xref: /kernel/linux/linux-5.10/net/ipv4/af_inet.c (revision 8c2ecf20)
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#ifdef CONFIG_LOWPOWER_PROTOCOL
121#include <net/lowpower_protocol.h>
122#endif /* CONFIG_LOWPOWER_PROTOCOL */
123
124#include <trace/events/sock.h>
125
126/* The inetsw table contains everything that inet_create needs to
127 * build a new socket.
128 */
129static struct list_head inetsw[SOCK_MAX];
130static DEFINE_SPINLOCK(inetsw_lock);
131
132/* New destruction routine */
133
134void inet_sock_destruct(struct sock *sk)
135{
136	struct inet_sock *inet = inet_sk(sk);
137
138	__skb_queue_purge(&sk->sk_receive_queue);
139	if (sk->sk_rx_skb_cache) {
140		__kfree_skb(sk->sk_rx_skb_cache);
141		sk->sk_rx_skb_cache = NULL;
142	}
143	__skb_queue_purge(&sk->sk_error_queue);
144
145	sk_mem_reclaim(sk);
146
147	if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
148		pr_err("Attempt to release TCP socket in state %d %p\n",
149		       sk->sk_state, sk);
150		return;
151	}
152	if (!sock_flag(sk, SOCK_DEAD)) {
153		pr_err("Attempt to release alive inet socket %p\n", sk);
154		return;
155	}
156
157	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
158	WARN_ON(refcount_read(&sk->sk_wmem_alloc));
159	WARN_ON(sk->sk_wmem_queued);
160	WARN_ON(sk->sk_forward_alloc);
161
162	kfree(rcu_dereference_protected(inet->inet_opt, 1));
163	dst_release(rcu_dereference_protected(sk->sk_dst_cache, 1));
164	dst_release(rcu_dereference_protected(sk->sk_rx_dst, 1));
165	sk_refcnt_debug_dec(sk);
166}
167EXPORT_SYMBOL(inet_sock_destruct);
168
169/*
170 *	The routines beyond this point handle the behaviour of an AF_INET
171 *	socket object. Mostly it punts to the subprotocols of IP to do
172 *	the work.
173 */
174
175/*
176 *	Automatically bind an unbound socket.
177 */
178
179static int inet_autobind(struct sock *sk)
180{
181	struct inet_sock *inet;
182	/* We may need to bind the socket. */
183	lock_sock(sk);
184	inet = inet_sk(sk);
185	if (!inet->inet_num) {
186		if (sk->sk_prot->get_port(sk, 0)) {
187			release_sock(sk);
188			return -EAGAIN;
189		}
190		inet->inet_sport = htons(inet->inet_num);
191	}
192	release_sock(sk);
193	return 0;
194}
195
196/*
197 *	Move a socket into listening state.
198 */
199int inet_listen(struct socket *sock, int backlog)
200{
201	struct sock *sk = sock->sk;
202	unsigned char old_state;
203	int err, tcp_fastopen;
204
205	lock_sock(sk);
206
207	err = -EINVAL;
208	if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
209		goto out;
210
211	old_state = sk->sk_state;
212	if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
213		goto out;
214
215	WRITE_ONCE(sk->sk_max_ack_backlog, backlog);
216	/* Really, if the socket is already in listen state
217	 * we can only allow the backlog to be adjusted.
218	 */
219	if (old_state != TCP_LISTEN) {
220		/* Enable TFO w/o requiring TCP_FASTOPEN socket option.
221		 * Note that only TCP sockets (SOCK_STREAM) will reach here.
222		 * Also fastopen backlog may already been set via the option
223		 * because the socket was in TCP_LISTEN state previously but
224		 * was shutdown() rather than close().
225		 */
226		tcp_fastopen = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fastopen);
227		if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
228		    (tcp_fastopen & TFO_SERVER_ENABLE) &&
229		    !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
230			fastopen_queue_tune(sk, backlog);
231			tcp_fastopen_init_key_once(sock_net(sk));
232		}
233
234		err = inet_csk_listen_start(sk, backlog);
235		if (err)
236			goto out;
237		tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL);
238	}
239	err = 0;
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 = -ENOBUFS;
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->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
338
339	inet->nodefrag = 0;
340
341	if (SOCK_RAW == sock->type) {
342		inet->inet_num = protocol;
343		if (IPPROTO_RAW == protocol)
344			inet->hdrincl = 1;
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	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
360	inet->uc_ttl	= -1;
361	inet->mc_loop	= 1;
362	inet->mc_ttl	= 1;
363	inet->mc_all	= 1;
364	inet->mc_index	= 0;
365	inet->mc_list	= NULL;
366	inet->rcv_tos	= 0;
367
368	sk_refcnt_debug_inc(sk);
369
370	if (inet->inet_num) {
371		/* It assumes that any protocol which allows
372		 * the user to assign a number at socket
373		 * creation time automatically
374		 * shares.
375		 */
376		inet->inet_sport = htons(inet->inet_num);
377		/* Add to protocol hash chains. */
378		err = sk->sk_prot->hash(sk);
379		if (err) {
380			sk_common_release(sk);
381			goto out;
382		}
383	}
384
385	if (sk->sk_prot->init) {
386		err = sk->sk_prot->init(sk);
387		if (err) {
388			sk_common_release(sk);
389			goto out;
390		}
391	}
392
393	if (!kern) {
394		err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
395		if (err) {
396			sk_common_release(sk);
397			goto out;
398		}
399	}
400out:
401	return err;
402out_rcu_unlock:
403	rcu_read_unlock();
404	goto out;
405}
406
407
408/*
409 *	The peer socket should always be NULL (or else). When we call this
410 *	function we are destroying the object and from then on nobody
411 *	should refer to it.
412 */
413int inet_release(struct socket *sock)
414{
415	struct sock *sk = sock->sk;
416
417	if (sk) {
418		long timeout;
419
420		if (!sk->sk_kern_sock)
421			BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk);
422
423		/* Applications forget to leave groups before exiting */
424		ip_mc_drop_socket(sk);
425
426		/* If linger is set, we don't return until the close
427		 * is complete.  Otherwise we return immediately. The
428		 * actually closing is done the same either way.
429		 *
430		 * If the close is due to the process exiting, we never
431		 * linger..
432		 */
433		timeout = 0;
434		if (sock_flag(sk, SOCK_LINGER) &&
435		    !(current->flags & PF_EXITING))
436			timeout = sk->sk_lingertime;
437		sk->sk_prot->close(sk, timeout);
438		sock->sk = NULL;
439	}
440	return 0;
441}
442EXPORT_SYMBOL(inet_release);
443
444int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
445{
446	struct sock *sk = sock->sk;
447	int err;
448
449	/* If the socket has its own bind function then use it. (RAW) */
450	if (sk->sk_prot->bind) {
451		return sk->sk_prot->bind(sk, uaddr, addr_len);
452	}
453	if (addr_len < sizeof(struct sockaddr_in))
454		return -EINVAL;
455
456	/* BPF prog is run before any checks are done so that if the prog
457	 * changes context in a wrong way it will be caught.
458	 */
459	err = BPF_CGROUP_RUN_PROG_INET4_BIND(sk, uaddr);
460	if (err)
461		return err;
462
463	return __inet_bind(sk, uaddr, addr_len, BIND_WITH_LOCK);
464}
465EXPORT_SYMBOL(inet_bind);
466
467int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
468		u32 flags)
469{
470	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
471	struct inet_sock *inet = inet_sk(sk);
472	struct net *net = sock_net(sk);
473	unsigned short snum;
474	int chk_addr_ret;
475	u32 tb_id = RT_TABLE_LOCAL;
476	int err;
477
478	if (addr->sin_family != AF_INET) {
479		/* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
480		 * only if s_addr is INADDR_ANY.
481		 */
482		err = -EAFNOSUPPORT;
483		if (addr->sin_family != AF_UNSPEC ||
484		    addr->sin_addr.s_addr != htonl(INADDR_ANY))
485			goto out;
486	}
487
488	tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
489	chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
490
491	/* Not specified by any standard per-se, however it breaks too
492	 * many applications when removed.  It is unfortunate since
493	 * allowing applications to make a non-local bind solves
494	 * several problems with systems using dynamic addressing.
495	 * (ie. your servers still start up even if your ISDN link
496	 *  is temporarily down)
497	 */
498	err = -EADDRNOTAVAIL;
499	if (!inet_can_nonlocal_bind(net, inet) &&
500	    addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
501	    chk_addr_ret != RTN_LOCAL &&
502	    chk_addr_ret != RTN_MULTICAST &&
503	    chk_addr_ret != RTN_BROADCAST)
504		goto out;
505
506	snum = ntohs(addr->sin_port);
507	err = -EACCES;
508	if (snum && inet_port_requires_bind_service(net, snum) &&
509	    !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
510		goto out;
511
512	/*      We keep a pair of addresses. rcv_saddr is the one
513	 *      used by hash lookups, and saddr is used for transmit.
514	 *
515	 *      In the BSD API these are the same except where it
516	 *      would be illegal to use them (multicast/broadcast) in
517	 *      which case the sending device address is used.
518	 */
519	if (flags & BIND_WITH_LOCK)
520		lock_sock(sk);
521
522	/* Check these errors (active socket, double bind). */
523	err = -EINVAL;
524	if (sk->sk_state != TCP_CLOSE || inet->inet_num)
525		goto out_release_sock;
526
527	inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
528	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
529		inet->inet_saddr = 0;  /* Use device */
530
531	/* Make sure we are allowed to bind here. */
532	if (snum || !(inet->bind_address_no_port ||
533		      (flags & BIND_FORCE_ADDRESS_NO_PORT))) {
534		if (sk->sk_prot->get_port(sk, snum)) {
535			inet->inet_saddr = inet->inet_rcv_saddr = 0;
536			err = -EADDRINUSE;
537			goto out_release_sock;
538		}
539		if (!(flags & BIND_FROM_BPF)) {
540			err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
541			if (err) {
542				inet->inet_saddr = inet->inet_rcv_saddr = 0;
543				goto out_release_sock;
544			}
545		}
546	}
547
548	if (inet->inet_rcv_saddr)
549		sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
550	if (snum)
551		sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
552	inet->inet_sport = htons(inet->inet_num);
553	inet->inet_daddr = 0;
554	inet->inet_dport = 0;
555	sk_dst_reset(sk);
556	err = 0;
557out_release_sock:
558	if (flags & BIND_WITH_LOCK)
559		release_sock(sk);
560out:
561	return err;
562}
563
564int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
565		       int addr_len, int flags)
566{
567	struct sock *sk = sock->sk;
568	const struct proto *prot;
569	int err;
570
571	if (addr_len < sizeof(uaddr->sa_family))
572		return -EINVAL;
573
574	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
575	prot = READ_ONCE(sk->sk_prot);
576
577	if (uaddr->sa_family == AF_UNSPEC)
578		return prot->disconnect(sk, flags);
579
580	if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
581		err = prot->pre_connect(sk, uaddr, addr_len);
582		if (err)
583			return err;
584	}
585
586	if (data_race(!inet_sk(sk)->inet_num) && inet_autobind(sk))
587		return -EAGAIN;
588	return prot->connect(sk, uaddr, addr_len);
589}
590EXPORT_SYMBOL(inet_dgram_connect);
591
592static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
593{
594	DEFINE_WAIT_FUNC(wait, woken_wake_function);
595
596	add_wait_queue(sk_sleep(sk), &wait);
597	sk->sk_write_pending += writebias;
598	sk->sk_wait_pending++;
599
600	/* Basic assumption: if someone sets sk->sk_err, he _must_
601	 * change state of the socket from TCP_SYN_*.
602	 * Connect() does not allow to get error notifications
603	 * without closing the socket.
604	 */
605	while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
606		release_sock(sk);
607		timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
608		lock_sock(sk);
609		if (signal_pending(current) || !timeo)
610			break;
611	}
612	remove_wait_queue(sk_sleep(sk), &wait);
613	sk->sk_write_pending -= writebias;
614	sk->sk_wait_pending--;
615	return timeo;
616}
617
618/*
619 *	Connect to a remote host. There is regrettably still a little
620 *	TCP 'magic' in here.
621 */
622int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
623			  int addr_len, int flags, int is_sendmsg)
624{
625	struct sock *sk = sock->sk;
626	int err;
627	long timeo;
628
629	/*
630	 * uaddr can be NULL and addr_len can be 0 if:
631	 * sk is a TCP fastopen active socket and
632	 * TCP_FASTOPEN_CONNECT sockopt is set and
633	 * we already have a valid cookie for this socket.
634	 * In this case, user can call write() after connect().
635	 * write() will invoke tcp_sendmsg_fastopen() which calls
636	 * __inet_stream_connect().
637	 */
638	if (uaddr) {
639		if (addr_len < sizeof(uaddr->sa_family))
640			return -EINVAL;
641
642		if (uaddr->sa_family == AF_UNSPEC) {
643			err = sk->sk_prot->disconnect(sk, flags);
644			sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
645			goto out;
646		}
647	}
648
649	switch (sock->state) {
650	default:
651		err = -EINVAL;
652		goto out;
653	case SS_CONNECTED:
654		err = -EISCONN;
655		goto out;
656	case SS_CONNECTING:
657		if (inet_sk(sk)->defer_connect)
658			err = is_sendmsg ? -EINPROGRESS : -EISCONN;
659		else
660			err = -EALREADY;
661		/* Fall out of switch with err, set for this state */
662		break;
663	case SS_UNCONNECTED:
664		err = -EISCONN;
665		if (sk->sk_state != TCP_CLOSE)
666			goto out;
667
668		if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
669			err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
670			if (err)
671				goto out;
672		}
673
674		err = sk->sk_prot->connect(sk, uaddr, addr_len);
675		if (err < 0)
676			goto out;
677
678		sock->state = SS_CONNECTING;
679
680		if (!err && inet_sk(sk)->defer_connect)
681			goto out;
682
683		/* Just entered SS_CONNECTING state; the only
684		 * difference is that return value in non-blocking
685		 * case is EINPROGRESS, rather than EALREADY.
686		 */
687		err = -EINPROGRESS;
688		break;
689	}
690
691	timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
692
693	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
694		int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
695				tcp_sk(sk)->fastopen_req &&
696				tcp_sk(sk)->fastopen_req->data ? 1 : 0;
697
698		/* Error code is set above */
699		if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
700			goto out;
701
702		err = sock_intr_errno(timeo);
703		if (signal_pending(current))
704			goto out;
705	}
706
707	/* Connection was closed by RST, timeout, ICMP error
708	 * or another process disconnected us.
709	 */
710	if (sk->sk_state == TCP_CLOSE)
711		goto sock_error;
712
713	/* sk->sk_err may be not zero now, if RECVERR was ordered by user
714	 * and error was received after socket entered established state.
715	 * Hence, it is handled normally after connect() return successfully.
716	 */
717
718	sock->state = SS_CONNECTED;
719	err = 0;
720out:
721	return err;
722
723sock_error:
724	err = sock_error(sk) ? : -ECONNABORTED;
725	sock->state = SS_UNCONNECTED;
726	if (sk->sk_prot->disconnect(sk, flags))
727		sock->state = SS_DISCONNECTING;
728	goto out;
729}
730EXPORT_SYMBOL(__inet_stream_connect);
731
732int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
733			int addr_len, int flags)
734{
735	int err;
736
737	lock_sock(sock->sk);
738	err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
739	release_sock(sock->sk);
740	return err;
741}
742EXPORT_SYMBOL(inet_stream_connect);
743
744/*
745 *	Accept a pending connection. The TCP layer now gives BSD semantics.
746 */
747
748int inet_accept(struct socket *sock, struct socket *newsock, int flags,
749		bool kern)
750{
751	struct sock *sk1 = sock->sk, *sk2;
752	int err = -EINVAL;
753
754	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
755	sk2 = READ_ONCE(sk1->sk_prot)->accept(sk1, flags, &err, kern);
756	if (!sk2)
757		goto do_err;
758
759	lock_sock(sk2);
760
761	sock_rps_record_flow(sk2);
762	WARN_ON(!((1 << sk2->sk_state) &
763		  (TCPF_ESTABLISHED | TCPF_SYN_RECV |
764		   TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2 |
765		   TCPF_CLOSING | TCPF_CLOSE_WAIT |
766		   TCPF_CLOSE)));
767
768	sock_graft(sk2, newsock);
769
770	newsock->state = SS_CONNECTED;
771	err = 0;
772	release_sock(sk2);
773do_err:
774	return err;
775}
776EXPORT_SYMBOL(inet_accept);
777
778/*
779 *	This does both peername and sockname.
780 */
781int inet_getname(struct socket *sock, struct sockaddr *uaddr,
782		 int peer)
783{
784	struct sock *sk		= sock->sk;
785	struct inet_sock *inet	= inet_sk(sk);
786	DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
787
788	sin->sin_family = AF_INET;
789	if (peer) {
790		if (!inet->inet_dport ||
791		    (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
792		     peer == 1))
793			return -ENOTCONN;
794		sin->sin_port = inet->inet_dport;
795		sin->sin_addr.s_addr = inet->inet_daddr;
796	} else {
797		__be32 addr = inet->inet_rcv_saddr;
798		if (!addr)
799			addr = inet->inet_saddr;
800		sin->sin_port = inet->inet_sport;
801		sin->sin_addr.s_addr = addr;
802	}
803	if (cgroup_bpf_enabled)
804		BPF_CGROUP_RUN_SA_PROG_LOCK(sk, (struct sockaddr *)sin,
805					    peer ? BPF_CGROUP_INET4_GETPEERNAME :
806						   BPF_CGROUP_INET4_GETSOCKNAME,
807					    NULL);
808	memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
809	return sizeof(*sin);
810}
811EXPORT_SYMBOL(inet_getname);
812
813int inet_send_prepare(struct sock *sk)
814{
815	sock_rps_record_flow(sk);
816
817	/* We may need to bind the socket. */
818	if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind &&
819	    inet_autobind(sk))
820		return -EAGAIN;
821
822	return 0;
823}
824EXPORT_SYMBOL_GPL(inet_send_prepare);
825
826int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
827{
828	struct sock *sk = sock->sk;
829
830	if (unlikely(inet_send_prepare(sk)))
831		return -EAGAIN;
832
833	return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
834			       sk, msg, size);
835}
836EXPORT_SYMBOL(inet_sendmsg);
837
838ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
839		      size_t size, int flags)
840{
841	struct sock *sk = sock->sk;
842	const struct proto *prot;
843
844	if (unlikely(inet_send_prepare(sk)))
845		return -EAGAIN;
846
847	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
848	prot = READ_ONCE(sk->sk_prot);
849	if (prot->sendpage)
850		return prot->sendpage(sk, page, offset, size, flags);
851	return sock_no_sendpage(sock, page, offset, size, flags);
852}
853EXPORT_SYMBOL(inet_sendpage);
854
855INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
856					  size_t, int, int, int *));
857int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
858		 int flags)
859{
860	struct sock *sk = sock->sk;
861	int addr_len = 0;
862	int err;
863
864	if (likely(!(flags & MSG_ERRQUEUE)))
865		sock_rps_record_flow(sk);
866
867	err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
868			      sk, msg, size, flags & MSG_DONTWAIT,
869			      flags & ~MSG_DONTWAIT, &addr_len);
870	if (err >= 0)
871		msg->msg_namelen = addr_len;
872	return err;
873}
874EXPORT_SYMBOL(inet_recvmsg);
875
876int inet_shutdown(struct socket *sock, int how)
877{
878	struct sock *sk = sock->sk;
879	int err = 0;
880
881	/* This should really check to make sure
882	 * the socket is a TCP socket. (WHY AC...)
883	 */
884	how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
885		       1->2 bit 2 snds.
886		       2->3 */
887	if ((how & ~SHUTDOWN_MASK) || !how)	/* MAXINT->0 */
888		return -EINVAL;
889
890	lock_sock(sk);
891	if (sock->state == SS_CONNECTING) {
892		if ((1 << sk->sk_state) &
893		    (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
894			sock->state = SS_DISCONNECTING;
895		else
896			sock->state = SS_CONNECTED;
897	}
898
899	switch (sk->sk_state) {
900	case TCP_CLOSE:
901		err = -ENOTCONN;
902		/* Hack to wake up other listeners, who can poll for
903		   EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
904		fallthrough;
905	default:
906		WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | how);
907		if (sk->sk_prot->shutdown)
908			sk->sk_prot->shutdown(sk, how);
909		break;
910
911	/* Remaining two branches are temporary solution for missing
912	 * close() in multithreaded environment. It is _not_ a good idea,
913	 * but we have no choice until close() is repaired at VFS level.
914	 */
915	case TCP_LISTEN:
916		if (!(how & RCV_SHUTDOWN))
917			break;
918		fallthrough;
919	case TCP_SYN_SENT:
920		err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
921		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
922		break;
923	}
924
925	/* Wake up anyone sleeping in poll. */
926	sk->sk_state_change(sk);
927	release_sock(sk);
928	return err;
929}
930EXPORT_SYMBOL(inet_shutdown);
931
932/*
933 *	ioctl() calls you can issue on an INET socket. Most of these are
934 *	device configuration and stuff and very rarely used. Some ioctls
935 *	pass on to the socket itself.
936 *
937 *	NOTE: I like the idea of a module for the config stuff. ie ifconfig
938 *	loads the devconfigure module does its configuring and unloads it.
939 *	There's a good 20K of config code hanging around the kernel.
940 */
941
942int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
943{
944	struct sock *sk = sock->sk;
945	int err = 0;
946	struct net *net = sock_net(sk);
947	void __user *p = (void __user *)arg;
948	struct ifreq ifr;
949	struct rtentry rt;
950
951	switch (cmd) {
952	case SIOCADDRT:
953	case SIOCDELRT:
954		if (copy_from_user(&rt, p, sizeof(struct rtentry)))
955			return -EFAULT;
956		err = ip_rt_ioctl(net, cmd, &rt);
957		break;
958	case SIOCRTMSG:
959		err = -EINVAL;
960		break;
961	case SIOCDARP:
962	case SIOCGARP:
963	case SIOCSARP:
964		err = arp_ioctl(net, cmd, (void __user *)arg);
965		break;
966	case SIOCGIFADDR:
967	case SIOCGIFBRDADDR:
968	case SIOCGIFNETMASK:
969	case SIOCGIFDSTADDR:
970	case SIOCGIFPFLAGS:
971		if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
972			return -EFAULT;
973		err = devinet_ioctl(net, cmd, &ifr);
974		if (!err && copy_to_user(p, &ifr, sizeof(struct ifreq)))
975			err = -EFAULT;
976		break;
977
978	case SIOCSIFADDR:
979	case SIOCSIFBRDADDR:
980	case SIOCSIFNETMASK:
981	case SIOCSIFDSTADDR:
982	case SIOCSIFPFLAGS:
983	case SIOCSIFFLAGS:
984		if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
985			return -EFAULT;
986		err = devinet_ioctl(net, cmd, &ifr);
987		break;
988	default:
989		if (sk->sk_prot->ioctl)
990			err = sk->sk_prot->ioctl(sk, cmd, arg);
991		else
992			err = -ENOIOCTLCMD;
993		break;
994	}
995	return err;
996}
997EXPORT_SYMBOL(inet_ioctl);
998
999#ifdef CONFIG_COMPAT
1000static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd,
1001		struct compat_rtentry __user *ur)
1002{
1003	compat_uptr_t rtdev;
1004	struct rtentry rt;
1005
1006	if (copy_from_user(&rt.rt_dst, &ur->rt_dst,
1007			3 * sizeof(struct sockaddr)) ||
1008	    get_user(rt.rt_flags, &ur->rt_flags) ||
1009	    get_user(rt.rt_metric, &ur->rt_metric) ||
1010	    get_user(rt.rt_mtu, &ur->rt_mtu) ||
1011	    get_user(rt.rt_window, &ur->rt_window) ||
1012	    get_user(rt.rt_irtt, &ur->rt_irtt) ||
1013	    get_user(rtdev, &ur->rt_dev))
1014		return -EFAULT;
1015
1016	rt.rt_dev = compat_ptr(rtdev);
1017	return ip_rt_ioctl(sock_net(sk), cmd, &rt);
1018}
1019
1020static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1021{
1022	void __user *argp = compat_ptr(arg);
1023	struct sock *sk = sock->sk;
1024
1025	switch (cmd) {
1026	case SIOCADDRT:
1027	case SIOCDELRT:
1028		return inet_compat_routing_ioctl(sk, cmd, argp);
1029	default:
1030		if (!sk->sk_prot->compat_ioctl)
1031			return -ENOIOCTLCMD;
1032		return sk->sk_prot->compat_ioctl(sk, cmd, arg);
1033	}
1034}
1035#endif /* CONFIG_COMPAT */
1036
1037const struct proto_ops inet_stream_ops = {
1038	.family		   = PF_INET,
1039	.owner		   = THIS_MODULE,
1040	.release	   = inet_release,
1041	.bind		   = inet_bind,
1042	.connect	   = inet_stream_connect,
1043	.socketpair	   = sock_no_socketpair,
1044	.accept		   = inet_accept,
1045	.getname	   = inet_getname,
1046	.poll		   = tcp_poll,
1047	.ioctl		   = inet_ioctl,
1048	.gettstamp	   = sock_gettstamp,
1049	.listen		   = inet_listen,
1050	.shutdown	   = inet_shutdown,
1051	.setsockopt	   = sock_common_setsockopt,
1052	.getsockopt	   = sock_common_getsockopt,
1053	.sendmsg	   = inet_sendmsg,
1054	.recvmsg	   = inet_recvmsg,
1055#ifdef CONFIG_MMU
1056	.mmap		   = tcp_mmap,
1057#endif
1058	.sendpage	   = inet_sendpage,
1059	.splice_read	   = tcp_splice_read,
1060	.read_sock	   = tcp_read_sock,
1061	.sendmsg_locked    = tcp_sendmsg_locked,
1062	.sendpage_locked   = tcp_sendpage_locked,
1063	.peek_len	   = tcp_peek_len,
1064#ifdef CONFIG_COMPAT
1065	.compat_ioctl	   = inet_compat_ioctl,
1066#endif
1067	.set_rcvlowat	   = tcp_set_rcvlowat,
1068};
1069EXPORT_SYMBOL(inet_stream_ops);
1070
1071const struct proto_ops inet_dgram_ops = {
1072	.family		   = PF_INET,
1073	.owner		   = THIS_MODULE,
1074	.release	   = inet_release,
1075	.bind		   = inet_bind,
1076	.connect	   = inet_dgram_connect,
1077	.socketpair	   = sock_no_socketpair,
1078	.accept		   = sock_no_accept,
1079	.getname	   = inet_getname,
1080	.poll		   = udp_poll,
1081	.ioctl		   = inet_ioctl,
1082	.gettstamp	   = sock_gettstamp,
1083	.listen		   = sock_no_listen,
1084	.shutdown	   = inet_shutdown,
1085	.setsockopt	   = sock_common_setsockopt,
1086	.getsockopt	   = sock_common_getsockopt,
1087	.sendmsg	   = inet_sendmsg,
1088	.recvmsg	   = inet_recvmsg,
1089	.mmap		   = sock_no_mmap,
1090	.sendpage	   = inet_sendpage,
1091	.set_peek_off	   = sk_set_peek_off,
1092#ifdef CONFIG_COMPAT
1093	.compat_ioctl	   = inet_compat_ioctl,
1094#endif
1095};
1096EXPORT_SYMBOL(inet_dgram_ops);
1097
1098/*
1099 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1100 * udp_poll
1101 */
1102static const struct proto_ops inet_sockraw_ops = {
1103	.family		   = PF_INET,
1104	.owner		   = THIS_MODULE,
1105	.release	   = inet_release,
1106	.bind		   = inet_bind,
1107	.connect	   = inet_dgram_connect,
1108	.socketpair	   = sock_no_socketpair,
1109	.accept		   = sock_no_accept,
1110	.getname	   = inet_getname,
1111	.poll		   = datagram_poll,
1112	.ioctl		   = inet_ioctl,
1113	.gettstamp	   = sock_gettstamp,
1114	.listen		   = sock_no_listen,
1115	.shutdown	   = inet_shutdown,
1116	.setsockopt	   = sock_common_setsockopt,
1117	.getsockopt	   = sock_common_getsockopt,
1118	.sendmsg	   = inet_sendmsg,
1119	.recvmsg	   = inet_recvmsg,
1120	.mmap		   = sock_no_mmap,
1121	.sendpage	   = inet_sendpage,
1122#ifdef CONFIG_COMPAT
1123	.compat_ioctl	   = inet_compat_ioctl,
1124#endif
1125};
1126
1127static const struct net_proto_family inet_family_ops = {
1128	.family = PF_INET,
1129	.create = inet_create,
1130	.owner	= THIS_MODULE,
1131};
1132
1133/* Upon startup we insert all the elements in inetsw_array[] into
1134 * the linked list inetsw.
1135 */
1136static struct inet_protosw inetsw_array[] =
1137{
1138	{
1139		.type =       SOCK_STREAM,
1140		.protocol =   IPPROTO_TCP,
1141		.prot =       &tcp_prot,
1142		.ops =        &inet_stream_ops,
1143		.flags =      INET_PROTOSW_PERMANENT |
1144			      INET_PROTOSW_ICSK,
1145	},
1146
1147	{
1148		.type =       SOCK_DGRAM,
1149		.protocol =   IPPROTO_UDP,
1150		.prot =       &udp_prot,
1151		.ops =        &inet_dgram_ops,
1152		.flags =      INET_PROTOSW_PERMANENT,
1153       },
1154
1155       {
1156		.type =       SOCK_DGRAM,
1157		.protocol =   IPPROTO_ICMP,
1158		.prot =       &ping_prot,
1159		.ops =        &inet_sockraw_ops,
1160		.flags =      INET_PROTOSW_REUSE,
1161       },
1162
1163       {
1164	       .type =       SOCK_RAW,
1165	       .protocol =   IPPROTO_IP,	/* wild card */
1166	       .prot =       &raw_prot,
1167	       .ops =        &inet_sockraw_ops,
1168	       .flags =      INET_PROTOSW_REUSE,
1169       }
1170};
1171
1172#define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1173
1174void inet_register_protosw(struct inet_protosw *p)
1175{
1176	struct list_head *lh;
1177	struct inet_protosw *answer;
1178	int protocol = p->protocol;
1179	struct list_head *last_perm;
1180
1181	spin_lock_bh(&inetsw_lock);
1182
1183	if (p->type >= SOCK_MAX)
1184		goto out_illegal;
1185
1186	/* If we are trying to override a permanent protocol, bail. */
1187	last_perm = &inetsw[p->type];
1188	list_for_each(lh, &inetsw[p->type]) {
1189		answer = list_entry(lh, struct inet_protosw, list);
1190		/* Check only the non-wild match. */
1191		if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1192			break;
1193		if (protocol == answer->protocol)
1194			goto out_permanent;
1195		last_perm = lh;
1196	}
1197
1198	/* Add the new entry after the last permanent entry if any, so that
1199	 * the new entry does not override a permanent entry when matched with
1200	 * a wild-card protocol. But it is allowed to override any existing
1201	 * non-permanent entry.  This means that when we remove this entry, the
1202	 * system automatically returns to the old behavior.
1203	 */
1204	list_add_rcu(&p->list, last_perm);
1205out:
1206	spin_unlock_bh(&inetsw_lock);
1207
1208	return;
1209
1210out_permanent:
1211	pr_err("Attempt to override permanent protocol %d\n", protocol);
1212	goto out;
1213
1214out_illegal:
1215	pr_err("Ignoring attempt to register invalid socket type %d\n",
1216	       p->type);
1217	goto out;
1218}
1219EXPORT_SYMBOL(inet_register_protosw);
1220
1221void inet_unregister_protosw(struct inet_protosw *p)
1222{
1223	if (INET_PROTOSW_PERMANENT & p->flags) {
1224		pr_err("Attempt to unregister permanent protocol %d\n",
1225		       p->protocol);
1226	} else {
1227		spin_lock_bh(&inetsw_lock);
1228		list_del_rcu(&p->list);
1229		spin_unlock_bh(&inetsw_lock);
1230
1231		synchronize_net();
1232	}
1233}
1234EXPORT_SYMBOL(inet_unregister_protosw);
1235
1236static int inet_sk_reselect_saddr(struct sock *sk)
1237{
1238	struct inet_sock *inet = inet_sk(sk);
1239	__be32 old_saddr = inet->inet_saddr;
1240	__be32 daddr = inet->inet_daddr;
1241	struct flowi4 *fl4;
1242	struct rtable *rt;
1243	__be32 new_saddr;
1244	struct ip_options_rcu *inet_opt;
1245
1246	inet_opt = rcu_dereference_protected(inet->inet_opt,
1247					     lockdep_sock_is_held(sk));
1248	if (inet_opt && inet_opt->opt.srr)
1249		daddr = inet_opt->opt.faddr;
1250
1251	/* Query new route. */
1252	fl4 = &inet->cork.fl.u.ip4;
1253	rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1254			      sk->sk_bound_dev_if, sk->sk_protocol,
1255			      inet->inet_sport, inet->inet_dport, sk);
1256	if (IS_ERR(rt))
1257		return PTR_ERR(rt);
1258
1259	sk_setup_caps(sk, &rt->dst);
1260
1261	new_saddr = fl4->saddr;
1262
1263	if (new_saddr == old_saddr)
1264		return 0;
1265
1266	if (READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) > 1) {
1267		pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1268			__func__, &old_saddr, &new_saddr);
1269	}
1270
1271	inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1272
1273	/*
1274	 * XXX The only one ugly spot where we need to
1275	 * XXX really change the sockets identity after
1276	 * XXX it has entered the hashes. -DaveM
1277	 *
1278	 * Besides that, it does not check for connection
1279	 * uniqueness. Wait for troubles.
1280	 */
1281	return __sk_prot_rehash(sk);
1282}
1283
1284int inet_sk_rebuild_header(struct sock *sk)
1285{
1286	struct inet_sock *inet = inet_sk(sk);
1287	struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1288	__be32 daddr;
1289	struct ip_options_rcu *inet_opt;
1290	struct flowi4 *fl4;
1291	int err;
1292
1293	/* Route is OK, nothing to do. */
1294	if (rt)
1295		return 0;
1296
1297	/* Reroute. */
1298	rcu_read_lock();
1299	inet_opt = rcu_dereference(inet->inet_opt);
1300	daddr = inet->inet_daddr;
1301	if (inet_opt && inet_opt->opt.srr)
1302		daddr = inet_opt->opt.faddr;
1303	rcu_read_unlock();
1304	fl4 = &inet->cork.fl.u.ip4;
1305	rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1306				   inet->inet_dport, inet->inet_sport,
1307				   sk->sk_protocol, RT_CONN_FLAGS(sk),
1308				   sk->sk_bound_dev_if);
1309	if (!IS_ERR(rt)) {
1310		err = 0;
1311		sk_setup_caps(sk, &rt->dst);
1312	} else {
1313		err = PTR_ERR(rt);
1314
1315		/* Routing failed... */
1316		sk->sk_route_caps = 0;
1317		/*
1318		 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1319		 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1320		 */
1321		if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) ||
1322		    sk->sk_state != TCP_SYN_SENT ||
1323		    (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1324		    (err = inet_sk_reselect_saddr(sk)) != 0)
1325			sk->sk_err_soft = -err;
1326	}
1327
1328	return err;
1329}
1330EXPORT_SYMBOL(inet_sk_rebuild_header);
1331
1332void inet_sk_set_state(struct sock *sk, int state)
1333{
1334	trace_inet_sock_set_state(sk, sk->sk_state, state);
1335	sk->sk_state = state;
1336}
1337EXPORT_SYMBOL(inet_sk_set_state);
1338
1339void inet_sk_state_store(struct sock *sk, int newstate)
1340{
1341	trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1342	smp_store_release(&sk->sk_state, newstate);
1343}
1344
1345struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1346				 netdev_features_t features)
1347{
1348	bool udpfrag = false, fixedid = false, gso_partial, encap;
1349	struct sk_buff *segs = ERR_PTR(-EINVAL);
1350	const struct net_offload *ops;
1351	unsigned int offset = 0;
1352	struct iphdr *iph;
1353	int proto, tot_len;
1354	int nhoff;
1355	int ihl;
1356	int id;
1357
1358	skb_reset_network_header(skb);
1359	nhoff = skb_network_header(skb) - skb_mac_header(skb);
1360	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1361		goto out;
1362
1363	iph = ip_hdr(skb);
1364	ihl = iph->ihl * 4;
1365	if (ihl < sizeof(*iph))
1366		goto out;
1367
1368	id = ntohs(iph->id);
1369	proto = iph->protocol;
1370
1371	/* Warning: after this point, iph might be no longer valid */
1372	if (unlikely(!pskb_may_pull(skb, ihl)))
1373		goto out;
1374	__skb_pull(skb, ihl);
1375
1376	encap = SKB_GSO_CB(skb)->encap_level > 0;
1377	if (encap)
1378		features &= skb->dev->hw_enc_features;
1379	SKB_GSO_CB(skb)->encap_level += ihl;
1380
1381	skb_reset_transport_header(skb);
1382
1383	segs = ERR_PTR(-EPROTONOSUPPORT);
1384
1385	if (!skb->encapsulation || encap) {
1386		udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1387		fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1388
1389		/* fixed ID is invalid if DF bit is not set */
1390		if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1391			goto out;
1392	}
1393
1394	ops = rcu_dereference(inet_offloads[proto]);
1395	if (likely(ops && ops->callbacks.gso_segment)) {
1396		segs = ops->callbacks.gso_segment(skb, features);
1397		if (!segs)
1398			skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
1399	}
1400
1401	if (IS_ERR_OR_NULL(segs))
1402		goto out;
1403
1404	gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1405
1406	skb = segs;
1407	do {
1408		iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1409		if (udpfrag) {
1410			iph->frag_off = htons(offset >> 3);
1411			if (skb->next)
1412				iph->frag_off |= htons(IP_MF);
1413			offset += skb->len - nhoff - ihl;
1414			tot_len = skb->len - nhoff;
1415		} else if (skb_is_gso(skb)) {
1416			if (!fixedid) {
1417				iph->id = htons(id);
1418				id += skb_shinfo(skb)->gso_segs;
1419			}
1420
1421			if (gso_partial)
1422				tot_len = skb_shinfo(skb)->gso_size +
1423					  SKB_GSO_CB(skb)->data_offset +
1424					  skb->head - (unsigned char *)iph;
1425			else
1426				tot_len = skb->len - nhoff;
1427		} else {
1428			if (!fixedid)
1429				iph->id = htons(id++);
1430			tot_len = skb->len - nhoff;
1431		}
1432		iph->tot_len = htons(tot_len);
1433		ip_send_check(iph);
1434		if (encap)
1435			skb_reset_inner_headers(skb);
1436		skb->network_header = (u8 *)iph - skb->head;
1437		skb_reset_mac_len(skb);
1438	} while ((skb = skb->next));
1439
1440out:
1441	return segs;
1442}
1443EXPORT_SYMBOL(inet_gso_segment);
1444
1445static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1446					netdev_features_t features)
1447{
1448	if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1449		return ERR_PTR(-EINVAL);
1450
1451	return inet_gso_segment(skb, features);
1452}
1453
1454struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1455{
1456	const struct net_offload *ops;
1457	struct sk_buff *pp = NULL;
1458	const struct iphdr *iph;
1459	struct sk_buff *p;
1460	unsigned int hlen;
1461	unsigned int off;
1462	unsigned int id;
1463	int flush = 1;
1464	int proto;
1465
1466	off = skb_gro_offset(skb);
1467	hlen = off + sizeof(*iph);
1468	iph = skb_gro_header_fast(skb, off);
1469	if (skb_gro_header_hard(skb, hlen)) {
1470		iph = skb_gro_header_slow(skb, hlen, off);
1471		if (unlikely(!iph))
1472			goto out;
1473	}
1474
1475	proto = iph->protocol;
1476
1477	rcu_read_lock();
1478	ops = rcu_dereference(inet_offloads[proto]);
1479	if (!ops || !ops->callbacks.gro_receive)
1480		goto out_unlock;
1481
1482	if (*(u8 *)iph != 0x45)
1483		goto out_unlock;
1484
1485	if (ip_is_fragment(iph))
1486		goto out_unlock;
1487
1488	if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1489		goto out_unlock;
1490
1491	id = ntohl(*(__be32 *)&iph->id);
1492	flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1493	id >>= 16;
1494
1495	list_for_each_entry(p, head, list) {
1496		struct iphdr *iph2;
1497		u16 flush_id;
1498
1499		if (!NAPI_GRO_CB(p)->same_flow)
1500			continue;
1501
1502		iph2 = (struct iphdr *)(p->data + off);
1503		/* The above works because, with the exception of the top
1504		 * (inner most) layer, we only aggregate pkts with the same
1505		 * hdr length so all the hdrs we'll need to verify will start
1506		 * at the same offset.
1507		 */
1508		if ((iph->protocol ^ iph2->protocol) |
1509		    ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1510		    ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1511			NAPI_GRO_CB(p)->same_flow = 0;
1512			continue;
1513		}
1514
1515		/* All fields must match except length and checksum. */
1516		NAPI_GRO_CB(p)->flush |=
1517			(iph->ttl ^ iph2->ttl) |
1518			(iph->tos ^ iph2->tos) |
1519			((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1520
1521		NAPI_GRO_CB(p)->flush |= flush;
1522
1523		/* We need to store of the IP ID check to be included later
1524		 * when we can verify that this packet does in fact belong
1525		 * to a given flow.
1526		 */
1527		flush_id = (u16)(id - ntohs(iph2->id));
1528
1529		/* This bit of code makes it much easier for us to identify
1530		 * the cases where we are doing atomic vs non-atomic IP ID
1531		 * checks.  Specifically an atomic check can return IP ID
1532		 * values 0 - 0xFFFF, while a non-atomic check can only
1533		 * return 0 or 0xFFFF.
1534		 */
1535		if (!NAPI_GRO_CB(p)->is_atomic ||
1536		    !(iph->frag_off & htons(IP_DF))) {
1537			flush_id ^= NAPI_GRO_CB(p)->count;
1538			flush_id = flush_id ? 0xFFFF : 0;
1539		}
1540
1541		/* If the previous IP ID value was based on an atomic
1542		 * datagram we can overwrite the value and ignore it.
1543		 */
1544		if (NAPI_GRO_CB(skb)->is_atomic)
1545			NAPI_GRO_CB(p)->flush_id = flush_id;
1546		else
1547			NAPI_GRO_CB(p)->flush_id |= flush_id;
1548	}
1549
1550	NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1551	NAPI_GRO_CB(skb)->flush |= flush;
1552	skb_set_network_header(skb, off);
1553	/* The above will be needed by the transport layer if there is one
1554	 * immediately following this IP hdr.
1555	 */
1556
1557	/* Note : No need to call skb_gro_postpull_rcsum() here,
1558	 * as we already checked checksum over ipv4 header was 0
1559	 */
1560	skb_gro_pull(skb, sizeof(*iph));
1561	skb_set_transport_header(skb, skb_gro_offset(skb));
1562
1563	pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1564				       ops->callbacks.gro_receive, head, skb);
1565
1566out_unlock:
1567	rcu_read_unlock();
1568
1569out:
1570	skb_gro_flush_final(skb, pp, flush);
1571
1572	return pp;
1573}
1574EXPORT_SYMBOL(inet_gro_receive);
1575
1576static struct sk_buff *ipip_gro_receive(struct list_head *head,
1577					struct sk_buff *skb)
1578{
1579	if (NAPI_GRO_CB(skb)->encap_mark) {
1580		NAPI_GRO_CB(skb)->flush = 1;
1581		return NULL;
1582	}
1583
1584	NAPI_GRO_CB(skb)->encap_mark = 1;
1585
1586	return inet_gro_receive(head, skb);
1587}
1588
1589#define SECONDS_PER_DAY	86400
1590
1591/* inet_current_timestamp - Return IP network timestamp
1592 *
1593 * Return milliseconds since midnight in network byte order.
1594 */
1595__be32 inet_current_timestamp(void)
1596{
1597	u32 secs;
1598	u32 msecs;
1599	struct timespec64 ts;
1600
1601	ktime_get_real_ts64(&ts);
1602
1603	/* Get secs since midnight. */
1604	(void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1605	/* Convert to msecs. */
1606	msecs = secs * MSEC_PER_SEC;
1607	/* Convert nsec to msec. */
1608	msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1609
1610	/* Convert to network byte order. */
1611	return htonl(msecs);
1612}
1613EXPORT_SYMBOL(inet_current_timestamp);
1614
1615int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1616{
1617	unsigned int family = READ_ONCE(sk->sk_family);
1618
1619	if (family == AF_INET)
1620		return ip_recv_error(sk, msg, len, addr_len);
1621#if IS_ENABLED(CONFIG_IPV6)
1622	if (family == AF_INET6)
1623		return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1624#endif
1625	return -EINVAL;
1626}
1627
1628int inet_gro_complete(struct sk_buff *skb, int nhoff)
1629{
1630	__be16 newlen = htons(skb->len - nhoff);
1631	struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1632	const struct net_offload *ops;
1633	int proto = iph->protocol;
1634	int err = -ENOSYS;
1635
1636	if (skb->encapsulation) {
1637		skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1638		skb_set_inner_network_header(skb, nhoff);
1639	}
1640
1641	csum_replace2(&iph->check, iph->tot_len, newlen);
1642	iph->tot_len = newlen;
1643
1644	rcu_read_lock();
1645	ops = rcu_dereference(inet_offloads[proto]);
1646	if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1647		goto out_unlock;
1648
1649	/* Only need to add sizeof(*iph) to get to the next hdr below
1650	 * because any hdr with option will have been flushed in
1651	 * inet_gro_receive().
1652	 */
1653	err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1654			      tcp4_gro_complete, udp4_gro_complete,
1655			      skb, nhoff + sizeof(*iph));
1656
1657out_unlock:
1658	rcu_read_unlock();
1659
1660	return err;
1661}
1662EXPORT_SYMBOL(inet_gro_complete);
1663
1664static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1665{
1666	skb->encapsulation = 1;
1667	skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1668	return inet_gro_complete(skb, nhoff);
1669}
1670
1671int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1672			 unsigned short type, unsigned char protocol,
1673			 struct net *net)
1674{
1675	struct socket *sock;
1676	int rc = sock_create_kern(net, family, type, protocol, &sock);
1677
1678	if (rc == 0) {
1679		*sk = sock->sk;
1680		(*sk)->sk_allocation = GFP_ATOMIC;
1681		/*
1682		 * Unhash it so that IP input processing does not even see it,
1683		 * we do not wish this socket to see incoming packets.
1684		 */
1685		(*sk)->sk_prot->unhash(*sk);
1686	}
1687	return rc;
1688}
1689EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1690
1691u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt)
1692{
1693	return  *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt);
1694}
1695EXPORT_SYMBOL_GPL(snmp_get_cpu_field);
1696
1697unsigned long snmp_fold_field(void __percpu *mib, int offt)
1698{
1699	unsigned long res = 0;
1700	int i;
1701
1702	for_each_possible_cpu(i)
1703		res += snmp_get_cpu_field(mib, i, offt);
1704	return res;
1705}
1706EXPORT_SYMBOL_GPL(snmp_fold_field);
1707
1708#if BITS_PER_LONG==32
1709
1710u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1711			 size_t syncp_offset)
1712{
1713	void *bhptr;
1714	struct u64_stats_sync *syncp;
1715	u64 v;
1716	unsigned int start;
1717
1718	bhptr = per_cpu_ptr(mib, cpu);
1719	syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1720	do {
1721		start = u64_stats_fetch_begin_irq(syncp);
1722		v = *(((u64 *)bhptr) + offt);
1723	} while (u64_stats_fetch_retry_irq(syncp, start));
1724
1725	return v;
1726}
1727EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1728
1729u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1730{
1731	u64 res = 0;
1732	int cpu;
1733
1734	for_each_possible_cpu(cpu) {
1735		res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1736	}
1737	return res;
1738}
1739EXPORT_SYMBOL_GPL(snmp_fold_field64);
1740#endif
1741
1742#ifdef CONFIG_IP_MULTICAST
1743static const struct net_protocol igmp_protocol = {
1744	.handler =	igmp_rcv,
1745	.netns_ok =	1,
1746};
1747#endif
1748
1749static const struct net_protocol tcp_protocol = {
1750	.handler	=	tcp_v4_rcv,
1751	.err_handler	=	tcp_v4_err,
1752	.no_policy	=	1,
1753	.netns_ok	=	1,
1754	.icmp_strict_tag_validation = 1,
1755};
1756
1757static const struct net_protocol udp_protocol = {
1758	.handler =	udp_rcv,
1759	.err_handler =	udp_err,
1760	.no_policy =	1,
1761	.netns_ok =	1,
1762};
1763
1764static const struct net_protocol icmp_protocol = {
1765	.handler =	icmp_rcv,
1766	.err_handler =	icmp_err,
1767	.no_policy =	1,
1768	.netns_ok =	1,
1769};
1770
1771static __net_init int ipv4_mib_init_net(struct net *net)
1772{
1773	int i;
1774
1775	net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1776	if (!net->mib.tcp_statistics)
1777		goto err_tcp_mib;
1778	net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1779	if (!net->mib.ip_statistics)
1780		goto err_ip_mib;
1781
1782	for_each_possible_cpu(i) {
1783		struct ipstats_mib *af_inet_stats;
1784		af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1785		u64_stats_init(&af_inet_stats->syncp);
1786	}
1787
1788	net->mib.net_statistics = alloc_percpu(struct linux_mib);
1789	if (!net->mib.net_statistics)
1790		goto err_net_mib;
1791	net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1792	if (!net->mib.udp_statistics)
1793		goto err_udp_mib;
1794	net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1795	if (!net->mib.udplite_statistics)
1796		goto err_udplite_mib;
1797	net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1798	if (!net->mib.icmp_statistics)
1799		goto err_icmp_mib;
1800	net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1801					      GFP_KERNEL);
1802	if (!net->mib.icmpmsg_statistics)
1803		goto err_icmpmsg_mib;
1804
1805	tcp_mib_init(net);
1806	return 0;
1807
1808err_icmpmsg_mib:
1809	free_percpu(net->mib.icmp_statistics);
1810err_icmp_mib:
1811	free_percpu(net->mib.udplite_statistics);
1812err_udplite_mib:
1813	free_percpu(net->mib.udp_statistics);
1814err_udp_mib:
1815	free_percpu(net->mib.net_statistics);
1816err_net_mib:
1817	free_percpu(net->mib.ip_statistics);
1818err_ip_mib:
1819	free_percpu(net->mib.tcp_statistics);
1820err_tcp_mib:
1821	return -ENOMEM;
1822}
1823
1824static __net_exit void ipv4_mib_exit_net(struct net *net)
1825{
1826	kfree(net->mib.icmpmsg_statistics);
1827	free_percpu(net->mib.icmp_statistics);
1828	free_percpu(net->mib.udplite_statistics);
1829	free_percpu(net->mib.udp_statistics);
1830	free_percpu(net->mib.net_statistics);
1831	free_percpu(net->mib.ip_statistics);
1832	free_percpu(net->mib.tcp_statistics);
1833#ifdef CONFIG_MPTCP
1834	/* allocated on demand, see mptcp_init_sock() */
1835	free_percpu(net->mib.mptcp_statistics);
1836#endif
1837}
1838
1839static __net_initdata struct pernet_operations ipv4_mib_ops = {
1840	.init = ipv4_mib_init_net,
1841	.exit = ipv4_mib_exit_net,
1842};
1843
1844static int __init init_ipv4_mibs(void)
1845{
1846	return register_pernet_subsys(&ipv4_mib_ops);
1847}
1848
1849static __net_init int inet_init_net(struct net *net)
1850{
1851	/*
1852	 * Set defaults for local port range
1853	 */
1854	seqlock_init(&net->ipv4.ip_local_ports.lock);
1855	net->ipv4.ip_local_ports.range[0] =  32768;
1856	net->ipv4.ip_local_ports.range[1] =  60999;
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#ifdef CONFIG_LOWPOWER_PROTOCOL
1888	lowpower_protocol_net_init(net);
1889#endif /* CONFIG_LOWPOWER_PROTOCOL */
1890	return 0;
1891}
1892
1893static __net_initdata struct pernet_operations af_inet_ops = {
1894	.init = inet_init_net,
1895};
1896
1897static int __init init_inet_pernet_ops(void)
1898{
1899	return register_pernet_subsys(&af_inet_ops);
1900}
1901
1902static int ipv4_proc_init(void);
1903
1904/*
1905 *	IP protocol layer initialiser
1906 */
1907
1908static struct packet_offload ip_packet_offload __read_mostly = {
1909	.type = cpu_to_be16(ETH_P_IP),
1910	.callbacks = {
1911		.gso_segment = inet_gso_segment,
1912		.gro_receive = inet_gro_receive,
1913		.gro_complete = inet_gro_complete,
1914	},
1915};
1916
1917static const struct net_offload ipip_offload = {
1918	.callbacks = {
1919		.gso_segment	= ipip_gso_segment,
1920		.gro_receive	= ipip_gro_receive,
1921		.gro_complete	= ipip_gro_complete,
1922	},
1923};
1924
1925static int __init ipip_offload_init(void)
1926{
1927	return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1928}
1929
1930static int __init ipv4_offload_init(void)
1931{
1932	/*
1933	 * Add offloads
1934	 */
1935	if (udpv4_offload_init() < 0)
1936		pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1937	if (tcpv4_offload_init() < 0)
1938		pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1939	if (ipip_offload_init() < 0)
1940		pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1941
1942	dev_add_offload(&ip_packet_offload);
1943	return 0;
1944}
1945
1946fs_initcall(ipv4_offload_init);
1947
1948static struct packet_type ip_packet_type __read_mostly = {
1949	.type = cpu_to_be16(ETH_P_IP),
1950	.func = ip_rcv,
1951	.list_func = ip_list_rcv,
1952};
1953
1954static int __init inet_init(void)
1955{
1956	struct inet_protosw *q;
1957	struct list_head *r;
1958	int rc;
1959
1960	sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1961
1962	rc = proto_register(&tcp_prot, 1);
1963	if (rc)
1964		goto out;
1965
1966	rc = proto_register(&udp_prot, 1);
1967	if (rc)
1968		goto out_unregister_tcp_proto;
1969
1970	rc = proto_register(&raw_prot, 1);
1971	if (rc)
1972		goto out_unregister_udp_proto;
1973
1974	rc = proto_register(&ping_prot, 1);
1975	if (rc)
1976		goto out_unregister_raw_proto;
1977
1978	/*
1979	 *	Tell SOCKET that we are alive...
1980	 */
1981
1982	(void)sock_register(&inet_family_ops);
1983
1984#ifdef CONFIG_SYSCTL
1985	ip_static_sysctl_init();
1986#endif
1987
1988	/*
1989	 *	Add all the base protocols.
1990	 */
1991
1992	if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1993		pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1994	if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1995		pr_crit("%s: Cannot add UDP protocol\n", __func__);
1996	if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1997		pr_crit("%s: Cannot add TCP protocol\n", __func__);
1998#ifdef CONFIG_IP_MULTICAST
1999	if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
2000		pr_crit("%s: Cannot add IGMP protocol\n", __func__);
2001#endif
2002
2003	/* Register the socket-side information for inet_create. */
2004	for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
2005		INIT_LIST_HEAD(r);
2006
2007	for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
2008		inet_register_protosw(q);
2009
2010	/*
2011	 *	Set the ARP module up
2012	 */
2013
2014	arp_init();
2015
2016	/*
2017	 *	Set the IP module up
2018	 */
2019
2020	ip_init();
2021
2022	/* Initialise per-cpu ipv4 mibs */
2023	if (init_ipv4_mibs())
2024		panic("%s: Cannot init ipv4 mibs\n", __func__);
2025
2026	/* Setup TCP slab cache for open requests. */
2027	tcp_init();
2028
2029	/* Setup UDP memory threshold */
2030	udp_init();
2031
2032	/* Add UDP-Lite (RFC 3828) */
2033	udplite4_register();
2034
2035	raw_init();
2036
2037	ping_init();
2038
2039	/*
2040	 *	Set the ICMP layer up
2041	 */
2042
2043	if (icmp_init() < 0)
2044		panic("Failed to create the ICMP control socket.\n");
2045
2046	/*
2047	 *	Initialise the multicast router
2048	 */
2049#if defined(CONFIG_IP_MROUTE)
2050	if (ip_mr_init())
2051		pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2052#endif
2053
2054	if (init_inet_pernet_ops())
2055		pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2056
2057	ipv4_proc_init();
2058
2059	ipfrag_init();
2060
2061	dev_add_pack(&ip_packet_type);
2062
2063	ip_tunnel_core_init();
2064
2065	rc = 0;
2066out:
2067	return rc;
2068out_unregister_raw_proto:
2069	proto_unregister(&raw_prot);
2070out_unregister_udp_proto:
2071	proto_unregister(&udp_prot);
2072out_unregister_tcp_proto:
2073	proto_unregister(&tcp_prot);
2074	goto out;
2075}
2076
2077fs_initcall(inet_init);
2078
2079/* ------------------------------------------------------------------------ */
2080
2081#ifdef CONFIG_PROC_FS
2082static int __init ipv4_proc_init(void)
2083{
2084	int rc = 0;
2085
2086	if (raw_proc_init())
2087		goto out_raw;
2088	if (tcp4_proc_init())
2089		goto out_tcp;
2090	if (udp4_proc_init())
2091		goto out_udp;
2092	if (ping_proc_init())
2093		goto out_ping;
2094	if (ip_misc_proc_init())
2095		goto out_misc;
2096out:
2097	return rc;
2098out_misc:
2099	ping_proc_exit();
2100out_ping:
2101	udp4_proc_exit();
2102out_udp:
2103	tcp4_proc_exit();
2104out_tcp:
2105	raw_proc_exit();
2106out_raw:
2107	rc = -ENOMEM;
2108	goto out;
2109}
2110
2111#else /* CONFIG_PROC_FS */
2112static int __init ipv4_proc_init(void)
2113{
2114	return 0;
2115}
2116#endif /* CONFIG_PROC_FS */
2117