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