xref: /openbmc/linux/net/ipv4/af_inet.c (revision 545e4006)
1 /*
2  * INET		An implementation of the TCP/IP protocol suite for the LINUX
3  *		operating system.  INET is implemented using the  BSD Socket
4  *		interface as the means of communication with the user level.
5  *
6  *		PF_INET protocol family socket handler.
7  *
8  * Authors:	Ross Biro
9  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10  *		Florian La Roche, <flla@stud.uni-sb.de>
11  *		Alan Cox, <A.Cox@swansea.ac.uk>
12  *
13  * Changes (see also sock.c)
14  *
15  *		piggy,
16  *		Karl Knutson	:	Socket protocol table
17  *		A.N.Kuznetsov	:	Socket death error in accept().
18  *		John Richardson :	Fix non blocking error in connect()
19  *					so sockets that fail to connect
20  *					don't return -EINPROGRESS.
21  *		Alan Cox	:	Asynchronous I/O support
22  *		Alan Cox	:	Keep correct socket pointer on sock
23  *					structures
24  *					when accept() ed
25  *		Alan Cox	:	Semantics of SO_LINGER aren't state
26  *					moved to close when you look carefully.
27  *					With this fixed and the accept bug fixed
28  *					some RPC stuff seems happier.
29  *		Niibe Yutaka	:	4.4BSD style write async I/O
30  *		Alan Cox,
31  *		Tony Gale 	:	Fixed reuse semantics.
32  *		Alan Cox	:	bind() shouldn't abort existing but dead
33  *					sockets. Stops FTP netin:.. I hope.
34  *		Alan Cox	:	bind() works correctly for RAW sockets.
35  *					Note that FreeBSD at least was broken
36  *					in this respect so be careful with
37  *					compatibility tests...
38  *		Alan Cox	:	routing cache support
39  *		Alan Cox	:	memzero the socket structure for
40  *					compactness.
41  *		Matt Day	:	nonblock connect error handler
42  *		Alan Cox	:	Allow large numbers of pending sockets
43  *					(eg for big web sites), but only if
44  *					specifically application requested.
45  *		Alan Cox	:	New buffering throughout IP. Used
46  *					dumbly.
47  *		Alan Cox	:	New buffering now used smartly.
48  *		Alan Cox	:	BSD rather than common sense
49  *					interpretation of listen.
50  *		Germano Caronni	:	Assorted small races.
51  *		Alan Cox	:	sendmsg/recvmsg basic support.
52  *		Alan Cox	:	Only sendmsg/recvmsg now supported.
53  *		Alan Cox	:	Locked down bind (see security list).
54  *		Alan Cox	:	Loosened bind a little.
55  *		Mike McLagan	:	ADD/DEL DLCI Ioctls
56  *	Willy Konynenberg	:	Transparent proxying support.
57  *		David S. Miller	:	New socket lookup architecture.
58  *					Some other random speedups.
59  *		Cyrus Durgin	:	Cleaned up file for kmod hacks.
60  *		Andi Kleen	:	Fix inet_stream_connect TCP race.
61  *
62  *		This program is free software; you can redistribute it and/or
63  *		modify it under the terms of the GNU General Public License
64  *		as published by the Free Software Foundation; either version
65  *		2 of the License, or (at your option) any later version.
66  */
67 
68 #include <linux/err.h>
69 #include <linux/errno.h>
70 #include <linux/types.h>
71 #include <linux/socket.h>
72 #include <linux/in.h>
73 #include <linux/kernel.h>
74 #include <linux/module.h>
75 #include <linux/sched.h>
76 #include <linux/timer.h>
77 #include <linux/string.h>
78 #include <linux/sockios.h>
79 #include <linux/net.h>
80 #include <linux/capability.h>
81 #include <linux/fcntl.h>
82 #include <linux/mm.h>
83 #include <linux/interrupt.h>
84 #include <linux/stat.h>
85 #include <linux/init.h>
86 #include <linux/poll.h>
87 #include <linux/netfilter_ipv4.h>
88 #include <linux/random.h>
89 
90 #include <asm/uaccess.h>
91 #include <asm/system.h>
92 
93 #include <linux/inet.h>
94 #include <linux/igmp.h>
95 #include <linux/inetdevice.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 <net/udplite.h>
106 #include <linux/skbuff.h>
107 #include <net/sock.h>
108 #include <net/raw.h>
109 #include <net/icmp.h>
110 #include <net/ipip.h>
111 #include <net/inet_common.h>
112 #include <net/xfrm.h>
113 #include <net/net_namespace.h>
114 #ifdef CONFIG_IP_MROUTE
115 #include <linux/mroute.h>
116 #endif
117 
118 extern 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  */
123 static struct list_head inetsw[SOCK_MAX];
124 static DEFINE_SPINLOCK(inetsw_lock);
125 
126 struct ipv4_config ipv4_config;
127 
128 EXPORT_SYMBOL(ipv4_config);
129 
130 /* New destruction routine */
131 
132 void 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 	sk_mem_reclaim(sk);
140 
141 	if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
142 		printk("Attempt to release TCP socket in state %d %p\n",
143 		       sk->sk_state, sk);
144 		return;
145 	}
146 	if (!sock_flag(sk, SOCK_DEAD)) {
147 		printk("Attempt to release alive inet socket %p\n", sk);
148 		return;
149 	}
150 
151 	BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
152 	BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
153 	BUG_TRAP(!sk->sk_wmem_queued);
154 	BUG_TRAP(!sk->sk_forward_alloc);
155 
156 	kfree(inet->opt);
157 	dst_release(sk->sk_dst_cache);
158 	sk_refcnt_debug_dec(sk);
159 }
160 
161 /*
162  *	The routines beyond this point handle the behaviour of an AF_INET
163  *	socket object. Mostly it punts to the subprotocols of IP to do
164  *	the work.
165  */
166 
167 /*
168  *	Automatically bind an unbound socket.
169  */
170 
171 static int inet_autobind(struct sock *sk)
172 {
173 	struct inet_sock *inet;
174 	/* We may need to bind the socket. */
175 	lock_sock(sk);
176 	inet = inet_sk(sk);
177 	if (!inet->num) {
178 		if (sk->sk_prot->get_port(sk, 0)) {
179 			release_sock(sk);
180 			return -EAGAIN;
181 		}
182 		inet->sport = htons(inet->num);
183 	}
184 	release_sock(sk);
185 	return 0;
186 }
187 
188 /*
189  *	Move a socket into listening state.
190  */
191 int inet_listen(struct socket *sock, int backlog)
192 {
193 	struct sock *sk = sock->sk;
194 	unsigned char old_state;
195 	int err;
196 
197 	lock_sock(sk);
198 
199 	err = -EINVAL;
200 	if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
201 		goto out;
202 
203 	old_state = sk->sk_state;
204 	if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
205 		goto out;
206 
207 	/* Really, if the socket is already in listen state
208 	 * we can only allow the backlog to be adjusted.
209 	 */
210 	if (old_state != TCP_LISTEN) {
211 		err = inet_csk_listen_start(sk, backlog);
212 		if (err)
213 			goto out;
214 	}
215 	sk->sk_max_ack_backlog = backlog;
216 	err = 0;
217 
218 out:
219 	release_sock(sk);
220 	return err;
221 }
222 
223 u32 inet_ehash_secret __read_mostly;
224 EXPORT_SYMBOL(inet_ehash_secret);
225 
226 /*
227  * inet_ehash_secret must be set exactly once
228  * Instead of using a dedicated spinlock, we (ab)use inetsw_lock
229  */
230 void build_ehash_secret(void)
231 {
232 	u32 rnd;
233 	do {
234 		get_random_bytes(&rnd, sizeof(rnd));
235 	} while (rnd == 0);
236 	spin_lock_bh(&inetsw_lock);
237 	if (!inet_ehash_secret)
238 		inet_ehash_secret = rnd;
239 	spin_unlock_bh(&inetsw_lock);
240 }
241 EXPORT_SYMBOL(build_ehash_secret);
242 
243 static inline int inet_netns_ok(struct net *net, int protocol)
244 {
245 	int hash;
246 	struct net_protocol *ipprot;
247 
248 	if (net == &init_net)
249 		return 1;
250 
251 	hash = protocol & (MAX_INET_PROTOS - 1);
252 	ipprot = rcu_dereference(inet_protos[hash]);
253 
254 	if (ipprot == NULL)
255 		/* raw IP is OK */
256 		return 1;
257 	return ipprot->netns_ok;
258 }
259 
260 /*
261  *	Create an inet socket.
262  */
263 
264 static int inet_create(struct net *net, struct socket *sock, int protocol)
265 {
266 	struct sock *sk;
267 	struct list_head *p;
268 	struct inet_protosw *answer;
269 	struct inet_sock *inet;
270 	struct proto *answer_prot;
271 	unsigned char answer_flags;
272 	char answer_no_check;
273 	int try_loading_module = 0;
274 	int err;
275 
276 	if (sock->type != SOCK_RAW &&
277 	    sock->type != SOCK_DGRAM &&
278 	    !inet_ehash_secret)
279 		build_ehash_secret();
280 
281 	sock->state = SS_UNCONNECTED;
282 
283 	/* Look for the requested type/protocol pair. */
284 	answer = NULL;
285 lookup_protocol:
286 	err = -ESOCKTNOSUPPORT;
287 	rcu_read_lock();
288 	list_for_each_rcu(p, &inetsw[sock->type]) {
289 		answer = list_entry(p, struct inet_protosw, list);
290 
291 		/* Check the non-wild match. */
292 		if (protocol == answer->protocol) {
293 			if (protocol != IPPROTO_IP)
294 				break;
295 		} else {
296 			/* Check for the two wild cases. */
297 			if (IPPROTO_IP == protocol) {
298 				protocol = answer->protocol;
299 				break;
300 			}
301 			if (IPPROTO_IP == answer->protocol)
302 				break;
303 		}
304 		err = -EPROTONOSUPPORT;
305 		answer = NULL;
306 	}
307 
308 	if (unlikely(answer == NULL)) {
309 		if (try_loading_module < 2) {
310 			rcu_read_unlock();
311 			/*
312 			 * Be more specific, e.g. net-pf-2-proto-132-type-1
313 			 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
314 			 */
315 			if (++try_loading_module == 1)
316 				request_module("net-pf-%d-proto-%d-type-%d",
317 					       PF_INET, protocol, sock->type);
318 			/*
319 			 * Fall back to generic, e.g. net-pf-2-proto-132
320 			 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
321 			 */
322 			else
323 				request_module("net-pf-%d-proto-%d",
324 					       PF_INET, protocol);
325 			goto lookup_protocol;
326 		} else
327 			goto out_rcu_unlock;
328 	}
329 
330 	err = -EPERM;
331 	if (answer->capability > 0 && !capable(answer->capability))
332 		goto out_rcu_unlock;
333 
334 	err = -EAFNOSUPPORT;
335 	if (!inet_netns_ok(net, protocol))
336 		goto out_rcu_unlock;
337 
338 	sock->ops = answer->ops;
339 	answer_prot = answer->prot;
340 	answer_no_check = answer->no_check;
341 	answer_flags = answer->flags;
342 	rcu_read_unlock();
343 
344 	BUG_TRAP(answer_prot->slab != NULL);
345 
346 	err = -ENOBUFS;
347 	sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
348 	if (sk == NULL)
349 		goto out;
350 
351 	err = 0;
352 	sk->sk_no_check = answer_no_check;
353 	if (INET_PROTOSW_REUSE & answer_flags)
354 		sk->sk_reuse = 1;
355 
356 	inet = inet_sk(sk);
357 	inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
358 
359 	if (SOCK_RAW == sock->type) {
360 		inet->num = protocol;
361 		if (IPPROTO_RAW == protocol)
362 			inet->hdrincl = 1;
363 	}
364 
365 	if (ipv4_config.no_pmtu_disc)
366 		inet->pmtudisc = IP_PMTUDISC_DONT;
367 	else
368 		inet->pmtudisc = IP_PMTUDISC_WANT;
369 
370 	inet->id = 0;
371 
372 	sock_init_data(sock, sk);
373 
374 	sk->sk_destruct	   = inet_sock_destruct;
375 	sk->sk_family	   = PF_INET;
376 	sk->sk_protocol	   = protocol;
377 	sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
378 
379 	inet->uc_ttl	= -1;
380 	inet->mc_loop	= 1;
381 	inet->mc_ttl	= 1;
382 	inet->mc_index	= 0;
383 	inet->mc_list	= NULL;
384 
385 	sk_refcnt_debug_inc(sk);
386 
387 	if (inet->num) {
388 		/* It assumes that any protocol which allows
389 		 * the user to assign a number at socket
390 		 * creation time automatically
391 		 * shares.
392 		 */
393 		inet->sport = htons(inet->num);
394 		/* Add to protocol hash chains. */
395 		sk->sk_prot->hash(sk);
396 	}
397 
398 	if (sk->sk_prot->init) {
399 		err = sk->sk_prot->init(sk);
400 		if (err)
401 			sk_common_release(sk);
402 	}
403 out:
404 	return err;
405 out_rcu_unlock:
406 	rcu_read_unlock();
407 	goto out;
408 }
409 
410 
411 /*
412  *	The peer socket should always be NULL (or else). When we call this
413  *	function we are destroying the object and from then on nobody
414  *	should refer to it.
415  */
416 int inet_release(struct socket *sock)
417 {
418 	struct sock *sk = sock->sk;
419 
420 	if (sk) {
421 		long timeout;
422 
423 		/* Applications forget to leave groups before exiting */
424 		ip_mc_drop_socket(sk);
425 
426 		/* If linger is set, we don't return until the close
427 		 * is complete.  Otherwise we return immediately. The
428 		 * actually closing is done the same either way.
429 		 *
430 		 * If the close is due to the process exiting, we never
431 		 * linger..
432 		 */
433 		timeout = 0;
434 		if (sock_flag(sk, SOCK_LINGER) &&
435 		    !(current->flags & PF_EXITING))
436 			timeout = sk->sk_lingertime;
437 		sock->sk = NULL;
438 		sk->sk_prot->close(sk, timeout);
439 	}
440 	return 0;
441 }
442 
443 /* It is off by default, see below. */
444 int sysctl_ip_nonlocal_bind __read_mostly;
445 
446 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
447 {
448 	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
449 	struct sock *sk = sock->sk;
450 	struct inet_sock *inet = inet_sk(sk);
451 	unsigned short snum;
452 	int chk_addr_ret;
453 	int err;
454 
455 	/* If the socket has its own bind function then use it. (RAW) */
456 	if (sk->sk_prot->bind) {
457 		err = sk->sk_prot->bind(sk, uaddr, addr_len);
458 		goto out;
459 	}
460 	err = -EINVAL;
461 	if (addr_len < sizeof(struct sockaddr_in))
462 		goto out;
463 
464 	chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr);
465 
466 	/* Not specified by any standard per-se, however it breaks too
467 	 * many applications when removed.  It is unfortunate since
468 	 * allowing applications to make a non-local bind solves
469 	 * several problems with systems using dynamic addressing.
470 	 * (ie. your servers still start up even if your ISDN link
471 	 *  is temporarily down)
472 	 */
473 	err = -EADDRNOTAVAIL;
474 	if (!sysctl_ip_nonlocal_bind &&
475 	    !inet->freebind &&
476 	    addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
477 	    chk_addr_ret != RTN_LOCAL &&
478 	    chk_addr_ret != RTN_MULTICAST &&
479 	    chk_addr_ret != RTN_BROADCAST)
480 		goto out;
481 
482 	snum = ntohs(addr->sin_port);
483 	err = -EACCES;
484 	if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
485 		goto out;
486 
487 	/*      We keep a pair of addresses. rcv_saddr is the one
488 	 *      used by hash lookups, and saddr is used for transmit.
489 	 *
490 	 *      In the BSD API these are the same except where it
491 	 *      would be illegal to use them (multicast/broadcast) in
492 	 *      which case the sending device address is used.
493 	 */
494 	lock_sock(sk);
495 
496 	/* Check these errors (active socket, double bind). */
497 	err = -EINVAL;
498 	if (sk->sk_state != TCP_CLOSE || inet->num)
499 		goto out_release_sock;
500 
501 	inet->rcv_saddr = inet->saddr = addr->sin_addr.s_addr;
502 	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
503 		inet->saddr = 0;  /* Use device */
504 
505 	/* Make sure we are allowed to bind here. */
506 	if (sk->sk_prot->get_port(sk, snum)) {
507 		inet->saddr = inet->rcv_saddr = 0;
508 		err = -EADDRINUSE;
509 		goto out_release_sock;
510 	}
511 
512 	if (inet->rcv_saddr)
513 		sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
514 	if (snum)
515 		sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
516 	inet->sport = htons(inet->num);
517 	inet->daddr = 0;
518 	inet->dport = 0;
519 	sk_dst_reset(sk);
520 	err = 0;
521 out_release_sock:
522 	release_sock(sk);
523 out:
524 	return err;
525 }
526 
527 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
528 		       int addr_len, int flags)
529 {
530 	struct sock *sk = sock->sk;
531 
532 	if (uaddr->sa_family == AF_UNSPEC)
533 		return sk->sk_prot->disconnect(sk, flags);
534 
535 	if (!inet_sk(sk)->num && inet_autobind(sk))
536 		return -EAGAIN;
537 	return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
538 }
539 
540 static long inet_wait_for_connect(struct sock *sk, long timeo)
541 {
542 	DEFINE_WAIT(wait);
543 
544 	prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
545 
546 	/* Basic assumption: if someone sets sk->sk_err, he _must_
547 	 * change state of the socket from TCP_SYN_*.
548 	 * Connect() does not allow to get error notifications
549 	 * without closing the socket.
550 	 */
551 	while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
552 		release_sock(sk);
553 		timeo = schedule_timeout(timeo);
554 		lock_sock(sk);
555 		if (signal_pending(current) || !timeo)
556 			break;
557 		prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
558 	}
559 	finish_wait(sk->sk_sleep, &wait);
560 	return timeo;
561 }
562 
563 /*
564  *	Connect to a remote host. There is regrettably still a little
565  *	TCP 'magic' in here.
566  */
567 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
568 			int addr_len, int flags)
569 {
570 	struct sock *sk = sock->sk;
571 	int err;
572 	long timeo;
573 
574 	lock_sock(sk);
575 
576 	if (uaddr->sa_family == AF_UNSPEC) {
577 		err = sk->sk_prot->disconnect(sk, flags);
578 		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
579 		goto out;
580 	}
581 
582 	switch (sock->state) {
583 	default:
584 		err = -EINVAL;
585 		goto out;
586 	case SS_CONNECTED:
587 		err = -EISCONN;
588 		goto out;
589 	case SS_CONNECTING:
590 		err = -EALREADY;
591 		/* Fall out of switch with err, set for this state */
592 		break;
593 	case SS_UNCONNECTED:
594 		err = -EISCONN;
595 		if (sk->sk_state != TCP_CLOSE)
596 			goto out;
597 
598 		err = sk->sk_prot->connect(sk, uaddr, addr_len);
599 		if (err < 0)
600 			goto out;
601 
602 		sock->state = SS_CONNECTING;
603 
604 		/* Just entered SS_CONNECTING state; the only
605 		 * difference is that return value in non-blocking
606 		 * case is EINPROGRESS, rather than EALREADY.
607 		 */
608 		err = -EINPROGRESS;
609 		break;
610 	}
611 
612 	timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
613 
614 	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
615 		/* Error code is set above */
616 		if (!timeo || !inet_wait_for_connect(sk, timeo))
617 			goto out;
618 
619 		err = sock_intr_errno(timeo);
620 		if (signal_pending(current))
621 			goto out;
622 	}
623 
624 	/* Connection was closed by RST, timeout, ICMP error
625 	 * or another process disconnected us.
626 	 */
627 	if (sk->sk_state == TCP_CLOSE)
628 		goto sock_error;
629 
630 	/* sk->sk_err may be not zero now, if RECVERR was ordered by user
631 	 * and error was received after socket entered established state.
632 	 * Hence, it is handled normally after connect() return successfully.
633 	 */
634 
635 	sock->state = SS_CONNECTED;
636 	err = 0;
637 out:
638 	release_sock(sk);
639 	return err;
640 
641 sock_error:
642 	err = sock_error(sk) ? : -ECONNABORTED;
643 	sock->state = SS_UNCONNECTED;
644 	if (sk->sk_prot->disconnect(sk, flags))
645 		sock->state = SS_DISCONNECTING;
646 	goto out;
647 }
648 
649 /*
650  *	Accept a pending connection. The TCP layer now gives BSD semantics.
651  */
652 
653 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
654 {
655 	struct sock *sk1 = sock->sk;
656 	int err = -EINVAL;
657 	struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
658 
659 	if (!sk2)
660 		goto do_err;
661 
662 	lock_sock(sk2);
663 
664 	BUG_TRAP((1 << sk2->sk_state) &
665 		 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE));
666 
667 	sock_graft(sk2, newsock);
668 
669 	newsock->state = SS_CONNECTED;
670 	err = 0;
671 	release_sock(sk2);
672 do_err:
673 	return err;
674 }
675 
676 
677 /*
678  *	This does both peername and sockname.
679  */
680 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
681 			int *uaddr_len, int peer)
682 {
683 	struct sock *sk		= sock->sk;
684 	struct inet_sock *inet	= inet_sk(sk);
685 	struct sockaddr_in *sin	= (struct sockaddr_in *)uaddr;
686 
687 	sin->sin_family = AF_INET;
688 	if (peer) {
689 		if (!inet->dport ||
690 		    (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
691 		     peer == 1))
692 			return -ENOTCONN;
693 		sin->sin_port = inet->dport;
694 		sin->sin_addr.s_addr = inet->daddr;
695 	} else {
696 		__be32 addr = inet->rcv_saddr;
697 		if (!addr)
698 			addr = inet->saddr;
699 		sin->sin_port = inet->sport;
700 		sin->sin_addr.s_addr = addr;
701 	}
702 	memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
703 	*uaddr_len = sizeof(*sin);
704 	return 0;
705 }
706 
707 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
708 		 size_t size)
709 {
710 	struct sock *sk = sock->sk;
711 
712 	/* We may need to bind the socket. */
713 	if (!inet_sk(sk)->num && inet_autobind(sk))
714 		return -EAGAIN;
715 
716 	return sk->sk_prot->sendmsg(iocb, sk, msg, size);
717 }
718 
719 
720 static ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
721 {
722 	struct sock *sk = sock->sk;
723 
724 	/* We may need to bind the socket. */
725 	if (!inet_sk(sk)->num && inet_autobind(sk))
726 		return -EAGAIN;
727 
728 	if (sk->sk_prot->sendpage)
729 		return sk->sk_prot->sendpage(sk, page, offset, size, flags);
730 	return sock_no_sendpage(sock, page, offset, size, flags);
731 }
732 
733 
734 int inet_shutdown(struct socket *sock, int how)
735 {
736 	struct sock *sk = sock->sk;
737 	int err = 0;
738 
739 	/* This should really check to make sure
740 	 * the socket is a TCP socket. (WHY AC...)
741 	 */
742 	how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
743 		       1->2 bit 2 snds.
744 		       2->3 */
745 	if ((how & ~SHUTDOWN_MASK) || !how)	/* MAXINT->0 */
746 		return -EINVAL;
747 
748 	lock_sock(sk);
749 	if (sock->state == SS_CONNECTING) {
750 		if ((1 << sk->sk_state) &
751 		    (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
752 			sock->state = SS_DISCONNECTING;
753 		else
754 			sock->state = SS_CONNECTED;
755 	}
756 
757 	switch (sk->sk_state) {
758 	case TCP_CLOSE:
759 		err = -ENOTCONN;
760 		/* Hack to wake up other listeners, who can poll for
761 		   POLLHUP, even on eg. unconnected UDP sockets -- RR */
762 	default:
763 		sk->sk_shutdown |= how;
764 		if (sk->sk_prot->shutdown)
765 			sk->sk_prot->shutdown(sk, how);
766 		break;
767 
768 	/* Remaining two branches are temporary solution for missing
769 	 * close() in multithreaded environment. It is _not_ a good idea,
770 	 * but we have no choice until close() is repaired at VFS level.
771 	 */
772 	case TCP_LISTEN:
773 		if (!(how & RCV_SHUTDOWN))
774 			break;
775 		/* Fall through */
776 	case TCP_SYN_SENT:
777 		err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
778 		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
779 		break;
780 	}
781 
782 	/* Wake up anyone sleeping in poll. */
783 	sk->sk_state_change(sk);
784 	release_sock(sk);
785 	return err;
786 }
787 
788 /*
789  *	ioctl() calls you can issue on an INET socket. Most of these are
790  *	device configuration and stuff and very rarely used. Some ioctls
791  *	pass on to the socket itself.
792  *
793  *	NOTE: I like the idea of a module for the config stuff. ie ifconfig
794  *	loads the devconfigure module does its configuring and unloads it.
795  *	There's a good 20K of config code hanging around the kernel.
796  */
797 
798 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
799 {
800 	struct sock *sk = sock->sk;
801 	int err = 0;
802 	struct net *net = sock_net(sk);
803 
804 	switch (cmd) {
805 		case SIOCGSTAMP:
806 			err = sock_get_timestamp(sk, (struct timeval __user *)arg);
807 			break;
808 		case SIOCGSTAMPNS:
809 			err = sock_get_timestampns(sk, (struct timespec __user *)arg);
810 			break;
811 		case SIOCADDRT:
812 		case SIOCDELRT:
813 		case SIOCRTMSG:
814 			err = ip_rt_ioctl(net, cmd, (void __user *)arg);
815 			break;
816 		case SIOCDARP:
817 		case SIOCGARP:
818 		case SIOCSARP:
819 			err = arp_ioctl(net, cmd, (void __user *)arg);
820 			break;
821 		case SIOCGIFADDR:
822 		case SIOCSIFADDR:
823 		case SIOCGIFBRDADDR:
824 		case SIOCSIFBRDADDR:
825 		case SIOCGIFNETMASK:
826 		case SIOCSIFNETMASK:
827 		case SIOCGIFDSTADDR:
828 		case SIOCSIFDSTADDR:
829 		case SIOCSIFPFLAGS:
830 		case SIOCGIFPFLAGS:
831 		case SIOCSIFFLAGS:
832 			err = devinet_ioctl(net, cmd, (void __user *)arg);
833 			break;
834 		default:
835 			if (sk->sk_prot->ioctl)
836 				err = sk->sk_prot->ioctl(sk, cmd, arg);
837 			else
838 				err = -ENOIOCTLCMD;
839 			break;
840 	}
841 	return err;
842 }
843 
844 const struct proto_ops inet_stream_ops = {
845 	.family		   = PF_INET,
846 	.owner		   = THIS_MODULE,
847 	.release	   = inet_release,
848 	.bind		   = inet_bind,
849 	.connect	   = inet_stream_connect,
850 	.socketpair	   = sock_no_socketpair,
851 	.accept		   = inet_accept,
852 	.getname	   = inet_getname,
853 	.poll		   = tcp_poll,
854 	.ioctl		   = inet_ioctl,
855 	.listen		   = inet_listen,
856 	.shutdown	   = inet_shutdown,
857 	.setsockopt	   = sock_common_setsockopt,
858 	.getsockopt	   = sock_common_getsockopt,
859 	.sendmsg	   = tcp_sendmsg,
860 	.recvmsg	   = sock_common_recvmsg,
861 	.mmap		   = sock_no_mmap,
862 	.sendpage	   = tcp_sendpage,
863 	.splice_read	   = tcp_splice_read,
864 #ifdef CONFIG_COMPAT
865 	.compat_setsockopt = compat_sock_common_setsockopt,
866 	.compat_getsockopt = compat_sock_common_getsockopt,
867 #endif
868 };
869 
870 const struct proto_ops inet_dgram_ops = {
871 	.family		   = PF_INET,
872 	.owner		   = THIS_MODULE,
873 	.release	   = inet_release,
874 	.bind		   = inet_bind,
875 	.connect	   = inet_dgram_connect,
876 	.socketpair	   = sock_no_socketpair,
877 	.accept		   = sock_no_accept,
878 	.getname	   = inet_getname,
879 	.poll		   = udp_poll,
880 	.ioctl		   = inet_ioctl,
881 	.listen		   = sock_no_listen,
882 	.shutdown	   = inet_shutdown,
883 	.setsockopt	   = sock_common_setsockopt,
884 	.getsockopt	   = sock_common_getsockopt,
885 	.sendmsg	   = inet_sendmsg,
886 	.recvmsg	   = sock_common_recvmsg,
887 	.mmap		   = sock_no_mmap,
888 	.sendpage	   = inet_sendpage,
889 #ifdef CONFIG_COMPAT
890 	.compat_setsockopt = compat_sock_common_setsockopt,
891 	.compat_getsockopt = compat_sock_common_getsockopt,
892 #endif
893 };
894 
895 /*
896  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
897  * udp_poll
898  */
899 static const struct proto_ops inet_sockraw_ops = {
900 	.family		   = PF_INET,
901 	.owner		   = THIS_MODULE,
902 	.release	   = inet_release,
903 	.bind		   = inet_bind,
904 	.connect	   = inet_dgram_connect,
905 	.socketpair	   = sock_no_socketpair,
906 	.accept		   = sock_no_accept,
907 	.getname	   = inet_getname,
908 	.poll		   = datagram_poll,
909 	.ioctl		   = inet_ioctl,
910 	.listen		   = sock_no_listen,
911 	.shutdown	   = inet_shutdown,
912 	.setsockopt	   = sock_common_setsockopt,
913 	.getsockopt	   = sock_common_getsockopt,
914 	.sendmsg	   = inet_sendmsg,
915 	.recvmsg	   = sock_common_recvmsg,
916 	.mmap		   = sock_no_mmap,
917 	.sendpage	   = inet_sendpage,
918 #ifdef CONFIG_COMPAT
919 	.compat_setsockopt = compat_sock_common_setsockopt,
920 	.compat_getsockopt = compat_sock_common_getsockopt,
921 #endif
922 };
923 
924 static struct net_proto_family inet_family_ops = {
925 	.family = PF_INET,
926 	.create = inet_create,
927 	.owner	= THIS_MODULE,
928 };
929 
930 /* Upon startup we insert all the elements in inetsw_array[] into
931  * the linked list inetsw.
932  */
933 static struct inet_protosw inetsw_array[] =
934 {
935 	{
936 		.type =       SOCK_STREAM,
937 		.protocol =   IPPROTO_TCP,
938 		.prot =       &tcp_prot,
939 		.ops =        &inet_stream_ops,
940 		.capability = -1,
941 		.no_check =   0,
942 		.flags =      INET_PROTOSW_PERMANENT |
943 			      INET_PROTOSW_ICSK,
944 	},
945 
946 	{
947 		.type =       SOCK_DGRAM,
948 		.protocol =   IPPROTO_UDP,
949 		.prot =       &udp_prot,
950 		.ops =        &inet_dgram_ops,
951 		.capability = -1,
952 		.no_check =   UDP_CSUM_DEFAULT,
953 		.flags =      INET_PROTOSW_PERMANENT,
954        },
955 
956 
957        {
958 	       .type =       SOCK_RAW,
959 	       .protocol =   IPPROTO_IP,	/* wild card */
960 	       .prot =       &raw_prot,
961 	       .ops =        &inet_sockraw_ops,
962 	       .capability = CAP_NET_RAW,
963 	       .no_check =   UDP_CSUM_DEFAULT,
964 	       .flags =      INET_PROTOSW_REUSE,
965        }
966 };
967 
968 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
969 
970 void inet_register_protosw(struct inet_protosw *p)
971 {
972 	struct list_head *lh;
973 	struct inet_protosw *answer;
974 	int protocol = p->protocol;
975 	struct list_head *last_perm;
976 
977 	spin_lock_bh(&inetsw_lock);
978 
979 	if (p->type >= SOCK_MAX)
980 		goto out_illegal;
981 
982 	/* If we are trying to override a permanent protocol, bail. */
983 	answer = NULL;
984 	last_perm = &inetsw[p->type];
985 	list_for_each(lh, &inetsw[p->type]) {
986 		answer = list_entry(lh, struct inet_protosw, list);
987 
988 		/* Check only the non-wild match. */
989 		if (INET_PROTOSW_PERMANENT & answer->flags) {
990 			if (protocol == answer->protocol)
991 				break;
992 			last_perm = lh;
993 		}
994 
995 		answer = NULL;
996 	}
997 	if (answer)
998 		goto out_permanent;
999 
1000 	/* Add the new entry after the last permanent entry if any, so that
1001 	 * the new entry does not override a permanent entry when matched with
1002 	 * a wild-card protocol. But it is allowed to override any existing
1003 	 * non-permanent entry.  This means that when we remove this entry, the
1004 	 * system automatically returns to the old behavior.
1005 	 */
1006 	list_add_rcu(&p->list, last_perm);
1007 out:
1008 	spin_unlock_bh(&inetsw_lock);
1009 
1010 	synchronize_net();
1011 
1012 	return;
1013 
1014 out_permanent:
1015 	printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
1016 	       protocol);
1017 	goto out;
1018 
1019 out_illegal:
1020 	printk(KERN_ERR
1021 	       "Ignoring attempt to register invalid socket type %d.\n",
1022 	       p->type);
1023 	goto out;
1024 }
1025 
1026 void inet_unregister_protosw(struct inet_protosw *p)
1027 {
1028 	if (INET_PROTOSW_PERMANENT & p->flags) {
1029 		printk(KERN_ERR
1030 		       "Attempt to unregister permanent protocol %d.\n",
1031 		       p->protocol);
1032 	} else {
1033 		spin_lock_bh(&inetsw_lock);
1034 		list_del_rcu(&p->list);
1035 		spin_unlock_bh(&inetsw_lock);
1036 
1037 		synchronize_net();
1038 	}
1039 }
1040 
1041 /*
1042  *      Shall we try to damage output packets if routing dev changes?
1043  */
1044 
1045 int sysctl_ip_dynaddr __read_mostly;
1046 
1047 static int inet_sk_reselect_saddr(struct sock *sk)
1048 {
1049 	struct inet_sock *inet = inet_sk(sk);
1050 	int err;
1051 	struct rtable *rt;
1052 	__be32 old_saddr = inet->saddr;
1053 	__be32 new_saddr;
1054 	__be32 daddr = inet->daddr;
1055 
1056 	if (inet->opt && inet->opt->srr)
1057 		daddr = inet->opt->faddr;
1058 
1059 	/* Query new route. */
1060 	err = ip_route_connect(&rt, daddr, 0,
1061 			       RT_CONN_FLAGS(sk),
1062 			       sk->sk_bound_dev_if,
1063 			       sk->sk_protocol,
1064 			       inet->sport, inet->dport, sk, 0);
1065 	if (err)
1066 		return err;
1067 
1068 	sk_setup_caps(sk, &rt->u.dst);
1069 
1070 	new_saddr = rt->rt_src;
1071 
1072 	if (new_saddr == old_saddr)
1073 		return 0;
1074 
1075 	if (sysctl_ip_dynaddr > 1) {
1076 		printk(KERN_INFO "%s(): shifting inet->"
1077 				 "saddr from " NIPQUAD_FMT " to " NIPQUAD_FMT "\n",
1078 		       __func__,
1079 		       NIPQUAD(old_saddr),
1080 		       NIPQUAD(new_saddr));
1081 	}
1082 
1083 	inet->saddr = inet->rcv_saddr = new_saddr;
1084 
1085 	/*
1086 	 * XXX The only one ugly spot where we need to
1087 	 * XXX really change the sockets identity after
1088 	 * XXX it has entered the hashes. -DaveM
1089 	 *
1090 	 * Besides that, it does not check for connection
1091 	 * uniqueness. Wait for troubles.
1092 	 */
1093 	__sk_prot_rehash(sk);
1094 	return 0;
1095 }
1096 
1097 int inet_sk_rebuild_header(struct sock *sk)
1098 {
1099 	struct inet_sock *inet = inet_sk(sk);
1100 	struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1101 	__be32 daddr;
1102 	int err;
1103 
1104 	/* Route is OK, nothing to do. */
1105 	if (rt)
1106 		return 0;
1107 
1108 	/* Reroute. */
1109 	daddr = inet->daddr;
1110 	if (inet->opt && inet->opt->srr)
1111 		daddr = inet->opt->faddr;
1112 {
1113 	struct flowi fl = {
1114 		.oif = sk->sk_bound_dev_if,
1115 		.nl_u = {
1116 			.ip4_u = {
1117 				.daddr	= daddr,
1118 				.saddr	= inet->saddr,
1119 				.tos	= RT_CONN_FLAGS(sk),
1120 			},
1121 		},
1122 		.proto = sk->sk_protocol,
1123 		.uli_u = {
1124 			.ports = {
1125 				.sport = inet->sport,
1126 				.dport = inet->dport,
1127 			},
1128 		},
1129 	};
1130 
1131 	security_sk_classify_flow(sk, &fl);
1132 	err = ip_route_output_flow(sock_net(sk), &rt, &fl, sk, 0);
1133 }
1134 	if (!err)
1135 		sk_setup_caps(sk, &rt->u.dst);
1136 	else {
1137 		/* Routing failed... */
1138 		sk->sk_route_caps = 0;
1139 		/*
1140 		 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1141 		 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1142 		 */
1143 		if (!sysctl_ip_dynaddr ||
1144 		    sk->sk_state != TCP_SYN_SENT ||
1145 		    (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1146 		    (err = inet_sk_reselect_saddr(sk)) != 0)
1147 			sk->sk_err_soft = -err;
1148 	}
1149 
1150 	return err;
1151 }
1152 
1153 EXPORT_SYMBOL(inet_sk_rebuild_header);
1154 
1155 static int inet_gso_send_check(struct sk_buff *skb)
1156 {
1157 	struct iphdr *iph;
1158 	struct net_protocol *ops;
1159 	int proto;
1160 	int ihl;
1161 	int err = -EINVAL;
1162 
1163 	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1164 		goto out;
1165 
1166 	iph = ip_hdr(skb);
1167 	ihl = iph->ihl * 4;
1168 	if (ihl < sizeof(*iph))
1169 		goto out;
1170 
1171 	if (unlikely(!pskb_may_pull(skb, ihl)))
1172 		goto out;
1173 
1174 	__skb_pull(skb, ihl);
1175 	skb_reset_transport_header(skb);
1176 	iph = ip_hdr(skb);
1177 	proto = iph->protocol & (MAX_INET_PROTOS - 1);
1178 	err = -EPROTONOSUPPORT;
1179 
1180 	rcu_read_lock();
1181 	ops = rcu_dereference(inet_protos[proto]);
1182 	if (likely(ops && ops->gso_send_check))
1183 		err = ops->gso_send_check(skb);
1184 	rcu_read_unlock();
1185 
1186 out:
1187 	return err;
1188 }
1189 
1190 static struct sk_buff *inet_gso_segment(struct sk_buff *skb, int features)
1191 {
1192 	struct sk_buff *segs = ERR_PTR(-EINVAL);
1193 	struct iphdr *iph;
1194 	struct net_protocol *ops;
1195 	int proto;
1196 	int ihl;
1197 	int id;
1198 
1199 	if (!(features & NETIF_F_V4_CSUM))
1200 		features &= ~NETIF_F_SG;
1201 
1202 	if (unlikely(skb_shinfo(skb)->gso_type &
1203 		     ~(SKB_GSO_TCPV4 |
1204 		       SKB_GSO_UDP |
1205 		       SKB_GSO_DODGY |
1206 		       SKB_GSO_TCP_ECN |
1207 		       0)))
1208 		goto out;
1209 
1210 	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1211 		goto out;
1212 
1213 	iph = ip_hdr(skb);
1214 	ihl = iph->ihl * 4;
1215 	if (ihl < sizeof(*iph))
1216 		goto out;
1217 
1218 	if (unlikely(!pskb_may_pull(skb, ihl)))
1219 		goto out;
1220 
1221 	__skb_pull(skb, ihl);
1222 	skb_reset_transport_header(skb);
1223 	iph = ip_hdr(skb);
1224 	id = ntohs(iph->id);
1225 	proto = iph->protocol & (MAX_INET_PROTOS - 1);
1226 	segs = ERR_PTR(-EPROTONOSUPPORT);
1227 
1228 	rcu_read_lock();
1229 	ops = rcu_dereference(inet_protos[proto]);
1230 	if (likely(ops && ops->gso_segment))
1231 		segs = ops->gso_segment(skb, features);
1232 	rcu_read_unlock();
1233 
1234 	if (!segs || IS_ERR(segs))
1235 		goto out;
1236 
1237 	skb = segs;
1238 	do {
1239 		iph = ip_hdr(skb);
1240 		iph->id = htons(id++);
1241 		iph->tot_len = htons(skb->len - skb->mac_len);
1242 		iph->check = 0;
1243 		iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1244 	} while ((skb = skb->next));
1245 
1246 out:
1247 	return segs;
1248 }
1249 
1250 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1251 			 unsigned short type, unsigned char protocol,
1252 			 struct net *net)
1253 {
1254 	struct socket *sock;
1255 	int rc = sock_create_kern(family, type, protocol, &sock);
1256 
1257 	if (rc == 0) {
1258 		*sk = sock->sk;
1259 		(*sk)->sk_allocation = GFP_ATOMIC;
1260 		/*
1261 		 * Unhash it so that IP input processing does not even see it,
1262 		 * we do not wish this socket to see incoming packets.
1263 		 */
1264 		(*sk)->sk_prot->unhash(*sk);
1265 
1266 		sk_change_net(*sk, net);
1267 	}
1268 	return rc;
1269 }
1270 
1271 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1272 
1273 unsigned long snmp_fold_field(void *mib[], int offt)
1274 {
1275 	unsigned long res = 0;
1276 	int i;
1277 
1278 	for_each_possible_cpu(i) {
1279 		res += *(((unsigned long *) per_cpu_ptr(mib[0], i)) + offt);
1280 		res += *(((unsigned long *) per_cpu_ptr(mib[1], i)) + offt);
1281 	}
1282 	return res;
1283 }
1284 EXPORT_SYMBOL_GPL(snmp_fold_field);
1285 
1286 int snmp_mib_init(void *ptr[2], size_t mibsize)
1287 {
1288 	BUG_ON(ptr == NULL);
1289 	ptr[0] = __alloc_percpu(mibsize);
1290 	if (!ptr[0])
1291 		goto err0;
1292 	ptr[1] = __alloc_percpu(mibsize);
1293 	if (!ptr[1])
1294 		goto err1;
1295 	return 0;
1296 err1:
1297 	free_percpu(ptr[0]);
1298 	ptr[0] = NULL;
1299 err0:
1300 	return -ENOMEM;
1301 }
1302 EXPORT_SYMBOL_GPL(snmp_mib_init);
1303 
1304 void snmp_mib_free(void *ptr[2])
1305 {
1306 	BUG_ON(ptr == NULL);
1307 	free_percpu(ptr[0]);
1308 	free_percpu(ptr[1]);
1309 	ptr[0] = ptr[1] = NULL;
1310 }
1311 EXPORT_SYMBOL_GPL(snmp_mib_free);
1312 
1313 #ifdef CONFIG_IP_MULTICAST
1314 static struct net_protocol igmp_protocol = {
1315 	.handler =	igmp_rcv,
1316 };
1317 #endif
1318 
1319 static struct net_protocol tcp_protocol = {
1320 	.handler =	tcp_v4_rcv,
1321 	.err_handler =	tcp_v4_err,
1322 	.gso_send_check = tcp_v4_gso_send_check,
1323 	.gso_segment =	tcp_tso_segment,
1324 	.no_policy =	1,
1325 	.netns_ok =	1,
1326 };
1327 
1328 static struct net_protocol udp_protocol = {
1329 	.handler =	udp_rcv,
1330 	.err_handler =	udp_err,
1331 	.no_policy =	1,
1332 	.netns_ok =	1,
1333 };
1334 
1335 static struct net_protocol icmp_protocol = {
1336 	.handler =	icmp_rcv,
1337 	.no_policy =	1,
1338 	.netns_ok =	1,
1339 };
1340 
1341 static __net_init int ipv4_mib_init_net(struct net *net)
1342 {
1343 	if (snmp_mib_init((void **)net->mib.tcp_statistics,
1344 			  sizeof(struct tcp_mib)) < 0)
1345 		goto err_tcp_mib;
1346 	if (snmp_mib_init((void **)net->mib.ip_statistics,
1347 			  sizeof(struct ipstats_mib)) < 0)
1348 		goto err_ip_mib;
1349 	if (snmp_mib_init((void **)net->mib.net_statistics,
1350 			  sizeof(struct linux_mib)) < 0)
1351 		goto err_net_mib;
1352 	if (snmp_mib_init((void **)net->mib.udp_statistics,
1353 			  sizeof(struct udp_mib)) < 0)
1354 		goto err_udp_mib;
1355 	if (snmp_mib_init((void **)net->mib.udplite_statistics,
1356 			  sizeof(struct udp_mib)) < 0)
1357 		goto err_udplite_mib;
1358 	if (snmp_mib_init((void **)net->mib.icmp_statistics,
1359 			  sizeof(struct icmp_mib)) < 0)
1360 		goto err_icmp_mib;
1361 	if (snmp_mib_init((void **)net->mib.icmpmsg_statistics,
1362 			  sizeof(struct icmpmsg_mib)) < 0)
1363 		goto err_icmpmsg_mib;
1364 
1365 	tcp_mib_init(net);
1366 	return 0;
1367 
1368 err_icmpmsg_mib:
1369 	snmp_mib_free((void **)net->mib.icmp_statistics);
1370 err_icmp_mib:
1371 	snmp_mib_free((void **)net->mib.udplite_statistics);
1372 err_udplite_mib:
1373 	snmp_mib_free((void **)net->mib.udp_statistics);
1374 err_udp_mib:
1375 	snmp_mib_free((void **)net->mib.net_statistics);
1376 err_net_mib:
1377 	snmp_mib_free((void **)net->mib.ip_statistics);
1378 err_ip_mib:
1379 	snmp_mib_free((void **)net->mib.tcp_statistics);
1380 err_tcp_mib:
1381 	return -ENOMEM;
1382 }
1383 
1384 static __net_exit void ipv4_mib_exit_net(struct net *net)
1385 {
1386 	snmp_mib_free((void **)net->mib.icmpmsg_statistics);
1387 	snmp_mib_free((void **)net->mib.icmp_statistics);
1388 	snmp_mib_free((void **)net->mib.udplite_statistics);
1389 	snmp_mib_free((void **)net->mib.udp_statistics);
1390 	snmp_mib_free((void **)net->mib.net_statistics);
1391 	snmp_mib_free((void **)net->mib.ip_statistics);
1392 	snmp_mib_free((void **)net->mib.tcp_statistics);
1393 }
1394 
1395 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1396 	.init = ipv4_mib_init_net,
1397 	.exit = ipv4_mib_exit_net,
1398 };
1399 
1400 static int __init init_ipv4_mibs(void)
1401 {
1402 	return register_pernet_subsys(&ipv4_mib_ops);
1403 }
1404 
1405 static int ipv4_proc_init(void);
1406 
1407 /*
1408  *	IP protocol layer initialiser
1409  */
1410 
1411 static struct packet_type ip_packet_type = {
1412 	.type = __constant_htons(ETH_P_IP),
1413 	.func = ip_rcv,
1414 	.gso_send_check = inet_gso_send_check,
1415 	.gso_segment = inet_gso_segment,
1416 };
1417 
1418 static int __init inet_init(void)
1419 {
1420 	struct sk_buff *dummy_skb;
1421 	struct inet_protosw *q;
1422 	struct list_head *r;
1423 	int rc = -EINVAL;
1424 
1425 	BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1426 
1427 	rc = proto_register(&tcp_prot, 1);
1428 	if (rc)
1429 		goto out;
1430 
1431 	rc = proto_register(&udp_prot, 1);
1432 	if (rc)
1433 		goto out_unregister_tcp_proto;
1434 
1435 	rc = proto_register(&raw_prot, 1);
1436 	if (rc)
1437 		goto out_unregister_udp_proto;
1438 
1439 	/*
1440 	 *	Tell SOCKET that we are alive...
1441 	 */
1442 
1443 	(void)sock_register(&inet_family_ops);
1444 
1445 	/*
1446 	 *	Add all the base protocols.
1447 	 */
1448 
1449 	if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1450 		printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1451 	if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1452 		printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1453 	if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1454 		printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1455 #ifdef CONFIG_IP_MULTICAST
1456 	if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1457 		printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1458 #endif
1459 
1460 	/* Register the socket-side information for inet_create. */
1461 	for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1462 		INIT_LIST_HEAD(r);
1463 
1464 	for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1465 		inet_register_protosw(q);
1466 
1467 	/*
1468 	 *	Set the ARP module up
1469 	 */
1470 
1471 	arp_init();
1472 
1473 	/*
1474 	 *	Set the IP module up
1475 	 */
1476 
1477 	ip_init();
1478 
1479 	tcp_v4_init();
1480 
1481 	/* Setup TCP slab cache for open requests. */
1482 	tcp_init();
1483 
1484 	/* Setup UDP memory threshold */
1485 	udp_init();
1486 
1487 	/* Add UDP-Lite (RFC 3828) */
1488 	udplite4_register();
1489 
1490 	/*
1491 	 *	Set the ICMP layer up
1492 	 */
1493 
1494 	if (icmp_init() < 0)
1495 		panic("Failed to create the ICMP control socket.\n");
1496 
1497 	/*
1498 	 *	Initialise the multicast router
1499 	 */
1500 #if defined(CONFIG_IP_MROUTE)
1501 	if (ip_mr_init())
1502 		printk(KERN_CRIT "inet_init: Cannot init ipv4 mroute\n");
1503 #endif
1504 	/*
1505 	 *	Initialise per-cpu ipv4 mibs
1506 	 */
1507 
1508 	if (init_ipv4_mibs())
1509 		printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n");
1510 
1511 	ipv4_proc_init();
1512 
1513 	ipfrag_init();
1514 
1515 	dev_add_pack(&ip_packet_type);
1516 
1517 	rc = 0;
1518 out:
1519 	return rc;
1520 out_unregister_udp_proto:
1521 	proto_unregister(&udp_prot);
1522 out_unregister_tcp_proto:
1523 	proto_unregister(&tcp_prot);
1524 	goto out;
1525 }
1526 
1527 fs_initcall(inet_init);
1528 
1529 /* ------------------------------------------------------------------------ */
1530 
1531 #ifdef CONFIG_PROC_FS
1532 static int __init ipv4_proc_init(void)
1533 {
1534 	int rc = 0;
1535 
1536 	if (raw_proc_init())
1537 		goto out_raw;
1538 	if (tcp4_proc_init())
1539 		goto out_tcp;
1540 	if (udp4_proc_init())
1541 		goto out_udp;
1542 	if (ip_misc_proc_init())
1543 		goto out_misc;
1544 out:
1545 	return rc;
1546 out_misc:
1547 	udp4_proc_exit();
1548 out_udp:
1549 	tcp4_proc_exit();
1550 out_tcp:
1551 	raw_proc_exit();
1552 out_raw:
1553 	rc = -ENOMEM;
1554 	goto out;
1555 }
1556 
1557 #else /* CONFIG_PROC_FS */
1558 static int __init ipv4_proc_init(void)
1559 {
1560 	return 0;
1561 }
1562 #endif /* CONFIG_PROC_FS */
1563 
1564 MODULE_ALIAS_NETPROTO(PF_INET);
1565 
1566 EXPORT_SYMBOL(inet_accept);
1567 EXPORT_SYMBOL(inet_bind);
1568 EXPORT_SYMBOL(inet_dgram_connect);
1569 EXPORT_SYMBOL(inet_dgram_ops);
1570 EXPORT_SYMBOL(inet_getname);
1571 EXPORT_SYMBOL(inet_ioctl);
1572 EXPORT_SYMBOL(inet_listen);
1573 EXPORT_SYMBOL(inet_register_protosw);
1574 EXPORT_SYMBOL(inet_release);
1575 EXPORT_SYMBOL(inet_sendmsg);
1576 EXPORT_SYMBOL(inet_shutdown);
1577 EXPORT_SYMBOL(inet_sock_destruct);
1578 EXPORT_SYMBOL(inet_stream_connect);
1579 EXPORT_SYMBOL(inet_stream_ops);
1580 EXPORT_SYMBOL(inet_unregister_protosw);
1581 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
1582