Linux Audio

Check our new training course

Loading...
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 */
v6.2
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * INET		An implementation of the TCP/IP protocol suite for the LINUX
   4 *		operating system.  INET is implemented using the  BSD Socket
   5 *		interface as the means of communication with the user level.
   6 *
   7 *		PF_INET protocol family socket handler.
   8 *
   9 * Authors:	Ross Biro
  10 *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  11 *		Florian La Roche, <flla@stud.uni-sb.de>
  12 *		Alan Cox, <A.Cox@swansea.ac.uk>
  13 *
  14 * Changes (see also sock.c)
  15 *
  16 *		piggy,
  17 *		Karl Knutson	:	Socket protocol table
  18 *		A.N.Kuznetsov	:	Socket death error in accept().
  19 *		John Richardson :	Fix non blocking error in connect()
  20 *					so sockets that fail to connect
  21 *					don't return -EINPROGRESS.
  22 *		Alan Cox	:	Asynchronous I/O support
  23 *		Alan Cox	:	Keep correct socket pointer on sock
  24 *					structures
  25 *					when accept() ed
  26 *		Alan Cox	:	Semantics of SO_LINGER aren't state
  27 *					moved to close when you look carefully.
  28 *					With this fixed and the accept bug fixed
  29 *					some RPC stuff seems happier.
  30 *		Niibe Yutaka	:	4.4BSD style write async I/O
  31 *		Alan Cox,
  32 *		Tony Gale 	:	Fixed reuse semantics.
  33 *		Alan Cox	:	bind() shouldn't abort existing but dead
  34 *					sockets. Stops FTP netin:.. I hope.
  35 *		Alan Cox	:	bind() works correctly for RAW sockets.
  36 *					Note that FreeBSD at least was broken
  37 *					in this respect so be careful with
  38 *					compatibility tests...
  39 *		Alan Cox	:	routing cache support
  40 *		Alan Cox	:	memzero the socket structure for
  41 *					compactness.
  42 *		Matt Day	:	nonblock connect error handler
  43 *		Alan Cox	:	Allow large numbers of pending sockets
  44 *					(eg for big web sites), but only if
  45 *					specifically application requested.
  46 *		Alan Cox	:	New buffering throughout IP. Used
  47 *					dumbly.
  48 *		Alan Cox	:	New buffering now used smartly.
  49 *		Alan Cox	:	BSD rather than common sense
  50 *					interpretation of listen.
  51 *		Germano Caronni	:	Assorted small races.
  52 *		Alan Cox	:	sendmsg/recvmsg basic support.
  53 *		Alan Cox	:	Only sendmsg/recvmsg now supported.
  54 *		Alan Cox	:	Locked down bind (see security list).
  55 *		Alan Cox	:	Loosened bind a little.
  56 *		Mike McLagan	:	ADD/DEL DLCI Ioctls
  57 *	Willy Konynenberg	:	Transparent proxying support.
  58 *		David S. Miller	:	New socket lookup architecture.
  59 *					Some other random speedups.
  60 *		Cyrus Durgin	:	Cleaned up file for kmod hacks.
  61 *		Andi Kleen	:	Fix inet_stream_connect TCP race.
 
 
 
 
 
  62 */
  63
  64#define pr_fmt(fmt) "IPv4: " fmt
  65
  66#include <linux/err.h>
  67#include <linux/errno.h>
  68#include <linux/types.h>
  69#include <linux/socket.h>
  70#include <linux/in.h>
  71#include <linux/kernel.h>
  72#include <linux/kmod.h>
  73#include <linux/sched.h>
  74#include <linux/timer.h>
  75#include <linux/string.h>
  76#include <linux/sockios.h>
  77#include <linux/net.h>
  78#include <linux/capability.h>
  79#include <linux/fcntl.h>
  80#include <linux/mm.h>
  81#include <linux/interrupt.h>
  82#include <linux/stat.h>
  83#include <linux/init.h>
  84#include <linux/poll.h>
  85#include <linux/netfilter_ipv4.h>
  86#include <linux/random.h>
  87#include <linux/slab.h>
  88
  89#include <linux/uaccess.h>
  90
  91#include <linux/inet.h>
  92#include <linux/igmp.h>
  93#include <linux/inetdevice.h>
  94#include <linux/netdevice.h>
  95#include <net/checksum.h>
  96#include <net/ip.h>
  97#include <net/protocol.h>
  98#include <net/arp.h>
  99#include <net/route.h>
 100#include <net/ip_fib.h>
 101#include <net/inet_connection_sock.h>
 102#include <net/gro.h>
 103#include <net/tcp.h>
 104#include <net/udp.h>
 105#include <net/udplite.h>
 106#include <net/ping.h>
 107#include <linux/skbuff.h>
 108#include <net/sock.h>
 109#include <net/raw.h>
 110#include <net/icmp.h>
 111#include <net/inet_common.h>
 112#include <net/ip_tunnels.h>
 113#include <net/xfrm.h>
 114#include <net/net_namespace.h>
 115#include <net/secure_seq.h>
 116#ifdef CONFIG_IP_MROUTE
 117#include <linux/mroute.h>
 118#endif
 119#include <net/l3mdev.h>
 120#include <net/compat.h>
 121
 122#include <trace/events/sock.h>
 123
 124/* The inetsw table contains everything that inet_create needs to
 125 * build a new socket.
 126 */
 127static struct list_head inetsw[SOCK_MAX];
 128static DEFINE_SPINLOCK(inetsw_lock);
 129
 130/* New destruction routine */
 131
 132void inet_sock_destruct(struct sock *sk)
 133{
 134	struct inet_sock *inet = inet_sk(sk);
 135
 136	__skb_queue_purge(&sk->sk_receive_queue);
 137	__skb_queue_purge(&sk->sk_error_queue);
 138
 139	sk_mem_reclaim_final(sk);
 140
 141	if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
 142		pr_err("Attempt to release TCP socket in state %d %p\n",
 143		       sk->sk_state, sk);
 144		return;
 145	}
 146	if (!sock_flag(sk, SOCK_DEAD)) {
 147		pr_err("Attempt to release alive inet socket %p\n", sk);
 148		return;
 149	}
 150
 151	WARN_ON_ONCE(atomic_read(&sk->sk_rmem_alloc));
 152	WARN_ON_ONCE(refcount_read(&sk->sk_wmem_alloc));
 153	WARN_ON_ONCE(sk->sk_wmem_queued);
 154	WARN_ON_ONCE(sk_forward_alloc_get(sk));
 155
 156	kfree(rcu_dereference_protected(inet->inet_opt, 1));
 157	dst_release(rcu_dereference_protected(sk->sk_dst_cache, 1));
 158	dst_release(rcu_dereference_protected(sk->sk_rx_dst, 1));
 159	sk_refcnt_debug_dec(sk);
 160}
 161EXPORT_SYMBOL(inet_sock_destruct);
 162
 163/*
 164 *	The routines beyond this point handle the behaviour of an AF_INET
 165 *	socket object. Mostly it punts to the subprotocols of IP to do
 166 *	the work.
 167 */
 168
 169/*
 170 *	Automatically bind an unbound socket.
 171 */
 172
 173static int inet_autobind(struct sock *sk)
 174{
 175	struct inet_sock *inet;
 176	/* We may need to bind the socket. */
 177	lock_sock(sk);
 178	inet = inet_sk(sk);
 179	if (!inet->inet_num) {
 180		if (sk->sk_prot->get_port(sk, 0)) {
 181			release_sock(sk);
 182			return -EAGAIN;
 183		}
 184		inet->inet_sport = htons(inet->inet_num);
 185	}
 186	release_sock(sk);
 187	return 0;
 188}
 189
 190/*
 191 *	Move a socket into listening state.
 192 */
 193int inet_listen(struct socket *sock, int backlog)
 194{
 195	struct sock *sk = sock->sk;
 196	unsigned char old_state;
 197	int err, tcp_fastopen;
 198
 199	lock_sock(sk);
 200
 201	err = -EINVAL;
 202	if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
 203		goto out;
 204
 205	old_state = sk->sk_state;
 206	if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
 207		goto out;
 208
 209	WRITE_ONCE(sk->sk_max_ack_backlog, backlog);
 210	/* Really, if the socket is already in listen state
 211	 * we can only allow the backlog to be adjusted.
 212	 */
 213	if (old_state != TCP_LISTEN) {
 214		/* Enable TFO w/o requiring TCP_FASTOPEN socket option.
 215		 * Note that only TCP sockets (SOCK_STREAM) will reach here.
 216		 * Also fastopen backlog may already been set via the option
 217		 * because the socket was in TCP_LISTEN state previously but
 218		 * was shutdown() rather than close().
 219		 */
 220		tcp_fastopen = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fastopen);
 221		if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
 222		    (tcp_fastopen & TFO_SERVER_ENABLE) &&
 223		    !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
 224			fastopen_queue_tune(sk, backlog);
 225			tcp_fastopen_init_key_once(sock_net(sk));
 226		}
 227
 228		err = inet_csk_listen_start(sk);
 229		if (err)
 230			goto out;
 231		tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL);
 232	}
 
 233	err = 0;
 234
 235out:
 236	release_sock(sk);
 237	return err;
 238}
 239EXPORT_SYMBOL(inet_listen);
 240
 241/*
 242 *	Create an inet socket.
 243 */
 244
 245static int inet_create(struct net *net, struct socket *sock, int protocol,
 246		       int kern)
 247{
 248	struct sock *sk;
 249	struct inet_protosw *answer;
 250	struct inet_sock *inet;
 251	struct proto *answer_prot;
 252	unsigned char answer_flags;
 253	int try_loading_module = 0;
 254	int err;
 255
 256	if (protocol < 0 || protocol >= IPPROTO_MAX)
 257		return -EINVAL;
 258
 259	sock->state = SS_UNCONNECTED;
 260
 261	/* Look for the requested type/protocol pair. */
 262lookup_protocol:
 263	err = -ESOCKTNOSUPPORT;
 264	rcu_read_lock();
 265	list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
 266
 267		err = 0;
 268		/* Check the non-wild match. */
 269		if (protocol == answer->protocol) {
 270			if (protocol != IPPROTO_IP)
 271				break;
 272		} else {
 273			/* Check for the two wild cases. */
 274			if (IPPROTO_IP == protocol) {
 275				protocol = answer->protocol;
 276				break;
 277			}
 278			if (IPPROTO_IP == answer->protocol)
 279				break;
 280		}
 281		err = -EPROTONOSUPPORT;
 282	}
 283
 284	if (unlikely(err)) {
 285		if (try_loading_module < 2) {
 286			rcu_read_unlock();
 287			/*
 288			 * Be more specific, e.g. net-pf-2-proto-132-type-1
 289			 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
 290			 */
 291			if (++try_loading_module == 1)
 292				request_module("net-pf-%d-proto-%d-type-%d",
 293					       PF_INET, protocol, sock->type);
 294			/*
 295			 * Fall back to generic, e.g. net-pf-2-proto-132
 296			 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
 297			 */
 298			else
 299				request_module("net-pf-%d-proto-%d",
 300					       PF_INET, protocol);
 301			goto lookup_protocol;
 302		} else
 303			goto out_rcu_unlock;
 304	}
 305
 306	err = -EPERM;
 307	if (sock->type == SOCK_RAW && !kern &&
 308	    !ns_capable(net->user_ns, CAP_NET_RAW))
 309		goto out_rcu_unlock;
 310
 311	sock->ops = answer->ops;
 312	answer_prot = answer->prot;
 313	answer_flags = answer->flags;
 314	rcu_read_unlock();
 315
 316	WARN_ON(!answer_prot->slab);
 317
 318	err = -ENOMEM;
 319	sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
 320	if (!sk)
 321		goto out;
 322
 323	err = 0;
 324	if (INET_PROTOSW_REUSE & answer_flags)
 325		sk->sk_reuse = SK_CAN_REUSE;
 326
 327	inet = inet_sk(sk);
 328	inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
 329
 330	inet->nodefrag = 0;
 331
 332	if (SOCK_RAW == sock->type) {
 333		inet->inet_num = protocol;
 334		if (IPPROTO_RAW == protocol)
 335			inet->hdrincl = 1;
 336	}
 337
 338	if (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc))
 339		inet->pmtudisc = IP_PMTUDISC_DONT;
 340	else
 341		inet->pmtudisc = IP_PMTUDISC_WANT;
 342
 343	inet->inet_id = 0;
 344
 345	sock_init_data(sock, sk);
 346
 347	sk->sk_destruct	   = inet_sock_destruct;
 348	sk->sk_protocol	   = protocol;
 349	sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
 350	sk->sk_txrehash = READ_ONCE(net->core.sysctl_txrehash);
 351
 352	inet->uc_ttl	= -1;
 353	inet->mc_loop	= 1;
 354	inet->mc_ttl	= 1;
 355	inet->mc_all	= 1;
 356	inet->mc_index	= 0;
 357	inet->mc_list	= NULL;
 358	inet->rcv_tos	= 0;
 359
 360	sk_refcnt_debug_inc(sk);
 361
 362	if (inet->inet_num) {
 363		/* It assumes that any protocol which allows
 364		 * the user to assign a number at socket
 365		 * creation time automatically
 366		 * shares.
 367		 */
 368		inet->inet_sport = htons(inet->inet_num);
 369		/* Add to protocol hash chains. */
 370		err = sk->sk_prot->hash(sk);
 371		if (err) {
 372			sk_common_release(sk);
 373			goto out;
 374		}
 375	}
 376
 377	if (sk->sk_prot->init) {
 378		err = sk->sk_prot->init(sk);
 379		if (err) {
 380			sk_common_release(sk);
 381			goto out;
 382		}
 383	}
 384
 385	if (!kern) {
 386		err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
 387		if (err) {
 388			sk_common_release(sk);
 389			goto out;
 390		}
 391	}
 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		if (!sk->sk_kern_sock)
 413			BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk);
 414
 415		/* Applications forget to leave groups before exiting */
 416		ip_mc_drop_socket(sk);
 417
 418		/* If linger is set, we don't return until the close
 419		 * is complete.  Otherwise we return immediately. The
 420		 * actually closing is done the same either way.
 421		 *
 422		 * If the close is due to the process exiting, we never
 423		 * linger..
 424		 */
 425		timeout = 0;
 426		if (sock_flag(sk, SOCK_LINGER) &&
 427		    !(current->flags & PF_EXITING))
 428			timeout = sk->sk_lingertime;
 
 429		sk->sk_prot->close(sk, timeout);
 430		sock->sk = NULL;
 431	}
 432	return 0;
 433}
 434EXPORT_SYMBOL(inet_release);
 435
 436int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 437{
 438	struct sock *sk = sock->sk;
 439	u32 flags = BIND_WITH_LOCK;
 440	int err;
 441
 442	/* If the socket has its own bind function then use it. (RAW) */
 443	if (sk->sk_prot->bind) {
 444		return sk->sk_prot->bind(sk, uaddr, addr_len);
 445	}
 446	if (addr_len < sizeof(struct sockaddr_in))
 447		return -EINVAL;
 448
 449	/* BPF prog is run before any checks are done so that if the prog
 450	 * changes context in a wrong way it will be caught.
 451	 */
 452	err = BPF_CGROUP_RUN_PROG_INET_BIND_LOCK(sk, uaddr,
 453						 CGROUP_INET4_BIND, &flags);
 454	if (err)
 455		return err;
 456
 457	return __inet_bind(sk, uaddr, addr_len, flags);
 458}
 459EXPORT_SYMBOL(inet_bind);
 460
 461int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
 462		u32 flags)
 463{
 464	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
 465	struct inet_sock *inet = inet_sk(sk);
 466	struct net *net = sock_net(sk);
 467	unsigned short snum;
 468	int chk_addr_ret;
 469	u32 tb_id = RT_TABLE_LOCAL;
 470	int err;
 471
 472	if (addr->sin_family != AF_INET) {
 473		/* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
 474		 * only if s_addr is INADDR_ANY.
 475		 */
 476		err = -EAFNOSUPPORT;
 477		if (addr->sin_family != AF_UNSPEC ||
 478		    addr->sin_addr.s_addr != htonl(INADDR_ANY))
 479			goto out;
 480	}
 481
 482	tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
 483	chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
 484
 485	/* Not specified by any standard per-se, however it breaks too
 486	 * many applications when removed.  It is unfortunate since
 487	 * allowing applications to make a non-local bind solves
 488	 * several problems with systems using dynamic addressing.
 489	 * (ie. your servers still start up even if your ISDN link
 490	 *  is temporarily down)
 491	 */
 492	err = -EADDRNOTAVAIL;
 493	if (!inet_addr_valid_or_nonlocal(net, inet, addr->sin_addr.s_addr,
 494	                                 chk_addr_ret))
 
 
 
 
 495		goto out;
 496
 497	snum = ntohs(addr->sin_port);
 498	err = -EACCES;
 499	if (!(flags & BIND_NO_CAP_NET_BIND_SERVICE) &&
 500	    snum && inet_port_requires_bind_service(net, snum) &&
 501	    !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
 502		goto out;
 503
 504	/*      We keep a pair of addresses. rcv_saddr is the one
 505	 *      used by hash lookups, and saddr is used for transmit.
 506	 *
 507	 *      In the BSD API these are the same except where it
 508	 *      would be illegal to use them (multicast/broadcast) in
 509	 *      which case the sending device address is used.
 510	 */
 511	if (flags & BIND_WITH_LOCK)
 512		lock_sock(sk);
 513
 514	/* Check these errors (active socket, double bind). */
 515	err = -EINVAL;
 516	if (sk->sk_state != TCP_CLOSE || inet->inet_num)
 517		goto out_release_sock;
 518
 519	inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
 520	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
 521		inet->inet_saddr = 0;  /* Use device */
 522
 523	/* Make sure we are allowed to bind here. */
 524	if (snum || !(inet->bind_address_no_port ||
 525		      (flags & BIND_FORCE_ADDRESS_NO_PORT))) {
 526		err = sk->sk_prot->get_port(sk, snum);
 
 
 
 
 
 527		if (err) {
 528			inet->inet_saddr = inet->inet_rcv_saddr = 0;
 529			goto out_release_sock;
 530		}
 531		if (!(flags & BIND_FROM_BPF)) {
 532			err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
 533			if (err) {
 534				inet->inet_saddr = inet->inet_rcv_saddr = 0;
 535				if (sk->sk_prot->put_port)
 536					sk->sk_prot->put_port(sk);
 537				goto out_release_sock;
 538			}
 539		}
 540	}
 541
 542	if (inet->inet_rcv_saddr)
 543		sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
 544	if (snum)
 545		sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
 546	inet->inet_sport = htons(inet->inet_num);
 547	inet->inet_daddr = 0;
 548	inet->inet_dport = 0;
 549	sk_dst_reset(sk);
 550	err = 0;
 551out_release_sock:
 552	if (flags & BIND_WITH_LOCK)
 553		release_sock(sk);
 554out:
 555	return err;
 556}
 557
 558int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
 559		       int addr_len, int flags)
 560{
 561	struct sock *sk = sock->sk;
 562	const struct proto *prot;
 563	int err;
 564
 565	if (addr_len < sizeof(uaddr->sa_family))
 566		return -EINVAL;
 567
 568	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
 569	prot = READ_ONCE(sk->sk_prot);
 570
 571	if (uaddr->sa_family == AF_UNSPEC)
 572		return prot->disconnect(sk, flags);
 573
 574	if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
 575		err = prot->pre_connect(sk, uaddr, addr_len);
 576		if (err)
 577			return err;
 578	}
 579
 580	if (data_race(!inet_sk(sk)->inet_num) && inet_autobind(sk))
 581		return -EAGAIN;
 582	return prot->connect(sk, uaddr, addr_len);
 583}
 584EXPORT_SYMBOL(inet_dgram_connect);
 585
 586static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
 587{
 588	DEFINE_WAIT_FUNC(wait, woken_wake_function);
 589
 590	add_wait_queue(sk_sleep(sk), &wait);
 591	sk->sk_write_pending += writebias;
 592
 593	/* Basic assumption: if someone sets sk->sk_err, he _must_
 594	 * change state of the socket from TCP_SYN_*.
 595	 * Connect() does not allow to get error notifications
 596	 * without closing the socket.
 597	 */
 598	while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 599		release_sock(sk);
 600		timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
 601		lock_sock(sk);
 602		if (signal_pending(current) || !timeo)
 603			break;
 604	}
 605	remove_wait_queue(sk_sleep(sk), &wait);
 606	sk->sk_write_pending -= writebias;
 607	return timeo;
 608}
 609
 610/*
 611 *	Connect to a remote host. There is regrettably still a little
 612 *	TCP 'magic' in here.
 613 */
 614int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 615			  int addr_len, int flags, int is_sendmsg)
 616{
 617	struct sock *sk = sock->sk;
 618	int err;
 619	long timeo;
 620
 621	/*
 622	 * uaddr can be NULL and addr_len can be 0 if:
 623	 * sk is a TCP fastopen active socket and
 624	 * TCP_FASTOPEN_CONNECT sockopt is set and
 625	 * we already have a valid cookie for this socket.
 626	 * In this case, user can call write() after connect().
 627	 * write() will invoke tcp_sendmsg_fastopen() which calls
 628	 * __inet_stream_connect().
 629	 */
 630	if (uaddr) {
 631		if (addr_len < sizeof(uaddr->sa_family))
 632			return -EINVAL;
 633
 634		if (uaddr->sa_family == AF_UNSPEC) {
 635			err = sk->sk_prot->disconnect(sk, flags);
 636			sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 637			goto out;
 638		}
 639	}
 640
 641	switch (sock->state) {
 642	default:
 643		err = -EINVAL;
 644		goto out;
 645	case SS_CONNECTED:
 646		err = -EISCONN;
 647		goto out;
 648	case SS_CONNECTING:
 649		if (inet_sk(sk)->defer_connect)
 650			err = is_sendmsg ? -EINPROGRESS : -EISCONN;
 651		else
 652			err = -EALREADY;
 653		/* Fall out of switch with err, set for this state */
 654		break;
 655	case SS_UNCONNECTED:
 656		err = -EISCONN;
 657		if (sk->sk_state != TCP_CLOSE)
 658			goto out;
 659
 660		if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
 661			err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
 662			if (err)
 663				goto out;
 664		}
 665
 666		err = sk->sk_prot->connect(sk, uaddr, addr_len);
 667		if (err < 0)
 668			goto out;
 669
 670		sock->state = SS_CONNECTING;
 671
 672		if (!err && inet_sk(sk)->defer_connect)
 673			goto out;
 674
 675		/* Just entered SS_CONNECTING state; the only
 676		 * difference is that return value in non-blocking
 677		 * case is EINPROGRESS, rather than EALREADY.
 678		 */
 679		err = -EINPROGRESS;
 680		break;
 681	}
 682
 683	timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
 684
 685	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 686		int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
 687				tcp_sk(sk)->fastopen_req &&
 688				tcp_sk(sk)->fastopen_req->data ? 1 : 0;
 689
 690		/* Error code is set above */
 691		if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
 692			goto out;
 693
 694		err = sock_intr_errno(timeo);
 695		if (signal_pending(current))
 696			goto out;
 697	}
 698
 699	/* Connection was closed by RST, timeout, ICMP error
 700	 * or another process disconnected us.
 701	 */
 702	if (sk->sk_state == TCP_CLOSE)
 703		goto sock_error;
 704
 705	/* sk->sk_err may be not zero now, if RECVERR was ordered by user
 706	 * and error was received after socket entered established state.
 707	 * Hence, it is handled normally after connect() return successfully.
 708	 */
 709
 710	sock->state = SS_CONNECTED;
 711	err = 0;
 712out:
 713	return err;
 714
 715sock_error:
 716	err = sock_error(sk) ? : -ECONNABORTED;
 717	sock->state = SS_UNCONNECTED;
 718	if (sk->sk_prot->disconnect(sk, flags))
 719		sock->state = SS_DISCONNECTING;
 720	goto out;
 721}
 722EXPORT_SYMBOL(__inet_stream_connect);
 723
 724int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 725			int addr_len, int flags)
 726{
 727	int err;
 728
 729	lock_sock(sock->sk);
 730	err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
 731	release_sock(sock->sk);
 732	return err;
 733}
 734EXPORT_SYMBOL(inet_stream_connect);
 735
 736/*
 737 *	Accept a pending connection. The TCP layer now gives BSD semantics.
 738 */
 739
 740int inet_accept(struct socket *sock, struct socket *newsock, int flags,
 741		bool kern)
 742{
 743	struct sock *sk1 = sock->sk, *sk2;
 744	int err = -EINVAL;
 
 745
 746	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
 747	sk2 = READ_ONCE(sk1->sk_prot)->accept(sk1, flags, &err, kern);
 748	if (!sk2)
 749		goto do_err;
 750
 751	lock_sock(sk2);
 752
 753	sock_rps_record_flow(sk2);
 754	WARN_ON(!((1 << sk2->sk_state) &
 755		  (TCPF_ESTABLISHED | TCPF_SYN_RECV |
 756		  TCPF_CLOSE_WAIT | TCPF_CLOSE)));
 757
 758	if (test_bit(SOCK_SUPPORT_ZC, &sock->flags))
 759		set_bit(SOCK_SUPPORT_ZC, &newsock->flags);
 760	sock_graft(sk2, newsock);
 761
 762	newsock->state = SS_CONNECTED;
 763	err = 0;
 764	release_sock(sk2);
 765do_err:
 766	return err;
 767}
 768EXPORT_SYMBOL(inet_accept);
 769
 
 770/*
 771 *	This does both peername and sockname.
 772 */
 773int inet_getname(struct socket *sock, struct sockaddr *uaddr,
 774		 int peer)
 775{
 776	struct sock *sk		= sock->sk;
 777	struct inet_sock *inet	= inet_sk(sk);
 778	DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
 779
 780	sin->sin_family = AF_INET;
 781	lock_sock(sk);
 782	if (peer) {
 783		if (!inet->inet_dport ||
 784		    (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
 785		     peer == 1)) {
 786			release_sock(sk);
 787			return -ENOTCONN;
 788		}
 789		sin->sin_port = inet->inet_dport;
 790		sin->sin_addr.s_addr = inet->inet_daddr;
 791		BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin,
 792				       CGROUP_INET4_GETPEERNAME);
 793	} else {
 794		__be32 addr = inet->inet_rcv_saddr;
 795		if (!addr)
 796			addr = inet->inet_saddr;
 797		sin->sin_port = inet->inet_sport;
 798		sin->sin_addr.s_addr = addr;
 799		BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin,
 800				       CGROUP_INET4_GETSOCKNAME);
 801	}
 802	release_sock(sk);
 803	memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 804	return sizeof(*sin);
 805}
 806EXPORT_SYMBOL(inet_getname);
 807
 808int inet_send_prepare(struct sock *sk)
 809{
 
 
 810	sock_rps_record_flow(sk);
 811
 812	/* We may need to bind the socket. */
 813	if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind &&
 814	    inet_autobind(sk))
 815		return -EAGAIN;
 816
 817	return 0;
 818}
 819EXPORT_SYMBOL_GPL(inet_send_prepare);
 820
 821int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
 822{
 823	struct sock *sk = sock->sk;
 824
 825	if (unlikely(inet_send_prepare(sk)))
 826		return -EAGAIN;
 827
 828	return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
 829			       sk, msg, size);
 830}
 831EXPORT_SYMBOL(inet_sendmsg);
 832
 833ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
 834		      size_t size, int flags)
 835{
 836	struct sock *sk = sock->sk;
 837	const struct proto *prot;
 838
 839	if (unlikely(inet_send_prepare(sk)))
 
 
 
 
 840		return -EAGAIN;
 841
 842	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
 843	prot = READ_ONCE(sk->sk_prot);
 844	if (prot->sendpage)
 845		return prot->sendpage(sk, page, offset, size, flags);
 846	return sock_no_sendpage(sock, page, offset, size, flags);
 847}
 848EXPORT_SYMBOL(inet_sendpage);
 849
 850INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
 851					  size_t, int, int *));
 852int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
 853		 int flags)
 854{
 855	struct sock *sk = sock->sk;
 856	int addr_len = 0;
 857	int err;
 858
 859	if (likely(!(flags & MSG_ERRQUEUE)))
 860		sock_rps_record_flow(sk);
 861
 862	err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
 863			      sk, msg, size, flags, &addr_len);
 864	if (err >= 0)
 865		msg->msg_namelen = addr_len;
 866	return err;
 867}
 868EXPORT_SYMBOL(inet_recvmsg);
 869
 870int inet_shutdown(struct socket *sock, int how)
 871{
 872	struct sock *sk = sock->sk;
 873	int err = 0;
 874
 875	/* This should really check to make sure
 876	 * the socket is a TCP socket. (WHY AC...)
 877	 */
 878	how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
 879		       1->2 bit 2 snds.
 880		       2->3 */
 881	if ((how & ~SHUTDOWN_MASK) || !how)	/* MAXINT->0 */
 882		return -EINVAL;
 883
 884	lock_sock(sk);
 885	if (sock->state == SS_CONNECTING) {
 886		if ((1 << sk->sk_state) &
 887		    (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
 888			sock->state = SS_DISCONNECTING;
 889		else
 890			sock->state = SS_CONNECTED;
 891	}
 892
 893	switch (sk->sk_state) {
 894	case TCP_CLOSE:
 895		err = -ENOTCONN;
 896		/* Hack to wake up other listeners, who can poll for
 897		   EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
 898		fallthrough;
 899	default:
 900		sk->sk_shutdown |= how;
 901		if (sk->sk_prot->shutdown)
 902			sk->sk_prot->shutdown(sk, how);
 903		break;
 904
 905	/* Remaining two branches are temporary solution for missing
 906	 * close() in multithreaded environment. It is _not_ a good idea,
 907	 * but we have no choice until close() is repaired at VFS level.
 908	 */
 909	case TCP_LISTEN:
 910		if (!(how & RCV_SHUTDOWN))
 911			break;
 912		fallthrough;
 913	case TCP_SYN_SENT:
 914		err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
 915		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 916		break;
 917	}
 918
 919	/* Wake up anyone sleeping in poll. */
 920	sk->sk_state_change(sk);
 921	release_sock(sk);
 922	return err;
 923}
 924EXPORT_SYMBOL(inet_shutdown);
 925
 926/*
 927 *	ioctl() calls you can issue on an INET socket. Most of these are
 928 *	device configuration and stuff and very rarely used. Some ioctls
 929 *	pass on to the socket itself.
 930 *
 931 *	NOTE: I like the idea of a module for the config stuff. ie ifconfig
 932 *	loads the devconfigure module does its configuring and unloads it.
 933 *	There's a good 20K of config code hanging around the kernel.
 934 */
 935
 936int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 937{
 938	struct sock *sk = sock->sk;
 939	int err = 0;
 940	struct net *net = sock_net(sk);
 941	void __user *p = (void __user *)arg;
 942	struct ifreq ifr;
 943	struct rtentry rt;
 944
 945	switch (cmd) {
 
 
 
 
 
 
 946	case SIOCADDRT:
 947	case SIOCDELRT:
 948		if (copy_from_user(&rt, p, sizeof(struct rtentry)))
 949			return -EFAULT;
 950		err = ip_rt_ioctl(net, cmd, &rt);
 951		break;
 952	case SIOCRTMSG:
 953		err = -EINVAL;
 954		break;
 955	case SIOCDARP:
 956	case SIOCGARP:
 957	case SIOCSARP:
 958		err = arp_ioctl(net, cmd, (void __user *)arg);
 959		break;
 960	case SIOCGIFADDR:
 961	case SIOCGIFBRDADDR:
 962	case SIOCGIFNETMASK:
 963	case SIOCGIFDSTADDR:
 964	case SIOCGIFPFLAGS:
 965		if (get_user_ifreq(&ifr, NULL, p))
 966			return -EFAULT;
 967		err = devinet_ioctl(net, cmd, &ifr);
 968		if (!err && put_user_ifreq(&ifr, p))
 969			err = -EFAULT;
 970		break;
 971
 972	case SIOCSIFADDR:
 973	case SIOCSIFBRDADDR:
 974	case SIOCSIFNETMASK:
 975	case SIOCSIFDSTADDR:
 976	case SIOCSIFPFLAGS:
 977	case SIOCSIFFLAGS:
 978		if (get_user_ifreq(&ifr, NULL, p))
 979			return -EFAULT;
 980		err = devinet_ioctl(net, cmd, &ifr);
 981		break;
 982	default:
 983		if (sk->sk_prot->ioctl)
 984			err = sk->sk_prot->ioctl(sk, cmd, arg);
 985		else
 986			err = -ENOIOCTLCMD;
 987		break;
 988	}
 989	return err;
 990}
 991EXPORT_SYMBOL(inet_ioctl);
 992
 993#ifdef CONFIG_COMPAT
 994static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd,
 995		struct compat_rtentry __user *ur)
 996{
 997	compat_uptr_t rtdev;
 998	struct rtentry rt;
 999
1000	if (copy_from_user(&rt.rt_dst, &ur->rt_dst,
1001			3 * sizeof(struct sockaddr)) ||
1002	    get_user(rt.rt_flags, &ur->rt_flags) ||
1003	    get_user(rt.rt_metric, &ur->rt_metric) ||
1004	    get_user(rt.rt_mtu, &ur->rt_mtu) ||
1005	    get_user(rt.rt_window, &ur->rt_window) ||
1006	    get_user(rt.rt_irtt, &ur->rt_irtt) ||
1007	    get_user(rtdev, &ur->rt_dev))
1008		return -EFAULT;
1009
1010	rt.rt_dev = compat_ptr(rtdev);
1011	return ip_rt_ioctl(sock_net(sk), cmd, &rt);
1012}
1013
1014static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1015{
1016	void __user *argp = compat_ptr(arg);
1017	struct sock *sk = sock->sk;
 
 
 
 
1018
1019	switch (cmd) {
1020	case SIOCADDRT:
1021	case SIOCDELRT:
1022		return inet_compat_routing_ioctl(sk, cmd, argp);
1023	default:
1024		if (!sk->sk_prot->compat_ioctl)
1025			return -ENOIOCTLCMD;
1026		return sk->sk_prot->compat_ioctl(sk, cmd, arg);
1027	}
1028}
1029#endif /* CONFIG_COMPAT */
1030
1031const struct proto_ops inet_stream_ops = {
1032	.family		   = PF_INET,
1033	.owner		   = THIS_MODULE,
1034	.release	   = inet_release,
1035	.bind		   = inet_bind,
1036	.connect	   = inet_stream_connect,
1037	.socketpair	   = sock_no_socketpair,
1038	.accept		   = inet_accept,
1039	.getname	   = inet_getname,
1040	.poll		   = tcp_poll,
1041	.ioctl		   = inet_ioctl,
1042	.gettstamp	   = sock_gettstamp,
1043	.listen		   = inet_listen,
1044	.shutdown	   = inet_shutdown,
1045	.setsockopt	   = sock_common_setsockopt,
1046	.getsockopt	   = sock_common_getsockopt,
1047	.sendmsg	   = inet_sendmsg,
1048	.recvmsg	   = inet_recvmsg,
1049#ifdef CONFIG_MMU
1050	.mmap		   = tcp_mmap,
1051#endif
1052	.sendpage	   = inet_sendpage,
1053	.splice_read	   = tcp_splice_read,
1054	.read_sock	   = tcp_read_sock,
1055	.read_skb	   = tcp_read_skb,
1056	.sendmsg_locked    = tcp_sendmsg_locked,
1057	.sendpage_locked   = tcp_sendpage_locked,
1058	.peek_len	   = tcp_peek_len,
1059#ifdef CONFIG_COMPAT
 
 
1060	.compat_ioctl	   = inet_compat_ioctl,
1061#endif
1062	.set_rcvlowat	   = tcp_set_rcvlowat,
1063};
1064EXPORT_SYMBOL(inet_stream_ops);
1065
1066const struct proto_ops inet_dgram_ops = {
1067	.family		   = PF_INET,
1068	.owner		   = THIS_MODULE,
1069	.release	   = inet_release,
1070	.bind		   = inet_bind,
1071	.connect	   = inet_dgram_connect,
1072	.socketpair	   = sock_no_socketpair,
1073	.accept		   = sock_no_accept,
1074	.getname	   = inet_getname,
1075	.poll		   = udp_poll,
1076	.ioctl		   = inet_ioctl,
1077	.gettstamp	   = sock_gettstamp,
1078	.listen		   = sock_no_listen,
1079	.shutdown	   = inet_shutdown,
1080	.setsockopt	   = sock_common_setsockopt,
1081	.getsockopt	   = sock_common_getsockopt,
1082	.sendmsg	   = inet_sendmsg,
1083	.read_skb	   = udp_read_skb,
1084	.recvmsg	   = inet_recvmsg,
1085	.mmap		   = sock_no_mmap,
1086	.sendpage	   = inet_sendpage,
1087	.set_peek_off	   = sk_set_peek_off,
1088#ifdef CONFIG_COMPAT
 
 
1089	.compat_ioctl	   = inet_compat_ioctl,
1090#endif
1091};
1092EXPORT_SYMBOL(inet_dgram_ops);
1093
1094/*
1095 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1096 * udp_poll
1097 */
1098static const struct proto_ops inet_sockraw_ops = {
1099	.family		   = PF_INET,
1100	.owner		   = THIS_MODULE,
1101	.release	   = inet_release,
1102	.bind		   = inet_bind,
1103	.connect	   = inet_dgram_connect,
1104	.socketpair	   = sock_no_socketpair,
1105	.accept		   = sock_no_accept,
1106	.getname	   = inet_getname,
1107	.poll		   = datagram_poll,
1108	.ioctl		   = inet_ioctl,
1109	.gettstamp	   = sock_gettstamp,
1110	.listen		   = sock_no_listen,
1111	.shutdown	   = inet_shutdown,
1112	.setsockopt	   = sock_common_setsockopt,
1113	.getsockopt	   = sock_common_getsockopt,
1114	.sendmsg	   = inet_sendmsg,
1115	.recvmsg	   = inet_recvmsg,
1116	.mmap		   = sock_no_mmap,
1117	.sendpage	   = inet_sendpage,
1118#ifdef CONFIG_COMPAT
 
 
1119	.compat_ioctl	   = inet_compat_ioctl,
1120#endif
1121};
1122
1123static const struct net_proto_family inet_family_ops = {
1124	.family = PF_INET,
1125	.create = inet_create,
1126	.owner	= THIS_MODULE,
1127};
1128
1129/* Upon startup we insert all the elements in inetsw_array[] into
1130 * the linked list inetsw.
1131 */
1132static struct inet_protosw inetsw_array[] =
1133{
1134	{
1135		.type =       SOCK_STREAM,
1136		.protocol =   IPPROTO_TCP,
1137		.prot =       &tcp_prot,
1138		.ops =        &inet_stream_ops,
1139		.flags =      INET_PROTOSW_PERMANENT |
1140			      INET_PROTOSW_ICSK,
1141	},
1142
1143	{
1144		.type =       SOCK_DGRAM,
1145		.protocol =   IPPROTO_UDP,
1146		.prot =       &udp_prot,
1147		.ops =        &inet_dgram_ops,
1148		.flags =      INET_PROTOSW_PERMANENT,
1149       },
1150
1151       {
1152		.type =       SOCK_DGRAM,
1153		.protocol =   IPPROTO_ICMP,
1154		.prot =       &ping_prot,
1155		.ops =        &inet_sockraw_ops,
1156		.flags =      INET_PROTOSW_REUSE,
1157       },
1158
1159       {
1160	       .type =       SOCK_RAW,
1161	       .protocol =   IPPROTO_IP,	/* wild card */
1162	       .prot =       &raw_prot,
1163	       .ops =        &inet_sockraw_ops,
1164	       .flags =      INET_PROTOSW_REUSE,
1165       }
1166};
1167
1168#define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1169
1170void inet_register_protosw(struct inet_protosw *p)
1171{
1172	struct list_head *lh;
1173	struct inet_protosw *answer;
1174	int protocol = p->protocol;
1175	struct list_head *last_perm;
1176
1177	spin_lock_bh(&inetsw_lock);
1178
1179	if (p->type >= SOCK_MAX)
1180		goto out_illegal;
1181
1182	/* If we are trying to override a permanent protocol, bail. */
1183	last_perm = &inetsw[p->type];
1184	list_for_each(lh, &inetsw[p->type]) {
1185		answer = list_entry(lh, struct inet_protosw, list);
1186		/* Check only the non-wild match. */
1187		if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1188			break;
1189		if (protocol == answer->protocol)
1190			goto out_permanent;
1191		last_perm = lh;
1192	}
1193
1194	/* Add the new entry after the last permanent entry if any, so that
1195	 * the new entry does not override a permanent entry when matched with
1196	 * a wild-card protocol. But it is allowed to override any existing
1197	 * non-permanent entry.  This means that when we remove this entry, the
1198	 * system automatically returns to the old behavior.
1199	 */
1200	list_add_rcu(&p->list, last_perm);
1201out:
1202	spin_unlock_bh(&inetsw_lock);
1203
1204	return;
1205
1206out_permanent:
1207	pr_err("Attempt to override permanent protocol %d\n", protocol);
1208	goto out;
1209
1210out_illegal:
1211	pr_err("Ignoring attempt to register invalid socket type %d\n",
1212	       p->type);
1213	goto out;
1214}
1215EXPORT_SYMBOL(inet_register_protosw);
1216
1217void inet_unregister_protosw(struct inet_protosw *p)
1218{
1219	if (INET_PROTOSW_PERMANENT & p->flags) {
1220		pr_err("Attempt to unregister permanent protocol %d\n",
1221		       p->protocol);
1222	} else {
1223		spin_lock_bh(&inetsw_lock);
1224		list_del_rcu(&p->list);
1225		spin_unlock_bh(&inetsw_lock);
1226
1227		synchronize_net();
1228	}
1229}
1230EXPORT_SYMBOL(inet_unregister_protosw);
1231
1232static int inet_sk_reselect_saddr(struct sock *sk)
1233{
1234	struct inet_sock *inet = inet_sk(sk);
1235	__be32 old_saddr = inet->inet_saddr;
1236	__be32 daddr = inet->inet_daddr;
1237	struct flowi4 *fl4;
1238	struct rtable *rt;
1239	__be32 new_saddr;
1240	struct ip_options_rcu *inet_opt;
1241	int err;
1242
1243	inet_opt = rcu_dereference_protected(inet->inet_opt,
1244					     lockdep_sock_is_held(sk));
1245	if (inet_opt && inet_opt->opt.srr)
1246		daddr = inet_opt->opt.faddr;
1247
1248	/* Query new route. */
1249	fl4 = &inet->cork.fl.u.ip4;
1250	rt = ip_route_connect(fl4, daddr, 0, sk->sk_bound_dev_if,
1251			      sk->sk_protocol, inet->inet_sport,
1252			      inet->inet_dport, sk);
1253	if (IS_ERR(rt))
1254		return PTR_ERR(rt);
1255
 
 
1256	new_saddr = fl4->saddr;
1257
1258	if (new_saddr == old_saddr) {
1259		sk_setup_caps(sk, &rt->dst);
1260		return 0;
1261	}
1262
1263	err = inet_bhash2_update_saddr(sk, &new_saddr, AF_INET);
1264	if (err) {
1265		ip_rt_put(rt);
1266		return err;
1267	}
1268
1269	sk_setup_caps(sk, &rt->dst);
1270
1271	if (READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) > 1) {
1272		pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1273			__func__, &old_saddr, &new_saddr);
1274	}
1275
 
 
1276	/*
1277	 * XXX The only one ugly spot where we need to
1278	 * XXX really change the sockets identity after
1279	 * XXX it has entered the hashes. -DaveM
1280	 *
1281	 * Besides that, it does not check for connection
1282	 * uniqueness. Wait for troubles.
1283	 */
1284	return __sk_prot_rehash(sk);
1285}
1286
1287int inet_sk_rebuild_header(struct sock *sk)
1288{
1289	struct inet_sock *inet = inet_sk(sk);
1290	struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1291	__be32 daddr;
1292	struct ip_options_rcu *inet_opt;
1293	struct flowi4 *fl4;
1294	int err;
1295
1296	/* Route is OK, nothing to do. */
1297	if (rt)
1298		return 0;
1299
1300	/* Reroute. */
1301	rcu_read_lock();
1302	inet_opt = rcu_dereference(inet->inet_opt);
1303	daddr = inet->inet_daddr;
1304	if (inet_opt && inet_opt->opt.srr)
1305		daddr = inet_opt->opt.faddr;
1306	rcu_read_unlock();
1307	fl4 = &inet->cork.fl.u.ip4;
1308	rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1309				   inet->inet_dport, inet->inet_sport,
1310				   sk->sk_protocol, RT_CONN_FLAGS(sk),
1311				   sk->sk_bound_dev_if);
1312	if (!IS_ERR(rt)) {
1313		err = 0;
1314		sk_setup_caps(sk, &rt->dst);
1315	} else {
1316		err = PTR_ERR(rt);
1317
1318		/* Routing failed... */
1319		sk->sk_route_caps = 0;
1320		/*
1321		 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1322		 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1323		 */
1324		if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) ||
1325		    sk->sk_state != TCP_SYN_SENT ||
1326		    (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1327		    (err = inet_sk_reselect_saddr(sk)) != 0)
1328			sk->sk_err_soft = -err;
1329	}
1330
1331	return err;
1332}
1333EXPORT_SYMBOL(inet_sk_rebuild_header);
1334
1335void inet_sk_set_state(struct sock *sk, int state)
1336{
1337	trace_inet_sock_set_state(sk, sk->sk_state, state);
1338	sk->sk_state = state;
1339}
1340EXPORT_SYMBOL(inet_sk_set_state);
1341
1342void inet_sk_state_store(struct sock *sk, int newstate)
1343{
1344	trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1345	smp_store_release(&sk->sk_state, newstate);
1346}
1347
1348struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1349				 netdev_features_t features)
1350{
1351	bool udpfrag = false, fixedid = false, gso_partial, encap;
1352	struct sk_buff *segs = ERR_PTR(-EINVAL);
1353	const struct net_offload *ops;
1354	unsigned int offset = 0;
1355	struct iphdr *iph;
1356	int proto, tot_len;
1357	int nhoff;
1358	int ihl;
1359	int id;
1360
1361	skb_reset_network_header(skb);
1362	nhoff = skb_network_header(skb) - skb_mac_header(skb);
1363	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1364		goto out;
1365
1366	iph = ip_hdr(skb);
1367	ihl = iph->ihl * 4;
1368	if (ihl < sizeof(*iph))
1369		goto out;
1370
1371	id = ntohs(iph->id);
1372	proto = iph->protocol;
1373
1374	/* Warning: after this point, iph might be no longer valid */
1375	if (unlikely(!pskb_may_pull(skb, ihl)))
1376		goto out;
1377	__skb_pull(skb, ihl);
1378
1379	encap = SKB_GSO_CB(skb)->encap_level > 0;
1380	if (encap)
1381		features &= skb->dev->hw_enc_features;
1382	SKB_GSO_CB(skb)->encap_level += ihl;
1383
1384	skb_reset_transport_header(skb);
1385
1386	segs = ERR_PTR(-EPROTONOSUPPORT);
1387
1388	if (!skb->encapsulation || encap) {
1389		udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1390		fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1391
1392		/* fixed ID is invalid if DF bit is not set */
1393		if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1394			goto out;
1395	}
1396
1397	ops = rcu_dereference(inet_offloads[proto]);
1398	if (likely(ops && ops->callbacks.gso_segment)) {
1399		segs = ops->callbacks.gso_segment(skb, features);
1400		if (!segs)
1401			skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
1402	}
1403
1404	if (IS_ERR_OR_NULL(segs))
1405		goto out;
1406
1407	gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1408
1409	skb = segs;
1410	do {
1411		iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1412		if (udpfrag) {
1413			iph->frag_off = htons(offset >> 3);
1414			if (skb->next)
1415				iph->frag_off |= htons(IP_MF);
1416			offset += skb->len - nhoff - ihl;
1417			tot_len = skb->len - nhoff;
1418		} else if (skb_is_gso(skb)) {
1419			if (!fixedid) {
1420				iph->id = htons(id);
1421				id += skb_shinfo(skb)->gso_segs;
1422			}
1423
1424			if (gso_partial)
1425				tot_len = skb_shinfo(skb)->gso_size +
1426					  SKB_GSO_CB(skb)->data_offset +
1427					  skb->head - (unsigned char *)iph;
1428			else
1429				tot_len = skb->len - nhoff;
1430		} else {
1431			if (!fixedid)
1432				iph->id = htons(id++);
1433			tot_len = skb->len - nhoff;
1434		}
1435		iph->tot_len = htons(tot_len);
1436		ip_send_check(iph);
1437		if (encap)
1438			skb_reset_inner_headers(skb);
1439		skb->network_header = (u8 *)iph - skb->head;
1440		skb_reset_mac_len(skb);
1441	} while ((skb = skb->next));
1442
1443out:
1444	return segs;
1445}
 
1446
1447static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1448					netdev_features_t features)
1449{
1450	if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1451		return ERR_PTR(-EINVAL);
1452
1453	return inet_gso_segment(skb, features);
1454}
1455
1456struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1457{
1458	const struct net_offload *ops;
1459	struct sk_buff *pp = NULL;
 
1460	const struct iphdr *iph;
1461	struct sk_buff *p;
1462	unsigned int hlen;
1463	unsigned int off;
1464	unsigned int id;
1465	int flush = 1;
1466	int proto;
1467
1468	off = skb_gro_offset(skb);
1469	hlen = off + sizeof(*iph);
1470	iph = skb_gro_header(skb, hlen, off);
1471	if (unlikely(!iph))
1472		goto out;
 
 
 
1473
1474	proto = iph->protocol;
1475
 
1476	ops = rcu_dereference(inet_offloads[proto]);
1477	if (!ops || !ops->callbacks.gro_receive)
1478		goto out;
1479
1480	if (*(u8 *)iph != 0x45)
1481		goto out;
1482
1483	if (ip_is_fragment(iph))
1484		goto out;
1485
1486	if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1487		goto out;
1488
1489	id = ntohl(*(__be32 *)&iph->id);
1490	flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1491	id >>= 16;
1492
1493	list_for_each_entry(p, head, list) {
1494		struct iphdr *iph2;
1495		u16 flush_id;
1496
1497		if (!NAPI_GRO_CB(p)->same_flow)
1498			continue;
1499
1500		iph2 = (struct iphdr *)(p->data + off);
1501		/* The above works because, with the exception of the top
1502		 * (inner most) layer, we only aggregate pkts with the same
1503		 * hdr length so all the hdrs we'll need to verify will start
1504		 * at the same offset.
1505		 */
1506		if ((iph->protocol ^ iph2->protocol) |
1507		    ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1508		    ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1509			NAPI_GRO_CB(p)->same_flow = 0;
1510			continue;
1511		}
1512
1513		/* All fields must match except length and checksum. */
1514		NAPI_GRO_CB(p)->flush |=
1515			(iph->ttl ^ iph2->ttl) |
1516			(iph->tos ^ iph2->tos) |
1517			((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1518
1519		NAPI_GRO_CB(p)->flush |= flush;
1520
1521		/* We need to store of the IP ID check to be included later
1522		 * when we can verify that this packet does in fact belong
1523		 * to a given flow.
1524		 */
1525		flush_id = (u16)(id - ntohs(iph2->id));
1526
1527		/* This bit of code makes it much easier for us to identify
1528		 * the cases where we are doing atomic vs non-atomic IP ID
1529		 * checks.  Specifically an atomic check can return IP ID
1530		 * values 0 - 0xFFFF, while a non-atomic check can only
1531		 * return 0 or 0xFFFF.
1532		 */
1533		if (!NAPI_GRO_CB(p)->is_atomic ||
1534		    !(iph->frag_off & htons(IP_DF))) {
1535			flush_id ^= NAPI_GRO_CB(p)->count;
1536			flush_id = flush_id ? 0xFFFF : 0;
1537		}
1538
1539		/* If the previous IP ID value was based on an atomic
1540		 * datagram we can overwrite the value and ignore it.
1541		 */
1542		if (NAPI_GRO_CB(skb)->is_atomic)
1543			NAPI_GRO_CB(p)->flush_id = flush_id;
1544		else
1545			NAPI_GRO_CB(p)->flush_id |= flush_id;
1546	}
1547
1548	NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1549	NAPI_GRO_CB(skb)->flush |= flush;
1550	skb_set_network_header(skb, off);
1551	/* The above will be needed by the transport layer if there is one
1552	 * immediately following this IP hdr.
1553	 */
1554
1555	/* Note : No need to call skb_gro_postpull_rcsum() here,
1556	 * as we already checked checksum over ipv4 header was 0
1557	 */
1558	skb_gro_pull(skb, sizeof(*iph));
1559	skb_set_transport_header(skb, skb_gro_offset(skb));
1560
1561	pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1562				       ops->callbacks.gro_receive, head, skb);
 
 
1563
1564out:
1565	skb_gro_flush_final(skb, pp, flush);
1566
1567	return pp;
1568}
 
1569
1570static struct sk_buff *ipip_gro_receive(struct list_head *head,
1571					struct sk_buff *skb)
1572{
1573	if (NAPI_GRO_CB(skb)->encap_mark) {
1574		NAPI_GRO_CB(skb)->flush = 1;
1575		return NULL;
1576	}
1577
1578	NAPI_GRO_CB(skb)->encap_mark = 1;
1579
1580	return inet_gro_receive(head, skb);
1581}
1582
1583#define SECONDS_PER_DAY	86400
1584
1585/* inet_current_timestamp - Return IP network timestamp
1586 *
1587 * Return milliseconds since midnight in network byte order.
1588 */
1589__be32 inet_current_timestamp(void)
1590{
1591	u32 secs;
1592	u32 msecs;
1593	struct timespec64 ts;
1594
1595	ktime_get_real_ts64(&ts);
1596
1597	/* Get secs since midnight. */
1598	(void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1599	/* Convert to msecs. */
1600	msecs = secs * MSEC_PER_SEC;
1601	/* Convert nsec to msec. */
1602	msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1603
1604	/* Convert to network byte order. */
1605	return htonl(msecs);
1606}
1607EXPORT_SYMBOL(inet_current_timestamp);
1608
1609int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1610{
1611	if (sk->sk_family == AF_INET)
1612		return ip_recv_error(sk, msg, len, addr_len);
1613#if IS_ENABLED(CONFIG_IPV6)
1614	if (sk->sk_family == AF_INET6)
1615		return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1616#endif
1617	return -EINVAL;
1618}
1619
1620int inet_gro_complete(struct sk_buff *skb, int nhoff)
1621{
1622	__be16 newlen = htons(skb->len - nhoff);
1623	struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1624	const struct net_offload *ops;
1625	int proto = iph->protocol;
1626	int err = -ENOSYS;
1627
1628	if (skb->encapsulation) {
1629		skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1630		skb_set_inner_network_header(skb, nhoff);
1631	}
1632
1633	csum_replace2(&iph->check, iph->tot_len, newlen);
1634	iph->tot_len = newlen;
1635
 
1636	ops = rcu_dereference(inet_offloads[proto]);
1637	if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1638		goto out;
1639
1640	/* Only need to add sizeof(*iph) to get to the next hdr below
1641	 * because any hdr with option will have been flushed in
1642	 * inet_gro_receive().
1643	 */
1644	err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1645			      tcp4_gro_complete, udp4_gro_complete,
1646			      skb, nhoff + sizeof(*iph));
 
1647
1648out:
1649	return err;
1650}
 
1651
1652static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1653{
1654	skb->encapsulation = 1;
1655	skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1656	return inet_gro_complete(skb, nhoff);
1657}
1658
1659int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1660			 unsigned short type, unsigned char protocol,
1661			 struct net *net)
1662{
1663	struct socket *sock;
1664	int rc = sock_create_kern(net, family, type, protocol, &sock);
1665
1666	if (rc == 0) {
1667		*sk = sock->sk;
1668		(*sk)->sk_allocation = GFP_ATOMIC;
1669		(*sk)->sk_use_task_frag = false;
1670		/*
1671		 * Unhash it so that IP input processing does not even see it,
1672		 * we do not wish this socket to see incoming packets.
1673		 */
1674		(*sk)->sk_prot->unhash(*sk);
1675	}
1676	return rc;
1677}
1678EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1679
 
 
 
 
 
 
1680unsigned long snmp_fold_field(void __percpu *mib, int offt)
1681{
1682	unsigned long res = 0;
1683	int i;
1684
1685	for_each_possible_cpu(i)
1686		res += snmp_get_cpu_field(mib, i, offt);
1687	return res;
1688}
1689EXPORT_SYMBOL_GPL(snmp_fold_field);
1690
1691#if BITS_PER_LONG==32
1692
1693u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1694			 size_t syncp_offset)
1695{
1696	void *bhptr;
1697	struct u64_stats_sync *syncp;
1698	u64 v;
1699	unsigned int start;
1700
1701	bhptr = per_cpu_ptr(mib, cpu);
1702	syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1703	do {
1704		start = u64_stats_fetch_begin(syncp);
1705		v = *(((u64 *)bhptr) + offt);
1706	} while (u64_stats_fetch_retry(syncp, start));
1707
1708	return v;
1709}
1710EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1711
1712u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1713{
1714	u64 res = 0;
1715	int cpu;
1716
1717	for_each_possible_cpu(cpu) {
1718		res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1719	}
1720	return res;
1721}
1722EXPORT_SYMBOL_GPL(snmp_fold_field64);
1723#endif
1724
1725#ifdef CONFIG_IP_MULTICAST
1726static const struct net_protocol igmp_protocol = {
1727	.handler =	igmp_rcv,
 
1728};
1729#endif
1730
1731static const struct net_protocol tcp_protocol = {
 
 
 
 
 
1732	.handler	=	tcp_v4_rcv,
1733	.err_handler	=	tcp_v4_err,
1734	.no_policy	=	1,
 
1735	.icmp_strict_tag_validation = 1,
1736};
1737
1738static const struct net_protocol udp_protocol = {
 
 
 
 
 
1739	.handler =	udp_rcv,
1740	.err_handler =	udp_err,
1741	.no_policy =	1,
 
1742};
1743
1744static const struct net_protocol icmp_protocol = {
1745	.handler =	icmp_rcv,
1746	.err_handler =	icmp_err,
1747	.no_policy =	1,
 
1748};
1749
1750static __net_init int ipv4_mib_init_net(struct net *net)
1751{
1752	int i;
1753
1754	net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1755	if (!net->mib.tcp_statistics)
1756		goto err_tcp_mib;
1757	net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1758	if (!net->mib.ip_statistics)
1759		goto err_ip_mib;
1760
1761	for_each_possible_cpu(i) {
1762		struct ipstats_mib *af_inet_stats;
1763		af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1764		u64_stats_init(&af_inet_stats->syncp);
1765	}
1766
1767	net->mib.net_statistics = alloc_percpu(struct linux_mib);
1768	if (!net->mib.net_statistics)
1769		goto err_net_mib;
1770	net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1771	if (!net->mib.udp_statistics)
1772		goto err_udp_mib;
1773	net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1774	if (!net->mib.udplite_statistics)
1775		goto err_udplite_mib;
1776	net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1777	if (!net->mib.icmp_statistics)
1778		goto err_icmp_mib;
1779	net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1780					      GFP_KERNEL);
1781	if (!net->mib.icmpmsg_statistics)
1782		goto err_icmpmsg_mib;
1783
1784	tcp_mib_init(net);
1785	return 0;
1786
1787err_icmpmsg_mib:
1788	free_percpu(net->mib.icmp_statistics);
1789err_icmp_mib:
1790	free_percpu(net->mib.udplite_statistics);
1791err_udplite_mib:
1792	free_percpu(net->mib.udp_statistics);
1793err_udp_mib:
1794	free_percpu(net->mib.net_statistics);
1795err_net_mib:
1796	free_percpu(net->mib.ip_statistics);
1797err_ip_mib:
1798	free_percpu(net->mib.tcp_statistics);
1799err_tcp_mib:
1800	return -ENOMEM;
1801}
1802
1803static __net_exit void ipv4_mib_exit_net(struct net *net)
1804{
1805	kfree(net->mib.icmpmsg_statistics);
1806	free_percpu(net->mib.icmp_statistics);
1807	free_percpu(net->mib.udplite_statistics);
1808	free_percpu(net->mib.udp_statistics);
1809	free_percpu(net->mib.net_statistics);
1810	free_percpu(net->mib.ip_statistics);
1811	free_percpu(net->mib.tcp_statistics);
1812#ifdef CONFIG_MPTCP
1813	/* allocated on demand, see mptcp_init_sock() */
1814	free_percpu(net->mib.mptcp_statistics);
1815#endif
1816}
1817
1818static __net_initdata struct pernet_operations ipv4_mib_ops = {
1819	.init = ipv4_mib_init_net,
1820	.exit = ipv4_mib_exit_net,
1821};
1822
1823static int __init init_ipv4_mibs(void)
1824{
1825	return register_pernet_subsys(&ipv4_mib_ops);
1826}
1827
1828static __net_init int inet_init_net(struct net *net)
1829{
1830	/*
1831	 * Set defaults for local port range
1832	 */
1833	seqlock_init(&net->ipv4.ip_local_ports.lock);
1834	net->ipv4.ip_local_ports.range[0] =  32768;
1835	net->ipv4.ip_local_ports.range[1] =  60999;
1836
1837	seqlock_init(&net->ipv4.ping_group_range.lock);
1838	/*
1839	 * Sane defaults - nobody may create ping sockets.
1840	 * Boot scripts should set this to distro-specific group.
1841	 */
1842	net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1843	net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1844
1845	/* Default values for sysctl-controlled parameters.
1846	 * We set them here, in case sysctl is not compiled.
1847	 */
1848	net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1849	net->ipv4.sysctl_ip_fwd_update_priority = 1;
1850	net->ipv4.sysctl_ip_dynaddr = 0;
1851	net->ipv4.sysctl_ip_early_demux = 1;
1852	net->ipv4.sysctl_udp_early_demux = 1;
1853	net->ipv4.sysctl_tcp_early_demux = 1;
1854	net->ipv4.sysctl_nexthop_compat_mode = 1;
1855#ifdef CONFIG_SYSCTL
1856	net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1857#endif
1858
1859	/* Some igmp sysctl, whose values are always used */
1860	net->ipv4.sysctl_igmp_max_memberships = 20;
1861	net->ipv4.sysctl_igmp_max_msf = 10;
1862	/* IGMP reports for link-local multicast groups are enabled by default */
1863	net->ipv4.sysctl_igmp_llm_reports = 1;
1864	net->ipv4.sysctl_igmp_qrv = 2;
1865
1866	net->ipv4.sysctl_fib_notify_on_flag_change = 0;
 
1867
1868	return 0;
 
1869}
1870
1871static __net_initdata struct pernet_operations af_inet_ops = {
1872	.init = inet_init_net,
 
1873};
1874
1875static int __init init_inet_pernet_ops(void)
1876{
1877	return register_pernet_subsys(&af_inet_ops);
1878}
1879
1880static int ipv4_proc_init(void);
1881
1882/*
1883 *	IP protocol layer initialiser
1884 */
1885
1886static struct packet_offload ip_packet_offload __read_mostly = {
1887	.type = cpu_to_be16(ETH_P_IP),
1888	.callbacks = {
1889		.gso_segment = inet_gso_segment,
1890		.gro_receive = inet_gro_receive,
1891		.gro_complete = inet_gro_complete,
1892	},
1893};
1894
1895static const struct net_offload ipip_offload = {
1896	.callbacks = {
1897		.gso_segment	= ipip_gso_segment,
1898		.gro_receive	= ipip_gro_receive,
1899		.gro_complete	= ipip_gro_complete,
1900	},
1901};
1902
1903static int __init ipip_offload_init(void)
1904{
1905	return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1906}
1907
1908static int __init ipv4_offload_init(void)
1909{
1910	/*
1911	 * Add offloads
1912	 */
1913	if (udpv4_offload_init() < 0)
1914		pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1915	if (tcpv4_offload_init() < 0)
1916		pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1917	if (ipip_offload_init() < 0)
1918		pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1919
1920	dev_add_offload(&ip_packet_offload);
1921	return 0;
1922}
1923
1924fs_initcall(ipv4_offload_init);
1925
1926static struct packet_type ip_packet_type __read_mostly = {
1927	.type = cpu_to_be16(ETH_P_IP),
1928	.func = ip_rcv,
1929	.list_func = ip_list_rcv,
1930};
1931
1932static int __init inet_init(void)
1933{
1934	struct inet_protosw *q;
1935	struct list_head *r;
1936	int rc;
1937
1938	sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1939
1940	raw_hashinfo_init(&raw_v4_hashinfo);
1941
1942	rc = proto_register(&tcp_prot, 1);
1943	if (rc)
1944		goto out;
1945
1946	rc = proto_register(&udp_prot, 1);
1947	if (rc)
1948		goto out_unregister_tcp_proto;
1949
1950	rc = proto_register(&raw_prot, 1);
1951	if (rc)
1952		goto out_unregister_udp_proto;
1953
1954	rc = proto_register(&ping_prot, 1);
1955	if (rc)
1956		goto out_unregister_raw_proto;
1957
1958	/*
1959	 *	Tell SOCKET that we are alive...
1960	 */
1961
1962	(void)sock_register(&inet_family_ops);
1963
1964#ifdef CONFIG_SYSCTL
1965	ip_static_sysctl_init();
1966#endif
1967
1968	/*
1969	 *	Add all the base protocols.
1970	 */
1971
1972	if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1973		pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1974	if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1975		pr_crit("%s: Cannot add UDP protocol\n", __func__);
1976	if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1977		pr_crit("%s: Cannot add TCP protocol\n", __func__);
1978#ifdef CONFIG_IP_MULTICAST
1979	if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1980		pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1981#endif
1982
1983	/* Register the socket-side information for inet_create. */
1984	for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1985		INIT_LIST_HEAD(r);
1986
1987	for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1988		inet_register_protosw(q);
1989
1990	/*
1991	 *	Set the ARP module up
1992	 */
1993
1994	arp_init();
1995
1996	/*
1997	 *	Set the IP module up
1998	 */
1999
2000	ip_init();
2001
2002	/* Initialise per-cpu ipv4 mibs */
2003	if (init_ipv4_mibs())
2004		panic("%s: Cannot init ipv4 mibs\n", __func__);
2005
2006	/* Setup TCP slab cache for open requests. */
2007	tcp_init();
2008
2009	/* Setup UDP memory threshold */
2010	udp_init();
2011
2012	/* Add UDP-Lite (RFC 3828) */
2013	udplite4_register();
2014
2015	raw_init();
2016
2017	ping_init();
2018
2019	/*
2020	 *	Set the ICMP layer up
2021	 */
2022
2023	if (icmp_init() < 0)
2024		panic("Failed to create the ICMP control socket.\n");
2025
2026	/*
2027	 *	Initialise the multicast router
2028	 */
2029#if defined(CONFIG_IP_MROUTE)
2030	if (ip_mr_init())
2031		pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2032#endif
2033
2034	if (init_inet_pernet_ops())
2035		pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
 
 
 
 
 
 
2036
2037	ipv4_proc_init();
2038
2039	ipfrag_init();
2040
2041	dev_add_pack(&ip_packet_type);
2042
2043	ip_tunnel_core_init();
2044
2045	rc = 0;
2046out:
2047	return rc;
2048out_unregister_raw_proto:
2049	proto_unregister(&raw_prot);
2050out_unregister_udp_proto:
2051	proto_unregister(&udp_prot);
2052out_unregister_tcp_proto:
2053	proto_unregister(&tcp_prot);
2054	goto out;
2055}
2056
2057fs_initcall(inet_init);
2058
2059/* ------------------------------------------------------------------------ */
2060
2061#ifdef CONFIG_PROC_FS
2062static int __init ipv4_proc_init(void)
2063{
2064	int rc = 0;
2065
2066	if (raw_proc_init())
2067		goto out_raw;
2068	if (tcp4_proc_init())
2069		goto out_tcp;
2070	if (udp4_proc_init())
2071		goto out_udp;
2072	if (ping_proc_init())
2073		goto out_ping;
2074	if (ip_misc_proc_init())
2075		goto out_misc;
2076out:
2077	return rc;
2078out_misc:
2079	ping_proc_exit();
2080out_ping:
2081	udp4_proc_exit();
2082out_udp:
2083	tcp4_proc_exit();
2084out_tcp:
2085	raw_proc_exit();
2086out_raw:
2087	rc = -ENOMEM;
2088	goto out;
2089}
2090
2091#else /* CONFIG_PROC_FS */
2092static int __init ipv4_proc_init(void)
2093{
2094	return 0;
2095}
2096#endif /* CONFIG_PROC_FS */