Linux Audio

Check our new training course

Loading...
v3.15
   1/*
   2 * INET		An implementation of the TCP/IP protocol suite for the LINUX
   3 *		operating system.  INET is implemented using the  BSD Socket
   4 *		interface as the means of communication with the user level.
   5 *
   6 *		PF_INET protocol family socket handler.
   7 *
   8 * Authors:	Ross Biro
   9 *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  10 *		Florian La Roche, <flla@stud.uni-sb.de>
  11 *		Alan Cox, <A.Cox@swansea.ac.uk>
  12 *
  13 * Changes (see also sock.c)
  14 *
  15 *		piggy,
  16 *		Karl Knutson	:	Socket protocol table
  17 *		A.N.Kuznetsov	:	Socket death error in accept().
  18 *		John Richardson :	Fix non blocking error in connect()
  19 *					so sockets that fail to connect
  20 *					don't return -EINPROGRESS.
  21 *		Alan Cox	:	Asynchronous I/O support
  22 *		Alan Cox	:	Keep correct socket pointer on sock
  23 *					structures
  24 *					when accept() ed
  25 *		Alan Cox	:	Semantics of SO_LINGER aren't state
  26 *					moved to close when you look carefully.
  27 *					With this fixed and the accept bug fixed
  28 *					some RPC stuff seems happier.
  29 *		Niibe Yutaka	:	4.4BSD style write async I/O
  30 *		Alan Cox,
  31 *		Tony Gale 	:	Fixed reuse semantics.
  32 *		Alan Cox	:	bind() shouldn't abort existing but dead
  33 *					sockets. Stops FTP netin:.. I hope.
  34 *		Alan Cox	:	bind() works correctly for RAW sockets.
  35 *					Note that FreeBSD at least was broken
  36 *					in this respect so be careful with
  37 *					compatibility tests...
  38 *		Alan Cox	:	routing cache support
  39 *		Alan Cox	:	memzero the socket structure for
  40 *					compactness.
  41 *		Matt Day	:	nonblock connect error handler
  42 *		Alan Cox	:	Allow large numbers of pending sockets
  43 *					(eg for big web sites), but only if
  44 *					specifically application requested.
  45 *		Alan Cox	:	New buffering throughout IP. Used
  46 *					dumbly.
  47 *		Alan Cox	:	New buffering now used smartly.
  48 *		Alan Cox	:	BSD rather than common sense
  49 *					interpretation of listen.
  50 *		Germano Caronni	:	Assorted small races.
  51 *		Alan Cox	:	sendmsg/recvmsg basic support.
  52 *		Alan Cox	:	Only sendmsg/recvmsg now supported.
  53 *		Alan Cox	:	Locked down bind (see security list).
  54 *		Alan Cox	:	Loosened bind a little.
  55 *		Mike McLagan	:	ADD/DEL DLCI Ioctls
  56 *	Willy Konynenberg	:	Transparent proxying support.
  57 *		David S. Miller	:	New socket lookup architecture.
  58 *					Some other random speedups.
  59 *		Cyrus Durgin	:	Cleaned up file for kmod hacks.
  60 *		Andi Kleen	:	Fix inet_stream_connect TCP race.
  61 *
  62 *		This program is free software; you can redistribute it and/or
  63 *		modify it under the terms of the GNU General Public License
  64 *		as published by the Free Software Foundation; either version
  65 *		2 of the License, or (at your option) any later version.
  66 */
  67
  68#define pr_fmt(fmt) "IPv4: " fmt
  69
  70#include <linux/err.h>
  71#include <linux/errno.h>
  72#include <linux/types.h>
  73#include <linux/socket.h>
  74#include <linux/in.h>
  75#include <linux/kernel.h>
  76#include <linux/module.h>
  77#include <linux/sched.h>
  78#include <linux/timer.h>
  79#include <linux/string.h>
  80#include <linux/sockios.h>
  81#include <linux/net.h>
  82#include <linux/capability.h>
  83#include <linux/fcntl.h>
  84#include <linux/mm.h>
  85#include <linux/interrupt.h>
  86#include <linux/stat.h>
  87#include <linux/init.h>
  88#include <linux/poll.h>
  89#include <linux/netfilter_ipv4.h>
  90#include <linux/random.h>
  91#include <linux/slab.h>
  92
  93#include <asm/uaccess.h>
  94
  95#include <linux/inet.h>
  96#include <linux/igmp.h>
  97#include <linux/inetdevice.h>
  98#include <linux/netdevice.h>
  99#include <net/checksum.h>
 100#include <net/ip.h>
 101#include <net/protocol.h>
 102#include <net/arp.h>
 103#include <net/route.h>
 104#include <net/ip_fib.h>
 105#include <net/inet_connection_sock.h>
 106#include <net/tcp.h>
 107#include <net/udp.h>
 108#include <net/udplite.h>
 109#include <net/ping.h>
 110#include <linux/skbuff.h>
 111#include <net/sock.h>
 112#include <net/raw.h>
 113#include <net/icmp.h>
 114#include <net/inet_common.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
 
 122
 123/* The inetsw table contains everything that inet_create needs to
 124 * build a new socket.
 125 */
 126static struct list_head inetsw[SOCK_MAX];
 127static DEFINE_SPINLOCK(inetsw_lock);
 128
 129/* New destruction routine */
 130
 131void inet_sock_destruct(struct sock *sk)
 132{
 133	struct inet_sock *inet = inet_sk(sk);
 134
 135	__skb_queue_purge(&sk->sk_receive_queue);
 136	__skb_queue_purge(&sk->sk_error_queue);
 137
 138	sk_mem_reclaim(sk);
 139
 140	if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
 141		pr_err("Attempt to release TCP socket in state %d %p\n",
 142		       sk->sk_state, sk);
 143		return;
 144	}
 145	if (!sock_flag(sk, SOCK_DEAD)) {
 146		pr_err("Attempt to release alive inet socket %p\n", sk);
 147		return;
 148	}
 149
 150	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
 151	WARN_ON(atomic_read(&sk->sk_wmem_alloc));
 152	WARN_ON(sk->sk_wmem_queued);
 153	WARN_ON(sk->sk_forward_alloc);
 154
 155	kfree(rcu_dereference_protected(inet->inet_opt, 1));
 156	dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
 157	dst_release(sk->sk_rx_dst);
 158	sk_refcnt_debug_dec(sk);
 159}
 160EXPORT_SYMBOL(inet_sock_destruct);
 161
 162/*
 163 *	The routines beyond this point handle the behaviour of an AF_INET
 164 *	socket object. Mostly it punts to the subprotocols of IP to do
 165 *	the work.
 166 */
 167
 168/*
 169 *	Automatically bind an unbound socket.
 170 */
 171
 172static int inet_autobind(struct sock *sk)
 173{
 174	struct inet_sock *inet;
 175	/* We may need to bind the socket. */
 176	lock_sock(sk);
 177	inet = inet_sk(sk);
 178	if (!inet->inet_num) {
 179		if (sk->sk_prot->get_port(sk, 0)) {
 180			release_sock(sk);
 181			return -EAGAIN;
 182		}
 183		inet->inet_sport = htons(inet->inet_num);
 184	}
 185	release_sock(sk);
 186	return 0;
 187}
 188
 189/*
 190 *	Move a socket into listening state.
 191 */
 192int inet_listen(struct socket *sock, int backlog)
 193{
 194	struct sock *sk = sock->sk;
 195	unsigned char old_state;
 196	int err;
 197
 198	lock_sock(sk);
 199
 200	err = -EINVAL;
 201	if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
 202		goto out;
 203
 204	old_state = sk->sk_state;
 205	if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
 206		goto out;
 207
 208	/* Really, if the socket is already in listen state
 209	 * we can only allow the backlog to be adjusted.
 210	 */
 211	if (old_state != TCP_LISTEN) {
 212		/* Check special setups for testing purpose to enable TFO w/o
 213		 * requiring TCP_FASTOPEN sockopt.
 214		 * Note that only TCP sockets (SOCK_STREAM) will reach here.
 215		 * Also fastopenq may already been allocated because this
 216		 * socket was in TCP_LISTEN state previously but was
 217		 * shutdown() (rather than close()).
 218		 */
 219		if ((sysctl_tcp_fastopen & TFO_SERVER_ENABLE) != 0 &&
 220		    inet_csk(sk)->icsk_accept_queue.fastopenq == NULL) {
 221			if ((sysctl_tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) != 0)
 222				err = fastopen_init_queue(sk, backlog);
 223			else if ((sysctl_tcp_fastopen &
 224				  TFO_SERVER_WO_SOCKOPT2) != 0)
 225				err = fastopen_init_queue(sk,
 226				    ((uint)sysctl_tcp_fastopen) >> 16);
 227			else
 228				err = 0;
 229			if (err)
 230				goto out;
 231		}
 
 232		err = inet_csk_listen_start(sk, backlog);
 233		if (err)
 234			goto out;
 235	}
 236	sk->sk_max_ack_backlog = backlog;
 237	err = 0;
 238
 239out:
 240	release_sock(sk);
 241	return err;
 242}
 243EXPORT_SYMBOL(inet_listen);
 244
 245/*
 246 *	Create an inet socket.
 247 */
 248
 249static int inet_create(struct net *net, struct socket *sock, int protocol,
 250		       int kern)
 251{
 252	struct sock *sk;
 253	struct inet_protosw *answer;
 254	struct inet_sock *inet;
 255	struct proto *answer_prot;
 256	unsigned char answer_flags;
 257	char answer_no_check;
 258	int try_loading_module = 0;
 259	int err;
 260
 
 
 
 261	sock->state = SS_UNCONNECTED;
 262
 263	/* Look for the requested type/protocol pair. */
 264lookup_protocol:
 265	err = -ESOCKTNOSUPPORT;
 266	rcu_read_lock();
 267	list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
 268
 269		err = 0;
 270		/* Check the non-wild match. */
 271		if (protocol == answer->protocol) {
 272			if (protocol != IPPROTO_IP)
 273				break;
 274		} else {
 275			/* Check for the two wild cases. */
 276			if (IPPROTO_IP == protocol) {
 277				protocol = answer->protocol;
 278				break;
 279			}
 280			if (IPPROTO_IP == answer->protocol)
 281				break;
 282		}
 283		err = -EPROTONOSUPPORT;
 284	}
 285
 286	if (unlikely(err)) {
 287		if (try_loading_module < 2) {
 288			rcu_read_unlock();
 289			/*
 290			 * Be more specific, e.g. net-pf-2-proto-132-type-1
 291			 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
 292			 */
 293			if (++try_loading_module == 1)
 294				request_module("net-pf-%d-proto-%d-type-%d",
 295					       PF_INET, protocol, sock->type);
 296			/*
 297			 * Fall back to generic, e.g. net-pf-2-proto-132
 298			 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
 299			 */
 300			else
 301				request_module("net-pf-%d-proto-%d",
 302					       PF_INET, protocol);
 303			goto lookup_protocol;
 304		} else
 305			goto out_rcu_unlock;
 306	}
 307
 308	err = -EPERM;
 309	if (sock->type == SOCK_RAW && !kern &&
 310	    !ns_capable(net->user_ns, CAP_NET_RAW))
 311		goto out_rcu_unlock;
 312
 313	sock->ops = answer->ops;
 314	answer_prot = answer->prot;
 315	answer_no_check = answer->no_check;
 316	answer_flags = answer->flags;
 317	rcu_read_unlock();
 318
 319	WARN_ON(answer_prot->slab == NULL);
 320
 321	err = -ENOBUFS;
 322	sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
 323	if (sk == NULL)
 324		goto out;
 325
 326	err = 0;
 327	sk->sk_no_check = answer_no_check;
 328	if (INET_PROTOSW_REUSE & answer_flags)
 329		sk->sk_reuse = SK_CAN_REUSE;
 330
 331	inet = inet_sk(sk);
 332	inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
 333
 334	inet->nodefrag = 0;
 335
 336	if (SOCK_RAW == sock->type) {
 337		inet->inet_num = protocol;
 338		if (IPPROTO_RAW == protocol)
 339			inet->hdrincl = 1;
 340	}
 341
 342	if (net->ipv4.sysctl_ip_no_pmtu_disc)
 343		inet->pmtudisc = IP_PMTUDISC_DONT;
 344	else
 345		inet->pmtudisc = IP_PMTUDISC_WANT;
 346
 347	inet->inet_id = 0;
 348
 349	sock_init_data(sock, sk);
 350
 351	sk->sk_destruct	   = inet_sock_destruct;
 352	sk->sk_protocol	   = protocol;
 353	sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
 354
 355	inet->uc_ttl	= -1;
 356	inet->mc_loop	= 1;
 357	inet->mc_ttl	= 1;
 358	inet->mc_all	= 1;
 359	inet->mc_index	= 0;
 360	inet->mc_list	= NULL;
 361	inet->rcv_tos	= 0;
 362
 363	sk_refcnt_debug_inc(sk);
 364
 365	if (inet->inet_num) {
 366		/* It assumes that any protocol which allows
 367		 * the user to assign a number at socket
 368		 * creation time automatically
 369		 * shares.
 370		 */
 371		inet->inet_sport = htons(inet->inet_num);
 372		/* Add to protocol hash chains. */
 373		sk->sk_prot->hash(sk);
 
 
 
 
 374	}
 375
 376	if (sk->sk_prot->init) {
 377		err = sk->sk_prot->init(sk);
 378		if (err)
 379			sk_common_release(sk);
 
 
 
 
 
 
 
 
 
 
 380	}
 381out:
 382	return err;
 383out_rcu_unlock:
 384	rcu_read_unlock();
 385	goto out;
 386}
 387
 388
 389/*
 390 *	The peer socket should always be NULL (or else). When we call this
 391 *	function we are destroying the object and from then on nobody
 392 *	should refer to it.
 393 */
 394int inet_release(struct socket *sock)
 395{
 396	struct sock *sk = sock->sk;
 397
 398	if (sk) {
 399		long timeout;
 400
 401		sock_rps_reset_flow(sk);
 402
 403		/* Applications forget to leave groups before exiting */
 404		ip_mc_drop_socket(sk);
 405
 406		/* If linger is set, we don't return until the close
 407		 * is complete.  Otherwise we return immediately. The
 408		 * actually closing is done the same either way.
 409		 *
 410		 * If the close is due to the process exiting, we never
 411		 * linger..
 412		 */
 413		timeout = 0;
 414		if (sock_flag(sk, SOCK_LINGER) &&
 415		    !(current->flags & PF_EXITING))
 416			timeout = sk->sk_lingertime;
 417		sock->sk = NULL;
 418		sk->sk_prot->close(sk, timeout);
 419	}
 420	return 0;
 421}
 422EXPORT_SYMBOL(inet_release);
 423
 424/* It is off by default, see below. */
 425int sysctl_ip_nonlocal_bind __read_mostly;
 426EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
 427
 428int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 429{
 430	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
 431	struct sock *sk = sock->sk;
 432	struct inet_sock *inet = inet_sk(sk);
 433	struct net *net = sock_net(sk);
 434	unsigned short snum;
 435	int chk_addr_ret;
 436	int err;
 437
 438	/* If the socket has its own bind function then use it. (RAW) */
 439	if (sk->sk_prot->bind) {
 440		err = sk->sk_prot->bind(sk, uaddr, addr_len);
 441		goto out;
 442	}
 443	err = -EINVAL;
 444	if (addr_len < sizeof(struct sockaddr_in))
 445		goto out;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 446
 447	if (addr->sin_family != AF_INET) {
 448		/* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
 449		 * only if s_addr is INADDR_ANY.
 450		 */
 451		err = -EAFNOSUPPORT;
 452		if (addr->sin_family != AF_UNSPEC ||
 453		    addr->sin_addr.s_addr != htonl(INADDR_ANY))
 454			goto out;
 455	}
 456
 457	chk_addr_ret = inet_addr_type(net, addr->sin_addr.s_addr);
 
 458
 459	/* Not specified by any standard per-se, however it breaks too
 460	 * many applications when removed.  It is unfortunate since
 461	 * allowing applications to make a non-local bind solves
 462	 * several problems with systems using dynamic addressing.
 463	 * (ie. your servers still start up even if your ISDN link
 464	 *  is temporarily down)
 465	 */
 466	err = -EADDRNOTAVAIL;
 467	if (!sysctl_ip_nonlocal_bind &&
 468	    !(inet->freebind || inet->transparent) &&
 469	    addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
 470	    chk_addr_ret != RTN_LOCAL &&
 471	    chk_addr_ret != RTN_MULTICAST &&
 472	    chk_addr_ret != RTN_BROADCAST)
 473		goto out;
 474
 475	snum = ntohs(addr->sin_port);
 476	err = -EACCES;
 477	if (snum && snum < PROT_SOCK &&
 478	    !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
 479		goto out;
 480
 481	/*      We keep a pair of addresses. rcv_saddr is the one
 482	 *      used by hash lookups, and saddr is used for transmit.
 483	 *
 484	 *      In the BSD API these are the same except where it
 485	 *      would be illegal to use them (multicast/broadcast) in
 486	 *      which case the sending device address is used.
 487	 */
 488	lock_sock(sk);
 
 489
 490	/* Check these errors (active socket, double bind). */
 491	err = -EINVAL;
 492	if (sk->sk_state != TCP_CLOSE || inet->inet_num)
 493		goto out_release_sock;
 494
 495	inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
 496	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
 497		inet->inet_saddr = 0;  /* Use device */
 498
 499	/* Make sure we are allowed to bind here. */
 500	if (sk->sk_prot->get_port(sk, snum)) {
 501		inet->inet_saddr = inet->inet_rcv_saddr = 0;
 502		err = -EADDRINUSE;
 503		goto out_release_sock;
 
 
 
 
 
 
 
 
 504	}
 505
 506	if (inet->inet_rcv_saddr)
 507		sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
 508	if (snum)
 509		sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
 510	inet->inet_sport = htons(inet->inet_num);
 511	inet->inet_daddr = 0;
 512	inet->inet_dport = 0;
 513	sk_dst_reset(sk);
 514	err = 0;
 515out_release_sock:
 516	release_sock(sk);
 
 517out:
 518	return err;
 519}
 520EXPORT_SYMBOL(inet_bind);
 521
 522int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
 523		       int addr_len, int flags)
 524{
 525	struct sock *sk = sock->sk;
 
 526
 527	if (addr_len < sizeof(uaddr->sa_family))
 528		return -EINVAL;
 529	if (uaddr->sa_family == AF_UNSPEC)
 530		return sk->sk_prot->disconnect(sk, flags);
 531
 
 
 
 
 
 
 532	if (!inet_sk(sk)->inet_num && inet_autobind(sk))
 533		return -EAGAIN;
 534	return sk->sk_prot->connect(sk, uaddr, addr_len);
 535}
 536EXPORT_SYMBOL(inet_dgram_connect);
 537
 538static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
 539{
 540	DEFINE_WAIT(wait);
 541
 542	prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
 543	sk->sk_write_pending += writebias;
 544
 545	/* Basic assumption: if someone sets sk->sk_err, he _must_
 546	 * change state of the socket from TCP_SYN_*.
 547	 * Connect() does not allow to get error notifications
 548	 * without closing the socket.
 549	 */
 550	while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 551		release_sock(sk);
 552		timeo = schedule_timeout(timeo);
 553		lock_sock(sk);
 554		if (signal_pending(current) || !timeo)
 555			break;
 556		prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
 557	}
 558	finish_wait(sk_sleep(sk), &wait);
 559	sk->sk_write_pending -= writebias;
 560	return timeo;
 561}
 562
 563/*
 564 *	Connect to a remote host. There is regrettably still a little
 565 *	TCP 'magic' in here.
 566 */
 567int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 568			  int addr_len, int flags)
 569{
 570	struct sock *sk = sock->sk;
 571	int err;
 572	long timeo;
 573
 574	if (addr_len < sizeof(uaddr->sa_family))
 575		return -EINVAL;
 576
 577	if (uaddr->sa_family == AF_UNSPEC) {
 578		err = sk->sk_prot->disconnect(sk, flags);
 579		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 580		goto out;
 
 
 
 
 
 
 
 
 
 
 
 581	}
 582
 583	switch (sock->state) {
 584	default:
 585		err = -EINVAL;
 586		goto out;
 587	case SS_CONNECTED:
 588		err = -EISCONN;
 589		goto out;
 590	case SS_CONNECTING:
 591		err = -EALREADY;
 
 
 
 592		/* Fall out of switch with err, set for this state */
 593		break;
 594	case SS_UNCONNECTED:
 595		err = -EISCONN;
 596		if (sk->sk_state != TCP_CLOSE)
 597			goto out;
 598
 
 
 
 
 
 
 599		err = sk->sk_prot->connect(sk, uaddr, addr_len);
 600		if (err < 0)
 601			goto out;
 602
 603		sock->state = SS_CONNECTING;
 604
 
 
 
 605		/* Just entered SS_CONNECTING state; the only
 606		 * difference is that return value in non-blocking
 607		 * case is EINPROGRESS, rather than EALREADY.
 608		 */
 609		err = -EINPROGRESS;
 610		break;
 611	}
 612
 613	timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
 614
 615	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 616		int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
 617				tcp_sk(sk)->fastopen_req &&
 618				tcp_sk(sk)->fastopen_req->data ? 1 : 0;
 619
 620		/* Error code is set above */
 621		if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
 622			goto out;
 623
 624		err = sock_intr_errno(timeo);
 625		if (signal_pending(current))
 626			goto out;
 627	}
 628
 629	/* Connection was closed by RST, timeout, ICMP error
 630	 * or another process disconnected us.
 631	 */
 632	if (sk->sk_state == TCP_CLOSE)
 633		goto sock_error;
 634
 635	/* sk->sk_err may be not zero now, if RECVERR was ordered by user
 636	 * and error was received after socket entered established state.
 637	 * Hence, it is handled normally after connect() return successfully.
 638	 */
 639
 640	sock->state = SS_CONNECTED;
 641	err = 0;
 642out:
 643	return err;
 644
 645sock_error:
 646	err = sock_error(sk) ? : -ECONNABORTED;
 647	sock->state = SS_UNCONNECTED;
 648	if (sk->sk_prot->disconnect(sk, flags))
 649		sock->state = SS_DISCONNECTING;
 650	goto out;
 651}
 652EXPORT_SYMBOL(__inet_stream_connect);
 653
 654int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 655			int addr_len, int flags)
 656{
 657	int err;
 658
 659	lock_sock(sock->sk);
 660	err = __inet_stream_connect(sock, uaddr, addr_len, flags);
 661	release_sock(sock->sk);
 662	return err;
 663}
 664EXPORT_SYMBOL(inet_stream_connect);
 665
 666/*
 667 *	Accept a pending connection. The TCP layer now gives BSD semantics.
 668 */
 669
 670int inet_accept(struct socket *sock, struct socket *newsock, int flags)
 
 671{
 672	struct sock *sk1 = sock->sk;
 673	int err = -EINVAL;
 674	struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
 675
 676	if (!sk2)
 677		goto do_err;
 678
 679	lock_sock(sk2);
 680
 681	sock_rps_record_flow(sk2);
 682	WARN_ON(!((1 << sk2->sk_state) &
 683		  (TCPF_ESTABLISHED | TCPF_SYN_RECV |
 684		  TCPF_CLOSE_WAIT | TCPF_CLOSE)));
 685
 686	sock_graft(sk2, newsock);
 687
 688	newsock->state = SS_CONNECTED;
 689	err = 0;
 690	release_sock(sk2);
 691do_err:
 692	return err;
 693}
 694EXPORT_SYMBOL(inet_accept);
 695
 696
 697/*
 698 *	This does both peername and sockname.
 699 */
 700int inet_getname(struct socket *sock, struct sockaddr *uaddr,
 701			int *uaddr_len, int peer)
 702{
 703	struct sock *sk		= sock->sk;
 704	struct inet_sock *inet	= inet_sk(sk);
 705	DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
 706
 707	sin->sin_family = AF_INET;
 708	if (peer) {
 709		if (!inet->inet_dport ||
 710		    (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
 711		     peer == 1))
 712			return -ENOTCONN;
 713		sin->sin_port = inet->inet_dport;
 714		sin->sin_addr.s_addr = inet->inet_daddr;
 715	} else {
 716		__be32 addr = inet->inet_rcv_saddr;
 717		if (!addr)
 718			addr = inet->inet_saddr;
 719		sin->sin_port = inet->inet_sport;
 720		sin->sin_addr.s_addr = addr;
 721	}
 722	memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 723	*uaddr_len = sizeof(*sin);
 724	return 0;
 725}
 726EXPORT_SYMBOL(inet_getname);
 727
 728int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
 729		 size_t size)
 730{
 731	struct sock *sk = sock->sk;
 732
 733	sock_rps_record_flow(sk);
 734
 735	/* We may need to bind the socket. */
 736	if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
 737	    inet_autobind(sk))
 738		return -EAGAIN;
 739
 740	return sk->sk_prot->sendmsg(iocb, sk, msg, size);
 741}
 742EXPORT_SYMBOL(inet_sendmsg);
 743
 744ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
 745		      size_t size, int flags)
 746{
 747	struct sock *sk = sock->sk;
 748
 749	sock_rps_record_flow(sk);
 750
 751	/* We may need to bind the socket. */
 752	if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
 753	    inet_autobind(sk))
 754		return -EAGAIN;
 755
 756	if (sk->sk_prot->sendpage)
 757		return sk->sk_prot->sendpage(sk, page, offset, size, flags);
 758	return sock_no_sendpage(sock, page, offset, size, flags);
 759}
 760EXPORT_SYMBOL(inet_sendpage);
 761
 762int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
 763		 size_t size, int flags)
 764{
 765	struct sock *sk = sock->sk;
 766	int addr_len = 0;
 767	int err;
 768
 769	sock_rps_record_flow(sk);
 
 770
 771	err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
 772				   flags & ~MSG_DONTWAIT, &addr_len);
 773	if (err >= 0)
 774		msg->msg_namelen = addr_len;
 775	return err;
 776}
 777EXPORT_SYMBOL(inet_recvmsg);
 778
 779int inet_shutdown(struct socket *sock, int how)
 780{
 781	struct sock *sk = sock->sk;
 782	int err = 0;
 783
 784	/* This should really check to make sure
 785	 * the socket is a TCP socket. (WHY AC...)
 786	 */
 787	how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
 788		       1->2 bit 2 snds.
 789		       2->3 */
 790	if ((how & ~SHUTDOWN_MASK) || !how)	/* MAXINT->0 */
 791		return -EINVAL;
 792
 793	lock_sock(sk);
 794	if (sock->state == SS_CONNECTING) {
 795		if ((1 << sk->sk_state) &
 796		    (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
 797			sock->state = SS_DISCONNECTING;
 798		else
 799			sock->state = SS_CONNECTED;
 800	}
 801
 802	switch (sk->sk_state) {
 803	case TCP_CLOSE:
 804		err = -ENOTCONN;
 805		/* Hack to wake up other listeners, who can poll for
 806		   POLLHUP, even on eg. unconnected UDP sockets -- RR */
 
 807	default:
 808		sk->sk_shutdown |= how;
 809		if (sk->sk_prot->shutdown)
 810			sk->sk_prot->shutdown(sk, how);
 811		break;
 812
 813	/* Remaining two branches are temporary solution for missing
 814	 * close() in multithreaded environment. It is _not_ a good idea,
 815	 * but we have no choice until close() is repaired at VFS level.
 816	 */
 817	case TCP_LISTEN:
 818		if (!(how & RCV_SHUTDOWN))
 819			break;
 820		/* Fall through */
 821	case TCP_SYN_SENT:
 822		err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
 823		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 824		break;
 825	}
 826
 827	/* Wake up anyone sleeping in poll. */
 828	sk->sk_state_change(sk);
 829	release_sock(sk);
 830	return err;
 831}
 832EXPORT_SYMBOL(inet_shutdown);
 833
 834/*
 835 *	ioctl() calls you can issue on an INET socket. Most of these are
 836 *	device configuration and stuff and very rarely used. Some ioctls
 837 *	pass on to the socket itself.
 838 *
 839 *	NOTE: I like the idea of a module for the config stuff. ie ifconfig
 840 *	loads the devconfigure module does its configuring and unloads it.
 841 *	There's a good 20K of config code hanging around the kernel.
 842 */
 843
 844int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 845{
 846	struct sock *sk = sock->sk;
 847	int err = 0;
 848	struct net *net = sock_net(sk);
 
 
 
 849
 850	switch (cmd) {
 851	case SIOCGSTAMP:
 852		err = sock_get_timestamp(sk, (struct timeval __user *)arg);
 853		break;
 854	case SIOCGSTAMPNS:
 855		err = sock_get_timestampns(sk, (struct timespec __user *)arg);
 856		break;
 857	case SIOCADDRT:
 858	case SIOCDELRT:
 
 
 
 
 859	case SIOCRTMSG:
 860		err = ip_rt_ioctl(net, cmd, (void __user *)arg);
 861		break;
 862	case SIOCDARP:
 863	case SIOCGARP:
 864	case SIOCSARP:
 865		err = arp_ioctl(net, cmd, (void __user *)arg);
 866		break;
 867	case SIOCGIFADDR:
 868	case SIOCSIFADDR:
 869	case SIOCGIFBRDADDR:
 870	case SIOCSIFBRDADDR:
 871	case SIOCGIFNETMASK:
 872	case SIOCSIFNETMASK:
 873	case SIOCGIFDSTADDR:
 
 
 
 
 
 
 
 
 
 
 
 874	case SIOCSIFDSTADDR:
 875	case SIOCSIFPFLAGS:
 876	case SIOCGIFPFLAGS:
 877	case SIOCSIFFLAGS:
 878		err = devinet_ioctl(net, cmd, (void __user *)arg);
 
 
 879		break;
 880	default:
 881		if (sk->sk_prot->ioctl)
 882			err = sk->sk_prot->ioctl(sk, cmd, arg);
 883		else
 884			err = -ENOIOCTLCMD;
 885		break;
 886	}
 887	return err;
 888}
 889EXPORT_SYMBOL(inet_ioctl);
 890
 891#ifdef CONFIG_COMPAT
 892static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 893{
 894	struct sock *sk = sock->sk;
 895	int err = -ENOIOCTLCMD;
 896
 897	if (sk->sk_prot->compat_ioctl)
 898		err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
 899
 900	return err;
 901}
 902#endif
 903
 904const struct proto_ops inet_stream_ops = {
 905	.family		   = PF_INET,
 906	.owner		   = THIS_MODULE,
 907	.release	   = inet_release,
 908	.bind		   = inet_bind,
 909	.connect	   = inet_stream_connect,
 910	.socketpair	   = sock_no_socketpair,
 911	.accept		   = inet_accept,
 912	.getname	   = inet_getname,
 913	.poll		   = tcp_poll,
 914	.ioctl		   = inet_ioctl,
 915	.listen		   = inet_listen,
 916	.shutdown	   = inet_shutdown,
 917	.setsockopt	   = sock_common_setsockopt,
 918	.getsockopt	   = sock_common_getsockopt,
 919	.sendmsg	   = inet_sendmsg,
 920	.recvmsg	   = inet_recvmsg,
 921	.mmap		   = sock_no_mmap,
 922	.sendpage	   = inet_sendpage,
 923	.splice_read	   = tcp_splice_read,
 
 
 
 
 924#ifdef CONFIG_COMPAT
 925	.compat_setsockopt = compat_sock_common_setsockopt,
 926	.compat_getsockopt = compat_sock_common_getsockopt,
 927	.compat_ioctl	   = inet_compat_ioctl,
 928#endif
 929};
 930EXPORT_SYMBOL(inet_stream_ops);
 931
 932const struct proto_ops inet_dgram_ops = {
 933	.family		   = PF_INET,
 934	.owner		   = THIS_MODULE,
 935	.release	   = inet_release,
 936	.bind		   = inet_bind,
 937	.connect	   = inet_dgram_connect,
 938	.socketpair	   = sock_no_socketpair,
 939	.accept		   = sock_no_accept,
 940	.getname	   = inet_getname,
 941	.poll		   = udp_poll,
 942	.ioctl		   = inet_ioctl,
 943	.listen		   = sock_no_listen,
 944	.shutdown	   = inet_shutdown,
 945	.setsockopt	   = sock_common_setsockopt,
 946	.getsockopt	   = sock_common_getsockopt,
 947	.sendmsg	   = inet_sendmsg,
 948	.recvmsg	   = inet_recvmsg,
 949	.mmap		   = sock_no_mmap,
 950	.sendpage	   = inet_sendpage,
 
 951#ifdef CONFIG_COMPAT
 952	.compat_setsockopt = compat_sock_common_setsockopt,
 953	.compat_getsockopt = compat_sock_common_getsockopt,
 954	.compat_ioctl	   = inet_compat_ioctl,
 955#endif
 956};
 957EXPORT_SYMBOL(inet_dgram_ops);
 958
 959/*
 960 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
 961 * udp_poll
 962 */
 963static const struct proto_ops inet_sockraw_ops = {
 964	.family		   = PF_INET,
 965	.owner		   = THIS_MODULE,
 966	.release	   = inet_release,
 967	.bind		   = inet_bind,
 968	.connect	   = inet_dgram_connect,
 969	.socketpair	   = sock_no_socketpair,
 970	.accept		   = sock_no_accept,
 971	.getname	   = inet_getname,
 972	.poll		   = datagram_poll,
 973	.ioctl		   = inet_ioctl,
 974	.listen		   = sock_no_listen,
 975	.shutdown	   = inet_shutdown,
 976	.setsockopt	   = sock_common_setsockopt,
 977	.getsockopt	   = sock_common_getsockopt,
 978	.sendmsg	   = inet_sendmsg,
 979	.recvmsg	   = inet_recvmsg,
 980	.mmap		   = sock_no_mmap,
 981	.sendpage	   = inet_sendpage,
 982#ifdef CONFIG_COMPAT
 983	.compat_setsockopt = compat_sock_common_setsockopt,
 984	.compat_getsockopt = compat_sock_common_getsockopt,
 985	.compat_ioctl	   = inet_compat_ioctl,
 986#endif
 987};
 988
 989static const struct net_proto_family inet_family_ops = {
 990	.family = PF_INET,
 991	.create = inet_create,
 992	.owner	= THIS_MODULE,
 993};
 994
 995/* Upon startup we insert all the elements in inetsw_array[] into
 996 * the linked list inetsw.
 997 */
 998static struct inet_protosw inetsw_array[] =
 999{
1000	{
1001		.type =       SOCK_STREAM,
1002		.protocol =   IPPROTO_TCP,
1003		.prot =       &tcp_prot,
1004		.ops =        &inet_stream_ops,
1005		.no_check =   0,
1006		.flags =      INET_PROTOSW_PERMANENT |
1007			      INET_PROTOSW_ICSK,
1008	},
1009
1010	{
1011		.type =       SOCK_DGRAM,
1012		.protocol =   IPPROTO_UDP,
1013		.prot =       &udp_prot,
1014		.ops =        &inet_dgram_ops,
1015		.no_check =   UDP_CSUM_DEFAULT,
1016		.flags =      INET_PROTOSW_PERMANENT,
1017       },
1018
1019       {
1020		.type =       SOCK_DGRAM,
1021		.protocol =   IPPROTO_ICMP,
1022		.prot =       &ping_prot,
1023		.ops =        &inet_dgram_ops,
1024		.no_check =   UDP_CSUM_DEFAULT,
1025		.flags =      INET_PROTOSW_REUSE,
1026       },
1027
1028       {
1029	       .type =       SOCK_RAW,
1030	       .protocol =   IPPROTO_IP,	/* wild card */
1031	       .prot =       &raw_prot,
1032	       .ops =        &inet_sockraw_ops,
1033	       .no_check =   UDP_CSUM_DEFAULT,
1034	       .flags =      INET_PROTOSW_REUSE,
1035       }
1036};
1037
1038#define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1039
1040void inet_register_protosw(struct inet_protosw *p)
1041{
1042	struct list_head *lh;
1043	struct inet_protosw *answer;
1044	int protocol = p->protocol;
1045	struct list_head *last_perm;
1046
1047	spin_lock_bh(&inetsw_lock);
1048
1049	if (p->type >= SOCK_MAX)
1050		goto out_illegal;
1051
1052	/* If we are trying to override a permanent protocol, bail. */
1053	answer = NULL;
1054	last_perm = &inetsw[p->type];
1055	list_for_each(lh, &inetsw[p->type]) {
1056		answer = list_entry(lh, struct inet_protosw, list);
1057
1058		/* Check only the non-wild match. */
1059		if (INET_PROTOSW_PERMANENT & answer->flags) {
1060			if (protocol == answer->protocol)
1061				break;
1062			last_perm = lh;
1063		}
1064
1065		answer = NULL;
1066	}
1067	if (answer)
1068		goto out_permanent;
1069
1070	/* Add the new entry after the last permanent entry if any, so that
1071	 * the new entry does not override a permanent entry when matched with
1072	 * a wild-card protocol. But it is allowed to override any existing
1073	 * non-permanent entry.  This means that when we remove this entry, the
1074	 * system automatically returns to the old behavior.
1075	 */
1076	list_add_rcu(&p->list, last_perm);
1077out:
1078	spin_unlock_bh(&inetsw_lock);
1079
1080	return;
1081
1082out_permanent:
1083	pr_err("Attempt to override permanent protocol %d\n", protocol);
1084	goto out;
1085
1086out_illegal:
1087	pr_err("Ignoring attempt to register invalid socket type %d\n",
1088	       p->type);
1089	goto out;
1090}
1091EXPORT_SYMBOL(inet_register_protosw);
1092
1093void inet_unregister_protosw(struct inet_protosw *p)
1094{
1095	if (INET_PROTOSW_PERMANENT & p->flags) {
1096		pr_err("Attempt to unregister permanent protocol %d\n",
1097		       p->protocol);
1098	} else {
1099		spin_lock_bh(&inetsw_lock);
1100		list_del_rcu(&p->list);
1101		spin_unlock_bh(&inetsw_lock);
1102
1103		synchronize_net();
1104	}
1105}
1106EXPORT_SYMBOL(inet_unregister_protosw);
1107
1108/*
1109 *      Shall we try to damage output packets if routing dev changes?
1110 */
1111
1112int sysctl_ip_dynaddr __read_mostly;
1113
1114static int inet_sk_reselect_saddr(struct sock *sk)
1115{
1116	struct inet_sock *inet = inet_sk(sk);
1117	__be32 old_saddr = inet->inet_saddr;
1118	__be32 daddr = inet->inet_daddr;
1119	struct flowi4 *fl4;
1120	struct rtable *rt;
1121	__be32 new_saddr;
1122	struct ip_options_rcu *inet_opt;
1123
1124	inet_opt = rcu_dereference_protected(inet->inet_opt,
1125					     sock_owned_by_user(sk));
1126	if (inet_opt && inet_opt->opt.srr)
1127		daddr = inet_opt->opt.faddr;
1128
1129	/* Query new route. */
1130	fl4 = &inet->cork.fl.u.ip4;
1131	rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1132			      sk->sk_bound_dev_if, sk->sk_protocol,
1133			      inet->inet_sport, inet->inet_dport, sk);
1134	if (IS_ERR(rt))
1135		return PTR_ERR(rt);
1136
1137	sk_setup_caps(sk, &rt->dst);
1138
1139	new_saddr = fl4->saddr;
1140
1141	if (new_saddr == old_saddr)
1142		return 0;
1143
1144	if (sysctl_ip_dynaddr > 1) {
1145		pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1146			__func__, &old_saddr, &new_saddr);
1147	}
1148
1149	inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1150
1151	/*
1152	 * XXX The only one ugly spot where we need to
1153	 * XXX really change the sockets identity after
1154	 * XXX it has entered the hashes. -DaveM
1155	 *
1156	 * Besides that, it does not check for connection
1157	 * uniqueness. Wait for troubles.
1158	 */
1159	__sk_prot_rehash(sk);
1160	return 0;
1161}
1162
1163int inet_sk_rebuild_header(struct sock *sk)
1164{
1165	struct inet_sock *inet = inet_sk(sk);
1166	struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1167	__be32 daddr;
1168	struct ip_options_rcu *inet_opt;
1169	struct flowi4 *fl4;
1170	int err;
1171
1172	/* Route is OK, nothing to do. */
1173	if (rt)
1174		return 0;
1175
1176	/* Reroute. */
1177	rcu_read_lock();
1178	inet_opt = rcu_dereference(inet->inet_opt);
1179	daddr = inet->inet_daddr;
1180	if (inet_opt && inet_opt->opt.srr)
1181		daddr = inet_opt->opt.faddr;
1182	rcu_read_unlock();
1183	fl4 = &inet->cork.fl.u.ip4;
1184	rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1185				   inet->inet_dport, inet->inet_sport,
1186				   sk->sk_protocol, RT_CONN_FLAGS(sk),
1187				   sk->sk_bound_dev_if);
1188	if (!IS_ERR(rt)) {
1189		err = 0;
1190		sk_setup_caps(sk, &rt->dst);
1191	} else {
1192		err = PTR_ERR(rt);
1193
1194		/* Routing failed... */
1195		sk->sk_route_caps = 0;
1196		/*
1197		 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1198		 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1199		 */
1200		if (!sysctl_ip_dynaddr ||
1201		    sk->sk_state != TCP_SYN_SENT ||
1202		    (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1203		    (err = inet_sk_reselect_saddr(sk)) != 0)
1204			sk->sk_err_soft = -err;
1205	}
1206
1207	return err;
1208}
1209EXPORT_SYMBOL(inet_sk_rebuild_header);
1210
1211static int inet_gso_send_check(struct sk_buff *skb)
1212{
1213	const struct net_offload *ops;
1214	const struct iphdr *iph;
1215	int proto;
1216	int ihl;
1217	int err = -EINVAL;
1218
1219	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1220		goto out;
1221
1222	iph = ip_hdr(skb);
1223	ihl = iph->ihl * 4;
1224	if (ihl < sizeof(*iph))
1225		goto out;
1226
1227	proto = iph->protocol;
1228
1229	/* Warning: after this point, iph might be no longer valid */
1230	if (unlikely(!pskb_may_pull(skb, ihl)))
1231		goto out;
1232	__skb_pull(skb, ihl);
1233
1234	skb_reset_transport_header(skb);
1235	err = -EPROTONOSUPPORT;
1236
1237	ops = rcu_dereference(inet_offloads[proto]);
1238	if (likely(ops && ops->callbacks.gso_send_check))
1239		err = ops->callbacks.gso_send_check(skb);
1240
1241out:
1242	return err;
 
 
1243}
1244
1245static struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1246					netdev_features_t features)
1247{
 
1248	struct sk_buff *segs = ERR_PTR(-EINVAL);
1249	const struct net_offload *ops;
1250	unsigned int offset = 0;
1251	bool udpfrag, encap;
1252	struct iphdr *iph;
1253	int proto;
1254	int nhoff;
1255	int ihl;
1256	int id;
1257
1258	if (unlikely(skb_shinfo(skb)->gso_type &
1259		     ~(SKB_GSO_TCPV4 |
1260		       SKB_GSO_UDP |
1261		       SKB_GSO_DODGY |
1262		       SKB_GSO_TCP_ECN |
1263		       SKB_GSO_GRE |
1264		       SKB_GSO_IPIP |
1265		       SKB_GSO_SIT |
1266		       SKB_GSO_TCPV6 |
1267		       SKB_GSO_UDP_TUNNEL |
1268		       SKB_GSO_MPLS |
1269		       0)))
1270		goto out;
1271
1272	skb_reset_network_header(skb);
1273	nhoff = skb_network_header(skb) - skb_mac_header(skb);
1274	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1275		goto out;
1276
1277	iph = ip_hdr(skb);
1278	ihl = iph->ihl * 4;
1279	if (ihl < sizeof(*iph))
1280		goto out;
1281
1282	id = ntohs(iph->id);
1283	proto = iph->protocol;
1284
1285	/* Warning: after this point, iph might be no longer valid */
1286	if (unlikely(!pskb_may_pull(skb, ihl)))
1287		goto out;
1288	__skb_pull(skb, ihl);
1289
1290	encap = SKB_GSO_CB(skb)->encap_level > 0;
1291	if (encap)
1292		features = skb->dev->hw_enc_features & netif_skb_features(skb);
1293	SKB_GSO_CB(skb)->encap_level += ihl;
1294
1295	skb_reset_transport_header(skb);
1296
1297	segs = ERR_PTR(-EPROTONOSUPPORT);
1298
1299	if (skb->encapsulation &&
1300	    skb_shinfo(skb)->gso_type & (SKB_GSO_SIT|SKB_GSO_IPIP))
1301		udpfrag = proto == IPPROTO_UDP && encap;
1302	else
1303		udpfrag = proto == IPPROTO_UDP && !skb->encapsulation;
 
 
 
1304
1305	ops = rcu_dereference(inet_offloads[proto]);
1306	if (likely(ops && ops->callbacks.gso_segment))
1307		segs = ops->callbacks.gso_segment(skb, features);
1308
1309	if (IS_ERR_OR_NULL(segs))
1310		goto out;
1311
 
 
1312	skb = segs;
1313	do {
1314		iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1315		if (udpfrag) {
1316			iph->id = htons(id);
1317			iph->frag_off = htons(offset >> 3);
1318			if (skb->next != NULL)
1319				iph->frag_off |= htons(IP_MF);
1320			offset += skb->len - nhoff - ihl;
 
 
 
 
 
 
 
 
 
 
 
 
 
1321		} else {
1322			iph->id = htons(id++);
 
 
1323		}
1324		iph->tot_len = htons(skb->len - nhoff);
1325		ip_send_check(iph);
1326		if (encap)
1327			skb_reset_inner_headers(skb);
1328		skb->network_header = (u8 *)iph - skb->head;
1329	} while ((skb = skb->next));
1330
1331out:
1332	return segs;
1333}
 
1334
1335static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1336					 struct sk_buff *skb)
1337{
1338	const struct net_offload *ops;
1339	struct sk_buff **pp = NULL;
1340	struct sk_buff *p;
1341	const struct iphdr *iph;
1342	unsigned int hlen;
1343	unsigned int off;
1344	unsigned int id;
1345	int flush = 1;
1346	int proto;
1347
1348	off = skb_gro_offset(skb);
1349	hlen = off + sizeof(*iph);
1350	iph = skb_gro_header_fast(skb, off);
1351	if (skb_gro_header_hard(skb, hlen)) {
1352		iph = skb_gro_header_slow(skb, hlen, off);
1353		if (unlikely(!iph))
1354			goto out;
1355	}
1356
1357	proto = iph->protocol;
1358
1359	rcu_read_lock();
1360	ops = rcu_dereference(inet_offloads[proto]);
1361	if (!ops || !ops->callbacks.gro_receive)
1362		goto out_unlock;
1363
1364	if (*(u8 *)iph != 0x45)
1365		goto out_unlock;
1366
 
 
 
1367	if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1368		goto out_unlock;
1369
1370	id = ntohl(*(__be32 *)&iph->id);
1371	flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1372	id >>= 16;
1373
1374	for (p = *head; p; p = p->next) {
1375		struct iphdr *iph2;
 
1376
1377		if (!NAPI_GRO_CB(p)->same_flow)
1378			continue;
1379
1380		iph2 = (struct iphdr *)(p->data + off);
1381		/* The above works because, with the exception of the top
1382		 * (inner most) layer, we only aggregate pkts with the same
1383		 * hdr length so all the hdrs we'll need to verify will start
1384		 * at the same offset.
1385		 */
1386		if ((iph->protocol ^ iph2->protocol) |
1387		    ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1388		    ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1389			NAPI_GRO_CB(p)->same_flow = 0;
1390			continue;
1391		}
1392
1393		/* All fields must match except length and checksum. */
1394		NAPI_GRO_CB(p)->flush |=
1395			(iph->ttl ^ iph2->ttl) |
1396			(iph->tos ^ iph2->tos) |
1397			((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1398
1399		/* Save the IP ID check to be included later when we get to
1400		 * the transport layer so only the inner most IP ID is checked.
1401		 * This is because some GSO/TSO implementations do not
1402		 * correctly increment the IP ID for the outer hdrs.
1403		 */
1404		NAPI_GRO_CB(p)->flush_id =
1405			    ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1406		NAPI_GRO_CB(p)->flush |= flush;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1407	}
1408
 
1409	NAPI_GRO_CB(skb)->flush |= flush;
1410	skb_set_network_header(skb, off);
1411	/* The above will be needed by the transport layer if there is one
1412	 * immediately following this IP hdr.
1413	 */
1414
 
 
 
1415	skb_gro_pull(skb, sizeof(*iph));
1416	skb_set_transport_header(skb, skb_gro_offset(skb));
1417
1418	pp = ops->callbacks.gro_receive(head, skb);
1419
1420out_unlock:
1421	rcu_read_unlock();
1422
1423out:
1424	NAPI_GRO_CB(skb)->flush |= flush;
1425
1426	return pp;
1427}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1428
1429static int inet_gro_complete(struct sk_buff *skb, int nhoff)
1430{
1431	__be16 newlen = htons(skb->len - nhoff);
1432	struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1433	const struct net_offload *ops;
1434	int proto = iph->protocol;
1435	int err = -ENOSYS;
1436
 
 
 
 
 
1437	csum_replace2(&iph->check, iph->tot_len, newlen);
1438	iph->tot_len = newlen;
1439
1440	rcu_read_lock();
1441	ops = rcu_dereference(inet_offloads[proto]);
1442	if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1443		goto out_unlock;
1444
1445	/* Only need to add sizeof(*iph) to get to the next hdr below
1446	 * because any hdr with option will have been flushed in
1447	 * inet_gro_receive().
1448	 */
1449	err = ops->callbacks.gro_complete(skb, nhoff + sizeof(*iph));
1450
1451out_unlock:
1452	rcu_read_unlock();
1453
1454	return err;
1455}
 
 
 
 
 
 
 
 
1456
1457int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1458			 unsigned short type, unsigned char protocol,
1459			 struct net *net)
1460{
1461	struct socket *sock;
1462	int rc = sock_create_kern(family, type, protocol, &sock);
1463
1464	if (rc == 0) {
1465		*sk = sock->sk;
1466		(*sk)->sk_allocation = GFP_ATOMIC;
1467		/*
1468		 * Unhash it so that IP input processing does not even see it,
1469		 * we do not wish this socket to see incoming packets.
1470		 */
1471		(*sk)->sk_prot->unhash(*sk);
1472
1473		sk_change_net(*sk, net);
1474	}
1475	return rc;
1476}
1477EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1478
1479unsigned long snmp_fold_field(void __percpu *mib[], int offt)
 
 
 
 
 
 
1480{
1481	unsigned long res = 0;
1482	int i, j;
1483
1484	for_each_possible_cpu(i) {
1485		for (j = 0; j < SNMP_ARRAY_SZ; j++)
1486			res += *(((unsigned long *) per_cpu_ptr(mib[j], i)) + offt);
1487	}
1488	return res;
1489}
1490EXPORT_SYMBOL_GPL(snmp_fold_field);
1491
1492#if BITS_PER_LONG==32
1493
1494u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1495{
1496	u64 res = 0;
1497	int cpu;
1498
1499	for_each_possible_cpu(cpu) {
1500		void *bhptr;
1501		struct u64_stats_sync *syncp;
1502		u64 v;
1503		unsigned int start;
1504
1505		bhptr = per_cpu_ptr(mib[0], cpu);
1506		syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1507		do {
1508			start = u64_stats_fetch_begin_irq(syncp);
1509			v = *(((u64 *) bhptr) + offt);
1510		} while (u64_stats_fetch_retry_irq(syncp, start));
1511
1512		res += v;
1513	}
1514	return res;
1515}
1516EXPORT_SYMBOL_GPL(snmp_fold_field64);
1517#endif
1518
1519int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align)
1520{
1521	BUG_ON(ptr == NULL);
1522	ptr[0] = __alloc_percpu(mibsize, align);
1523	if (!ptr[0])
1524		return -ENOMEM;
1525
1526#if SNMP_ARRAY_SZ == 2
1527	ptr[1] = __alloc_percpu(mibsize, align);
1528	if (!ptr[1]) {
1529		free_percpu(ptr[0]);
1530		ptr[0] = NULL;
1531		return -ENOMEM;
1532	}
1533#endif
1534	return 0;
1535}
1536EXPORT_SYMBOL_GPL(snmp_mib_init);
1537
1538#ifdef CONFIG_IP_MULTICAST
1539static const struct net_protocol igmp_protocol = {
1540	.handler =	igmp_rcv,
1541	.netns_ok =	1,
1542};
1543#endif
1544
1545static const struct net_protocol tcp_protocol = {
 
 
 
1546	.early_demux	=	tcp_v4_early_demux,
 
1547	.handler	=	tcp_v4_rcv,
1548	.err_handler	=	tcp_v4_err,
1549	.no_policy	=	1,
1550	.netns_ok	=	1,
1551	.icmp_strict_tag_validation = 1,
1552};
1553
1554static const struct net_protocol udp_protocol = {
 
 
 
1555	.early_demux =	udp_v4_early_demux,
 
1556	.handler =	udp_rcv,
1557	.err_handler =	udp_err,
1558	.no_policy =	1,
1559	.netns_ok =	1,
1560};
1561
1562static const struct net_protocol icmp_protocol = {
1563	.handler =	icmp_rcv,
1564	.err_handler =	icmp_err,
1565	.no_policy =	1,
1566	.netns_ok =	1,
1567};
1568
1569static __net_init int ipv4_mib_init_net(struct net *net)
1570{
1571	int i;
1572
1573	if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1574			  sizeof(struct tcp_mib),
1575			  __alignof__(struct tcp_mib)) < 0)
1576		goto err_tcp_mib;
1577	if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1578			  sizeof(struct ipstats_mib),
1579			  __alignof__(struct ipstats_mib)) < 0)
1580		goto err_ip_mib;
1581
1582	for_each_possible_cpu(i) {
1583		struct ipstats_mib *af_inet_stats;
1584		af_inet_stats = per_cpu_ptr(net->mib.ip_statistics[0], i);
1585		u64_stats_init(&af_inet_stats->syncp);
1586#if SNMP_ARRAY_SZ == 2
1587		af_inet_stats = per_cpu_ptr(net->mib.ip_statistics[1], i);
1588		u64_stats_init(&af_inet_stats->syncp);
1589#endif
1590	}
1591
1592	if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1593			  sizeof(struct linux_mib),
1594			  __alignof__(struct linux_mib)) < 0)
1595		goto err_net_mib;
1596	if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1597			  sizeof(struct udp_mib),
1598			  __alignof__(struct udp_mib)) < 0)
1599		goto err_udp_mib;
1600	if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1601			  sizeof(struct udp_mib),
1602			  __alignof__(struct udp_mib)) < 0)
1603		goto err_udplite_mib;
1604	if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1605			  sizeof(struct icmp_mib),
1606			  __alignof__(struct icmp_mib)) < 0)
1607		goto err_icmp_mib;
1608	net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1609					      GFP_KERNEL);
1610	if (!net->mib.icmpmsg_statistics)
1611		goto err_icmpmsg_mib;
1612
1613	tcp_mib_init(net);
1614	return 0;
1615
1616err_icmpmsg_mib:
1617	snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1618err_icmp_mib:
1619	snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1620err_udplite_mib:
1621	snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1622err_udp_mib:
1623	snmp_mib_free((void __percpu **)net->mib.net_statistics);
1624err_net_mib:
1625	snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1626err_ip_mib:
1627	snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1628err_tcp_mib:
1629	return -ENOMEM;
1630}
1631
1632static __net_exit void ipv4_mib_exit_net(struct net *net)
1633{
1634	kfree(net->mib.icmpmsg_statistics);
1635	snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1636	snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1637	snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1638	snmp_mib_free((void __percpu **)net->mib.net_statistics);
1639	snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1640	snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1641}
1642
1643static __net_initdata struct pernet_operations ipv4_mib_ops = {
1644	.init = ipv4_mib_init_net,
1645	.exit = ipv4_mib_exit_net,
1646};
1647
1648static int __init init_ipv4_mibs(void)
1649{
1650	return register_pernet_subsys(&ipv4_mib_ops);
1651}
1652
1653static __net_init int inet_init_net(struct net *net)
1654{
1655	/*
1656	 * Set defaults for local port range
1657	 */
1658	seqlock_init(&net->ipv4.ip_local_ports.lock);
1659	net->ipv4.ip_local_ports.range[0] =  32768;
1660	net->ipv4.ip_local_ports.range[1] =  61000;
1661
1662	seqlock_init(&net->ipv4.ping_group_range.lock);
1663	/*
1664	 * Sane defaults - nobody may create ping sockets.
1665	 * Boot scripts should set this to distro-specific group.
1666	 */
1667	net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1668	net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1669	return 0;
1670}
1671
1672static __net_exit void inet_exit_net(struct net *net)
1673{
1674}
1675
1676static __net_initdata struct pernet_operations af_inet_ops = {
1677	.init = inet_init_net,
1678	.exit = inet_exit_net,
1679};
1680
1681static int __init init_inet_pernet_ops(void)
1682{
1683	return register_pernet_subsys(&af_inet_ops);
1684}
1685
1686static int ipv4_proc_init(void);
1687
1688/*
1689 *	IP protocol layer initialiser
1690 */
1691
1692static struct packet_offload ip_packet_offload __read_mostly = {
1693	.type = cpu_to_be16(ETH_P_IP),
1694	.callbacks = {
1695		.gso_send_check = inet_gso_send_check,
1696		.gso_segment = inet_gso_segment,
1697		.gro_receive = inet_gro_receive,
1698		.gro_complete = inet_gro_complete,
1699	},
1700};
1701
1702static const struct net_offload ipip_offload = {
1703	.callbacks = {
1704		.gso_send_check = inet_gso_send_check,
1705		.gso_segment	= inet_gso_segment,
 
 
1706	},
1707};
1708
 
 
 
 
 
1709static int __init ipv4_offload_init(void)
1710{
1711	/*
1712	 * Add offloads
1713	 */
1714	if (udpv4_offload_init() < 0)
1715		pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1716	if (tcpv4_offload_init() < 0)
1717		pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
 
 
1718
1719	dev_add_offload(&ip_packet_offload);
1720	inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1721	return 0;
1722}
1723
1724fs_initcall(ipv4_offload_init);
1725
1726static struct packet_type ip_packet_type __read_mostly = {
1727	.type = cpu_to_be16(ETH_P_IP),
1728	.func = ip_rcv,
1729};
1730
1731static int __init inet_init(void)
1732{
1733	struct inet_protosw *q;
1734	struct list_head *r;
1735	int rc = -EINVAL;
1736
1737	BUILD_BUG_ON(sizeof(struct inet_skb_parm) > FIELD_SIZEOF(struct sk_buff, cb));
1738
1739	sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL);
1740	if (!sysctl_local_reserved_ports)
1741		goto out;
1742
1743	rc = proto_register(&tcp_prot, 1);
1744	if (rc)
1745		goto out_free_reserved_ports;
1746
1747	rc = proto_register(&udp_prot, 1);
1748	if (rc)
1749		goto out_unregister_tcp_proto;
1750
1751	rc = proto_register(&raw_prot, 1);
1752	if (rc)
1753		goto out_unregister_udp_proto;
1754
1755	rc = proto_register(&ping_prot, 1);
1756	if (rc)
1757		goto out_unregister_raw_proto;
1758
1759	/*
1760	 *	Tell SOCKET that we are alive...
1761	 */
1762
1763	(void)sock_register(&inet_family_ops);
1764
1765#ifdef CONFIG_SYSCTL
1766	ip_static_sysctl_init();
1767#endif
1768
1769	/*
1770	 *	Add all the base protocols.
1771	 */
1772
1773	if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1774		pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1775	if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1776		pr_crit("%s: Cannot add UDP protocol\n", __func__);
1777	if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1778		pr_crit("%s: Cannot add TCP protocol\n", __func__);
1779#ifdef CONFIG_IP_MULTICAST
1780	if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1781		pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1782#endif
1783
1784	/* Register the socket-side information for inet_create. */
1785	for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1786		INIT_LIST_HEAD(r);
1787
1788	for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1789		inet_register_protosw(q);
1790
1791	/*
1792	 *	Set the ARP module up
1793	 */
1794
1795	arp_init();
1796
1797	/*
1798	 *	Set the IP module up
1799	 */
1800
1801	ip_init();
1802
1803	tcp_v4_init();
1804
1805	/* Setup TCP slab cache for open requests. */
1806	tcp_init();
1807
1808	/* Setup UDP memory threshold */
1809	udp_init();
1810
1811	/* Add UDP-Lite (RFC 3828) */
1812	udplite4_register();
1813
1814	ping_init();
1815
1816	/*
1817	 *	Set the ICMP layer up
1818	 */
1819
1820	if (icmp_init() < 0)
1821		panic("Failed to create the ICMP control socket.\n");
1822
1823	/*
1824	 *	Initialise the multicast router
1825	 */
1826#if defined(CONFIG_IP_MROUTE)
1827	if (ip_mr_init())
1828		pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1829#endif
1830
1831	if (init_inet_pernet_ops())
1832		pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
1833	/*
1834	 *	Initialise per-cpu ipv4 mibs
1835	 */
1836
1837	if (init_ipv4_mibs())
1838		pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1839
1840	ipv4_proc_init();
1841
1842	ipfrag_init();
1843
1844	dev_add_pack(&ip_packet_type);
1845
 
 
1846	rc = 0;
1847out:
1848	return rc;
1849out_unregister_raw_proto:
1850	proto_unregister(&raw_prot);
1851out_unregister_udp_proto:
1852	proto_unregister(&udp_prot);
1853out_unregister_tcp_proto:
1854	proto_unregister(&tcp_prot);
1855out_free_reserved_ports:
1856	kfree(sysctl_local_reserved_ports);
1857	goto out;
1858}
1859
1860fs_initcall(inet_init);
1861
1862/* ------------------------------------------------------------------------ */
1863
1864#ifdef CONFIG_PROC_FS
1865static int __init ipv4_proc_init(void)
1866{
1867	int rc = 0;
1868
1869	if (raw_proc_init())
1870		goto out_raw;
1871	if (tcp4_proc_init())
1872		goto out_tcp;
1873	if (udp4_proc_init())
1874		goto out_udp;
1875	if (ping_proc_init())
1876		goto out_ping;
1877	if (ip_misc_proc_init())
1878		goto out_misc;
1879out:
1880	return rc;
1881out_misc:
1882	ping_proc_exit();
1883out_ping:
1884	udp4_proc_exit();
1885out_udp:
1886	tcp4_proc_exit();
1887out_tcp:
1888	raw_proc_exit();
1889out_raw:
1890	rc = -ENOMEM;
1891	goto out;
1892}
1893
1894#else /* CONFIG_PROC_FS */
1895static int __init ipv4_proc_init(void)
1896{
1897	return 0;
1898}
1899#endif /* CONFIG_PROC_FS */
1900
1901MODULE_ALIAS_NETPROTO(PF_INET);
1902
v4.17
   1/*
   2 * INET		An implementation of the TCP/IP protocol suite for the LINUX
   3 *		operating system.  INET is implemented using the  BSD Socket
   4 *		interface as the means of communication with the user level.
   5 *
   6 *		PF_INET protocol family socket handler.
   7 *
   8 * Authors:	Ross Biro
   9 *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  10 *		Florian La Roche, <flla@stud.uni-sb.de>
  11 *		Alan Cox, <A.Cox@swansea.ac.uk>
  12 *
  13 * Changes (see also sock.c)
  14 *
  15 *		piggy,
  16 *		Karl Knutson	:	Socket protocol table
  17 *		A.N.Kuznetsov	:	Socket death error in accept().
  18 *		John Richardson :	Fix non blocking error in connect()
  19 *					so sockets that fail to connect
  20 *					don't return -EINPROGRESS.
  21 *		Alan Cox	:	Asynchronous I/O support
  22 *		Alan Cox	:	Keep correct socket pointer on sock
  23 *					structures
  24 *					when accept() ed
  25 *		Alan Cox	:	Semantics of SO_LINGER aren't state
  26 *					moved to close when you look carefully.
  27 *					With this fixed and the accept bug fixed
  28 *					some RPC stuff seems happier.
  29 *		Niibe Yutaka	:	4.4BSD style write async I/O
  30 *		Alan Cox,
  31 *		Tony Gale 	:	Fixed reuse semantics.
  32 *		Alan Cox	:	bind() shouldn't abort existing but dead
  33 *					sockets. Stops FTP netin:.. I hope.
  34 *		Alan Cox	:	bind() works correctly for RAW sockets.
  35 *					Note that FreeBSD at least was broken
  36 *					in this respect so be careful with
  37 *					compatibility tests...
  38 *		Alan Cox	:	routing cache support
  39 *		Alan Cox	:	memzero the socket structure for
  40 *					compactness.
  41 *		Matt Day	:	nonblock connect error handler
  42 *		Alan Cox	:	Allow large numbers of pending sockets
  43 *					(eg for big web sites), but only if
  44 *					specifically application requested.
  45 *		Alan Cox	:	New buffering throughout IP. Used
  46 *					dumbly.
  47 *		Alan Cox	:	New buffering now used smartly.
  48 *		Alan Cox	:	BSD rather than common sense
  49 *					interpretation of listen.
  50 *		Germano Caronni	:	Assorted small races.
  51 *		Alan Cox	:	sendmsg/recvmsg basic support.
  52 *		Alan Cox	:	Only sendmsg/recvmsg now supported.
  53 *		Alan Cox	:	Locked down bind (see security list).
  54 *		Alan Cox	:	Loosened bind a little.
  55 *		Mike McLagan	:	ADD/DEL DLCI Ioctls
  56 *	Willy Konynenberg	:	Transparent proxying support.
  57 *		David S. Miller	:	New socket lookup architecture.
  58 *					Some other random speedups.
  59 *		Cyrus Durgin	:	Cleaned up file for kmod hacks.
  60 *		Andi Kleen	:	Fix inet_stream_connect TCP race.
  61 *
  62 *		This program is free software; you can redistribute it and/or
  63 *		modify it under the terms of the GNU General Public License
  64 *		as published by the Free Software Foundation; either version
  65 *		2 of the License, or (at your option) any later version.
  66 */
  67
  68#define pr_fmt(fmt) "IPv4: " fmt
  69
  70#include <linux/err.h>
  71#include <linux/errno.h>
  72#include <linux/types.h>
  73#include <linux/socket.h>
  74#include <linux/in.h>
  75#include <linux/kernel.h>
  76#include <linux/kmod.h>
  77#include <linux/sched.h>
  78#include <linux/timer.h>
  79#include <linux/string.h>
  80#include <linux/sockios.h>
  81#include <linux/net.h>
  82#include <linux/capability.h>
  83#include <linux/fcntl.h>
  84#include <linux/mm.h>
  85#include <linux/interrupt.h>
  86#include <linux/stat.h>
  87#include <linux/init.h>
  88#include <linux/poll.h>
  89#include <linux/netfilter_ipv4.h>
  90#include <linux/random.h>
  91#include <linux/slab.h>
  92
  93#include <linux/uaccess.h>
  94
  95#include <linux/inet.h>
  96#include <linux/igmp.h>
  97#include <linux/inetdevice.h>
  98#include <linux/netdevice.h>
  99#include <net/checksum.h>
 100#include <net/ip.h>
 101#include <net/protocol.h>
 102#include <net/arp.h>
 103#include <net/route.h>
 104#include <net/ip_fib.h>
 105#include <net/inet_connection_sock.h>
 106#include <net/tcp.h>
 107#include <net/udp.h>
 108#include <net/udplite.h>
 109#include <net/ping.h>
 110#include <linux/skbuff.h>
 111#include <net/sock.h>
 112#include <net/raw.h>
 113#include <net/icmp.h>
 114#include <net/inet_common.h>
 115#include <net/ip_tunnels.h>
 116#include <net/xfrm.h>
 117#include <net/net_namespace.h>
 118#include <net/secure_seq.h>
 119#ifdef CONFIG_IP_MROUTE
 120#include <linux/mroute.h>
 121#endif
 122#include <net/l3mdev.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(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(atomic_read(&sk->sk_rmem_alloc));
 154	WARN_ON(refcount_read(&sk->sk_wmem_alloc));
 155	WARN_ON(sk->sk_wmem_queued);
 156	WARN_ON(sk->sk_forward_alloc);
 157
 158	kfree(rcu_dereference_protected(inet->inet_opt, 1));
 159	dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
 160	dst_release(sk->sk_rx_dst);
 161	sk_refcnt_debug_dec(sk);
 162}
 163EXPORT_SYMBOL(inet_sock_destruct);
 164
 165/*
 166 *	The routines beyond this point handle the behaviour of an AF_INET
 167 *	socket object. Mostly it punts to the subprotocols of IP to do
 168 *	the work.
 169 */
 170
 171/*
 172 *	Automatically bind an unbound socket.
 173 */
 174
 175static int inet_autobind(struct sock *sk)
 176{
 177	struct inet_sock *inet;
 178	/* We may need to bind the socket. */
 179	lock_sock(sk);
 180	inet = inet_sk(sk);
 181	if (!inet->inet_num) {
 182		if (sk->sk_prot->get_port(sk, 0)) {
 183			release_sock(sk);
 184			return -EAGAIN;
 185		}
 186		inet->inet_sport = htons(inet->inet_num);
 187	}
 188	release_sock(sk);
 189	return 0;
 190}
 191
 192/*
 193 *	Move a socket into listening state.
 194 */
 195int inet_listen(struct socket *sock, int backlog)
 196{
 197	struct sock *sk = sock->sk;
 198	unsigned char old_state;
 199	int err, tcp_fastopen;
 200
 201	lock_sock(sk);
 202
 203	err = -EINVAL;
 204	if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
 205		goto out;
 206
 207	old_state = sk->sk_state;
 208	if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
 209		goto out;
 210
 211	/* Really, if the socket is already in listen state
 212	 * we can only allow the backlog to be adjusted.
 213	 */
 214	if (old_state != TCP_LISTEN) {
 215		/* Enable TFO w/o requiring TCP_FASTOPEN socket option.
 
 216		 * Note that only TCP sockets (SOCK_STREAM) will reach here.
 217		 * Also fastopen backlog may already been set via the option
 218		 * because the socket was in TCP_LISTEN state previously but
 219		 * was shutdown() rather than close().
 220		 */
 221		tcp_fastopen = sock_net(sk)->ipv4.sysctl_tcp_fastopen;
 222		if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
 223		    (tcp_fastopen & TFO_SERVER_ENABLE) &&
 224		    !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
 225			fastopen_queue_tune(sk, backlog);
 226			tcp_fastopen_init_key_once(sock_net(sk));
 
 
 
 
 
 
 227		}
 228
 229		err = inet_csk_listen_start(sk, backlog);
 230		if (err)
 231			goto out;
 232	}
 233	sk->sk_max_ack_backlog = backlog;
 234	err = 0;
 235
 236out:
 237	release_sock(sk);
 238	return err;
 239}
 240EXPORT_SYMBOL(inet_listen);
 241
 242/*
 243 *	Create an inet socket.
 244 */
 245
 246static int inet_create(struct net *net, struct socket *sock, int protocol,
 247		       int kern)
 248{
 249	struct sock *sk;
 250	struct inet_protosw *answer;
 251	struct inet_sock *inet;
 252	struct proto *answer_prot;
 253	unsigned char answer_flags;
 
 254	int try_loading_module = 0;
 255	int err;
 256
 257	if (protocol < 0 || protocol >= IPPROTO_MAX)
 258		return -EINVAL;
 259
 260	sock->state = SS_UNCONNECTED;
 261
 262	/* Look for the requested type/protocol pair. */
 263lookup_protocol:
 264	err = -ESOCKTNOSUPPORT;
 265	rcu_read_lock();
 266	list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
 267
 268		err = 0;
 269		/* Check the non-wild match. */
 270		if (protocol == answer->protocol) {
 271			if (protocol != IPPROTO_IP)
 272				break;
 273		} else {
 274			/* Check for the two wild cases. */
 275			if (IPPROTO_IP == protocol) {
 276				protocol = answer->protocol;
 277				break;
 278			}
 279			if (IPPROTO_IP == answer->protocol)
 280				break;
 281		}
 282		err = -EPROTONOSUPPORT;
 283	}
 284
 285	if (unlikely(err)) {
 286		if (try_loading_module < 2) {
 287			rcu_read_unlock();
 288			/*
 289			 * Be more specific, e.g. net-pf-2-proto-132-type-1
 290			 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
 291			 */
 292			if (++try_loading_module == 1)
 293				request_module("net-pf-%d-proto-%d-type-%d",
 294					       PF_INET, protocol, sock->type);
 295			/*
 296			 * Fall back to generic, e.g. net-pf-2-proto-132
 297			 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
 298			 */
 299			else
 300				request_module("net-pf-%d-proto-%d",
 301					       PF_INET, protocol);
 302			goto lookup_protocol;
 303		} else
 304			goto out_rcu_unlock;
 305	}
 306
 307	err = -EPERM;
 308	if (sock->type == SOCK_RAW && !kern &&
 309	    !ns_capable(net->user_ns, CAP_NET_RAW))
 310		goto out_rcu_unlock;
 311
 312	sock->ops = answer->ops;
 313	answer_prot = answer->prot;
 
 314	answer_flags = answer->flags;
 315	rcu_read_unlock();
 316
 317	WARN_ON(!answer_prot->slab);
 318
 319	err = -ENOBUFS;
 320	sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
 321	if (!sk)
 322		goto out;
 323
 324	err = 0;
 
 325	if (INET_PROTOSW_REUSE & answer_flags)
 326		sk->sk_reuse = SK_CAN_REUSE;
 327
 328	inet = inet_sk(sk);
 329	inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
 330
 331	inet->nodefrag = 0;
 332
 333	if (SOCK_RAW == sock->type) {
 334		inet->inet_num = protocol;
 335		if (IPPROTO_RAW == protocol)
 336			inet->hdrincl = 1;
 337	}
 338
 339	if (net->ipv4.sysctl_ip_no_pmtu_disc)
 340		inet->pmtudisc = IP_PMTUDISC_DONT;
 341	else
 342		inet->pmtudisc = IP_PMTUDISC_WANT;
 343
 344	inet->inet_id = 0;
 345
 346	sock_init_data(sock, sk);
 347
 348	sk->sk_destruct	   = inet_sock_destruct;
 349	sk->sk_protocol	   = protocol;
 350	sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
 351
 352	inet->uc_ttl	= -1;
 353	inet->mc_loop	= 1;
 354	inet->mc_ttl	= 1;
 355	inet->mc_all	= 1;
 356	inet->mc_index	= 0;
 357	inet->mc_list	= NULL;
 358	inet->rcv_tos	= 0;
 359
 360	sk_refcnt_debug_inc(sk);
 361
 362	if (inet->inet_num) {
 363		/* It assumes that any protocol which allows
 364		 * the user to assign a number at socket
 365		 * creation time automatically
 366		 * shares.
 367		 */
 368		inet->inet_sport = htons(inet->inet_num);
 369		/* Add to protocol hash chains. */
 370		err = sk->sk_prot->hash(sk);
 371		if (err) {
 372			sk_common_release(sk);
 373			goto out;
 374		}
 375	}
 376
 377	if (sk->sk_prot->init) {
 378		err = sk->sk_prot->init(sk);
 379		if (err) {
 380			sk_common_release(sk);
 381			goto out;
 382		}
 383	}
 384
 385	if (!kern) {
 386		err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
 387		if (err) {
 388			sk_common_release(sk);
 389			goto out;
 390		}
 391	}
 392out:
 393	return err;
 394out_rcu_unlock:
 395	rcu_read_unlock();
 396	goto out;
 397}
 398
 399
 400/*
 401 *	The peer socket should always be NULL (or else). When we call this
 402 *	function we are destroying the object and from then on nobody
 403 *	should refer to it.
 404 */
 405int inet_release(struct socket *sock)
 406{
 407	struct sock *sk = sock->sk;
 408
 409	if (sk) {
 410		long timeout;
 411
 
 
 412		/* Applications forget to leave groups before exiting */
 413		ip_mc_drop_socket(sk);
 414
 415		/* If linger is set, we don't return until the close
 416		 * is complete.  Otherwise we return immediately. The
 417		 * actually closing is done the same either way.
 418		 *
 419		 * If the close is due to the process exiting, we never
 420		 * linger..
 421		 */
 422		timeout = 0;
 423		if (sock_flag(sk, SOCK_LINGER) &&
 424		    !(current->flags & PF_EXITING))
 425			timeout = sk->sk_lingertime;
 426		sock->sk = NULL;
 427		sk->sk_prot->close(sk, timeout);
 428	}
 429	return 0;
 430}
 431EXPORT_SYMBOL(inet_release);
 432
 
 
 
 
 433int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 434{
 
 435	struct sock *sk = sock->sk;
 
 
 
 
 436	int err;
 437
 438	/* If the socket has its own bind function then use it. (RAW) */
 439	if (sk->sk_prot->bind) {
 440		return sk->sk_prot->bind(sk, uaddr, addr_len);
 
 441	}
 
 442	if (addr_len < sizeof(struct sockaddr_in))
 443		return -EINVAL;
 444
 445	/* BPF prog is run before any checks are done so that if the prog
 446	 * changes context in a wrong way it will be caught.
 447	 */
 448	err = BPF_CGROUP_RUN_PROG_INET4_BIND(sk, uaddr);
 449	if (err)
 450		return err;
 451
 452	return __inet_bind(sk, uaddr, addr_len, false, true);
 453}
 454EXPORT_SYMBOL(inet_bind);
 455
 456int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
 457		bool force_bind_address_no_port, bool with_lock)
 458{
 459	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
 460	struct inet_sock *inet = inet_sk(sk);
 461	struct net *net = sock_net(sk);
 462	unsigned short snum;
 463	int chk_addr_ret;
 464	u32 tb_id = RT_TABLE_LOCAL;
 465	int err;
 466
 467	if (addr->sin_family != AF_INET) {
 468		/* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
 469		 * only if s_addr is INADDR_ANY.
 470		 */
 471		err = -EAFNOSUPPORT;
 472		if (addr->sin_family != AF_UNSPEC ||
 473		    addr->sin_addr.s_addr != htonl(INADDR_ANY))
 474			goto out;
 475	}
 476
 477	tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
 478	chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
 479
 480	/* Not specified by any standard per-se, however it breaks too
 481	 * many applications when removed.  It is unfortunate since
 482	 * allowing applications to make a non-local bind solves
 483	 * several problems with systems using dynamic addressing.
 484	 * (ie. your servers still start up even if your ISDN link
 485	 *  is temporarily down)
 486	 */
 487	err = -EADDRNOTAVAIL;
 488	if (!net->ipv4.sysctl_ip_nonlocal_bind &&
 489	    !(inet->freebind || inet->transparent) &&
 490	    addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
 491	    chk_addr_ret != RTN_LOCAL &&
 492	    chk_addr_ret != RTN_MULTICAST &&
 493	    chk_addr_ret != RTN_BROADCAST)
 494		goto out;
 495
 496	snum = ntohs(addr->sin_port);
 497	err = -EACCES;
 498	if (snum && snum < inet_prot_sock(net) &&
 499	    !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
 500		goto out;
 501
 502	/*      We keep a pair of addresses. rcv_saddr is the one
 503	 *      used by hash lookups, and saddr is used for transmit.
 504	 *
 505	 *      In the BSD API these are the same except where it
 506	 *      would be illegal to use them (multicast/broadcast) in
 507	 *      which case the sending device address is used.
 508	 */
 509	if (with_lock)
 510		lock_sock(sk);
 511
 512	/* Check these errors (active socket, double bind). */
 513	err = -EINVAL;
 514	if (sk->sk_state != TCP_CLOSE || inet->inet_num)
 515		goto out_release_sock;
 516
 517	inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
 518	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
 519		inet->inet_saddr = 0;  /* Use device */
 520
 521	/* Make sure we are allowed to bind here. */
 522	if (snum || !(inet->bind_address_no_port ||
 523		      force_bind_address_no_port)) {
 524		if (sk->sk_prot->get_port(sk, snum)) {
 525			inet->inet_saddr = inet->inet_rcv_saddr = 0;
 526			err = -EADDRINUSE;
 527			goto out_release_sock;
 528		}
 529		err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
 530		if (err) {
 531			inet->inet_saddr = inet->inet_rcv_saddr = 0;
 532			goto out_release_sock;
 533		}
 534	}
 535
 536	if (inet->inet_rcv_saddr)
 537		sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
 538	if (snum)
 539		sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
 540	inet->inet_sport = htons(inet->inet_num);
 541	inet->inet_daddr = 0;
 542	inet->inet_dport = 0;
 543	sk_dst_reset(sk);
 544	err = 0;
 545out_release_sock:
 546	if (with_lock)
 547		release_sock(sk);
 548out:
 549	return err;
 550}
 
 551
 552int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
 553		       int addr_len, int flags)
 554{
 555	struct sock *sk = sock->sk;
 556	int err;
 557
 558	if (addr_len < sizeof(uaddr->sa_family))
 559		return -EINVAL;
 560	if (uaddr->sa_family == AF_UNSPEC)
 561		return sk->sk_prot->disconnect(sk, flags);
 562
 563	if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
 564		err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
 565		if (err)
 566			return err;
 567	}
 568
 569	if (!inet_sk(sk)->inet_num && inet_autobind(sk))
 570		return -EAGAIN;
 571	return sk->sk_prot->connect(sk, uaddr, addr_len);
 572}
 573EXPORT_SYMBOL(inet_dgram_connect);
 574
 575static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
 576{
 577	DEFINE_WAIT_FUNC(wait, woken_wake_function);
 578
 579	add_wait_queue(sk_sleep(sk), &wait);
 580	sk->sk_write_pending += writebias;
 581
 582	/* Basic assumption: if someone sets sk->sk_err, he _must_
 583	 * change state of the socket from TCP_SYN_*.
 584	 * Connect() does not allow to get error notifications
 585	 * without closing the socket.
 586	 */
 587	while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 588		release_sock(sk);
 589		timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
 590		lock_sock(sk);
 591		if (signal_pending(current) || !timeo)
 592			break;
 
 593	}
 594	remove_wait_queue(sk_sleep(sk), &wait);
 595	sk->sk_write_pending -= writebias;
 596	return timeo;
 597}
 598
 599/*
 600 *	Connect to a remote host. There is regrettably still a little
 601 *	TCP 'magic' in here.
 602 */
 603int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 604			  int addr_len, int flags, int is_sendmsg)
 605{
 606	struct sock *sk = sock->sk;
 607	int err;
 608	long timeo;
 609
 610	/*
 611	 * uaddr can be NULL and addr_len can be 0 if:
 612	 * sk is a TCP fastopen active socket and
 613	 * TCP_FASTOPEN_CONNECT sockopt is set and
 614	 * we already have a valid cookie for this socket.
 615	 * In this case, user can call write() after connect().
 616	 * write() will invoke tcp_sendmsg_fastopen() which calls
 617	 * __inet_stream_connect().
 618	 */
 619	if (uaddr) {
 620		if (addr_len < sizeof(uaddr->sa_family))
 621			return -EINVAL;
 622
 623		if (uaddr->sa_family == AF_UNSPEC) {
 624			err = sk->sk_prot->disconnect(sk, flags);
 625			sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 626			goto out;
 627		}
 628	}
 629
 630	switch (sock->state) {
 631	default:
 632		err = -EINVAL;
 633		goto out;
 634	case SS_CONNECTED:
 635		err = -EISCONN;
 636		goto out;
 637	case SS_CONNECTING:
 638		if (inet_sk(sk)->defer_connect)
 639			err = is_sendmsg ? -EINPROGRESS : -EISCONN;
 640		else
 641			err = -EALREADY;
 642		/* Fall out of switch with err, set for this state */
 643		break;
 644	case SS_UNCONNECTED:
 645		err = -EISCONN;
 646		if (sk->sk_state != TCP_CLOSE)
 647			goto out;
 648
 649		if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
 650			err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
 651			if (err)
 652				goto out;
 653		}
 654
 655		err = sk->sk_prot->connect(sk, uaddr, addr_len);
 656		if (err < 0)
 657			goto out;
 658
 659		sock->state = SS_CONNECTING;
 660
 661		if (!err && inet_sk(sk)->defer_connect)
 662			goto out;
 663
 664		/* Just entered SS_CONNECTING state; the only
 665		 * difference is that return value in non-blocking
 666		 * case is EINPROGRESS, rather than EALREADY.
 667		 */
 668		err = -EINPROGRESS;
 669		break;
 670	}
 671
 672	timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
 673
 674	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 675		int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
 676				tcp_sk(sk)->fastopen_req &&
 677				tcp_sk(sk)->fastopen_req->data ? 1 : 0;
 678
 679		/* Error code is set above */
 680		if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
 681			goto out;
 682
 683		err = sock_intr_errno(timeo);
 684		if (signal_pending(current))
 685			goto out;
 686	}
 687
 688	/* Connection was closed by RST, timeout, ICMP error
 689	 * or another process disconnected us.
 690	 */
 691	if (sk->sk_state == TCP_CLOSE)
 692		goto sock_error;
 693
 694	/* sk->sk_err may be not zero now, if RECVERR was ordered by user
 695	 * and error was received after socket entered established state.
 696	 * Hence, it is handled normally after connect() return successfully.
 697	 */
 698
 699	sock->state = SS_CONNECTED;
 700	err = 0;
 701out:
 702	return err;
 703
 704sock_error:
 705	err = sock_error(sk) ? : -ECONNABORTED;
 706	sock->state = SS_UNCONNECTED;
 707	if (sk->sk_prot->disconnect(sk, flags))
 708		sock->state = SS_DISCONNECTING;
 709	goto out;
 710}
 711EXPORT_SYMBOL(__inet_stream_connect);
 712
 713int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 714			int addr_len, int flags)
 715{
 716	int err;
 717
 718	lock_sock(sock->sk);
 719	err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
 720	release_sock(sock->sk);
 721	return err;
 722}
 723EXPORT_SYMBOL(inet_stream_connect);
 724
 725/*
 726 *	Accept a pending connection. The TCP layer now gives BSD semantics.
 727 */
 728
 729int inet_accept(struct socket *sock, struct socket *newsock, int flags,
 730		bool kern)
 731{
 732	struct sock *sk1 = sock->sk;
 733	int err = -EINVAL;
 734	struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err, kern);
 735
 736	if (!sk2)
 737		goto do_err;
 738
 739	lock_sock(sk2);
 740
 741	sock_rps_record_flow(sk2);
 742	WARN_ON(!((1 << sk2->sk_state) &
 743		  (TCPF_ESTABLISHED | TCPF_SYN_RECV |
 744		  TCPF_CLOSE_WAIT | TCPF_CLOSE)));
 745
 746	sock_graft(sk2, newsock);
 747
 748	newsock->state = SS_CONNECTED;
 749	err = 0;
 750	release_sock(sk2);
 751do_err:
 752	return err;
 753}
 754EXPORT_SYMBOL(inet_accept);
 755
 756
 757/*
 758 *	This does both peername and sockname.
 759 */
 760int inet_getname(struct socket *sock, struct sockaddr *uaddr,
 761			int peer)
 762{
 763	struct sock *sk		= sock->sk;
 764	struct inet_sock *inet	= inet_sk(sk);
 765	DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
 766
 767	sin->sin_family = AF_INET;
 768	if (peer) {
 769		if (!inet->inet_dport ||
 770		    (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
 771		     peer == 1))
 772			return -ENOTCONN;
 773		sin->sin_port = inet->inet_dport;
 774		sin->sin_addr.s_addr = inet->inet_daddr;
 775	} else {
 776		__be32 addr = inet->inet_rcv_saddr;
 777		if (!addr)
 778			addr = inet->inet_saddr;
 779		sin->sin_port = inet->inet_sport;
 780		sin->sin_addr.s_addr = addr;
 781	}
 782	memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 783	return sizeof(*sin);
 
 784}
 785EXPORT_SYMBOL(inet_getname);
 786
 787int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
 
 788{
 789	struct sock *sk = sock->sk;
 790
 791	sock_rps_record_flow(sk);
 792
 793	/* We may need to bind the socket. */
 794	if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
 795	    inet_autobind(sk))
 796		return -EAGAIN;
 797
 798	return sk->sk_prot->sendmsg(sk, msg, size);
 799}
 800EXPORT_SYMBOL(inet_sendmsg);
 801
 802ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
 803		      size_t size, int flags)
 804{
 805	struct sock *sk = sock->sk;
 806
 807	sock_rps_record_flow(sk);
 808
 809	/* We may need to bind the socket. */
 810	if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
 811	    inet_autobind(sk))
 812		return -EAGAIN;
 813
 814	if (sk->sk_prot->sendpage)
 815		return sk->sk_prot->sendpage(sk, page, offset, size, flags);
 816	return sock_no_sendpage(sock, page, offset, size, flags);
 817}
 818EXPORT_SYMBOL(inet_sendpage);
 819
 820int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
 821		 int flags)
 822{
 823	struct sock *sk = sock->sk;
 824	int addr_len = 0;
 825	int err;
 826
 827	if (likely(!(flags & MSG_ERRQUEUE)))
 828		sock_rps_record_flow(sk);
 829
 830	err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
 831				   flags & ~MSG_DONTWAIT, &addr_len);
 832	if (err >= 0)
 833		msg->msg_namelen = addr_len;
 834	return err;
 835}
 836EXPORT_SYMBOL(inet_recvmsg);
 837
 838int inet_shutdown(struct socket *sock, int how)
 839{
 840	struct sock *sk = sock->sk;
 841	int err = 0;
 842
 843	/* This should really check to make sure
 844	 * the socket is a TCP socket. (WHY AC...)
 845	 */
 846	how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
 847		       1->2 bit 2 snds.
 848		       2->3 */
 849	if ((how & ~SHUTDOWN_MASK) || !how)	/* MAXINT->0 */
 850		return -EINVAL;
 851
 852	lock_sock(sk);
 853	if (sock->state == SS_CONNECTING) {
 854		if ((1 << sk->sk_state) &
 855		    (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
 856			sock->state = SS_DISCONNECTING;
 857		else
 858			sock->state = SS_CONNECTED;
 859	}
 860
 861	switch (sk->sk_state) {
 862	case TCP_CLOSE:
 863		err = -ENOTCONN;
 864		/* Hack to wake up other listeners, who can poll for
 865		   EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
 866		/* fall through */
 867	default:
 868		sk->sk_shutdown |= how;
 869		if (sk->sk_prot->shutdown)
 870			sk->sk_prot->shutdown(sk, how);
 871		break;
 872
 873	/* Remaining two branches are temporary solution for missing
 874	 * close() in multithreaded environment. It is _not_ a good idea,
 875	 * but we have no choice until close() is repaired at VFS level.
 876	 */
 877	case TCP_LISTEN:
 878		if (!(how & RCV_SHUTDOWN))
 879			break;
 880		/* fall through */
 881	case TCP_SYN_SENT:
 882		err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
 883		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 884		break;
 885	}
 886
 887	/* Wake up anyone sleeping in poll. */
 888	sk->sk_state_change(sk);
 889	release_sock(sk);
 890	return err;
 891}
 892EXPORT_SYMBOL(inet_shutdown);
 893
 894/*
 895 *	ioctl() calls you can issue on an INET socket. Most of these are
 896 *	device configuration and stuff and very rarely used. Some ioctls
 897 *	pass on to the socket itself.
 898 *
 899 *	NOTE: I like the idea of a module for the config stuff. ie ifconfig
 900 *	loads the devconfigure module does its configuring and unloads it.
 901 *	There's a good 20K of config code hanging around the kernel.
 902 */
 903
 904int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 905{
 906	struct sock *sk = sock->sk;
 907	int err = 0;
 908	struct net *net = sock_net(sk);
 909	void __user *p = (void __user *)arg;
 910	struct ifreq ifr;
 911	struct rtentry rt;
 912
 913	switch (cmd) {
 914	case SIOCGSTAMP:
 915		err = sock_get_timestamp(sk, (struct timeval __user *)arg);
 916		break;
 917	case SIOCGSTAMPNS:
 918		err = sock_get_timestampns(sk, (struct timespec __user *)arg);
 919		break;
 920	case SIOCADDRT:
 921	case SIOCDELRT:
 922		if (copy_from_user(&rt, p, sizeof(struct rtentry)))
 923			return -EFAULT;
 924		err = ip_rt_ioctl(net, cmd, &rt);
 925		break;
 926	case SIOCRTMSG:
 927		err = -EINVAL;
 928		break;
 929	case SIOCDARP:
 930	case SIOCGARP:
 931	case SIOCSARP:
 932		err = arp_ioctl(net, cmd, (void __user *)arg);
 933		break;
 934	case SIOCGIFADDR:
 
 935	case SIOCGIFBRDADDR:
 
 936	case SIOCGIFNETMASK:
 
 937	case SIOCGIFDSTADDR:
 938	case SIOCGIFPFLAGS:
 939		if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
 940			return -EFAULT;
 941		err = devinet_ioctl(net, cmd, &ifr);
 942		if (!err && copy_to_user(p, &ifr, sizeof(struct ifreq)))
 943			err = -EFAULT;
 944		break;
 945
 946	case SIOCSIFADDR:
 947	case SIOCSIFBRDADDR:
 948	case SIOCSIFNETMASK:
 949	case SIOCSIFDSTADDR:
 950	case SIOCSIFPFLAGS:
 
 951	case SIOCSIFFLAGS:
 952		if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
 953			return -EFAULT;
 954		err = devinet_ioctl(net, cmd, &ifr);
 955		break;
 956	default:
 957		if (sk->sk_prot->ioctl)
 958			err = sk->sk_prot->ioctl(sk, cmd, arg);
 959		else
 960			err = -ENOIOCTLCMD;
 961		break;
 962	}
 963	return err;
 964}
 965EXPORT_SYMBOL(inet_ioctl);
 966
 967#ifdef CONFIG_COMPAT
 968static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 969{
 970	struct sock *sk = sock->sk;
 971	int err = -ENOIOCTLCMD;
 972
 973	if (sk->sk_prot->compat_ioctl)
 974		err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
 975
 976	return err;
 977}
 978#endif
 979
 980const struct proto_ops inet_stream_ops = {
 981	.family		   = PF_INET,
 982	.owner		   = THIS_MODULE,
 983	.release	   = inet_release,
 984	.bind		   = inet_bind,
 985	.connect	   = inet_stream_connect,
 986	.socketpair	   = sock_no_socketpair,
 987	.accept		   = inet_accept,
 988	.getname	   = inet_getname,
 989	.poll		   = tcp_poll,
 990	.ioctl		   = inet_ioctl,
 991	.listen		   = inet_listen,
 992	.shutdown	   = inet_shutdown,
 993	.setsockopt	   = sock_common_setsockopt,
 994	.getsockopt	   = sock_common_getsockopt,
 995	.sendmsg	   = inet_sendmsg,
 996	.recvmsg	   = inet_recvmsg,
 997	.mmap		   = sock_no_mmap,
 998	.sendpage	   = inet_sendpage,
 999	.splice_read	   = tcp_splice_read,
1000	.read_sock	   = tcp_read_sock,
1001	.sendmsg_locked    = tcp_sendmsg_locked,
1002	.sendpage_locked   = tcp_sendpage_locked,
1003	.peek_len	   = tcp_peek_len,
1004#ifdef CONFIG_COMPAT
1005	.compat_setsockopt = compat_sock_common_setsockopt,
1006	.compat_getsockopt = compat_sock_common_getsockopt,
1007	.compat_ioctl	   = inet_compat_ioctl,
1008#endif
1009};
1010EXPORT_SYMBOL(inet_stream_ops);
1011
1012const struct proto_ops inet_dgram_ops = {
1013	.family		   = PF_INET,
1014	.owner		   = THIS_MODULE,
1015	.release	   = inet_release,
1016	.bind		   = inet_bind,
1017	.connect	   = inet_dgram_connect,
1018	.socketpair	   = sock_no_socketpair,
1019	.accept		   = sock_no_accept,
1020	.getname	   = inet_getname,
1021	.poll		   = udp_poll,
1022	.ioctl		   = inet_ioctl,
1023	.listen		   = sock_no_listen,
1024	.shutdown	   = inet_shutdown,
1025	.setsockopt	   = sock_common_setsockopt,
1026	.getsockopt	   = sock_common_getsockopt,
1027	.sendmsg	   = inet_sendmsg,
1028	.recvmsg	   = inet_recvmsg,
1029	.mmap		   = sock_no_mmap,
1030	.sendpage	   = inet_sendpage,
1031	.set_peek_off	   = sk_set_peek_off,
1032#ifdef CONFIG_COMPAT
1033	.compat_setsockopt = compat_sock_common_setsockopt,
1034	.compat_getsockopt = compat_sock_common_getsockopt,
1035	.compat_ioctl	   = inet_compat_ioctl,
1036#endif
1037};
1038EXPORT_SYMBOL(inet_dgram_ops);
1039
1040/*
1041 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1042 * udp_poll
1043 */
1044static const struct proto_ops inet_sockraw_ops = {
1045	.family		   = PF_INET,
1046	.owner		   = THIS_MODULE,
1047	.release	   = inet_release,
1048	.bind		   = inet_bind,
1049	.connect	   = inet_dgram_connect,
1050	.socketpair	   = sock_no_socketpair,
1051	.accept		   = sock_no_accept,
1052	.getname	   = inet_getname,
1053	.poll		   = datagram_poll,
1054	.ioctl		   = inet_ioctl,
1055	.listen		   = sock_no_listen,
1056	.shutdown	   = inet_shutdown,
1057	.setsockopt	   = sock_common_setsockopt,
1058	.getsockopt	   = sock_common_getsockopt,
1059	.sendmsg	   = inet_sendmsg,
1060	.recvmsg	   = inet_recvmsg,
1061	.mmap		   = sock_no_mmap,
1062	.sendpage	   = inet_sendpage,
1063#ifdef CONFIG_COMPAT
1064	.compat_setsockopt = compat_sock_common_setsockopt,
1065	.compat_getsockopt = compat_sock_common_getsockopt,
1066	.compat_ioctl	   = inet_compat_ioctl,
1067#endif
1068};
1069
1070static const struct net_proto_family inet_family_ops = {
1071	.family = PF_INET,
1072	.create = inet_create,
1073	.owner	= THIS_MODULE,
1074};
1075
1076/* Upon startup we insert all the elements in inetsw_array[] into
1077 * the linked list inetsw.
1078 */
1079static struct inet_protosw inetsw_array[] =
1080{
1081	{
1082		.type =       SOCK_STREAM,
1083		.protocol =   IPPROTO_TCP,
1084		.prot =       &tcp_prot,
1085		.ops =        &inet_stream_ops,
 
1086		.flags =      INET_PROTOSW_PERMANENT |
1087			      INET_PROTOSW_ICSK,
1088	},
1089
1090	{
1091		.type =       SOCK_DGRAM,
1092		.protocol =   IPPROTO_UDP,
1093		.prot =       &udp_prot,
1094		.ops =        &inet_dgram_ops,
 
1095		.flags =      INET_PROTOSW_PERMANENT,
1096       },
1097
1098       {
1099		.type =       SOCK_DGRAM,
1100		.protocol =   IPPROTO_ICMP,
1101		.prot =       &ping_prot,
1102		.ops =        &inet_sockraw_ops,
 
1103		.flags =      INET_PROTOSW_REUSE,
1104       },
1105
1106       {
1107	       .type =       SOCK_RAW,
1108	       .protocol =   IPPROTO_IP,	/* wild card */
1109	       .prot =       &raw_prot,
1110	       .ops =        &inet_sockraw_ops,
 
1111	       .flags =      INET_PROTOSW_REUSE,
1112       }
1113};
1114
1115#define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1116
1117void inet_register_protosw(struct inet_protosw *p)
1118{
1119	struct list_head *lh;
1120	struct inet_protosw *answer;
1121	int protocol = p->protocol;
1122	struct list_head *last_perm;
1123
1124	spin_lock_bh(&inetsw_lock);
1125
1126	if (p->type >= SOCK_MAX)
1127		goto out_illegal;
1128
1129	/* If we are trying to override a permanent protocol, bail. */
 
1130	last_perm = &inetsw[p->type];
1131	list_for_each(lh, &inetsw[p->type]) {
1132		answer = list_entry(lh, struct inet_protosw, list);
 
1133		/* Check only the non-wild match. */
1134		if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1135			break;
1136		if (protocol == answer->protocol)
1137			goto out_permanent;
1138		last_perm = lh;
 
 
1139	}
 
 
1140
1141	/* Add the new entry after the last permanent entry if any, so that
1142	 * the new entry does not override a permanent entry when matched with
1143	 * a wild-card protocol. But it is allowed to override any existing
1144	 * non-permanent entry.  This means that when we remove this entry, the
1145	 * system automatically returns to the old behavior.
1146	 */
1147	list_add_rcu(&p->list, last_perm);
1148out:
1149	spin_unlock_bh(&inetsw_lock);
1150
1151	return;
1152
1153out_permanent:
1154	pr_err("Attempt to override permanent protocol %d\n", protocol);
1155	goto out;
1156
1157out_illegal:
1158	pr_err("Ignoring attempt to register invalid socket type %d\n",
1159	       p->type);
1160	goto out;
1161}
1162EXPORT_SYMBOL(inet_register_protosw);
1163
1164void inet_unregister_protosw(struct inet_protosw *p)
1165{
1166	if (INET_PROTOSW_PERMANENT & p->flags) {
1167		pr_err("Attempt to unregister permanent protocol %d\n",
1168		       p->protocol);
1169	} else {
1170		spin_lock_bh(&inetsw_lock);
1171		list_del_rcu(&p->list);
1172		spin_unlock_bh(&inetsw_lock);
1173
1174		synchronize_net();
1175	}
1176}
1177EXPORT_SYMBOL(inet_unregister_protosw);
1178
 
 
 
 
 
 
1179static int inet_sk_reselect_saddr(struct sock *sk)
1180{
1181	struct inet_sock *inet = inet_sk(sk);
1182	__be32 old_saddr = inet->inet_saddr;
1183	__be32 daddr = inet->inet_daddr;
1184	struct flowi4 *fl4;
1185	struct rtable *rt;
1186	__be32 new_saddr;
1187	struct ip_options_rcu *inet_opt;
1188
1189	inet_opt = rcu_dereference_protected(inet->inet_opt,
1190					     lockdep_sock_is_held(sk));
1191	if (inet_opt && inet_opt->opt.srr)
1192		daddr = inet_opt->opt.faddr;
1193
1194	/* Query new route. */
1195	fl4 = &inet->cork.fl.u.ip4;
1196	rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1197			      sk->sk_bound_dev_if, sk->sk_protocol,
1198			      inet->inet_sport, inet->inet_dport, sk);
1199	if (IS_ERR(rt))
1200		return PTR_ERR(rt);
1201
1202	sk_setup_caps(sk, &rt->dst);
1203
1204	new_saddr = fl4->saddr;
1205
1206	if (new_saddr == old_saddr)
1207		return 0;
1208
1209	if (sock_net(sk)->ipv4.sysctl_ip_dynaddr > 1) {
1210		pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1211			__func__, &old_saddr, &new_saddr);
1212	}
1213
1214	inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1215
1216	/*
1217	 * XXX The only one ugly spot where we need to
1218	 * XXX really change the sockets identity after
1219	 * XXX it has entered the hashes. -DaveM
1220	 *
1221	 * Besides that, it does not check for connection
1222	 * uniqueness. Wait for troubles.
1223	 */
1224	return __sk_prot_rehash(sk);
 
1225}
1226
1227int inet_sk_rebuild_header(struct sock *sk)
1228{
1229	struct inet_sock *inet = inet_sk(sk);
1230	struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1231	__be32 daddr;
1232	struct ip_options_rcu *inet_opt;
1233	struct flowi4 *fl4;
1234	int err;
1235
1236	/* Route is OK, nothing to do. */
1237	if (rt)
1238		return 0;
1239
1240	/* Reroute. */
1241	rcu_read_lock();
1242	inet_opt = rcu_dereference(inet->inet_opt);
1243	daddr = inet->inet_daddr;
1244	if (inet_opt && inet_opt->opt.srr)
1245		daddr = inet_opt->opt.faddr;
1246	rcu_read_unlock();
1247	fl4 = &inet->cork.fl.u.ip4;
1248	rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1249				   inet->inet_dport, inet->inet_sport,
1250				   sk->sk_protocol, RT_CONN_FLAGS(sk),
1251				   sk->sk_bound_dev_if);
1252	if (!IS_ERR(rt)) {
1253		err = 0;
1254		sk_setup_caps(sk, &rt->dst);
1255	} else {
1256		err = PTR_ERR(rt);
1257
1258		/* Routing failed... */
1259		sk->sk_route_caps = 0;
1260		/*
1261		 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1262		 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1263		 */
1264		if (!sock_net(sk)->ipv4.sysctl_ip_dynaddr ||
1265		    sk->sk_state != TCP_SYN_SENT ||
1266		    (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1267		    (err = inet_sk_reselect_saddr(sk)) != 0)
1268			sk->sk_err_soft = -err;
1269	}
1270
1271	return err;
1272}
1273EXPORT_SYMBOL(inet_sk_rebuild_header);
1274
1275void inet_sk_set_state(struct sock *sk, int state)
1276{
1277	trace_inet_sock_set_state(sk, sk->sk_state, state);
1278	sk->sk_state = state;
1279}
1280EXPORT_SYMBOL(inet_sk_set_state);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1281
1282void inet_sk_state_store(struct sock *sk, int newstate)
1283{
1284	trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1285	smp_store_release(&sk->sk_state, newstate);
1286}
1287
1288struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1289				 netdev_features_t features)
1290{
1291	bool udpfrag = false, fixedid = false, gso_partial, encap;
1292	struct sk_buff *segs = ERR_PTR(-EINVAL);
1293	const struct net_offload *ops;
1294	unsigned int offset = 0;
 
1295	struct iphdr *iph;
1296	int proto, tot_len;
1297	int nhoff;
1298	int ihl;
1299	int id;
1300
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1301	skb_reset_network_header(skb);
1302	nhoff = skb_network_header(skb) - skb_mac_header(skb);
1303	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1304		goto out;
1305
1306	iph = ip_hdr(skb);
1307	ihl = iph->ihl * 4;
1308	if (ihl < sizeof(*iph))
1309		goto out;
1310
1311	id = ntohs(iph->id);
1312	proto = iph->protocol;
1313
1314	/* Warning: after this point, iph might be no longer valid */
1315	if (unlikely(!pskb_may_pull(skb, ihl)))
1316		goto out;
1317	__skb_pull(skb, ihl);
1318
1319	encap = SKB_GSO_CB(skb)->encap_level > 0;
1320	if (encap)
1321		features &= skb->dev->hw_enc_features;
1322	SKB_GSO_CB(skb)->encap_level += ihl;
1323
1324	skb_reset_transport_header(skb);
1325
1326	segs = ERR_PTR(-EPROTONOSUPPORT);
1327
1328	if (!skb->encapsulation || encap) {
1329		udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1330		fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1331
1332		/* fixed ID is invalid if DF bit is not set */
1333		if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1334			goto out;
1335	}
1336
1337	ops = rcu_dereference(inet_offloads[proto]);
1338	if (likely(ops && ops->callbacks.gso_segment))
1339		segs = ops->callbacks.gso_segment(skb, features);
1340
1341	if (IS_ERR_OR_NULL(segs))
1342		goto out;
1343
1344	gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1345
1346	skb = segs;
1347	do {
1348		iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1349		if (udpfrag) {
 
1350			iph->frag_off = htons(offset >> 3);
1351			if (skb->next)
1352				iph->frag_off |= htons(IP_MF);
1353			offset += skb->len - nhoff - ihl;
1354			tot_len = skb->len - nhoff;
1355		} else if (skb_is_gso(skb)) {
1356			if (!fixedid) {
1357				iph->id = htons(id);
1358				id += skb_shinfo(skb)->gso_segs;
1359			}
1360
1361			if (gso_partial)
1362				tot_len = skb_shinfo(skb)->gso_size +
1363					  SKB_GSO_CB(skb)->data_offset +
1364					  skb->head - (unsigned char *)iph;
1365			else
1366				tot_len = skb->len - nhoff;
1367		} else {
1368			if (!fixedid)
1369				iph->id = htons(id++);
1370			tot_len = skb->len - nhoff;
1371		}
1372		iph->tot_len = htons(tot_len);
1373		ip_send_check(iph);
1374		if (encap)
1375			skb_reset_inner_headers(skb);
1376		skb->network_header = (u8 *)iph - skb->head;
1377	} while ((skb = skb->next));
1378
1379out:
1380	return segs;
1381}
1382EXPORT_SYMBOL(inet_gso_segment);
1383
1384struct sk_buff **inet_gro_receive(struct sk_buff **head, struct sk_buff *skb)
 
1385{
1386	const struct net_offload *ops;
1387	struct sk_buff **pp = NULL;
1388	struct sk_buff *p;
1389	const struct iphdr *iph;
1390	unsigned int hlen;
1391	unsigned int off;
1392	unsigned int id;
1393	int flush = 1;
1394	int proto;
1395
1396	off = skb_gro_offset(skb);
1397	hlen = off + sizeof(*iph);
1398	iph = skb_gro_header_fast(skb, off);
1399	if (skb_gro_header_hard(skb, hlen)) {
1400		iph = skb_gro_header_slow(skb, hlen, off);
1401		if (unlikely(!iph))
1402			goto out;
1403	}
1404
1405	proto = iph->protocol;
1406
1407	rcu_read_lock();
1408	ops = rcu_dereference(inet_offloads[proto]);
1409	if (!ops || !ops->callbacks.gro_receive)
1410		goto out_unlock;
1411
1412	if (*(u8 *)iph != 0x45)
1413		goto out_unlock;
1414
1415	if (ip_is_fragment(iph))
1416		goto out_unlock;
1417
1418	if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1419		goto out_unlock;
1420
1421	id = ntohl(*(__be32 *)&iph->id);
1422	flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1423	id >>= 16;
1424
1425	for (p = *head; p; p = p->next) {
1426		struct iphdr *iph2;
1427		u16 flush_id;
1428
1429		if (!NAPI_GRO_CB(p)->same_flow)
1430			continue;
1431
1432		iph2 = (struct iphdr *)(p->data + off);
1433		/* The above works because, with the exception of the top
1434		 * (inner most) layer, we only aggregate pkts with the same
1435		 * hdr length so all the hdrs we'll need to verify will start
1436		 * at the same offset.
1437		 */
1438		if ((iph->protocol ^ iph2->protocol) |
1439		    ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1440		    ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1441			NAPI_GRO_CB(p)->same_flow = 0;
1442			continue;
1443		}
1444
1445		/* All fields must match except length and checksum. */
1446		NAPI_GRO_CB(p)->flush |=
1447			(iph->ttl ^ iph2->ttl) |
1448			(iph->tos ^ iph2->tos) |
1449			((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1450
 
 
 
 
 
 
 
1451		NAPI_GRO_CB(p)->flush |= flush;
1452
1453		/* We need to store of the IP ID check to be included later
1454		 * when we can verify that this packet does in fact belong
1455		 * to a given flow.
1456		 */
1457		flush_id = (u16)(id - ntohs(iph2->id));
1458
1459		/* This bit of code makes it much easier for us to identify
1460		 * the cases where we are doing atomic vs non-atomic IP ID
1461		 * checks.  Specifically an atomic check can return IP ID
1462		 * values 0 - 0xFFFF, while a non-atomic check can only
1463		 * return 0 or 0xFFFF.
1464		 */
1465		if (!NAPI_GRO_CB(p)->is_atomic ||
1466		    !(iph->frag_off & htons(IP_DF))) {
1467			flush_id ^= NAPI_GRO_CB(p)->count;
1468			flush_id = flush_id ? 0xFFFF : 0;
1469		}
1470
1471		/* If the previous IP ID value was based on an atomic
1472		 * datagram we can overwrite the value and ignore it.
1473		 */
1474		if (NAPI_GRO_CB(skb)->is_atomic)
1475			NAPI_GRO_CB(p)->flush_id = flush_id;
1476		else
1477			NAPI_GRO_CB(p)->flush_id |= flush_id;
1478	}
1479
1480	NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1481	NAPI_GRO_CB(skb)->flush |= flush;
1482	skb_set_network_header(skb, off);
1483	/* The above will be needed by the transport layer if there is one
1484	 * immediately following this IP hdr.
1485	 */
1486
1487	/* Note : No need to call skb_gro_postpull_rcsum() here,
1488	 * as we already checked checksum over ipv4 header was 0
1489	 */
1490	skb_gro_pull(skb, sizeof(*iph));
1491	skb_set_transport_header(skb, skb_gro_offset(skb));
1492
1493	pp = call_gro_receive(ops->callbacks.gro_receive, head, skb);
1494
1495out_unlock:
1496	rcu_read_unlock();
1497
1498out:
1499	skb_gro_flush_final(skb, pp, flush);
1500
1501	return pp;
1502}
1503EXPORT_SYMBOL(inet_gro_receive);
1504
1505static struct sk_buff **ipip_gro_receive(struct sk_buff **head,
1506					 struct sk_buff *skb)
1507{
1508	if (NAPI_GRO_CB(skb)->encap_mark) {
1509		NAPI_GRO_CB(skb)->flush = 1;
1510		return NULL;
1511	}
1512
1513	NAPI_GRO_CB(skb)->encap_mark = 1;
1514
1515	return inet_gro_receive(head, skb);
1516}
1517
1518#define SECONDS_PER_DAY	86400
1519
1520/* inet_current_timestamp - Return IP network timestamp
1521 *
1522 * Return milliseconds since midnight in network byte order.
1523 */
1524__be32 inet_current_timestamp(void)
1525{
1526	u32 secs;
1527	u32 msecs;
1528	struct timespec64 ts;
1529
1530	ktime_get_real_ts64(&ts);
1531
1532	/* Get secs since midnight. */
1533	(void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1534	/* Convert to msecs. */
1535	msecs = secs * MSEC_PER_SEC;
1536	/* Convert nsec to msec. */
1537	msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1538
1539	/* Convert to network byte order. */
1540	return htonl(msecs);
1541}
1542EXPORT_SYMBOL(inet_current_timestamp);
1543
1544int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1545{
1546	if (sk->sk_family == AF_INET)
1547		return ip_recv_error(sk, msg, len, addr_len);
1548#if IS_ENABLED(CONFIG_IPV6)
1549	if (sk->sk_family == AF_INET6)
1550		return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1551#endif
1552	return -EINVAL;
1553}
1554
1555int inet_gro_complete(struct sk_buff *skb, int nhoff)
1556{
1557	__be16 newlen = htons(skb->len - nhoff);
1558	struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1559	const struct net_offload *ops;
1560	int proto = iph->protocol;
1561	int err = -ENOSYS;
1562
1563	if (skb->encapsulation) {
1564		skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1565		skb_set_inner_network_header(skb, nhoff);
1566	}
1567
1568	csum_replace2(&iph->check, iph->tot_len, newlen);
1569	iph->tot_len = newlen;
1570
1571	rcu_read_lock();
1572	ops = rcu_dereference(inet_offloads[proto]);
1573	if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1574		goto out_unlock;
1575
1576	/* Only need to add sizeof(*iph) to get to the next hdr below
1577	 * because any hdr with option will have been flushed in
1578	 * inet_gro_receive().
1579	 */
1580	err = ops->callbacks.gro_complete(skb, nhoff + sizeof(*iph));
1581
1582out_unlock:
1583	rcu_read_unlock();
1584
1585	return err;
1586}
1587EXPORT_SYMBOL(inet_gro_complete);
1588
1589static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1590{
1591	skb->encapsulation = 1;
1592	skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1593	return inet_gro_complete(skb, nhoff);
1594}
1595
1596int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1597			 unsigned short type, unsigned char protocol,
1598			 struct net *net)
1599{
1600	struct socket *sock;
1601	int rc = sock_create_kern(net, family, type, protocol, &sock);
1602
1603	if (rc == 0) {
1604		*sk = sock->sk;
1605		(*sk)->sk_allocation = GFP_ATOMIC;
1606		/*
1607		 * Unhash it so that IP input processing does not even see it,
1608		 * we do not wish this socket to see incoming packets.
1609		 */
1610		(*sk)->sk_prot->unhash(*sk);
 
 
1611	}
1612	return rc;
1613}
1614EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1615
1616u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt)
1617{
1618	return  *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt);
1619}
1620EXPORT_SYMBOL_GPL(snmp_get_cpu_field);
1621
1622unsigned long snmp_fold_field(void __percpu *mib, int offt)
1623{
1624	unsigned long res = 0;
1625	int i;
1626
1627	for_each_possible_cpu(i)
1628		res += snmp_get_cpu_field(mib, i, offt);
 
 
1629	return res;
1630}
1631EXPORT_SYMBOL_GPL(snmp_fold_field);
1632
1633#if BITS_PER_LONG==32
1634
1635u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1636			 size_t syncp_offset)
1637{
1638	void *bhptr;
1639	struct u64_stats_sync *syncp;
1640	u64 v;
1641	unsigned int start;
1642
1643	bhptr = per_cpu_ptr(mib, cpu);
1644	syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1645	do {
1646		start = u64_stats_fetch_begin_irq(syncp);
1647		v = *(((u64 *)bhptr) + offt);
1648	} while (u64_stats_fetch_retry_irq(syncp, start));
1649
1650	return v;
1651}
1652EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1653
1654u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1655{
1656	u64 res = 0;
1657	int cpu;
1658
1659	for_each_possible_cpu(cpu) {
1660		res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
 
 
 
 
 
 
 
 
 
 
 
 
1661	}
1662	return res;
1663}
1664EXPORT_SYMBOL_GPL(snmp_fold_field64);
1665#endif
1666
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1667#ifdef CONFIG_IP_MULTICAST
1668static const struct net_protocol igmp_protocol = {
1669	.handler =	igmp_rcv,
1670	.netns_ok =	1,
1671};
1672#endif
1673
1674/* thinking of making this const? Don't.
1675 * early_demux can change based on sysctl.
1676 */
1677static struct net_protocol tcp_protocol = {
1678	.early_demux	=	tcp_v4_early_demux,
1679	.early_demux_handler =  tcp_v4_early_demux,
1680	.handler	=	tcp_v4_rcv,
1681	.err_handler	=	tcp_v4_err,
1682	.no_policy	=	1,
1683	.netns_ok	=	1,
1684	.icmp_strict_tag_validation = 1,
1685};
1686
1687/* thinking of making this const? Don't.
1688 * early_demux can change based on sysctl.
1689 */
1690static struct net_protocol udp_protocol = {
1691	.early_demux =	udp_v4_early_demux,
1692	.early_demux_handler =	udp_v4_early_demux,
1693	.handler =	udp_rcv,
1694	.err_handler =	udp_err,
1695	.no_policy =	1,
1696	.netns_ok =	1,
1697};
1698
1699static const struct net_protocol icmp_protocol = {
1700	.handler =	icmp_rcv,
1701	.err_handler =	icmp_err,
1702	.no_policy =	1,
1703	.netns_ok =	1,
1704};
1705
1706static __net_init int ipv4_mib_init_net(struct net *net)
1707{
1708	int i;
1709
1710	net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1711	if (!net->mib.tcp_statistics)
 
1712		goto err_tcp_mib;
1713	net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1714	if (!net->mib.ip_statistics)
 
1715		goto err_ip_mib;
1716
1717	for_each_possible_cpu(i) {
1718		struct ipstats_mib *af_inet_stats;
1719		af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
 
 
 
1720		u64_stats_init(&af_inet_stats->syncp);
 
1721	}
1722
1723	net->mib.net_statistics = alloc_percpu(struct linux_mib);
1724	if (!net->mib.net_statistics)
 
1725		goto err_net_mib;
1726	net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1727	if (!net->mib.udp_statistics)
 
1728		goto err_udp_mib;
1729	net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1730	if (!net->mib.udplite_statistics)
 
1731		goto err_udplite_mib;
1732	net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1733	if (!net->mib.icmp_statistics)
 
1734		goto err_icmp_mib;
1735	net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1736					      GFP_KERNEL);
1737	if (!net->mib.icmpmsg_statistics)
1738		goto err_icmpmsg_mib;
1739
1740	tcp_mib_init(net);
1741	return 0;
1742
1743err_icmpmsg_mib:
1744	free_percpu(net->mib.icmp_statistics);
1745err_icmp_mib:
1746	free_percpu(net->mib.udplite_statistics);
1747err_udplite_mib:
1748	free_percpu(net->mib.udp_statistics);
1749err_udp_mib:
1750	free_percpu(net->mib.net_statistics);
1751err_net_mib:
1752	free_percpu(net->mib.ip_statistics);
1753err_ip_mib:
1754	free_percpu(net->mib.tcp_statistics);
1755err_tcp_mib:
1756	return -ENOMEM;
1757}
1758
1759static __net_exit void ipv4_mib_exit_net(struct net *net)
1760{
1761	kfree(net->mib.icmpmsg_statistics);
1762	free_percpu(net->mib.icmp_statistics);
1763	free_percpu(net->mib.udplite_statistics);
1764	free_percpu(net->mib.udp_statistics);
1765	free_percpu(net->mib.net_statistics);
1766	free_percpu(net->mib.ip_statistics);
1767	free_percpu(net->mib.tcp_statistics);
1768}
1769
1770static __net_initdata struct pernet_operations ipv4_mib_ops = {
1771	.init = ipv4_mib_init_net,
1772	.exit = ipv4_mib_exit_net,
1773};
1774
1775static int __init init_ipv4_mibs(void)
1776{
1777	return register_pernet_subsys(&ipv4_mib_ops);
1778}
1779
1780static __net_init int inet_init_net(struct net *net)
1781{
1782	/*
1783	 * Set defaults for local port range
1784	 */
1785	seqlock_init(&net->ipv4.ip_local_ports.lock);
1786	net->ipv4.ip_local_ports.range[0] =  32768;
1787	net->ipv4.ip_local_ports.range[1] =  60999;
1788
1789	seqlock_init(&net->ipv4.ping_group_range.lock);
1790	/*
1791	 * Sane defaults - nobody may create ping sockets.
1792	 * Boot scripts should set this to distro-specific group.
1793	 */
1794	net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1795	net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1796
1797	/* Default values for sysctl-controlled parameters.
1798	 * We set them here, in case sysctl is not compiled.
1799	 */
1800	net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1801	net->ipv4.sysctl_ip_dynaddr = 0;
1802	net->ipv4.sysctl_ip_early_demux = 1;
1803	net->ipv4.sysctl_udp_early_demux = 1;
1804	net->ipv4.sysctl_tcp_early_demux = 1;
1805#ifdef CONFIG_SYSCTL
1806	net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1807#endif
1808
1809	/* Some igmp sysctl, whose values are always used */
1810	net->ipv4.sysctl_igmp_max_memberships = 20;
1811	net->ipv4.sysctl_igmp_max_msf = 10;
1812	/* IGMP reports for link-local multicast groups are enabled by default */
1813	net->ipv4.sysctl_igmp_llm_reports = 1;
1814	net->ipv4.sysctl_igmp_qrv = 2;
1815
1816	return 0;
1817}
1818
1819static __net_exit void inet_exit_net(struct net *net)
1820{
1821}
1822
1823static __net_initdata struct pernet_operations af_inet_ops = {
1824	.init = inet_init_net,
1825	.exit = inet_exit_net,
1826};
1827
1828static int __init init_inet_pernet_ops(void)
1829{
1830	return register_pernet_subsys(&af_inet_ops);
1831}
1832
1833static int ipv4_proc_init(void);
1834
1835/*
1836 *	IP protocol layer initialiser
1837 */
1838
1839static struct packet_offload ip_packet_offload __read_mostly = {
1840	.type = cpu_to_be16(ETH_P_IP),
1841	.callbacks = {
 
1842		.gso_segment = inet_gso_segment,
1843		.gro_receive = inet_gro_receive,
1844		.gro_complete = inet_gro_complete,
1845	},
1846};
1847
1848static const struct net_offload ipip_offload = {
1849	.callbacks = {
 
1850		.gso_segment	= inet_gso_segment,
1851		.gro_receive	= ipip_gro_receive,
1852		.gro_complete	= ipip_gro_complete,
1853	},
1854};
1855
1856static int __init ipip_offload_init(void)
1857{
1858	return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1859}
1860
1861static int __init ipv4_offload_init(void)
1862{
1863	/*
1864	 * Add offloads
1865	 */
1866	if (udpv4_offload_init() < 0)
1867		pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1868	if (tcpv4_offload_init() < 0)
1869		pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1870	if (ipip_offload_init() < 0)
1871		pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1872
1873	dev_add_offload(&ip_packet_offload);
 
1874	return 0;
1875}
1876
1877fs_initcall(ipv4_offload_init);
1878
1879static struct packet_type ip_packet_type __read_mostly = {
1880	.type = cpu_to_be16(ETH_P_IP),
1881	.func = ip_rcv,
1882};
1883
1884static int __init inet_init(void)
1885{
1886	struct inet_protosw *q;
1887	struct list_head *r;
1888	int rc = -EINVAL;
1889
1890	sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
 
 
 
 
1891
1892	rc = proto_register(&tcp_prot, 1);
1893	if (rc)
1894		goto out;
1895
1896	rc = proto_register(&udp_prot, 1);
1897	if (rc)
1898		goto out_unregister_tcp_proto;
1899
1900	rc = proto_register(&raw_prot, 1);
1901	if (rc)
1902		goto out_unregister_udp_proto;
1903
1904	rc = proto_register(&ping_prot, 1);
1905	if (rc)
1906		goto out_unregister_raw_proto;
1907
1908	/*
1909	 *	Tell SOCKET that we are alive...
1910	 */
1911
1912	(void)sock_register(&inet_family_ops);
1913
1914#ifdef CONFIG_SYSCTL
1915	ip_static_sysctl_init();
1916#endif
1917
1918	/*
1919	 *	Add all the base protocols.
1920	 */
1921
1922	if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1923		pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1924	if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1925		pr_crit("%s: Cannot add UDP protocol\n", __func__);
1926	if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1927		pr_crit("%s: Cannot add TCP protocol\n", __func__);
1928#ifdef CONFIG_IP_MULTICAST
1929	if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1930		pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1931#endif
1932
1933	/* Register the socket-side information for inet_create. */
1934	for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1935		INIT_LIST_HEAD(r);
1936
1937	for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1938		inet_register_protosw(q);
1939
1940	/*
1941	 *	Set the ARP module up
1942	 */
1943
1944	arp_init();
1945
1946	/*
1947	 *	Set the IP module up
1948	 */
1949
1950	ip_init();
1951
 
 
1952	/* Setup TCP slab cache for open requests. */
1953	tcp_init();
1954
1955	/* Setup UDP memory threshold */
1956	udp_init();
1957
1958	/* Add UDP-Lite (RFC 3828) */
1959	udplite4_register();
1960
1961	ping_init();
1962
1963	/*
1964	 *	Set the ICMP layer up
1965	 */
1966
1967	if (icmp_init() < 0)
1968		panic("Failed to create the ICMP control socket.\n");
1969
1970	/*
1971	 *	Initialise the multicast router
1972	 */
1973#if defined(CONFIG_IP_MROUTE)
1974	if (ip_mr_init())
1975		pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1976#endif
1977
1978	if (init_inet_pernet_ops())
1979		pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
1980	/*
1981	 *	Initialise per-cpu ipv4 mibs
1982	 */
1983
1984	if (init_ipv4_mibs())
1985		pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1986
1987	ipv4_proc_init();
1988
1989	ipfrag_init();
1990
1991	dev_add_pack(&ip_packet_type);
1992
1993	ip_tunnel_core_init();
1994
1995	rc = 0;
1996out:
1997	return rc;
1998out_unregister_raw_proto:
1999	proto_unregister(&raw_prot);
2000out_unregister_udp_proto:
2001	proto_unregister(&udp_prot);
2002out_unregister_tcp_proto:
2003	proto_unregister(&tcp_prot);
 
 
2004	goto out;
2005}
2006
2007fs_initcall(inet_init);
2008
2009/* ------------------------------------------------------------------------ */
2010
2011#ifdef CONFIG_PROC_FS
2012static int __init ipv4_proc_init(void)
2013{
2014	int rc = 0;
2015
2016	if (raw_proc_init())
2017		goto out_raw;
2018	if (tcp4_proc_init())
2019		goto out_tcp;
2020	if (udp4_proc_init())
2021		goto out_udp;
2022	if (ping_proc_init())
2023		goto out_ping;
2024	if (ip_misc_proc_init())
2025		goto out_misc;
2026out:
2027	return rc;
2028out_misc:
2029	ping_proc_exit();
2030out_ping:
2031	udp4_proc_exit();
2032out_udp:
2033	tcp4_proc_exit();
2034out_tcp:
2035	raw_proc_exit();
2036out_raw:
2037	rc = -ENOMEM;
2038	goto out;
2039}
2040
2041#else /* CONFIG_PROC_FS */
2042static int __init ipv4_proc_init(void)
2043{
2044	return 0;
2045}
2046#endif /* CONFIG_PROC_FS */