xref: /linux/net/ipv4/af_inet.c (revision ec2212088c42ff7d1362629ec26dda4f3e8bdad3)
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 #define pr_fmt(fmt) "IPv4: " fmt
69 
70 #include <linux/err.h>
71 #include <linux/errno.h>
72 #include <linux/types.h>
73 #include <linux/socket.h>
74 #include <linux/in.h>
75 #include <linux/kernel.h>
76 #include <linux/module.h>
77 #include <linux/sched.h>
78 #include <linux/timer.h>
79 #include <linux/string.h>
80 #include <linux/sockios.h>
81 #include <linux/net.h>
82 #include <linux/capability.h>
83 #include <linux/fcntl.h>
84 #include <linux/mm.h>
85 #include <linux/interrupt.h>
86 #include <linux/stat.h>
87 #include <linux/init.h>
88 #include <linux/poll.h>
89 #include <linux/netfilter_ipv4.h>
90 #include <linux/random.h>
91 #include <linux/slab.h>
92 
93 #include <asm/uaccess.h>
94 #include <asm/system.h>
95 
96 #include <linux/inet.h>
97 #include <linux/igmp.h>
98 #include <linux/inetdevice.h>
99 #include <linux/netdevice.h>
100 #include <net/checksum.h>
101 #include <net/ip.h>
102 #include <net/protocol.h>
103 #include <net/arp.h>
104 #include <net/route.h>
105 #include <net/ip_fib.h>
106 #include <net/inet_connection_sock.h>
107 #include <net/tcp.h>
108 #include <net/udp.h>
109 #include <net/udplite.h>
110 #include <net/ping.h>
111 #include <linux/skbuff.h>
112 #include <net/sock.h>
113 #include <net/raw.h>
114 #include <net/icmp.h>
115 #include <net/ipip.h>
116 #include <net/inet_common.h>
117 #include <net/xfrm.h>
118 #include <net/net_namespace.h>
119 #ifdef CONFIG_IP_MROUTE
120 #include <linux/mroute.h>
121 #endif
122 
123 
124 /* The inetsw table contains everything that inet_create needs to
125  * build a new socket.
126  */
127 static struct list_head inetsw[SOCK_MAX];
128 static DEFINE_SPINLOCK(inetsw_lock);
129 
130 struct ipv4_config ipv4_config;
131 EXPORT_SYMBOL(ipv4_config);
132 
133 /* New destruction routine */
134 
135 void inet_sock_destruct(struct sock *sk)
136 {
137 	struct inet_sock *inet = inet_sk(sk);
138 
139 	__skb_queue_purge(&sk->sk_receive_queue);
140 	__skb_queue_purge(&sk->sk_error_queue);
141 
142 	sk_mem_reclaim(sk);
143 
144 	if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
145 		pr_err("Attempt to release TCP socket in state %d %p\n",
146 		       sk->sk_state, sk);
147 		return;
148 	}
149 	if (!sock_flag(sk, SOCK_DEAD)) {
150 		pr_err("Attempt to release alive inet socket %p\n", sk);
151 		return;
152 	}
153 
154 	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
155 	WARN_ON(atomic_read(&sk->sk_wmem_alloc));
156 	WARN_ON(sk->sk_wmem_queued);
157 	WARN_ON(sk->sk_forward_alloc);
158 
159 	kfree(rcu_dereference_protected(inet->inet_opt, 1));
160 	dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
161 	sk_refcnt_debug_dec(sk);
162 }
163 EXPORT_SYMBOL(inet_sock_destruct);
164 
165 /*
166  *	The routines beyond this point handle the behaviour of an AF_INET
167  *	socket object. Mostly it punts to the subprotocols of IP to do
168  *	the work.
169  */
170 
171 /*
172  *	Automatically bind an unbound socket.
173  */
174 
175 static int inet_autobind(struct sock *sk)
176 {
177 	struct inet_sock *inet;
178 	/* We may need to bind the socket. */
179 	lock_sock(sk);
180 	inet = inet_sk(sk);
181 	if (!inet->inet_num) {
182 		if (sk->sk_prot->get_port(sk, 0)) {
183 			release_sock(sk);
184 			return -EAGAIN;
185 		}
186 		inet->inet_sport = htons(inet->inet_num);
187 	}
188 	release_sock(sk);
189 	return 0;
190 }
191 
192 /*
193  *	Move a socket into listening state.
194  */
195 int inet_listen(struct socket *sock, int backlog)
196 {
197 	struct sock *sk = sock->sk;
198 	unsigned char old_state;
199 	int err;
200 
201 	lock_sock(sk);
202 
203 	err = -EINVAL;
204 	if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
205 		goto out;
206 
207 	old_state = sk->sk_state;
208 	if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
209 		goto out;
210 
211 	/* Really, if the socket is already in listen state
212 	 * we can only allow the backlog to be adjusted.
213 	 */
214 	if (old_state != TCP_LISTEN) {
215 		err = inet_csk_listen_start(sk, backlog);
216 		if (err)
217 			goto out;
218 	}
219 	sk->sk_max_ack_backlog = backlog;
220 	err = 0;
221 
222 out:
223 	release_sock(sk);
224 	return err;
225 }
226 EXPORT_SYMBOL(inet_listen);
227 
228 u32 inet_ehash_secret __read_mostly;
229 EXPORT_SYMBOL(inet_ehash_secret);
230 
231 /*
232  * inet_ehash_secret must be set exactly once
233  */
234 void build_ehash_secret(void)
235 {
236 	u32 rnd;
237 
238 	do {
239 		get_random_bytes(&rnd, sizeof(rnd));
240 	} while (rnd == 0);
241 
242 	cmpxchg(&inet_ehash_secret, 0, rnd);
243 }
244 EXPORT_SYMBOL(build_ehash_secret);
245 
246 static inline int inet_netns_ok(struct net *net, int protocol)
247 {
248 	int hash;
249 	const struct net_protocol *ipprot;
250 
251 	if (net_eq(net, &init_net))
252 		return 1;
253 
254 	hash = protocol & (MAX_INET_PROTOS - 1);
255 	ipprot = rcu_dereference(inet_protos[hash]);
256 
257 	if (ipprot == NULL)
258 		/* raw IP is OK */
259 		return 1;
260 	return ipprot->netns_ok;
261 }
262 
263 /*
264  *	Create an inet socket.
265  */
266 
267 static int inet_create(struct net *net, struct socket *sock, int protocol,
268 		       int kern)
269 {
270 	struct sock *sk;
271 	struct inet_protosw *answer;
272 	struct inet_sock *inet;
273 	struct proto *answer_prot;
274 	unsigned char answer_flags;
275 	char answer_no_check;
276 	int try_loading_module = 0;
277 	int err;
278 
279 	if (unlikely(!inet_ehash_secret))
280 		if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
281 			build_ehash_secret();
282 
283 	sock->state = SS_UNCONNECTED;
284 
285 	/* Look for the requested type/protocol pair. */
286 lookup_protocol:
287 	err = -ESOCKTNOSUPPORT;
288 	rcu_read_lock();
289 	list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
290 
291 		err = 0;
292 		/* Check the non-wild match. */
293 		if (protocol == answer->protocol) {
294 			if (protocol != IPPROTO_IP)
295 				break;
296 		} else {
297 			/* Check for the two wild cases. */
298 			if (IPPROTO_IP == protocol) {
299 				protocol = answer->protocol;
300 				break;
301 			}
302 			if (IPPROTO_IP == answer->protocol)
303 				break;
304 		}
305 		err = -EPROTONOSUPPORT;
306 	}
307 
308 	if (unlikely(err)) {
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 (sock->type == SOCK_RAW && !kern && !capable(CAP_NET_RAW))
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 	WARN_ON(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 	inet->nodefrag = 0;
360 
361 	if (SOCK_RAW == sock->type) {
362 		inet->inet_num = protocol;
363 		if (IPPROTO_RAW == protocol)
364 			inet->hdrincl = 1;
365 	}
366 
367 	if (ipv4_config.no_pmtu_disc)
368 		inet->pmtudisc = IP_PMTUDISC_DONT;
369 	else
370 		inet->pmtudisc = IP_PMTUDISC_WANT;
371 
372 	inet->inet_id = 0;
373 
374 	sock_init_data(sock, sk);
375 
376 	sk->sk_destruct	   = inet_sock_destruct;
377 	sk->sk_protocol	   = protocol;
378 	sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
379 
380 	inet->uc_ttl	= -1;
381 	inet->mc_loop	= 1;
382 	inet->mc_ttl	= 1;
383 	inet->mc_all	= 1;
384 	inet->mc_index	= 0;
385 	inet->mc_list	= NULL;
386 	inet->rcv_tos	= 0;
387 
388 	sk_refcnt_debug_inc(sk);
389 
390 	if (inet->inet_num) {
391 		/* It assumes that any protocol which allows
392 		 * the user to assign a number at socket
393 		 * creation time automatically
394 		 * shares.
395 		 */
396 		inet->inet_sport = htons(inet->inet_num);
397 		/* Add to protocol hash chains. */
398 		sk->sk_prot->hash(sk);
399 	}
400 
401 	if (sk->sk_prot->init) {
402 		err = sk->sk_prot->init(sk);
403 		if (err)
404 			sk_common_release(sk);
405 	}
406 out:
407 	return err;
408 out_rcu_unlock:
409 	rcu_read_unlock();
410 	goto out;
411 }
412 
413 
414 /*
415  *	The peer socket should always be NULL (or else). When we call this
416  *	function we are destroying the object and from then on nobody
417  *	should refer to it.
418  */
419 int inet_release(struct socket *sock)
420 {
421 	struct sock *sk = sock->sk;
422 
423 	if (sk) {
424 		long timeout;
425 
426 		sock_rps_reset_flow(sk);
427 
428 		/* Applications forget to leave groups before exiting */
429 		ip_mc_drop_socket(sk);
430 
431 		/* If linger is set, we don't return until the close
432 		 * is complete.  Otherwise we return immediately. The
433 		 * actually closing is done the same either way.
434 		 *
435 		 * If the close is due to the process exiting, we never
436 		 * linger..
437 		 */
438 		timeout = 0;
439 		if (sock_flag(sk, SOCK_LINGER) &&
440 		    !(current->flags & PF_EXITING))
441 			timeout = sk->sk_lingertime;
442 		sock->sk = NULL;
443 		sk->sk_prot->close(sk, timeout);
444 	}
445 	return 0;
446 }
447 EXPORT_SYMBOL(inet_release);
448 
449 /* It is off by default, see below. */
450 int sysctl_ip_nonlocal_bind __read_mostly;
451 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
452 
453 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
454 {
455 	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
456 	struct sock *sk = sock->sk;
457 	struct inet_sock *inet = inet_sk(sk);
458 	unsigned short snum;
459 	int chk_addr_ret;
460 	int err;
461 
462 	/* If the socket has its own bind function then use it. (RAW) */
463 	if (sk->sk_prot->bind) {
464 		err = sk->sk_prot->bind(sk, uaddr, addr_len);
465 		goto out;
466 	}
467 	err = -EINVAL;
468 	if (addr_len < sizeof(struct sockaddr_in))
469 		goto out;
470 
471 	if (addr->sin_family != AF_INET) {
472 		/* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
473 		 * only if s_addr is INADDR_ANY.
474 		 */
475 		err = -EAFNOSUPPORT;
476 		if (addr->sin_family != AF_UNSPEC ||
477 		    addr->sin_addr.s_addr != htonl(INADDR_ANY))
478 			goto out;
479 	}
480 
481 	chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr);
482 
483 	/* Not specified by any standard per-se, however it breaks too
484 	 * many applications when removed.  It is unfortunate since
485 	 * allowing applications to make a non-local bind solves
486 	 * several problems with systems using dynamic addressing.
487 	 * (ie. your servers still start up even if your ISDN link
488 	 *  is temporarily down)
489 	 */
490 	err = -EADDRNOTAVAIL;
491 	if (!sysctl_ip_nonlocal_bind &&
492 	    !(inet->freebind || inet->transparent) &&
493 	    addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
494 	    chk_addr_ret != RTN_LOCAL &&
495 	    chk_addr_ret != RTN_MULTICAST &&
496 	    chk_addr_ret != RTN_BROADCAST)
497 		goto out;
498 
499 	snum = ntohs(addr->sin_port);
500 	err = -EACCES;
501 	if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
502 		goto out;
503 
504 	/*      We keep a pair of addresses. rcv_saddr is the one
505 	 *      used by hash lookups, and saddr is used for transmit.
506 	 *
507 	 *      In the BSD API these are the same except where it
508 	 *      would be illegal to use them (multicast/broadcast) in
509 	 *      which case the sending device address is used.
510 	 */
511 	lock_sock(sk);
512 
513 	/* Check these errors (active socket, double bind). */
514 	err = -EINVAL;
515 	if (sk->sk_state != TCP_CLOSE || inet->inet_num)
516 		goto out_release_sock;
517 
518 	inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
519 	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
520 		inet->inet_saddr = 0;  /* Use device */
521 
522 	/* Make sure we are allowed to bind here. */
523 	if (sk->sk_prot->get_port(sk, snum)) {
524 		inet->inet_saddr = inet->inet_rcv_saddr = 0;
525 		err = -EADDRINUSE;
526 		goto out_release_sock;
527 	}
528 
529 	if (inet->inet_rcv_saddr)
530 		sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
531 	if (snum)
532 		sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
533 	inet->inet_sport = htons(inet->inet_num);
534 	inet->inet_daddr = 0;
535 	inet->inet_dport = 0;
536 	sk_dst_reset(sk);
537 	err = 0;
538 out_release_sock:
539 	release_sock(sk);
540 out:
541 	return err;
542 }
543 EXPORT_SYMBOL(inet_bind);
544 
545 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
546 		       int addr_len, int flags)
547 {
548 	struct sock *sk = sock->sk;
549 
550 	if (addr_len < sizeof(uaddr->sa_family))
551 		return -EINVAL;
552 	if (uaddr->sa_family == AF_UNSPEC)
553 		return sk->sk_prot->disconnect(sk, flags);
554 
555 	if (!inet_sk(sk)->inet_num && inet_autobind(sk))
556 		return -EAGAIN;
557 	return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
558 }
559 EXPORT_SYMBOL(inet_dgram_connect);
560 
561 static long inet_wait_for_connect(struct sock *sk, long timeo)
562 {
563 	DEFINE_WAIT(wait);
564 
565 	prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
566 
567 	/* Basic assumption: if someone sets sk->sk_err, he _must_
568 	 * change state of the socket from TCP_SYN_*.
569 	 * Connect() does not allow to get error notifications
570 	 * without closing the socket.
571 	 */
572 	while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
573 		release_sock(sk);
574 		timeo = schedule_timeout(timeo);
575 		lock_sock(sk);
576 		if (signal_pending(current) || !timeo)
577 			break;
578 		prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
579 	}
580 	finish_wait(sk_sleep(sk), &wait);
581 	return timeo;
582 }
583 
584 /*
585  *	Connect to a remote host. There is regrettably still a little
586  *	TCP 'magic' in here.
587  */
588 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
589 			int addr_len, int flags)
590 {
591 	struct sock *sk = sock->sk;
592 	int err;
593 	long timeo;
594 
595 	if (addr_len < sizeof(uaddr->sa_family))
596 		return -EINVAL;
597 
598 	lock_sock(sk);
599 
600 	if (uaddr->sa_family == AF_UNSPEC) {
601 		err = sk->sk_prot->disconnect(sk, flags);
602 		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
603 		goto out;
604 	}
605 
606 	switch (sock->state) {
607 	default:
608 		err = -EINVAL;
609 		goto out;
610 	case SS_CONNECTED:
611 		err = -EISCONN;
612 		goto out;
613 	case SS_CONNECTING:
614 		err = -EALREADY;
615 		/* Fall out of switch with err, set for this state */
616 		break;
617 	case SS_UNCONNECTED:
618 		err = -EISCONN;
619 		if (sk->sk_state != TCP_CLOSE)
620 			goto out;
621 
622 		err = sk->sk_prot->connect(sk, uaddr, addr_len);
623 		if (err < 0)
624 			goto out;
625 
626 		sock->state = SS_CONNECTING;
627 
628 		/* Just entered SS_CONNECTING state; the only
629 		 * difference is that return value in non-blocking
630 		 * case is EINPROGRESS, rather than EALREADY.
631 		 */
632 		err = -EINPROGRESS;
633 		break;
634 	}
635 
636 	timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
637 
638 	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
639 		/* Error code is set above */
640 		if (!timeo || !inet_wait_for_connect(sk, timeo))
641 			goto out;
642 
643 		err = sock_intr_errno(timeo);
644 		if (signal_pending(current))
645 			goto out;
646 	}
647 
648 	/* Connection was closed by RST, timeout, ICMP error
649 	 * or another process disconnected us.
650 	 */
651 	if (sk->sk_state == TCP_CLOSE)
652 		goto sock_error;
653 
654 	/* sk->sk_err may be not zero now, if RECVERR was ordered by user
655 	 * and error was received after socket entered established state.
656 	 * Hence, it is handled normally after connect() return successfully.
657 	 */
658 
659 	sock->state = SS_CONNECTED;
660 	err = 0;
661 out:
662 	release_sock(sk);
663 	return err;
664 
665 sock_error:
666 	err = sock_error(sk) ? : -ECONNABORTED;
667 	sock->state = SS_UNCONNECTED;
668 	if (sk->sk_prot->disconnect(sk, flags))
669 		sock->state = SS_DISCONNECTING;
670 	goto out;
671 }
672 EXPORT_SYMBOL(inet_stream_connect);
673 
674 /*
675  *	Accept a pending connection. The TCP layer now gives BSD semantics.
676  */
677 
678 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
679 {
680 	struct sock *sk1 = sock->sk;
681 	int err = -EINVAL;
682 	struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
683 
684 	if (!sk2)
685 		goto do_err;
686 
687 	lock_sock(sk2);
688 
689 	sock_rps_record_flow(sk2);
690 	WARN_ON(!((1 << sk2->sk_state) &
691 		  (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE)));
692 
693 	sock_graft(sk2, newsock);
694 
695 	newsock->state = SS_CONNECTED;
696 	err = 0;
697 	release_sock(sk2);
698 do_err:
699 	return err;
700 }
701 EXPORT_SYMBOL(inet_accept);
702 
703 
704 /*
705  *	This does both peername and sockname.
706  */
707 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
708 			int *uaddr_len, int peer)
709 {
710 	struct sock *sk		= sock->sk;
711 	struct inet_sock *inet	= inet_sk(sk);
712 	DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
713 
714 	sin->sin_family = AF_INET;
715 	if (peer) {
716 		if (!inet->inet_dport ||
717 		    (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
718 		     peer == 1))
719 			return -ENOTCONN;
720 		sin->sin_port = inet->inet_dport;
721 		sin->sin_addr.s_addr = inet->inet_daddr;
722 	} else {
723 		__be32 addr = inet->inet_rcv_saddr;
724 		if (!addr)
725 			addr = inet->inet_saddr;
726 		sin->sin_port = inet->inet_sport;
727 		sin->sin_addr.s_addr = addr;
728 	}
729 	memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
730 	*uaddr_len = sizeof(*sin);
731 	return 0;
732 }
733 EXPORT_SYMBOL(inet_getname);
734 
735 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
736 		 size_t size)
737 {
738 	struct sock *sk = sock->sk;
739 
740 	sock_rps_record_flow(sk);
741 
742 	/* We may need to bind the socket. */
743 	if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
744 	    inet_autobind(sk))
745 		return -EAGAIN;
746 
747 	return sk->sk_prot->sendmsg(iocb, sk, msg, size);
748 }
749 EXPORT_SYMBOL(inet_sendmsg);
750 
751 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
752 		      size_t size, int flags)
753 {
754 	struct sock *sk = sock->sk;
755 
756 	sock_rps_record_flow(sk);
757 
758 	/* We may need to bind the socket. */
759 	if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
760 	    inet_autobind(sk))
761 		return -EAGAIN;
762 
763 	if (sk->sk_prot->sendpage)
764 		return sk->sk_prot->sendpage(sk, page, offset, size, flags);
765 	return sock_no_sendpage(sock, page, offset, size, flags);
766 }
767 EXPORT_SYMBOL(inet_sendpage);
768 
769 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
770 		 size_t size, int flags)
771 {
772 	struct sock *sk = sock->sk;
773 	int addr_len = 0;
774 	int err;
775 
776 	sock_rps_record_flow(sk);
777 
778 	err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
779 				   flags & ~MSG_DONTWAIT, &addr_len);
780 	if (err >= 0)
781 		msg->msg_namelen = addr_len;
782 	return err;
783 }
784 EXPORT_SYMBOL(inet_recvmsg);
785 
786 int inet_shutdown(struct socket *sock, int how)
787 {
788 	struct sock *sk = sock->sk;
789 	int err = 0;
790 
791 	/* This should really check to make sure
792 	 * the socket is a TCP socket. (WHY AC...)
793 	 */
794 	how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
795 		       1->2 bit 2 snds.
796 		       2->3 */
797 	if ((how & ~SHUTDOWN_MASK) || !how)	/* MAXINT->0 */
798 		return -EINVAL;
799 
800 	lock_sock(sk);
801 	if (sock->state == SS_CONNECTING) {
802 		if ((1 << sk->sk_state) &
803 		    (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
804 			sock->state = SS_DISCONNECTING;
805 		else
806 			sock->state = SS_CONNECTED;
807 	}
808 
809 	switch (sk->sk_state) {
810 	case TCP_CLOSE:
811 		err = -ENOTCONN;
812 		/* Hack to wake up other listeners, who can poll for
813 		   POLLHUP, even on eg. unconnected UDP sockets -- RR */
814 	default:
815 		sk->sk_shutdown |= how;
816 		if (sk->sk_prot->shutdown)
817 			sk->sk_prot->shutdown(sk, how);
818 		break;
819 
820 	/* Remaining two branches are temporary solution for missing
821 	 * close() in multithreaded environment. It is _not_ a good idea,
822 	 * but we have no choice until close() is repaired at VFS level.
823 	 */
824 	case TCP_LISTEN:
825 		if (!(how & RCV_SHUTDOWN))
826 			break;
827 		/* Fall through */
828 	case TCP_SYN_SENT:
829 		err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
830 		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
831 		break;
832 	}
833 
834 	/* Wake up anyone sleeping in poll. */
835 	sk->sk_state_change(sk);
836 	release_sock(sk);
837 	return err;
838 }
839 EXPORT_SYMBOL(inet_shutdown);
840 
841 /*
842  *	ioctl() calls you can issue on an INET socket. Most of these are
843  *	device configuration and stuff and very rarely used. Some ioctls
844  *	pass on to the socket itself.
845  *
846  *	NOTE: I like the idea of a module for the config stuff. ie ifconfig
847  *	loads the devconfigure module does its configuring and unloads it.
848  *	There's a good 20K of config code hanging around the kernel.
849  */
850 
851 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
852 {
853 	struct sock *sk = sock->sk;
854 	int err = 0;
855 	struct net *net = sock_net(sk);
856 
857 	switch (cmd) {
858 	case SIOCGSTAMP:
859 		err = sock_get_timestamp(sk, (struct timeval __user *)arg);
860 		break;
861 	case SIOCGSTAMPNS:
862 		err = sock_get_timestampns(sk, (struct timespec __user *)arg);
863 		break;
864 	case SIOCADDRT:
865 	case SIOCDELRT:
866 	case SIOCRTMSG:
867 		err = ip_rt_ioctl(net, cmd, (void __user *)arg);
868 		break;
869 	case SIOCDARP:
870 	case SIOCGARP:
871 	case SIOCSARP:
872 		err = arp_ioctl(net, cmd, (void __user *)arg);
873 		break;
874 	case SIOCGIFADDR:
875 	case SIOCSIFADDR:
876 	case SIOCGIFBRDADDR:
877 	case SIOCSIFBRDADDR:
878 	case SIOCGIFNETMASK:
879 	case SIOCSIFNETMASK:
880 	case SIOCGIFDSTADDR:
881 	case SIOCSIFDSTADDR:
882 	case SIOCSIFPFLAGS:
883 	case SIOCGIFPFLAGS:
884 	case SIOCSIFFLAGS:
885 		err = devinet_ioctl(net, cmd, (void __user *)arg);
886 		break;
887 	default:
888 		if (sk->sk_prot->ioctl)
889 			err = sk->sk_prot->ioctl(sk, cmd, arg);
890 		else
891 			err = -ENOIOCTLCMD;
892 		break;
893 	}
894 	return err;
895 }
896 EXPORT_SYMBOL(inet_ioctl);
897 
898 #ifdef CONFIG_COMPAT
899 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
900 {
901 	struct sock *sk = sock->sk;
902 	int err = -ENOIOCTLCMD;
903 
904 	if (sk->sk_prot->compat_ioctl)
905 		err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
906 
907 	return err;
908 }
909 #endif
910 
911 const struct proto_ops inet_stream_ops = {
912 	.family		   = PF_INET,
913 	.owner		   = THIS_MODULE,
914 	.release	   = inet_release,
915 	.bind		   = inet_bind,
916 	.connect	   = inet_stream_connect,
917 	.socketpair	   = sock_no_socketpair,
918 	.accept		   = inet_accept,
919 	.getname	   = inet_getname,
920 	.poll		   = tcp_poll,
921 	.ioctl		   = inet_ioctl,
922 	.listen		   = inet_listen,
923 	.shutdown	   = inet_shutdown,
924 	.setsockopt	   = sock_common_setsockopt,
925 	.getsockopt	   = sock_common_getsockopt,
926 	.sendmsg	   = inet_sendmsg,
927 	.recvmsg	   = inet_recvmsg,
928 	.mmap		   = sock_no_mmap,
929 	.sendpage	   = inet_sendpage,
930 	.splice_read	   = tcp_splice_read,
931 #ifdef CONFIG_COMPAT
932 	.compat_setsockopt = compat_sock_common_setsockopt,
933 	.compat_getsockopt = compat_sock_common_getsockopt,
934 	.compat_ioctl	   = inet_compat_ioctl,
935 #endif
936 };
937 EXPORT_SYMBOL(inet_stream_ops);
938 
939 const struct proto_ops inet_dgram_ops = {
940 	.family		   = PF_INET,
941 	.owner		   = THIS_MODULE,
942 	.release	   = inet_release,
943 	.bind		   = inet_bind,
944 	.connect	   = inet_dgram_connect,
945 	.socketpair	   = sock_no_socketpair,
946 	.accept		   = sock_no_accept,
947 	.getname	   = inet_getname,
948 	.poll		   = udp_poll,
949 	.ioctl		   = inet_ioctl,
950 	.listen		   = sock_no_listen,
951 	.shutdown	   = inet_shutdown,
952 	.setsockopt	   = sock_common_setsockopt,
953 	.getsockopt	   = sock_common_getsockopt,
954 	.sendmsg	   = inet_sendmsg,
955 	.recvmsg	   = inet_recvmsg,
956 	.mmap		   = sock_no_mmap,
957 	.sendpage	   = inet_sendpage,
958 #ifdef CONFIG_COMPAT
959 	.compat_setsockopt = compat_sock_common_setsockopt,
960 	.compat_getsockopt = compat_sock_common_getsockopt,
961 	.compat_ioctl	   = inet_compat_ioctl,
962 #endif
963 };
964 EXPORT_SYMBOL(inet_dgram_ops);
965 
966 /*
967  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
968  * udp_poll
969  */
970 static const struct proto_ops inet_sockraw_ops = {
971 	.family		   = PF_INET,
972 	.owner		   = THIS_MODULE,
973 	.release	   = inet_release,
974 	.bind		   = inet_bind,
975 	.connect	   = inet_dgram_connect,
976 	.socketpair	   = sock_no_socketpair,
977 	.accept		   = sock_no_accept,
978 	.getname	   = inet_getname,
979 	.poll		   = datagram_poll,
980 	.ioctl		   = inet_ioctl,
981 	.listen		   = sock_no_listen,
982 	.shutdown	   = inet_shutdown,
983 	.setsockopt	   = sock_common_setsockopt,
984 	.getsockopt	   = sock_common_getsockopt,
985 	.sendmsg	   = inet_sendmsg,
986 	.recvmsg	   = inet_recvmsg,
987 	.mmap		   = sock_no_mmap,
988 	.sendpage	   = inet_sendpage,
989 #ifdef CONFIG_COMPAT
990 	.compat_setsockopt = compat_sock_common_setsockopt,
991 	.compat_getsockopt = compat_sock_common_getsockopt,
992 	.compat_ioctl	   = inet_compat_ioctl,
993 #endif
994 };
995 
996 static const struct net_proto_family inet_family_ops = {
997 	.family = PF_INET,
998 	.create = inet_create,
999 	.owner	= THIS_MODULE,
1000 };
1001 
1002 /* Upon startup we insert all the elements in inetsw_array[] into
1003  * the linked list inetsw.
1004  */
1005 static struct inet_protosw inetsw_array[] =
1006 {
1007 	{
1008 		.type =       SOCK_STREAM,
1009 		.protocol =   IPPROTO_TCP,
1010 		.prot =       &tcp_prot,
1011 		.ops =        &inet_stream_ops,
1012 		.no_check =   0,
1013 		.flags =      INET_PROTOSW_PERMANENT |
1014 			      INET_PROTOSW_ICSK,
1015 	},
1016 
1017 	{
1018 		.type =       SOCK_DGRAM,
1019 		.protocol =   IPPROTO_UDP,
1020 		.prot =       &udp_prot,
1021 		.ops =        &inet_dgram_ops,
1022 		.no_check =   UDP_CSUM_DEFAULT,
1023 		.flags =      INET_PROTOSW_PERMANENT,
1024        },
1025 
1026        {
1027 		.type =       SOCK_DGRAM,
1028 		.protocol =   IPPROTO_ICMP,
1029 		.prot =       &ping_prot,
1030 		.ops =        &inet_dgram_ops,
1031 		.no_check =   UDP_CSUM_DEFAULT,
1032 		.flags =      INET_PROTOSW_REUSE,
1033        },
1034 
1035        {
1036 	       .type =       SOCK_RAW,
1037 	       .protocol =   IPPROTO_IP,	/* wild card */
1038 	       .prot =       &raw_prot,
1039 	       .ops =        &inet_sockraw_ops,
1040 	       .no_check =   UDP_CSUM_DEFAULT,
1041 	       .flags =      INET_PROTOSW_REUSE,
1042        }
1043 };
1044 
1045 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1046 
1047 void inet_register_protosw(struct inet_protosw *p)
1048 {
1049 	struct list_head *lh;
1050 	struct inet_protosw *answer;
1051 	int protocol = p->protocol;
1052 	struct list_head *last_perm;
1053 
1054 	spin_lock_bh(&inetsw_lock);
1055 
1056 	if (p->type >= SOCK_MAX)
1057 		goto out_illegal;
1058 
1059 	/* If we are trying to override a permanent protocol, bail. */
1060 	answer = NULL;
1061 	last_perm = &inetsw[p->type];
1062 	list_for_each(lh, &inetsw[p->type]) {
1063 		answer = list_entry(lh, struct inet_protosw, list);
1064 
1065 		/* Check only the non-wild match. */
1066 		if (INET_PROTOSW_PERMANENT & answer->flags) {
1067 			if (protocol == answer->protocol)
1068 				break;
1069 			last_perm = lh;
1070 		}
1071 
1072 		answer = NULL;
1073 	}
1074 	if (answer)
1075 		goto out_permanent;
1076 
1077 	/* Add the new entry after the last permanent entry if any, so that
1078 	 * the new entry does not override a permanent entry when matched with
1079 	 * a wild-card protocol. But it is allowed to override any existing
1080 	 * non-permanent entry.  This means that when we remove this entry, the
1081 	 * system automatically returns to the old behavior.
1082 	 */
1083 	list_add_rcu(&p->list, last_perm);
1084 out:
1085 	spin_unlock_bh(&inetsw_lock);
1086 
1087 	return;
1088 
1089 out_permanent:
1090 	pr_err("Attempt to override permanent protocol %d\n", protocol);
1091 	goto out;
1092 
1093 out_illegal:
1094 	pr_err("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 		pr_err("Attempt to unregister permanent protocol %d\n",
1104 		       p->protocol);
1105 	} else {
1106 		spin_lock_bh(&inetsw_lock);
1107 		list_del_rcu(&p->list);
1108 		spin_unlock_bh(&inetsw_lock);
1109 
1110 		synchronize_net();
1111 	}
1112 }
1113 EXPORT_SYMBOL(inet_unregister_protosw);
1114 
1115 /*
1116  *      Shall we try to damage output packets if routing dev changes?
1117  */
1118 
1119 int sysctl_ip_dynaddr __read_mostly;
1120 
1121 static int inet_sk_reselect_saddr(struct sock *sk)
1122 {
1123 	struct inet_sock *inet = inet_sk(sk);
1124 	__be32 old_saddr = inet->inet_saddr;
1125 	__be32 daddr = inet->inet_daddr;
1126 	struct flowi4 *fl4;
1127 	struct rtable *rt;
1128 	__be32 new_saddr;
1129 	struct ip_options_rcu *inet_opt;
1130 
1131 	inet_opt = rcu_dereference_protected(inet->inet_opt,
1132 					     sock_owned_by_user(sk));
1133 	if (inet_opt && inet_opt->opt.srr)
1134 		daddr = inet_opt->opt.faddr;
1135 
1136 	/* Query new route. */
1137 	fl4 = &inet->cork.fl.u.ip4;
1138 	rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1139 			      sk->sk_bound_dev_if, sk->sk_protocol,
1140 			      inet->inet_sport, inet->inet_dport, sk, false);
1141 	if (IS_ERR(rt))
1142 		return PTR_ERR(rt);
1143 
1144 	sk_setup_caps(sk, &rt->dst);
1145 
1146 	new_saddr = fl4->saddr;
1147 
1148 	if (new_saddr == old_saddr)
1149 		return 0;
1150 
1151 	if (sysctl_ip_dynaddr > 1) {
1152 		pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1153 			__func__, &old_saddr, &new_saddr);
1154 	}
1155 
1156 	inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1157 
1158 	/*
1159 	 * XXX The only one ugly spot where we need to
1160 	 * XXX really change the sockets identity after
1161 	 * XXX it has entered the hashes. -DaveM
1162 	 *
1163 	 * Besides that, it does not check for connection
1164 	 * uniqueness. Wait for troubles.
1165 	 */
1166 	__sk_prot_rehash(sk);
1167 	return 0;
1168 }
1169 
1170 int inet_sk_rebuild_header(struct sock *sk)
1171 {
1172 	struct inet_sock *inet = inet_sk(sk);
1173 	struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1174 	__be32 daddr;
1175 	struct ip_options_rcu *inet_opt;
1176 	struct flowi4 *fl4;
1177 	int err;
1178 
1179 	/* Route is OK, nothing to do. */
1180 	if (rt)
1181 		return 0;
1182 
1183 	/* Reroute. */
1184 	rcu_read_lock();
1185 	inet_opt = rcu_dereference(inet->inet_opt);
1186 	daddr = inet->inet_daddr;
1187 	if (inet_opt && inet_opt->opt.srr)
1188 		daddr = inet_opt->opt.faddr;
1189 	rcu_read_unlock();
1190 	fl4 = &inet->cork.fl.u.ip4;
1191 	rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1192 				   inet->inet_dport, inet->inet_sport,
1193 				   sk->sk_protocol, RT_CONN_FLAGS(sk),
1194 				   sk->sk_bound_dev_if);
1195 	if (!IS_ERR(rt)) {
1196 		err = 0;
1197 		sk_setup_caps(sk, &rt->dst);
1198 	} else {
1199 		err = PTR_ERR(rt);
1200 
1201 		/* Routing failed... */
1202 		sk->sk_route_caps = 0;
1203 		/*
1204 		 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1205 		 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1206 		 */
1207 		if (!sysctl_ip_dynaddr ||
1208 		    sk->sk_state != TCP_SYN_SENT ||
1209 		    (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1210 		    (err = inet_sk_reselect_saddr(sk)) != 0)
1211 			sk->sk_err_soft = -err;
1212 	}
1213 
1214 	return err;
1215 }
1216 EXPORT_SYMBOL(inet_sk_rebuild_header);
1217 
1218 static int inet_gso_send_check(struct sk_buff *skb)
1219 {
1220 	const struct iphdr *iph;
1221 	const struct net_protocol *ops;
1222 	int proto;
1223 	int ihl;
1224 	int err = -EINVAL;
1225 
1226 	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1227 		goto out;
1228 
1229 	iph = ip_hdr(skb);
1230 	ihl = iph->ihl * 4;
1231 	if (ihl < sizeof(*iph))
1232 		goto out;
1233 
1234 	if (unlikely(!pskb_may_pull(skb, ihl)))
1235 		goto out;
1236 
1237 	__skb_pull(skb, ihl);
1238 	skb_reset_transport_header(skb);
1239 	iph = ip_hdr(skb);
1240 	proto = iph->protocol & (MAX_INET_PROTOS - 1);
1241 	err = -EPROTONOSUPPORT;
1242 
1243 	rcu_read_lock();
1244 	ops = rcu_dereference(inet_protos[proto]);
1245 	if (likely(ops && ops->gso_send_check))
1246 		err = ops->gso_send_check(skb);
1247 	rcu_read_unlock();
1248 
1249 out:
1250 	return err;
1251 }
1252 
1253 static struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1254 	netdev_features_t features)
1255 {
1256 	struct sk_buff *segs = ERR_PTR(-EINVAL);
1257 	struct iphdr *iph;
1258 	const struct net_protocol *ops;
1259 	int proto;
1260 	int ihl;
1261 	int id;
1262 	unsigned int offset = 0;
1263 
1264 	if (!(features & NETIF_F_V4_CSUM))
1265 		features &= ~NETIF_F_SG;
1266 
1267 	if (unlikely(skb_shinfo(skb)->gso_type &
1268 		     ~(SKB_GSO_TCPV4 |
1269 		       SKB_GSO_UDP |
1270 		       SKB_GSO_DODGY |
1271 		       SKB_GSO_TCP_ECN |
1272 		       0)))
1273 		goto out;
1274 
1275 	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1276 		goto out;
1277 
1278 	iph = ip_hdr(skb);
1279 	ihl = iph->ihl * 4;
1280 	if (ihl < sizeof(*iph))
1281 		goto out;
1282 
1283 	if (unlikely(!pskb_may_pull(skb, ihl)))
1284 		goto out;
1285 
1286 	__skb_pull(skb, ihl);
1287 	skb_reset_transport_header(skb);
1288 	iph = ip_hdr(skb);
1289 	id = ntohs(iph->id);
1290 	proto = iph->protocol & (MAX_INET_PROTOS - 1);
1291 	segs = ERR_PTR(-EPROTONOSUPPORT);
1292 
1293 	rcu_read_lock();
1294 	ops = rcu_dereference(inet_protos[proto]);
1295 	if (likely(ops && ops->gso_segment))
1296 		segs = ops->gso_segment(skb, features);
1297 	rcu_read_unlock();
1298 
1299 	if (!segs || IS_ERR(segs))
1300 		goto out;
1301 
1302 	skb = segs;
1303 	do {
1304 		iph = ip_hdr(skb);
1305 		if (proto == IPPROTO_UDP) {
1306 			iph->id = htons(id);
1307 			iph->frag_off = htons(offset >> 3);
1308 			if (skb->next != NULL)
1309 				iph->frag_off |= htons(IP_MF);
1310 			offset += (skb->len - skb->mac_len - iph->ihl * 4);
1311 		} else
1312 			iph->id = htons(id++);
1313 		iph->tot_len = htons(skb->len - skb->mac_len);
1314 		iph->check = 0;
1315 		iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1316 	} while ((skb = skb->next));
1317 
1318 out:
1319 	return segs;
1320 }
1321 
1322 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1323 					 struct sk_buff *skb)
1324 {
1325 	const struct net_protocol *ops;
1326 	struct sk_buff **pp = NULL;
1327 	struct sk_buff *p;
1328 	const struct iphdr *iph;
1329 	unsigned int hlen;
1330 	unsigned int off;
1331 	unsigned int id;
1332 	int flush = 1;
1333 	int proto;
1334 
1335 	off = skb_gro_offset(skb);
1336 	hlen = off + sizeof(*iph);
1337 	iph = skb_gro_header_fast(skb, off);
1338 	if (skb_gro_header_hard(skb, hlen)) {
1339 		iph = skb_gro_header_slow(skb, hlen, off);
1340 		if (unlikely(!iph))
1341 			goto out;
1342 	}
1343 
1344 	proto = iph->protocol & (MAX_INET_PROTOS - 1);
1345 
1346 	rcu_read_lock();
1347 	ops = rcu_dereference(inet_protos[proto]);
1348 	if (!ops || !ops->gro_receive)
1349 		goto out_unlock;
1350 
1351 	if (*(u8 *)iph != 0x45)
1352 		goto out_unlock;
1353 
1354 	if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1355 		goto out_unlock;
1356 
1357 	id = ntohl(*(__be32 *)&iph->id);
1358 	flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF));
1359 	id >>= 16;
1360 
1361 	for (p = *head; p; p = p->next) {
1362 		struct iphdr *iph2;
1363 
1364 		if (!NAPI_GRO_CB(p)->same_flow)
1365 			continue;
1366 
1367 		iph2 = ip_hdr(p);
1368 
1369 		if ((iph->protocol ^ iph2->protocol) |
1370 		    (iph->tos ^ iph2->tos) |
1371 		    ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1372 		    ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1373 			NAPI_GRO_CB(p)->same_flow = 0;
1374 			continue;
1375 		}
1376 
1377 		/* All fields must match except length and checksum. */
1378 		NAPI_GRO_CB(p)->flush |=
1379 			(iph->ttl ^ iph2->ttl) |
1380 			((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1381 
1382 		NAPI_GRO_CB(p)->flush |= flush;
1383 	}
1384 
1385 	NAPI_GRO_CB(skb)->flush |= flush;
1386 	skb_gro_pull(skb, sizeof(*iph));
1387 	skb_set_transport_header(skb, skb_gro_offset(skb));
1388 
1389 	pp = ops->gro_receive(head, skb);
1390 
1391 out_unlock:
1392 	rcu_read_unlock();
1393 
1394 out:
1395 	NAPI_GRO_CB(skb)->flush |= flush;
1396 
1397 	return pp;
1398 }
1399 
1400 static int inet_gro_complete(struct sk_buff *skb)
1401 {
1402 	const struct net_protocol *ops;
1403 	struct iphdr *iph = ip_hdr(skb);
1404 	int proto = iph->protocol & (MAX_INET_PROTOS - 1);
1405 	int err = -ENOSYS;
1406 	__be16 newlen = htons(skb->len - skb_network_offset(skb));
1407 
1408 	csum_replace2(&iph->check, iph->tot_len, newlen);
1409 	iph->tot_len = newlen;
1410 
1411 	rcu_read_lock();
1412 	ops = rcu_dereference(inet_protos[proto]);
1413 	if (WARN_ON(!ops || !ops->gro_complete))
1414 		goto out_unlock;
1415 
1416 	err = ops->gro_complete(skb);
1417 
1418 out_unlock:
1419 	rcu_read_unlock();
1420 
1421 	return err;
1422 }
1423 
1424 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1425 			 unsigned short type, unsigned char protocol,
1426 			 struct net *net)
1427 {
1428 	struct socket *sock;
1429 	int rc = sock_create_kern(family, type, protocol, &sock);
1430 
1431 	if (rc == 0) {
1432 		*sk = sock->sk;
1433 		(*sk)->sk_allocation = GFP_ATOMIC;
1434 		/*
1435 		 * Unhash it so that IP input processing does not even see it,
1436 		 * we do not wish this socket to see incoming packets.
1437 		 */
1438 		(*sk)->sk_prot->unhash(*sk);
1439 
1440 		sk_change_net(*sk, net);
1441 	}
1442 	return rc;
1443 }
1444 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1445 
1446 unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1447 {
1448 	unsigned long res = 0;
1449 	int i, j;
1450 
1451 	for_each_possible_cpu(i) {
1452 		for (j = 0; j < SNMP_ARRAY_SZ; j++)
1453 			res += *(((unsigned long *) per_cpu_ptr(mib[j], i)) + offt);
1454 	}
1455 	return res;
1456 }
1457 EXPORT_SYMBOL_GPL(snmp_fold_field);
1458 
1459 #if BITS_PER_LONG==32
1460 
1461 u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset)
1462 {
1463 	u64 res = 0;
1464 	int cpu;
1465 
1466 	for_each_possible_cpu(cpu) {
1467 		void *bhptr;
1468 		struct u64_stats_sync *syncp;
1469 		u64 v;
1470 		unsigned int start;
1471 
1472 		bhptr = per_cpu_ptr(mib[0], cpu);
1473 		syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1474 		do {
1475 			start = u64_stats_fetch_begin_bh(syncp);
1476 			v = *(((u64 *) bhptr) + offt);
1477 		} while (u64_stats_fetch_retry_bh(syncp, start));
1478 
1479 		res += v;
1480 	}
1481 	return res;
1482 }
1483 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1484 #endif
1485 
1486 int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align)
1487 {
1488 	BUG_ON(ptr == NULL);
1489 	ptr[0] = __alloc_percpu(mibsize, align);
1490 	if (!ptr[0])
1491 		return -ENOMEM;
1492 #if SNMP_ARRAY_SZ == 2
1493 	ptr[1] = __alloc_percpu(mibsize, align);
1494 	if (!ptr[1]) {
1495 		free_percpu(ptr[0]);
1496 		ptr[0] = NULL;
1497 		return -ENOMEM;
1498 	}
1499 #endif
1500 	return 0;
1501 }
1502 EXPORT_SYMBOL_GPL(snmp_mib_init);
1503 
1504 void snmp_mib_free(void __percpu *ptr[SNMP_ARRAY_SZ])
1505 {
1506 	int i;
1507 
1508 	BUG_ON(ptr == NULL);
1509 	for (i = 0; i < SNMP_ARRAY_SZ; i++) {
1510 		free_percpu(ptr[i]);
1511 		ptr[i] = NULL;
1512 	}
1513 }
1514 EXPORT_SYMBOL_GPL(snmp_mib_free);
1515 
1516 #ifdef CONFIG_IP_MULTICAST
1517 static const struct net_protocol igmp_protocol = {
1518 	.handler =	igmp_rcv,
1519 	.netns_ok =	1,
1520 };
1521 #endif
1522 
1523 static const struct net_protocol tcp_protocol = {
1524 	.handler =	tcp_v4_rcv,
1525 	.err_handler =	tcp_v4_err,
1526 	.gso_send_check = tcp_v4_gso_send_check,
1527 	.gso_segment =	tcp_tso_segment,
1528 	.gro_receive =	tcp4_gro_receive,
1529 	.gro_complete =	tcp4_gro_complete,
1530 	.no_policy =	1,
1531 	.netns_ok =	1,
1532 };
1533 
1534 static const struct net_protocol udp_protocol = {
1535 	.handler =	udp_rcv,
1536 	.err_handler =	udp_err,
1537 	.gso_send_check = udp4_ufo_send_check,
1538 	.gso_segment = udp4_ufo_fragment,
1539 	.no_policy =	1,
1540 	.netns_ok =	1,
1541 };
1542 
1543 static const struct net_protocol icmp_protocol = {
1544 	.handler =	icmp_rcv,
1545 	.err_handler =	ping_err,
1546 	.no_policy =	1,
1547 	.netns_ok =	1,
1548 };
1549 
1550 static __net_init int ipv4_mib_init_net(struct net *net)
1551 {
1552 	if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1553 			  sizeof(struct tcp_mib),
1554 			  __alignof__(struct tcp_mib)) < 0)
1555 		goto err_tcp_mib;
1556 	if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1557 			  sizeof(struct ipstats_mib),
1558 			  __alignof__(struct ipstats_mib)) < 0)
1559 		goto err_ip_mib;
1560 	if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1561 			  sizeof(struct linux_mib),
1562 			  __alignof__(struct linux_mib)) < 0)
1563 		goto err_net_mib;
1564 	if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1565 			  sizeof(struct udp_mib),
1566 			  __alignof__(struct udp_mib)) < 0)
1567 		goto err_udp_mib;
1568 	if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1569 			  sizeof(struct udp_mib),
1570 			  __alignof__(struct udp_mib)) < 0)
1571 		goto err_udplite_mib;
1572 	if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1573 			  sizeof(struct icmp_mib),
1574 			  __alignof__(struct icmp_mib)) < 0)
1575 		goto err_icmp_mib;
1576 	net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1577 					      GFP_KERNEL);
1578 	if (!net->mib.icmpmsg_statistics)
1579 		goto err_icmpmsg_mib;
1580 
1581 	tcp_mib_init(net);
1582 	return 0;
1583 
1584 err_icmpmsg_mib:
1585 	snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1586 err_icmp_mib:
1587 	snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1588 err_udplite_mib:
1589 	snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1590 err_udp_mib:
1591 	snmp_mib_free((void __percpu **)net->mib.net_statistics);
1592 err_net_mib:
1593 	snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1594 err_ip_mib:
1595 	snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1596 err_tcp_mib:
1597 	return -ENOMEM;
1598 }
1599 
1600 static __net_exit void ipv4_mib_exit_net(struct net *net)
1601 {
1602 	kfree(net->mib.icmpmsg_statistics);
1603 	snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1604 	snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1605 	snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1606 	snmp_mib_free((void __percpu **)net->mib.net_statistics);
1607 	snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1608 	snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1609 }
1610 
1611 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1612 	.init = ipv4_mib_init_net,
1613 	.exit = ipv4_mib_exit_net,
1614 };
1615 
1616 static int __init init_ipv4_mibs(void)
1617 {
1618 	return register_pernet_subsys(&ipv4_mib_ops);
1619 }
1620 
1621 static int ipv4_proc_init(void);
1622 
1623 /*
1624  *	IP protocol layer initialiser
1625  */
1626 
1627 static struct packet_type ip_packet_type __read_mostly = {
1628 	.type = cpu_to_be16(ETH_P_IP),
1629 	.func = ip_rcv,
1630 	.gso_send_check = inet_gso_send_check,
1631 	.gso_segment = inet_gso_segment,
1632 	.gro_receive = inet_gro_receive,
1633 	.gro_complete = inet_gro_complete,
1634 };
1635 
1636 static int __init inet_init(void)
1637 {
1638 	struct sk_buff *dummy_skb;
1639 	struct inet_protosw *q;
1640 	struct list_head *r;
1641 	int rc = -EINVAL;
1642 
1643 	BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1644 
1645 	sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL);
1646 	if (!sysctl_local_reserved_ports)
1647 		goto out;
1648 
1649 	rc = proto_register(&tcp_prot, 1);
1650 	if (rc)
1651 		goto out_free_reserved_ports;
1652 
1653 	rc = proto_register(&udp_prot, 1);
1654 	if (rc)
1655 		goto out_unregister_tcp_proto;
1656 
1657 	rc = proto_register(&raw_prot, 1);
1658 	if (rc)
1659 		goto out_unregister_udp_proto;
1660 
1661 	rc = proto_register(&ping_prot, 1);
1662 	if (rc)
1663 		goto out_unregister_raw_proto;
1664 
1665 	/*
1666 	 *	Tell SOCKET that we are alive...
1667 	 */
1668 
1669 	(void)sock_register(&inet_family_ops);
1670 
1671 #ifdef CONFIG_SYSCTL
1672 	ip_static_sysctl_init();
1673 #endif
1674 
1675 	tcp_prot.sysctl_mem = init_net.ipv4.sysctl_tcp_mem;
1676 
1677 	/*
1678 	 *	Add all the base protocols.
1679 	 */
1680 
1681 	if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1682 		pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1683 	if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1684 		pr_crit("%s: Cannot add UDP protocol\n", __func__);
1685 	if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1686 		pr_crit("%s: Cannot add TCP protocol\n", __func__);
1687 #ifdef CONFIG_IP_MULTICAST
1688 	if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1689 		pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1690 #endif
1691 
1692 	/* Register the socket-side information for inet_create. */
1693 	for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1694 		INIT_LIST_HEAD(r);
1695 
1696 	for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1697 		inet_register_protosw(q);
1698 
1699 	/*
1700 	 *	Set the ARP module up
1701 	 */
1702 
1703 	arp_init();
1704 
1705 	/*
1706 	 *	Set the IP module up
1707 	 */
1708 
1709 	ip_init();
1710 
1711 	tcp_v4_init();
1712 
1713 	/* Setup TCP slab cache for open requests. */
1714 	tcp_init();
1715 
1716 	/* Setup UDP memory threshold */
1717 	udp_init();
1718 
1719 	/* Add UDP-Lite (RFC 3828) */
1720 	udplite4_register();
1721 
1722 	ping_init();
1723 
1724 	/*
1725 	 *	Set the ICMP layer up
1726 	 */
1727 
1728 	if (icmp_init() < 0)
1729 		panic("Failed to create the ICMP control socket.\n");
1730 
1731 	/*
1732 	 *	Initialise the multicast router
1733 	 */
1734 #if defined(CONFIG_IP_MROUTE)
1735 	if (ip_mr_init())
1736 		pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1737 #endif
1738 	/*
1739 	 *	Initialise per-cpu ipv4 mibs
1740 	 */
1741 
1742 	if (init_ipv4_mibs())
1743 		pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1744 
1745 	ipv4_proc_init();
1746 
1747 	ipfrag_init();
1748 
1749 	dev_add_pack(&ip_packet_type);
1750 
1751 	rc = 0;
1752 out:
1753 	return rc;
1754 out_unregister_raw_proto:
1755 	proto_unregister(&raw_prot);
1756 out_unregister_udp_proto:
1757 	proto_unregister(&udp_prot);
1758 out_unregister_tcp_proto:
1759 	proto_unregister(&tcp_prot);
1760 out_free_reserved_ports:
1761 	kfree(sysctl_local_reserved_ports);
1762 	goto out;
1763 }
1764 
1765 fs_initcall(inet_init);
1766 
1767 /* ------------------------------------------------------------------------ */
1768 
1769 #ifdef CONFIG_PROC_FS
1770 static int __init ipv4_proc_init(void)
1771 {
1772 	int rc = 0;
1773 
1774 	if (raw_proc_init())
1775 		goto out_raw;
1776 	if (tcp4_proc_init())
1777 		goto out_tcp;
1778 	if (udp4_proc_init())
1779 		goto out_udp;
1780 	if (ping_proc_init())
1781 		goto out_ping;
1782 	if (ip_misc_proc_init())
1783 		goto out_misc;
1784 out:
1785 	return rc;
1786 out_misc:
1787 	ping_proc_exit();
1788 out_ping:
1789 	udp4_proc_exit();
1790 out_udp:
1791 	tcp4_proc_exit();
1792 out_tcp:
1793 	raw_proc_exit();
1794 out_raw:
1795 	rc = -ENOMEM;
1796 	goto out;
1797 }
1798 
1799 #else /* CONFIG_PROC_FS */
1800 static int __init ipv4_proc_init(void)
1801 {
1802 	return 0;
1803 }
1804 #endif /* CONFIG_PROC_FS */
1805 
1806 MODULE_ALIAS_NETPROTO(PF_INET);
1807 
1808