Linux Audio

Check our new training course

Loading...
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * INET		An implementation of the TCP/IP protocol suite for the LINUX
   4 *		operating system.  INET is implemented using the  BSD Socket
   5 *		interface as the means of communication with the user level.
   6 *
   7 *		PF_INET protocol family socket handler.
   8 *
   9 * Authors:	Ross Biro
  10 *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  11 *		Florian La Roche, <flla@stud.uni-sb.de>
  12 *		Alan Cox, <A.Cox@swansea.ac.uk>
  13 *
  14 * Changes (see also sock.c)
  15 *
  16 *		piggy,
  17 *		Karl Knutson	:	Socket protocol table
  18 *		A.N.Kuznetsov	:	Socket death error in accept().
  19 *		John Richardson :	Fix non blocking error in connect()
  20 *					so sockets that fail to connect
  21 *					don't return -EINPROGRESS.
  22 *		Alan Cox	:	Asynchronous I/O support
  23 *		Alan Cox	:	Keep correct socket pointer on sock
  24 *					structures
  25 *					when accept() ed
  26 *		Alan Cox	:	Semantics of SO_LINGER aren't state
  27 *					moved to close when you look carefully.
  28 *					With this fixed and the accept bug fixed
  29 *					some RPC stuff seems happier.
  30 *		Niibe Yutaka	:	4.4BSD style write async I/O
  31 *		Alan Cox,
  32 *		Tony Gale 	:	Fixed reuse semantics.
  33 *		Alan Cox	:	bind() shouldn't abort existing but dead
  34 *					sockets. Stops FTP netin:.. I hope.
  35 *		Alan Cox	:	bind() works correctly for RAW sockets.
  36 *					Note that FreeBSD at least was broken
  37 *					in this respect so be careful with
  38 *					compatibility tests...
  39 *		Alan Cox	:	routing cache support
  40 *		Alan Cox	:	memzero the socket structure for
  41 *					compactness.
  42 *		Matt Day	:	nonblock connect error handler
  43 *		Alan Cox	:	Allow large numbers of pending sockets
  44 *					(eg for big web sites), but only if
  45 *					specifically application requested.
  46 *		Alan Cox	:	New buffering throughout IP. Used
  47 *					dumbly.
  48 *		Alan Cox	:	New buffering now used smartly.
  49 *		Alan Cox	:	BSD rather than common sense
  50 *					interpretation of listen.
  51 *		Germano Caronni	:	Assorted small races.
  52 *		Alan Cox	:	sendmsg/recvmsg basic support.
  53 *		Alan Cox	:	Only sendmsg/recvmsg now supported.
  54 *		Alan Cox	:	Locked down bind (see security list).
  55 *		Alan Cox	:	Loosened bind a little.
  56 *		Mike McLagan	:	ADD/DEL DLCI Ioctls
  57 *	Willy Konynenberg	:	Transparent proxying support.
  58 *		David S. Miller	:	New socket lookup architecture.
  59 *					Some other random speedups.
  60 *		Cyrus Durgin	:	Cleaned up file for kmod hacks.
  61 *		Andi Kleen	:	Fix inet_stream_connect TCP race.
  62 */
  63
  64#define pr_fmt(fmt) "IPv4: " fmt
  65
  66#include <linux/err.h>
  67#include <linux/errno.h>
  68#include <linux/types.h>
  69#include <linux/socket.h>
  70#include <linux/in.h>
  71#include <linux/kernel.h>
  72#include <linux/kmod.h>
  73#include <linux/sched.h>
  74#include <linux/timer.h>
  75#include <linux/string.h>
  76#include <linux/sockios.h>
  77#include <linux/net.h>
  78#include <linux/capability.h>
  79#include <linux/fcntl.h>
  80#include <linux/mm.h>
  81#include <linux/interrupt.h>
  82#include <linux/stat.h>
  83#include <linux/init.h>
  84#include <linux/poll.h>
  85#include <linux/netfilter_ipv4.h>
  86#include <linux/random.h>
  87#include <linux/slab.h>
  88
  89#include <linux/uaccess.h>
  90
  91#include <linux/inet.h>
  92#include <linux/igmp.h>
  93#include <linux/inetdevice.h>
  94#include <linux/netdevice.h>
  95#include <net/checksum.h>
  96#include <net/ip.h>
  97#include <net/protocol.h>
  98#include <net/arp.h>
  99#include <net/route.h>
 100#include <net/ip_fib.h>
 101#include <net/inet_connection_sock.h>
 102#include <net/gro.h>
 103#include <net/gso.h>
 104#include <net/tcp.h>
 105#include <net/udp.h>
 106#include <net/udplite.h>
 107#include <net/ping.h>
 108#include <linux/skbuff.h>
 109#include <net/sock.h>
 110#include <net/raw.h>
 111#include <net/icmp.h>
 112#include <net/inet_common.h>
 113#include <net/ip_tunnels.h>
 114#include <net/xfrm.h>
 115#include <net/net_namespace.h>
 116#include <net/secure_seq.h>
 117#ifdef CONFIG_IP_MROUTE
 118#include <linux/mroute.h>
 119#endif
 120#include <net/l3mdev.h>
 121#include <net/compat.h>
 122
 123#include <trace/events/sock.h>
 124
 125/* The inetsw table contains everything that inet_create needs to
 126 * build a new socket.
 127 */
 128static struct list_head inetsw[SOCK_MAX];
 129static DEFINE_SPINLOCK(inetsw_lock);
 130
 131/* New destruction routine */
 132
 133void inet_sock_destruct(struct sock *sk)
 134{
 135	struct inet_sock *inet = inet_sk(sk);
 136
 137	__skb_queue_purge(&sk->sk_receive_queue);
 
 
 
 
 138	__skb_queue_purge(&sk->sk_error_queue);
 139
 140	sk_mem_reclaim_final(sk);
 141
 142	if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
 143		pr_err("Attempt to release TCP socket in state %d %p\n",
 144		       sk->sk_state, sk);
 145		return;
 146	}
 147	if (!sock_flag(sk, SOCK_DEAD)) {
 148		pr_err("Attempt to release alive inet socket %p\n", sk);
 149		return;
 150	}
 151
 152	WARN_ON_ONCE(atomic_read(&sk->sk_rmem_alloc));
 153	WARN_ON_ONCE(refcount_read(&sk->sk_wmem_alloc));
 154	WARN_ON_ONCE(sk->sk_wmem_queued);
 155	WARN_ON_ONCE(sk_forward_alloc_get(sk));
 156
 157	kfree(rcu_dereference_protected(inet->inet_opt, 1));
 158	dst_release(rcu_dereference_protected(sk->sk_dst_cache, 1));
 159	dst_release(rcu_dereference_protected(sk->sk_rx_dst, 1));
 
 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
 190int __inet_listen_sk(struct sock *sk, int backlog)
 
 
 
 191{
 192	unsigned char old_state = sk->sk_state;
 
 193	int err, tcp_fastopen;
 194
 
 
 
 
 
 
 
 195	if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
 196		return -EINVAL;
 197
 198	WRITE_ONCE(sk->sk_max_ack_backlog, backlog);
 199	/* Really, if the socket is already in listen state
 200	 * we can only allow the backlog to be adjusted.
 201	 */
 202	if (old_state != TCP_LISTEN) {
 203		/* Enable TFO w/o requiring TCP_FASTOPEN socket option.
 204		 * Note that only TCP sockets (SOCK_STREAM) will reach here.
 205		 * Also fastopen backlog may already been set via the option
 206		 * because the socket was in TCP_LISTEN state previously but
 207		 * was shutdown() rather than close().
 208		 */
 209		tcp_fastopen = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fastopen);
 210		if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
 211		    (tcp_fastopen & TFO_SERVER_ENABLE) &&
 212		    !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
 213			fastopen_queue_tune(sk, backlog);
 214			tcp_fastopen_init_key_once(sock_net(sk));
 215		}
 216
 217		err = inet_csk_listen_start(sk);
 218		if (err)
 219			return err;
 220
 221		tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL);
 222	}
 223	return 0;
 224}
 225
 226/*
 227 *	Move a socket into listening state.
 228 */
 229int inet_listen(struct socket *sock, int backlog)
 230{
 231	struct sock *sk = sock->sk;
 232	int err = -EINVAL;
 233
 234	lock_sock(sk);
 235
 236	if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
 237		goto out;
 238
 239	err = __inet_listen_sk(sk, backlog);
 240
 241out:
 242	release_sock(sk);
 243	return err;
 244}
 245EXPORT_SYMBOL(inet_listen);
 246
 247/*
 248 *	Create an inet socket.
 249 */
 250
 251static int inet_create(struct net *net, struct socket *sock, int protocol,
 252		       int kern)
 253{
 254	struct sock *sk;
 255	struct inet_protosw *answer;
 256	struct inet_sock *inet;
 257	struct proto *answer_prot;
 258	unsigned char answer_flags;
 259	int try_loading_module = 0;
 260	int err;
 261
 262	if (protocol < 0 || protocol >= IPPROTO_MAX)
 263		return -EINVAL;
 264
 265	sock->state = SS_UNCONNECTED;
 266
 267	/* Look for the requested type/protocol pair. */
 268lookup_protocol:
 269	err = -ESOCKTNOSUPPORT;
 270	rcu_read_lock();
 271	list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
 272
 273		err = 0;
 274		/* Check the non-wild match. */
 275		if (protocol == answer->protocol) {
 276			if (protocol != IPPROTO_IP)
 277				break;
 278		} else {
 279			/* Check for the two wild cases. */
 280			if (IPPROTO_IP == protocol) {
 281				protocol = answer->protocol;
 282				break;
 283			}
 284			if (IPPROTO_IP == answer->protocol)
 285				break;
 286		}
 287		err = -EPROTONOSUPPORT;
 288	}
 289
 290	if (unlikely(err)) {
 291		if (try_loading_module < 2) {
 292			rcu_read_unlock();
 293			/*
 294			 * Be more specific, e.g. net-pf-2-proto-132-type-1
 295			 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
 296			 */
 297			if (++try_loading_module == 1)
 298				request_module("net-pf-%d-proto-%d-type-%d",
 299					       PF_INET, protocol, sock->type);
 300			/*
 301			 * Fall back to generic, e.g. net-pf-2-proto-132
 302			 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
 303			 */
 304			else
 305				request_module("net-pf-%d-proto-%d",
 306					       PF_INET, protocol);
 307			goto lookup_protocol;
 308		} else
 309			goto out_rcu_unlock;
 310	}
 311
 312	err = -EPERM;
 313	if (sock->type == SOCK_RAW && !kern &&
 314	    !ns_capable(net->user_ns, CAP_NET_RAW))
 315		goto out_rcu_unlock;
 316
 317	sock->ops = answer->ops;
 318	answer_prot = answer->prot;
 319	answer_flags = answer->flags;
 320	rcu_read_unlock();
 321
 322	WARN_ON(!answer_prot->slab);
 323
 324	err = -ENOMEM;
 325	sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
 326	if (!sk)
 327		goto out;
 328
 329	err = 0;
 330	if (INET_PROTOSW_REUSE & answer_flags)
 331		sk->sk_reuse = SK_CAN_REUSE;
 332
 333	if (INET_PROTOSW_ICSK & answer_flags)
 334		inet_init_csk_locks(sk);
 335
 336	inet = inet_sk(sk);
 337	inet_assign_bit(IS_ICSK, sk, INET_PROTOSW_ICSK & answer_flags);
 338
 339	inet_clear_bit(NODEFRAG, sk);
 340
 341	if (SOCK_RAW == sock->type) {
 342		inet->inet_num = protocol;
 343		if (IPPROTO_RAW == protocol)
 344			inet_set_bit(HDRINCL, sk);
 345	}
 346
 347	if (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc))
 348		inet->pmtudisc = IP_PMTUDISC_DONT;
 349	else
 350		inet->pmtudisc = IP_PMTUDISC_WANT;
 351
 352	atomic_set(&inet->inet_id, 0);
 353
 354	sock_init_data(sock, sk);
 355
 356	sk->sk_destruct	   = inet_sock_destruct;
 357	sk->sk_protocol	   = protocol;
 358	sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
 359	sk->sk_txrehash = READ_ONCE(net->core.sysctl_txrehash);
 360
 361	inet->uc_ttl	= -1;
 362	inet_set_bit(MC_LOOP, sk);
 363	inet->mc_ttl	= 1;
 364	inet_set_bit(MC_ALL, sk);
 365	inet->mc_index	= 0;
 366	inet->mc_list	= NULL;
 367	inet->rcv_tos	= 0;
 368
 
 
 369	if (inet->inet_num) {
 370		/* It assumes that any protocol which allows
 371		 * the user to assign a number at socket
 372		 * creation time automatically
 373		 * shares.
 374		 */
 375		inet->inet_sport = htons(inet->inet_num);
 376		/* Add to protocol hash chains. */
 377		err = sk->sk_prot->hash(sk);
 378		if (err) {
 379			sk_common_release(sk);
 380			goto out;
 381		}
 382	}
 383
 384	if (sk->sk_prot->init) {
 385		err = sk->sk_prot->init(sk);
 386		if (err) {
 387			sk_common_release(sk);
 388			goto out;
 389		}
 390	}
 391
 392	if (!kern) {
 393		err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
 394		if (err) {
 395			sk_common_release(sk);
 396			goto out;
 397		}
 398	}
 399out:
 400	return err;
 401out_rcu_unlock:
 402	rcu_read_unlock();
 403	goto out;
 404}
 405
 406
 407/*
 408 *	The peer socket should always be NULL (or else). When we call this
 409 *	function we are destroying the object and from then on nobody
 410 *	should refer to it.
 411 */
 412int inet_release(struct socket *sock)
 413{
 414	struct sock *sk = sock->sk;
 415
 416	if (sk) {
 417		long timeout;
 418
 419		if (!sk->sk_kern_sock)
 420			BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk);
 421
 422		/* Applications forget to leave groups before exiting */
 423		ip_mc_drop_socket(sk);
 424
 425		/* If linger is set, we don't return until the close
 426		 * is complete.  Otherwise we return immediately. The
 427		 * actually closing is done the same either way.
 428		 *
 429		 * If the close is due to the process exiting, we never
 430		 * linger..
 431		 */
 432		timeout = 0;
 433		if (sock_flag(sk, SOCK_LINGER) &&
 434		    !(current->flags & PF_EXITING))
 435			timeout = sk->sk_lingertime;
 436		sk->sk_prot->close(sk, timeout);
 437		sock->sk = NULL;
 438	}
 439	return 0;
 440}
 441EXPORT_SYMBOL(inet_release);
 442
 443int inet_bind_sk(struct sock *sk, struct sockaddr *uaddr, int addr_len)
 444{
 445	u32 flags = BIND_WITH_LOCK;
 446	int err;
 447
 448	/* If the socket has its own bind function then use it. (RAW) */
 449	if (sk->sk_prot->bind) {
 450		return sk->sk_prot->bind(sk, uaddr, addr_len);
 451	}
 452	if (addr_len < sizeof(struct sockaddr_in))
 453		return -EINVAL;
 454
 455	/* BPF prog is run before any checks are done so that if the prog
 456	 * changes context in a wrong way it will be caught.
 457	 */
 458	err = BPF_CGROUP_RUN_PROG_INET_BIND_LOCK(sk, uaddr, &addr_len,
 459						 CGROUP_INET4_BIND, &flags);
 460	if (err)
 461		return err;
 462
 463	return __inet_bind(sk, uaddr, addr_len, flags);
 464}
 465
 466int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 467{
 468	return inet_bind_sk(sock->sk, uaddr, addr_len);
 469}
 470EXPORT_SYMBOL(inet_bind);
 471
 472int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
 473		u32 flags)
 474{
 475	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
 476	struct inet_sock *inet = inet_sk(sk);
 477	struct net *net = sock_net(sk);
 478	unsigned short snum;
 479	int chk_addr_ret;
 480	u32 tb_id = RT_TABLE_LOCAL;
 481	int err;
 482
 483	if (addr->sin_family != AF_INET) {
 484		/* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
 485		 * only if s_addr is INADDR_ANY.
 486		 */
 487		err = -EAFNOSUPPORT;
 488		if (addr->sin_family != AF_UNSPEC ||
 489		    addr->sin_addr.s_addr != htonl(INADDR_ANY))
 490			goto out;
 491	}
 492
 493	tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
 494	chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
 495
 496	/* Not specified by any standard per-se, however it breaks too
 497	 * many applications when removed.  It is unfortunate since
 498	 * allowing applications to make a non-local bind solves
 499	 * several problems with systems using dynamic addressing.
 500	 * (ie. your servers still start up even if your ISDN link
 501	 *  is temporarily down)
 502	 */
 503	err = -EADDRNOTAVAIL;
 504	if (!inet_addr_valid_or_nonlocal(net, inet, addr->sin_addr.s_addr,
 505	                                 chk_addr_ret))
 
 
 
 506		goto out;
 507
 508	snum = ntohs(addr->sin_port);
 509	err = -EACCES;
 510	if (!(flags & BIND_NO_CAP_NET_BIND_SERVICE) &&
 511	    snum && inet_port_requires_bind_service(net, snum) &&
 512	    !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
 513		goto out;
 514
 515	/*      We keep a pair of addresses. rcv_saddr is the one
 516	 *      used by hash lookups, and saddr is used for transmit.
 517	 *
 518	 *      In the BSD API these are the same except where it
 519	 *      would be illegal to use them (multicast/broadcast) in
 520	 *      which case the sending device address is used.
 521	 */
 522	if (flags & BIND_WITH_LOCK)
 523		lock_sock(sk);
 524
 525	/* Check these errors (active socket, double bind). */
 526	err = -EINVAL;
 527	if (sk->sk_state != TCP_CLOSE || inet->inet_num)
 528		goto out_release_sock;
 529
 530	inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
 531	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
 532		inet->inet_saddr = 0;  /* Use device */
 533
 534	/* Make sure we are allowed to bind here. */
 535	if (snum || !(inet_test_bit(BIND_ADDRESS_NO_PORT, sk) ||
 536		      (flags & BIND_FORCE_ADDRESS_NO_PORT))) {
 537		err = sk->sk_prot->get_port(sk, snum);
 538		if (err) {
 539			inet->inet_saddr = inet->inet_rcv_saddr = 0;
 
 540			goto out_release_sock;
 541		}
 542		if (!(flags & BIND_FROM_BPF)) {
 543			err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
 544			if (err) {
 545				inet->inet_saddr = inet->inet_rcv_saddr = 0;
 546				if (sk->sk_prot->put_port)
 547					sk->sk_prot->put_port(sk);
 548				goto out_release_sock;
 549			}
 550		}
 551	}
 552
 553	if (inet->inet_rcv_saddr)
 554		sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
 555	if (snum)
 556		sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
 557	inet->inet_sport = htons(inet->inet_num);
 558	inet->inet_daddr = 0;
 559	inet->inet_dport = 0;
 560	sk_dst_reset(sk);
 561	err = 0;
 562out_release_sock:
 563	if (flags & BIND_WITH_LOCK)
 564		release_sock(sk);
 565out:
 566	return err;
 567}
 568
 569int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
 570		       int addr_len, int flags)
 571{
 572	struct sock *sk = sock->sk;
 573	const struct proto *prot;
 574	int err;
 575
 576	if (addr_len < sizeof(uaddr->sa_family))
 577		return -EINVAL;
 578
 579	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
 580	prot = READ_ONCE(sk->sk_prot);
 581
 582	if (uaddr->sa_family == AF_UNSPEC)
 583		return prot->disconnect(sk, flags);
 584
 585	if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
 586		err = prot->pre_connect(sk, uaddr, addr_len);
 587		if (err)
 588			return err;
 589	}
 590
 591	if (data_race(!inet_sk(sk)->inet_num) && inet_autobind(sk))
 592		return -EAGAIN;
 593	return prot->connect(sk, uaddr, addr_len);
 594}
 595EXPORT_SYMBOL(inet_dgram_connect);
 596
 597static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
 598{
 599	DEFINE_WAIT_FUNC(wait, woken_wake_function);
 600
 601	add_wait_queue(sk_sleep(sk), &wait);
 602	sk->sk_write_pending += writebias;
 603
 604	/* Basic assumption: if someone sets sk->sk_err, he _must_
 605	 * change state of the socket from TCP_SYN_*.
 606	 * Connect() does not allow to get error notifications
 607	 * without closing the socket.
 608	 */
 609	while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 610		release_sock(sk);
 611		timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
 612		lock_sock(sk);
 613		if (signal_pending(current) || !timeo)
 614			break;
 615	}
 616	remove_wait_queue(sk_sleep(sk), &wait);
 617	sk->sk_write_pending -= writebias;
 618	return timeo;
 619}
 620
 621/*
 622 *	Connect to a remote host. There is regrettably still a little
 623 *	TCP 'magic' in here.
 624 */
 625int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 626			  int addr_len, int flags, int is_sendmsg)
 627{
 628	struct sock *sk = sock->sk;
 629	int err;
 630	long timeo;
 631
 632	/*
 633	 * uaddr can be NULL and addr_len can be 0 if:
 634	 * sk is a TCP fastopen active socket and
 635	 * TCP_FASTOPEN_CONNECT sockopt is set and
 636	 * we already have a valid cookie for this socket.
 637	 * In this case, user can call write() after connect().
 638	 * write() will invoke tcp_sendmsg_fastopen() which calls
 639	 * __inet_stream_connect().
 640	 */
 641	if (uaddr) {
 642		if (addr_len < sizeof(uaddr->sa_family))
 643			return -EINVAL;
 644
 645		if (uaddr->sa_family == AF_UNSPEC) {
 646			sk->sk_disconnects++;
 647			err = sk->sk_prot->disconnect(sk, flags);
 648			sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 649			goto out;
 650		}
 651	}
 652
 653	switch (sock->state) {
 654	default:
 655		err = -EINVAL;
 656		goto out;
 657	case SS_CONNECTED:
 658		err = -EISCONN;
 659		goto out;
 660	case SS_CONNECTING:
 661		if (inet_test_bit(DEFER_CONNECT, sk))
 662			err = is_sendmsg ? -EINPROGRESS : -EISCONN;
 663		else
 664			err = -EALREADY;
 665		/* Fall out of switch with err, set for this state */
 666		break;
 667	case SS_UNCONNECTED:
 668		err = -EISCONN;
 669		if (sk->sk_state != TCP_CLOSE)
 670			goto out;
 671
 672		if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
 673			err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
 674			if (err)
 675				goto out;
 676		}
 677
 678		err = sk->sk_prot->connect(sk, uaddr, addr_len);
 679		if (err < 0)
 680			goto out;
 681
 682		sock->state = SS_CONNECTING;
 683
 684		if (!err && inet_test_bit(DEFER_CONNECT, sk))
 685			goto out;
 686
 687		/* Just entered SS_CONNECTING state; the only
 688		 * difference is that return value in non-blocking
 689		 * case is EINPROGRESS, rather than EALREADY.
 690		 */
 691		err = -EINPROGRESS;
 692		break;
 693	}
 694
 695	timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
 696
 697	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 698		int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
 699				tcp_sk(sk)->fastopen_req &&
 700				tcp_sk(sk)->fastopen_req->data ? 1 : 0;
 701		int dis = sk->sk_disconnects;
 702
 703		/* Error code is set above */
 704		if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
 705			goto out;
 706
 707		err = sock_intr_errno(timeo);
 708		if (signal_pending(current))
 709			goto out;
 710
 711		if (dis != sk->sk_disconnects) {
 712			err = -EPIPE;
 713			goto out;
 714		}
 715	}
 716
 717	/* Connection was closed by RST, timeout, ICMP error
 718	 * or another process disconnected us.
 719	 */
 720	if (sk->sk_state == TCP_CLOSE)
 721		goto sock_error;
 722
 723	/* sk->sk_err may be not zero now, if RECVERR was ordered by user
 724	 * and error was received after socket entered established state.
 725	 * Hence, it is handled normally after connect() return successfully.
 726	 */
 727
 728	sock->state = SS_CONNECTED;
 729	err = 0;
 730out:
 731	return err;
 732
 733sock_error:
 734	err = sock_error(sk) ? : -ECONNABORTED;
 735	sock->state = SS_UNCONNECTED;
 736	sk->sk_disconnects++;
 737	if (sk->sk_prot->disconnect(sk, flags))
 738		sock->state = SS_DISCONNECTING;
 739	goto out;
 740}
 741EXPORT_SYMBOL(__inet_stream_connect);
 742
 743int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 744			int addr_len, int flags)
 745{
 746	int err;
 747
 748	lock_sock(sock->sk);
 749	err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
 750	release_sock(sock->sk);
 751	return err;
 752}
 753EXPORT_SYMBOL(inet_stream_connect);
 754
 755void __inet_accept(struct socket *sock, struct socket *newsock, struct sock *newsk)
 756{
 757	sock_rps_record_flow(newsk);
 758	WARN_ON(!((1 << newsk->sk_state) &
 759		  (TCPF_ESTABLISHED | TCPF_SYN_RECV |
 760		  TCPF_CLOSE_WAIT | TCPF_CLOSE)));
 761
 762	if (test_bit(SOCK_SUPPORT_ZC, &sock->flags))
 763		set_bit(SOCK_SUPPORT_ZC, &newsock->flags);
 764	sock_graft(newsk, newsock);
 765
 766	newsock->state = SS_CONNECTED;
 767}
 768
 769/*
 770 *	Accept a pending connection. The TCP layer now gives BSD semantics.
 771 */
 772
 773int inet_accept(struct socket *sock, struct socket *newsock, int flags,
 774		bool kern)
 775{
 776	struct sock *sk1 = sock->sk, *sk2;
 777	int err = -EINVAL;
 
 778
 779	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
 780	sk2 = READ_ONCE(sk1->sk_prot)->accept(sk1, flags, &err, kern);
 781	if (!sk2)
 782		return err;
 783
 784	lock_sock(sk2);
 785	__inet_accept(sock, newsock, sk2);
 
 
 
 
 
 
 
 
 
 786	release_sock(sk2);
 787	return 0;
 
 788}
 789EXPORT_SYMBOL(inet_accept);
 790
 791/*
 792 *	This does both peername and sockname.
 793 */
 794int inet_getname(struct socket *sock, struct sockaddr *uaddr,
 795		 int peer)
 796{
 797	struct sock *sk		= sock->sk;
 798	struct inet_sock *inet	= inet_sk(sk);
 799	DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
 800	int sin_addr_len = sizeof(*sin);
 801
 802	sin->sin_family = AF_INET;
 803	lock_sock(sk);
 804	if (peer) {
 805		if (!inet->inet_dport ||
 806		    (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
 807		     peer == 1)) {
 808			release_sock(sk);
 809			return -ENOTCONN;
 810		}
 811		sin->sin_port = inet->inet_dport;
 812		sin->sin_addr.s_addr = inet->inet_daddr;
 813		BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin, &sin_addr_len,
 814				       CGROUP_INET4_GETPEERNAME);
 815	} else {
 816		__be32 addr = inet->inet_rcv_saddr;
 817		if (!addr)
 818			addr = inet->inet_saddr;
 819		sin->sin_port = inet->inet_sport;
 820		sin->sin_addr.s_addr = addr;
 821		BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin, &sin_addr_len,
 822				       CGROUP_INET4_GETSOCKNAME);
 823	}
 824	release_sock(sk);
 
 
 
 
 825	memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 826	return sin_addr_len;
 827}
 828EXPORT_SYMBOL(inet_getname);
 829
 830int inet_send_prepare(struct sock *sk)
 831{
 832	sock_rps_record_flow(sk);
 833
 834	/* We may need to bind the socket. */
 835	if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind &&
 836	    inet_autobind(sk))
 837		return -EAGAIN;
 838
 839	return 0;
 840}
 841EXPORT_SYMBOL_GPL(inet_send_prepare);
 842
 843int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
 844{
 845	struct sock *sk = sock->sk;
 846
 847	if (unlikely(inet_send_prepare(sk)))
 848		return -EAGAIN;
 849
 850	return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
 851			       sk, msg, size);
 852}
 853EXPORT_SYMBOL(inet_sendmsg);
 854
 855void inet_splice_eof(struct socket *sock)
 
 856{
 857	const struct proto *prot;
 858	struct sock *sk = sock->sk;
 859
 860	if (unlikely(inet_send_prepare(sk)))
 861		return;
 862
 863	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
 864	prot = READ_ONCE(sk->sk_prot);
 865	if (prot->splice_eof)
 866		prot->splice_eof(sock);
 867}
 868EXPORT_SYMBOL_GPL(inet_splice_eof);
 869
 870INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
 871					  size_t, int, int *));
 872int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
 873		 int flags)
 874{
 875	struct sock *sk = sock->sk;
 876	int addr_len = 0;
 877	int err;
 878
 879	if (likely(!(flags & MSG_ERRQUEUE)))
 880		sock_rps_record_flow(sk);
 881
 882	err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
 883			      sk, msg, size, flags, &addr_len);
 
 884	if (err >= 0)
 885		msg->msg_namelen = addr_len;
 886	return err;
 887}
 888EXPORT_SYMBOL(inet_recvmsg);
 889
 890int inet_shutdown(struct socket *sock, int how)
 891{
 892	struct sock *sk = sock->sk;
 893	int err = 0;
 894
 895	/* This should really check to make sure
 896	 * the socket is a TCP socket. (WHY AC...)
 897	 */
 898	how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
 899		       1->2 bit 2 snds.
 900		       2->3 */
 901	if ((how & ~SHUTDOWN_MASK) || !how)	/* MAXINT->0 */
 902		return -EINVAL;
 903
 904	lock_sock(sk);
 905	if (sock->state == SS_CONNECTING) {
 906		if ((1 << sk->sk_state) &
 907		    (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
 908			sock->state = SS_DISCONNECTING;
 909		else
 910			sock->state = SS_CONNECTED;
 911	}
 912
 913	switch (sk->sk_state) {
 914	case TCP_CLOSE:
 915		err = -ENOTCONN;
 916		/* Hack to wake up other listeners, who can poll for
 917		   EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
 918		fallthrough;
 919	default:
 920		WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | how);
 921		if (sk->sk_prot->shutdown)
 922			sk->sk_prot->shutdown(sk, how);
 923		break;
 924
 925	/* Remaining two branches are temporary solution for missing
 926	 * close() in multithreaded environment. It is _not_ a good idea,
 927	 * but we have no choice until close() is repaired at VFS level.
 928	 */
 929	case TCP_LISTEN:
 930		if (!(how & RCV_SHUTDOWN))
 931			break;
 932		fallthrough;
 933	case TCP_SYN_SENT:
 934		err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
 935		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 936		break;
 937	}
 938
 939	/* Wake up anyone sleeping in poll. */
 940	sk->sk_state_change(sk);
 941	release_sock(sk);
 942	return err;
 943}
 944EXPORT_SYMBOL(inet_shutdown);
 945
 946/*
 947 *	ioctl() calls you can issue on an INET socket. Most of these are
 948 *	device configuration and stuff and very rarely used. Some ioctls
 949 *	pass on to the socket itself.
 950 *
 951 *	NOTE: I like the idea of a module for the config stuff. ie ifconfig
 952 *	loads the devconfigure module does its configuring and unloads it.
 953 *	There's a good 20K of config code hanging around the kernel.
 954 */
 955
 956int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 957{
 958	struct sock *sk = sock->sk;
 959	int err = 0;
 960	struct net *net = sock_net(sk);
 961	void __user *p = (void __user *)arg;
 962	struct ifreq ifr;
 963	struct rtentry rt;
 964
 965	switch (cmd) {
 966	case SIOCADDRT:
 967	case SIOCDELRT:
 968		if (copy_from_user(&rt, p, sizeof(struct rtentry)))
 969			return -EFAULT;
 970		err = ip_rt_ioctl(net, cmd, &rt);
 971		break;
 972	case SIOCRTMSG:
 973		err = -EINVAL;
 974		break;
 975	case SIOCDARP:
 976	case SIOCGARP:
 977	case SIOCSARP:
 978		err = arp_ioctl(net, cmd, (void __user *)arg);
 979		break;
 980	case SIOCGIFADDR:
 981	case SIOCGIFBRDADDR:
 982	case SIOCGIFNETMASK:
 983	case SIOCGIFDSTADDR:
 984	case SIOCGIFPFLAGS:
 985		if (get_user_ifreq(&ifr, NULL, p))
 986			return -EFAULT;
 987		err = devinet_ioctl(net, cmd, &ifr);
 988		if (!err && put_user_ifreq(&ifr, p))
 989			err = -EFAULT;
 990		break;
 991
 992	case SIOCSIFADDR:
 993	case SIOCSIFBRDADDR:
 994	case SIOCSIFNETMASK:
 995	case SIOCSIFDSTADDR:
 996	case SIOCSIFPFLAGS:
 997	case SIOCSIFFLAGS:
 998		if (get_user_ifreq(&ifr, NULL, p))
 999			return -EFAULT;
1000		err = devinet_ioctl(net, cmd, &ifr);
1001		break;
1002	default:
1003		if (sk->sk_prot->ioctl)
1004			err = sk_ioctl(sk, cmd, (void __user *)arg);
1005		else
1006			err = -ENOIOCTLCMD;
1007		break;
1008	}
1009	return err;
1010}
1011EXPORT_SYMBOL(inet_ioctl);
1012
1013#ifdef CONFIG_COMPAT
1014static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd,
1015		struct compat_rtentry __user *ur)
1016{
1017	compat_uptr_t rtdev;
1018	struct rtentry rt;
1019
1020	if (copy_from_user(&rt.rt_dst, &ur->rt_dst,
1021			3 * sizeof(struct sockaddr)) ||
1022	    get_user(rt.rt_flags, &ur->rt_flags) ||
1023	    get_user(rt.rt_metric, &ur->rt_metric) ||
1024	    get_user(rt.rt_mtu, &ur->rt_mtu) ||
1025	    get_user(rt.rt_window, &ur->rt_window) ||
1026	    get_user(rt.rt_irtt, &ur->rt_irtt) ||
1027	    get_user(rtdev, &ur->rt_dev))
1028		return -EFAULT;
1029
1030	rt.rt_dev = compat_ptr(rtdev);
1031	return ip_rt_ioctl(sock_net(sk), cmd, &rt);
1032}
1033
1034static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1035{
1036	void __user *argp = compat_ptr(arg);
1037	struct sock *sk = sock->sk;
1038
1039	switch (cmd) {
1040	case SIOCADDRT:
1041	case SIOCDELRT:
1042		return inet_compat_routing_ioctl(sk, cmd, argp);
1043	default:
1044		if (!sk->sk_prot->compat_ioctl)
1045			return -ENOIOCTLCMD;
1046		return sk->sk_prot->compat_ioctl(sk, cmd, arg);
1047	}
1048}
1049#endif /* CONFIG_COMPAT */
1050
1051const struct proto_ops inet_stream_ops = {
1052	.family		   = PF_INET,
1053	.owner		   = THIS_MODULE,
1054	.release	   = inet_release,
1055	.bind		   = inet_bind,
1056	.connect	   = inet_stream_connect,
1057	.socketpair	   = sock_no_socketpair,
1058	.accept		   = inet_accept,
1059	.getname	   = inet_getname,
1060	.poll		   = tcp_poll,
1061	.ioctl		   = inet_ioctl,
1062	.gettstamp	   = sock_gettstamp,
1063	.listen		   = inet_listen,
1064	.shutdown	   = inet_shutdown,
1065	.setsockopt	   = sock_common_setsockopt,
1066	.getsockopt	   = sock_common_getsockopt,
1067	.sendmsg	   = inet_sendmsg,
1068	.recvmsg	   = inet_recvmsg,
1069#ifdef CONFIG_MMU
1070	.mmap		   = tcp_mmap,
1071#endif
1072	.splice_eof	   = inet_splice_eof,
1073	.splice_read	   = tcp_splice_read,
1074	.read_sock	   = tcp_read_sock,
1075	.read_skb	   = tcp_read_skb,
1076	.sendmsg_locked    = tcp_sendmsg_locked,
 
1077	.peek_len	   = tcp_peek_len,
1078#ifdef CONFIG_COMPAT
1079	.compat_ioctl	   = inet_compat_ioctl,
1080#endif
1081	.set_rcvlowat	   = tcp_set_rcvlowat,
1082};
1083EXPORT_SYMBOL(inet_stream_ops);
1084
1085const struct proto_ops inet_dgram_ops = {
1086	.family		   = PF_INET,
1087	.owner		   = THIS_MODULE,
1088	.release	   = inet_release,
1089	.bind		   = inet_bind,
1090	.connect	   = inet_dgram_connect,
1091	.socketpair	   = sock_no_socketpair,
1092	.accept		   = sock_no_accept,
1093	.getname	   = inet_getname,
1094	.poll		   = udp_poll,
1095	.ioctl		   = inet_ioctl,
1096	.gettstamp	   = sock_gettstamp,
1097	.listen		   = sock_no_listen,
1098	.shutdown	   = inet_shutdown,
1099	.setsockopt	   = sock_common_setsockopt,
1100	.getsockopt	   = sock_common_getsockopt,
1101	.sendmsg	   = inet_sendmsg,
1102	.read_skb	   = udp_read_skb,
1103	.recvmsg	   = inet_recvmsg,
1104	.mmap		   = sock_no_mmap,
1105	.splice_eof	   = inet_splice_eof,
1106	.set_peek_off	   = sk_set_peek_off,
1107#ifdef CONFIG_COMPAT
1108	.compat_ioctl	   = inet_compat_ioctl,
1109#endif
1110};
1111EXPORT_SYMBOL(inet_dgram_ops);
1112
1113/*
1114 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1115 * udp_poll
1116 */
1117static const struct proto_ops inet_sockraw_ops = {
1118	.family		   = PF_INET,
1119	.owner		   = THIS_MODULE,
1120	.release	   = inet_release,
1121	.bind		   = inet_bind,
1122	.connect	   = inet_dgram_connect,
1123	.socketpair	   = sock_no_socketpair,
1124	.accept		   = sock_no_accept,
1125	.getname	   = inet_getname,
1126	.poll		   = datagram_poll,
1127	.ioctl		   = inet_ioctl,
1128	.gettstamp	   = sock_gettstamp,
1129	.listen		   = sock_no_listen,
1130	.shutdown	   = inet_shutdown,
1131	.setsockopt	   = sock_common_setsockopt,
1132	.getsockopt	   = sock_common_getsockopt,
1133	.sendmsg	   = inet_sendmsg,
1134	.recvmsg	   = inet_recvmsg,
1135	.mmap		   = sock_no_mmap,
1136	.splice_eof	   = inet_splice_eof,
1137#ifdef CONFIG_COMPAT
1138	.compat_ioctl	   = inet_compat_ioctl,
1139#endif
1140};
1141
1142static const struct net_proto_family inet_family_ops = {
1143	.family = PF_INET,
1144	.create = inet_create,
1145	.owner	= THIS_MODULE,
1146};
1147
1148/* Upon startup we insert all the elements in inetsw_array[] into
1149 * the linked list inetsw.
1150 */
1151static struct inet_protosw inetsw_array[] =
1152{
1153	{
1154		.type =       SOCK_STREAM,
1155		.protocol =   IPPROTO_TCP,
1156		.prot =       &tcp_prot,
1157		.ops =        &inet_stream_ops,
1158		.flags =      INET_PROTOSW_PERMANENT |
1159			      INET_PROTOSW_ICSK,
1160	},
1161
1162	{
1163		.type =       SOCK_DGRAM,
1164		.protocol =   IPPROTO_UDP,
1165		.prot =       &udp_prot,
1166		.ops =        &inet_dgram_ops,
1167		.flags =      INET_PROTOSW_PERMANENT,
1168       },
1169
1170       {
1171		.type =       SOCK_DGRAM,
1172		.protocol =   IPPROTO_ICMP,
1173		.prot =       &ping_prot,
1174		.ops =        &inet_sockraw_ops,
1175		.flags =      INET_PROTOSW_REUSE,
1176       },
1177
1178       {
1179	       .type =       SOCK_RAW,
1180	       .protocol =   IPPROTO_IP,	/* wild card */
1181	       .prot =       &raw_prot,
1182	       .ops =        &inet_sockraw_ops,
1183	       .flags =      INET_PROTOSW_REUSE,
1184       }
1185};
1186
1187#define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1188
1189void inet_register_protosw(struct inet_protosw *p)
1190{
1191	struct list_head *lh;
1192	struct inet_protosw *answer;
1193	int protocol = p->protocol;
1194	struct list_head *last_perm;
1195
1196	spin_lock_bh(&inetsw_lock);
1197
1198	if (p->type >= SOCK_MAX)
1199		goto out_illegal;
1200
1201	/* If we are trying to override a permanent protocol, bail. */
1202	last_perm = &inetsw[p->type];
1203	list_for_each(lh, &inetsw[p->type]) {
1204		answer = list_entry(lh, struct inet_protosw, list);
1205		/* Check only the non-wild match. */
1206		if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1207			break;
1208		if (protocol == answer->protocol)
1209			goto out_permanent;
1210		last_perm = lh;
1211	}
1212
1213	/* Add the new entry after the last permanent entry if any, so that
1214	 * the new entry does not override a permanent entry when matched with
1215	 * a wild-card protocol. But it is allowed to override any existing
1216	 * non-permanent entry.  This means that when we remove this entry, the
1217	 * system automatically returns to the old behavior.
1218	 */
1219	list_add_rcu(&p->list, last_perm);
1220out:
1221	spin_unlock_bh(&inetsw_lock);
1222
1223	return;
1224
1225out_permanent:
1226	pr_err("Attempt to override permanent protocol %d\n", protocol);
1227	goto out;
1228
1229out_illegal:
1230	pr_err("Ignoring attempt to register invalid socket type %d\n",
1231	       p->type);
1232	goto out;
1233}
1234EXPORT_SYMBOL(inet_register_protosw);
1235
1236void inet_unregister_protosw(struct inet_protosw *p)
1237{
1238	if (INET_PROTOSW_PERMANENT & p->flags) {
1239		pr_err("Attempt to unregister permanent protocol %d\n",
1240		       p->protocol);
1241	} else {
1242		spin_lock_bh(&inetsw_lock);
1243		list_del_rcu(&p->list);
1244		spin_unlock_bh(&inetsw_lock);
1245
1246		synchronize_net();
1247	}
1248}
1249EXPORT_SYMBOL(inet_unregister_protosw);
1250
1251static int inet_sk_reselect_saddr(struct sock *sk)
1252{
1253	struct inet_sock *inet = inet_sk(sk);
1254	__be32 old_saddr = inet->inet_saddr;
1255	__be32 daddr = inet->inet_daddr;
1256	struct flowi4 *fl4;
1257	struct rtable *rt;
1258	__be32 new_saddr;
1259	struct ip_options_rcu *inet_opt;
1260	int err;
1261
1262	inet_opt = rcu_dereference_protected(inet->inet_opt,
1263					     lockdep_sock_is_held(sk));
1264	if (inet_opt && inet_opt->opt.srr)
1265		daddr = inet_opt->opt.faddr;
1266
1267	/* Query new route. */
1268	fl4 = &inet->cork.fl.u.ip4;
1269	rt = ip_route_connect(fl4, daddr, 0, sk->sk_bound_dev_if,
1270			      sk->sk_protocol, inet->inet_sport,
1271			      inet->inet_dport, sk);
1272	if (IS_ERR(rt))
1273		return PTR_ERR(rt);
1274
 
 
1275	new_saddr = fl4->saddr;
1276
1277	if (new_saddr == old_saddr) {
1278		sk_setup_caps(sk, &rt->dst);
1279		return 0;
1280	}
1281
1282	err = inet_bhash2_update_saddr(sk, &new_saddr, AF_INET);
1283	if (err) {
1284		ip_rt_put(rt);
1285		return err;
1286	}
1287
1288	sk_setup_caps(sk, &rt->dst);
1289
1290	if (READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) > 1) {
1291		pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1292			__func__, &old_saddr, &new_saddr);
1293	}
1294
 
 
1295	/*
1296	 * XXX The only one ugly spot where we need to
1297	 * XXX really change the sockets identity after
1298	 * XXX it has entered the hashes. -DaveM
1299	 *
1300	 * Besides that, it does not check for connection
1301	 * uniqueness. Wait for troubles.
1302	 */
1303	return __sk_prot_rehash(sk);
1304}
1305
1306int inet_sk_rebuild_header(struct sock *sk)
1307{
1308	struct inet_sock *inet = inet_sk(sk);
1309	struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1310	__be32 daddr;
1311	struct ip_options_rcu *inet_opt;
1312	struct flowi4 *fl4;
1313	int err;
1314
1315	/* Route is OK, nothing to do. */
1316	if (rt)
1317		return 0;
1318
1319	/* Reroute. */
1320	rcu_read_lock();
1321	inet_opt = rcu_dereference(inet->inet_opt);
1322	daddr = inet->inet_daddr;
1323	if (inet_opt && inet_opt->opt.srr)
1324		daddr = inet_opt->opt.faddr;
1325	rcu_read_unlock();
1326	fl4 = &inet->cork.fl.u.ip4;
1327	rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1328				   inet->inet_dport, inet->inet_sport,
1329				   sk->sk_protocol, RT_CONN_FLAGS(sk),
1330				   sk->sk_bound_dev_if);
1331	if (!IS_ERR(rt)) {
1332		err = 0;
1333		sk_setup_caps(sk, &rt->dst);
1334	} else {
1335		err = PTR_ERR(rt);
1336
1337		/* Routing failed... */
1338		sk->sk_route_caps = 0;
1339		/*
1340		 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1341		 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1342		 */
1343		if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) ||
1344		    sk->sk_state != TCP_SYN_SENT ||
1345		    (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1346		    (err = inet_sk_reselect_saddr(sk)) != 0)
1347			WRITE_ONCE(sk->sk_err_soft, -err);
1348	}
1349
1350	return err;
1351}
1352EXPORT_SYMBOL(inet_sk_rebuild_header);
1353
1354void inet_sk_set_state(struct sock *sk, int state)
1355{
1356	trace_inet_sock_set_state(sk, sk->sk_state, state);
1357	sk->sk_state = state;
1358}
1359EXPORT_SYMBOL(inet_sk_set_state);
1360
1361void inet_sk_state_store(struct sock *sk, int newstate)
1362{
1363	trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1364	smp_store_release(&sk->sk_state, newstate);
1365}
1366
1367struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1368				 netdev_features_t features)
1369{
1370	bool udpfrag = false, fixedid = false, gso_partial, encap;
1371	struct sk_buff *segs = ERR_PTR(-EINVAL);
1372	const struct net_offload *ops;
1373	unsigned int offset = 0;
1374	struct iphdr *iph;
1375	int proto, tot_len;
1376	int nhoff;
1377	int ihl;
1378	int id;
1379
1380	skb_reset_network_header(skb);
1381	nhoff = skb_network_header(skb) - skb_mac_header(skb);
1382	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1383		goto out;
1384
1385	iph = ip_hdr(skb);
1386	ihl = iph->ihl * 4;
1387	if (ihl < sizeof(*iph))
1388		goto out;
1389
1390	id = ntohs(iph->id);
1391	proto = iph->protocol;
1392
1393	/* Warning: after this point, iph might be no longer valid */
1394	if (unlikely(!pskb_may_pull(skb, ihl)))
1395		goto out;
1396	__skb_pull(skb, ihl);
1397
1398	encap = SKB_GSO_CB(skb)->encap_level > 0;
1399	if (encap)
1400		features &= skb->dev->hw_enc_features;
1401	SKB_GSO_CB(skb)->encap_level += ihl;
1402
1403	skb_reset_transport_header(skb);
1404
1405	segs = ERR_PTR(-EPROTONOSUPPORT);
1406
1407	if (!skb->encapsulation || encap) {
1408		udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1409		fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1410
1411		/* fixed ID is invalid if DF bit is not set */
1412		if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1413			goto out;
1414	}
1415
1416	ops = rcu_dereference(inet_offloads[proto]);
1417	if (likely(ops && ops->callbacks.gso_segment)) {
1418		segs = ops->callbacks.gso_segment(skb, features);
1419		if (!segs)
1420			skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
1421	}
1422
1423	if (IS_ERR_OR_NULL(segs))
1424		goto out;
1425
1426	gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1427
1428	skb = segs;
1429	do {
1430		iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1431		if (udpfrag) {
1432			iph->frag_off = htons(offset >> 3);
1433			if (skb->next)
1434				iph->frag_off |= htons(IP_MF);
1435			offset += skb->len - nhoff - ihl;
1436			tot_len = skb->len - nhoff;
1437		} else if (skb_is_gso(skb)) {
1438			if (!fixedid) {
1439				iph->id = htons(id);
1440				id += skb_shinfo(skb)->gso_segs;
1441			}
1442
1443			if (gso_partial)
1444				tot_len = skb_shinfo(skb)->gso_size +
1445					  SKB_GSO_CB(skb)->data_offset +
1446					  skb->head - (unsigned char *)iph;
1447			else
1448				tot_len = skb->len - nhoff;
1449		} else {
1450			if (!fixedid)
1451				iph->id = htons(id++);
1452			tot_len = skb->len - nhoff;
1453		}
1454		iph->tot_len = htons(tot_len);
1455		ip_send_check(iph);
1456		if (encap)
1457			skb_reset_inner_headers(skb);
1458		skb->network_header = (u8 *)iph - skb->head;
1459		skb_reset_mac_len(skb);
1460	} while ((skb = skb->next));
1461
1462out:
1463	return segs;
1464}
 
1465
1466static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1467					netdev_features_t features)
1468{
1469	if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1470		return ERR_PTR(-EINVAL);
1471
1472	return inet_gso_segment(skb, features);
1473}
1474
1475struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1476{
1477	const struct net_offload *ops;
1478	struct sk_buff *pp = NULL;
1479	const struct iphdr *iph;
1480	struct sk_buff *p;
1481	unsigned int hlen;
1482	unsigned int off;
1483	unsigned int id;
1484	int flush = 1;
1485	int proto;
1486
1487	off = skb_gro_offset(skb);
1488	hlen = off + sizeof(*iph);
1489	iph = skb_gro_header(skb, hlen, off);
1490	if (unlikely(!iph))
1491		goto out;
 
 
 
1492
1493	proto = iph->protocol;
1494
 
1495	ops = rcu_dereference(inet_offloads[proto]);
1496	if (!ops || !ops->callbacks.gro_receive)
1497		goto out;
1498
1499	if (*(u8 *)iph != 0x45)
1500		goto out;
1501
1502	if (ip_is_fragment(iph))
1503		goto out;
1504
1505	if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1506		goto out;
1507
1508	NAPI_GRO_CB(skb)->proto = proto;
1509	id = ntohl(*(__be32 *)&iph->id);
1510	flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1511	id >>= 16;
1512
1513	list_for_each_entry(p, head, list) {
1514		struct iphdr *iph2;
1515		u16 flush_id;
1516
1517		if (!NAPI_GRO_CB(p)->same_flow)
1518			continue;
1519
1520		iph2 = (struct iphdr *)(p->data + off);
1521		/* The above works because, with the exception of the top
1522		 * (inner most) layer, we only aggregate pkts with the same
1523		 * hdr length so all the hdrs we'll need to verify will start
1524		 * at the same offset.
1525		 */
1526		if ((iph->protocol ^ iph2->protocol) |
1527		    ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1528		    ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1529			NAPI_GRO_CB(p)->same_flow = 0;
1530			continue;
1531		}
1532
1533		/* All fields must match except length and checksum. */
1534		NAPI_GRO_CB(p)->flush |=
1535			(iph->ttl ^ iph2->ttl) |
1536			(iph->tos ^ iph2->tos) |
1537			((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1538
1539		NAPI_GRO_CB(p)->flush |= flush;
1540
1541		/* We need to store of the IP ID check to be included later
1542		 * when we can verify that this packet does in fact belong
1543		 * to a given flow.
1544		 */
1545		flush_id = (u16)(id - ntohs(iph2->id));
1546
1547		/* This bit of code makes it much easier for us to identify
1548		 * the cases where we are doing atomic vs non-atomic IP ID
1549		 * checks.  Specifically an atomic check can return IP ID
1550		 * values 0 - 0xFFFF, while a non-atomic check can only
1551		 * return 0 or 0xFFFF.
1552		 */
1553		if (!NAPI_GRO_CB(p)->is_atomic ||
1554		    !(iph->frag_off & htons(IP_DF))) {
1555			flush_id ^= NAPI_GRO_CB(p)->count;
1556			flush_id = flush_id ? 0xFFFF : 0;
1557		}
1558
1559		/* If the previous IP ID value was based on an atomic
1560		 * datagram we can overwrite the value and ignore it.
1561		 */
1562		if (NAPI_GRO_CB(skb)->is_atomic)
1563			NAPI_GRO_CB(p)->flush_id = flush_id;
1564		else
1565			NAPI_GRO_CB(p)->flush_id |= flush_id;
1566	}
1567
1568	NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1569	NAPI_GRO_CB(skb)->flush |= flush;
1570	skb_set_network_header(skb, off);
1571	/* The above will be needed by the transport layer if there is one
1572	 * immediately following this IP hdr.
1573	 */
1574
1575	/* Note : No need to call skb_gro_postpull_rcsum() here,
1576	 * as we already checked checksum over ipv4 header was 0
1577	 */
1578	skb_gro_pull(skb, sizeof(*iph));
1579	skb_set_transport_header(skb, skb_gro_offset(skb));
1580
1581	pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1582				       ops->callbacks.gro_receive, head, skb);
1583
 
 
 
1584out:
1585	skb_gro_flush_final(skb, pp, flush);
1586
1587	return pp;
1588}
 
1589
1590static struct sk_buff *ipip_gro_receive(struct list_head *head,
1591					struct sk_buff *skb)
1592{
1593	if (NAPI_GRO_CB(skb)->encap_mark) {
1594		NAPI_GRO_CB(skb)->flush = 1;
1595		return NULL;
1596	}
1597
1598	NAPI_GRO_CB(skb)->encap_mark = 1;
1599
1600	return inet_gro_receive(head, skb);
1601}
1602
1603#define SECONDS_PER_DAY	86400
1604
1605/* inet_current_timestamp - Return IP network timestamp
1606 *
1607 * Return milliseconds since midnight in network byte order.
1608 */
1609__be32 inet_current_timestamp(void)
1610{
1611	u32 secs;
1612	u32 msecs;
1613	struct timespec64 ts;
1614
1615	ktime_get_real_ts64(&ts);
1616
1617	/* Get secs since midnight. */
1618	(void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1619	/* Convert to msecs. */
1620	msecs = secs * MSEC_PER_SEC;
1621	/* Convert nsec to msec. */
1622	msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1623
1624	/* Convert to network byte order. */
1625	return htonl(msecs);
1626}
1627EXPORT_SYMBOL(inet_current_timestamp);
1628
1629int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1630{
1631	unsigned int family = READ_ONCE(sk->sk_family);
1632
1633	if (family == AF_INET)
1634		return ip_recv_error(sk, msg, len, addr_len);
1635#if IS_ENABLED(CONFIG_IPV6)
1636	if (family == AF_INET6)
1637		return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1638#endif
1639	return -EINVAL;
1640}
1641EXPORT_SYMBOL(inet_recv_error);
1642
1643int inet_gro_complete(struct sk_buff *skb, int nhoff)
1644{
 
1645	struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1646	const struct net_offload *ops;
1647	__be16 totlen = iph->tot_len;
1648	int proto = iph->protocol;
1649	int err = -ENOSYS;
1650
1651	if (skb->encapsulation) {
1652		skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1653		skb_set_inner_network_header(skb, nhoff);
1654	}
1655
1656	iph_set_totlen(iph, skb->len - nhoff);
1657	csum_replace2(&iph->check, totlen, iph->tot_len);
1658
 
1659	ops = rcu_dereference(inet_offloads[proto]);
1660	if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1661		goto out;
1662
1663	/* Only need to add sizeof(*iph) to get to the next hdr below
1664	 * because any hdr with option will have been flushed in
1665	 * inet_gro_receive().
1666	 */
1667	err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1668			      tcp4_gro_complete, udp4_gro_complete,
1669			      skb, nhoff + sizeof(*iph));
1670
1671out:
 
 
1672	return err;
1673}
 
1674
1675static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1676{
1677	skb->encapsulation = 1;
1678	skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1679	return inet_gro_complete(skb, nhoff);
1680}
1681
1682int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1683			 unsigned short type, unsigned char protocol,
1684			 struct net *net)
1685{
1686	struct socket *sock;
1687	int rc = sock_create_kern(net, family, type, protocol, &sock);
1688
1689	if (rc == 0) {
1690		*sk = sock->sk;
1691		(*sk)->sk_allocation = GFP_ATOMIC;
1692		(*sk)->sk_use_task_frag = false;
1693		/*
1694		 * Unhash it so that IP input processing does not even see it,
1695		 * we do not wish this socket to see incoming packets.
1696		 */
1697		(*sk)->sk_prot->unhash(*sk);
1698	}
1699	return rc;
1700}
1701EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1702
 
 
 
 
 
 
1703unsigned long snmp_fold_field(void __percpu *mib, int offt)
1704{
1705	unsigned long res = 0;
1706	int i;
1707
1708	for_each_possible_cpu(i)
1709		res += snmp_get_cpu_field(mib, i, offt);
1710	return res;
1711}
1712EXPORT_SYMBOL_GPL(snmp_fold_field);
1713
1714#if BITS_PER_LONG==32
1715
1716u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1717			 size_t syncp_offset)
1718{
1719	void *bhptr;
1720	struct u64_stats_sync *syncp;
1721	u64 v;
1722	unsigned int start;
1723
1724	bhptr = per_cpu_ptr(mib, cpu);
1725	syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1726	do {
1727		start = u64_stats_fetch_begin(syncp);
1728		v = *(((u64 *)bhptr) + offt);
1729	} while (u64_stats_fetch_retry(syncp, start));
1730
1731	return v;
1732}
1733EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1734
1735u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1736{
1737	u64 res = 0;
1738	int cpu;
1739
1740	for_each_possible_cpu(cpu) {
1741		res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1742	}
1743	return res;
1744}
1745EXPORT_SYMBOL_GPL(snmp_fold_field64);
1746#endif
1747
1748#ifdef CONFIG_IP_MULTICAST
1749static const struct net_protocol igmp_protocol = {
1750	.handler =	igmp_rcv,
 
1751};
1752#endif
1753
1754static const struct net_protocol tcp_protocol = {
 
 
 
 
 
1755	.handler	=	tcp_v4_rcv,
1756	.err_handler	=	tcp_v4_err,
1757	.no_policy	=	1,
 
1758	.icmp_strict_tag_validation = 1,
1759};
1760
1761static const struct net_protocol udp_protocol = {
 
 
 
 
 
1762	.handler =	udp_rcv,
1763	.err_handler =	udp_err,
1764	.no_policy =	1,
 
1765};
1766
1767static const struct net_protocol icmp_protocol = {
1768	.handler =	icmp_rcv,
1769	.err_handler =	icmp_err,
1770	.no_policy =	1,
 
1771};
1772
1773static __net_init int ipv4_mib_init_net(struct net *net)
1774{
1775	int i;
1776
1777	net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1778	if (!net->mib.tcp_statistics)
1779		goto err_tcp_mib;
1780	net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1781	if (!net->mib.ip_statistics)
1782		goto err_ip_mib;
1783
1784	for_each_possible_cpu(i) {
1785		struct ipstats_mib *af_inet_stats;
1786		af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1787		u64_stats_init(&af_inet_stats->syncp);
1788	}
1789
1790	net->mib.net_statistics = alloc_percpu(struct linux_mib);
1791	if (!net->mib.net_statistics)
1792		goto err_net_mib;
1793	net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1794	if (!net->mib.udp_statistics)
1795		goto err_udp_mib;
1796	net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1797	if (!net->mib.udplite_statistics)
1798		goto err_udplite_mib;
1799	net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1800	if (!net->mib.icmp_statistics)
1801		goto err_icmp_mib;
1802	net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1803					      GFP_KERNEL);
1804	if (!net->mib.icmpmsg_statistics)
1805		goto err_icmpmsg_mib;
1806
1807	tcp_mib_init(net);
1808	return 0;
1809
1810err_icmpmsg_mib:
1811	free_percpu(net->mib.icmp_statistics);
1812err_icmp_mib:
1813	free_percpu(net->mib.udplite_statistics);
1814err_udplite_mib:
1815	free_percpu(net->mib.udp_statistics);
1816err_udp_mib:
1817	free_percpu(net->mib.net_statistics);
1818err_net_mib:
1819	free_percpu(net->mib.ip_statistics);
1820err_ip_mib:
1821	free_percpu(net->mib.tcp_statistics);
1822err_tcp_mib:
1823	return -ENOMEM;
1824}
1825
1826static __net_exit void ipv4_mib_exit_net(struct net *net)
1827{
1828	kfree(net->mib.icmpmsg_statistics);
1829	free_percpu(net->mib.icmp_statistics);
1830	free_percpu(net->mib.udplite_statistics);
1831	free_percpu(net->mib.udp_statistics);
1832	free_percpu(net->mib.net_statistics);
1833	free_percpu(net->mib.ip_statistics);
1834	free_percpu(net->mib.tcp_statistics);
1835#ifdef CONFIG_MPTCP
1836	/* allocated on demand, see mptcp_init_sock() */
1837	free_percpu(net->mib.mptcp_statistics);
1838#endif
1839}
1840
1841static __net_initdata struct pernet_operations ipv4_mib_ops = {
1842	.init = ipv4_mib_init_net,
1843	.exit = ipv4_mib_exit_net,
1844};
1845
1846static int __init init_ipv4_mibs(void)
1847{
1848	return register_pernet_subsys(&ipv4_mib_ops);
1849}
1850
1851static __net_init int inet_init_net(struct net *net)
1852{
1853	/*
1854	 * Set defaults for local port range
1855	 */
1856	net->ipv4.ip_local_ports.range = 60999u << 16 | 32768u;
 
 
1857
1858	seqlock_init(&net->ipv4.ping_group_range.lock);
1859	/*
1860	 * Sane defaults - nobody may create ping sockets.
1861	 * Boot scripts should set this to distro-specific group.
1862	 */
1863	net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1864	net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1865
1866	/* Default values for sysctl-controlled parameters.
1867	 * We set them here, in case sysctl is not compiled.
1868	 */
1869	net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1870	net->ipv4.sysctl_ip_fwd_update_priority = 1;
1871	net->ipv4.sysctl_ip_dynaddr = 0;
1872	net->ipv4.sysctl_ip_early_demux = 1;
1873	net->ipv4.sysctl_udp_early_demux = 1;
1874	net->ipv4.sysctl_tcp_early_demux = 1;
1875	net->ipv4.sysctl_nexthop_compat_mode = 1;
1876#ifdef CONFIG_SYSCTL
1877	net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1878#endif
1879
1880	/* Some igmp sysctl, whose values are always used */
1881	net->ipv4.sysctl_igmp_max_memberships = 20;
1882	net->ipv4.sysctl_igmp_max_msf = 10;
1883	/* IGMP reports for link-local multicast groups are enabled by default */
1884	net->ipv4.sysctl_igmp_llm_reports = 1;
1885	net->ipv4.sysctl_igmp_qrv = 2;
1886
1887	net->ipv4.sysctl_fib_notify_on_flag_change = 0;
1888
1889	return 0;
1890}
1891
1892static __net_initdata struct pernet_operations af_inet_ops = {
1893	.init = inet_init_net,
1894};
1895
1896static int __init init_inet_pernet_ops(void)
1897{
1898	return register_pernet_subsys(&af_inet_ops);
1899}
1900
1901static int ipv4_proc_init(void);
1902
1903/*
1904 *	IP protocol layer initialiser
1905 */
1906
1907static struct packet_offload ip_packet_offload __read_mostly = {
1908	.type = cpu_to_be16(ETH_P_IP),
1909	.callbacks = {
1910		.gso_segment = inet_gso_segment,
1911		.gro_receive = inet_gro_receive,
1912		.gro_complete = inet_gro_complete,
1913	},
1914};
1915
1916static const struct net_offload ipip_offload = {
1917	.callbacks = {
1918		.gso_segment	= ipip_gso_segment,
1919		.gro_receive	= ipip_gro_receive,
1920		.gro_complete	= ipip_gro_complete,
1921	},
1922};
1923
1924static int __init ipip_offload_init(void)
1925{
1926	return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1927}
1928
1929static int __init ipv4_offload_init(void)
1930{
1931	/*
1932	 * Add offloads
1933	 */
1934	if (udpv4_offload_init() < 0)
1935		pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1936	if (tcpv4_offload_init() < 0)
1937		pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1938	if (ipip_offload_init() < 0)
1939		pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1940
1941	dev_add_offload(&ip_packet_offload);
1942	return 0;
1943}
1944
1945fs_initcall(ipv4_offload_init);
1946
1947static struct packet_type ip_packet_type __read_mostly = {
1948	.type = cpu_to_be16(ETH_P_IP),
1949	.func = ip_rcv,
1950	.list_func = ip_list_rcv,
1951};
1952
1953static int __init inet_init(void)
1954{
1955	struct inet_protosw *q;
1956	struct list_head *r;
1957	int rc;
1958
1959	sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1960
1961	raw_hashinfo_init(&raw_v4_hashinfo);
1962
1963	rc = proto_register(&tcp_prot, 1);
1964	if (rc)
1965		goto out;
1966
1967	rc = proto_register(&udp_prot, 1);
1968	if (rc)
1969		goto out_unregister_tcp_proto;
1970
1971	rc = proto_register(&raw_prot, 1);
1972	if (rc)
1973		goto out_unregister_udp_proto;
1974
1975	rc = proto_register(&ping_prot, 1);
1976	if (rc)
1977		goto out_unregister_raw_proto;
1978
1979	/*
1980	 *	Tell SOCKET that we are alive...
1981	 */
1982
1983	(void)sock_register(&inet_family_ops);
1984
1985#ifdef CONFIG_SYSCTL
1986	ip_static_sysctl_init();
1987#endif
1988
1989	/*
1990	 *	Add all the base protocols.
1991	 */
1992
1993	if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1994		pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1995	if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1996		pr_crit("%s: Cannot add UDP protocol\n", __func__);
1997	if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1998		pr_crit("%s: Cannot add TCP protocol\n", __func__);
1999#ifdef CONFIG_IP_MULTICAST
2000	if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
2001		pr_crit("%s: Cannot add IGMP protocol\n", __func__);
2002#endif
2003
2004	/* Register the socket-side information for inet_create. */
2005	for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
2006		INIT_LIST_HEAD(r);
2007
2008	for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
2009		inet_register_protosw(q);
2010
2011	/*
2012	 *	Set the ARP module up
2013	 */
2014
2015	arp_init();
2016
2017	/*
2018	 *	Set the IP module up
2019	 */
2020
2021	ip_init();
2022
2023	/* Initialise per-cpu ipv4 mibs */
2024	if (init_ipv4_mibs())
2025		panic("%s: Cannot init ipv4 mibs\n", __func__);
2026
2027	/* Setup TCP slab cache for open requests. */
2028	tcp_init();
2029
2030	/* Setup UDP memory threshold */
2031	udp_init();
2032
2033	/* Add UDP-Lite (RFC 3828) */
2034	udplite4_register();
2035
2036	raw_init();
2037
2038	ping_init();
2039
2040	/*
2041	 *	Set the ICMP layer up
2042	 */
2043
2044	if (icmp_init() < 0)
2045		panic("Failed to create the ICMP control socket.\n");
2046
2047	/*
2048	 *	Initialise the multicast router
2049	 */
2050#if defined(CONFIG_IP_MROUTE)
2051	if (ip_mr_init())
2052		pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2053#endif
2054
2055	if (init_inet_pernet_ops())
2056		pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
 
 
 
 
 
 
2057
2058	ipv4_proc_init();
2059
2060	ipfrag_init();
2061
2062	dev_add_pack(&ip_packet_type);
2063
2064	ip_tunnel_core_init();
2065
2066	rc = 0;
2067out:
2068	return rc;
2069out_unregister_raw_proto:
2070	proto_unregister(&raw_prot);
2071out_unregister_udp_proto:
2072	proto_unregister(&udp_prot);
2073out_unregister_tcp_proto:
2074	proto_unregister(&tcp_prot);
2075	goto out;
2076}
2077
2078fs_initcall(inet_init);
2079
2080/* ------------------------------------------------------------------------ */
2081
2082#ifdef CONFIG_PROC_FS
2083static int __init ipv4_proc_init(void)
2084{
2085	int rc = 0;
2086
2087	if (raw_proc_init())
2088		goto out_raw;
2089	if (tcp4_proc_init())
2090		goto out_tcp;
2091	if (udp4_proc_init())
2092		goto out_udp;
2093	if (ping_proc_init())
2094		goto out_ping;
2095	if (ip_misc_proc_init())
2096		goto out_misc;
2097out:
2098	return rc;
2099out_misc:
2100	ping_proc_exit();
2101out_ping:
2102	udp4_proc_exit();
2103out_udp:
2104	tcp4_proc_exit();
2105out_tcp:
2106	raw_proc_exit();
2107out_raw:
2108	rc = -ENOMEM;
2109	goto out;
2110}
2111
2112#else /* CONFIG_PROC_FS */
2113static int __init ipv4_proc_init(void)
2114{
2115	return 0;
2116}
2117#endif /* CONFIG_PROC_FS */
   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/tcp.h>
 103#include <net/udp.h>
 104#include <net/udplite.h>
 105#include <net/ping.h>
 106#include <linux/skbuff.h>
 107#include <net/sock.h>
 108#include <net/raw.h>
 109#include <net/icmp.h>
 110#include <net/inet_common.h>
 111#include <net/ip_tunnels.h>
 112#include <net/xfrm.h>
 113#include <net/net_namespace.h>
 114#include <net/secure_seq.h>
 115#ifdef CONFIG_IP_MROUTE
 116#include <linux/mroute.h>
 117#endif
 118#include <net/l3mdev.h>
 119#include <net/compat.h>
 120
 121#include <trace/events/sock.h>
 122
 123/* The inetsw table contains everything that inet_create needs to
 124 * build a new socket.
 125 */
 126static struct list_head inetsw[SOCK_MAX];
 127static DEFINE_SPINLOCK(inetsw_lock);
 128
 129/* New destruction routine */
 130
 131void inet_sock_destruct(struct sock *sk)
 132{
 133	struct inet_sock *inet = inet_sk(sk);
 134
 135	__skb_queue_purge(&sk->sk_receive_queue);
 136	if (sk->sk_rx_skb_cache) {
 137		__kfree_skb(sk->sk_rx_skb_cache);
 138		sk->sk_rx_skb_cache = NULL;
 139	}
 140	__skb_queue_purge(&sk->sk_error_queue);
 141
 142	sk_mem_reclaim(sk);
 143
 144	if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
 145		pr_err("Attempt to release TCP socket in state %d %p\n",
 146		       sk->sk_state, sk);
 147		return;
 148	}
 149	if (!sock_flag(sk, SOCK_DEAD)) {
 150		pr_err("Attempt to release alive inet socket %p\n", sk);
 151		return;
 152	}
 153
 154	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
 155	WARN_ON(refcount_read(&sk->sk_wmem_alloc));
 156	WARN_ON(sk->sk_wmem_queued);
 157	WARN_ON(sk->sk_forward_alloc);
 158
 159	kfree(rcu_dereference_protected(inet->inet_opt, 1));
 160	dst_release(rcu_dereference_protected(sk->sk_dst_cache, 1));
 161	dst_release(sk->sk_rx_dst);
 162	sk_refcnt_debug_dec(sk);
 163}
 164EXPORT_SYMBOL(inet_sock_destruct);
 165
 166/*
 167 *	The routines beyond this point handle the behaviour of an AF_INET
 168 *	socket object. Mostly it punts to the subprotocols of IP to do
 169 *	the work.
 170 */
 171
 172/*
 173 *	Automatically bind an unbound socket.
 174 */
 175
 176static int inet_autobind(struct sock *sk)
 177{
 178	struct inet_sock *inet;
 179	/* We may need to bind the socket. */
 180	lock_sock(sk);
 181	inet = inet_sk(sk);
 182	if (!inet->inet_num) {
 183		if (sk->sk_prot->get_port(sk, 0)) {
 184			release_sock(sk);
 185			return -EAGAIN;
 186		}
 187		inet->inet_sport = htons(inet->inet_num);
 188	}
 189	release_sock(sk);
 190	return 0;
 191}
 192
 193/*
 194 *	Move a socket into listening state.
 195 */
 196int inet_listen(struct socket *sock, int backlog)
 197{
 198	struct sock *sk = sock->sk;
 199	unsigned char old_state;
 200	int err, tcp_fastopen;
 201
 202	lock_sock(sk);
 203
 204	err = -EINVAL;
 205	if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
 206		goto out;
 207
 208	old_state = sk->sk_state;
 209	if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
 210		goto out;
 211
 212	WRITE_ONCE(sk->sk_max_ack_backlog, backlog);
 213	/* Really, if the socket is already in listen state
 214	 * we can only allow the backlog to be adjusted.
 215	 */
 216	if (old_state != TCP_LISTEN) {
 217		/* Enable TFO w/o requiring TCP_FASTOPEN socket option.
 218		 * Note that only TCP sockets (SOCK_STREAM) will reach here.
 219		 * Also fastopen backlog may already been set via the option
 220		 * because the socket was in TCP_LISTEN state previously but
 221		 * was shutdown() rather than close().
 222		 */
 223		tcp_fastopen = sock_net(sk)->ipv4.sysctl_tcp_fastopen;
 224		if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
 225		    (tcp_fastopen & TFO_SERVER_ENABLE) &&
 226		    !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
 227			fastopen_queue_tune(sk, backlog);
 228			tcp_fastopen_init_key_once(sock_net(sk));
 229		}
 230
 231		err = inet_csk_listen_start(sk, backlog);
 232		if (err)
 233			goto out;
 
 234		tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL);
 235	}
 236	err = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 237
 238out:
 239	release_sock(sk);
 240	return err;
 241}
 242EXPORT_SYMBOL(inet_listen);
 243
 244/*
 245 *	Create an inet socket.
 246 */
 247
 248static int inet_create(struct net *net, struct socket *sock, int protocol,
 249		       int kern)
 250{
 251	struct sock *sk;
 252	struct inet_protosw *answer;
 253	struct inet_sock *inet;
 254	struct proto *answer_prot;
 255	unsigned char answer_flags;
 256	int try_loading_module = 0;
 257	int err;
 258
 259	if (protocol < 0 || protocol >= IPPROTO_MAX)
 260		return -EINVAL;
 261
 262	sock->state = SS_UNCONNECTED;
 263
 264	/* Look for the requested type/protocol pair. */
 265lookup_protocol:
 266	err = -ESOCKTNOSUPPORT;
 267	rcu_read_lock();
 268	list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
 269
 270		err = 0;
 271		/* Check the non-wild match. */
 272		if (protocol == answer->protocol) {
 273			if (protocol != IPPROTO_IP)
 274				break;
 275		} else {
 276			/* Check for the two wild cases. */
 277			if (IPPROTO_IP == protocol) {
 278				protocol = answer->protocol;
 279				break;
 280			}
 281			if (IPPROTO_IP == answer->protocol)
 282				break;
 283		}
 284		err = -EPROTONOSUPPORT;
 285	}
 286
 287	if (unlikely(err)) {
 288		if (try_loading_module < 2) {
 289			rcu_read_unlock();
 290			/*
 291			 * Be more specific, e.g. net-pf-2-proto-132-type-1
 292			 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
 293			 */
 294			if (++try_loading_module == 1)
 295				request_module("net-pf-%d-proto-%d-type-%d",
 296					       PF_INET, protocol, sock->type);
 297			/*
 298			 * Fall back to generic, e.g. net-pf-2-proto-132
 299			 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
 300			 */
 301			else
 302				request_module("net-pf-%d-proto-%d",
 303					       PF_INET, protocol);
 304			goto lookup_protocol;
 305		} else
 306			goto out_rcu_unlock;
 307	}
 308
 309	err = -EPERM;
 310	if (sock->type == SOCK_RAW && !kern &&
 311	    !ns_capable(net->user_ns, CAP_NET_RAW))
 312		goto out_rcu_unlock;
 313
 314	sock->ops = answer->ops;
 315	answer_prot = answer->prot;
 316	answer_flags = answer->flags;
 317	rcu_read_unlock();
 318
 319	WARN_ON(!answer_prot->slab);
 320
 321	err = -ENOBUFS;
 322	sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
 323	if (!sk)
 324		goto out;
 325
 326	err = 0;
 327	if (INET_PROTOSW_REUSE & answer_flags)
 328		sk->sk_reuse = SK_CAN_REUSE;
 329
 
 
 
 330	inet = inet_sk(sk);
 331	inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
 332
 333	inet->nodefrag = 0;
 334
 335	if (SOCK_RAW == sock->type) {
 336		inet->inet_num = protocol;
 337		if (IPPROTO_RAW == protocol)
 338			inet->hdrincl = 1;
 339	}
 340
 341	if (net->ipv4.sysctl_ip_no_pmtu_disc)
 342		inet->pmtudisc = IP_PMTUDISC_DONT;
 343	else
 344		inet->pmtudisc = IP_PMTUDISC_WANT;
 345
 346	inet->inet_id = 0;
 347
 348	sock_init_data(sock, sk);
 349
 350	sk->sk_destruct	   = inet_sock_destruct;
 351	sk->sk_protocol	   = protocol;
 352	sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
 
 353
 354	inet->uc_ttl	= -1;
 355	inet->mc_loop	= 1;
 356	inet->mc_ttl	= 1;
 357	inet->mc_all	= 1;
 358	inet->mc_index	= 0;
 359	inet->mc_list	= NULL;
 360	inet->rcv_tos	= 0;
 361
 362	sk_refcnt_debug_inc(sk);
 363
 364	if (inet->inet_num) {
 365		/* It assumes that any protocol which allows
 366		 * the user to assign a number at socket
 367		 * creation time automatically
 368		 * shares.
 369		 */
 370		inet->inet_sport = htons(inet->inet_num);
 371		/* Add to protocol hash chains. */
 372		err = sk->sk_prot->hash(sk);
 373		if (err) {
 374			sk_common_release(sk);
 375			goto out;
 376		}
 377	}
 378
 379	if (sk->sk_prot->init) {
 380		err = sk->sk_prot->init(sk);
 381		if (err) {
 382			sk_common_release(sk);
 383			goto out;
 384		}
 385	}
 386
 387	if (!kern) {
 388		err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
 389		if (err) {
 390			sk_common_release(sk);
 391			goto out;
 392		}
 393	}
 394out:
 395	return err;
 396out_rcu_unlock:
 397	rcu_read_unlock();
 398	goto out;
 399}
 400
 401
 402/*
 403 *	The peer socket should always be NULL (or else). When we call this
 404 *	function we are destroying the object and from then on nobody
 405 *	should refer to it.
 406 */
 407int inet_release(struct socket *sock)
 408{
 409	struct sock *sk = sock->sk;
 410
 411	if (sk) {
 412		long timeout;
 413
 414		if (!sk->sk_kern_sock)
 415			BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk);
 416
 417		/* Applications forget to leave groups before exiting */
 418		ip_mc_drop_socket(sk);
 419
 420		/* If linger is set, we don't return until the close
 421		 * is complete.  Otherwise we return immediately. The
 422		 * actually closing is done the same either way.
 423		 *
 424		 * If the close is due to the process exiting, we never
 425		 * linger..
 426		 */
 427		timeout = 0;
 428		if (sock_flag(sk, SOCK_LINGER) &&
 429		    !(current->flags & PF_EXITING))
 430			timeout = sk->sk_lingertime;
 431		sk->sk_prot->close(sk, timeout);
 432		sock->sk = NULL;
 433	}
 434	return 0;
 435}
 436EXPORT_SYMBOL(inet_release);
 437
 438int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 439{
 440	struct sock *sk = sock->sk;
 441	int err;
 442
 443	/* If the socket has its own bind function then use it. (RAW) */
 444	if (sk->sk_prot->bind) {
 445		return sk->sk_prot->bind(sk, uaddr, addr_len);
 446	}
 447	if (addr_len < sizeof(struct sockaddr_in))
 448		return -EINVAL;
 449
 450	/* BPF prog is run before any checks are done so that if the prog
 451	 * changes context in a wrong way it will be caught.
 452	 */
 453	err = BPF_CGROUP_RUN_PROG_INET4_BIND(sk, uaddr);
 
 454	if (err)
 455		return err;
 456
 457	return __inet_bind(sk, uaddr, addr_len, BIND_WITH_LOCK);
 
 
 
 
 
 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_can_nonlocal_bind(net, inet) &&
 494	    addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
 495	    chk_addr_ret != RTN_LOCAL &&
 496	    chk_addr_ret != RTN_MULTICAST &&
 497	    chk_addr_ret != RTN_BROADCAST)
 498		goto out;
 499
 500	snum = ntohs(addr->sin_port);
 501	err = -EACCES;
 502	if (snum && inet_port_requires_bind_service(net, snum) &&
 
 503	    !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
 504		goto out;
 505
 506	/*      We keep a pair of addresses. rcv_saddr is the one
 507	 *      used by hash lookups, and saddr is used for transmit.
 508	 *
 509	 *      In the BSD API these are the same except where it
 510	 *      would be illegal to use them (multicast/broadcast) in
 511	 *      which case the sending device address is used.
 512	 */
 513	if (flags & BIND_WITH_LOCK)
 514		lock_sock(sk);
 515
 516	/* Check these errors (active socket, double bind). */
 517	err = -EINVAL;
 518	if (sk->sk_state != TCP_CLOSE || inet->inet_num)
 519		goto out_release_sock;
 520
 521	inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
 522	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
 523		inet->inet_saddr = 0;  /* Use device */
 524
 525	/* Make sure we are allowed to bind here. */
 526	if (snum || !(inet->bind_address_no_port ||
 527		      (flags & BIND_FORCE_ADDRESS_NO_PORT))) {
 528		if (sk->sk_prot->get_port(sk, snum)) {
 
 529			inet->inet_saddr = inet->inet_rcv_saddr = 0;
 530			err = -EADDRINUSE;
 531			goto out_release_sock;
 532		}
 533		if (!(flags & BIND_FROM_BPF)) {
 534			err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
 535			if (err) {
 536				inet->inet_saddr = inet->inet_rcv_saddr = 0;
 
 
 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	int err;
 563
 564	if (addr_len < sizeof(uaddr->sa_family))
 565		return -EINVAL;
 
 
 
 
 566	if (uaddr->sa_family == AF_UNSPEC)
 567		return sk->sk_prot->disconnect(sk, flags);
 568
 569	if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
 570		err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
 571		if (err)
 572			return err;
 573	}
 574
 575	if (!inet_sk(sk)->inet_num && inet_autobind(sk))
 576		return -EAGAIN;
 577	return sk->sk_prot->connect(sk, uaddr, addr_len);
 578}
 579EXPORT_SYMBOL(inet_dgram_connect);
 580
 581static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
 582{
 583	DEFINE_WAIT_FUNC(wait, woken_wake_function);
 584
 585	add_wait_queue(sk_sleep(sk), &wait);
 586	sk->sk_write_pending += writebias;
 587
 588	/* Basic assumption: if someone sets sk->sk_err, he _must_
 589	 * change state of the socket from TCP_SYN_*.
 590	 * Connect() does not allow to get error notifications
 591	 * without closing the socket.
 592	 */
 593	while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 594		release_sock(sk);
 595		timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
 596		lock_sock(sk);
 597		if (signal_pending(current) || !timeo)
 598			break;
 599	}
 600	remove_wait_queue(sk_sleep(sk), &wait);
 601	sk->sk_write_pending -= writebias;
 602	return timeo;
 603}
 604
 605/*
 606 *	Connect to a remote host. There is regrettably still a little
 607 *	TCP 'magic' in here.
 608 */
 609int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 610			  int addr_len, int flags, int is_sendmsg)
 611{
 612	struct sock *sk = sock->sk;
 613	int err;
 614	long timeo;
 615
 616	/*
 617	 * uaddr can be NULL and addr_len can be 0 if:
 618	 * sk is a TCP fastopen active socket and
 619	 * TCP_FASTOPEN_CONNECT sockopt is set and
 620	 * we already have a valid cookie for this socket.
 621	 * In this case, user can call write() after connect().
 622	 * write() will invoke tcp_sendmsg_fastopen() which calls
 623	 * __inet_stream_connect().
 624	 */
 625	if (uaddr) {
 626		if (addr_len < sizeof(uaddr->sa_family))
 627			return -EINVAL;
 628
 629		if (uaddr->sa_family == AF_UNSPEC) {
 
 630			err = sk->sk_prot->disconnect(sk, flags);
 631			sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 632			goto out;
 633		}
 634	}
 635
 636	switch (sock->state) {
 637	default:
 638		err = -EINVAL;
 639		goto out;
 640	case SS_CONNECTED:
 641		err = -EISCONN;
 642		goto out;
 643	case SS_CONNECTING:
 644		if (inet_sk(sk)->defer_connect)
 645			err = is_sendmsg ? -EINPROGRESS : -EISCONN;
 646		else
 647			err = -EALREADY;
 648		/* Fall out of switch with err, set for this state */
 649		break;
 650	case SS_UNCONNECTED:
 651		err = -EISCONN;
 652		if (sk->sk_state != TCP_CLOSE)
 653			goto out;
 654
 655		if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
 656			err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
 657			if (err)
 658				goto out;
 659		}
 660
 661		err = sk->sk_prot->connect(sk, uaddr, addr_len);
 662		if (err < 0)
 663			goto out;
 664
 665		sock->state = SS_CONNECTING;
 666
 667		if (!err && inet_sk(sk)->defer_connect)
 668			goto out;
 669
 670		/* Just entered SS_CONNECTING state; the only
 671		 * difference is that return value in non-blocking
 672		 * case is EINPROGRESS, rather than EALREADY.
 673		 */
 674		err = -EINPROGRESS;
 675		break;
 676	}
 677
 678	timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
 679
 680	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
 681		int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
 682				tcp_sk(sk)->fastopen_req &&
 683				tcp_sk(sk)->fastopen_req->data ? 1 : 0;
 
 684
 685		/* Error code is set above */
 686		if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
 687			goto out;
 688
 689		err = sock_intr_errno(timeo);
 690		if (signal_pending(current))
 691			goto out;
 
 
 
 
 
 692	}
 693
 694	/* Connection was closed by RST, timeout, ICMP error
 695	 * or another process disconnected us.
 696	 */
 697	if (sk->sk_state == TCP_CLOSE)
 698		goto sock_error;
 699
 700	/* sk->sk_err may be not zero now, if RECVERR was ordered by user
 701	 * and error was received after socket entered established state.
 702	 * Hence, it is handled normally after connect() return successfully.
 703	 */
 704
 705	sock->state = SS_CONNECTED;
 706	err = 0;
 707out:
 708	return err;
 709
 710sock_error:
 711	err = sock_error(sk) ? : -ECONNABORTED;
 712	sock->state = SS_UNCONNECTED;
 
 713	if (sk->sk_prot->disconnect(sk, flags))
 714		sock->state = SS_DISCONNECTING;
 715	goto out;
 716}
 717EXPORT_SYMBOL(__inet_stream_connect);
 718
 719int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 720			int addr_len, int flags)
 721{
 722	int err;
 723
 724	lock_sock(sock->sk);
 725	err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
 726	release_sock(sock->sk);
 727	return err;
 728}
 729EXPORT_SYMBOL(inet_stream_connect);
 730
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 731/*
 732 *	Accept a pending connection. The TCP layer now gives BSD semantics.
 733 */
 734
 735int inet_accept(struct socket *sock, struct socket *newsock, int flags,
 736		bool kern)
 737{
 738	struct sock *sk1 = sock->sk;
 739	int err = -EINVAL;
 740	struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err, kern);
 741
 
 
 742	if (!sk2)
 743		goto do_err;
 744
 745	lock_sock(sk2);
 746
 747	sock_rps_record_flow(sk2);
 748	WARN_ON(!((1 << sk2->sk_state) &
 749		  (TCPF_ESTABLISHED | TCPF_SYN_RECV |
 750		  TCPF_CLOSE_WAIT | TCPF_CLOSE)));
 751
 752	sock_graft(sk2, newsock);
 753
 754	newsock->state = SS_CONNECTED;
 755	err = 0;
 756	release_sock(sk2);
 757do_err:
 758	return err;
 759}
 760EXPORT_SYMBOL(inet_accept);
 761
 762/*
 763 *	This does both peername and sockname.
 764 */
 765int inet_getname(struct socket *sock, struct sockaddr *uaddr,
 766		 int peer)
 767{
 768	struct sock *sk		= sock->sk;
 769	struct inet_sock *inet	= inet_sk(sk);
 770	DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
 
 771
 772	sin->sin_family = AF_INET;
 
 773	if (peer) {
 774		if (!inet->inet_dport ||
 775		    (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
 776		     peer == 1))
 
 777			return -ENOTCONN;
 
 778		sin->sin_port = inet->inet_dport;
 779		sin->sin_addr.s_addr = inet->inet_daddr;
 
 
 780	} else {
 781		__be32 addr = inet->inet_rcv_saddr;
 782		if (!addr)
 783			addr = inet->inet_saddr;
 784		sin->sin_port = inet->inet_sport;
 785		sin->sin_addr.s_addr = addr;
 
 
 786	}
 787	if (cgroup_bpf_enabled)
 788		BPF_CGROUP_RUN_SA_PROG_LOCK(sk, (struct sockaddr *)sin,
 789					    peer ? BPF_CGROUP_INET4_GETPEERNAME :
 790						   BPF_CGROUP_INET4_GETSOCKNAME,
 791					    NULL);
 792	memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 793	return sizeof(*sin);
 794}
 795EXPORT_SYMBOL(inet_getname);
 796
 797int inet_send_prepare(struct sock *sk)
 798{
 799	sock_rps_record_flow(sk);
 800
 801	/* We may need to bind the socket. */
 802	if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
 803	    inet_autobind(sk))
 804		return -EAGAIN;
 805
 806	return 0;
 807}
 808EXPORT_SYMBOL_GPL(inet_send_prepare);
 809
 810int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
 811{
 812	struct sock *sk = sock->sk;
 813
 814	if (unlikely(inet_send_prepare(sk)))
 815		return -EAGAIN;
 816
 817	return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
 818			       sk, msg, size);
 819}
 820EXPORT_SYMBOL(inet_sendmsg);
 821
 822ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
 823		      size_t size, int flags)
 824{
 
 825	struct sock *sk = sock->sk;
 826
 827	if (unlikely(inet_send_prepare(sk)))
 828		return -EAGAIN;
 829
 830	if (sk->sk_prot->sendpage)
 831		return sk->sk_prot->sendpage(sk, page, offset, size, flags);
 832	return sock_no_sendpage(sock, page, offset, size, flags);
 
 833}
 834EXPORT_SYMBOL(inet_sendpage);
 835
 836INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
 837					  size_t, int, int, int *));
 838int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
 839		 int flags)
 840{
 841	struct sock *sk = sock->sk;
 842	int addr_len = 0;
 843	int err;
 844
 845	if (likely(!(flags & MSG_ERRQUEUE)))
 846		sock_rps_record_flow(sk);
 847
 848	err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
 849			      sk, msg, size, flags & MSG_DONTWAIT,
 850			      flags & ~MSG_DONTWAIT, &addr_len);
 851	if (err >= 0)
 852		msg->msg_namelen = addr_len;
 853	return err;
 854}
 855EXPORT_SYMBOL(inet_recvmsg);
 856
 857int inet_shutdown(struct socket *sock, int how)
 858{
 859	struct sock *sk = sock->sk;
 860	int err = 0;
 861
 862	/* This should really check to make sure
 863	 * the socket is a TCP socket. (WHY AC...)
 864	 */
 865	how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
 866		       1->2 bit 2 snds.
 867		       2->3 */
 868	if ((how & ~SHUTDOWN_MASK) || !how)	/* MAXINT->0 */
 869		return -EINVAL;
 870
 871	lock_sock(sk);
 872	if (sock->state == SS_CONNECTING) {
 873		if ((1 << sk->sk_state) &
 874		    (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
 875			sock->state = SS_DISCONNECTING;
 876		else
 877			sock->state = SS_CONNECTED;
 878	}
 879
 880	switch (sk->sk_state) {
 881	case TCP_CLOSE:
 882		err = -ENOTCONN;
 883		/* Hack to wake up other listeners, who can poll for
 884		   EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
 885		fallthrough;
 886	default:
 887		sk->sk_shutdown |= how;
 888		if (sk->sk_prot->shutdown)
 889			sk->sk_prot->shutdown(sk, how);
 890		break;
 891
 892	/* Remaining two branches are temporary solution for missing
 893	 * close() in multithreaded environment. It is _not_ a good idea,
 894	 * but we have no choice until close() is repaired at VFS level.
 895	 */
 896	case TCP_LISTEN:
 897		if (!(how & RCV_SHUTDOWN))
 898			break;
 899		fallthrough;
 900	case TCP_SYN_SENT:
 901		err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
 902		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
 903		break;
 904	}
 905
 906	/* Wake up anyone sleeping in poll. */
 907	sk->sk_state_change(sk);
 908	release_sock(sk);
 909	return err;
 910}
 911EXPORT_SYMBOL(inet_shutdown);
 912
 913/*
 914 *	ioctl() calls you can issue on an INET socket. Most of these are
 915 *	device configuration and stuff and very rarely used. Some ioctls
 916 *	pass on to the socket itself.
 917 *
 918 *	NOTE: I like the idea of a module for the config stuff. ie ifconfig
 919 *	loads the devconfigure module does its configuring and unloads it.
 920 *	There's a good 20K of config code hanging around the kernel.
 921 */
 922
 923int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 924{
 925	struct sock *sk = sock->sk;
 926	int err = 0;
 927	struct net *net = sock_net(sk);
 928	void __user *p = (void __user *)arg;
 929	struct ifreq ifr;
 930	struct rtentry rt;
 931
 932	switch (cmd) {
 933	case SIOCADDRT:
 934	case SIOCDELRT:
 935		if (copy_from_user(&rt, p, sizeof(struct rtentry)))
 936			return -EFAULT;
 937		err = ip_rt_ioctl(net, cmd, &rt);
 938		break;
 939	case SIOCRTMSG:
 940		err = -EINVAL;
 941		break;
 942	case SIOCDARP:
 943	case SIOCGARP:
 944	case SIOCSARP:
 945		err = arp_ioctl(net, cmd, (void __user *)arg);
 946		break;
 947	case SIOCGIFADDR:
 948	case SIOCGIFBRDADDR:
 949	case SIOCGIFNETMASK:
 950	case SIOCGIFDSTADDR:
 951	case SIOCGIFPFLAGS:
 952		if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
 953			return -EFAULT;
 954		err = devinet_ioctl(net, cmd, &ifr);
 955		if (!err && copy_to_user(p, &ifr, sizeof(struct ifreq)))
 956			err = -EFAULT;
 957		break;
 958
 959	case SIOCSIFADDR:
 960	case SIOCSIFBRDADDR:
 961	case SIOCSIFNETMASK:
 962	case SIOCSIFDSTADDR:
 963	case SIOCSIFPFLAGS:
 964	case SIOCSIFFLAGS:
 965		if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
 966			return -EFAULT;
 967		err = devinet_ioctl(net, cmd, &ifr);
 968		break;
 969	default:
 970		if (sk->sk_prot->ioctl)
 971			err = sk->sk_prot->ioctl(sk, cmd, arg);
 972		else
 973			err = -ENOIOCTLCMD;
 974		break;
 975	}
 976	return err;
 977}
 978EXPORT_SYMBOL(inet_ioctl);
 979
 980#ifdef CONFIG_COMPAT
 981static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd,
 982		struct compat_rtentry __user *ur)
 983{
 984	compat_uptr_t rtdev;
 985	struct rtentry rt;
 986
 987	if (copy_from_user(&rt.rt_dst, &ur->rt_dst,
 988			3 * sizeof(struct sockaddr)) ||
 989	    get_user(rt.rt_flags, &ur->rt_flags) ||
 990	    get_user(rt.rt_metric, &ur->rt_metric) ||
 991	    get_user(rt.rt_mtu, &ur->rt_mtu) ||
 992	    get_user(rt.rt_window, &ur->rt_window) ||
 993	    get_user(rt.rt_irtt, &ur->rt_irtt) ||
 994	    get_user(rtdev, &ur->rt_dev))
 995		return -EFAULT;
 996
 997	rt.rt_dev = compat_ptr(rtdev);
 998	return ip_rt_ioctl(sock_net(sk), cmd, &rt);
 999}
1000
1001static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1002{
1003	void __user *argp = compat_ptr(arg);
1004	struct sock *sk = sock->sk;
1005
1006	switch (cmd) {
1007	case SIOCADDRT:
1008	case SIOCDELRT:
1009		return inet_compat_routing_ioctl(sk, cmd, argp);
1010	default:
1011		if (!sk->sk_prot->compat_ioctl)
1012			return -ENOIOCTLCMD;
1013		return sk->sk_prot->compat_ioctl(sk, cmd, arg);
1014	}
1015}
1016#endif /* CONFIG_COMPAT */
1017
1018const struct proto_ops inet_stream_ops = {
1019	.family		   = PF_INET,
1020	.owner		   = THIS_MODULE,
1021	.release	   = inet_release,
1022	.bind		   = inet_bind,
1023	.connect	   = inet_stream_connect,
1024	.socketpair	   = sock_no_socketpair,
1025	.accept		   = inet_accept,
1026	.getname	   = inet_getname,
1027	.poll		   = tcp_poll,
1028	.ioctl		   = inet_ioctl,
1029	.gettstamp	   = sock_gettstamp,
1030	.listen		   = inet_listen,
1031	.shutdown	   = inet_shutdown,
1032	.setsockopt	   = sock_common_setsockopt,
1033	.getsockopt	   = sock_common_getsockopt,
1034	.sendmsg	   = inet_sendmsg,
1035	.recvmsg	   = inet_recvmsg,
1036#ifdef CONFIG_MMU
1037	.mmap		   = tcp_mmap,
1038#endif
1039	.sendpage	   = inet_sendpage,
1040	.splice_read	   = tcp_splice_read,
1041	.read_sock	   = tcp_read_sock,
 
1042	.sendmsg_locked    = tcp_sendmsg_locked,
1043	.sendpage_locked   = tcp_sendpage_locked,
1044	.peek_len	   = tcp_peek_len,
1045#ifdef CONFIG_COMPAT
1046	.compat_ioctl	   = inet_compat_ioctl,
1047#endif
1048	.set_rcvlowat	   = tcp_set_rcvlowat,
1049};
1050EXPORT_SYMBOL(inet_stream_ops);
1051
1052const struct proto_ops inet_dgram_ops = {
1053	.family		   = PF_INET,
1054	.owner		   = THIS_MODULE,
1055	.release	   = inet_release,
1056	.bind		   = inet_bind,
1057	.connect	   = inet_dgram_connect,
1058	.socketpair	   = sock_no_socketpair,
1059	.accept		   = sock_no_accept,
1060	.getname	   = inet_getname,
1061	.poll		   = udp_poll,
1062	.ioctl		   = inet_ioctl,
1063	.gettstamp	   = sock_gettstamp,
1064	.listen		   = sock_no_listen,
1065	.shutdown	   = inet_shutdown,
1066	.setsockopt	   = sock_common_setsockopt,
1067	.getsockopt	   = sock_common_getsockopt,
1068	.sendmsg	   = inet_sendmsg,
 
1069	.recvmsg	   = inet_recvmsg,
1070	.mmap		   = sock_no_mmap,
1071	.sendpage	   = inet_sendpage,
1072	.set_peek_off	   = sk_set_peek_off,
1073#ifdef CONFIG_COMPAT
1074	.compat_ioctl	   = inet_compat_ioctl,
1075#endif
1076};
1077EXPORT_SYMBOL(inet_dgram_ops);
1078
1079/*
1080 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1081 * udp_poll
1082 */
1083static const struct proto_ops inet_sockraw_ops = {
1084	.family		   = PF_INET,
1085	.owner		   = THIS_MODULE,
1086	.release	   = inet_release,
1087	.bind		   = inet_bind,
1088	.connect	   = inet_dgram_connect,
1089	.socketpair	   = sock_no_socketpair,
1090	.accept		   = sock_no_accept,
1091	.getname	   = inet_getname,
1092	.poll		   = datagram_poll,
1093	.ioctl		   = inet_ioctl,
1094	.gettstamp	   = sock_gettstamp,
1095	.listen		   = sock_no_listen,
1096	.shutdown	   = inet_shutdown,
1097	.setsockopt	   = sock_common_setsockopt,
1098	.getsockopt	   = sock_common_getsockopt,
1099	.sendmsg	   = inet_sendmsg,
1100	.recvmsg	   = inet_recvmsg,
1101	.mmap		   = sock_no_mmap,
1102	.sendpage	   = inet_sendpage,
1103#ifdef CONFIG_COMPAT
1104	.compat_ioctl	   = inet_compat_ioctl,
1105#endif
1106};
1107
1108static const struct net_proto_family inet_family_ops = {
1109	.family = PF_INET,
1110	.create = inet_create,
1111	.owner	= THIS_MODULE,
1112};
1113
1114/* Upon startup we insert all the elements in inetsw_array[] into
1115 * the linked list inetsw.
1116 */
1117static struct inet_protosw inetsw_array[] =
1118{
1119	{
1120		.type =       SOCK_STREAM,
1121		.protocol =   IPPROTO_TCP,
1122		.prot =       &tcp_prot,
1123		.ops =        &inet_stream_ops,
1124		.flags =      INET_PROTOSW_PERMANENT |
1125			      INET_PROTOSW_ICSK,
1126	},
1127
1128	{
1129		.type =       SOCK_DGRAM,
1130		.protocol =   IPPROTO_UDP,
1131		.prot =       &udp_prot,
1132		.ops =        &inet_dgram_ops,
1133		.flags =      INET_PROTOSW_PERMANENT,
1134       },
1135
1136       {
1137		.type =       SOCK_DGRAM,
1138		.protocol =   IPPROTO_ICMP,
1139		.prot =       &ping_prot,
1140		.ops =        &inet_sockraw_ops,
1141		.flags =      INET_PROTOSW_REUSE,
1142       },
1143
1144       {
1145	       .type =       SOCK_RAW,
1146	       .protocol =   IPPROTO_IP,	/* wild card */
1147	       .prot =       &raw_prot,
1148	       .ops =        &inet_sockraw_ops,
1149	       .flags =      INET_PROTOSW_REUSE,
1150       }
1151};
1152
1153#define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1154
1155void inet_register_protosw(struct inet_protosw *p)
1156{
1157	struct list_head *lh;
1158	struct inet_protosw *answer;
1159	int protocol = p->protocol;
1160	struct list_head *last_perm;
1161
1162	spin_lock_bh(&inetsw_lock);
1163
1164	if (p->type >= SOCK_MAX)
1165		goto out_illegal;
1166
1167	/* If we are trying to override a permanent protocol, bail. */
1168	last_perm = &inetsw[p->type];
1169	list_for_each(lh, &inetsw[p->type]) {
1170		answer = list_entry(lh, struct inet_protosw, list);
1171		/* Check only the non-wild match. */
1172		if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1173			break;
1174		if (protocol == answer->protocol)
1175			goto out_permanent;
1176		last_perm = lh;
1177	}
1178
1179	/* Add the new entry after the last permanent entry if any, so that
1180	 * the new entry does not override a permanent entry when matched with
1181	 * a wild-card protocol. But it is allowed to override any existing
1182	 * non-permanent entry.  This means that when we remove this entry, the
1183	 * system automatically returns to the old behavior.
1184	 */
1185	list_add_rcu(&p->list, last_perm);
1186out:
1187	spin_unlock_bh(&inetsw_lock);
1188
1189	return;
1190
1191out_permanent:
1192	pr_err("Attempt to override permanent protocol %d\n", protocol);
1193	goto out;
1194
1195out_illegal:
1196	pr_err("Ignoring attempt to register invalid socket type %d\n",
1197	       p->type);
1198	goto out;
1199}
1200EXPORT_SYMBOL(inet_register_protosw);
1201
1202void inet_unregister_protosw(struct inet_protosw *p)
1203{
1204	if (INET_PROTOSW_PERMANENT & p->flags) {
1205		pr_err("Attempt to unregister permanent protocol %d\n",
1206		       p->protocol);
1207	} else {
1208		spin_lock_bh(&inetsw_lock);
1209		list_del_rcu(&p->list);
1210		spin_unlock_bh(&inetsw_lock);
1211
1212		synchronize_net();
1213	}
1214}
1215EXPORT_SYMBOL(inet_unregister_protosw);
1216
1217static int inet_sk_reselect_saddr(struct sock *sk)
1218{
1219	struct inet_sock *inet = inet_sk(sk);
1220	__be32 old_saddr = inet->inet_saddr;
1221	__be32 daddr = inet->inet_daddr;
1222	struct flowi4 *fl4;
1223	struct rtable *rt;
1224	__be32 new_saddr;
1225	struct ip_options_rcu *inet_opt;
 
1226
1227	inet_opt = rcu_dereference_protected(inet->inet_opt,
1228					     lockdep_sock_is_held(sk));
1229	if (inet_opt && inet_opt->opt.srr)
1230		daddr = inet_opt->opt.faddr;
1231
1232	/* Query new route. */
1233	fl4 = &inet->cork.fl.u.ip4;
1234	rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1235			      sk->sk_bound_dev_if, sk->sk_protocol,
1236			      inet->inet_sport, inet->inet_dport, sk);
1237	if (IS_ERR(rt))
1238		return PTR_ERR(rt);
1239
1240	sk_setup_caps(sk, &rt->dst);
1241
1242	new_saddr = fl4->saddr;
1243
1244	if (new_saddr == old_saddr)
 
1245		return 0;
 
 
 
 
 
 
 
 
 
1246
1247	if (sock_net(sk)->ipv4.sysctl_ip_dynaddr > 1) {
1248		pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1249			__func__, &old_saddr, &new_saddr);
1250	}
1251
1252	inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1253
1254	/*
1255	 * XXX The only one ugly spot where we need to
1256	 * XXX really change the sockets identity after
1257	 * XXX it has entered the hashes. -DaveM
1258	 *
1259	 * Besides that, it does not check for connection
1260	 * uniqueness. Wait for troubles.
1261	 */
1262	return __sk_prot_rehash(sk);
1263}
1264
1265int inet_sk_rebuild_header(struct sock *sk)
1266{
1267	struct inet_sock *inet = inet_sk(sk);
1268	struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1269	__be32 daddr;
1270	struct ip_options_rcu *inet_opt;
1271	struct flowi4 *fl4;
1272	int err;
1273
1274	/* Route is OK, nothing to do. */
1275	if (rt)
1276		return 0;
1277
1278	/* Reroute. */
1279	rcu_read_lock();
1280	inet_opt = rcu_dereference(inet->inet_opt);
1281	daddr = inet->inet_daddr;
1282	if (inet_opt && inet_opt->opt.srr)
1283		daddr = inet_opt->opt.faddr;
1284	rcu_read_unlock();
1285	fl4 = &inet->cork.fl.u.ip4;
1286	rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1287				   inet->inet_dport, inet->inet_sport,
1288				   sk->sk_protocol, RT_CONN_FLAGS(sk),
1289				   sk->sk_bound_dev_if);
1290	if (!IS_ERR(rt)) {
1291		err = 0;
1292		sk_setup_caps(sk, &rt->dst);
1293	} else {
1294		err = PTR_ERR(rt);
1295
1296		/* Routing failed... */
1297		sk->sk_route_caps = 0;
1298		/*
1299		 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1300		 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1301		 */
1302		if (!sock_net(sk)->ipv4.sysctl_ip_dynaddr ||
1303		    sk->sk_state != TCP_SYN_SENT ||
1304		    (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1305		    (err = inet_sk_reselect_saddr(sk)) != 0)
1306			sk->sk_err_soft = -err;
1307	}
1308
1309	return err;
1310}
1311EXPORT_SYMBOL(inet_sk_rebuild_header);
1312
1313void inet_sk_set_state(struct sock *sk, int state)
1314{
1315	trace_inet_sock_set_state(sk, sk->sk_state, state);
1316	sk->sk_state = state;
1317}
1318EXPORT_SYMBOL(inet_sk_set_state);
1319
1320void inet_sk_state_store(struct sock *sk, int newstate)
1321{
1322	trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1323	smp_store_release(&sk->sk_state, newstate);
1324}
1325
1326struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1327				 netdev_features_t features)
1328{
1329	bool udpfrag = false, fixedid = false, gso_partial, encap;
1330	struct sk_buff *segs = ERR_PTR(-EINVAL);
1331	const struct net_offload *ops;
1332	unsigned int offset = 0;
1333	struct iphdr *iph;
1334	int proto, tot_len;
1335	int nhoff;
1336	int ihl;
1337	int id;
1338
1339	skb_reset_network_header(skb);
1340	nhoff = skb_network_header(skb) - skb_mac_header(skb);
1341	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1342		goto out;
1343
1344	iph = ip_hdr(skb);
1345	ihl = iph->ihl * 4;
1346	if (ihl < sizeof(*iph))
1347		goto out;
1348
1349	id = ntohs(iph->id);
1350	proto = iph->protocol;
1351
1352	/* Warning: after this point, iph might be no longer valid */
1353	if (unlikely(!pskb_may_pull(skb, ihl)))
1354		goto out;
1355	__skb_pull(skb, ihl);
1356
1357	encap = SKB_GSO_CB(skb)->encap_level > 0;
1358	if (encap)
1359		features &= skb->dev->hw_enc_features;
1360	SKB_GSO_CB(skb)->encap_level += ihl;
1361
1362	skb_reset_transport_header(skb);
1363
1364	segs = ERR_PTR(-EPROTONOSUPPORT);
1365
1366	if (!skb->encapsulation || encap) {
1367		udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1368		fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1369
1370		/* fixed ID is invalid if DF bit is not set */
1371		if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1372			goto out;
1373	}
1374
1375	ops = rcu_dereference(inet_offloads[proto]);
1376	if (likely(ops && ops->callbacks.gso_segment))
1377		segs = ops->callbacks.gso_segment(skb, features);
 
 
 
1378
1379	if (IS_ERR_OR_NULL(segs))
1380		goto out;
1381
1382	gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1383
1384	skb = segs;
1385	do {
1386		iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1387		if (udpfrag) {
1388			iph->frag_off = htons(offset >> 3);
1389			if (skb->next)
1390				iph->frag_off |= htons(IP_MF);
1391			offset += skb->len - nhoff - ihl;
1392			tot_len = skb->len - nhoff;
1393		} else if (skb_is_gso(skb)) {
1394			if (!fixedid) {
1395				iph->id = htons(id);
1396				id += skb_shinfo(skb)->gso_segs;
1397			}
1398
1399			if (gso_partial)
1400				tot_len = skb_shinfo(skb)->gso_size +
1401					  SKB_GSO_CB(skb)->data_offset +
1402					  skb->head - (unsigned char *)iph;
1403			else
1404				tot_len = skb->len - nhoff;
1405		} else {
1406			if (!fixedid)
1407				iph->id = htons(id++);
1408			tot_len = skb->len - nhoff;
1409		}
1410		iph->tot_len = htons(tot_len);
1411		ip_send_check(iph);
1412		if (encap)
1413			skb_reset_inner_headers(skb);
1414		skb->network_header = (u8 *)iph - skb->head;
1415		skb_reset_mac_len(skb);
1416	} while ((skb = skb->next));
1417
1418out:
1419	return segs;
1420}
1421EXPORT_SYMBOL(inet_gso_segment);
1422
1423static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1424					netdev_features_t features)
1425{
1426	if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1427		return ERR_PTR(-EINVAL);
1428
1429	return inet_gso_segment(skb, features);
1430}
1431
1432struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1433{
1434	const struct net_offload *ops;
1435	struct sk_buff *pp = NULL;
1436	const struct iphdr *iph;
1437	struct sk_buff *p;
1438	unsigned int hlen;
1439	unsigned int off;
1440	unsigned int id;
1441	int flush = 1;
1442	int proto;
1443
1444	off = skb_gro_offset(skb);
1445	hlen = off + sizeof(*iph);
1446	iph = skb_gro_header_fast(skb, off);
1447	if (skb_gro_header_hard(skb, hlen)) {
1448		iph = skb_gro_header_slow(skb, hlen, off);
1449		if (unlikely(!iph))
1450			goto out;
1451	}
1452
1453	proto = iph->protocol;
1454
1455	rcu_read_lock();
1456	ops = rcu_dereference(inet_offloads[proto]);
1457	if (!ops || !ops->callbacks.gro_receive)
1458		goto out_unlock;
1459
1460	if (*(u8 *)iph != 0x45)
1461		goto out_unlock;
1462
1463	if (ip_is_fragment(iph))
1464		goto out_unlock;
1465
1466	if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1467		goto out_unlock;
1468
 
1469	id = ntohl(*(__be32 *)&iph->id);
1470	flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1471	id >>= 16;
1472
1473	list_for_each_entry(p, head, list) {
1474		struct iphdr *iph2;
1475		u16 flush_id;
1476
1477		if (!NAPI_GRO_CB(p)->same_flow)
1478			continue;
1479
1480		iph2 = (struct iphdr *)(p->data + off);
1481		/* The above works because, with the exception of the top
1482		 * (inner most) layer, we only aggregate pkts with the same
1483		 * hdr length so all the hdrs we'll need to verify will start
1484		 * at the same offset.
1485		 */
1486		if ((iph->protocol ^ iph2->protocol) |
1487		    ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1488		    ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1489			NAPI_GRO_CB(p)->same_flow = 0;
1490			continue;
1491		}
1492
1493		/* All fields must match except length and checksum. */
1494		NAPI_GRO_CB(p)->flush |=
1495			(iph->ttl ^ iph2->ttl) |
1496			(iph->tos ^ iph2->tos) |
1497			((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1498
1499		NAPI_GRO_CB(p)->flush |= flush;
1500
1501		/* We need to store of the IP ID check to be included later
1502		 * when we can verify that this packet does in fact belong
1503		 * to a given flow.
1504		 */
1505		flush_id = (u16)(id - ntohs(iph2->id));
1506
1507		/* This bit of code makes it much easier for us to identify
1508		 * the cases where we are doing atomic vs non-atomic IP ID
1509		 * checks.  Specifically an atomic check can return IP ID
1510		 * values 0 - 0xFFFF, while a non-atomic check can only
1511		 * return 0 or 0xFFFF.
1512		 */
1513		if (!NAPI_GRO_CB(p)->is_atomic ||
1514		    !(iph->frag_off & htons(IP_DF))) {
1515			flush_id ^= NAPI_GRO_CB(p)->count;
1516			flush_id = flush_id ? 0xFFFF : 0;
1517		}
1518
1519		/* If the previous IP ID value was based on an atomic
1520		 * datagram we can overwrite the value and ignore it.
1521		 */
1522		if (NAPI_GRO_CB(skb)->is_atomic)
1523			NAPI_GRO_CB(p)->flush_id = flush_id;
1524		else
1525			NAPI_GRO_CB(p)->flush_id |= flush_id;
1526	}
1527
1528	NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1529	NAPI_GRO_CB(skb)->flush |= flush;
1530	skb_set_network_header(skb, off);
1531	/* The above will be needed by the transport layer if there is one
1532	 * immediately following this IP hdr.
1533	 */
1534
1535	/* Note : No need to call skb_gro_postpull_rcsum() here,
1536	 * as we already checked checksum over ipv4 header was 0
1537	 */
1538	skb_gro_pull(skb, sizeof(*iph));
1539	skb_set_transport_header(skb, skb_gro_offset(skb));
1540
1541	pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1542				       ops->callbacks.gro_receive, head, skb);
1543
1544out_unlock:
1545	rcu_read_unlock();
1546
1547out:
1548	skb_gro_flush_final(skb, pp, flush);
1549
1550	return pp;
1551}
1552EXPORT_SYMBOL(inet_gro_receive);
1553
1554static struct sk_buff *ipip_gro_receive(struct list_head *head,
1555					struct sk_buff *skb)
1556{
1557	if (NAPI_GRO_CB(skb)->encap_mark) {
1558		NAPI_GRO_CB(skb)->flush = 1;
1559		return NULL;
1560	}
1561
1562	NAPI_GRO_CB(skb)->encap_mark = 1;
1563
1564	return inet_gro_receive(head, skb);
1565}
1566
1567#define SECONDS_PER_DAY	86400
1568
1569/* inet_current_timestamp - Return IP network timestamp
1570 *
1571 * Return milliseconds since midnight in network byte order.
1572 */
1573__be32 inet_current_timestamp(void)
1574{
1575	u32 secs;
1576	u32 msecs;
1577	struct timespec64 ts;
1578
1579	ktime_get_real_ts64(&ts);
1580
1581	/* Get secs since midnight. */
1582	(void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1583	/* Convert to msecs. */
1584	msecs = secs * MSEC_PER_SEC;
1585	/* Convert nsec to msec. */
1586	msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1587
1588	/* Convert to network byte order. */
1589	return htonl(msecs);
1590}
1591EXPORT_SYMBOL(inet_current_timestamp);
1592
1593int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1594{
1595	if (sk->sk_family == AF_INET)
 
 
1596		return ip_recv_error(sk, msg, len, addr_len);
1597#if IS_ENABLED(CONFIG_IPV6)
1598	if (sk->sk_family == AF_INET6)
1599		return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1600#endif
1601	return -EINVAL;
1602}
 
1603
1604int inet_gro_complete(struct sk_buff *skb, int nhoff)
1605{
1606	__be16 newlen = htons(skb->len - nhoff);
1607	struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1608	const struct net_offload *ops;
 
1609	int proto = iph->protocol;
1610	int err = -ENOSYS;
1611
1612	if (skb->encapsulation) {
1613		skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1614		skb_set_inner_network_header(skb, nhoff);
1615	}
1616
1617	csum_replace2(&iph->check, iph->tot_len, newlen);
1618	iph->tot_len = newlen;
1619
1620	rcu_read_lock();
1621	ops = rcu_dereference(inet_offloads[proto]);
1622	if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1623		goto out_unlock;
1624
1625	/* Only need to add sizeof(*iph) to get to the next hdr below
1626	 * because any hdr with option will have been flushed in
1627	 * inet_gro_receive().
1628	 */
1629	err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1630			      tcp4_gro_complete, udp4_gro_complete,
1631			      skb, nhoff + sizeof(*iph));
1632
1633out_unlock:
1634	rcu_read_unlock();
1635
1636	return err;
1637}
1638EXPORT_SYMBOL(inet_gro_complete);
1639
1640static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1641{
1642	skb->encapsulation = 1;
1643	skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1644	return inet_gro_complete(skb, nhoff);
1645}
1646
1647int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1648			 unsigned short type, unsigned char protocol,
1649			 struct net *net)
1650{
1651	struct socket *sock;
1652	int rc = sock_create_kern(net, family, type, protocol, &sock);
1653
1654	if (rc == 0) {
1655		*sk = sock->sk;
1656		(*sk)->sk_allocation = GFP_ATOMIC;
 
1657		/*
1658		 * Unhash it so that IP input processing does not even see it,
1659		 * we do not wish this socket to see incoming packets.
1660		 */
1661		(*sk)->sk_prot->unhash(*sk);
1662	}
1663	return rc;
1664}
1665EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1666
1667u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt)
1668{
1669	return  *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt);
1670}
1671EXPORT_SYMBOL_GPL(snmp_get_cpu_field);
1672
1673unsigned long snmp_fold_field(void __percpu *mib, int offt)
1674{
1675	unsigned long res = 0;
1676	int i;
1677
1678	for_each_possible_cpu(i)
1679		res += snmp_get_cpu_field(mib, i, offt);
1680	return res;
1681}
1682EXPORT_SYMBOL_GPL(snmp_fold_field);
1683
1684#if BITS_PER_LONG==32
1685
1686u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1687			 size_t syncp_offset)
1688{
1689	void *bhptr;
1690	struct u64_stats_sync *syncp;
1691	u64 v;
1692	unsigned int start;
1693
1694	bhptr = per_cpu_ptr(mib, cpu);
1695	syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1696	do {
1697		start = u64_stats_fetch_begin_irq(syncp);
1698		v = *(((u64 *)bhptr) + offt);
1699	} while (u64_stats_fetch_retry_irq(syncp, start));
1700
1701	return v;
1702}
1703EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1704
1705u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1706{
1707	u64 res = 0;
1708	int cpu;
1709
1710	for_each_possible_cpu(cpu) {
1711		res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1712	}
1713	return res;
1714}
1715EXPORT_SYMBOL_GPL(snmp_fold_field64);
1716#endif
1717
1718#ifdef CONFIG_IP_MULTICAST
1719static const struct net_protocol igmp_protocol = {
1720	.handler =	igmp_rcv,
1721	.netns_ok =	1,
1722};
1723#endif
1724
1725/* thinking of making this const? Don't.
1726 * early_demux can change based on sysctl.
1727 */
1728static struct net_protocol tcp_protocol = {
1729	.early_demux	=	tcp_v4_early_demux,
1730	.early_demux_handler =  tcp_v4_early_demux,
1731	.handler	=	tcp_v4_rcv,
1732	.err_handler	=	tcp_v4_err,
1733	.no_policy	=	1,
1734	.netns_ok	=	1,
1735	.icmp_strict_tag_validation = 1,
1736};
1737
1738/* thinking of making this const? Don't.
1739 * early_demux can change based on sysctl.
1740 */
1741static struct net_protocol udp_protocol = {
1742	.early_demux =	udp_v4_early_demux,
1743	.early_demux_handler =	udp_v4_early_demux,
1744	.handler =	udp_rcv,
1745	.err_handler =	udp_err,
1746	.no_policy =	1,
1747	.netns_ok =	1,
1748};
1749
1750static const struct net_protocol icmp_protocol = {
1751	.handler =	icmp_rcv,
1752	.err_handler =	icmp_err,
1753	.no_policy =	1,
1754	.netns_ok =	1,
1755};
1756
1757static __net_init int ipv4_mib_init_net(struct net *net)
1758{
1759	int i;
1760
1761	net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1762	if (!net->mib.tcp_statistics)
1763		goto err_tcp_mib;
1764	net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1765	if (!net->mib.ip_statistics)
1766		goto err_ip_mib;
1767
1768	for_each_possible_cpu(i) {
1769		struct ipstats_mib *af_inet_stats;
1770		af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1771		u64_stats_init(&af_inet_stats->syncp);
1772	}
1773
1774	net->mib.net_statistics = alloc_percpu(struct linux_mib);
1775	if (!net->mib.net_statistics)
1776		goto err_net_mib;
1777	net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1778	if (!net->mib.udp_statistics)
1779		goto err_udp_mib;
1780	net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1781	if (!net->mib.udplite_statistics)
1782		goto err_udplite_mib;
1783	net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1784	if (!net->mib.icmp_statistics)
1785		goto err_icmp_mib;
1786	net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1787					      GFP_KERNEL);
1788	if (!net->mib.icmpmsg_statistics)
1789		goto err_icmpmsg_mib;
1790
1791	tcp_mib_init(net);
1792	return 0;
1793
1794err_icmpmsg_mib:
1795	free_percpu(net->mib.icmp_statistics);
1796err_icmp_mib:
1797	free_percpu(net->mib.udplite_statistics);
1798err_udplite_mib:
1799	free_percpu(net->mib.udp_statistics);
1800err_udp_mib:
1801	free_percpu(net->mib.net_statistics);
1802err_net_mib:
1803	free_percpu(net->mib.ip_statistics);
1804err_ip_mib:
1805	free_percpu(net->mib.tcp_statistics);
1806err_tcp_mib:
1807	return -ENOMEM;
1808}
1809
1810static __net_exit void ipv4_mib_exit_net(struct net *net)
1811{
1812	kfree(net->mib.icmpmsg_statistics);
1813	free_percpu(net->mib.icmp_statistics);
1814	free_percpu(net->mib.udplite_statistics);
1815	free_percpu(net->mib.udp_statistics);
1816	free_percpu(net->mib.net_statistics);
1817	free_percpu(net->mib.ip_statistics);
1818	free_percpu(net->mib.tcp_statistics);
1819#ifdef CONFIG_MPTCP
1820	/* allocated on demand, see mptcp_init_sock() */
1821	free_percpu(net->mib.mptcp_statistics);
1822#endif
1823}
1824
1825static __net_initdata struct pernet_operations ipv4_mib_ops = {
1826	.init = ipv4_mib_init_net,
1827	.exit = ipv4_mib_exit_net,
1828};
1829
1830static int __init init_ipv4_mibs(void)
1831{
1832	return register_pernet_subsys(&ipv4_mib_ops);
1833}
1834
1835static __net_init int inet_init_net(struct net *net)
1836{
1837	/*
1838	 * Set defaults for local port range
1839	 */
1840	seqlock_init(&net->ipv4.ip_local_ports.lock);
1841	net->ipv4.ip_local_ports.range[0] =  32768;
1842	net->ipv4.ip_local_ports.range[1] =  60999;
1843
1844	seqlock_init(&net->ipv4.ping_group_range.lock);
1845	/*
1846	 * Sane defaults - nobody may create ping sockets.
1847	 * Boot scripts should set this to distro-specific group.
1848	 */
1849	net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1850	net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1851
1852	/* Default values for sysctl-controlled parameters.
1853	 * We set them here, in case sysctl is not compiled.
1854	 */
1855	net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1856	net->ipv4.sysctl_ip_fwd_update_priority = 1;
1857	net->ipv4.sysctl_ip_dynaddr = 0;
1858	net->ipv4.sysctl_ip_early_demux = 1;
1859	net->ipv4.sysctl_udp_early_demux = 1;
1860	net->ipv4.sysctl_tcp_early_demux = 1;
1861	net->ipv4.sysctl_nexthop_compat_mode = 1;
1862#ifdef CONFIG_SYSCTL
1863	net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1864#endif
1865
1866	/* Some igmp sysctl, whose values are always used */
1867	net->ipv4.sysctl_igmp_max_memberships = 20;
1868	net->ipv4.sysctl_igmp_max_msf = 10;
1869	/* IGMP reports for link-local multicast groups are enabled by default */
1870	net->ipv4.sysctl_igmp_llm_reports = 1;
1871	net->ipv4.sysctl_igmp_qrv = 2;
1872
 
 
1873	return 0;
1874}
1875
1876static __net_initdata struct pernet_operations af_inet_ops = {
1877	.init = inet_init_net,
1878};
1879
1880static int __init init_inet_pernet_ops(void)
1881{
1882	return register_pernet_subsys(&af_inet_ops);
1883}
1884
1885static int ipv4_proc_init(void);
1886
1887/*
1888 *	IP protocol layer initialiser
1889 */
1890
1891static struct packet_offload ip_packet_offload __read_mostly = {
1892	.type = cpu_to_be16(ETH_P_IP),
1893	.callbacks = {
1894		.gso_segment = inet_gso_segment,
1895		.gro_receive = inet_gro_receive,
1896		.gro_complete = inet_gro_complete,
1897	},
1898};
1899
1900static const struct net_offload ipip_offload = {
1901	.callbacks = {
1902		.gso_segment	= ipip_gso_segment,
1903		.gro_receive	= ipip_gro_receive,
1904		.gro_complete	= ipip_gro_complete,
1905	},
1906};
1907
1908static int __init ipip_offload_init(void)
1909{
1910	return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1911}
1912
1913static int __init ipv4_offload_init(void)
1914{
1915	/*
1916	 * Add offloads
1917	 */
1918	if (udpv4_offload_init() < 0)
1919		pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1920	if (tcpv4_offload_init() < 0)
1921		pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1922	if (ipip_offload_init() < 0)
1923		pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1924
1925	dev_add_offload(&ip_packet_offload);
1926	return 0;
1927}
1928
1929fs_initcall(ipv4_offload_init);
1930
1931static struct packet_type ip_packet_type __read_mostly = {
1932	.type = cpu_to_be16(ETH_P_IP),
1933	.func = ip_rcv,
1934	.list_func = ip_list_rcv,
1935};
1936
1937static int __init inet_init(void)
1938{
1939	struct inet_protosw *q;
1940	struct list_head *r;
1941	int rc;
1942
1943	sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1944
 
 
1945	rc = proto_register(&tcp_prot, 1);
1946	if (rc)
1947		goto out;
1948
1949	rc = proto_register(&udp_prot, 1);
1950	if (rc)
1951		goto out_unregister_tcp_proto;
1952
1953	rc = proto_register(&raw_prot, 1);
1954	if (rc)
1955		goto out_unregister_udp_proto;
1956
1957	rc = proto_register(&ping_prot, 1);
1958	if (rc)
1959		goto out_unregister_raw_proto;
1960
1961	/*
1962	 *	Tell SOCKET that we are alive...
1963	 */
1964
1965	(void)sock_register(&inet_family_ops);
1966
1967#ifdef CONFIG_SYSCTL
1968	ip_static_sysctl_init();
1969#endif
1970
1971	/*
1972	 *	Add all the base protocols.
1973	 */
1974
1975	if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1976		pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1977	if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1978		pr_crit("%s: Cannot add UDP protocol\n", __func__);
1979	if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1980		pr_crit("%s: Cannot add TCP protocol\n", __func__);
1981#ifdef CONFIG_IP_MULTICAST
1982	if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1983		pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1984#endif
1985
1986	/* Register the socket-side information for inet_create. */
1987	for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1988		INIT_LIST_HEAD(r);
1989
1990	for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1991		inet_register_protosw(q);
1992
1993	/*
1994	 *	Set the ARP module up
1995	 */
1996
1997	arp_init();
1998
1999	/*
2000	 *	Set the IP module up
2001	 */
2002
2003	ip_init();
2004
 
 
 
 
2005	/* Setup TCP slab cache for open requests. */
2006	tcp_init();
2007
2008	/* Setup UDP memory threshold */
2009	udp_init();
2010
2011	/* Add UDP-Lite (RFC 3828) */
2012	udplite4_register();
2013
2014	raw_init();
2015
2016	ping_init();
2017
2018	/*
2019	 *	Set the ICMP layer up
2020	 */
2021
2022	if (icmp_init() < 0)
2023		panic("Failed to create the ICMP control socket.\n");
2024
2025	/*
2026	 *	Initialise the multicast router
2027	 */
2028#if defined(CONFIG_IP_MROUTE)
2029	if (ip_mr_init())
2030		pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2031#endif
2032
2033	if (init_inet_pernet_ops())
2034		pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2035	/*
2036	 *	Initialise per-cpu ipv4 mibs
2037	 */
2038
2039	if (init_ipv4_mibs())
2040		pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
2041
2042	ipv4_proc_init();
2043
2044	ipfrag_init();
2045
2046	dev_add_pack(&ip_packet_type);
2047
2048	ip_tunnel_core_init();
2049
2050	rc = 0;
2051out:
2052	return rc;
2053out_unregister_raw_proto:
2054	proto_unregister(&raw_prot);
2055out_unregister_udp_proto:
2056	proto_unregister(&udp_prot);
2057out_unregister_tcp_proto:
2058	proto_unregister(&tcp_prot);
2059	goto out;
2060}
2061
2062fs_initcall(inet_init);
2063
2064/* ------------------------------------------------------------------------ */
2065
2066#ifdef CONFIG_PROC_FS
2067static int __init ipv4_proc_init(void)
2068{
2069	int rc = 0;
2070
2071	if (raw_proc_init())
2072		goto out_raw;
2073	if (tcp4_proc_init())
2074		goto out_tcp;
2075	if (udp4_proc_init())
2076		goto out_udp;
2077	if (ping_proc_init())
2078		goto out_ping;
2079	if (ip_misc_proc_init())
2080		goto out_misc;
2081out:
2082	return rc;
2083out_misc:
2084	ping_proc_exit();
2085out_ping:
2086	udp4_proc_exit();
2087out_udp:
2088	tcp4_proc_exit();
2089out_tcp:
2090	raw_proc_exit();
2091out_raw:
2092	rc = -ENOMEM;
2093	goto out;
2094}
2095
2096#else /* CONFIG_PROC_FS */
2097static int __init ipv4_proc_init(void)
2098{
2099	return 0;
2100}
2101#endif /* CONFIG_PROC_FS */