xref: /linux/net/ipv4/af_inet.c (revision 82492a355fac112908271faa74f473a38c1fb647)
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 #include <linux/slab.h>
90 
91 #include <asm/uaccess.h>
92 #include <asm/system.h>
93 
94 #include <linux/inet.h>
95 #include <linux/igmp.h>
96 #include <linux/inetdevice.h>
97 #include <linux/netdevice.h>
98 #include <net/checksum.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 <net/ping.h>
109 #include <linux/skbuff.h>
110 #include <net/sock.h>
111 #include <net/raw.h>
112 #include <net/icmp.h>
113 #include <net/ipip.h>
114 #include <net/inet_common.h>
115 #include <net/xfrm.h>
116 #include <net/net_namespace.h>
117 #ifdef CONFIG_IP_MROUTE
118 #include <linux/mroute.h>
119 #endif
120 
121 
122 /* The inetsw table contains everything that inet_create needs to
123  * build a new socket.
124  */
125 static struct list_head inetsw[SOCK_MAX];
126 static DEFINE_SPINLOCK(inetsw_lock);
127 
128 struct ipv4_config ipv4_config;
129 EXPORT_SYMBOL(ipv4_config);
130 
131 /* New destruction routine */
132 
133 void inet_sock_destruct(struct sock *sk)
134 {
135 	struct inet_sock *inet = inet_sk(sk);
136 
137 	__skb_queue_purge(&sk->sk_receive_queue);
138 	__skb_queue_purge(&sk->sk_error_queue);
139 
140 	sk_mem_reclaim(sk);
141 
142 	if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
143 		pr_err("Attempt to release TCP socket in state %d %p\n",
144 		       sk->sk_state, sk);
145 		return;
146 	}
147 	if (!sock_flag(sk, SOCK_DEAD)) {
148 		pr_err("Attempt to release alive inet socket %p\n", sk);
149 		return;
150 	}
151 
152 	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
153 	WARN_ON(atomic_read(&sk->sk_wmem_alloc));
154 	WARN_ON(sk->sk_wmem_queued);
155 	WARN_ON(sk->sk_forward_alloc);
156 
157 	kfree(rcu_dereference_protected(inet->inet_opt, 1));
158 	dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
159 	sk_refcnt_debug_dec(sk);
160 }
161 EXPORT_SYMBOL(inet_sock_destruct);
162 
163 /*
164  *	The routines beyond this point handle the behaviour of an AF_INET
165  *	socket object. Mostly it punts to the subprotocols of IP to do
166  *	the work.
167  */
168 
169 /*
170  *	Automatically bind an unbound socket.
171  */
172 
173 static int inet_autobind(struct sock *sk)
174 {
175 	struct inet_sock *inet;
176 	/* We may need to bind the socket. */
177 	lock_sock(sk);
178 	inet = inet_sk(sk);
179 	if (!inet->inet_num) {
180 		if (sk->sk_prot->get_port(sk, 0)) {
181 			release_sock(sk);
182 			return -EAGAIN;
183 		}
184 		inet->inet_sport = htons(inet->inet_num);
185 	}
186 	release_sock(sk);
187 	return 0;
188 }
189 
190 /*
191  *	Move a socket into listening state.
192  */
193 int inet_listen(struct socket *sock, int backlog)
194 {
195 	struct sock *sk = sock->sk;
196 	unsigned char old_state;
197 	int err;
198 
199 	lock_sock(sk);
200 
201 	err = -EINVAL;
202 	if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
203 		goto out;
204 
205 	old_state = sk->sk_state;
206 	if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
207 		goto out;
208 
209 	/* Really, if the socket is already in listen state
210 	 * we can only allow the backlog to be adjusted.
211 	 */
212 	if (old_state != TCP_LISTEN) {
213 		err = inet_csk_listen_start(sk, backlog);
214 		if (err)
215 			goto out;
216 	}
217 	sk->sk_max_ack_backlog = backlog;
218 	err = 0;
219 
220 out:
221 	release_sock(sk);
222 	return err;
223 }
224 EXPORT_SYMBOL(inet_listen);
225 
226 u32 inet_ehash_secret __read_mostly;
227 EXPORT_SYMBOL(inet_ehash_secret);
228 
229 /*
230  * inet_ehash_secret must be set exactly once
231  */
232 void build_ehash_secret(void)
233 {
234 	u32 rnd;
235 
236 	do {
237 		get_random_bytes(&rnd, sizeof(rnd));
238 	} while (rnd == 0);
239 
240 	cmpxchg(&inet_ehash_secret, 0, rnd);
241 }
242 EXPORT_SYMBOL(build_ehash_secret);
243 
244 static inline int inet_netns_ok(struct net *net, int protocol)
245 {
246 	int hash;
247 	const struct net_protocol *ipprot;
248 
249 	if (net_eq(net, &init_net))
250 		return 1;
251 
252 	hash = protocol & (MAX_INET_PROTOS - 1);
253 	ipprot = rcu_dereference(inet_protos[hash]);
254 
255 	if (ipprot == NULL)
256 		/* raw IP is OK */
257 		return 1;
258 	return ipprot->netns_ok;
259 }
260 
261 /*
262  *	Create an inet socket.
263  */
264 
265 static int inet_create(struct net *net, struct socket *sock, int protocol,
266 		       int kern)
267 {
268 	struct sock *sk;
269 	struct inet_protosw *answer;
270 	struct inet_sock *inet;
271 	struct proto *answer_prot;
272 	unsigned char answer_flags;
273 	char answer_no_check;
274 	int try_loading_module = 0;
275 	int err;
276 
277 	if (unlikely(!inet_ehash_secret))
278 		if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
279 			build_ehash_secret();
280 
281 	sock->state = SS_UNCONNECTED;
282 
283 	/* Look for the requested type/protocol pair. */
284 lookup_protocol:
285 	err = -ESOCKTNOSUPPORT;
286 	rcu_read_lock();
287 	list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
288 
289 		err = 0;
290 		/* Check the non-wild match. */
291 		if (protocol == answer->protocol) {
292 			if (protocol != IPPROTO_IP)
293 				break;
294 		} else {
295 			/* Check for the two wild cases. */
296 			if (IPPROTO_IP == protocol) {
297 				protocol = answer->protocol;
298 				break;
299 			}
300 			if (IPPROTO_IP == answer->protocol)
301 				break;
302 		}
303 		err = -EPROTONOSUPPORT;
304 	}
305 
306 	if (unlikely(err)) {
307 		if (try_loading_module < 2) {
308 			rcu_read_unlock();
309 			/*
310 			 * Be more specific, e.g. net-pf-2-proto-132-type-1
311 			 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
312 			 */
313 			if (++try_loading_module == 1)
314 				request_module("net-pf-%d-proto-%d-type-%d",
315 					       PF_INET, protocol, sock->type);
316 			/*
317 			 * Fall back to generic, e.g. net-pf-2-proto-132
318 			 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
319 			 */
320 			else
321 				request_module("net-pf-%d-proto-%d",
322 					       PF_INET, protocol);
323 			goto lookup_protocol;
324 		} else
325 			goto out_rcu_unlock;
326 	}
327 
328 	err = -EPERM;
329 	if (sock->type == SOCK_RAW && !kern && !capable(CAP_NET_RAW))
330 		goto out_rcu_unlock;
331 
332 	err = -EAFNOSUPPORT;
333 	if (!inet_netns_ok(net, protocol))
334 		goto out_rcu_unlock;
335 
336 	sock->ops = answer->ops;
337 	answer_prot = answer->prot;
338 	answer_no_check = answer->no_check;
339 	answer_flags = answer->flags;
340 	rcu_read_unlock();
341 
342 	WARN_ON(answer_prot->slab == NULL);
343 
344 	err = -ENOBUFS;
345 	sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
346 	if (sk == NULL)
347 		goto out;
348 
349 	err = 0;
350 	sk->sk_no_check = answer_no_check;
351 	if (INET_PROTOSW_REUSE & answer_flags)
352 		sk->sk_reuse = 1;
353 
354 	inet = inet_sk(sk);
355 	inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
356 
357 	inet->nodefrag = 0;
358 
359 	if (SOCK_RAW == sock->type) {
360 		inet->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->inet_id = 0;
371 
372 	sock_init_data(sock, sk);
373 
374 	sk->sk_destruct	   = inet_sock_destruct;
375 	sk->sk_protocol	   = protocol;
376 	sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
377 
378 	inet->uc_ttl	= -1;
379 	inet->mc_loop	= 1;
380 	inet->mc_ttl	= 1;
381 	inet->mc_all	= 1;
382 	inet->mc_index	= 0;
383 	inet->mc_list	= NULL;
384 	inet->rcv_tos	= 0;
385 
386 	sk_refcnt_debug_inc(sk);
387 
388 	if (inet->inet_num) {
389 		/* It assumes that any protocol which allows
390 		 * the user to assign a number at socket
391 		 * creation time automatically
392 		 * shares.
393 		 */
394 		inet->inet_sport = htons(inet->inet_num);
395 		/* Add to protocol hash chains. */
396 		sk->sk_prot->hash(sk);
397 	}
398 
399 	if (sk->sk_prot->init) {
400 		err = sk->sk_prot->init(sk);
401 		if (err)
402 			sk_common_release(sk);
403 	}
404 out:
405 	return err;
406 out_rcu_unlock:
407 	rcu_read_unlock();
408 	goto out;
409 }
410 
411 
412 /*
413  *	The peer socket should always be NULL (or else). When we call this
414  *	function we are destroying the object and from then on nobody
415  *	should refer to it.
416  */
417 int inet_release(struct socket *sock)
418 {
419 	struct sock *sk = sock->sk;
420 
421 	if (sk) {
422 		long timeout;
423 
424 		sock_rps_reset_flow(sk);
425 
426 		/* Applications forget to leave groups before exiting */
427 		ip_mc_drop_socket(sk);
428 
429 		/* If linger is set, we don't return until the close
430 		 * is complete.  Otherwise we return immediately. The
431 		 * actually closing is done the same either way.
432 		 *
433 		 * If the close is due to the process exiting, we never
434 		 * linger..
435 		 */
436 		timeout = 0;
437 		if (sock_flag(sk, SOCK_LINGER) &&
438 		    !(current->flags & PF_EXITING))
439 			timeout = sk->sk_lingertime;
440 		sock->sk = NULL;
441 		sk->sk_prot->close(sk, timeout);
442 	}
443 	return 0;
444 }
445 EXPORT_SYMBOL(inet_release);
446 
447 /* It is off by default, see below. */
448 int sysctl_ip_nonlocal_bind __read_mostly;
449 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
450 
451 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
452 {
453 	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
454 	struct sock *sk = sock->sk;
455 	struct inet_sock *inet = inet_sk(sk);
456 	unsigned short snum;
457 	int chk_addr_ret;
458 	int err;
459 
460 	/* If the socket has its own bind function then use it. (RAW) */
461 	if (sk->sk_prot->bind) {
462 		err = sk->sk_prot->bind(sk, uaddr, addr_len);
463 		goto out;
464 	}
465 	err = -EINVAL;
466 	if (addr_len < sizeof(struct sockaddr_in))
467 		goto out;
468 
469 	if (addr->sin_family != AF_INET) {
470 		/* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
471 		 * only if s_addr is INADDR_ANY.
472 		 */
473 		err = -EAFNOSUPPORT;
474 		if (addr->sin_family != AF_UNSPEC ||
475 		    addr->sin_addr.s_addr != htonl(INADDR_ANY))
476 			goto out;
477 	}
478 
479 	chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr);
480 
481 	/* Not specified by any standard per-se, however it breaks too
482 	 * many applications when removed.  It is unfortunate since
483 	 * allowing applications to make a non-local bind solves
484 	 * several problems with systems using dynamic addressing.
485 	 * (ie. your servers still start up even if your ISDN link
486 	 *  is temporarily down)
487 	 */
488 	err = -EADDRNOTAVAIL;
489 	if (!sysctl_ip_nonlocal_bind &&
490 	    !(inet->freebind || inet->transparent) &&
491 	    addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
492 	    chk_addr_ret != RTN_LOCAL &&
493 	    chk_addr_ret != RTN_MULTICAST &&
494 	    chk_addr_ret != RTN_BROADCAST)
495 		goto out;
496 
497 	snum = ntohs(addr->sin_port);
498 	err = -EACCES;
499 	if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
500 		goto out;
501 
502 	/*      We keep a pair of addresses. rcv_saddr is the one
503 	 *      used by hash lookups, and saddr is used for transmit.
504 	 *
505 	 *      In the BSD API these are the same except where it
506 	 *      would be illegal to use them (multicast/broadcast) in
507 	 *      which case the sending device address is used.
508 	 */
509 	lock_sock(sk);
510 
511 	/* Check these errors (active socket, double bind). */
512 	err = -EINVAL;
513 	if (sk->sk_state != TCP_CLOSE || inet->inet_num)
514 		goto out_release_sock;
515 
516 	inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
517 	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
518 		inet->inet_saddr = 0;  /* Use device */
519 
520 	/* Make sure we are allowed to bind here. */
521 	if (sk->sk_prot->get_port(sk, snum)) {
522 		inet->inet_saddr = inet->inet_rcv_saddr = 0;
523 		err = -EADDRINUSE;
524 		goto out_release_sock;
525 	}
526 
527 	if (inet->inet_rcv_saddr)
528 		sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
529 	if (snum)
530 		sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
531 	inet->inet_sport = htons(inet->inet_num);
532 	inet->inet_daddr = 0;
533 	inet->inet_dport = 0;
534 	sk_dst_reset(sk);
535 	err = 0;
536 out_release_sock:
537 	release_sock(sk);
538 out:
539 	return err;
540 }
541 EXPORT_SYMBOL(inet_bind);
542 
543 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
544 		       int addr_len, int flags)
545 {
546 	struct sock *sk = sock->sk;
547 
548 	if (addr_len < sizeof(uaddr->sa_family))
549 		return -EINVAL;
550 	if (uaddr->sa_family == AF_UNSPEC)
551 		return sk->sk_prot->disconnect(sk, flags);
552 
553 	if (!inet_sk(sk)->inet_num && inet_autobind(sk))
554 		return -EAGAIN;
555 	return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
556 }
557 EXPORT_SYMBOL(inet_dgram_connect);
558 
559 static long inet_wait_for_connect(struct sock *sk, long timeo)
560 {
561 	DEFINE_WAIT(wait);
562 
563 	prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
564 
565 	/* Basic assumption: if someone sets sk->sk_err, he _must_
566 	 * change state of the socket from TCP_SYN_*.
567 	 * Connect() does not allow to get error notifications
568 	 * without closing the socket.
569 	 */
570 	while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
571 		release_sock(sk);
572 		timeo = schedule_timeout(timeo);
573 		lock_sock(sk);
574 		if (signal_pending(current) || !timeo)
575 			break;
576 		prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
577 	}
578 	finish_wait(sk_sleep(sk), &wait);
579 	return timeo;
580 }
581 
582 /*
583  *	Connect to a remote host. There is regrettably still a little
584  *	TCP 'magic' in here.
585  */
586 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
587 			int addr_len, int flags)
588 {
589 	struct sock *sk = sock->sk;
590 	int err;
591 	long timeo;
592 
593 	if (addr_len < sizeof(uaddr->sa_family))
594 		return -EINVAL;
595 
596 	lock_sock(sk);
597 
598 	if (uaddr->sa_family == AF_UNSPEC) {
599 		err = sk->sk_prot->disconnect(sk, flags);
600 		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
601 		goto out;
602 	}
603 
604 	switch (sock->state) {
605 	default:
606 		err = -EINVAL;
607 		goto out;
608 	case SS_CONNECTED:
609 		err = -EISCONN;
610 		goto out;
611 	case SS_CONNECTING:
612 		err = -EALREADY;
613 		/* Fall out of switch with err, set for this state */
614 		break;
615 	case SS_UNCONNECTED:
616 		err = -EISCONN;
617 		if (sk->sk_state != TCP_CLOSE)
618 			goto out;
619 
620 		err = sk->sk_prot->connect(sk, uaddr, addr_len);
621 		if (err < 0)
622 			goto out;
623 
624 		sock->state = SS_CONNECTING;
625 
626 		/* Just entered SS_CONNECTING state; the only
627 		 * difference is that return value in non-blocking
628 		 * case is EINPROGRESS, rather than EALREADY.
629 		 */
630 		err = -EINPROGRESS;
631 		break;
632 	}
633 
634 	timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
635 
636 	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
637 		/* Error code is set above */
638 		if (!timeo || !inet_wait_for_connect(sk, timeo))
639 			goto out;
640 
641 		err = sock_intr_errno(timeo);
642 		if (signal_pending(current))
643 			goto out;
644 	}
645 
646 	/* Connection was closed by RST, timeout, ICMP error
647 	 * or another process disconnected us.
648 	 */
649 	if (sk->sk_state == TCP_CLOSE)
650 		goto sock_error;
651 
652 	/* sk->sk_err may be not zero now, if RECVERR was ordered by user
653 	 * and error was received after socket entered established state.
654 	 * Hence, it is handled normally after connect() return successfully.
655 	 */
656 
657 	sock->state = SS_CONNECTED;
658 	err = 0;
659 out:
660 	release_sock(sk);
661 	return err;
662 
663 sock_error:
664 	err = sock_error(sk) ? : -ECONNABORTED;
665 	sock->state = SS_UNCONNECTED;
666 	if (sk->sk_prot->disconnect(sk, flags))
667 		sock->state = SS_DISCONNECTING;
668 	goto out;
669 }
670 EXPORT_SYMBOL(inet_stream_connect);
671 
672 /*
673  *	Accept a pending connection. The TCP layer now gives BSD semantics.
674  */
675 
676 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
677 {
678 	struct sock *sk1 = sock->sk;
679 	int err = -EINVAL;
680 	struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
681 
682 	if (!sk2)
683 		goto do_err;
684 
685 	lock_sock(sk2);
686 
687 	sock_rps_record_flow(sk2);
688 	WARN_ON(!((1 << sk2->sk_state) &
689 		  (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE)));
690 
691 	sock_graft(sk2, newsock);
692 
693 	newsock->state = SS_CONNECTED;
694 	err = 0;
695 	release_sock(sk2);
696 do_err:
697 	return err;
698 }
699 EXPORT_SYMBOL(inet_accept);
700 
701 
702 /*
703  *	This does both peername and sockname.
704  */
705 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
706 			int *uaddr_len, int peer)
707 {
708 	struct sock *sk		= sock->sk;
709 	struct inet_sock *inet	= inet_sk(sk);
710 	DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
711 
712 	sin->sin_family = AF_INET;
713 	if (peer) {
714 		if (!inet->inet_dport ||
715 		    (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
716 		     peer == 1))
717 			return -ENOTCONN;
718 		sin->sin_port = inet->inet_dport;
719 		sin->sin_addr.s_addr = inet->inet_daddr;
720 	} else {
721 		__be32 addr = inet->inet_rcv_saddr;
722 		if (!addr)
723 			addr = inet->inet_saddr;
724 		sin->sin_port = inet->inet_sport;
725 		sin->sin_addr.s_addr = addr;
726 	}
727 	memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
728 	*uaddr_len = sizeof(*sin);
729 	return 0;
730 }
731 EXPORT_SYMBOL(inet_getname);
732 
733 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
734 		 size_t size)
735 {
736 	struct sock *sk = sock->sk;
737 
738 	sock_rps_record_flow(sk);
739 
740 	/* We may need to bind the socket. */
741 	if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
742 	    inet_autobind(sk))
743 		return -EAGAIN;
744 
745 	return sk->sk_prot->sendmsg(iocb, sk, msg, size);
746 }
747 EXPORT_SYMBOL(inet_sendmsg);
748 
749 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
750 		      size_t size, int flags)
751 {
752 	struct sock *sk = sock->sk;
753 
754 	sock_rps_record_flow(sk);
755 
756 	/* We may need to bind the socket. */
757 	if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
758 	    inet_autobind(sk))
759 		return -EAGAIN;
760 
761 	if (sk->sk_prot->sendpage)
762 		return sk->sk_prot->sendpage(sk, page, offset, size, flags);
763 	return sock_no_sendpage(sock, page, offset, size, flags);
764 }
765 EXPORT_SYMBOL(inet_sendpage);
766 
767 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
768 		 size_t size, int flags)
769 {
770 	struct sock *sk = sock->sk;
771 	int addr_len = 0;
772 	int err;
773 
774 	sock_rps_record_flow(sk);
775 
776 	err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
777 				   flags & ~MSG_DONTWAIT, &addr_len);
778 	if (err >= 0)
779 		msg->msg_namelen = addr_len;
780 	return err;
781 }
782 EXPORT_SYMBOL(inet_recvmsg);
783 
784 int inet_shutdown(struct socket *sock, int how)
785 {
786 	struct sock *sk = sock->sk;
787 	int err = 0;
788 
789 	/* This should really check to make sure
790 	 * the socket is a TCP socket. (WHY AC...)
791 	 */
792 	how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
793 		       1->2 bit 2 snds.
794 		       2->3 */
795 	if ((how & ~SHUTDOWN_MASK) || !how)	/* MAXINT->0 */
796 		return -EINVAL;
797 
798 	lock_sock(sk);
799 	if (sock->state == SS_CONNECTING) {
800 		if ((1 << sk->sk_state) &
801 		    (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
802 			sock->state = SS_DISCONNECTING;
803 		else
804 			sock->state = SS_CONNECTED;
805 	}
806 
807 	switch (sk->sk_state) {
808 	case TCP_CLOSE:
809 		err = -ENOTCONN;
810 		/* Hack to wake up other listeners, who can poll for
811 		   POLLHUP, even on eg. unconnected UDP sockets -- RR */
812 	default:
813 		sk->sk_shutdown |= how;
814 		if (sk->sk_prot->shutdown)
815 			sk->sk_prot->shutdown(sk, how);
816 		break;
817 
818 	/* Remaining two branches are temporary solution for missing
819 	 * close() in multithreaded environment. It is _not_ a good idea,
820 	 * but we have no choice until close() is repaired at VFS level.
821 	 */
822 	case TCP_LISTEN:
823 		if (!(how & RCV_SHUTDOWN))
824 			break;
825 		/* Fall through */
826 	case TCP_SYN_SENT:
827 		err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
828 		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
829 		break;
830 	}
831 
832 	/* Wake up anyone sleeping in poll. */
833 	sk->sk_state_change(sk);
834 	release_sock(sk);
835 	return err;
836 }
837 EXPORT_SYMBOL(inet_shutdown);
838 
839 /*
840  *	ioctl() calls you can issue on an INET socket. Most of these are
841  *	device configuration and stuff and very rarely used. Some ioctls
842  *	pass on to the socket itself.
843  *
844  *	NOTE: I like the idea of a module for the config stuff. ie ifconfig
845  *	loads the devconfigure module does its configuring and unloads it.
846  *	There's a good 20K of config code hanging around the kernel.
847  */
848 
849 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
850 {
851 	struct sock *sk = sock->sk;
852 	int err = 0;
853 	struct net *net = sock_net(sk);
854 
855 	switch (cmd) {
856 	case SIOCGSTAMP:
857 		err = sock_get_timestamp(sk, (struct timeval __user *)arg);
858 		break;
859 	case SIOCGSTAMPNS:
860 		err = sock_get_timestampns(sk, (struct timespec __user *)arg);
861 		break;
862 	case SIOCADDRT:
863 	case SIOCDELRT:
864 	case SIOCRTMSG:
865 		err = ip_rt_ioctl(net, cmd, (void __user *)arg);
866 		break;
867 	case SIOCDARP:
868 	case SIOCGARP:
869 	case SIOCSARP:
870 		err = arp_ioctl(net, cmd, (void __user *)arg);
871 		break;
872 	case SIOCGIFADDR:
873 	case SIOCSIFADDR:
874 	case SIOCGIFBRDADDR:
875 	case SIOCSIFBRDADDR:
876 	case SIOCGIFNETMASK:
877 	case SIOCSIFNETMASK:
878 	case SIOCGIFDSTADDR:
879 	case SIOCSIFDSTADDR:
880 	case SIOCSIFPFLAGS:
881 	case SIOCGIFPFLAGS:
882 	case SIOCSIFFLAGS:
883 		err = devinet_ioctl(net, cmd, (void __user *)arg);
884 		break;
885 	default:
886 		if (sk->sk_prot->ioctl)
887 			err = sk->sk_prot->ioctl(sk, cmd, arg);
888 		else
889 			err = -ENOIOCTLCMD;
890 		break;
891 	}
892 	return err;
893 }
894 EXPORT_SYMBOL(inet_ioctl);
895 
896 #ifdef CONFIG_COMPAT
897 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
898 {
899 	struct sock *sk = sock->sk;
900 	int err = -ENOIOCTLCMD;
901 
902 	if (sk->sk_prot->compat_ioctl)
903 		err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
904 
905 	return err;
906 }
907 #endif
908 
909 const struct proto_ops inet_stream_ops = {
910 	.family		   = PF_INET,
911 	.owner		   = THIS_MODULE,
912 	.release	   = inet_release,
913 	.bind		   = inet_bind,
914 	.connect	   = inet_stream_connect,
915 	.socketpair	   = sock_no_socketpair,
916 	.accept		   = inet_accept,
917 	.getname	   = inet_getname,
918 	.poll		   = tcp_poll,
919 	.ioctl		   = inet_ioctl,
920 	.listen		   = inet_listen,
921 	.shutdown	   = inet_shutdown,
922 	.setsockopt	   = sock_common_setsockopt,
923 	.getsockopt	   = sock_common_getsockopt,
924 	.sendmsg	   = inet_sendmsg,
925 	.recvmsg	   = inet_recvmsg,
926 	.mmap		   = sock_no_mmap,
927 	.sendpage	   = inet_sendpage,
928 	.splice_read	   = tcp_splice_read,
929 #ifdef CONFIG_COMPAT
930 	.compat_setsockopt = compat_sock_common_setsockopt,
931 	.compat_getsockopt = compat_sock_common_getsockopt,
932 	.compat_ioctl	   = inet_compat_ioctl,
933 #endif
934 };
935 EXPORT_SYMBOL(inet_stream_ops);
936 
937 const struct proto_ops inet_dgram_ops = {
938 	.family		   = PF_INET,
939 	.owner		   = THIS_MODULE,
940 	.release	   = inet_release,
941 	.bind		   = inet_bind,
942 	.connect	   = inet_dgram_connect,
943 	.socketpair	   = sock_no_socketpair,
944 	.accept		   = sock_no_accept,
945 	.getname	   = inet_getname,
946 	.poll		   = udp_poll,
947 	.ioctl		   = inet_ioctl,
948 	.listen		   = sock_no_listen,
949 	.shutdown	   = inet_shutdown,
950 	.setsockopt	   = sock_common_setsockopt,
951 	.getsockopt	   = sock_common_getsockopt,
952 	.sendmsg	   = inet_sendmsg,
953 	.recvmsg	   = inet_recvmsg,
954 	.mmap		   = sock_no_mmap,
955 	.sendpage	   = inet_sendpage,
956 #ifdef CONFIG_COMPAT
957 	.compat_setsockopt = compat_sock_common_setsockopt,
958 	.compat_getsockopt = compat_sock_common_getsockopt,
959 	.compat_ioctl	   = inet_compat_ioctl,
960 #endif
961 };
962 EXPORT_SYMBOL(inet_dgram_ops);
963 
964 /*
965  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
966  * udp_poll
967  */
968 static const struct proto_ops inet_sockraw_ops = {
969 	.family		   = PF_INET,
970 	.owner		   = THIS_MODULE,
971 	.release	   = inet_release,
972 	.bind		   = inet_bind,
973 	.connect	   = inet_dgram_connect,
974 	.socketpair	   = sock_no_socketpair,
975 	.accept		   = sock_no_accept,
976 	.getname	   = inet_getname,
977 	.poll		   = datagram_poll,
978 	.ioctl		   = inet_ioctl,
979 	.listen		   = sock_no_listen,
980 	.shutdown	   = inet_shutdown,
981 	.setsockopt	   = sock_common_setsockopt,
982 	.getsockopt	   = sock_common_getsockopt,
983 	.sendmsg	   = inet_sendmsg,
984 	.recvmsg	   = inet_recvmsg,
985 	.mmap		   = sock_no_mmap,
986 	.sendpage	   = inet_sendpage,
987 #ifdef CONFIG_COMPAT
988 	.compat_setsockopt = compat_sock_common_setsockopt,
989 	.compat_getsockopt = compat_sock_common_getsockopt,
990 	.compat_ioctl	   = inet_compat_ioctl,
991 #endif
992 };
993 
994 static const struct net_proto_family inet_family_ops = {
995 	.family = PF_INET,
996 	.create = inet_create,
997 	.owner	= THIS_MODULE,
998 };
999 
1000 /* Upon startup we insert all the elements in inetsw_array[] into
1001  * the linked list inetsw.
1002  */
1003 static struct inet_protosw inetsw_array[] =
1004 {
1005 	{
1006 		.type =       SOCK_STREAM,
1007 		.protocol =   IPPROTO_TCP,
1008 		.prot =       &tcp_prot,
1009 		.ops =        &inet_stream_ops,
1010 		.no_check =   0,
1011 		.flags =      INET_PROTOSW_PERMANENT |
1012 			      INET_PROTOSW_ICSK,
1013 	},
1014 
1015 	{
1016 		.type =       SOCK_DGRAM,
1017 		.protocol =   IPPROTO_UDP,
1018 		.prot =       &udp_prot,
1019 		.ops =        &inet_dgram_ops,
1020 		.no_check =   UDP_CSUM_DEFAULT,
1021 		.flags =      INET_PROTOSW_PERMANENT,
1022        },
1023 
1024        {
1025 		.type =       SOCK_DGRAM,
1026 		.protocol =   IPPROTO_ICMP,
1027 		.prot =       &ping_prot,
1028 		.ops =        &inet_dgram_ops,
1029 		.no_check =   UDP_CSUM_DEFAULT,
1030 		.flags =      INET_PROTOSW_REUSE,
1031        },
1032 
1033        {
1034 	       .type =       SOCK_RAW,
1035 	       .protocol =   IPPROTO_IP,	/* wild card */
1036 	       .prot =       &raw_prot,
1037 	       .ops =        &inet_sockraw_ops,
1038 	       .no_check =   UDP_CSUM_DEFAULT,
1039 	       .flags =      INET_PROTOSW_REUSE,
1040        }
1041 };
1042 
1043 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1044 
1045 void inet_register_protosw(struct inet_protosw *p)
1046 {
1047 	struct list_head *lh;
1048 	struct inet_protosw *answer;
1049 	int protocol = p->protocol;
1050 	struct list_head *last_perm;
1051 
1052 	spin_lock_bh(&inetsw_lock);
1053 
1054 	if (p->type >= SOCK_MAX)
1055 		goto out_illegal;
1056 
1057 	/* If we are trying to override a permanent protocol, bail. */
1058 	answer = NULL;
1059 	last_perm = &inetsw[p->type];
1060 	list_for_each(lh, &inetsw[p->type]) {
1061 		answer = list_entry(lh, struct inet_protosw, list);
1062 
1063 		/* Check only the non-wild match. */
1064 		if (INET_PROTOSW_PERMANENT & answer->flags) {
1065 			if (protocol == answer->protocol)
1066 				break;
1067 			last_perm = lh;
1068 		}
1069 
1070 		answer = NULL;
1071 	}
1072 	if (answer)
1073 		goto out_permanent;
1074 
1075 	/* Add the new entry after the last permanent entry if any, so that
1076 	 * the new entry does not override a permanent entry when matched with
1077 	 * a wild-card protocol. But it is allowed to override any existing
1078 	 * non-permanent entry.  This means that when we remove this entry, the
1079 	 * system automatically returns to the old behavior.
1080 	 */
1081 	list_add_rcu(&p->list, last_perm);
1082 out:
1083 	spin_unlock_bh(&inetsw_lock);
1084 
1085 	return;
1086 
1087 out_permanent:
1088 	printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
1089 	       protocol);
1090 	goto out;
1091 
1092 out_illegal:
1093 	printk(KERN_ERR
1094 	       "Ignoring attempt to register invalid socket type %d.\n",
1095 	       p->type);
1096 	goto out;
1097 }
1098 EXPORT_SYMBOL(inet_register_protosw);
1099 
1100 void inet_unregister_protosw(struct inet_protosw *p)
1101 {
1102 	if (INET_PROTOSW_PERMANENT & p->flags) {
1103 		printk(KERN_ERR
1104 		       "Attempt to unregister permanent protocol %d.\n",
1105 		       p->protocol);
1106 	} else {
1107 		spin_lock_bh(&inetsw_lock);
1108 		list_del_rcu(&p->list);
1109 		spin_unlock_bh(&inetsw_lock);
1110 
1111 		synchronize_net();
1112 	}
1113 }
1114 EXPORT_SYMBOL(inet_unregister_protosw);
1115 
1116 /*
1117  *      Shall we try to damage output packets if routing dev changes?
1118  */
1119 
1120 int sysctl_ip_dynaddr __read_mostly;
1121 
1122 static int inet_sk_reselect_saddr(struct sock *sk)
1123 {
1124 	struct inet_sock *inet = inet_sk(sk);
1125 	__be32 old_saddr = inet->inet_saddr;
1126 	__be32 daddr = inet->inet_daddr;
1127 	struct flowi4 *fl4;
1128 	struct rtable *rt;
1129 	__be32 new_saddr;
1130 	struct ip_options_rcu *inet_opt;
1131 
1132 	inet_opt = rcu_dereference_protected(inet->inet_opt,
1133 					     sock_owned_by_user(sk));
1134 	if (inet_opt && inet_opt->opt.srr)
1135 		daddr = inet_opt->opt.faddr;
1136 
1137 	/* Query new route. */
1138 	fl4 = &inet->cork.fl.u.ip4;
1139 	rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1140 			      sk->sk_bound_dev_if, sk->sk_protocol,
1141 			      inet->inet_sport, inet->inet_dport, sk, false);
1142 	if (IS_ERR(rt))
1143 		return PTR_ERR(rt);
1144 
1145 	sk_setup_caps(sk, &rt->dst);
1146 
1147 	new_saddr = fl4->saddr;
1148 
1149 	if (new_saddr == old_saddr)
1150 		return 0;
1151 
1152 	if (sysctl_ip_dynaddr > 1) {
1153 		printk(KERN_INFO "%s(): shifting inet->saddr from %pI4 to %pI4\n",
1154 		       __func__, &old_saddr, &new_saddr);
1155 	}
1156 
1157 	inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1158 
1159 	/*
1160 	 * XXX The only one ugly spot where we need to
1161 	 * XXX really change the sockets identity after
1162 	 * XXX it has entered the hashes. -DaveM
1163 	 *
1164 	 * Besides that, it does not check for connection
1165 	 * uniqueness. Wait for troubles.
1166 	 */
1167 	__sk_prot_rehash(sk);
1168 	return 0;
1169 }
1170 
1171 int inet_sk_rebuild_header(struct sock *sk)
1172 {
1173 	struct inet_sock *inet = inet_sk(sk);
1174 	struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1175 	__be32 daddr;
1176 	struct ip_options_rcu *inet_opt;
1177 	struct flowi4 *fl4;
1178 	int err;
1179 
1180 	/* Route is OK, nothing to do. */
1181 	if (rt)
1182 		return 0;
1183 
1184 	/* Reroute. */
1185 	rcu_read_lock();
1186 	inet_opt = rcu_dereference(inet->inet_opt);
1187 	daddr = inet->inet_daddr;
1188 	if (inet_opt && inet_opt->opt.srr)
1189 		daddr = inet_opt->opt.faddr;
1190 	rcu_read_unlock();
1191 	fl4 = &inet->cork.fl.u.ip4;
1192 	rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1193 				   inet->inet_dport, inet->inet_sport,
1194 				   sk->sk_protocol, RT_CONN_FLAGS(sk),
1195 				   sk->sk_bound_dev_if);
1196 	if (!IS_ERR(rt)) {
1197 		err = 0;
1198 		sk_setup_caps(sk, &rt->dst);
1199 	} else {
1200 		err = PTR_ERR(rt);
1201 
1202 		/* Routing failed... */
1203 		sk->sk_route_caps = 0;
1204 		/*
1205 		 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1206 		 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1207 		 */
1208 		if (!sysctl_ip_dynaddr ||
1209 		    sk->sk_state != TCP_SYN_SENT ||
1210 		    (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1211 		    (err = inet_sk_reselect_saddr(sk)) != 0)
1212 			sk->sk_err_soft = -err;
1213 	}
1214 
1215 	return err;
1216 }
1217 EXPORT_SYMBOL(inet_sk_rebuild_header);
1218 
1219 static int inet_gso_send_check(struct sk_buff *skb)
1220 {
1221 	const struct iphdr *iph;
1222 	const struct net_protocol *ops;
1223 	int proto;
1224 	int ihl;
1225 	int err = -EINVAL;
1226 
1227 	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1228 		goto out;
1229 
1230 	iph = ip_hdr(skb);
1231 	ihl = iph->ihl * 4;
1232 	if (ihl < sizeof(*iph))
1233 		goto out;
1234 
1235 	if (unlikely(!pskb_may_pull(skb, ihl)))
1236 		goto out;
1237 
1238 	__skb_pull(skb, ihl);
1239 	skb_reset_transport_header(skb);
1240 	iph = ip_hdr(skb);
1241 	proto = iph->protocol & (MAX_INET_PROTOS - 1);
1242 	err = -EPROTONOSUPPORT;
1243 
1244 	rcu_read_lock();
1245 	ops = rcu_dereference(inet_protos[proto]);
1246 	if (likely(ops && ops->gso_send_check))
1247 		err = ops->gso_send_check(skb);
1248 	rcu_read_unlock();
1249 
1250 out:
1251 	return err;
1252 }
1253 
1254 static struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1255 	netdev_features_t features)
1256 {
1257 	struct sk_buff *segs = ERR_PTR(-EINVAL);
1258 	struct iphdr *iph;
1259 	const struct net_protocol *ops;
1260 	int proto;
1261 	int ihl;
1262 	int id;
1263 	unsigned int offset = 0;
1264 
1265 	if (!(features & NETIF_F_V4_CSUM))
1266 		features &= ~NETIF_F_SG;
1267 
1268 	if (unlikely(skb_shinfo(skb)->gso_type &
1269 		     ~(SKB_GSO_TCPV4 |
1270 		       SKB_GSO_UDP |
1271 		       SKB_GSO_DODGY |
1272 		       SKB_GSO_TCP_ECN |
1273 		       0)))
1274 		goto out;
1275 
1276 	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1277 		goto out;
1278 
1279 	iph = ip_hdr(skb);
1280 	ihl = iph->ihl * 4;
1281 	if (ihl < sizeof(*iph))
1282 		goto out;
1283 
1284 	if (unlikely(!pskb_may_pull(skb, ihl)))
1285 		goto out;
1286 
1287 	__skb_pull(skb, ihl);
1288 	skb_reset_transport_header(skb);
1289 	iph = ip_hdr(skb);
1290 	id = ntohs(iph->id);
1291 	proto = iph->protocol & (MAX_INET_PROTOS - 1);
1292 	segs = ERR_PTR(-EPROTONOSUPPORT);
1293 
1294 	rcu_read_lock();
1295 	ops = rcu_dereference(inet_protos[proto]);
1296 	if (likely(ops && ops->gso_segment))
1297 		segs = ops->gso_segment(skb, features);
1298 	rcu_read_unlock();
1299 
1300 	if (!segs || IS_ERR(segs))
1301 		goto out;
1302 
1303 	skb = segs;
1304 	do {
1305 		iph = ip_hdr(skb);
1306 		if (proto == IPPROTO_UDP) {
1307 			iph->id = htons(id);
1308 			iph->frag_off = htons(offset >> 3);
1309 			if (skb->next != NULL)
1310 				iph->frag_off |= htons(IP_MF);
1311 			offset += (skb->len - skb->mac_len - iph->ihl * 4);
1312 		} else
1313 			iph->id = htons(id++);
1314 		iph->tot_len = htons(skb->len - skb->mac_len);
1315 		iph->check = 0;
1316 		iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1317 	} while ((skb = skb->next));
1318 
1319 out:
1320 	return segs;
1321 }
1322 
1323 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1324 					 struct sk_buff *skb)
1325 {
1326 	const struct net_protocol *ops;
1327 	struct sk_buff **pp = NULL;
1328 	struct sk_buff *p;
1329 	const struct iphdr *iph;
1330 	unsigned int hlen;
1331 	unsigned int off;
1332 	unsigned int id;
1333 	int flush = 1;
1334 	int proto;
1335 
1336 	off = skb_gro_offset(skb);
1337 	hlen = off + sizeof(*iph);
1338 	iph = skb_gro_header_fast(skb, off);
1339 	if (skb_gro_header_hard(skb, hlen)) {
1340 		iph = skb_gro_header_slow(skb, hlen, off);
1341 		if (unlikely(!iph))
1342 			goto out;
1343 	}
1344 
1345 	proto = iph->protocol & (MAX_INET_PROTOS - 1);
1346 
1347 	rcu_read_lock();
1348 	ops = rcu_dereference(inet_protos[proto]);
1349 	if (!ops || !ops->gro_receive)
1350 		goto out_unlock;
1351 
1352 	if (*(u8 *)iph != 0x45)
1353 		goto out_unlock;
1354 
1355 	if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1356 		goto out_unlock;
1357 
1358 	id = ntohl(*(__be32 *)&iph->id);
1359 	flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF));
1360 	id >>= 16;
1361 
1362 	for (p = *head; p; p = p->next) {
1363 		struct iphdr *iph2;
1364 
1365 		if (!NAPI_GRO_CB(p)->same_flow)
1366 			continue;
1367 
1368 		iph2 = ip_hdr(p);
1369 
1370 		if ((iph->protocol ^ iph2->protocol) |
1371 		    (iph->tos ^ iph2->tos) |
1372 		    ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1373 		    ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1374 			NAPI_GRO_CB(p)->same_flow = 0;
1375 			continue;
1376 		}
1377 
1378 		/* All fields must match except length and checksum. */
1379 		NAPI_GRO_CB(p)->flush |=
1380 			(iph->ttl ^ iph2->ttl) |
1381 			((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1382 
1383 		NAPI_GRO_CB(p)->flush |= flush;
1384 	}
1385 
1386 	NAPI_GRO_CB(skb)->flush |= flush;
1387 	skb_gro_pull(skb, sizeof(*iph));
1388 	skb_set_transport_header(skb, skb_gro_offset(skb));
1389 
1390 	pp = ops->gro_receive(head, skb);
1391 
1392 out_unlock:
1393 	rcu_read_unlock();
1394 
1395 out:
1396 	NAPI_GRO_CB(skb)->flush |= flush;
1397 
1398 	return pp;
1399 }
1400 
1401 static int inet_gro_complete(struct sk_buff *skb)
1402 {
1403 	const struct net_protocol *ops;
1404 	struct iphdr *iph = ip_hdr(skb);
1405 	int proto = iph->protocol & (MAX_INET_PROTOS - 1);
1406 	int err = -ENOSYS;
1407 	__be16 newlen = htons(skb->len - skb_network_offset(skb));
1408 
1409 	csum_replace2(&iph->check, iph->tot_len, newlen);
1410 	iph->tot_len = newlen;
1411 
1412 	rcu_read_lock();
1413 	ops = rcu_dereference(inet_protos[proto]);
1414 	if (WARN_ON(!ops || !ops->gro_complete))
1415 		goto out_unlock;
1416 
1417 	err = ops->gro_complete(skb);
1418 
1419 out_unlock:
1420 	rcu_read_unlock();
1421 
1422 	return err;
1423 }
1424 
1425 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1426 			 unsigned short type, unsigned char protocol,
1427 			 struct net *net)
1428 {
1429 	struct socket *sock;
1430 	int rc = sock_create_kern(family, type, protocol, &sock);
1431 
1432 	if (rc == 0) {
1433 		*sk = sock->sk;
1434 		(*sk)->sk_allocation = GFP_ATOMIC;
1435 		/*
1436 		 * Unhash it so that IP input processing does not even see it,
1437 		 * we do not wish this socket to see incoming packets.
1438 		 */
1439 		(*sk)->sk_prot->unhash(*sk);
1440 
1441 		sk_change_net(*sk, net);
1442 	}
1443 	return rc;
1444 }
1445 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1446 
1447 unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1448 {
1449 	unsigned long res = 0;
1450 	int i, j;
1451 
1452 	for_each_possible_cpu(i) {
1453 		for (j = 0; j < SNMP_ARRAY_SZ; j++)
1454 			res += *(((unsigned long *) per_cpu_ptr(mib[j], i)) + offt);
1455 	}
1456 	return res;
1457 }
1458 EXPORT_SYMBOL_GPL(snmp_fold_field);
1459 
1460 #if BITS_PER_LONG==32
1461 
1462 u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset)
1463 {
1464 	u64 res = 0;
1465 	int cpu;
1466 
1467 	for_each_possible_cpu(cpu) {
1468 		void *bhptr;
1469 		struct u64_stats_sync *syncp;
1470 		u64 v;
1471 		unsigned int start;
1472 
1473 		bhptr = per_cpu_ptr(mib[0], cpu);
1474 		syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1475 		do {
1476 			start = u64_stats_fetch_begin_bh(syncp);
1477 			v = *(((u64 *) bhptr) + offt);
1478 		} while (u64_stats_fetch_retry_bh(syncp, start));
1479 
1480 		res += v;
1481 	}
1482 	return res;
1483 }
1484 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1485 #endif
1486 
1487 int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align)
1488 {
1489 	BUG_ON(ptr == NULL);
1490 	ptr[0] = __alloc_percpu(mibsize, align);
1491 	if (!ptr[0])
1492 		return -ENOMEM;
1493 #if SNMP_ARRAY_SZ == 2
1494 	ptr[1] = __alloc_percpu(mibsize, align);
1495 	if (!ptr[1]) {
1496 		free_percpu(ptr[0]);
1497 		ptr[0] = NULL;
1498 		return -ENOMEM;
1499 	}
1500 #endif
1501 	return 0;
1502 }
1503 EXPORT_SYMBOL_GPL(snmp_mib_init);
1504 
1505 void snmp_mib_free(void __percpu *ptr[SNMP_ARRAY_SZ])
1506 {
1507 	int i;
1508 
1509 	BUG_ON(ptr == NULL);
1510 	for (i = 0; i < SNMP_ARRAY_SZ; i++) {
1511 		free_percpu(ptr[i]);
1512 		ptr[i] = NULL;
1513 	}
1514 }
1515 EXPORT_SYMBOL_GPL(snmp_mib_free);
1516 
1517 #ifdef CONFIG_IP_MULTICAST
1518 static const struct net_protocol igmp_protocol = {
1519 	.handler =	igmp_rcv,
1520 	.netns_ok =	1,
1521 };
1522 #endif
1523 
1524 static const struct net_protocol tcp_protocol = {
1525 	.handler =	tcp_v4_rcv,
1526 	.err_handler =	tcp_v4_err,
1527 	.gso_send_check = tcp_v4_gso_send_check,
1528 	.gso_segment =	tcp_tso_segment,
1529 	.gro_receive =	tcp4_gro_receive,
1530 	.gro_complete =	tcp4_gro_complete,
1531 	.no_policy =	1,
1532 	.netns_ok =	1,
1533 };
1534 
1535 static const struct net_protocol udp_protocol = {
1536 	.handler =	udp_rcv,
1537 	.err_handler =	udp_err,
1538 	.gso_send_check = udp4_ufo_send_check,
1539 	.gso_segment = udp4_ufo_fragment,
1540 	.no_policy =	1,
1541 	.netns_ok =	1,
1542 };
1543 
1544 static const struct net_protocol icmp_protocol = {
1545 	.handler =	icmp_rcv,
1546 	.err_handler =	ping_err,
1547 	.no_policy =	1,
1548 	.netns_ok =	1,
1549 };
1550 
1551 static __net_init int ipv4_mib_init_net(struct net *net)
1552 {
1553 	if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1554 			  sizeof(struct tcp_mib),
1555 			  __alignof__(struct tcp_mib)) < 0)
1556 		goto err_tcp_mib;
1557 	if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1558 			  sizeof(struct ipstats_mib),
1559 			  __alignof__(struct ipstats_mib)) < 0)
1560 		goto err_ip_mib;
1561 	if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1562 			  sizeof(struct linux_mib),
1563 			  __alignof__(struct linux_mib)) < 0)
1564 		goto err_net_mib;
1565 	if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1566 			  sizeof(struct udp_mib),
1567 			  __alignof__(struct udp_mib)) < 0)
1568 		goto err_udp_mib;
1569 	if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1570 			  sizeof(struct udp_mib),
1571 			  __alignof__(struct udp_mib)) < 0)
1572 		goto err_udplite_mib;
1573 	if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1574 			  sizeof(struct icmp_mib),
1575 			  __alignof__(struct icmp_mib)) < 0)
1576 		goto err_icmp_mib;
1577 	net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1578 					      GFP_KERNEL);
1579 	if (!net->mib.icmpmsg_statistics)
1580 		goto err_icmpmsg_mib;
1581 
1582 	tcp_mib_init(net);
1583 	return 0;
1584 
1585 err_icmpmsg_mib:
1586 	snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1587 err_icmp_mib:
1588 	snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1589 err_udplite_mib:
1590 	snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1591 err_udp_mib:
1592 	snmp_mib_free((void __percpu **)net->mib.net_statistics);
1593 err_net_mib:
1594 	snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1595 err_ip_mib:
1596 	snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1597 err_tcp_mib:
1598 	return -ENOMEM;
1599 }
1600 
1601 static __net_exit void ipv4_mib_exit_net(struct net *net)
1602 {
1603 	kfree(net->mib.icmpmsg_statistics);
1604 	snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1605 	snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1606 	snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1607 	snmp_mib_free((void __percpu **)net->mib.net_statistics);
1608 	snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1609 	snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1610 }
1611 
1612 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1613 	.init = ipv4_mib_init_net,
1614 	.exit = ipv4_mib_exit_net,
1615 };
1616 
1617 static int __init init_ipv4_mibs(void)
1618 {
1619 	return register_pernet_subsys(&ipv4_mib_ops);
1620 }
1621 
1622 static int ipv4_proc_init(void);
1623 
1624 /*
1625  *	IP protocol layer initialiser
1626  */
1627 
1628 static struct packet_type ip_packet_type __read_mostly = {
1629 	.type = cpu_to_be16(ETH_P_IP),
1630 	.func = ip_rcv,
1631 	.gso_send_check = inet_gso_send_check,
1632 	.gso_segment = inet_gso_segment,
1633 	.gro_receive = inet_gro_receive,
1634 	.gro_complete = inet_gro_complete,
1635 };
1636 
1637 static int __init inet_init(void)
1638 {
1639 	struct sk_buff *dummy_skb;
1640 	struct inet_protosw *q;
1641 	struct list_head *r;
1642 	int rc = -EINVAL;
1643 
1644 	BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1645 
1646 	sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL);
1647 	if (!sysctl_local_reserved_ports)
1648 		goto out;
1649 
1650 	rc = proto_register(&tcp_prot, 1);
1651 	if (rc)
1652 		goto out_free_reserved_ports;
1653 
1654 	rc = proto_register(&udp_prot, 1);
1655 	if (rc)
1656 		goto out_unregister_tcp_proto;
1657 
1658 	rc = proto_register(&raw_prot, 1);
1659 	if (rc)
1660 		goto out_unregister_udp_proto;
1661 
1662 	rc = proto_register(&ping_prot, 1);
1663 	if (rc)
1664 		goto out_unregister_raw_proto;
1665 
1666 	/*
1667 	 *	Tell SOCKET that we are alive...
1668 	 */
1669 
1670 	(void)sock_register(&inet_family_ops);
1671 
1672 #ifdef CONFIG_SYSCTL
1673 	ip_static_sysctl_init();
1674 #endif
1675 
1676 	tcp_prot.sysctl_mem = init_net.ipv4.sysctl_tcp_mem;
1677 
1678 	/*
1679 	 *	Add all the base protocols.
1680 	 */
1681 
1682 	if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1683 		printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1684 	if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1685 		printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1686 	if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1687 		printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1688 #ifdef CONFIG_IP_MULTICAST
1689 	if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1690 		printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1691 #endif
1692 
1693 	/* Register the socket-side information for inet_create. */
1694 	for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1695 		INIT_LIST_HEAD(r);
1696 
1697 	for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1698 		inet_register_protosw(q);
1699 
1700 	/*
1701 	 *	Set the ARP module up
1702 	 */
1703 
1704 	arp_init();
1705 
1706 	/*
1707 	 *	Set the IP module up
1708 	 */
1709 
1710 	ip_init();
1711 
1712 	tcp_v4_init();
1713 
1714 	/* Setup TCP slab cache for open requests. */
1715 	tcp_init();
1716 
1717 	/* Setup UDP memory threshold */
1718 	udp_init();
1719 
1720 	/* Add UDP-Lite (RFC 3828) */
1721 	udplite4_register();
1722 
1723 	ping_init();
1724 
1725 	/*
1726 	 *	Set the ICMP layer up
1727 	 */
1728 
1729 	if (icmp_init() < 0)
1730 		panic("Failed to create the ICMP control socket.\n");
1731 
1732 	/*
1733 	 *	Initialise the multicast router
1734 	 */
1735 #if defined(CONFIG_IP_MROUTE)
1736 	if (ip_mr_init())
1737 		printk(KERN_CRIT "inet_init: Cannot init ipv4 mroute\n");
1738 #endif
1739 	/*
1740 	 *	Initialise per-cpu ipv4 mibs
1741 	 */
1742 
1743 	if (init_ipv4_mibs())
1744 		printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n");
1745 
1746 	ipv4_proc_init();
1747 
1748 	ipfrag_init();
1749 
1750 	dev_add_pack(&ip_packet_type);
1751 
1752 	rc = 0;
1753 out:
1754 	return rc;
1755 out_unregister_raw_proto:
1756 	proto_unregister(&raw_prot);
1757 out_unregister_udp_proto:
1758 	proto_unregister(&udp_prot);
1759 out_unregister_tcp_proto:
1760 	proto_unregister(&tcp_prot);
1761 out_free_reserved_ports:
1762 	kfree(sysctl_local_reserved_ports);
1763 	goto out;
1764 }
1765 
1766 fs_initcall(inet_init);
1767 
1768 /* ------------------------------------------------------------------------ */
1769 
1770 #ifdef CONFIG_PROC_FS
1771 static int __init ipv4_proc_init(void)
1772 {
1773 	int rc = 0;
1774 
1775 	if (raw_proc_init())
1776 		goto out_raw;
1777 	if (tcp4_proc_init())
1778 		goto out_tcp;
1779 	if (udp4_proc_init())
1780 		goto out_udp;
1781 	if (ping_proc_init())
1782 		goto out_ping;
1783 	if (ip_misc_proc_init())
1784 		goto out_misc;
1785 out:
1786 	return rc;
1787 out_misc:
1788 	ping_proc_exit();
1789 out_ping:
1790 	udp4_proc_exit();
1791 out_udp:
1792 	tcp4_proc_exit();
1793 out_tcp:
1794 	raw_proc_exit();
1795 out_raw:
1796 	rc = -ENOMEM;
1797 	goto out;
1798 }
1799 
1800 #else /* CONFIG_PROC_FS */
1801 static int __init ipv4_proc_init(void)
1802 {
1803 	return 0;
1804 }
1805 #endif /* CONFIG_PROC_FS */
1806 
1807 MODULE_ALIAS_NETPROTO(PF_INET);
1808 
1809