af_inet.c revision 1da177e4c3f41524e886b7f1b8a0c1fc7321cac2 
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, <bir7@leland.Stanford.Edu>
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/major.h>
77#include <linux/module.h>
78#include <linux/sched.h>
79#include <linux/timer.h>
80#include <linux/string.h>
81#include <linux/sockios.h>
82#include <linux/net.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
91#include <asm/uaccess.h>
92#include <asm/system.h>
93
94#include <linux/smp_lock.h>
95#include <linux/inet.h>
96#include <linux/igmp.h>
97#include <linux/netdevice.h>
98#include <net/ip.h>
99#include <net/protocol.h>
100#include <net/arp.h>
101#include <net/route.h>
102#include <net/ip_fib.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);
117
118#ifdef INET_REFCNT_DEBUG
119atomic_t inet_sock_nr;
120#endif
121
122extern void ip_mc_drop_socket(struct sock *sk);
123
124/* The inetsw table contains everything that inet_create needs to
125 * build a new socket.
126 */
127static struct list_head inetsw[SOCK_MAX];
128static DEFINE_SPINLOCK(inetsw_lock);
129
130/* New destruction routine */
131
132void inet_sock_destruct(struct sock *sk)
133{
134	struct inet_sock *inet = inet_sk(sk);
135
136	__skb_queue_purge(&sk->sk_receive_queue);
137	__skb_queue_purge(&sk->sk_error_queue);
138
139	if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
140		printk("Attempt to release TCP socket in state %d %p\n",
141		       sk->sk_state, sk);
142		return;
143	}
144	if (!sock_flag(sk, SOCK_DEAD)) {
145		printk("Attempt to release alive inet socket %p\n", sk);
146		return;
147	}
148
149	BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
150	BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
151	BUG_TRAP(!sk->sk_wmem_queued);
152	BUG_TRAP(!sk->sk_forward_alloc);
153
154	if (inet->opt)
155		kfree(inet->opt);
156	dst_release(sk->sk_dst_cache);
157#ifdef INET_REFCNT_DEBUG
158	atomic_dec(&inet_sock_nr);
159	printk(KERN_DEBUG "INET socket %p released, %d are still alive\n",
160	       sk, atomic_read(&inet_sock_nr));
161#endif
162}
163
164/*
165 *	The routines beyond this point handle the behaviour of an AF_INET
166 *	socket object. Mostly it punts to the subprotocols of IP to do
167 *	the work.
168 */
169
170/*
171 *	Automatically bind an unbound socket.
172 */
173
174static int inet_autobind(struct sock *sk)
175{
176	struct inet_sock *inet;
177	/* We may need to bind the socket. */
178	lock_sock(sk);
179	inet = inet_sk(sk);
180	if (!inet->num) {
181		if (sk->sk_prot->get_port(sk, 0)) {
182			release_sock(sk);
183			return -EAGAIN;
184		}
185		inet->sport = htons(inet->num);
186	}
187	release_sock(sk);
188	return 0;
189}
190
191/*
192 *	Move a socket into listening state.
193 */
194int inet_listen(struct socket *sock, int backlog)
195{
196	struct sock *sk = sock->sk;
197	unsigned char old_state;
198	int err;
199
200	lock_sock(sk);
201
202	err = -EINVAL;
203	if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
204		goto out;
205
206	old_state = sk->sk_state;
207	if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
208		goto out;
209
210	/* Really, if the socket is already in listen state
211	 * we can only allow the backlog to be adjusted.
212	 */
213	if (old_state != TCP_LISTEN) {
214		err = tcp_listen_start(sk);
215		if (err)
216			goto out;
217	}
218	sk->sk_max_ack_backlog = backlog;
219	err = 0;
220
221out:
222	release_sock(sk);
223	return err;
224}
225
226/*
227 *	Create an inet socket.
228 */
229
230static int inet_create(struct socket *sock, int protocol)
231{
232	struct sock *sk;
233	struct list_head *p;
234	struct inet_protosw *answer;
235	struct inet_sock *inet;
236	struct proto *answer_prot;
237	unsigned char answer_flags;
238	char answer_no_check;
239	int err;
240
241	sock->state = SS_UNCONNECTED;
242
243	/* Look for the requested type/protocol pair. */
244	answer = NULL;
245	rcu_read_lock();
246	list_for_each_rcu(p, &inetsw[sock->type]) {
247		answer = list_entry(p, struct inet_protosw, list);
248
249		/* Check the non-wild match. */
250		if (protocol == answer->protocol) {
251			if (protocol != IPPROTO_IP)
252				break;
253		} else {
254			/* Check for the two wild cases. */
255			if (IPPROTO_IP == protocol) {
256				protocol = answer->protocol;
257				break;
258			}
259			if (IPPROTO_IP == answer->protocol)
260				break;
261		}
262		answer = NULL;
263	}
264
265	err = -ESOCKTNOSUPPORT;
266	if (!answer)
267		goto out_rcu_unlock;
268	err = -EPERM;
269	if (answer->capability > 0 && !capable(answer->capability))
270		goto out_rcu_unlock;
271	err = -EPROTONOSUPPORT;
272	if (!protocol)
273		goto out_rcu_unlock;
274
275	sock->ops = answer->ops;
276	answer_prot = answer->prot;
277	answer_no_check = answer->no_check;
278	answer_flags = answer->flags;
279	rcu_read_unlock();
280
281	BUG_TRAP(answer_prot->slab != NULL);
282
283	err = -ENOBUFS;
284	sk = sk_alloc(PF_INET, GFP_KERNEL, answer_prot, 1);
285	if (sk == NULL)
286		goto out;
287
288	err = 0;
289	sk->sk_no_check = answer_no_check;
290	if (INET_PROTOSW_REUSE & answer_flags)
291		sk->sk_reuse = 1;
292
293	inet = inet_sk(sk);
294
295	if (SOCK_RAW == sock->type) {
296		inet->num = protocol;
297		if (IPPROTO_RAW == protocol)
298			inet->hdrincl = 1;
299	}
300
301	if (ipv4_config.no_pmtu_disc)
302		inet->pmtudisc = IP_PMTUDISC_DONT;
303	else
304		inet->pmtudisc = IP_PMTUDISC_WANT;
305
306	inet->id = 0;
307
308	sock_init_data(sock, sk);
309
310	sk->sk_destruct	   = inet_sock_destruct;
311	sk->sk_family	   = PF_INET;
312	sk->sk_protocol	   = protocol;
313	sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
314
315	inet->uc_ttl	= -1;
316	inet->mc_loop	= 1;
317	inet->mc_ttl	= 1;
318	inet->mc_index	= 0;
319	inet->mc_list	= NULL;
320
321#ifdef INET_REFCNT_DEBUG
322	atomic_inc(&inet_sock_nr);
323#endif
324
325	if (inet->num) {
326		/* It assumes that any protocol which allows
327		 * the user to assign a number at socket
328		 * creation time automatically
329		 * shares.
330		 */
331		inet->sport = htons(inet->num);
332		/* Add to protocol hash chains. */
333		sk->sk_prot->hash(sk);
334	}
335
336	if (sk->sk_prot->init) {
337		err = sk->sk_prot->init(sk);
338		if (err)
339			sk_common_release(sk);
340	}
341out:
342	return err;
343out_rcu_unlock:
344	rcu_read_unlock();
345	goto out;
346}
347
348
349/*
350 *	The peer socket should always be NULL (or else). When we call this
351 *	function we are destroying the object and from then on nobody
352 *	should refer to it.
353 */
354int inet_release(struct socket *sock)
355{
356	struct sock *sk = sock->sk;
357
358	if (sk) {
359		long timeout;
360
361		/* Applications forget to leave groups before exiting */
362		ip_mc_drop_socket(sk);
363
364		/* If linger is set, we don't return until the close
365		 * is complete.  Otherwise we return immediately. The
366		 * actually closing is done the same either way.
367		 *
368		 * If the close is due to the process exiting, we never
369		 * linger..
370		 */
371		timeout = 0;
372		if (sock_flag(sk, SOCK_LINGER) &&
373		    !(current->flags & PF_EXITING))
374			timeout = sk->sk_lingertime;
375		sock->sk = NULL;
376		sk->sk_prot->close(sk, timeout);
377	}
378	return 0;
379}
380
381/* It is off by default, see below. */
382int sysctl_ip_nonlocal_bind;
383
384int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
385{
386	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
387	struct sock *sk = sock->sk;
388	struct inet_sock *inet = inet_sk(sk);
389	unsigned short snum;
390	int chk_addr_ret;
391	int err;
392
393	/* If the socket has its own bind function then use it. (RAW) */
394	if (sk->sk_prot->bind) {
395		err = sk->sk_prot->bind(sk, uaddr, addr_len);
396		goto out;
397	}
398	err = -EINVAL;
399	if (addr_len < sizeof(struct sockaddr_in))
400		goto out;
401
402	chk_addr_ret = inet_addr_type(addr->sin_addr.s_addr);
403
404	/* Not specified by any standard per-se, however it breaks too
405	 * many applications when removed.  It is unfortunate since
406	 * allowing applications to make a non-local bind solves
407	 * several problems with systems using dynamic addressing.
408	 * (ie. your servers still start up even if your ISDN link
409	 *  is temporarily down)
410	 */
411	err = -EADDRNOTAVAIL;
412	if (!sysctl_ip_nonlocal_bind &&
413	    !inet->freebind &&
414	    addr->sin_addr.s_addr != INADDR_ANY &&
415	    chk_addr_ret != RTN_LOCAL &&
416	    chk_addr_ret != RTN_MULTICAST &&
417	    chk_addr_ret != RTN_BROADCAST)
418		goto out;
419
420	snum = ntohs(addr->sin_port);
421	err = -EACCES;
422	if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
423		goto out;
424
425	/*      We keep a pair of addresses. rcv_saddr is the one
426	 *      used by hash lookups, and saddr is used for transmit.
427	 *
428	 *      In the BSD API these are the same except where it
429	 *      would be illegal to use them (multicast/broadcast) in
430	 *      which case the sending device address is used.
431	 */
432	lock_sock(sk);
433
434	/* Check these errors (active socket, double bind). */
435	err = -EINVAL;
436	if (sk->sk_state != TCP_CLOSE || inet->num)
437		goto out_release_sock;
438
439	inet->rcv_saddr = inet->saddr = addr->sin_addr.s_addr;
440	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
441		inet->saddr = 0;  /* Use device */
442
443	/* Make sure we are allowed to bind here. */
444	if (sk->sk_prot->get_port(sk, snum)) {
445		inet->saddr = inet->rcv_saddr = 0;
446		err = -EADDRINUSE;
447		goto out_release_sock;
448	}
449
450	if (inet->rcv_saddr)
451		sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
452	if (snum)
453		sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
454	inet->sport = htons(inet->num);
455	inet->daddr = 0;
456	inet->dport = 0;
457	sk_dst_reset(sk);
458	err = 0;
459out_release_sock:
460	release_sock(sk);
461out:
462	return err;
463}
464
465int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
466		       int addr_len, int flags)
467{
468	struct sock *sk = sock->sk;
469
470	if (uaddr->sa_family == AF_UNSPEC)
471		return sk->sk_prot->disconnect(sk, flags);
472
473	if (!inet_sk(sk)->num && inet_autobind(sk))
474		return -EAGAIN;
475	return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
476}
477
478static long inet_wait_for_connect(struct sock *sk, long timeo)
479{
480	DEFINE_WAIT(wait);
481
482	prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
483
484	/* Basic assumption: if someone sets sk->sk_err, he _must_
485	 * change state of the socket from TCP_SYN_*.
486	 * Connect() does not allow to get error notifications
487	 * without closing the socket.
488	 */
489	while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
490		release_sock(sk);
491		timeo = schedule_timeout(timeo);
492		lock_sock(sk);
493		if (signal_pending(current) || !timeo)
494			break;
495		prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
496	}
497	finish_wait(sk->sk_sleep, &wait);
498	return timeo;
499}
500
501/*
502 *	Connect to a remote host. There is regrettably still a little
503 *	TCP 'magic' in here.
504 */
505int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
506			int addr_len, int flags)
507{
508	struct sock *sk = sock->sk;
509	int err;
510	long timeo;
511
512	lock_sock(sk);
513
514	if (uaddr->sa_family == AF_UNSPEC) {
515		err = sk->sk_prot->disconnect(sk, flags);
516		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
517		goto out;
518	}
519
520	switch (sock->state) {
521	default:
522		err = -EINVAL;
523		goto out;
524	case SS_CONNECTED:
525		err = -EISCONN;
526		goto out;
527	case SS_CONNECTING:
528		err = -EALREADY;
529		/* Fall out of switch with err, set for this state */
530		break;
531	case SS_UNCONNECTED:
532		err = -EISCONN;
533		if (sk->sk_state != TCP_CLOSE)
534			goto out;
535
536		err = sk->sk_prot->connect(sk, uaddr, addr_len);
537		if (err < 0)
538			goto out;
539
540  		sock->state = SS_CONNECTING;
541
542		/* Just entered SS_CONNECTING state; the only
543		 * difference is that return value in non-blocking
544		 * case is EINPROGRESS, rather than EALREADY.
545		 */
546		err = -EINPROGRESS;
547		break;
548	}
549
550	timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
551
552	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
553		/* Error code is set above */
554		if (!timeo || !inet_wait_for_connect(sk, timeo))
555			goto out;
556
557		err = sock_intr_errno(timeo);
558		if (signal_pending(current))
559			goto out;
560	}
561
562	/* Connection was closed by RST, timeout, ICMP error
563	 * or another process disconnected us.
564	 */
565	if (sk->sk_state == TCP_CLOSE)
566		goto sock_error;
567
568	/* sk->sk_err may be not zero now, if RECVERR was ordered by user
569	 * and error was received after socket entered established state.
570	 * Hence, it is handled normally after connect() return successfully.
571	 */
572
573	sock->state = SS_CONNECTED;
574	err = 0;
575out:
576	release_sock(sk);
577	return err;
578
579sock_error:
580	err = sock_error(sk) ? : -ECONNABORTED;
581	sock->state = SS_UNCONNECTED;
582	if (sk->sk_prot->disconnect(sk, flags))
583		sock->state = SS_DISCONNECTING;
584	goto out;
585}
586
587/*
588 *	Accept a pending connection. The TCP layer now gives BSD semantics.
589 */
590
591int inet_accept(struct socket *sock, struct socket *newsock, int flags)
592{
593	struct sock *sk1 = sock->sk;
594	int err = -EINVAL;
595	struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
596
597	if (!sk2)
598		goto do_err;
599
600	lock_sock(sk2);
601
602	BUG_TRAP((1 << sk2->sk_state) &
603		 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE));
604
605	sock_graft(sk2, newsock);
606
607	newsock->state = SS_CONNECTED;
608	err = 0;
609	release_sock(sk2);
610do_err:
611	return err;
612}
613
614
615/*
616 *	This does both peername and sockname.
617 */
618int inet_getname(struct socket *sock, struct sockaddr *uaddr,
619			int *uaddr_len, int peer)
620{
621	struct sock *sk		= sock->sk;
622	struct inet_sock *inet	= inet_sk(sk);
623	struct sockaddr_in *sin	= (struct sockaddr_in *)uaddr;
624
625	sin->sin_family = AF_INET;
626	if (peer) {
627		if (!inet->dport ||
628		    (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
629		     peer == 1))
630			return -ENOTCONN;
631		sin->sin_port = inet->dport;
632		sin->sin_addr.s_addr = inet->daddr;
633	} else {
634		__u32 addr = inet->rcv_saddr;
635		if (!addr)
636			addr = inet->saddr;
637		sin->sin_port = inet->sport;
638		sin->sin_addr.s_addr = addr;
639	}
640	memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
641	*uaddr_len = sizeof(*sin);
642	return 0;
643}
644
645int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
646		 size_t size)
647{
648	struct sock *sk = sock->sk;
649
650	/* We may need to bind the socket. */
651	if (!inet_sk(sk)->num && inet_autobind(sk))
652		return -EAGAIN;
653
654	return sk->sk_prot->sendmsg(iocb, sk, msg, size);
655}
656
657
658static ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
659{
660	struct sock *sk = sock->sk;
661
662	/* We may need to bind the socket. */
663	if (!inet_sk(sk)->num && inet_autobind(sk))
664		return -EAGAIN;
665
666	if (sk->sk_prot->sendpage)
667		return sk->sk_prot->sendpage(sk, page, offset, size, flags);
668	return sock_no_sendpage(sock, page, offset, size, flags);
669}
670
671
672int inet_shutdown(struct socket *sock, int how)
673{
674	struct sock *sk = sock->sk;
675	int err = 0;
676
677	/* This should really check to make sure
678	 * the socket is a TCP socket. (WHY AC...)
679	 */
680	how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
681		       1->2 bit 2 snds.
682		       2->3 */
683	if ((how & ~SHUTDOWN_MASK) || !how)	/* MAXINT->0 */
684		return -EINVAL;
685
686	lock_sock(sk);
687	if (sock->state == SS_CONNECTING) {
688		if ((1 << sk->sk_state) &
689		    (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
690			sock->state = SS_DISCONNECTING;
691		else
692			sock->state = SS_CONNECTED;
693	}
694
695	switch (sk->sk_state) {
696	case TCP_CLOSE:
697		err = -ENOTCONN;
698		/* Hack to wake up other listeners, who can poll for
699		   POLLHUP, even on eg. unconnected UDP sockets -- RR */
700	default:
701		sk->sk_shutdown |= how;
702		if (sk->sk_prot->shutdown)
703			sk->sk_prot->shutdown(sk, how);
704		break;
705
706	/* Remaining two branches are temporary solution for missing
707	 * close() in multithreaded environment. It is _not_ a good idea,
708	 * but we have no choice until close() is repaired at VFS level.
709	 */
710	case TCP_LISTEN:
711		if (!(how & RCV_SHUTDOWN))
712			break;
713		/* Fall through */
714	case TCP_SYN_SENT:
715		err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
716		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
717		break;
718	}
719
720	/* Wake up anyone sleeping in poll. */
721	sk->sk_state_change(sk);
722	release_sock(sk);
723	return err;
724}
725
726/*
727 *	ioctl() calls you can issue on an INET socket. Most of these are
728 *	device configuration and stuff and very rarely used. Some ioctls
729 *	pass on to the socket itself.
730 *
731 *	NOTE: I like the idea of a module for the config stuff. ie ifconfig
732 *	loads the devconfigure module does its configuring and unloads it.
733 *	There's a good 20K of config code hanging around the kernel.
734 */
735
736int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
737{
738	struct sock *sk = sock->sk;
739	int err = 0;
740
741	switch (cmd) {
742		case SIOCGSTAMP:
743			err = sock_get_timestamp(sk, (struct timeval __user *)arg);
744			break;
745		case SIOCADDRT:
746		case SIOCDELRT:
747		case SIOCRTMSG:
748			err = ip_rt_ioctl(cmd, (void __user *)arg);
749			break;
750		case SIOCDARP:
751		case SIOCGARP:
752		case SIOCSARP:
753			err = arp_ioctl(cmd, (void __user *)arg);
754			break;
755		case SIOCGIFADDR:
756		case SIOCSIFADDR:
757		case SIOCGIFBRDADDR:
758		case SIOCSIFBRDADDR:
759		case SIOCGIFNETMASK:
760		case SIOCSIFNETMASK:
761		case SIOCGIFDSTADDR:
762		case SIOCSIFDSTADDR:
763		case SIOCSIFPFLAGS:
764		case SIOCGIFPFLAGS:
765		case SIOCSIFFLAGS:
766			err = devinet_ioctl(cmd, (void __user *)arg);
767			break;
768		default:
769			if (!sk->sk_prot->ioctl ||
770			    (err = sk->sk_prot->ioctl(sk, cmd, arg)) ==
771			    					-ENOIOCTLCMD)
772				err = dev_ioctl(cmd, (void __user *)arg);
773			break;
774	}
775	return err;
776}
777
778struct proto_ops inet_stream_ops = {
779	.family =	PF_INET,
780	.owner =	THIS_MODULE,
781	.release =	inet_release,
782	.bind =		inet_bind,
783	.connect =	inet_stream_connect,
784	.socketpair =	sock_no_socketpair,
785	.accept =	inet_accept,
786	.getname =	inet_getname,
787	.poll =		tcp_poll,
788	.ioctl =	inet_ioctl,
789	.listen =	inet_listen,
790	.shutdown =	inet_shutdown,
791	.setsockopt =	sock_common_setsockopt,
792	.getsockopt =	sock_common_getsockopt,
793	.sendmsg =	inet_sendmsg,
794	.recvmsg =	sock_common_recvmsg,
795	.mmap =		sock_no_mmap,
796	.sendpage =	tcp_sendpage
797};
798
799struct proto_ops inet_dgram_ops = {
800	.family =	PF_INET,
801	.owner =	THIS_MODULE,
802	.release =	inet_release,
803	.bind =		inet_bind,
804	.connect =	inet_dgram_connect,
805	.socketpair =	sock_no_socketpair,
806	.accept =	sock_no_accept,
807	.getname =	inet_getname,
808	.poll =		udp_poll,
809	.ioctl =	inet_ioctl,
810	.listen =	sock_no_listen,
811	.shutdown =	inet_shutdown,
812	.setsockopt =	sock_common_setsockopt,
813	.getsockopt =	sock_common_getsockopt,
814	.sendmsg =	inet_sendmsg,
815	.recvmsg =	sock_common_recvmsg,
816	.mmap =		sock_no_mmap,
817	.sendpage =	inet_sendpage,
818};
819
820/*
821 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
822 * udp_poll
823 */
824static struct proto_ops inet_sockraw_ops = {
825	.family =	PF_INET,
826	.owner =	THIS_MODULE,
827	.release =	inet_release,
828	.bind =		inet_bind,
829	.connect =	inet_dgram_connect,
830	.socketpair =	sock_no_socketpair,
831	.accept =	sock_no_accept,
832	.getname =	inet_getname,
833	.poll =		datagram_poll,
834	.ioctl =	inet_ioctl,
835	.listen =	sock_no_listen,
836	.shutdown =	inet_shutdown,
837	.setsockopt =	sock_common_setsockopt,
838	.getsockopt =	sock_common_getsockopt,
839	.sendmsg =	inet_sendmsg,
840	.recvmsg =	sock_common_recvmsg,
841	.mmap =		sock_no_mmap,
842	.sendpage =	inet_sendpage,
843};
844
845static struct net_proto_family inet_family_ops = {
846	.family = PF_INET,
847	.create = inet_create,
848	.owner	= THIS_MODULE,
849};
850
851
852extern void tcp_init(void);
853extern void tcp_v4_init(struct net_proto_family *);
854
855/* Upon startup we insert all the elements in inetsw_array[] into
856 * the linked list inetsw.
857 */
858static struct inet_protosw inetsw_array[] =
859{
860        {
861                .type =       SOCK_STREAM,
862                .protocol =   IPPROTO_TCP,
863                .prot =       &tcp_prot,
864                .ops =        &inet_stream_ops,
865                .capability = -1,
866                .no_check =   0,
867                .flags =      INET_PROTOSW_PERMANENT,
868        },
869
870        {
871                .type =       SOCK_DGRAM,
872                .protocol =   IPPROTO_UDP,
873                .prot =       &udp_prot,
874                .ops =        &inet_dgram_ops,
875                .capability = -1,
876                .no_check =   UDP_CSUM_DEFAULT,
877                .flags =      INET_PROTOSW_PERMANENT,
878       },
879
880
881       {
882               .type =       SOCK_RAW,
883               .protocol =   IPPROTO_IP,	/* wild card */
884               .prot =       &raw_prot,
885               .ops =        &inet_sockraw_ops,
886               .capability = CAP_NET_RAW,
887               .no_check =   UDP_CSUM_DEFAULT,
888               .flags =      INET_PROTOSW_REUSE,
889       }
890};
891
892#define INETSW_ARRAY_LEN (sizeof(inetsw_array) / sizeof(struct inet_protosw))
893
894void inet_register_protosw(struct inet_protosw *p)
895{
896	struct list_head *lh;
897	struct inet_protosw *answer;
898	int protocol = p->protocol;
899	struct list_head *last_perm;
900
901	spin_lock_bh(&inetsw_lock);
902
903	if (p->type >= SOCK_MAX)
904		goto out_illegal;
905
906	/* If we are trying to override a permanent protocol, bail. */
907	answer = NULL;
908	last_perm = &inetsw[p->type];
909	list_for_each(lh, &inetsw[p->type]) {
910		answer = list_entry(lh, struct inet_protosw, list);
911
912		/* Check only the non-wild match. */
913		if (INET_PROTOSW_PERMANENT & answer->flags) {
914			if (protocol == answer->protocol)
915				break;
916			last_perm = lh;
917		}
918
919		answer = NULL;
920	}
921	if (answer)
922		goto out_permanent;
923
924	/* Add the new entry after the last permanent entry if any, so that
925	 * the new entry does not override a permanent entry when matched with
926	 * a wild-card protocol. But it is allowed to override any existing
927	 * non-permanent entry.  This means that when we remove this entry, the
928	 * system automatically returns to the old behavior.
929	 */
930	list_add_rcu(&p->list, last_perm);
931out:
932	spin_unlock_bh(&inetsw_lock);
933
934	synchronize_net();
935
936	return;
937
938out_permanent:
939	printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
940	       protocol);
941	goto out;
942
943out_illegal:
944	printk(KERN_ERR
945	       "Ignoring attempt to register invalid socket type %d.\n",
946	       p->type);
947	goto out;
948}
949
950void inet_unregister_protosw(struct inet_protosw *p)
951{
952	if (INET_PROTOSW_PERMANENT & p->flags) {
953		printk(KERN_ERR
954		       "Attempt to unregister permanent protocol %d.\n",
955		       p->protocol);
956	} else {
957		spin_lock_bh(&inetsw_lock);
958		list_del_rcu(&p->list);
959		spin_unlock_bh(&inetsw_lock);
960
961		synchronize_net();
962	}
963}
964
965#ifdef CONFIG_IP_MULTICAST
966static struct net_protocol igmp_protocol = {
967	.handler =	igmp_rcv,
968};
969#endif
970
971static struct net_protocol tcp_protocol = {
972	.handler =	tcp_v4_rcv,
973	.err_handler =	tcp_v4_err,
974	.no_policy =	1,
975};
976
977static struct net_protocol udp_protocol = {
978	.handler =	udp_rcv,
979	.err_handler =	udp_err,
980	.no_policy =	1,
981};
982
983static struct net_protocol icmp_protocol = {
984	.handler =	icmp_rcv,
985};
986
987static int __init init_ipv4_mibs(void)
988{
989	net_statistics[0] = alloc_percpu(struct linux_mib);
990	net_statistics[1] = alloc_percpu(struct linux_mib);
991	ip_statistics[0] = alloc_percpu(struct ipstats_mib);
992	ip_statistics[1] = alloc_percpu(struct ipstats_mib);
993	icmp_statistics[0] = alloc_percpu(struct icmp_mib);
994	icmp_statistics[1] = alloc_percpu(struct icmp_mib);
995	tcp_statistics[0] = alloc_percpu(struct tcp_mib);
996	tcp_statistics[1] = alloc_percpu(struct tcp_mib);
997	udp_statistics[0] = alloc_percpu(struct udp_mib);
998	udp_statistics[1] = alloc_percpu(struct udp_mib);
999	if (!
1000	    (net_statistics[0] && net_statistics[1] && ip_statistics[0]
1001	     && ip_statistics[1] && tcp_statistics[0] && tcp_statistics[1]
1002	     && udp_statistics[0] && udp_statistics[1]))
1003		return -ENOMEM;
1004
1005	(void) tcp_mib_init();
1006
1007	return 0;
1008}
1009
1010static int ipv4_proc_init(void);
1011extern void ipfrag_init(void);
1012
1013static int __init inet_init(void)
1014{
1015	struct sk_buff *dummy_skb;
1016	struct inet_protosw *q;
1017	struct list_head *r;
1018	int rc = -EINVAL;
1019
1020	if (sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb)) {
1021		printk(KERN_CRIT "%s: panic\n", __FUNCTION__);
1022		goto out;
1023	}
1024
1025	rc = proto_register(&tcp_prot, 1);
1026	if (rc)
1027		goto out;
1028
1029	rc = proto_register(&udp_prot, 1);
1030	if (rc)
1031		goto out_unregister_tcp_proto;
1032
1033	rc = proto_register(&raw_prot, 1);
1034	if (rc)
1035		goto out_unregister_udp_proto;
1036
1037	/*
1038	 *	Tell SOCKET that we are alive...
1039	 */
1040
1041  	(void)sock_register(&inet_family_ops);
1042
1043	/*
1044	 *	Add all the base protocols.
1045	 */
1046
1047	if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1048		printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1049	if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1050		printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1051	if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1052		printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1053#ifdef CONFIG_IP_MULTICAST
1054	if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1055		printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1056#endif
1057
1058	/* Register the socket-side information for inet_create. */
1059	for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1060		INIT_LIST_HEAD(r);
1061
1062	for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1063		inet_register_protosw(q);
1064
1065	/*
1066	 *	Set the ARP module up
1067	 */
1068
1069	arp_init();
1070
1071  	/*
1072  	 *	Set the IP module up
1073  	 */
1074
1075	ip_init();
1076
1077	tcp_v4_init(&inet_family_ops);
1078
1079	/* Setup TCP slab cache for open requests. */
1080	tcp_init();
1081
1082
1083	/*
1084	 *	Set the ICMP layer up
1085	 */
1086
1087	icmp_init(&inet_family_ops);
1088
1089	/*
1090	 *	Initialise the multicast router
1091	 */
1092#if defined(CONFIG_IP_MROUTE)
1093	ip_mr_init();
1094#endif
1095	/*
1096	 *	Initialise per-cpu ipv4 mibs
1097	 */
1098
1099	if(init_ipv4_mibs())
1100		printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n"); ;
1101
1102	ipv4_proc_init();
1103
1104	ipfrag_init();
1105
1106	rc = 0;
1107out:
1108	return rc;
1109out_unregister_tcp_proto:
1110	proto_unregister(&tcp_prot);
1111out_unregister_udp_proto:
1112	proto_unregister(&udp_prot);
1113	goto out;
1114}
1115
1116module_init(inet_init);
1117
1118/* ------------------------------------------------------------------------ */
1119
1120#ifdef CONFIG_PROC_FS
1121extern int  fib_proc_init(void);
1122extern void fib_proc_exit(void);
1123extern int  ip_misc_proc_init(void);
1124extern int  raw_proc_init(void);
1125extern void raw_proc_exit(void);
1126extern int  tcp4_proc_init(void);
1127extern void tcp4_proc_exit(void);
1128extern int  udp4_proc_init(void);
1129extern void udp4_proc_exit(void);
1130
1131static int __init ipv4_proc_init(void)
1132{
1133	int rc = 0;
1134
1135	if (raw_proc_init())
1136		goto out_raw;
1137	if (tcp4_proc_init())
1138		goto out_tcp;
1139	if (udp4_proc_init())
1140		goto out_udp;
1141	if (fib_proc_init())
1142		goto out_fib;
1143	if (ip_misc_proc_init())
1144		goto out_misc;
1145out:
1146	return rc;
1147out_misc:
1148	fib_proc_exit();
1149out_fib:
1150	udp4_proc_exit();
1151out_udp:
1152	tcp4_proc_exit();
1153out_tcp:
1154	raw_proc_exit();
1155out_raw:
1156	rc = -ENOMEM;
1157	goto out;
1158}
1159
1160#else /* CONFIG_PROC_FS */
1161static int __init ipv4_proc_init(void)
1162{
1163	return 0;
1164}
1165#endif /* CONFIG_PROC_FS */
1166
1167MODULE_ALIAS_NETPROTO(PF_INET);
1168
1169EXPORT_SYMBOL(inet_accept);
1170EXPORT_SYMBOL(inet_bind);
1171EXPORT_SYMBOL(inet_dgram_connect);
1172EXPORT_SYMBOL(inet_dgram_ops);
1173EXPORT_SYMBOL(inet_getname);
1174EXPORT_SYMBOL(inet_ioctl);
1175EXPORT_SYMBOL(inet_listen);
1176EXPORT_SYMBOL(inet_register_protosw);
1177EXPORT_SYMBOL(inet_release);
1178EXPORT_SYMBOL(inet_sendmsg);
1179EXPORT_SYMBOL(inet_shutdown);
1180EXPORT_SYMBOL(inet_sock_destruct);
1181EXPORT_SYMBOL(inet_stream_connect);
1182EXPORT_SYMBOL(inet_stream_ops);
1183EXPORT_SYMBOL(inet_unregister_protosw);
1184EXPORT_SYMBOL(net_statistics);
1185
1186#ifdef INET_REFCNT_DEBUG
1187EXPORT_SYMBOL(inet_sock_nr);
1188#endif
1189