xref: /openbmc/linux/net/ipv4/af_inet.c (revision e868d61272caa648214046a096e5a6bfc068dc8c)
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/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 
92 #include <asm/uaccess.h>
93 #include <asm/system.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/ip.h>
100 #include <net/protocol.h>
101 #include <net/arp.h>
102 #include <net/route.h>
103 #include <net/ip_fib.h>
104 #include <net/inet_connection_sock.h>
105 #include <net/tcp.h>
106 #include <net/udp.h>
107 #include <net/udplite.h>
108 #include <linux/skbuff.h>
109 #include <net/sock.h>
110 #include <net/raw.h>
111 #include <net/icmp.h>
112 #include <net/ipip.h>
113 #include <net/inet_common.h>
114 #include <net/xfrm.h>
115 #ifdef CONFIG_IP_MROUTE
116 #include <linux/mroute.h>
117 #endif
118 
119 DEFINE_SNMP_STAT(struct linux_mib, net_statistics) __read_mostly;
120 
121 extern void ip_mc_drop_socket(struct sock *sk);
122 
123 /* The inetsw table contains everything that inet_create needs to
124  * build a new socket.
125  */
126 static struct list_head inetsw[SOCK_MAX];
127 static DEFINE_SPINLOCK(inetsw_lock);
128 
129 /* New destruction routine */
130 
131 void 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 	if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
139 		printk("Attempt to release TCP socket in state %d %p\n",
140 		       sk->sk_state, sk);
141 		return;
142 	}
143 	if (!sock_flag(sk, SOCK_DEAD)) {
144 		printk("Attempt to release alive inet socket %p\n", sk);
145 		return;
146 	}
147 
148 	BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
149 	BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
150 	BUG_TRAP(!sk->sk_wmem_queued);
151 	BUG_TRAP(!sk->sk_forward_alloc);
152 
153 	kfree(inet->opt);
154 	dst_release(sk->sk_dst_cache);
155 	sk_refcnt_debug_dec(sk);
156 }
157 
158 /*
159  *	The routines beyond this point handle the behaviour of an AF_INET
160  *	socket object. Mostly it punts to the subprotocols of IP to do
161  *	the work.
162  */
163 
164 /*
165  *	Automatically bind an unbound socket.
166  */
167 
168 static int inet_autobind(struct sock *sk)
169 {
170 	struct inet_sock *inet;
171 	/* We may need to bind the socket. */
172 	lock_sock(sk);
173 	inet = inet_sk(sk);
174 	if (!inet->num) {
175 		if (sk->sk_prot->get_port(sk, 0)) {
176 			release_sock(sk);
177 			return -EAGAIN;
178 		}
179 		inet->sport = htons(inet->num);
180 	}
181 	release_sock(sk);
182 	return 0;
183 }
184 
185 /*
186  *	Move a socket into listening state.
187  */
188 int inet_listen(struct socket *sock, int backlog)
189 {
190 	struct sock *sk = sock->sk;
191 	unsigned char old_state;
192 	int err;
193 
194 	lock_sock(sk);
195 
196 	err = -EINVAL;
197 	if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
198 		goto out;
199 
200 	old_state = sk->sk_state;
201 	if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
202 		goto out;
203 
204 	/* Really, if the socket is already in listen state
205 	 * we can only allow the backlog to be adjusted.
206 	 */
207 	if (old_state != TCP_LISTEN) {
208 		err = inet_csk_listen_start(sk, backlog);
209 		if (err)
210 			goto out;
211 	}
212 	sk->sk_max_ack_backlog = backlog;
213 	err = 0;
214 
215 out:
216 	release_sock(sk);
217 	return err;
218 }
219 
220 u32 inet_ehash_secret __read_mostly;
221 EXPORT_SYMBOL(inet_ehash_secret);
222 
223 /*
224  * inet_ehash_secret must be set exactly once
225  * Instead of using a dedicated spinlock, we (ab)use inetsw_lock
226  */
227 void build_ehash_secret(void)
228 {
229 	u32 rnd;
230 	do {
231 		get_random_bytes(&rnd, sizeof(rnd));
232 	} while (rnd == 0);
233 	spin_lock_bh(&inetsw_lock);
234 	if (!inet_ehash_secret)
235 		inet_ehash_secret = rnd;
236 	spin_unlock_bh(&inetsw_lock);
237 }
238 EXPORT_SYMBOL(build_ehash_secret);
239 
240 /*
241  *	Create an inet socket.
242  */
243 
244 static int inet_create(struct socket *sock, int protocol)
245 {
246 	struct sock *sk;
247 	struct list_head *p;
248 	struct inet_protosw *answer;
249 	struct inet_sock *inet;
250 	struct proto *answer_prot;
251 	unsigned char answer_flags;
252 	char answer_no_check;
253 	int try_loading_module = 0;
254 	int err;
255 
256 	if (sock->type != SOCK_RAW &&
257 	    sock->type != SOCK_DGRAM &&
258 	    !inet_ehash_secret)
259 		build_ehash_secret();
260 
261 	sock->state = SS_UNCONNECTED;
262 
263 	/* Look for the requested type/protocol pair. */
264 	answer = NULL;
265 lookup_protocol:
266 	err = -ESOCKTNOSUPPORT;
267 	rcu_read_lock();
268 	list_for_each_rcu(p, &inetsw[sock->type]) {
269 		answer = list_entry(p, struct inet_protosw, list);
270 
271 		/* Check the non-wild match. */
272 		if (protocol == answer->protocol) {
273 			if (protocol != IPPROTO_IP)
274 				break;
275 		} else {
276 			/* Check for the two wild cases. */
277 			if (IPPROTO_IP == protocol) {
278 				protocol = answer->protocol;
279 				break;
280 			}
281 			if (IPPROTO_IP == answer->protocol)
282 				break;
283 		}
284 		err = -EPROTONOSUPPORT;
285 		answer = NULL;
286 	}
287 
288 	if (unlikely(answer == NULL)) {
289 		if (try_loading_module < 2) {
290 			rcu_read_unlock();
291 			/*
292 			 * Be more specific, e.g. net-pf-2-proto-132-type-1
293 			 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
294 			 */
295 			if (++try_loading_module == 1)
296 				request_module("net-pf-%d-proto-%d-type-%d",
297 					       PF_INET, protocol, sock->type);
298 			/*
299 			 * Fall back to generic, e.g. net-pf-2-proto-132
300 			 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
301 			 */
302 			else
303 				request_module("net-pf-%d-proto-%d",
304 					       PF_INET, protocol);
305 			goto lookup_protocol;
306 		} else
307 			goto out_rcu_unlock;
308 	}
309 
310 	err = -EPERM;
311 	if (answer->capability > 0 && !capable(answer->capability))
312 		goto out_rcu_unlock;
313 
314 	sock->ops = answer->ops;
315 	answer_prot = answer->prot;
316 	answer_no_check = answer->no_check;
317 	answer_flags = answer->flags;
318 	rcu_read_unlock();
319 
320 	BUG_TRAP(answer_prot->slab != NULL);
321 
322 	err = -ENOBUFS;
323 	sk = sk_alloc(PF_INET, GFP_KERNEL, answer_prot, 1);
324 	if (sk == NULL)
325 		goto out;
326 
327 	err = 0;
328 	sk->sk_no_check = answer_no_check;
329 	if (INET_PROTOSW_REUSE & answer_flags)
330 		sk->sk_reuse = 1;
331 
332 	inet = inet_sk(sk);
333 	inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
334 
335 	if (SOCK_RAW == sock->type) {
336 		inet->num = protocol;
337 		if (IPPROTO_RAW == protocol)
338 			inet->hdrincl = 1;
339 	}
340 
341 	if (ipv4_config.no_pmtu_disc)
342 		inet->pmtudisc = IP_PMTUDISC_DONT;
343 	else
344 		inet->pmtudisc = IP_PMTUDISC_WANT;
345 
346 	inet->id = 0;
347 
348 	sock_init_data(sock, sk);
349 
350 	sk->sk_destruct	   = inet_sock_destruct;
351 	sk->sk_family	   = PF_INET;
352 	sk->sk_protocol	   = protocol;
353 	sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
354 
355 	inet->uc_ttl	= -1;
356 	inet->mc_loop	= 1;
357 	inet->mc_ttl	= 1;
358 	inet->mc_index	= 0;
359 	inet->mc_list	= NULL;
360 
361 	sk_refcnt_debug_inc(sk);
362 
363 	if (inet->num) {
364 		/* It assumes that any protocol which allows
365 		 * the user to assign a number at socket
366 		 * creation time automatically
367 		 * shares.
368 		 */
369 		inet->sport = htons(inet->num);
370 		/* Add to protocol hash chains. */
371 		sk->sk_prot->hash(sk);
372 	}
373 
374 	if (sk->sk_prot->init) {
375 		err = sk->sk_prot->init(sk);
376 		if (err)
377 			sk_common_release(sk);
378 	}
379 out:
380 	return err;
381 out_rcu_unlock:
382 	rcu_read_unlock();
383 	goto out;
384 }
385 
386 
387 /*
388  *	The peer socket should always be NULL (or else). When we call this
389  *	function we are destroying the object and from then on nobody
390  *	should refer to it.
391  */
392 int inet_release(struct socket *sock)
393 {
394 	struct sock *sk = sock->sk;
395 
396 	if (sk) {
397 		long timeout;
398 
399 		/* Applications forget to leave groups before exiting */
400 		ip_mc_drop_socket(sk);
401 
402 		/* If linger is set, we don't return until the close
403 		 * is complete.  Otherwise we return immediately. The
404 		 * actually closing is done the same either way.
405 		 *
406 		 * If the close is due to the process exiting, we never
407 		 * linger..
408 		 */
409 		timeout = 0;
410 		if (sock_flag(sk, SOCK_LINGER) &&
411 		    !(current->flags & PF_EXITING))
412 			timeout = sk->sk_lingertime;
413 		sock->sk = NULL;
414 		sk->sk_prot->close(sk, timeout);
415 	}
416 	return 0;
417 }
418 
419 /* It is off by default, see below. */
420 int sysctl_ip_nonlocal_bind __read_mostly;
421 
422 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
423 {
424 	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
425 	struct sock *sk = sock->sk;
426 	struct inet_sock *inet = inet_sk(sk);
427 	unsigned short snum;
428 	int chk_addr_ret;
429 	int err;
430 
431 	/* If the socket has its own bind function then use it. (RAW) */
432 	if (sk->sk_prot->bind) {
433 		err = sk->sk_prot->bind(sk, uaddr, addr_len);
434 		goto out;
435 	}
436 	err = -EINVAL;
437 	if (addr_len < sizeof(struct sockaddr_in))
438 		goto out;
439 
440 	chk_addr_ret = inet_addr_type(addr->sin_addr.s_addr);
441 
442 	/* Not specified by any standard per-se, however it breaks too
443 	 * many applications when removed.  It is unfortunate since
444 	 * allowing applications to make a non-local bind solves
445 	 * several problems with systems using dynamic addressing.
446 	 * (ie. your servers still start up even if your ISDN link
447 	 *  is temporarily down)
448 	 */
449 	err = -EADDRNOTAVAIL;
450 	if (!sysctl_ip_nonlocal_bind &&
451 	    !inet->freebind &&
452 	    addr->sin_addr.s_addr != INADDR_ANY &&
453 	    chk_addr_ret != RTN_LOCAL &&
454 	    chk_addr_ret != RTN_MULTICAST &&
455 	    chk_addr_ret != RTN_BROADCAST)
456 		goto out;
457 
458 	snum = ntohs(addr->sin_port);
459 	err = -EACCES;
460 	if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
461 		goto out;
462 
463 	/*      We keep a pair of addresses. rcv_saddr is the one
464 	 *      used by hash lookups, and saddr is used for transmit.
465 	 *
466 	 *      In the BSD API these are the same except where it
467 	 *      would be illegal to use them (multicast/broadcast) in
468 	 *      which case the sending device address is used.
469 	 */
470 	lock_sock(sk);
471 
472 	/* Check these errors (active socket, double bind). */
473 	err = -EINVAL;
474 	if (sk->sk_state != TCP_CLOSE || inet->num)
475 		goto out_release_sock;
476 
477 	inet->rcv_saddr = inet->saddr = addr->sin_addr.s_addr;
478 	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
479 		inet->saddr = 0;  /* Use device */
480 
481 	/* Make sure we are allowed to bind here. */
482 	if (sk->sk_prot->get_port(sk, snum)) {
483 		inet->saddr = inet->rcv_saddr = 0;
484 		err = -EADDRINUSE;
485 		goto out_release_sock;
486 	}
487 
488 	if (inet->rcv_saddr)
489 		sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
490 	if (snum)
491 		sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
492 	inet->sport = htons(inet->num);
493 	inet->daddr = 0;
494 	inet->dport = 0;
495 	sk_dst_reset(sk);
496 	err = 0;
497 out_release_sock:
498 	release_sock(sk);
499 out:
500 	return err;
501 }
502 
503 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
504 		       int addr_len, int flags)
505 {
506 	struct sock *sk = sock->sk;
507 
508 	if (uaddr->sa_family == AF_UNSPEC)
509 		return sk->sk_prot->disconnect(sk, flags);
510 
511 	if (!inet_sk(sk)->num && inet_autobind(sk))
512 		return -EAGAIN;
513 	return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
514 }
515 
516 static long inet_wait_for_connect(struct sock *sk, long timeo)
517 {
518 	DEFINE_WAIT(wait);
519 
520 	prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
521 
522 	/* Basic assumption: if someone sets sk->sk_err, he _must_
523 	 * change state of the socket from TCP_SYN_*.
524 	 * Connect() does not allow to get error notifications
525 	 * without closing the socket.
526 	 */
527 	while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
528 		release_sock(sk);
529 		timeo = schedule_timeout(timeo);
530 		lock_sock(sk);
531 		if (signal_pending(current) || !timeo)
532 			break;
533 		prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
534 	}
535 	finish_wait(sk->sk_sleep, &wait);
536 	return timeo;
537 }
538 
539 /*
540  *	Connect to a remote host. There is regrettably still a little
541  *	TCP 'magic' in here.
542  */
543 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
544 			int addr_len, int flags)
545 {
546 	struct sock *sk = sock->sk;
547 	int err;
548 	long timeo;
549 
550 	lock_sock(sk);
551 
552 	if (uaddr->sa_family == AF_UNSPEC) {
553 		err = sk->sk_prot->disconnect(sk, flags);
554 		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
555 		goto out;
556 	}
557 
558 	switch (sock->state) {
559 	default:
560 		err = -EINVAL;
561 		goto out;
562 	case SS_CONNECTED:
563 		err = -EISCONN;
564 		goto out;
565 	case SS_CONNECTING:
566 		err = -EALREADY;
567 		/* Fall out of switch with err, set for this state */
568 		break;
569 	case SS_UNCONNECTED:
570 		err = -EISCONN;
571 		if (sk->sk_state != TCP_CLOSE)
572 			goto out;
573 
574 		err = sk->sk_prot->connect(sk, uaddr, addr_len);
575 		if (err < 0)
576 			goto out;
577 
578 		sock->state = SS_CONNECTING;
579 
580 		/* Just entered SS_CONNECTING state; the only
581 		 * difference is that return value in non-blocking
582 		 * case is EINPROGRESS, rather than EALREADY.
583 		 */
584 		err = -EINPROGRESS;
585 		break;
586 	}
587 
588 	timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
589 
590 	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
591 		/* Error code is set above */
592 		if (!timeo || !inet_wait_for_connect(sk, timeo))
593 			goto out;
594 
595 		err = sock_intr_errno(timeo);
596 		if (signal_pending(current))
597 			goto out;
598 	}
599 
600 	/* Connection was closed by RST, timeout, ICMP error
601 	 * or another process disconnected us.
602 	 */
603 	if (sk->sk_state == TCP_CLOSE)
604 		goto sock_error;
605 
606 	/* sk->sk_err may be not zero now, if RECVERR was ordered by user
607 	 * and error was received after socket entered established state.
608 	 * Hence, it is handled normally after connect() return successfully.
609 	 */
610 
611 	sock->state = SS_CONNECTED;
612 	err = 0;
613 out:
614 	release_sock(sk);
615 	return err;
616 
617 sock_error:
618 	err = sock_error(sk) ? : -ECONNABORTED;
619 	sock->state = SS_UNCONNECTED;
620 	if (sk->sk_prot->disconnect(sk, flags))
621 		sock->state = SS_DISCONNECTING;
622 	goto out;
623 }
624 
625 /*
626  *	Accept a pending connection. The TCP layer now gives BSD semantics.
627  */
628 
629 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
630 {
631 	struct sock *sk1 = sock->sk;
632 	int err = -EINVAL;
633 	struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
634 
635 	if (!sk2)
636 		goto do_err;
637 
638 	lock_sock(sk2);
639 
640 	BUG_TRAP((1 << sk2->sk_state) &
641 		 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE));
642 
643 	sock_graft(sk2, newsock);
644 
645 	newsock->state = SS_CONNECTED;
646 	err = 0;
647 	release_sock(sk2);
648 do_err:
649 	return err;
650 }
651 
652 
653 /*
654  *	This does both peername and sockname.
655  */
656 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
657 			int *uaddr_len, int peer)
658 {
659 	struct sock *sk		= sock->sk;
660 	struct inet_sock *inet	= inet_sk(sk);
661 	struct sockaddr_in *sin	= (struct sockaddr_in *)uaddr;
662 
663 	sin->sin_family = AF_INET;
664 	if (peer) {
665 		if (!inet->dport ||
666 		    (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
667 		     peer == 1))
668 			return -ENOTCONN;
669 		sin->sin_port = inet->dport;
670 		sin->sin_addr.s_addr = inet->daddr;
671 	} else {
672 		__be32 addr = inet->rcv_saddr;
673 		if (!addr)
674 			addr = inet->saddr;
675 		sin->sin_port = inet->sport;
676 		sin->sin_addr.s_addr = addr;
677 	}
678 	memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
679 	*uaddr_len = sizeof(*sin);
680 	return 0;
681 }
682 
683 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
684 		 size_t size)
685 {
686 	struct sock *sk = sock->sk;
687 
688 	/* We may need to bind the socket. */
689 	if (!inet_sk(sk)->num && inet_autobind(sk))
690 		return -EAGAIN;
691 
692 	return sk->sk_prot->sendmsg(iocb, sk, msg, size);
693 }
694 
695 
696 static ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
697 {
698 	struct sock *sk = sock->sk;
699 
700 	/* We may need to bind the socket. */
701 	if (!inet_sk(sk)->num && inet_autobind(sk))
702 		return -EAGAIN;
703 
704 	if (sk->sk_prot->sendpage)
705 		return sk->sk_prot->sendpage(sk, page, offset, size, flags);
706 	return sock_no_sendpage(sock, page, offset, size, flags);
707 }
708 
709 
710 int inet_shutdown(struct socket *sock, int how)
711 {
712 	struct sock *sk = sock->sk;
713 	int err = 0;
714 
715 	/* This should really check to make sure
716 	 * the socket is a TCP socket. (WHY AC...)
717 	 */
718 	how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
719 		       1->2 bit 2 snds.
720 		       2->3 */
721 	if ((how & ~SHUTDOWN_MASK) || !how)	/* MAXINT->0 */
722 		return -EINVAL;
723 
724 	lock_sock(sk);
725 	if (sock->state == SS_CONNECTING) {
726 		if ((1 << sk->sk_state) &
727 		    (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
728 			sock->state = SS_DISCONNECTING;
729 		else
730 			sock->state = SS_CONNECTED;
731 	}
732 
733 	switch (sk->sk_state) {
734 	case TCP_CLOSE:
735 		err = -ENOTCONN;
736 		/* Hack to wake up other listeners, who can poll for
737 		   POLLHUP, even on eg. unconnected UDP sockets -- RR */
738 	default:
739 		sk->sk_shutdown |= how;
740 		if (sk->sk_prot->shutdown)
741 			sk->sk_prot->shutdown(sk, how);
742 		break;
743 
744 	/* Remaining two branches are temporary solution for missing
745 	 * close() in multithreaded environment. It is _not_ a good idea,
746 	 * but we have no choice until close() is repaired at VFS level.
747 	 */
748 	case TCP_LISTEN:
749 		if (!(how & RCV_SHUTDOWN))
750 			break;
751 		/* Fall through */
752 	case TCP_SYN_SENT:
753 		err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
754 		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
755 		break;
756 	}
757 
758 	/* Wake up anyone sleeping in poll. */
759 	sk->sk_state_change(sk);
760 	release_sock(sk);
761 	return err;
762 }
763 
764 /*
765  *	ioctl() calls you can issue on an INET socket. Most of these are
766  *	device configuration and stuff and very rarely used. Some ioctls
767  *	pass on to the socket itself.
768  *
769  *	NOTE: I like the idea of a module for the config stuff. ie ifconfig
770  *	loads the devconfigure module does its configuring and unloads it.
771  *	There's a good 20K of config code hanging around the kernel.
772  */
773 
774 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
775 {
776 	struct sock *sk = sock->sk;
777 	int err = 0;
778 
779 	switch (cmd) {
780 		case SIOCGSTAMP:
781 			err = sock_get_timestamp(sk, (struct timeval __user *)arg);
782 			break;
783 		case SIOCGSTAMPNS:
784 			err = sock_get_timestampns(sk, (struct timespec __user *)arg);
785 			break;
786 		case SIOCADDRT:
787 		case SIOCDELRT:
788 		case SIOCRTMSG:
789 			err = ip_rt_ioctl(cmd, (void __user *)arg);
790 			break;
791 		case SIOCDARP:
792 		case SIOCGARP:
793 		case SIOCSARP:
794 			err = arp_ioctl(cmd, (void __user *)arg);
795 			break;
796 		case SIOCGIFADDR:
797 		case SIOCSIFADDR:
798 		case SIOCGIFBRDADDR:
799 		case SIOCSIFBRDADDR:
800 		case SIOCGIFNETMASK:
801 		case SIOCSIFNETMASK:
802 		case SIOCGIFDSTADDR:
803 		case SIOCSIFDSTADDR:
804 		case SIOCSIFPFLAGS:
805 		case SIOCGIFPFLAGS:
806 		case SIOCSIFFLAGS:
807 			err = devinet_ioctl(cmd, (void __user *)arg);
808 			break;
809 		default:
810 			if (sk->sk_prot->ioctl)
811 				err = sk->sk_prot->ioctl(sk, cmd, arg);
812 			else
813 				err = -ENOIOCTLCMD;
814 			break;
815 	}
816 	return err;
817 }
818 
819 const struct proto_ops inet_stream_ops = {
820 	.family		   = PF_INET,
821 	.owner		   = THIS_MODULE,
822 	.release	   = inet_release,
823 	.bind		   = inet_bind,
824 	.connect	   = inet_stream_connect,
825 	.socketpair	   = sock_no_socketpair,
826 	.accept		   = inet_accept,
827 	.getname	   = inet_getname,
828 	.poll		   = tcp_poll,
829 	.ioctl		   = inet_ioctl,
830 	.listen		   = inet_listen,
831 	.shutdown	   = inet_shutdown,
832 	.setsockopt	   = sock_common_setsockopt,
833 	.getsockopt	   = sock_common_getsockopt,
834 	.sendmsg	   = inet_sendmsg,
835 	.recvmsg	   = sock_common_recvmsg,
836 	.mmap		   = sock_no_mmap,
837 	.sendpage	   = tcp_sendpage,
838 #ifdef CONFIG_COMPAT
839 	.compat_setsockopt = compat_sock_common_setsockopt,
840 	.compat_getsockopt = compat_sock_common_getsockopt,
841 #endif
842 };
843 
844 const struct proto_ops inet_dgram_ops = {
845 	.family		   = PF_INET,
846 	.owner		   = THIS_MODULE,
847 	.release	   = inet_release,
848 	.bind		   = inet_bind,
849 	.connect	   = inet_dgram_connect,
850 	.socketpair	   = sock_no_socketpair,
851 	.accept		   = sock_no_accept,
852 	.getname	   = inet_getname,
853 	.poll		   = udp_poll,
854 	.ioctl		   = inet_ioctl,
855 	.listen		   = sock_no_listen,
856 	.shutdown	   = inet_shutdown,
857 	.setsockopt	   = sock_common_setsockopt,
858 	.getsockopt	   = sock_common_getsockopt,
859 	.sendmsg	   = inet_sendmsg,
860 	.recvmsg	   = sock_common_recvmsg,
861 	.mmap		   = sock_no_mmap,
862 	.sendpage	   = inet_sendpage,
863 #ifdef CONFIG_COMPAT
864 	.compat_setsockopt = compat_sock_common_setsockopt,
865 	.compat_getsockopt = compat_sock_common_getsockopt,
866 #endif
867 };
868 
869 /*
870  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
871  * udp_poll
872  */
873 static const struct proto_ops inet_sockraw_ops = {
874 	.family		   = PF_INET,
875 	.owner		   = THIS_MODULE,
876 	.release	   = inet_release,
877 	.bind		   = inet_bind,
878 	.connect	   = inet_dgram_connect,
879 	.socketpair	   = sock_no_socketpair,
880 	.accept		   = sock_no_accept,
881 	.getname	   = inet_getname,
882 	.poll		   = datagram_poll,
883 	.ioctl		   = inet_ioctl,
884 	.listen		   = sock_no_listen,
885 	.shutdown	   = inet_shutdown,
886 	.setsockopt	   = sock_common_setsockopt,
887 	.getsockopt	   = sock_common_getsockopt,
888 	.sendmsg	   = inet_sendmsg,
889 	.recvmsg	   = sock_common_recvmsg,
890 	.mmap		   = sock_no_mmap,
891 	.sendpage	   = inet_sendpage,
892 #ifdef CONFIG_COMPAT
893 	.compat_setsockopt = compat_sock_common_setsockopt,
894 	.compat_getsockopt = compat_sock_common_getsockopt,
895 #endif
896 };
897 
898 static struct net_proto_family inet_family_ops = {
899 	.family = PF_INET,
900 	.create = inet_create,
901 	.owner	= THIS_MODULE,
902 };
903 
904 /* Upon startup we insert all the elements in inetsw_array[] into
905  * the linked list inetsw.
906  */
907 static struct inet_protosw inetsw_array[] =
908 {
909 	{
910 		.type =       SOCK_STREAM,
911 		.protocol =   IPPROTO_TCP,
912 		.prot =       &tcp_prot,
913 		.ops =        &inet_stream_ops,
914 		.capability = -1,
915 		.no_check =   0,
916 		.flags =      INET_PROTOSW_PERMANENT |
917 			      INET_PROTOSW_ICSK,
918 	},
919 
920 	{
921 		.type =       SOCK_DGRAM,
922 		.protocol =   IPPROTO_UDP,
923 		.prot =       &udp_prot,
924 		.ops =        &inet_dgram_ops,
925 		.capability = -1,
926 		.no_check =   UDP_CSUM_DEFAULT,
927 		.flags =      INET_PROTOSW_PERMANENT,
928        },
929 
930 
931        {
932 	       .type =       SOCK_RAW,
933 	       .protocol =   IPPROTO_IP,	/* wild card */
934 	       .prot =       &raw_prot,
935 	       .ops =        &inet_sockraw_ops,
936 	       .capability = CAP_NET_RAW,
937 	       .no_check =   UDP_CSUM_DEFAULT,
938 	       .flags =      INET_PROTOSW_REUSE,
939        }
940 };
941 
942 #define INETSW_ARRAY_LEN (sizeof(inetsw_array) / sizeof(struct inet_protosw))
943 
944 void inet_register_protosw(struct inet_protosw *p)
945 {
946 	struct list_head *lh;
947 	struct inet_protosw *answer;
948 	int protocol = p->protocol;
949 	struct list_head *last_perm;
950 
951 	spin_lock_bh(&inetsw_lock);
952 
953 	if (p->type >= SOCK_MAX)
954 		goto out_illegal;
955 
956 	/* If we are trying to override a permanent protocol, bail. */
957 	answer = NULL;
958 	last_perm = &inetsw[p->type];
959 	list_for_each(lh, &inetsw[p->type]) {
960 		answer = list_entry(lh, struct inet_protosw, list);
961 
962 		/* Check only the non-wild match. */
963 		if (INET_PROTOSW_PERMANENT & answer->flags) {
964 			if (protocol == answer->protocol)
965 				break;
966 			last_perm = lh;
967 		}
968 
969 		answer = NULL;
970 	}
971 	if (answer)
972 		goto out_permanent;
973 
974 	/* Add the new entry after the last permanent entry if any, so that
975 	 * the new entry does not override a permanent entry when matched with
976 	 * a wild-card protocol. But it is allowed to override any existing
977 	 * non-permanent entry.  This means that when we remove this entry, the
978 	 * system automatically returns to the old behavior.
979 	 */
980 	list_add_rcu(&p->list, last_perm);
981 out:
982 	spin_unlock_bh(&inetsw_lock);
983 
984 	synchronize_net();
985 
986 	return;
987 
988 out_permanent:
989 	printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
990 	       protocol);
991 	goto out;
992 
993 out_illegal:
994 	printk(KERN_ERR
995 	       "Ignoring attempt to register invalid socket type %d.\n",
996 	       p->type);
997 	goto out;
998 }
999 
1000 void inet_unregister_protosw(struct inet_protosw *p)
1001 {
1002 	if (INET_PROTOSW_PERMANENT & p->flags) {
1003 		printk(KERN_ERR
1004 		       "Attempt to unregister permanent protocol %d.\n",
1005 		       p->protocol);
1006 	} else {
1007 		spin_lock_bh(&inetsw_lock);
1008 		list_del_rcu(&p->list);
1009 		spin_unlock_bh(&inetsw_lock);
1010 
1011 		synchronize_net();
1012 	}
1013 }
1014 
1015 /*
1016  *      Shall we try to damage output packets if routing dev changes?
1017  */
1018 
1019 int sysctl_ip_dynaddr __read_mostly;
1020 
1021 static int inet_sk_reselect_saddr(struct sock *sk)
1022 {
1023 	struct inet_sock *inet = inet_sk(sk);
1024 	int err;
1025 	struct rtable *rt;
1026 	__be32 old_saddr = inet->saddr;
1027 	__be32 new_saddr;
1028 	__be32 daddr = inet->daddr;
1029 
1030 	if (inet->opt && inet->opt->srr)
1031 		daddr = inet->opt->faddr;
1032 
1033 	/* Query new route. */
1034 	err = ip_route_connect(&rt, daddr, 0,
1035 			       RT_CONN_FLAGS(sk),
1036 			       sk->sk_bound_dev_if,
1037 			       sk->sk_protocol,
1038 			       inet->sport, inet->dport, sk, 0);
1039 	if (err)
1040 		return err;
1041 
1042 	sk_setup_caps(sk, &rt->u.dst);
1043 
1044 	new_saddr = rt->rt_src;
1045 
1046 	if (new_saddr == old_saddr)
1047 		return 0;
1048 
1049 	if (sysctl_ip_dynaddr > 1) {
1050 		printk(KERN_INFO "%s(): shifting inet->"
1051 				 "saddr from %d.%d.%d.%d to %d.%d.%d.%d\n",
1052 		       __FUNCTION__,
1053 		       NIPQUAD(old_saddr),
1054 		       NIPQUAD(new_saddr));
1055 	}
1056 
1057 	inet->saddr = inet->rcv_saddr = new_saddr;
1058 
1059 	/*
1060 	 * XXX The only one ugly spot where we need to
1061 	 * XXX really change the sockets identity after
1062 	 * XXX it has entered the hashes. -DaveM
1063 	 *
1064 	 * Besides that, it does not check for connection
1065 	 * uniqueness. Wait for troubles.
1066 	 */
1067 	__sk_prot_rehash(sk);
1068 	return 0;
1069 }
1070 
1071 int inet_sk_rebuild_header(struct sock *sk)
1072 {
1073 	struct inet_sock *inet = inet_sk(sk);
1074 	struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1075 	__be32 daddr;
1076 	int err;
1077 
1078 	/* Route is OK, nothing to do. */
1079 	if (rt)
1080 		return 0;
1081 
1082 	/* Reroute. */
1083 	daddr = inet->daddr;
1084 	if (inet->opt && inet->opt->srr)
1085 		daddr = inet->opt->faddr;
1086 {
1087 	struct flowi fl = {
1088 		.oif = sk->sk_bound_dev_if,
1089 		.nl_u = {
1090 			.ip4_u = {
1091 				.daddr	= daddr,
1092 				.saddr	= inet->saddr,
1093 				.tos	= RT_CONN_FLAGS(sk),
1094 			},
1095 		},
1096 		.proto = sk->sk_protocol,
1097 		.uli_u = {
1098 			.ports = {
1099 				.sport = inet->sport,
1100 				.dport = inet->dport,
1101 			},
1102 		},
1103 	};
1104 
1105 	security_sk_classify_flow(sk, &fl);
1106 	err = ip_route_output_flow(&rt, &fl, sk, 0);
1107 }
1108 	if (!err)
1109 		sk_setup_caps(sk, &rt->u.dst);
1110 	else {
1111 		/* Routing failed... */
1112 		sk->sk_route_caps = 0;
1113 		/*
1114 		 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1115 		 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1116 		 */
1117 		if (!sysctl_ip_dynaddr ||
1118 		    sk->sk_state != TCP_SYN_SENT ||
1119 		    (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1120 		    (err = inet_sk_reselect_saddr(sk)) != 0)
1121 			sk->sk_err_soft = -err;
1122 	}
1123 
1124 	return err;
1125 }
1126 
1127 EXPORT_SYMBOL(inet_sk_rebuild_header);
1128 
1129 static int inet_gso_send_check(struct sk_buff *skb)
1130 {
1131 	struct iphdr *iph;
1132 	struct net_protocol *ops;
1133 	int proto;
1134 	int ihl;
1135 	int err = -EINVAL;
1136 
1137 	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1138 		goto out;
1139 
1140 	iph = ip_hdr(skb);
1141 	ihl = iph->ihl * 4;
1142 	if (ihl < sizeof(*iph))
1143 		goto out;
1144 
1145 	if (unlikely(!pskb_may_pull(skb, ihl)))
1146 		goto out;
1147 
1148 	__skb_pull(skb, ihl);
1149 	skb_reset_transport_header(skb);
1150 	iph = ip_hdr(skb);
1151 	proto = iph->protocol & (MAX_INET_PROTOS - 1);
1152 	err = -EPROTONOSUPPORT;
1153 
1154 	rcu_read_lock();
1155 	ops = rcu_dereference(inet_protos[proto]);
1156 	if (likely(ops && ops->gso_send_check))
1157 		err = ops->gso_send_check(skb);
1158 	rcu_read_unlock();
1159 
1160 out:
1161 	return err;
1162 }
1163 
1164 static struct sk_buff *inet_gso_segment(struct sk_buff *skb, int features)
1165 {
1166 	struct sk_buff *segs = ERR_PTR(-EINVAL);
1167 	struct iphdr *iph;
1168 	struct net_protocol *ops;
1169 	int proto;
1170 	int ihl;
1171 	int id;
1172 
1173 	if (unlikely(skb_shinfo(skb)->gso_type &
1174 		     ~(SKB_GSO_TCPV4 |
1175 		       SKB_GSO_UDP |
1176 		       SKB_GSO_DODGY |
1177 		       SKB_GSO_TCP_ECN |
1178 		       0)))
1179 		goto out;
1180 
1181 	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1182 		goto out;
1183 
1184 	iph = ip_hdr(skb);
1185 	ihl = iph->ihl * 4;
1186 	if (ihl < sizeof(*iph))
1187 		goto out;
1188 
1189 	if (unlikely(!pskb_may_pull(skb, ihl)))
1190 		goto out;
1191 
1192 	__skb_pull(skb, ihl);
1193 	skb_reset_transport_header(skb);
1194 	iph = ip_hdr(skb);
1195 	id = ntohs(iph->id);
1196 	proto = iph->protocol & (MAX_INET_PROTOS - 1);
1197 	segs = ERR_PTR(-EPROTONOSUPPORT);
1198 
1199 	rcu_read_lock();
1200 	ops = rcu_dereference(inet_protos[proto]);
1201 	if (likely(ops && ops->gso_segment))
1202 		segs = ops->gso_segment(skb, features);
1203 	rcu_read_unlock();
1204 
1205 	if (!segs || unlikely(IS_ERR(segs)))
1206 		goto out;
1207 
1208 	skb = segs;
1209 	do {
1210 		iph = ip_hdr(skb);
1211 		iph->id = htons(id++);
1212 		iph->tot_len = htons(skb->len - skb->mac_len);
1213 		iph->check = 0;
1214 		iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1215 	} while ((skb = skb->next));
1216 
1217 out:
1218 	return segs;
1219 }
1220 
1221 unsigned long snmp_fold_field(void *mib[], int offt)
1222 {
1223 	unsigned long res = 0;
1224 	int i;
1225 
1226 	for_each_possible_cpu(i) {
1227 		res += *(((unsigned long *) per_cpu_ptr(mib[0], i)) + offt);
1228 		res += *(((unsigned long *) per_cpu_ptr(mib[1], i)) + offt);
1229 	}
1230 	return res;
1231 }
1232 EXPORT_SYMBOL_GPL(snmp_fold_field);
1233 
1234 int snmp_mib_init(void *ptr[2], size_t mibsize, size_t mibalign)
1235 {
1236 	BUG_ON(ptr == NULL);
1237 	ptr[0] = __alloc_percpu(mibsize);
1238 	if (!ptr[0])
1239 		goto err0;
1240 	ptr[1] = __alloc_percpu(mibsize);
1241 	if (!ptr[1])
1242 		goto err1;
1243 	return 0;
1244 err1:
1245 	free_percpu(ptr[0]);
1246 	ptr[0] = NULL;
1247 err0:
1248 	return -ENOMEM;
1249 }
1250 EXPORT_SYMBOL_GPL(snmp_mib_init);
1251 
1252 void snmp_mib_free(void *ptr[2])
1253 {
1254 	BUG_ON(ptr == NULL);
1255 	free_percpu(ptr[0]);
1256 	free_percpu(ptr[1]);
1257 	ptr[0] = ptr[1] = NULL;
1258 }
1259 EXPORT_SYMBOL_GPL(snmp_mib_free);
1260 
1261 #ifdef CONFIG_IP_MULTICAST
1262 static struct net_protocol igmp_protocol = {
1263 	.handler =	igmp_rcv,
1264 };
1265 #endif
1266 
1267 static struct net_protocol tcp_protocol = {
1268 	.handler =	tcp_v4_rcv,
1269 	.err_handler =	tcp_v4_err,
1270 	.gso_send_check = tcp_v4_gso_send_check,
1271 	.gso_segment =	tcp_tso_segment,
1272 	.no_policy =	1,
1273 };
1274 
1275 static struct net_protocol udp_protocol = {
1276 	.handler =	udp_rcv,
1277 	.err_handler =	udp_err,
1278 	.no_policy =	1,
1279 };
1280 
1281 static struct net_protocol icmp_protocol = {
1282 	.handler =	icmp_rcv,
1283 };
1284 
1285 static int __init init_ipv4_mibs(void)
1286 {
1287 	if (snmp_mib_init((void **)net_statistics,
1288 			  sizeof(struct linux_mib),
1289 			  __alignof__(struct linux_mib)) < 0)
1290 		goto err_net_mib;
1291 	if (snmp_mib_init((void **)ip_statistics,
1292 			  sizeof(struct ipstats_mib),
1293 			  __alignof__(struct ipstats_mib)) < 0)
1294 		goto err_ip_mib;
1295 	if (snmp_mib_init((void **)icmp_statistics,
1296 			  sizeof(struct icmp_mib),
1297 			  __alignof__(struct icmp_mib)) < 0)
1298 		goto err_icmp_mib;
1299 	if (snmp_mib_init((void **)tcp_statistics,
1300 			  sizeof(struct tcp_mib),
1301 			  __alignof__(struct tcp_mib)) < 0)
1302 		goto err_tcp_mib;
1303 	if (snmp_mib_init((void **)udp_statistics,
1304 			  sizeof(struct udp_mib),
1305 			  __alignof__(struct udp_mib)) < 0)
1306 		goto err_udp_mib;
1307 	if (snmp_mib_init((void **)udplite_statistics,
1308 			  sizeof(struct udp_mib),
1309 			  __alignof__(struct udp_mib)) < 0)
1310 		goto err_udplite_mib;
1311 
1312 	tcp_mib_init();
1313 
1314 	return 0;
1315 
1316 err_udplite_mib:
1317 	snmp_mib_free((void **)udp_statistics);
1318 err_udp_mib:
1319 	snmp_mib_free((void **)tcp_statistics);
1320 err_tcp_mib:
1321 	snmp_mib_free((void **)icmp_statistics);
1322 err_icmp_mib:
1323 	snmp_mib_free((void **)ip_statistics);
1324 err_ip_mib:
1325 	snmp_mib_free((void **)net_statistics);
1326 err_net_mib:
1327 	return -ENOMEM;
1328 }
1329 
1330 static int ipv4_proc_init(void);
1331 
1332 /*
1333  *	IP protocol layer initialiser
1334  */
1335 
1336 static struct packet_type ip_packet_type = {
1337 	.type = __constant_htons(ETH_P_IP),
1338 	.func = ip_rcv,
1339 	.gso_send_check = inet_gso_send_check,
1340 	.gso_segment = inet_gso_segment,
1341 };
1342 
1343 static int __init inet_init(void)
1344 {
1345 	struct sk_buff *dummy_skb;
1346 	struct inet_protosw *q;
1347 	struct list_head *r;
1348 	int rc = -EINVAL;
1349 
1350 	BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1351 
1352 	rc = proto_register(&tcp_prot, 1);
1353 	if (rc)
1354 		goto out;
1355 
1356 	rc = proto_register(&udp_prot, 1);
1357 	if (rc)
1358 		goto out_unregister_tcp_proto;
1359 
1360 	rc = proto_register(&raw_prot, 1);
1361 	if (rc)
1362 		goto out_unregister_udp_proto;
1363 
1364 	/*
1365 	 *	Tell SOCKET that we are alive...
1366 	 */
1367 
1368 	(void)sock_register(&inet_family_ops);
1369 
1370 	/*
1371 	 *	Add all the base protocols.
1372 	 */
1373 
1374 	if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1375 		printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1376 	if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1377 		printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1378 	if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1379 		printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1380 #ifdef CONFIG_IP_MULTICAST
1381 	if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1382 		printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1383 #endif
1384 
1385 	/* Register the socket-side information for inet_create. */
1386 	for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1387 		INIT_LIST_HEAD(r);
1388 
1389 	for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1390 		inet_register_protosw(q);
1391 
1392 	/*
1393 	 *	Set the ARP module up
1394 	 */
1395 
1396 	arp_init();
1397 
1398 	/*
1399 	 *	Set the IP module up
1400 	 */
1401 
1402 	ip_init();
1403 
1404 	tcp_v4_init(&inet_family_ops);
1405 
1406 	/* Setup TCP slab cache for open requests. */
1407 	tcp_init();
1408 
1409 	/* Add UDP-Lite (RFC 3828) */
1410 	udplite4_register();
1411 
1412 	/*
1413 	 *	Set the ICMP layer up
1414 	 */
1415 
1416 	icmp_init(&inet_family_ops);
1417 
1418 	/*
1419 	 *	Initialise the multicast router
1420 	 */
1421 #if defined(CONFIG_IP_MROUTE)
1422 	ip_mr_init();
1423 #endif
1424 	/*
1425 	 *	Initialise per-cpu ipv4 mibs
1426 	 */
1427 
1428 	if (init_ipv4_mibs())
1429 		printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n"); ;
1430 
1431 	ipv4_proc_init();
1432 
1433 	ipfrag_init();
1434 
1435 	dev_add_pack(&ip_packet_type);
1436 
1437 	rc = 0;
1438 out:
1439 	return rc;
1440 out_unregister_udp_proto:
1441 	proto_unregister(&udp_prot);
1442 out_unregister_tcp_proto:
1443 	proto_unregister(&tcp_prot);
1444 	goto out;
1445 }
1446 
1447 fs_initcall(inet_init);
1448 
1449 /* ------------------------------------------------------------------------ */
1450 
1451 #ifdef CONFIG_PROC_FS
1452 static int __init ipv4_proc_init(void)
1453 {
1454 	int rc = 0;
1455 
1456 	if (raw_proc_init())
1457 		goto out_raw;
1458 	if (tcp4_proc_init())
1459 		goto out_tcp;
1460 	if (udp4_proc_init())
1461 		goto out_udp;
1462 	if (fib_proc_init())
1463 		goto out_fib;
1464 	if (ip_misc_proc_init())
1465 		goto out_misc;
1466 out:
1467 	return rc;
1468 out_misc:
1469 	fib_proc_exit();
1470 out_fib:
1471 	udp4_proc_exit();
1472 out_udp:
1473 	tcp4_proc_exit();
1474 out_tcp:
1475 	raw_proc_exit();
1476 out_raw:
1477 	rc = -ENOMEM;
1478 	goto out;
1479 }
1480 
1481 #else /* CONFIG_PROC_FS */
1482 static int __init ipv4_proc_init(void)
1483 {
1484 	return 0;
1485 }
1486 #endif /* CONFIG_PROC_FS */
1487 
1488 MODULE_ALIAS_NETPROTO(PF_INET);
1489 
1490 EXPORT_SYMBOL(inet_accept);
1491 EXPORT_SYMBOL(inet_bind);
1492 EXPORT_SYMBOL(inet_dgram_connect);
1493 EXPORT_SYMBOL(inet_dgram_ops);
1494 EXPORT_SYMBOL(inet_getname);
1495 EXPORT_SYMBOL(inet_ioctl);
1496 EXPORT_SYMBOL(inet_listen);
1497 EXPORT_SYMBOL(inet_register_protosw);
1498 EXPORT_SYMBOL(inet_release);
1499 EXPORT_SYMBOL(inet_sendmsg);
1500 EXPORT_SYMBOL(inet_shutdown);
1501 EXPORT_SYMBOL(inet_sock_destruct);
1502 EXPORT_SYMBOL(inet_stream_connect);
1503 EXPORT_SYMBOL(inet_stream_ops);
1504 EXPORT_SYMBOL(inet_unregister_protosw);
1505 EXPORT_SYMBOL(net_statistics);
1506 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
1507