlikely(inet_send_prepare(sk)))
 862		return;
 863
 864	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
 865	prot = READ_ONCE(sk->sk_prot);
 866	if (prot->splice_eof)
 867		prot->splice_eof(sock);
 868}
 869EXPORT_SYMBOL_GPL(inet_splice_eof);
 870
 871INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
 872					  size_t, int, int *));
 873int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
 874		 int flags)
 875{
 876	struct sock *sk = sock->sk;
 877	int addr_len = 0;
 878	int err;
 879
 880	if (likely(!(flags & MSG_ERRQUEUE)))
 881		sock_rps_record_flow(sk);
 882
 883	err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
 884			      sk, msg, size, flags, &addr_len);
 885	if (err >= 0)
 886		msg->msg_namelen = addr_len;
 887	return err;
 888}
 889EXPORT_SYMBOL(inet_recvmsg);
 890
 891int inet_shutdown(struct socket *sock, int how)
 892{
 893	struct sock *sk = sock->sk;
 894	int err = 0;
 895
 896	/* This should really check to make sure
 897	 * the socket is a TCP socket. (WHY AC...)
 898	 */
 899	how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
 900		       1->2 bit 2 snds.
 901		       2->3 */
 902	if ((how & ~SHUTDOWN_MASK) || !how)	/* MAXINT->0 */
 903		return -EINVAL;
 904
 905	lock_sock(sk);
 906	if (sock->state == SS_CONNECTING) {
 907		if ((1 << sk->sk_state) &
 908		    (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
 909			sock->state = SS_DISCONNECTING;
 910		else
 911			sock->state = SS_CONNECTED;
 912	}
 913
 914	switch (sk->sk_state) {
 915	case TCP_CLOSE:
 916		err = -ENOTCONN;
 917		/* Hack to wake up other listeners, who can poll for
 918		   EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
 919		fallthrough;
 920	default:
 921		WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | how);
 922		if (sk->sk_prot->shutdown)
 923			sk->sk_prot->shutdown(sk, how);
 924		break;
 925
 926	/* Remaining two branches are temporary solution for missing
 927	 * close() in multithreaded environment. It is _not_ a good idea,
 928	 * but we have no choice until close() is repaired at VFS level.
 929	 */
 930	case TCP_LISTEN:
 931		if (!(how & RCV_SHUTDOWN))
 932			break;
 933		fallthrough;
 934	case TCP_SYN_SENT:
 935		err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
 936		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 937		break;
 938	}
 939
 940	/* Wake up anyone sleeping in poll. */
 941	sk->sk_state_change(sk);
 942	release_sock(sk);
 943	return err;
 944}
 945EXPORT_SYMBOL(inet_shutdown);
 946
 947/*
 948 *	ioctl() calls you can issue on an INET socket. Most of these are
 949 *	device configuration and stuff and very rarely used. Some ioctls
 950 *	pass on to the socket itself.
 951 *
 952 *	NOTE: I like the idea of a module for the config stuff. ie ifconfig
 953 *	loads the devconfigure module does its configuring and unloads it.
 954 *	There's a good 20K of config code hanging around the kernel.
 955 */
 956
 957int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 958{
 959	struct sock *sk = sock->sk;
 960	int err = 0;
 961	struct net *net = sock_net(sk);
 962	void __user *p = (void __user *)arg;
 963	struct ifreq ifr;
 964	struct rtentry rt;
 965
 966	switch (cmd) {
 967	case SIOCADDRT:
 968	case SIOCDELRT:
 969		if (copy_from_user(&rt, p, sizeof(struct rtentry)))
 970			return -EFAULT;
 971		err = ip_rt_ioctl(net, cmd, &rt);
 972		break;
 973	case SIOCRTMSG:
 974		err = -EINVAL;
 975		break;
 976	case SIOCDARP:
 977	case SIOCGARP:
 978	case SIOCSARP:
 979		err = arp_ioctl(net, cmd, (void __user *)arg);
 980		break;
 981	case SIOCGIFADDR:
 982	case SIOCGIFBRDADDR:
 983	case SIOCGIFNETMASK:
 984	case SIOCGIFDSTADDR:
 985	case SIOCGIFPFLAGS:
 986		if (get_user_ifreq(&ifr, NULL, p))
 987			return -EFAULT;
 988		err = devinet_ioctl(net, cmd, &ifr);
 989		if (!err && put_user_ifreq(&ifr, p))
 990			err = -EFAULT;
 991		break;
 992
 993	case SIOCSIFADDR:
 994	case SIOCSIFBRDADDR:
 995	case SIOCSIFNETMASK:
 996	case SIOCSIFDSTADDR:
 997	case SIOCSIFPFLAGS:
 998	case SIOCSIFFLAGS:
 999		if (get_user_ifreq(&ifr, NULL, p))
1000			return -EFAULT;
1001		err = devinet_ioctl(net, cmd, &ifr);
1002		break;
1003	default:
1004		if (sk->sk_prot->ioctl)
1005			err = sk_ioctl(sk, cmd, (void __user *)arg);
1006		else
1007			err = -ENOIOCTLCMD;
1008		break;
1009	}
1010	return err;
1011}
1012EXPORT_SYMBOL(inet_ioctl);
1013
1014#ifdef CONFIG_COMPAT
1015static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd,
1016		struct compat_rtentry __user *ur)
1017{
1018	compat_uptr_t rtdev;
1019	struct rtentry rt;
1020
1021	if (copy_from_user(&rt.rt_dst, &ur->rt_dst,
1022			3 * sizeof(struct sockaddr)) ||
1023	    get_user(rt.rt_flags, &ur->rt_flags) ||
1024	    get_user(rt.rt_metric, &ur->rt_metric) ||
1025	    get_user(rt.rt_mtu, &ur->rt_mtu) ||
1026	    get_user(rt.rt_window, &ur->rt_window) ||
1027	    get_user(rt.rt_irtt, &ur->rt_irtt) ||
1028	    get_user(rtdev, &ur->rt_dev))
1029		return -EFAULT;
1030
1031	rt.rt_dev = compat_ptr(rtdev);
1032	return ip_rt_ioctl(sock_net(sk), cmd, &rt);
1033}
1034
1035static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1036{
1037	void __user *argp = compat_ptr(arg);
1038	struct sock *sk = sock->sk;
1039
1040	switch (cmd) {
1041	case SIOCADDRT:
1042	case SIOCDELRT:
1043		return inet_compat_routing_ioctl(sk, cmd, argp);
1044	default:
1045		if (!sk->sk_prot->compat_ioctl)
1046			return -ENOIOCTLCMD;
1047		return sk->sk_prot->compat_ioctl(sk, cmd, arg);
1048	}
1049}
1050#endif /* CONFIG_COMPAT */
1051
1052const struct proto_ops inet_stream_ops = {
1053	.family		   = PF_INET,
1054	.owner		   = THIS_MODULE,
1055	.release	   = inet_release,
1056	.bind		   = inet_bind,
1057	.connect	   = inet_stream_connect,
1058	.socketpair	   = sock_no_socketpair,
1059	.accept		   = inet_accept,
1060	.getname	   = inet_getname,
1061	.poll		   = tcp_poll,
1062	.ioctl		   = inet_ioctl,
1063	.gettstamp	   = sock_gettstamp,
1064	.listen		   = inet_listen,
1065	.shutdown	   = inet_shutdown,
1066	.setsockopt	   = sock_common_setsockopt,
1067	.getsockopt	   = sock_common_getsockopt,
1068	.sendmsg	   = inet_sendmsg,
1069	.recvmsg	   = inet_recvmsg,
1070#ifdef CONFIG_MMU
1071	.mmap		   = tcp_mmap,
1072#endif
1073	.splice_eof	   = inet_splice_eof,
1074	.splice_read	   = tcp_splice_read,
1075	.set_peek_off      = sk_set_peek_off,
1076	.read_sock	   = tcp_read_sock,
1077	.read_skb	   = tcp_read_skb,
1078	.sendmsg_locked    = tcp_sendmsg_locked,
1079	.peek_len	   = tcp_peek_len,
1080#ifdef CONFIG_COMPAT
1081	.compat_ioctl	   = inet_compat_ioctl,
1082#endif
1083	.set_rcvlowat	   = tcp_set_rcvlowat,
1084};
1085EXPORT_SYMBOL(inet_stream_ops);
1086
1087const struct proto_ops inet_dgram_ops = {
1088	.family		   = PF_INET,
1089	.owner		   = THIS_MODULE,
1090	.release	   = inet_release,
1091	.bind		   = inet_bind,
1092	.connect	   = inet_dgram_connect,
1093	.socketpair	   = sock_no_socketpair,
1094	.accept		   = sock_no_accept,
1095	.getname	   = inet_getname,
1096	.poll		   = udp_poll,
1097	.ioctl		   = inet_ioctl,
1098	.gettstamp	   = sock_gettstamp,
1099	.listen		   = sock_no_listen,
1100	.shutdown	   = inet_shutdown,
1101	.setsockopt	   = sock_common_setsockopt,
1102	.getsockopt	   = sock_common_getsockopt,
1103	.sendmsg	   = inet_sendmsg,
1104	.read_skb	   = udp_read_skb,
1105	.recvmsg	   = inet_recvmsg,
1106	.mmap		   = sock_no_mmap,
1107	.splice_eof	   = inet_splice_eof,
1108	.set_peek_off	   = udp_set_peek_off,
1109#ifdef CONFIG_COMPAT
1110	.compat_ioctl	   = inet_compat_ioctl,
1111#endif
1112};
1113EXPORT_SYMBOL(inet_dgram_ops);
1114
1115/*
1116 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1117 * udp_poll
1118 */
1119static const struct proto_ops inet_sockraw_ops = {
1120	.family		   = PF_INET,
1121	.owner		   = THIS_MODULE,
1122	.release	   = inet_release,
1123	.bind		   = inet_bind,
1124	.connect	   = inet_dgram_connect,
1125	.socketpair	   = sock_no_socketpair,
1126	.accept		   = sock_no_accept,
1127	.getname	   = inet_getname,
1128	.poll		   = datagram_poll,
1129	.ioctl		   = inet_ioctl,
1130	.gettstamp	   = sock_gettstamp,
1131	.listen		   = sock_no_listen,
1132	.shutdown	   = inet_shutdown,
1133	.setsockopt	   = sock_common_setsockopt,
1134	.getsockopt	   = sock_common_getsockopt,
1135	.sendmsg	   = inet_sendmsg,
1136	.recvmsg	   = inet_recvmsg,
1137	.mmap		   = sock_no_mmap,
1138	.splice_eof	   = inet_splice_eof,
1139#ifdef CONFIG_COMPAT
1140	.compat_ioctl	   = inet_compat_ioctl,
1141#endif
1142};
1143
1144static const struct net_proto_family inet_family_ops = {
1145	.family = PF_INET,
1146	.create = inet_create,
1147	.owner	= THIS_MODULE,
1148};
1149
1150/* Upon startup we insert all the elements in inetsw_array[] into
1151 * the linked list inetsw.
1152 */
1153static struct inet_protosw inetsw_array[] =
1154{
1155	{
1156		.type =       SOCK_STREAM,
1157		.protocol =   IPPROTO_TCP,
1158		.prot =       &tcp_prot,
1159		.ops =        &inet_stream_ops,
1160		.flags =      INET_PROTOSW_PERMANENT |
1161			      INET_PROTOSW_ICSK,
1162	},
1163
1164	{
1165		.type =       SOCK_DGRAM,
1166		.protocol =   IPPROTO_UDP,
1167		.prot =       &udp_prot,
1168		.ops =        &inet_dgram_ops,
1169		.flags =      INET_PROTOSW_PERMANENT,
1170       },
1171
1172       {
1173		.type =       SOCK_DGRAM,
1174		.protocol =   IPPROTO_ICMP,
1175		.prot =       &ping_prot,
1176		.ops =        &inet_sockraw_ops,
1177		.flags =      INET_PROTOSW_REUSE,
1178       },
1179
1180       {
1181	       .type =       SOCK_RAW,
1182	       .protocol =   IPPROTO_IP,	/* wild card */
1183	       .prot =       &raw_prot,
1184	       .ops =        &inet_sockraw_ops,
1185	       .flags =      INET_PROTOSW_REUSE,
1186       }
1187};
1188
1189#define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1190
1191void inet_register_protosw(struct inet_protosw *p)
1192{
1193	struct list_head *lh;
1194	struct inet_protosw *answer;
1195	int protocol = p->protocol;
1196	struct list_head *last_perm;
1197
1198	spin_lock_bh(&inetsw_lock);
1199
1200	if (p->type >= SOCK_MAX)
1201		goto out_illegal;
1202
1203	/* If we are trying to override a permanent protocol, bail. */
1204	last_perm = &inetsw[p->type];
1205	list_for_each(lh, &inetsw[p->type]) {
1206		answer = list_entry(lh, struct inet_protosw, list);
1207		/* Check only the non-wild match. */
1208		if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1209			break;
1210		if (protocol == answer->protocol)
1211			goto out_permanent;
1212		last_perm = lh;
1213	}
1214
1215	/* Add the new entry after the last permanent entry if any, so that
1216	 * the new entry does not override a permanent entry when matched with
1217	 * a wild-card protocol. But it is allowed to override any existing
1218	 * non-permanent entry.  This means that when we remove this entry, the
1219	 * system automatically returns to the old behavior.
1220	 */
1221	list_add_rcu(&p->list, last_perm);
1222out:
1223	spin_unlock_bh(&inetsw_lock);
1224
1225	return;
1226
1227out_permanent:
1228	pr_err("Attempt to override permanent protocol %d\n", protocol);
1229	goto out;
1230
1231out_illegal:
1232	pr_err("Ignoring attempt to register invalid socket type %d\n",
1233	       p->type);
1234	goto out;
1235}
1236EXPORT_SYMBOL(inet_register_protosw);
1237
1238void inet_unregister_protosw(struct inet_protosw *p)
1239{
1240	if (INET_PROTOSW_PERMANENT & p->flags) {
1241		pr_err("Attempt to unregister permanent protocol %d\n",
1242		       p->protocol);
1243	} else {
1244		spin_lock_bh(&inetsw_lock);
1245		list_del_rcu(&p->list);
1246		spin_unlock_bh(&inetsw_lock);
1247
1248		synchronize_net();
1249	}
1250}
1251EXPORT_SYMBOL(inet_unregister_protosw);
1252
1253static int inet_sk_reselect_saddr(struct sock *sk)
1254{
1255	struct inet_sock *inet = inet_sk(sk);
1256	__be32 old_saddr = inet->inet_saddr;
1257	__be32 daddr = inet->inet_daddr;
1258	struct flowi4 *fl4;
1259	struct rtable *rt;
1260	__be32 new_saddr;
1261	struct ip_options_rcu *inet_opt;
1262	int err;
1263
1264	inet_opt = rcu_dereference_protected(inet->inet_opt,
1265					     lockdep_sock_is_held(sk));
1266	if (inet_opt && inet_opt->opt.srr)
1267		daddr = inet_opt->opt.faddr;
1268
1269	/* Query new route. */
1270	fl4 = &inet->cork.fl.u.ip4;
1271	rt = ip_route_connect(fl4, daddr, 0, sk->sk_bound_dev_if,
1272			      sk->sk_protocol, inet->inet_sport,
1273			      inet->inet_dport, sk);
1274	if (IS_ERR(rt))
1275		return PTR_ERR(rt);
1276
1277	new_saddr = fl4->saddr;
1278
1279	if (new_saddr == old_saddr) {
1280		sk_setup_caps(sk, &rt->dst);
1281		return 0;
1282	}
1283
1284	err = inet_bhash2_update_saddr(sk, &new_saddr, AF_INET);
1285	if (err) {
1286		ip_rt_put(rt);
1287		return err;
1288	}
1289
1290	sk_setup_caps(sk, &rt->dst);
1291
1292	if (READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) > 1) {
1293		pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1294			__func__, &old_saddr, &new_saddr);
1295	}
1296
1297	/*
1298	 * XXX The only one ugly spot where we need to
1299	 * XXX really change the sockets identity after
1300	 * XXX it has entered the hashes. -DaveM
1301	 *
1302	 * Besides that, it does not check for connection
1303	 * uniqueness. Wait for troubles.
1304	 */
1305	return __sk_prot_rehash(sk);
1306}
1307
1308int inet_sk_rebuild_header(struct sock *sk)
1309{
1310	struct rtable *rt = dst_rtable(__sk_dst_check(sk, 0));
1311	struct inet_sock *inet = inet_sk(sk);
1312	__be32 daddr;
1313	struct ip_options_rcu *inet_opt;
1314	struct flowi4 *fl4;
1315	int err;
1316
1317	/* Route is OK, nothing to do. */
1318	if (rt)
1319		return 0;
1320
1321	/* Reroute. */
1322	rcu_read_lock();
1323	inet_opt = rcu_dereference(inet->inet_opt);
1324	daddr = inet->inet_daddr;
1325	if (inet_opt && inet_opt->opt.srr)
1326		daddr = inet_opt->opt.faddr;
1327	rcu_read_unlock();
1328	fl4 = &inet->cork.fl.u.ip4;
1329	rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1330				   inet->inet_dport, inet->inet_sport,
1331				   sk->sk_protocol, ip_sock_rt_tos(sk),
1332				   sk->sk_bound_dev_if);
1333	if (!IS_ERR(rt)) {
1334		err = 0;
1335		sk_setup_caps(sk, &rt->dst);
1336	} else {
1337		err = PTR_ERR(rt);
1338
1339		/* Routing failed... */
1340		sk->sk_route_caps = 0;
1341		/*
1342		 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1343		 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1344		 */
1345		if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) ||
1346		    sk->sk_state != TCP_SYN_SENT ||
1347		    (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1348		    (err = inet_sk_reselect_saddr(sk)) != 0)
1349			WRITE_ONCE(sk->sk_err_soft, -err);
1350	}
1351
1352	return err;
1353}
1354EXPORT_SYMBOL(inet_sk_rebuild_header);
1355
1356void inet_sk_set_state(struct sock *sk, int state)
1357{
1358	trace_inet_sock_set_state(sk, sk->sk_state, state);
1359	sk->sk_state = state;
1360}
1361EXPORT_SYMBOL(inet_sk_set_state);
1362
1363void inet_sk_state_store(struct sock *sk, int newstate)
1364{
1365	trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1366	smp_store_release(&sk->sk_state, newstate);
1367}
1368
1369struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1370				 netdev_features_t features)
1371{
1372	bool udpfrag = false, fixedid = false, gso_partial, encap;
1373	struct sk_buff *segs = ERR_PTR(-EINVAL);
1374	const struct net_offload *ops;
1375	unsigned int offset = 0;
1376	struct iphdr *iph;
1377	int proto, tot_len;
1378	int nhoff;
1379	int ihl;
1380	int id;
1381
1382	skb_reset_network_header(skb);
1383	nhoff = skb_network_header(skb) - skb_mac_header(skb);
1384	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1385		goto out;
1386
1387	iph = ip_hdr(skb);
1388	ihl = iph->ihl * 4;
1389	if (ihl < sizeof(*iph))
1390		goto out;
1391
1392	id = ntohs(iph->id);
1393	proto = iph->protocol;
1394
1395	/* Warning: after this point, iph might be no longer valid */
1396	if (unlikely(!pskb_may_pull(skb, ihl)))
1397		goto out;
1398	__skb_pull(skb, ihl);
1399
1400	encap = SKB_GSO_CB(skb)->encap_level > 0;
1401	if (encap)
1402		features &= skb->dev->hw_enc_features;
1403	SKB_GSO_CB(skb)->encap_level += ihl;
1404
1405	skb_reset_transport_header(skb);
1406
1407	segs = ERR_PTR(-EPROTONOSUPPORT);
1408
1409	if (!skb->encapsulation || encap) {
1410		udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1411		fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1412
1413		/* fixed ID is invalid if DF bit is not set */
1414		if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1415			goto out;
1416	}
1417
1418	ops = rcu_dereference(inet_offloads[proto]);
1419	if (likely(ops && ops->callbacks.gso_segment)) {
1420		segs = ops->callbacks.gso_segment(skb, features);
1421		if (!segs)
1422			skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
1423	}
1424
1425	if (IS_ERR_OR_NULL(segs))
1426		goto out;
1427
1428	gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1429
1430	skb = segs;
1431	do {
1432		iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1433		if (udpfrag) {
1434			iph->frag_off = htons(offset >> 3);
1435			if (skb->next)
1436				iph->frag_off |= htons(IP_MF);
1437			offset += skb->len - nhoff - ihl;
1438			tot_len = skb->len - nhoff;
1439		} else if (skb_is_gso(skb)) {
1440			if (!fixedid) {
1441				iph->id = htons(id);
1442				id += skb_shinfo(skb)->gso_segs;
1443			}
1444
1445			if (gso_partial)
1446				tot_len = skb_shinfo(skb)->gso_size +
1447					  SKB_GSO_CB(skb)->data_offset +
1448					  skb->head - (unsigned char *)iph;
1449			else
1450				tot_len = skb->len - nhoff;
1451		} else {
1452			if (!fixedid)
1453				iph->id = htons(id++);
1454			tot_len = skb->len - nhoff;
1455		}
1456		iph->tot_len = htons(tot_len);
1457		ip_send_check(iph);
1458		if (encap)
1459			skb_reset_inner_headers(skb);
1460		skb->network_header = (u8 *)iph - skb->head;
1461		skb_reset_mac_len(skb);
1462	} while ((skb = skb->next));
1463
1464out:
1465	return segs;
1466}
1467
1468static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1469					netdev_features_t features)
1470{
1471	if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1472		return ERR_PTR(-EINVAL);
1473
1474	return inet_gso_segment(skb, features);
1475}
1476
1477struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1478{
1479	const struct net_offload *ops;
1480	struct sk_buff *pp = NULL;
1481	const struct iphdr *iph;
1482	struct sk_buff *p;
1483	unsigned int hlen;
1484	unsigned int off;
 
1485	int flush = 1;
1486	int proto;
1487
1488	off = skb_gro_offset(skb);
1489	hlen = off + sizeof(*iph);
1490	iph = skb_gro_header(skb, hlen, off);
1491	if (unlikely(!iph))
1492		goto out;
1493
1494	proto = iph->protocol;
1495
1496	ops = rcu_dereference(inet_offloads[proto]);
1497	if (!ops || !ops->callbacks.gro_receive)
1498		goto out;
1499
1500	if (*(u8 *)iph != 0x45)
1501		goto out;
1502
1503	if (ip_is_fragment(iph))
1504		goto out;
1505
1506	if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1507		goto out;
1508
1509	NAPI_GRO_CB(skb)->proto = proto;
1510	flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (ntohl(*(__be32 *)&iph->id) & ~IP_DF));
 
 
1511
1512	list_for_each_entry(p, head, list) {
1513		struct iphdr *iph2;
 
1514
1515		if (!NAPI_GRO_CB(p)->same_flow)
1516			continue;
1517
1518		iph2 = (struct iphdr *)(p->data + off);
1519		/* The above works because, with the exception of the top
1520		 * (inner most) layer, we only aggregate pkts with the same
1521		 * hdr length so all the hdrs we'll need to verify will start
1522		 * at the same offset.
1523		 */
1524		if ((iph->protocol ^ iph2->protocol) |
1525		    ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1526		    ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1527			NAPI_GRO_CB(p)->same_flow = 0;
1528			continue;
1529		}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1530	}
1531
 
1532	NAPI_GRO_CB(skb)->flush |= flush;
1533	NAPI_GRO_CB(skb)->network_offsets[NAPI_GRO_CB(skb)->encap_mark] = off;
 
 
 
 
1534
1535	/* Note : No need to call skb_gro_postpull_rcsum() here,
1536	 * as we already checked checksum over ipv4 header was 0
1537	 */
1538	skb_gro_pull(skb, sizeof(*iph));
1539	skb_set_transport_header(skb, skb_gro_offset(skb));
1540
1541	pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1542				       ops->callbacks.gro_receive, head, skb);
1543
1544out:
1545	skb_gro_flush_final(skb, pp, flush);
1546
1547	return pp;
1548}
1549
1550static struct sk_buff *ipip_gro_receive(struct list_head *head,
1551					struct sk_buff *skb)
1552{
1553	if (NAPI_GRO_CB(skb)->encap_mark) {
1554		NAPI_GRO_CB(skb)->flush = 1;
1555		return NULL;
1556	}
1557
1558	NAPI_GRO_CB(skb)->encap_mark = 1;
1559
1560	return inet_gro_receive(head, skb);
1561}
1562
1563#define SECONDS_PER_DAY	86400
1564
1565/* inet_current_timestamp - Return IP network timestamp
1566 *
1567 * Return milliseconds since midnight in network byte order.
1568 */
1569__be32 inet_current_timestamp(void)
1570{
1571	u32 secs;
1572	u32 msecs;
1573	struct timespec64 ts;
1574
1575	ktime_get_real_ts64(&ts);
1576
1577	/* Get secs since midnight. */
1578	(void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1579	/* Convert to msecs. */
1580	msecs = secs * MSEC_PER_SEC;
1581	/* Convert nsec to msec. */
1582	msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1583
1584	/* Convert to network byte order. */
1585	return htonl(msecs);
1586}
1587EXPORT_SYMBOL(inet_current_timestamp);
1588
1589int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1590{
1591	unsigned int family = READ_ONCE(sk->sk_family);
1592
1593	if (family == AF_INET)
1594		return ip_recv_error(sk, msg, len, addr_len);
1595#if IS_ENABLED(CONFIG_IPV6)
1596	if (family == AF_INET6)
1597		return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1598#endif
1599	return -EINVAL;
1600}
1601EXPORT_SYMBOL(inet_recv_error);
1602
1603int inet_gro_complete(struct sk_buff *skb, int nhoff)
1604{
1605	struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1606	const struct net_offload *ops;
1607	__be16 totlen = iph->tot_len;
1608	int proto = iph->protocol;
1609	int err = -ENOSYS;
1610
1611	if (skb->encapsulation) {
1612		skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1613		skb_set_inner_network_header(skb, nhoff);
1614	}
1615
1616	iph_set_totlen(iph, skb->len - nhoff);
1617	csum_replace2(&iph->check, totlen, iph->tot_len);
1618
1619	ops = rcu_dereference(inet_offloads[proto]);
1620	if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1621		goto out;
1622
1623	/* Only need to add sizeof(*iph) to get to the next hdr below
1624	 * because any hdr with option will have been flushed in
1625	 * inet_gro_receive().
1626	 */
1627	err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1628			      tcp4_gro_complete, udp4_gro_complete,
1629			      skb, nhoff + sizeof(*iph));
1630
1631out:
1632	return err;
1633}
1634
1635static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1636{
1637	skb->encapsulation = 1;
1638	skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1639	return inet_gro_complete(skb, nhoff);
1640}
1641
1642int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1643			 unsigned short type, unsigned char protocol,
1644			 struct net *net)
1645{
1646	struct socket *sock;
1647	int rc = sock_create_kern(net, family, type, protocol, &sock);
1648
1649	if (rc == 0) {
1650		*sk = sock->sk;
1651		(*sk)->sk_allocation = GFP_ATOMIC;
1652		(*sk)->sk_use_task_frag = false;
1653		/*
1654		 * Unhash it so that IP input processing does not even see it,
1655		 * we do not wish this socket to see incoming packets.
1656		 */
1657		(*sk)->sk_prot->unhash(*sk);
1658	}
1659	return rc;
1660}
1661EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1662
1663unsigned long snmp_fold_field(void __percpu *mib, int offt)
1664{
1665	unsigned long res = 0;
1666	int i;
1667
1668	for_each_possible_cpu(i)
1669		res += snmp_get_cpu_field(mib, i, offt);
1670	return res;
1671}
1672EXPORT_SYMBOL_GPL(snmp_fold_field);
1673
1674#if BITS_PER_LONG==32
1675
1676u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1677			 size_t syncp_offset)
1678{
1679	void *bhptr;
1680	struct u64_stats_sync *syncp;
1681	u64 v;
1682	unsigned int start;
1683
1684	bhptr = per_cpu_ptr(mib, cpu);
1685	syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1686	do {
1687		start = u64_stats_fetch_begin(syncp);
1688		v = *(((u64 *)bhptr) + offt);
1689	} while (u64_stats_fetch_retry(syncp, start));
1690
1691	return v;
1692}
1693EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1694
1695u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1696{
1697	u64 res = 0;
1698	int cpu;
1699
1700	for_each_possible_cpu(cpu) {
1701		res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1702	}
1703	return res;
1704}
1705EXPORT_SYMBOL_GPL(snmp_fold_field64);
1706#endif
1707
1708#ifdef CONFIG_IP_MULTICAST
1709static const struct net_protocol igmp_protocol = {
1710	.handler =	igmp_rcv,
1711};
1712#endif
1713
1714static const struct net_protocol icmp_protocol = {
1715	.handler =	icmp_rcv,
1716	.err_handler =	icmp_err,
1717	.no_policy =	1,
1718};
1719
1720static __net_init int ipv4_mib_init_net(struct net *net)
1721{
1722	int i;
1723
1724	net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1725	if (!net->mib.tcp_statistics)
1726		goto err_tcp_mib;
1727	net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1728	if (!net->mib.ip_statistics)
1729		goto err_ip_mib;
1730
1731	for_each_possible_cpu(i) {
1732		struct ipstats_mib *af_inet_stats;
1733		af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1734		u64_stats_init(&af_inet_stats->syncp);
1735	}
1736
1737	net->mib.net_statistics = alloc_percpu(struct linux_mib);
1738	if (!net->mib.net_statistics)
1739		goto err_net_mib;
1740	net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1741	if (!net->mib.udp_statistics)
1742		goto err_udp_mib;
1743	net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1744	if (!net->mib.udplite_statistics)
1745		goto err_udplite_mib;
1746	net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1747	if (!net->mib.icmp_statistics)
1748		goto err_icmp_mib;
1749	net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1750					      GFP_KERNEL);
1751	if (!net->mib.icmpmsg_statistics)
1752		goto err_icmpmsg_mib;
1753
1754	tcp_mib_init(net);
1755	return 0;
1756
1757err_icmpmsg_mib:
1758	free_percpu(net->mib.icmp_statistics);
1759err_icmp_mib:
1760	free_percpu(net->mib.udplite_statistics);
1761err_udplite_mib:
1762	free_percpu(net->mib.udp_statistics);
1763err_udp_mib:
1764	free_percpu(net->mib.net_statistics);
1765err_net_mib:
1766	free_percpu(net->mib.ip_statistics);
1767err_ip_mib:
1768	free_percpu(net->mib.tcp_statistics);
1769err_tcp_mib:
1770	return -ENOMEM;
1771}
1772
1773static __net_exit void ipv4_mib_exit_net(struct net *net)
1774{
1775	kfree(net->mib.icmpmsg_statistics);
1776	free_percpu(net->mib.icmp_statistics);
1777	free_percpu(net->mib.udplite_statistics);
1778	free_percpu(net->mib.udp_statistics);
1779	free_percpu(net->mib.net_statistics);
1780	free_percpu(net->mib.ip_statistics);
1781	free_percpu(net->mib.tcp_statistics);
1782#ifdef CONFIG_MPTCP
1783	/* allocated on demand, see mptcp_init_sock() */
1784	free_percpu(net->mib.mptcp_statistics);
1785#endif
1786}
1787
1788static __net_initdata struct pernet_operations ipv4_mib_ops = {
1789	.init = ipv4_mib_init_net,
1790	.exit = ipv4_mib_exit_net,
1791};
1792
1793static int __init init_ipv4_mibs(void)
1794{
1795	return register_pernet_subsys(&ipv4_mib_ops);
1796}
1797
1798static __net_init int inet_init_net(struct net *net)
1799{
1800	/*
1801	 * Set defaults for local port range
1802	 */
1803	net->ipv4.ip_local_ports.range = 60999u << 16 | 32768u;
1804
1805	seqlock_init(&net->ipv4.ping_group_range.lock);
1806	/*
1807	 * Sane defaults - nobody may create ping sockets.
1808	 * Boot scripts should set this to distro-specific group.
1809	 */
1810	net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1811	net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1812
1813	/* Default values for sysctl-controlled parameters.
1814	 * We set them here, in case sysctl is not compiled.
1815	 */
1816	net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1817	net->ipv4.sysctl_ip_fwd_update_priority = 1;
1818	net->ipv4.sysctl_ip_dynaddr = 0;
1819	net->ipv4.sysctl_ip_early_demux = 1;
1820	net->ipv4.sysctl_udp_early_demux = 1;
1821	net->ipv4.sysctl_tcp_early_demux = 1;
1822	net->ipv4.sysctl_nexthop_compat_mode = 1;
1823#ifdef CONFIG_SYSCTL
1824	net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1825#endif
1826
1827	/* Some igmp sysctl, whose values are always used */
1828	net->ipv4.sysctl_igmp_max_memberships = 20;
1829	net->ipv4.sysctl_igmp_max_msf = 10;
1830	/* IGMP reports for link-local multicast groups are enabled by default */
1831	net->ipv4.sysctl_igmp_llm_reports = 1;
1832	net->ipv4.sysctl_igmp_qrv = 2;
1833
1834	net->ipv4.sysctl_fib_notify_on_flag_change = 0;
1835
1836	return 0;
1837}
1838
1839static __net_initdata struct pernet_operations af_inet_ops = {
1840	.init = inet_init_net,
1841};
1842
1843static int __init init_inet_pernet_ops(void)
1844{
1845	return register_pernet_subsys(&af_inet_ops);
1846}
1847
1848static int ipv4_proc_init(void);
1849
1850/*
1851 *	IP protocol layer initialiser
1852 */
1853
1854
1855static const struct net_offload ipip_offload = {
1856	.callbacks = {
1857		.gso_segment	= ipip_gso_segment,
1858		.gro_receive	= ipip_gro_receive,
1859		.gro_complete	= ipip_gro_complete,
1860	},
1861};
1862
1863static int __init ipip_offload_init(void)
1864{
1865	return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1866}
1867
1868static int __init ipv4_offload_init(void)
1869{
1870	/*
1871	 * Add offloads
1872	 */
1873	if (udpv4_offload_init() < 0)
1874		pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1875	if (tcpv4_offload_init() < 0)
1876		pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1877	if (ipip_offload_init() < 0)
1878		pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1879
1880	net_hotdata.ip_packet_offload = (struct packet_offload) {
1881		.type = cpu_to_be16(ETH_P_IP),
1882		.callbacks = {
1883			.gso_segment = inet_gso_segment,
1884			.gro_receive = inet_gro_receive,
1885			.gro_complete = inet_gro_complete,
1886		},
1887	};
1888	dev_add_offload(&net_hotdata.ip_packet_offload);
1889	return 0;
1890}
1891
1892fs_initcall(ipv4_offload_init);
1893
1894static struct packet_type ip_packet_type __read_mostly = {
1895	.type = cpu_to_be16(ETH_P_IP),
1896	.func = ip_rcv,
1897	.list_func = ip_list_rcv,
1898};
1899
1900static int __init inet_init(void)
1901{
1902	struct inet_protosw *q;
1903	struct list_head *r;
1904	int rc;
1905
1906	sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1907
1908	raw_hashinfo_init(&raw_v4_hashinfo);
1909
1910	rc = proto_register(&tcp_prot, 1);
1911	if (rc)
1912		goto out;
1913
1914	rc = proto_register(&udp_prot, 1);
1915	if (rc)
1916		goto out_unregister_tcp_proto;
1917
1918	rc = proto_register(&raw_prot, 1);
1919	if (rc)
1920		goto out_unregister_udp_proto;
1921
1922	rc = proto_register(&ping_prot, 1);
1923	if (rc)
1924		goto out_unregister_raw_proto;
1925
1926	/*
1927	 *	Tell SOCKET that we are alive...
1928	 */
1929
1930	(void)sock_register(&inet_family_ops);
1931
1932#ifdef CONFIG_SYSCTL
1933	ip_static_sysctl_init();
1934#endif
1935
1936	/*
1937	 *	Add all the base protocols.
1938	 */
1939
1940	if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1941		pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1942
1943	net_hotdata.udp_protocol = (struct net_protocol) {
1944		.handler =	udp_rcv,
1945		.err_handler =	udp_err,
1946		.no_policy =	1,
1947	};
1948	if (inet_add_protocol(&net_hotdata.udp_protocol, IPPROTO_UDP) < 0)
1949		pr_crit("%s: Cannot add UDP protocol\n", __func__);
1950
1951	net_hotdata.tcp_protocol = (struct net_protocol) {
1952		.handler	=	tcp_v4_rcv,
1953		.err_handler	=	tcp_v4_err,
1954		.no_policy	=	1,
1955		.icmp_strict_tag_validation = 1,
1956	};
1957	if (inet_add_protocol(&net_hotdata.tcp_protocol, IPPROTO_TCP) < 0)
1958		pr_crit("%s: Cannot add TCP protocol\n", __func__);
1959#ifdef CONFIG_IP_MULTICAST
1960	if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1961		pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1962#endif
1963
1964	/* Register the socket-side information for inet_create. */
1965	for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1966		INIT_LIST_HEAD(r);
1967
1968	for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1969		inet_register_protosw(q);
1970
1971	/*
1972	 *	Set the ARP module up
1973	 */
1974
1975	arp_init();
1976
1977	/*
1978	 *	Set the IP module up
1979	 */
1980
1981	ip_init();
1982
1983	/* Initialise per-cpu ipv4 mibs */
1984	if (init_ipv4_mibs())
1985		panic("%s: Cannot init ipv4 mibs\n", __func__);
1986
1987	/* Setup TCP slab cache for open requests. */
1988	tcp_init();
1989
1990	/* Setup UDP memory threshold */
1991	udp_init();
1992
1993	/* Add UDP-Lite (RFC 3828) */
1994	udplite4_register();
1995
1996	raw_init();
1997
1998	ping_init();
1999
2000	/*
2001	 *	Set the ICMP layer up
2002	 */
2003
2004	if (icmp_init() < 0)
2005		panic("Failed to create the ICMP control socket.\n");
2006
2007	/*
2008	 *	Initialise the multicast router
2009	 */
2010#if defined(CONFIG_IP_MROUTE)
2011	if (ip_mr_init())
2012		pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2013#endif
2014
2015	if (init_inet_pernet_ops())
2016		pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2017
2018	ipv4_proc_init();
2019
2020	ipfrag_init();
2021
2022	dev_add_pack(&ip_packet_type);
2023
2024	ip_tunnel_core_init();
2025
2026	rc = 0;
2027out:
2028	return rc;
2029out_unregister_raw_proto:
2030	proto_unregister(&raw_prot);
2031out_unregister_udp_proto:
2032	proto_unregister(&udp_prot);
2033out_unregister_tcp_proto:
2034	proto_unregister(&tcp_prot);
2035	goto out;
2036}
2037
2038fs_initcall(inet_init);
2039
2040/* ------------------------------------------------------------------------ */
2041
2042#ifdef CONFIG_PROC_FS
2043static int __init ipv4_proc_init(void)
2044{
2045	int rc = 0;
2046
2047	if (raw_proc_init())
2048		goto out_raw;
2049	if (tcp4_proc_init())
2050		goto out_tcp;
2051	if (udp4_proc_init())
2052		goto out_udp;
2053	if (ping_proc_init())
2054		goto out_ping;
2055	if (ip_misc_proc_init())
2056		goto out_misc;
2057out:
2058	return rc;
2059out_misc:
2060	ping_proc_exit();
2061out_ping:
2062	udp4_proc_exit();
2063out_udp:
2064	tcp4_proc_exit();
2065out_tcp:
2066	raw_proc_exit();
2067out_raw:
2068	rc = -ENOMEM;
2069	goto out;
2070}
2071
2072#else /* CONFIG_PROC_FS */
2073static int __init ipv4_proc_init(void)
2074{
2075	return 0;
2076}
2077#endif /* CONFIG_PROC_FS */