af_inet.c revision ba89966c1984513f4f2cc0a6c182266be44ddd03 
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 * Version:	$Id: af_inet.c,v 1.137 2002/02/01 22:01:03 davem Exp $
9 *
10 * Authors:	Ross Biro
11 *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 *		Florian La Roche, <flla@stud.uni-sb.de>
13 *		Alan Cox, <A.Cox@swansea.ac.uk>
14 *
15 * Changes (see also sock.c)
16 *
17 *		piggy,
18 *		Karl Knutson	:	Socket protocol table
19 *		A.N.Kuznetsov	:	Socket death error in accept().
20 *		John Richardson :	Fix non blocking error in connect()
21 *					so sockets that fail to connect
22 *					don't return -EINPROGRESS.
23 *		Alan Cox	:	Asynchronous I/O support
24 *		Alan Cox	:	Keep correct socket pointer on sock
25 *					structures
26 *					when accept() ed
27 *		Alan Cox	:	Semantics of SO_LINGER aren't state
28 *					moved to close when you look carefully.
29 *					With this fixed and the accept bug fixed
30 *					some RPC stuff seems happier.
31 *		Niibe Yutaka	:	4.4BSD style write async I/O
32 *		Alan Cox,
33 *		Tony Gale 	:	Fixed reuse semantics.
34 *		Alan Cox	:	bind() shouldn't abort existing but dead
35 *					sockets. Stops FTP netin:.. I hope.
36 *		Alan Cox	:	bind() works correctly for RAW sockets.
37 *					Note that FreeBSD at least was broken
38 *					in this respect so be careful with
39 *					compatibility tests...
40 *		Alan Cox	:	routing cache support
41 *		Alan Cox	:	memzero the socket structure for
42 *					compactness.
43 *		Matt Day	:	nonblock connect error handler
44 *		Alan Cox	:	Allow large numbers of pending sockets
45 *					(eg for big web sites), but only if
46 *					specifically application requested.
47 *		Alan Cox	:	New buffering throughout IP. Used
48 *					dumbly.
49 *		Alan Cox	:	New buffering now used smartly.
50 *		Alan Cox	:	BSD rather than common sense
51 *					interpretation of listen.
52 *		Germano Caronni	:	Assorted small races.
53 *		Alan Cox	:	sendmsg/recvmsg basic support.
54 *		Alan Cox	:	Only sendmsg/recvmsg now supported.
55 *		Alan Cox	:	Locked down bind (see security list).
56 *		Alan Cox	:	Loosened bind a little.
57 *		Mike McLagan	:	ADD/DEL DLCI Ioctls
58 *	Willy Konynenberg	:	Transparent proxying support.
59 *		David S. Miller	:	New socket lookup architecture.
60 *					Some other random speedups.
61 *		Cyrus Durgin	:	Cleaned up file for kmod hacks.
62 *		Andi Kleen	:	Fix inet_stream_connect TCP race.
63 *
64 *		This program is free software; you can redistribute it and/or
65 *		modify it under the terms of the GNU General Public License
66 *		as published by the Free Software Foundation; either version
67 *		2 of the License, or (at your option) any later version.
68 */
69
70#include <linux/config.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/fcntl.h>
83#include <linux/mm.h>
84#include <linux/interrupt.h>
85#include <linux/stat.h>
86#include <linux/init.h>
87#include <linux/poll.h>
88#include <linux/netfilter_ipv4.h>
89
90#include <asm/uaccess.h>
91#include <asm/system.h>
92
93#include <linux/smp_lock.h>
94#include <linux/inet.h>
95#include <linux/igmp.h>
96#include <linux/netdevice.h>
97#include <net/ip.h>
98#include <net/protocol.h>
99#include <net/arp.h>
100#include <net/route.h>
101#include <net/ip_fib.h>
102#include <net/inet_connection_sock.h>
103#include <net/tcp.h>
104#include <net/udp.h>
105#include <linux/skbuff.h>
106#include <net/sock.h>
107#include <net/raw.h>
108#include <net/icmp.h>
109#include <net/ipip.h>
110#include <net/inet_common.h>
111#include <net/xfrm.h>
112#ifdef CONFIG_IP_MROUTE
113#include <linux/mroute.h>
114#endif
115
116DEFINE_SNMP_STAT(struct linux_mib, net_statistics) __read_mostly;
117
118extern void ip_mc_drop_socket(struct sock *sk);
119
120/* The inetsw table contains everything that inet_create needs to
121 * build a new socket.
122 */
123static struct list_head inetsw[SOCK_MAX];
124static DEFINE_SPINLOCK(inetsw_lock);
125
126/* New destruction routine */
127
128void inet_sock_destruct(struct sock *sk)
129{
130	struct inet_sock *inet = inet_sk(sk);
131
132	__skb_queue_purge(&sk->sk_receive_queue);
133	__skb_queue_purge(&sk->sk_error_queue);
134
135	if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
136		printk("Attempt to release TCP socket in state %d %p\n",
137		       sk->sk_state, sk);
138		return;
139	}
140	if (!sock_flag(sk, SOCK_DEAD)) {
141		printk("Attempt to release alive inet socket %p\n", sk);
142		return;
143	}
144
145	BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
146	BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
147	BUG_TRAP(!sk->sk_wmem_queued);
148	BUG_TRAP(!sk->sk_forward_alloc);
149
150	if (inet->opt)
151		kfree(inet->opt);
152	dst_release(sk->sk_dst_cache);
153	sk_refcnt_debug_dec(sk);
154}
155
156/*
157 *	The routines beyond this point handle the behaviour of an AF_INET
158 *	socket object. Mostly it punts to the subprotocols of IP to do
159 *	the work.
160 */
161
162/*
163 *	Automatically bind an unbound socket.
164 */
165
166static int inet_autobind(struct sock *sk)
167{
168	struct inet_sock *inet;
169	/* We may need to bind the socket. */
170	lock_sock(sk);
171	inet = inet_sk(sk);
172	if (!inet->num) {
173		if (sk->sk_prot->get_port(sk, 0)) {
174			release_sock(sk);
175			return -EAGAIN;
176		}
177		inet->sport = htons(inet->num);
178	}
179	release_sock(sk);
180	return 0;
181}
182
183/*
184 *	Move a socket into listening state.
185 */
186int inet_listen(struct socket *sock, int backlog)
187{
188	struct sock *sk = sock->sk;
189	unsigned char old_state;
190	int err;
191
192	lock_sock(sk);
193
194	err = -EINVAL;
195	if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
196		goto out;
197
198	old_state = sk->sk_state;
199	if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
200		goto out;
201
202	/* Really, if the socket is already in listen state
203	 * we can only allow the backlog to be adjusted.
204	 */
205	if (old_state != TCP_LISTEN) {
206		err = inet_csk_listen_start(sk, TCP_SYNQ_HSIZE);
207		if (err)
208			goto out;
209	}
210	sk->sk_max_ack_backlog = backlog;
211	err = 0;
212
213out:
214	release_sock(sk);
215	return err;
216}
217
218/*
219 *	Create an inet socket.
220 */
221
222static int inet_create(struct socket *sock, int protocol)
223{
224	struct sock *sk;
225	struct list_head *p;
226	struct inet_protosw *answer;
227	struct inet_sock *inet;
228	struct proto *answer_prot;
229	unsigned char answer_flags;
230	char answer_no_check;
231	int try_loading_module = 0;
232	int err = -ESOCKTNOSUPPORT;
233
234	sock->state = SS_UNCONNECTED;
235
236	/* Look for the requested type/protocol pair. */
237	answer = NULL;
238lookup_protocol:
239	rcu_read_lock();
240	list_for_each_rcu(p, &inetsw[sock->type]) {
241		answer = list_entry(p, struct inet_protosw, list);
242
243		/* Check the non-wild match. */
244		if (protocol == answer->protocol) {
245			if (protocol != IPPROTO_IP)
246				break;
247		} else {
248			/* Check for the two wild cases. */
249			if (IPPROTO_IP == protocol) {
250				protocol = answer->protocol;
251				break;
252			}
253			if (IPPROTO_IP == answer->protocol)
254				break;
255		}
256		answer = NULL;
257	}
258
259	if (unlikely(answer == NULL)) {
260		if (try_loading_module < 2) {
261			rcu_read_unlock();
262			/*
263			 * Be more specific, e.g. net-pf-2-proto-132-type-1
264			 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
265			 */
266			if (++try_loading_module == 1)
267				request_module("net-pf-%d-proto-%d-type-%d",
268					       PF_INET, protocol, sock->type);
269			/*
270			 * Fall back to generic, e.g. net-pf-2-proto-132
271			 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
272			 */
273			else
274				request_module("net-pf-%d-proto-%d",
275					       PF_INET, protocol);
276			goto lookup_protocol;
277		} else
278			goto out_rcu_unlock;
279	}
280
281	err = -EPERM;
282	if (answer->capability > 0 && !capable(answer->capability))
283		goto out_rcu_unlock;
284	err = -EPROTONOSUPPORT;
285	if (!protocol)
286		goto out_rcu_unlock;
287
288	sock->ops = answer->ops;
289	answer_prot = answer->prot;
290	answer_no_check = answer->no_check;
291	answer_flags = answer->flags;
292	rcu_read_unlock();
293
294	BUG_TRAP(answer_prot->slab != NULL);
295
296	err = -ENOBUFS;
297	sk = sk_alloc(PF_INET, GFP_KERNEL, answer_prot, 1);
298	if (sk == NULL)
299		goto out;
300
301	err = 0;
302	sk->sk_no_check = answer_no_check;
303	if (INET_PROTOSW_REUSE & answer_flags)
304		sk->sk_reuse = 1;
305
306	inet = inet_sk(sk);
307
308	if (SOCK_RAW == sock->type) {
309		inet->num = protocol;
310		if (IPPROTO_RAW == protocol)
311			inet->hdrincl = 1;
312	}
313
314	if (ipv4_config.no_pmtu_disc)
315		inet->pmtudisc = IP_PMTUDISC_DONT;
316	else
317		inet->pmtudisc = IP_PMTUDISC_WANT;
318
319	inet->id = 0;
320
321	sock_init_data(sock, sk);
322
323	sk->sk_destruct	   = inet_sock_destruct;
324	sk->sk_family	   = PF_INET;
325	sk->sk_protocol	   = protocol;
326	sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
327
328	inet->uc_ttl	= -1;
329	inet->mc_loop	= 1;
330	inet->mc_ttl	= 1;
331	inet->mc_index	= 0;
332	inet->mc_list	= NULL;
333
334	sk_refcnt_debug_inc(sk);
335
336	if (inet->num) {
337		/* It assumes that any protocol which allows
338		 * the user to assign a number at socket
339		 * creation time automatically
340		 * shares.
341		 */
342		inet->sport = htons(inet->num);
343		/* Add to protocol hash chains. */
344		sk->sk_prot->hash(sk);
345	}
346
347	if (sk->sk_prot->init) {
348		err = sk->sk_prot->init(sk);
349		if (err)
350			sk_common_release(sk);
351	}
352out:
353	return err;
354out_rcu_unlock:
355	rcu_read_unlock();
356	goto out;
357}
358
359
360/*
361 *	The peer socket should always be NULL (or else). When we call this
362 *	function we are destroying the object and from then on nobody
363 *	should refer to it.
364 */
365int inet_release(struct socket *sock)
366{
367	struct sock *sk = sock->sk;
368
369	if (sk) {
370		long timeout;
371
372		/* Applications forget to leave groups before exiting */
373		ip_mc_drop_socket(sk);
374
375		/* If linger is set, we don't return until the close
376		 * is complete.  Otherwise we return immediately. The
377		 * actually closing is done the same either way.
378		 *
379		 * If the close is due to the process exiting, we never
380		 * linger..
381		 */
382		timeout = 0;
383		if (sock_flag(sk, SOCK_LINGER) &&
384		    !(current->flags & PF_EXITING))
385			timeout = sk->sk_lingertime;
386		sock->sk = NULL;
387		sk->sk_prot->close(sk, timeout);
388	}
389	return 0;
390}
391
392/* It is off by default, see below. */
393int sysctl_ip_nonlocal_bind;
394
395int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
396{
397	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
398	struct sock *sk = sock->sk;
399	struct inet_sock *inet = inet_sk(sk);
400	unsigned short snum;
401	int chk_addr_ret;
402	int err;
403
404	/* If the socket has its own bind function then use it. (RAW) */
405	if (sk->sk_prot->bind) {
406		err = sk->sk_prot->bind(sk, uaddr, addr_len);
407		goto out;
408	}
409	err = -EINVAL;
410	if (addr_len < sizeof(struct sockaddr_in))
411		goto out;
412
413	chk_addr_ret = inet_addr_type(addr->sin_addr.s_addr);
414
415	/* Not specified by any standard per-se, however it breaks too
416	 * many applications when removed.  It is unfortunate since
417	 * allowing applications to make a non-local bind solves
418	 * several problems with systems using dynamic addressing.
419	 * (ie. your servers still start up even if your ISDN link
420	 *  is temporarily down)
421	 */
422	err = -EADDRNOTAVAIL;
423	if (!sysctl_ip_nonlocal_bind &&
424	    !inet->freebind &&
425	    addr->sin_addr.s_addr != INADDR_ANY &&
426	    chk_addr_ret != RTN_LOCAL &&
427	    chk_addr_ret != RTN_MULTICAST &&
428	    chk_addr_ret != RTN_BROADCAST)
429		goto out;
430
431	snum = ntohs(addr->sin_port);
432	err = -EACCES;
433	if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
434		goto out;
435
436	/*      We keep a pair of addresses. rcv_saddr is the one
437	 *      used by hash lookups, and saddr is used for transmit.
438	 *
439	 *      In the BSD API these are the same except where it
440	 *      would be illegal to use them (multicast/broadcast) in
441	 *      which case the sending device address is used.
442	 */
443	lock_sock(sk);
444
445	/* Check these errors (active socket, double bind). */
446	err = -EINVAL;
447	if (sk->sk_state != TCP_CLOSE || inet->num)
448		goto out_release_sock;
449
450	inet->rcv_saddr = inet->saddr = addr->sin_addr.s_addr;
451	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
452		inet->saddr = 0;  /* Use device */
453
454	/* Make sure we are allowed to bind here. */
455	if (sk->sk_prot->get_port(sk, snum)) {
456		inet->saddr = inet->rcv_saddr = 0;
457		err = -EADDRINUSE;
458		goto out_release_sock;
459	}
460
461	if (inet->rcv_saddr)
462		sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
463	if (snum)
464		sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
465	inet->sport = htons(inet->num);
466	inet->daddr = 0;
467	inet->dport = 0;
468	sk_dst_reset(sk);
469	err = 0;
470out_release_sock:
471	release_sock(sk);
472out:
473	return err;
474}
475
476int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
477		       int addr_len, int flags)
478{
479	struct sock *sk = sock->sk;
480
481	if (uaddr->sa_family == AF_UNSPEC)
482		return sk->sk_prot->disconnect(sk, flags);
483
484	if (!inet_sk(sk)->num && inet_autobind(sk))
485		return -EAGAIN;
486	return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
487}
488
489static long inet_wait_for_connect(struct sock *sk, long timeo)
490{
491	DEFINE_WAIT(wait);
492
493	prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
494
495	/* Basic assumption: if someone sets sk->sk_err, he _must_
496	 * change state of the socket from TCP_SYN_*.
497	 * Connect() does not allow to get error notifications
498	 * without closing the socket.
499	 */
500	while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
501		release_sock(sk);
502		timeo = schedule_timeout(timeo);
503		lock_sock(sk);
504		if (signal_pending(current) || !timeo)
505			break;
506		prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
507	}
508	finish_wait(sk->sk_sleep, &wait);
509	return timeo;
510}
511
512/*
513 *	Connect to a remote host. There is regrettably still a little
514 *	TCP 'magic' in here.
515 */
516int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
517			int addr_len, int flags)
518{
519	struct sock *sk = sock->sk;
520	int err;
521	long timeo;
522
523	lock_sock(sk);
524
525	if (uaddr->sa_family == AF_UNSPEC) {
526		err = sk->sk_prot->disconnect(sk, flags);
527		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
528		goto out;
529	}
530
531	switch (sock->state) {
532	default:
533		err = -EINVAL;
534		goto out;
535	case SS_CONNECTED:
536		err = -EISCONN;
537		goto out;
538	case SS_CONNECTING:
539		err = -EALREADY;
540		/* Fall out of switch with err, set for this state */
541		break;
542	case SS_UNCONNECTED:
543		err = -EISCONN;
544		if (sk->sk_state != TCP_CLOSE)
545			goto out;
546
547		err = sk->sk_prot->connect(sk, uaddr, addr_len);
548		if (err < 0)
549			goto out;
550
551  		sock->state = SS_CONNECTING;
552
553		/* Just entered SS_CONNECTING state; the only
554		 * difference is that return value in non-blocking
555		 * case is EINPROGRESS, rather than EALREADY.
556		 */
557		err = -EINPROGRESS;
558		break;
559	}
560
561	timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
562
563	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
564		/* Error code is set above */
565		if (!timeo || !inet_wait_for_connect(sk, timeo))
566			goto out;
567
568		err = sock_intr_errno(timeo);
569		if (signal_pending(current))
570			goto out;
571	}
572
573	/* Connection was closed by RST, timeout, ICMP error
574	 * or another process disconnected us.
575	 */
576	if (sk->sk_state == TCP_CLOSE)
577		goto sock_error;
578
579	/* sk->sk_err may be not zero now, if RECVERR was ordered by user
580	 * and error was received after socket entered established state.
581	 * Hence, it is handled normally after connect() return successfully.
582	 */
583
584	sock->state = SS_CONNECTED;
585	err = 0;
586out:
587	release_sock(sk);
588	return err;
589
590sock_error:
591	err = sock_error(sk) ? : -ECONNABORTED;
592	sock->state = SS_UNCONNECTED;
593	if (sk->sk_prot->disconnect(sk, flags))
594		sock->state = SS_DISCONNECTING;
595	goto out;
596}
597
598/*
599 *	Accept a pending connection. The TCP layer now gives BSD semantics.
600 */
601
602int inet_accept(struct socket *sock, struct socket *newsock, int flags)
603{
604	struct sock *sk1 = sock->sk;
605	int err = -EINVAL;
606	struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
607
608	if (!sk2)
609		goto do_err;
610
611	lock_sock(sk2);
612
613	BUG_TRAP((1 << sk2->sk_state) &
614		 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE));
615
616	sock_graft(sk2, newsock);
617
618	newsock->state = SS_CONNECTED;
619	err = 0;
620	release_sock(sk2);
621do_err:
622	return err;
623}
624
625
626/*
627 *	This does both peername and sockname.
628 */
629int inet_getname(struct socket *sock, struct sockaddr *uaddr,
630			int *uaddr_len, int peer)
631{
632	struct sock *sk		= sock->sk;
633	struct inet_sock *inet	= inet_sk(sk);
634	struct sockaddr_in *sin	= (struct sockaddr_in *)uaddr;
635
636	sin->sin_family = AF_INET;
637	if (peer) {
638		if (!inet->dport ||
639		    (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
640		     peer == 1))
641			return -ENOTCONN;
642		sin->sin_port = inet->dport;
643		sin->sin_addr.s_addr = inet->daddr;
644	} else {
645		__u32 addr = inet->rcv_saddr;
646		if (!addr)
647			addr = inet->saddr;
648		sin->sin_port = inet->sport;
649		sin->sin_addr.s_addr = addr;
650	}
651	memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
652	*uaddr_len = sizeof(*sin);
653	return 0;
654}
655
656int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
657		 size_t size)
658{
659	struct sock *sk = sock->sk;
660
661	/* We may need to bind the socket. */
662	if (!inet_sk(sk)->num && inet_autobind(sk))
663		return -EAGAIN;
664
665	return sk->sk_prot->sendmsg(iocb, sk, msg, size);
666}
667
668
669static ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
670{
671	struct sock *sk = sock->sk;
672
673	/* We may need to bind the socket. */
674	if (!inet_sk(sk)->num && inet_autobind(sk))
675		return -EAGAIN;
676
677	if (sk->sk_prot->sendpage)
678		return sk->sk_prot->sendpage(sk, page, offset, size, flags);
679	return sock_no_sendpage(sock, page, offset, size, flags);
680}
681
682
683int inet_shutdown(struct socket *sock, int how)
684{
685	struct sock *sk = sock->sk;
686	int err = 0;
687
688	/* This should really check to make sure
689	 * the socket is a TCP socket. (WHY AC...)
690	 */
691	how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
692		       1->2 bit 2 snds.
693		       2->3 */
694	if ((how & ~SHUTDOWN_MASK) || !how)	/* MAXINT->0 */
695		return -EINVAL;
696
697	lock_sock(sk);
698	if (sock->state == SS_CONNECTING) {
699		if ((1 << sk->sk_state) &
700		    (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
701			sock->state = SS_DISCONNECTING;
702		else
703			sock->state = SS_CONNECTED;
704	}
705
706	switch (sk->sk_state) {
707	case TCP_CLOSE:
708		err = -ENOTCONN;
709		/* Hack to wake up other listeners, who can poll for
710		   POLLHUP, even on eg. unconnected UDP sockets -- RR */
711	default:
712		sk->sk_shutdown |= how;
713		if (sk->sk_prot->shutdown)
714			sk->sk_prot->shutdown(sk, how);
715		break;
716
717	/* Remaining two branches are temporary solution for missing
718	 * close() in multithreaded environment. It is _not_ a good idea,
719	 * but we have no choice until close() is repaired at VFS level.
720	 */
721	case TCP_LISTEN:
722		if (!(how & RCV_SHUTDOWN))
723			break;
724		/* Fall through */
725	case TCP_SYN_SENT:
726		err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
727		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
728		break;
729	}
730
731	/* Wake up anyone sleeping in poll. */
732	sk->sk_state_change(sk);
733	release_sock(sk);
734	return err;
735}
736
737/*
738 *	ioctl() calls you can issue on an INET socket. Most of these are
739 *	device configuration and stuff and very rarely used. Some ioctls
740 *	pass on to the socket itself.
741 *
742 *	NOTE: I like the idea of a module for the config stuff. ie ifconfig
743 *	loads the devconfigure module does its configuring and unloads it.
744 *	There's a good 20K of config code hanging around the kernel.
745 */
746
747int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
748{
749	struct sock *sk = sock->sk;
750	int err = 0;
751
752	switch (cmd) {
753		case SIOCGSTAMP:
754			err = sock_get_timestamp(sk, (struct timeval __user *)arg);
755			break;
756		case SIOCADDRT:
757		case SIOCDELRT:
758		case SIOCRTMSG:
759			err = ip_rt_ioctl(cmd, (void __user *)arg);
760			break;
761		case SIOCDARP:
762		case SIOCGARP:
763		case SIOCSARP:
764			err = arp_ioctl(cmd, (void __user *)arg);
765			break;
766		case SIOCGIFADDR:
767		case SIOCSIFADDR:
768		case SIOCGIFBRDADDR:
769		case SIOCSIFBRDADDR:
770		case SIOCGIFNETMASK:
771		case SIOCSIFNETMASK:
772		case SIOCGIFDSTADDR:
773		case SIOCSIFDSTADDR:
774		case SIOCSIFPFLAGS:
775		case SIOCGIFPFLAGS:
776		case SIOCSIFFLAGS:
777			err = devinet_ioctl(cmd, (void __user *)arg);
778			break;
779		default:
780			if (!sk->sk_prot->ioctl ||
781			    (err = sk->sk_prot->ioctl(sk, cmd, arg)) ==
782			    					-ENOIOCTLCMD)
783				err = dev_ioctl(cmd, (void __user *)arg);
784			break;
785	}
786	return err;
787}
788
789struct proto_ops inet_stream_ops = {
790	.family =	PF_INET,
791	.owner =	THIS_MODULE,
792	.release =	inet_release,
793	.bind =		inet_bind,
794	.connect =	inet_stream_connect,
795	.socketpair =	sock_no_socketpair,
796	.accept =	inet_accept,
797	.getname =	inet_getname,
798	.poll =		tcp_poll,
799	.ioctl =	inet_ioctl,
800	.listen =	inet_listen,
801	.shutdown =	inet_shutdown,
802	.setsockopt =	sock_common_setsockopt,
803	.getsockopt =	sock_common_getsockopt,
804	.sendmsg =	inet_sendmsg,
805	.recvmsg =	sock_common_recvmsg,
806	.mmap =		sock_no_mmap,
807	.sendpage =	tcp_sendpage
808};
809
810struct proto_ops inet_dgram_ops = {
811	.family =	PF_INET,
812	.owner =	THIS_MODULE,
813	.release =	inet_release,
814	.bind =		inet_bind,
815	.connect =	inet_dgram_connect,
816	.socketpair =	sock_no_socketpair,
817	.accept =	sock_no_accept,
818	.getname =	inet_getname,
819	.poll =		udp_poll,
820	.ioctl =	inet_ioctl,
821	.listen =	sock_no_listen,
822	.shutdown =	inet_shutdown,
823	.setsockopt =	sock_common_setsockopt,
824	.getsockopt =	sock_common_getsockopt,
825	.sendmsg =	inet_sendmsg,
826	.recvmsg =	sock_common_recvmsg,
827	.mmap =		sock_no_mmap,
828	.sendpage =	inet_sendpage,
829};
830
831/*
832 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
833 * udp_poll
834 */
835static struct proto_ops inet_sockraw_ops = {
836	.family =	PF_INET,
837	.owner =	THIS_MODULE,
838	.release =	inet_release,
839	.bind =		inet_bind,
840	.connect =	inet_dgram_connect,
841	.socketpair =	sock_no_socketpair,
842	.accept =	sock_no_accept,
843	.getname =	inet_getname,
844	.poll =		datagram_poll,
845	.ioctl =	inet_ioctl,
846	.listen =	sock_no_listen,
847	.shutdown =	inet_shutdown,
848	.setsockopt =	sock_common_setsockopt,
849	.getsockopt =	sock_common_getsockopt,
850	.sendmsg =	inet_sendmsg,
851	.recvmsg =	sock_common_recvmsg,
852	.mmap =		sock_no_mmap,
853	.sendpage =	inet_sendpage,
854};
855
856static struct net_proto_family inet_family_ops = {
857	.family = PF_INET,
858	.create = inet_create,
859	.owner	= THIS_MODULE,
860};
861
862/* Upon startup we insert all the elements in inetsw_array[] into
863 * the linked list inetsw.
864 */
865static struct inet_protosw inetsw_array[] =
866{
867        {
868                .type =       SOCK_STREAM,
869                .protocol =   IPPROTO_TCP,
870                .prot =       &tcp_prot,
871                .ops =        &inet_stream_ops,
872                .capability = -1,
873                .no_check =   0,
874                .flags =      INET_PROTOSW_PERMANENT,
875        },
876
877        {
878                .type =       SOCK_DGRAM,
879                .protocol =   IPPROTO_UDP,
880                .prot =       &udp_prot,
881                .ops =        &inet_dgram_ops,
882                .capability = -1,
883                .no_check =   UDP_CSUM_DEFAULT,
884                .flags =      INET_PROTOSW_PERMANENT,
885       },
886
887
888       {
889               .type =       SOCK_RAW,
890               .protocol =   IPPROTO_IP,	/* wild card */
891               .prot =       &raw_prot,
892               .ops =        &inet_sockraw_ops,
893               .capability = CAP_NET_RAW,
894               .no_check =   UDP_CSUM_DEFAULT,
895               .flags =      INET_PROTOSW_REUSE,
896       }
897};
898
899#define INETSW_ARRAY_LEN (sizeof(inetsw_array) / sizeof(struct inet_protosw))
900
901void inet_register_protosw(struct inet_protosw *p)
902{
903	struct list_head *lh;
904	struct inet_protosw *answer;
905	int protocol = p->protocol;
906	struct list_head *last_perm;
907
908	spin_lock_bh(&inetsw_lock);
909
910	if (p->type >= SOCK_MAX)
911		goto out_illegal;
912
913	/* If we are trying to override a permanent protocol, bail. */
914	answer = NULL;
915	last_perm = &inetsw[p->type];
916	list_for_each(lh, &inetsw[p->type]) {
917		answer = list_entry(lh, struct inet_protosw, list);
918
919		/* Check only the non-wild match. */
920		if (INET_PROTOSW_PERMANENT & answer->flags) {
921			if (protocol == answer->protocol)
922				break;
923			last_perm = lh;
924		}
925
926		answer = NULL;
927	}
928	if (answer)
929		goto out_permanent;
930
931	/* Add the new entry after the last permanent entry if any, so that
932	 * the new entry does not override a permanent entry when matched with
933	 * a wild-card protocol. But it is allowed to override any existing
934	 * non-permanent entry.  This means that when we remove this entry, the
935	 * system automatically returns to the old behavior.
936	 */
937	list_add_rcu(&p->list, last_perm);
938out:
939	spin_unlock_bh(&inetsw_lock);
940
941	synchronize_net();
942
943	return;
944
945out_permanent:
946	printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
947	       protocol);
948	goto out;
949
950out_illegal:
951	printk(KERN_ERR
952	       "Ignoring attempt to register invalid socket type %d.\n",
953	       p->type);
954	goto out;
955}
956
957void inet_unregister_protosw(struct inet_protosw *p)
958{
959	if (INET_PROTOSW_PERMANENT & p->flags) {
960		printk(KERN_ERR
961		       "Attempt to unregister permanent protocol %d.\n",
962		       p->protocol);
963	} else {
964		spin_lock_bh(&inetsw_lock);
965		list_del_rcu(&p->list);
966		spin_unlock_bh(&inetsw_lock);
967
968		synchronize_net();
969	}
970}
971
972/*
973 *      Shall we try to damage output packets if routing dev changes?
974 */
975
976int sysctl_ip_dynaddr;
977
978static int inet_sk_reselect_saddr(struct sock *sk)
979{
980	struct inet_sock *inet = inet_sk(sk);
981	int err;
982	struct rtable *rt;
983	__u32 old_saddr = inet->saddr;
984	__u32 new_saddr;
985	__u32 daddr = inet->daddr;
986
987	if (inet->opt && inet->opt->srr)
988		daddr = inet->opt->faddr;
989
990	/* Query new route. */
991	err = ip_route_connect(&rt, daddr, 0,
992			       RT_CONN_FLAGS(sk),
993			       sk->sk_bound_dev_if,
994			       sk->sk_protocol,
995			       inet->sport, inet->dport, sk);
996	if (err)
997		return err;
998
999	sk_setup_caps(sk, &rt->u.dst);
1000
1001	new_saddr = rt->rt_src;
1002
1003	if (new_saddr == old_saddr)
1004		return 0;
1005
1006	if (sysctl_ip_dynaddr > 1) {
1007		printk(KERN_INFO "%s(): shifting inet->"
1008				 "saddr from %d.%d.%d.%d to %d.%d.%d.%d\n",
1009		       __FUNCTION__,
1010		       NIPQUAD(old_saddr),
1011		       NIPQUAD(new_saddr));
1012	}
1013
1014	inet->saddr = inet->rcv_saddr = new_saddr;
1015
1016	/*
1017	 * XXX The only one ugly spot where we need to
1018	 * XXX really change the sockets identity after
1019	 * XXX it has entered the hashes. -DaveM
1020	 *
1021	 * Besides that, it does not check for connection
1022	 * uniqueness. Wait for troubles.
1023	 */
1024	__sk_prot_rehash(sk);
1025	return 0;
1026}
1027
1028int inet_sk_rebuild_header(struct sock *sk)
1029{
1030	struct inet_sock *inet = inet_sk(sk);
1031	struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1032	u32 daddr;
1033	int err;
1034
1035	/* Route is OK, nothing to do. */
1036	if (rt)
1037		return 0;
1038
1039	/* Reroute. */
1040	daddr = inet->daddr;
1041	if (inet->opt && inet->opt->srr)
1042		daddr = inet->opt->faddr;
1043{
1044	struct flowi fl = {
1045		.oif = sk->sk_bound_dev_if,
1046		.nl_u = {
1047			.ip4_u = {
1048				.daddr	= daddr,
1049				.saddr	= inet->saddr,
1050				.tos	= RT_CONN_FLAGS(sk),
1051			},
1052		},
1053		.proto = sk->sk_protocol,
1054		.uli_u = {
1055			.ports = {
1056				.sport = inet->sport,
1057				.dport = inet->dport,
1058			},
1059		},
1060	};
1061
1062	err = ip_route_output_flow(&rt, &fl, sk, 0);
1063}
1064	if (!err)
1065		sk_setup_caps(sk, &rt->u.dst);
1066	else {
1067		/* Routing failed... */
1068		sk->sk_route_caps = 0;
1069		/*
1070		 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1071		 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1072		 */
1073		if (!sysctl_ip_dynaddr ||
1074		    sk->sk_state != TCP_SYN_SENT ||
1075		    (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1076		    (err = inet_sk_reselect_saddr(sk)) != 0)
1077			sk->sk_err_soft = -err;
1078	}
1079
1080	return err;
1081}
1082
1083EXPORT_SYMBOL(inet_sk_rebuild_header);
1084
1085#ifdef CONFIG_IP_MULTICAST
1086static struct net_protocol igmp_protocol = {
1087	.handler =	igmp_rcv,
1088};
1089#endif
1090
1091static struct net_protocol tcp_protocol = {
1092	.handler =	tcp_v4_rcv,
1093	.err_handler =	tcp_v4_err,
1094	.no_policy =	1,
1095};
1096
1097static struct net_protocol udp_protocol = {
1098	.handler =	udp_rcv,
1099	.err_handler =	udp_err,
1100	.no_policy =	1,
1101};
1102
1103static struct net_protocol icmp_protocol = {
1104	.handler =	icmp_rcv,
1105};
1106
1107static int __init init_ipv4_mibs(void)
1108{
1109	net_statistics[0] = alloc_percpu(struct linux_mib);
1110	net_statistics[1] = alloc_percpu(struct linux_mib);
1111	ip_statistics[0] = alloc_percpu(struct ipstats_mib);
1112	ip_statistics[1] = alloc_percpu(struct ipstats_mib);
1113	icmp_statistics[0] = alloc_percpu(struct icmp_mib);
1114	icmp_statistics[1] = alloc_percpu(struct icmp_mib);
1115	tcp_statistics[0] = alloc_percpu(struct tcp_mib);
1116	tcp_statistics[1] = alloc_percpu(struct tcp_mib);
1117	udp_statistics[0] = alloc_percpu(struct udp_mib);
1118	udp_statistics[1] = alloc_percpu(struct udp_mib);
1119	if (!
1120	    (net_statistics[0] && net_statistics[1] && ip_statistics[0]
1121	     && ip_statistics[1] && tcp_statistics[0] && tcp_statistics[1]
1122	     && udp_statistics[0] && udp_statistics[1]))
1123		return -ENOMEM;
1124
1125	(void) tcp_mib_init();
1126
1127	return 0;
1128}
1129
1130static int ipv4_proc_init(void);
1131
1132/*
1133 *	IP protocol layer initialiser
1134 */
1135
1136static struct packet_type ip_packet_type = {
1137	.type = __constant_htons(ETH_P_IP),
1138	.func = ip_rcv,
1139};
1140
1141static int __init inet_init(void)
1142{
1143	struct sk_buff *dummy_skb;
1144	struct inet_protosw *q;
1145	struct list_head *r;
1146	int rc = -EINVAL;
1147
1148	if (sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb)) {
1149		printk(KERN_CRIT "%s: panic\n", __FUNCTION__);
1150		goto out;
1151	}
1152
1153	rc = proto_register(&tcp_prot, 1);
1154	if (rc)
1155		goto out;
1156
1157	rc = proto_register(&udp_prot, 1);
1158	if (rc)
1159		goto out_unregister_tcp_proto;
1160
1161	rc = proto_register(&raw_prot, 1);
1162	if (rc)
1163		goto out_unregister_udp_proto;
1164
1165	/*
1166	 *	Tell SOCKET that we are alive...
1167	 */
1168
1169  	(void)sock_register(&inet_family_ops);
1170
1171	/*
1172	 *	Add all the base protocols.
1173	 */
1174
1175	if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1176		printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1177	if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1178		printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1179	if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1180		printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1181#ifdef CONFIG_IP_MULTICAST
1182	if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1183		printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1184#endif
1185
1186	/* Register the socket-side information for inet_create. */
1187	for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1188		INIT_LIST_HEAD(r);
1189
1190	for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1191		inet_register_protosw(q);
1192
1193	/*
1194	 *	Set the ARP module up
1195	 */
1196
1197	arp_init();
1198
1199  	/*
1200  	 *	Set the IP module up
1201  	 */
1202
1203	ip_init();
1204
1205	tcp_v4_init(&inet_family_ops);
1206
1207	/* Setup TCP slab cache for open requests. */
1208	tcp_init();
1209
1210
1211	/*
1212	 *	Set the ICMP layer up
1213	 */
1214
1215	icmp_init(&inet_family_ops);
1216
1217	/*
1218	 *	Initialise the multicast router
1219	 */
1220#if defined(CONFIG_IP_MROUTE)
1221	ip_mr_init();
1222#endif
1223	/*
1224	 *	Initialise per-cpu ipv4 mibs
1225	 */
1226
1227	if(init_ipv4_mibs())
1228		printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n"); ;
1229
1230	ipv4_proc_init();
1231
1232	ipfrag_init();
1233
1234	dev_add_pack(&ip_packet_type);
1235
1236	rc = 0;
1237out:
1238	return rc;
1239out_unregister_tcp_proto:
1240	proto_unregister(&tcp_prot);
1241out_unregister_udp_proto:
1242	proto_unregister(&udp_prot);
1243	goto out;
1244}
1245
1246module_init(inet_init);
1247
1248/* ------------------------------------------------------------------------ */
1249
1250#ifdef CONFIG_PROC_FS
1251#ifdef CONFIG_IP_FIB_TRIE
1252extern int  fib_stat_proc_init(void);
1253extern void fib_stat_proc_exit(void);
1254#endif
1255
1256static int __init ipv4_proc_init(void)
1257{
1258	int rc = 0;
1259
1260	if (raw_proc_init())
1261		goto out_raw;
1262	if (tcp4_proc_init())
1263		goto out_tcp;
1264	if (udp4_proc_init())
1265		goto out_udp;
1266	if (fib_proc_init())
1267		goto out_fib;
1268#ifdef CONFIG_IP_FIB_TRIE
1269         if (fib_stat_proc_init())
1270                 goto out_fib_stat;
1271#endif
1272	if (ip_misc_proc_init())
1273		goto out_misc;
1274out:
1275	return rc;
1276out_misc:
1277#ifdef CONFIG_IP_FIB_TRIE
1278 	fib_stat_proc_exit();
1279out_fib_stat:
1280#endif
1281	fib_proc_exit();
1282out_fib:
1283	udp4_proc_exit();
1284out_udp:
1285	tcp4_proc_exit();
1286out_tcp:
1287	raw_proc_exit();
1288out_raw:
1289	rc = -ENOMEM;
1290	goto out;
1291}
1292
1293#else /* CONFIG_PROC_FS */
1294static int __init ipv4_proc_init(void)
1295{
1296	return 0;
1297}
1298#endif /* CONFIG_PROC_FS */
1299
1300MODULE_ALIAS_NETPROTO(PF_INET);
1301
1302EXPORT_SYMBOL(inet_accept);
1303EXPORT_SYMBOL(inet_bind);
1304EXPORT_SYMBOL(inet_dgram_connect);
1305EXPORT_SYMBOL(inet_dgram_ops);
1306EXPORT_SYMBOL(inet_getname);
1307EXPORT_SYMBOL(inet_ioctl);
1308EXPORT_SYMBOL(inet_listen);
1309EXPORT_SYMBOL(inet_register_protosw);
1310EXPORT_SYMBOL(inet_release);
1311EXPORT_SYMBOL(inet_sendmsg);
1312EXPORT_SYMBOL(inet_shutdown);
1313EXPORT_SYMBOL(inet_sock_destruct);
1314EXPORT_SYMBOL(inet_stream_connect);
1315EXPORT_SYMBOL(inet_stream_ops);
1316EXPORT_SYMBOL(inet_unregister_protosw);
1317EXPORT_SYMBOL(net_statistics);
1318EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
1319