xref: /linux/net/ipv4/af_inet.c (revision 4413e16d9d21673bb5048a2e542f1aaa00015c2e)
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 
95 #include <linux/inet.h>
96 #include <linux/igmp.h>
97 #include <linux/inetdevice.h>
98 #include <linux/netdevice.h>
99 #include <net/checksum.h>
100 #include <net/ip.h>
101 #include <net/protocol.h>
102 #include <net/arp.h>
103 #include <net/route.h>
104 #include <net/ip_fib.h>
105 #include <net/inet_connection_sock.h>
106 #include <net/tcp.h>
107 #include <net/udp.h>
108 #include <net/udplite.h>
109 #include <net/ping.h>
110 #include <linux/skbuff.h>
111 #include <net/sock.h>
112 #include <net/raw.h>
113 #include <net/icmp.h>
114 #include <net/ipip.h>
115 #include <net/inet_common.h>
116 #include <net/xfrm.h>
117 #include <net/net_namespace.h>
118 #ifdef CONFIG_IP_MROUTE
119 #include <linux/mroute.h>
120 #endif
121 
122 
123 /* The inetsw table contains everything that inet_create needs to
124  * build a new socket.
125  */
126 static struct list_head inetsw[SOCK_MAX];
127 static DEFINE_SPINLOCK(inetsw_lock);
128 
129 struct ipv4_config ipv4_config;
130 EXPORT_SYMBOL(ipv4_config);
131 
132 /* New destruction routine */
133 
134 void inet_sock_destruct(struct sock *sk)
135 {
136 	struct inet_sock *inet = inet_sk(sk);
137 
138 	__skb_queue_purge(&sk->sk_receive_queue);
139 	__skb_queue_purge(&sk->sk_error_queue);
140 
141 	sk_mem_reclaim(sk);
142 
143 	if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
144 		pr_err("Attempt to release TCP socket in state %d %p\n",
145 		       sk->sk_state, sk);
146 		return;
147 	}
148 	if (!sock_flag(sk, SOCK_DEAD)) {
149 		pr_err("Attempt to release alive inet socket %p\n", sk);
150 		return;
151 	}
152 
153 	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
154 	WARN_ON(atomic_read(&sk->sk_wmem_alloc));
155 	WARN_ON(sk->sk_wmem_queued);
156 	WARN_ON(sk->sk_forward_alloc);
157 
158 	kfree(rcu_dereference_protected(inet->inet_opt, 1));
159 	dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
160 	dst_release(sk->sk_rx_dst);
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, __u8 protocol)
247 {
248 	const struct net_protocol *ipprot;
249 
250 	if (net_eq(net, &init_net))
251 		return 1;
252 
253 	ipprot = rcu_dereference(inet_protos[protocol]);
254 	if (ipprot == NULL) {
255 		/* raw IP is OK */
256 		return 1;
257 	}
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 = SK_CAN_REUSE;
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, uaddr, addr_len);
556 }
557 EXPORT_SYMBOL(inet_dgram_connect);
558 
559 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
560 {
561 	DEFINE_WAIT(wait);
562 
563 	prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
564 	sk->sk_write_pending += writebias;
565 
566 	/* Basic assumption: if someone sets sk->sk_err, he _must_
567 	 * change state of the socket from TCP_SYN_*.
568 	 * Connect() does not allow to get error notifications
569 	 * without closing the socket.
570 	 */
571 	while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
572 		release_sock(sk);
573 		timeo = schedule_timeout(timeo);
574 		lock_sock(sk);
575 		if (signal_pending(current) || !timeo)
576 			break;
577 		prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
578 	}
579 	finish_wait(sk_sleep(sk), &wait);
580 	sk->sk_write_pending -= writebias;
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 	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 		int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
638 				tcp_sk(sk)->fastopen_req &&
639 				tcp_sk(sk)->fastopen_req->data ? 1 : 0;
640 
641 		/* Error code is set above */
642 		if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
643 			goto out;
644 
645 		err = sock_intr_errno(timeo);
646 		if (signal_pending(current))
647 			goto out;
648 	}
649 
650 	/* Connection was closed by RST, timeout, ICMP error
651 	 * or another process disconnected us.
652 	 */
653 	if (sk->sk_state == TCP_CLOSE)
654 		goto sock_error;
655 
656 	/* sk->sk_err may be not zero now, if RECVERR was ordered by user
657 	 * and error was received after socket entered established state.
658 	 * Hence, it is handled normally after connect() return successfully.
659 	 */
660 
661 	sock->state = SS_CONNECTED;
662 	err = 0;
663 out:
664 	return err;
665 
666 sock_error:
667 	err = sock_error(sk) ? : -ECONNABORTED;
668 	sock->state = SS_UNCONNECTED;
669 	if (sk->sk_prot->disconnect(sk, flags))
670 		sock->state = SS_DISCONNECTING;
671 	goto out;
672 }
673 EXPORT_SYMBOL(__inet_stream_connect);
674 
675 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
676 			int addr_len, int flags)
677 {
678 	int err;
679 
680 	lock_sock(sock->sk);
681 	err = __inet_stream_connect(sock, uaddr, addr_len, flags);
682 	release_sock(sock->sk);
683 	return err;
684 }
685 EXPORT_SYMBOL(inet_stream_connect);
686 
687 /*
688  *	Accept a pending connection. The TCP layer now gives BSD semantics.
689  */
690 
691 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
692 {
693 	struct sock *sk1 = sock->sk;
694 	int err = -EINVAL;
695 	struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
696 
697 	if (!sk2)
698 		goto do_err;
699 
700 	lock_sock(sk2);
701 
702 	sock_rps_record_flow(sk2);
703 	WARN_ON(!((1 << sk2->sk_state) &
704 		  (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE)));
705 
706 	sock_graft(sk2, newsock);
707 
708 	newsock->state = SS_CONNECTED;
709 	err = 0;
710 	release_sock(sk2);
711 do_err:
712 	return err;
713 }
714 EXPORT_SYMBOL(inet_accept);
715 
716 
717 /*
718  *	This does both peername and sockname.
719  */
720 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
721 			int *uaddr_len, int peer)
722 {
723 	struct sock *sk		= sock->sk;
724 	struct inet_sock *inet	= inet_sk(sk);
725 	DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
726 
727 	sin->sin_family = AF_INET;
728 	if (peer) {
729 		if (!inet->inet_dport ||
730 		    (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
731 		     peer == 1))
732 			return -ENOTCONN;
733 		sin->sin_port = inet->inet_dport;
734 		sin->sin_addr.s_addr = inet->inet_daddr;
735 	} else {
736 		__be32 addr = inet->inet_rcv_saddr;
737 		if (!addr)
738 			addr = inet->inet_saddr;
739 		sin->sin_port = inet->inet_sport;
740 		sin->sin_addr.s_addr = addr;
741 	}
742 	memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
743 	*uaddr_len = sizeof(*sin);
744 	return 0;
745 }
746 EXPORT_SYMBOL(inet_getname);
747 
748 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
749 		 size_t size)
750 {
751 	struct sock *sk = sock->sk;
752 
753 	sock_rps_record_flow(sk);
754 
755 	/* We may need to bind the socket. */
756 	if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
757 	    inet_autobind(sk))
758 		return -EAGAIN;
759 
760 	return sk->sk_prot->sendmsg(iocb, sk, msg, size);
761 }
762 EXPORT_SYMBOL(inet_sendmsg);
763 
764 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
765 		      size_t size, int flags)
766 {
767 	struct sock *sk = sock->sk;
768 
769 	sock_rps_record_flow(sk);
770 
771 	/* We may need to bind the socket. */
772 	if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
773 	    inet_autobind(sk))
774 		return -EAGAIN;
775 
776 	if (sk->sk_prot->sendpage)
777 		return sk->sk_prot->sendpage(sk, page, offset, size, flags);
778 	return sock_no_sendpage(sock, page, offset, size, flags);
779 }
780 EXPORT_SYMBOL(inet_sendpage);
781 
782 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
783 		 size_t size, int flags)
784 {
785 	struct sock *sk = sock->sk;
786 	int addr_len = 0;
787 	int err;
788 
789 	sock_rps_record_flow(sk);
790 
791 	err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
792 				   flags & ~MSG_DONTWAIT, &addr_len);
793 	if (err >= 0)
794 		msg->msg_namelen = addr_len;
795 	return err;
796 }
797 EXPORT_SYMBOL(inet_recvmsg);
798 
799 int inet_shutdown(struct socket *sock, int how)
800 {
801 	struct sock *sk = sock->sk;
802 	int err = 0;
803 
804 	/* This should really check to make sure
805 	 * the socket is a TCP socket. (WHY AC...)
806 	 */
807 	how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
808 		       1->2 bit 2 snds.
809 		       2->3 */
810 	if ((how & ~SHUTDOWN_MASK) || !how)	/* MAXINT->0 */
811 		return -EINVAL;
812 
813 	lock_sock(sk);
814 	if (sock->state == SS_CONNECTING) {
815 		if ((1 << sk->sk_state) &
816 		    (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
817 			sock->state = SS_DISCONNECTING;
818 		else
819 			sock->state = SS_CONNECTED;
820 	}
821 
822 	switch (sk->sk_state) {
823 	case TCP_CLOSE:
824 		err = -ENOTCONN;
825 		/* Hack to wake up other listeners, who can poll for
826 		   POLLHUP, even on eg. unconnected UDP sockets -- RR */
827 	default:
828 		sk->sk_shutdown |= how;
829 		if (sk->sk_prot->shutdown)
830 			sk->sk_prot->shutdown(sk, how);
831 		break;
832 
833 	/* Remaining two branches are temporary solution for missing
834 	 * close() in multithreaded environment. It is _not_ a good idea,
835 	 * but we have no choice until close() is repaired at VFS level.
836 	 */
837 	case TCP_LISTEN:
838 		if (!(how & RCV_SHUTDOWN))
839 			break;
840 		/* Fall through */
841 	case TCP_SYN_SENT:
842 		err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
843 		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
844 		break;
845 	}
846 
847 	/* Wake up anyone sleeping in poll. */
848 	sk->sk_state_change(sk);
849 	release_sock(sk);
850 	return err;
851 }
852 EXPORT_SYMBOL(inet_shutdown);
853 
854 /*
855  *	ioctl() calls you can issue on an INET socket. Most of these are
856  *	device configuration and stuff and very rarely used. Some ioctls
857  *	pass on to the socket itself.
858  *
859  *	NOTE: I like the idea of a module for the config stuff. ie ifconfig
860  *	loads the devconfigure module does its configuring and unloads it.
861  *	There's a good 20K of config code hanging around the kernel.
862  */
863 
864 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
865 {
866 	struct sock *sk = sock->sk;
867 	int err = 0;
868 	struct net *net = sock_net(sk);
869 
870 	switch (cmd) {
871 	case SIOCGSTAMP:
872 		err = sock_get_timestamp(sk, (struct timeval __user *)arg);
873 		break;
874 	case SIOCGSTAMPNS:
875 		err = sock_get_timestampns(sk, (struct timespec __user *)arg);
876 		break;
877 	case SIOCADDRT:
878 	case SIOCDELRT:
879 	case SIOCRTMSG:
880 		err = ip_rt_ioctl(net, cmd, (void __user *)arg);
881 		break;
882 	case SIOCDARP:
883 	case SIOCGARP:
884 	case SIOCSARP:
885 		err = arp_ioctl(net, cmd, (void __user *)arg);
886 		break;
887 	case SIOCGIFADDR:
888 	case SIOCSIFADDR:
889 	case SIOCGIFBRDADDR:
890 	case SIOCSIFBRDADDR:
891 	case SIOCGIFNETMASK:
892 	case SIOCSIFNETMASK:
893 	case SIOCGIFDSTADDR:
894 	case SIOCSIFDSTADDR:
895 	case SIOCSIFPFLAGS:
896 	case SIOCGIFPFLAGS:
897 	case SIOCSIFFLAGS:
898 		err = devinet_ioctl(net, cmd, (void __user *)arg);
899 		break;
900 	default:
901 		if (sk->sk_prot->ioctl)
902 			err = sk->sk_prot->ioctl(sk, cmd, arg);
903 		else
904 			err = -ENOIOCTLCMD;
905 		break;
906 	}
907 	return err;
908 }
909 EXPORT_SYMBOL(inet_ioctl);
910 
911 #ifdef CONFIG_COMPAT
912 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
913 {
914 	struct sock *sk = sock->sk;
915 	int err = -ENOIOCTLCMD;
916 
917 	if (sk->sk_prot->compat_ioctl)
918 		err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
919 
920 	return err;
921 }
922 #endif
923 
924 const struct proto_ops inet_stream_ops = {
925 	.family		   = PF_INET,
926 	.owner		   = THIS_MODULE,
927 	.release	   = inet_release,
928 	.bind		   = inet_bind,
929 	.connect	   = inet_stream_connect,
930 	.socketpair	   = sock_no_socketpair,
931 	.accept		   = inet_accept,
932 	.getname	   = inet_getname,
933 	.poll		   = tcp_poll,
934 	.ioctl		   = inet_ioctl,
935 	.listen		   = inet_listen,
936 	.shutdown	   = inet_shutdown,
937 	.setsockopt	   = sock_common_setsockopt,
938 	.getsockopt	   = sock_common_getsockopt,
939 	.sendmsg	   = inet_sendmsg,
940 	.recvmsg	   = inet_recvmsg,
941 	.mmap		   = sock_no_mmap,
942 	.sendpage	   = inet_sendpage,
943 	.splice_read	   = tcp_splice_read,
944 #ifdef CONFIG_COMPAT
945 	.compat_setsockopt = compat_sock_common_setsockopt,
946 	.compat_getsockopt = compat_sock_common_getsockopt,
947 	.compat_ioctl	   = inet_compat_ioctl,
948 #endif
949 };
950 EXPORT_SYMBOL(inet_stream_ops);
951 
952 const struct proto_ops inet_dgram_ops = {
953 	.family		   = PF_INET,
954 	.owner		   = THIS_MODULE,
955 	.release	   = inet_release,
956 	.bind		   = inet_bind,
957 	.connect	   = inet_dgram_connect,
958 	.socketpair	   = sock_no_socketpair,
959 	.accept		   = sock_no_accept,
960 	.getname	   = inet_getname,
961 	.poll		   = udp_poll,
962 	.ioctl		   = inet_ioctl,
963 	.listen		   = sock_no_listen,
964 	.shutdown	   = inet_shutdown,
965 	.setsockopt	   = sock_common_setsockopt,
966 	.getsockopt	   = sock_common_getsockopt,
967 	.sendmsg	   = inet_sendmsg,
968 	.recvmsg	   = inet_recvmsg,
969 	.mmap		   = sock_no_mmap,
970 	.sendpage	   = inet_sendpage,
971 #ifdef CONFIG_COMPAT
972 	.compat_setsockopt = compat_sock_common_setsockopt,
973 	.compat_getsockopt = compat_sock_common_getsockopt,
974 	.compat_ioctl	   = inet_compat_ioctl,
975 #endif
976 };
977 EXPORT_SYMBOL(inet_dgram_ops);
978 
979 /*
980  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
981  * udp_poll
982  */
983 static const struct proto_ops inet_sockraw_ops = {
984 	.family		   = PF_INET,
985 	.owner		   = THIS_MODULE,
986 	.release	   = inet_release,
987 	.bind		   = inet_bind,
988 	.connect	   = inet_dgram_connect,
989 	.socketpair	   = sock_no_socketpair,
990 	.accept		   = sock_no_accept,
991 	.getname	   = inet_getname,
992 	.poll		   = datagram_poll,
993 	.ioctl		   = inet_ioctl,
994 	.listen		   = sock_no_listen,
995 	.shutdown	   = inet_shutdown,
996 	.setsockopt	   = sock_common_setsockopt,
997 	.getsockopt	   = sock_common_getsockopt,
998 	.sendmsg	   = inet_sendmsg,
999 	.recvmsg	   = inet_recvmsg,
1000 	.mmap		   = sock_no_mmap,
1001 	.sendpage	   = inet_sendpage,
1002 #ifdef CONFIG_COMPAT
1003 	.compat_setsockopt = compat_sock_common_setsockopt,
1004 	.compat_getsockopt = compat_sock_common_getsockopt,
1005 	.compat_ioctl	   = inet_compat_ioctl,
1006 #endif
1007 };
1008 
1009 static const struct net_proto_family inet_family_ops = {
1010 	.family = PF_INET,
1011 	.create = inet_create,
1012 	.owner	= THIS_MODULE,
1013 };
1014 
1015 /* Upon startup we insert all the elements in inetsw_array[] into
1016  * the linked list inetsw.
1017  */
1018 static struct inet_protosw inetsw_array[] =
1019 {
1020 	{
1021 		.type =       SOCK_STREAM,
1022 		.protocol =   IPPROTO_TCP,
1023 		.prot =       &tcp_prot,
1024 		.ops =        &inet_stream_ops,
1025 		.no_check =   0,
1026 		.flags =      INET_PROTOSW_PERMANENT |
1027 			      INET_PROTOSW_ICSK,
1028 	},
1029 
1030 	{
1031 		.type =       SOCK_DGRAM,
1032 		.protocol =   IPPROTO_UDP,
1033 		.prot =       &udp_prot,
1034 		.ops =        &inet_dgram_ops,
1035 		.no_check =   UDP_CSUM_DEFAULT,
1036 		.flags =      INET_PROTOSW_PERMANENT,
1037        },
1038 
1039        {
1040 		.type =       SOCK_DGRAM,
1041 		.protocol =   IPPROTO_ICMP,
1042 		.prot =       &ping_prot,
1043 		.ops =        &inet_dgram_ops,
1044 		.no_check =   UDP_CSUM_DEFAULT,
1045 		.flags =      INET_PROTOSW_REUSE,
1046        },
1047 
1048        {
1049 	       .type =       SOCK_RAW,
1050 	       .protocol =   IPPROTO_IP,	/* wild card */
1051 	       .prot =       &raw_prot,
1052 	       .ops =        &inet_sockraw_ops,
1053 	       .no_check =   UDP_CSUM_DEFAULT,
1054 	       .flags =      INET_PROTOSW_REUSE,
1055        }
1056 };
1057 
1058 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1059 
1060 void inet_register_protosw(struct inet_protosw *p)
1061 {
1062 	struct list_head *lh;
1063 	struct inet_protosw *answer;
1064 	int protocol = p->protocol;
1065 	struct list_head *last_perm;
1066 
1067 	spin_lock_bh(&inetsw_lock);
1068 
1069 	if (p->type >= SOCK_MAX)
1070 		goto out_illegal;
1071 
1072 	/* If we are trying to override a permanent protocol, bail. */
1073 	answer = NULL;
1074 	last_perm = &inetsw[p->type];
1075 	list_for_each(lh, &inetsw[p->type]) {
1076 		answer = list_entry(lh, struct inet_protosw, list);
1077 
1078 		/* Check only the non-wild match. */
1079 		if (INET_PROTOSW_PERMANENT & answer->flags) {
1080 			if (protocol == answer->protocol)
1081 				break;
1082 			last_perm = lh;
1083 		}
1084 
1085 		answer = NULL;
1086 	}
1087 	if (answer)
1088 		goto out_permanent;
1089 
1090 	/* Add the new entry after the last permanent entry if any, so that
1091 	 * the new entry does not override a permanent entry when matched with
1092 	 * a wild-card protocol. But it is allowed to override any existing
1093 	 * non-permanent entry.  This means that when we remove this entry, the
1094 	 * system automatically returns to the old behavior.
1095 	 */
1096 	list_add_rcu(&p->list, last_perm);
1097 out:
1098 	spin_unlock_bh(&inetsw_lock);
1099 
1100 	return;
1101 
1102 out_permanent:
1103 	pr_err("Attempt to override permanent protocol %d\n", protocol);
1104 	goto out;
1105 
1106 out_illegal:
1107 	pr_err("Ignoring attempt to register invalid socket type %d\n",
1108 	       p->type);
1109 	goto out;
1110 }
1111 EXPORT_SYMBOL(inet_register_protosw);
1112 
1113 void inet_unregister_protosw(struct inet_protosw *p)
1114 {
1115 	if (INET_PROTOSW_PERMANENT & p->flags) {
1116 		pr_err("Attempt to unregister permanent protocol %d\n",
1117 		       p->protocol);
1118 	} else {
1119 		spin_lock_bh(&inetsw_lock);
1120 		list_del_rcu(&p->list);
1121 		spin_unlock_bh(&inetsw_lock);
1122 
1123 		synchronize_net();
1124 	}
1125 }
1126 EXPORT_SYMBOL(inet_unregister_protosw);
1127 
1128 /*
1129  *      Shall we try to damage output packets if routing dev changes?
1130  */
1131 
1132 int sysctl_ip_dynaddr __read_mostly;
1133 
1134 static int inet_sk_reselect_saddr(struct sock *sk)
1135 {
1136 	struct inet_sock *inet = inet_sk(sk);
1137 	__be32 old_saddr = inet->inet_saddr;
1138 	__be32 daddr = inet->inet_daddr;
1139 	struct flowi4 *fl4;
1140 	struct rtable *rt;
1141 	__be32 new_saddr;
1142 	struct ip_options_rcu *inet_opt;
1143 
1144 	inet_opt = rcu_dereference_protected(inet->inet_opt,
1145 					     sock_owned_by_user(sk));
1146 	if (inet_opt && inet_opt->opt.srr)
1147 		daddr = inet_opt->opt.faddr;
1148 
1149 	/* Query new route. */
1150 	fl4 = &inet->cork.fl.u.ip4;
1151 	rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1152 			      sk->sk_bound_dev_if, sk->sk_protocol,
1153 			      inet->inet_sport, inet->inet_dport, sk, false);
1154 	if (IS_ERR(rt))
1155 		return PTR_ERR(rt);
1156 
1157 	sk_setup_caps(sk, &rt->dst);
1158 
1159 	new_saddr = fl4->saddr;
1160 
1161 	if (new_saddr == old_saddr)
1162 		return 0;
1163 
1164 	if (sysctl_ip_dynaddr > 1) {
1165 		pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1166 			__func__, &old_saddr, &new_saddr);
1167 	}
1168 
1169 	inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1170 
1171 	/*
1172 	 * XXX The only one ugly spot where we need to
1173 	 * XXX really change the sockets identity after
1174 	 * XXX it has entered the hashes. -DaveM
1175 	 *
1176 	 * Besides that, it does not check for connection
1177 	 * uniqueness. Wait for troubles.
1178 	 */
1179 	__sk_prot_rehash(sk);
1180 	return 0;
1181 }
1182 
1183 int inet_sk_rebuild_header(struct sock *sk)
1184 {
1185 	struct inet_sock *inet = inet_sk(sk);
1186 	struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1187 	__be32 daddr;
1188 	struct ip_options_rcu *inet_opt;
1189 	struct flowi4 *fl4;
1190 	int err;
1191 
1192 	/* Route is OK, nothing to do. */
1193 	if (rt)
1194 		return 0;
1195 
1196 	/* Reroute. */
1197 	rcu_read_lock();
1198 	inet_opt = rcu_dereference(inet->inet_opt);
1199 	daddr = inet->inet_daddr;
1200 	if (inet_opt && inet_opt->opt.srr)
1201 		daddr = inet_opt->opt.faddr;
1202 	rcu_read_unlock();
1203 	fl4 = &inet->cork.fl.u.ip4;
1204 	rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1205 				   inet->inet_dport, inet->inet_sport,
1206 				   sk->sk_protocol, RT_CONN_FLAGS(sk),
1207 				   sk->sk_bound_dev_if);
1208 	if (!IS_ERR(rt)) {
1209 		err = 0;
1210 		sk_setup_caps(sk, &rt->dst);
1211 	} else {
1212 		err = PTR_ERR(rt);
1213 
1214 		/* Routing failed... */
1215 		sk->sk_route_caps = 0;
1216 		/*
1217 		 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1218 		 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1219 		 */
1220 		if (!sysctl_ip_dynaddr ||
1221 		    sk->sk_state != TCP_SYN_SENT ||
1222 		    (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1223 		    (err = inet_sk_reselect_saddr(sk)) != 0)
1224 			sk->sk_err_soft = -err;
1225 	}
1226 
1227 	return err;
1228 }
1229 EXPORT_SYMBOL(inet_sk_rebuild_header);
1230 
1231 static int inet_gso_send_check(struct sk_buff *skb)
1232 {
1233 	const struct net_protocol *ops;
1234 	const struct iphdr *iph;
1235 	int proto;
1236 	int ihl;
1237 	int err = -EINVAL;
1238 
1239 	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1240 		goto out;
1241 
1242 	iph = ip_hdr(skb);
1243 	ihl = iph->ihl * 4;
1244 	if (ihl < sizeof(*iph))
1245 		goto out;
1246 
1247 	if (unlikely(!pskb_may_pull(skb, ihl)))
1248 		goto out;
1249 
1250 	__skb_pull(skb, ihl);
1251 	skb_reset_transport_header(skb);
1252 	iph = ip_hdr(skb);
1253 	proto = iph->protocol;
1254 	err = -EPROTONOSUPPORT;
1255 
1256 	rcu_read_lock();
1257 	ops = rcu_dereference(inet_protos[proto]);
1258 	if (likely(ops && ops->gso_send_check))
1259 		err = ops->gso_send_check(skb);
1260 	rcu_read_unlock();
1261 
1262 out:
1263 	return err;
1264 }
1265 
1266 static struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1267 	netdev_features_t features)
1268 {
1269 	struct sk_buff *segs = ERR_PTR(-EINVAL);
1270 	const struct net_protocol *ops;
1271 	struct iphdr *iph;
1272 	int proto;
1273 	int ihl;
1274 	int id;
1275 	unsigned int offset = 0;
1276 
1277 	if (!(features & NETIF_F_V4_CSUM))
1278 		features &= ~NETIF_F_SG;
1279 
1280 	if (unlikely(skb_shinfo(skb)->gso_type &
1281 		     ~(SKB_GSO_TCPV4 |
1282 		       SKB_GSO_UDP |
1283 		       SKB_GSO_DODGY |
1284 		       SKB_GSO_TCP_ECN |
1285 		       0)))
1286 		goto out;
1287 
1288 	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1289 		goto out;
1290 
1291 	iph = ip_hdr(skb);
1292 	ihl = iph->ihl * 4;
1293 	if (ihl < sizeof(*iph))
1294 		goto out;
1295 
1296 	if (unlikely(!pskb_may_pull(skb, ihl)))
1297 		goto out;
1298 
1299 	__skb_pull(skb, ihl);
1300 	skb_reset_transport_header(skb);
1301 	iph = ip_hdr(skb);
1302 	id = ntohs(iph->id);
1303 	proto = iph->protocol;
1304 	segs = ERR_PTR(-EPROTONOSUPPORT);
1305 
1306 	rcu_read_lock();
1307 	ops = rcu_dereference(inet_protos[proto]);
1308 	if (likely(ops && ops->gso_segment))
1309 		segs = ops->gso_segment(skb, features);
1310 	rcu_read_unlock();
1311 
1312 	if (!segs || IS_ERR(segs))
1313 		goto out;
1314 
1315 	skb = segs;
1316 	do {
1317 		iph = ip_hdr(skb);
1318 		if (proto == IPPROTO_UDP) {
1319 			iph->id = htons(id);
1320 			iph->frag_off = htons(offset >> 3);
1321 			if (skb->next != NULL)
1322 				iph->frag_off |= htons(IP_MF);
1323 			offset += (skb->len - skb->mac_len - iph->ihl * 4);
1324 		} else
1325 			iph->id = htons(id++);
1326 		iph->tot_len = htons(skb->len - skb->mac_len);
1327 		iph->check = 0;
1328 		iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1329 	} while ((skb = skb->next));
1330 
1331 out:
1332 	return segs;
1333 }
1334 
1335 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1336 					 struct sk_buff *skb)
1337 {
1338 	const struct net_protocol *ops;
1339 	struct sk_buff **pp = NULL;
1340 	struct sk_buff *p;
1341 	const struct iphdr *iph;
1342 	unsigned int hlen;
1343 	unsigned int off;
1344 	unsigned int id;
1345 	int flush = 1;
1346 	int proto;
1347 
1348 	off = skb_gro_offset(skb);
1349 	hlen = off + sizeof(*iph);
1350 	iph = skb_gro_header_fast(skb, off);
1351 	if (skb_gro_header_hard(skb, hlen)) {
1352 		iph = skb_gro_header_slow(skb, hlen, off);
1353 		if (unlikely(!iph))
1354 			goto out;
1355 	}
1356 
1357 	proto = iph->protocol;
1358 
1359 	rcu_read_lock();
1360 	ops = rcu_dereference(inet_protos[proto]);
1361 	if (!ops || !ops->gro_receive)
1362 		goto out_unlock;
1363 
1364 	if (*(u8 *)iph != 0x45)
1365 		goto out_unlock;
1366 
1367 	if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1368 		goto out_unlock;
1369 
1370 	id = ntohl(*(__be32 *)&iph->id);
1371 	flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF));
1372 	id >>= 16;
1373 
1374 	for (p = *head; p; p = p->next) {
1375 		struct iphdr *iph2;
1376 
1377 		if (!NAPI_GRO_CB(p)->same_flow)
1378 			continue;
1379 
1380 		iph2 = ip_hdr(p);
1381 
1382 		if ((iph->protocol ^ iph2->protocol) |
1383 		    (iph->tos ^ iph2->tos) |
1384 		    ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1385 		    ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1386 			NAPI_GRO_CB(p)->same_flow = 0;
1387 			continue;
1388 		}
1389 
1390 		/* All fields must match except length and checksum. */
1391 		NAPI_GRO_CB(p)->flush |=
1392 			(iph->ttl ^ iph2->ttl) |
1393 			((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1394 
1395 		NAPI_GRO_CB(p)->flush |= flush;
1396 	}
1397 
1398 	NAPI_GRO_CB(skb)->flush |= flush;
1399 	skb_gro_pull(skb, sizeof(*iph));
1400 	skb_set_transport_header(skb, skb_gro_offset(skb));
1401 
1402 	pp = ops->gro_receive(head, skb);
1403 
1404 out_unlock:
1405 	rcu_read_unlock();
1406 
1407 out:
1408 	NAPI_GRO_CB(skb)->flush |= flush;
1409 
1410 	return pp;
1411 }
1412 
1413 static int inet_gro_complete(struct sk_buff *skb)
1414 {
1415 	__be16 newlen = htons(skb->len - skb_network_offset(skb));
1416 	struct iphdr *iph = ip_hdr(skb);
1417 	const struct net_protocol *ops;
1418 	int proto = iph->protocol;
1419 	int err = -ENOSYS;
1420 
1421 	csum_replace2(&iph->check, iph->tot_len, newlen);
1422 	iph->tot_len = newlen;
1423 
1424 	rcu_read_lock();
1425 	ops = rcu_dereference(inet_protos[proto]);
1426 	if (WARN_ON(!ops || !ops->gro_complete))
1427 		goto out_unlock;
1428 
1429 	err = ops->gro_complete(skb);
1430 
1431 out_unlock:
1432 	rcu_read_unlock();
1433 
1434 	return err;
1435 }
1436 
1437 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1438 			 unsigned short type, unsigned char protocol,
1439 			 struct net *net)
1440 {
1441 	struct socket *sock;
1442 	int rc = sock_create_kern(family, type, protocol, &sock);
1443 
1444 	if (rc == 0) {
1445 		*sk = sock->sk;
1446 		(*sk)->sk_allocation = GFP_ATOMIC;
1447 		/*
1448 		 * Unhash it so that IP input processing does not even see it,
1449 		 * we do not wish this socket to see incoming packets.
1450 		 */
1451 		(*sk)->sk_prot->unhash(*sk);
1452 
1453 		sk_change_net(*sk, net);
1454 	}
1455 	return rc;
1456 }
1457 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1458 
1459 unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1460 {
1461 	unsigned long res = 0;
1462 	int i, j;
1463 
1464 	for_each_possible_cpu(i) {
1465 		for (j = 0; j < SNMP_ARRAY_SZ; j++)
1466 			res += *(((unsigned long *) per_cpu_ptr(mib[j], i)) + offt);
1467 	}
1468 	return res;
1469 }
1470 EXPORT_SYMBOL_GPL(snmp_fold_field);
1471 
1472 #if BITS_PER_LONG==32
1473 
1474 u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset)
1475 {
1476 	u64 res = 0;
1477 	int cpu;
1478 
1479 	for_each_possible_cpu(cpu) {
1480 		void *bhptr;
1481 		struct u64_stats_sync *syncp;
1482 		u64 v;
1483 		unsigned int start;
1484 
1485 		bhptr = per_cpu_ptr(mib[0], cpu);
1486 		syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1487 		do {
1488 			start = u64_stats_fetch_begin_bh(syncp);
1489 			v = *(((u64 *) bhptr) + offt);
1490 		} while (u64_stats_fetch_retry_bh(syncp, start));
1491 
1492 		res += v;
1493 	}
1494 	return res;
1495 }
1496 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1497 #endif
1498 
1499 int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align)
1500 {
1501 	BUG_ON(ptr == NULL);
1502 	ptr[0] = __alloc_percpu(mibsize, align);
1503 	if (!ptr[0])
1504 		return -ENOMEM;
1505 #if SNMP_ARRAY_SZ == 2
1506 	ptr[1] = __alloc_percpu(mibsize, align);
1507 	if (!ptr[1]) {
1508 		free_percpu(ptr[0]);
1509 		ptr[0] = NULL;
1510 		return -ENOMEM;
1511 	}
1512 #endif
1513 	return 0;
1514 }
1515 EXPORT_SYMBOL_GPL(snmp_mib_init);
1516 
1517 void snmp_mib_free(void __percpu *ptr[SNMP_ARRAY_SZ])
1518 {
1519 	int i;
1520 
1521 	BUG_ON(ptr == NULL);
1522 	for (i = 0; i < SNMP_ARRAY_SZ; i++) {
1523 		free_percpu(ptr[i]);
1524 		ptr[i] = NULL;
1525 	}
1526 }
1527 EXPORT_SYMBOL_GPL(snmp_mib_free);
1528 
1529 #ifdef CONFIG_IP_MULTICAST
1530 static const struct net_protocol igmp_protocol = {
1531 	.handler =	igmp_rcv,
1532 	.netns_ok =	1,
1533 };
1534 #endif
1535 
1536 static const struct net_protocol tcp_protocol = {
1537 	.early_demux	=	tcp_v4_early_demux,
1538 	.handler	=	tcp_v4_rcv,
1539 	.err_handler	=	tcp_v4_err,
1540 	.gso_send_check	=	tcp_v4_gso_send_check,
1541 	.gso_segment	=	tcp_tso_segment,
1542 	.gro_receive	=	tcp4_gro_receive,
1543 	.gro_complete	=	tcp4_gro_complete,
1544 	.no_policy	=	1,
1545 	.netns_ok	=	1,
1546 };
1547 
1548 static const struct net_protocol udp_protocol = {
1549 	.handler =	udp_rcv,
1550 	.err_handler =	udp_err,
1551 	.gso_send_check = udp4_ufo_send_check,
1552 	.gso_segment = udp4_ufo_fragment,
1553 	.no_policy =	1,
1554 	.netns_ok =	1,
1555 };
1556 
1557 static const struct net_protocol icmp_protocol = {
1558 	.handler =	icmp_rcv,
1559 	.err_handler =	ping_err,
1560 	.no_policy =	1,
1561 	.netns_ok =	1,
1562 };
1563 
1564 static __net_init int ipv4_mib_init_net(struct net *net)
1565 {
1566 	if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1567 			  sizeof(struct tcp_mib),
1568 			  __alignof__(struct tcp_mib)) < 0)
1569 		goto err_tcp_mib;
1570 	if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1571 			  sizeof(struct ipstats_mib),
1572 			  __alignof__(struct ipstats_mib)) < 0)
1573 		goto err_ip_mib;
1574 	if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1575 			  sizeof(struct linux_mib),
1576 			  __alignof__(struct linux_mib)) < 0)
1577 		goto err_net_mib;
1578 	if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1579 			  sizeof(struct udp_mib),
1580 			  __alignof__(struct udp_mib)) < 0)
1581 		goto err_udp_mib;
1582 	if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1583 			  sizeof(struct udp_mib),
1584 			  __alignof__(struct udp_mib)) < 0)
1585 		goto err_udplite_mib;
1586 	if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1587 			  sizeof(struct icmp_mib),
1588 			  __alignof__(struct icmp_mib)) < 0)
1589 		goto err_icmp_mib;
1590 	net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1591 					      GFP_KERNEL);
1592 	if (!net->mib.icmpmsg_statistics)
1593 		goto err_icmpmsg_mib;
1594 
1595 	tcp_mib_init(net);
1596 	return 0;
1597 
1598 err_icmpmsg_mib:
1599 	snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1600 err_icmp_mib:
1601 	snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1602 err_udplite_mib:
1603 	snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1604 err_udp_mib:
1605 	snmp_mib_free((void __percpu **)net->mib.net_statistics);
1606 err_net_mib:
1607 	snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1608 err_ip_mib:
1609 	snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1610 err_tcp_mib:
1611 	return -ENOMEM;
1612 }
1613 
1614 static __net_exit void ipv4_mib_exit_net(struct net *net)
1615 {
1616 	kfree(net->mib.icmpmsg_statistics);
1617 	snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1618 	snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1619 	snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1620 	snmp_mib_free((void __percpu **)net->mib.net_statistics);
1621 	snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1622 	snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1623 }
1624 
1625 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1626 	.init = ipv4_mib_init_net,
1627 	.exit = ipv4_mib_exit_net,
1628 };
1629 
1630 static int __init init_ipv4_mibs(void)
1631 {
1632 	return register_pernet_subsys(&ipv4_mib_ops);
1633 }
1634 
1635 static int ipv4_proc_init(void);
1636 
1637 /*
1638  *	IP protocol layer initialiser
1639  */
1640 
1641 static struct packet_type ip_packet_type __read_mostly = {
1642 	.type = cpu_to_be16(ETH_P_IP),
1643 	.func = ip_rcv,
1644 	.gso_send_check = inet_gso_send_check,
1645 	.gso_segment = inet_gso_segment,
1646 	.gro_receive = inet_gro_receive,
1647 	.gro_complete = inet_gro_complete,
1648 };
1649 
1650 static int __init inet_init(void)
1651 {
1652 	struct sk_buff *dummy_skb;
1653 	struct inet_protosw *q;
1654 	struct list_head *r;
1655 	int rc = -EINVAL;
1656 
1657 	BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1658 
1659 	sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL);
1660 	if (!sysctl_local_reserved_ports)
1661 		goto out;
1662 
1663 	rc = proto_register(&tcp_prot, 1);
1664 	if (rc)
1665 		goto out_free_reserved_ports;
1666 
1667 	rc = proto_register(&udp_prot, 1);
1668 	if (rc)
1669 		goto out_unregister_tcp_proto;
1670 
1671 	rc = proto_register(&raw_prot, 1);
1672 	if (rc)
1673 		goto out_unregister_udp_proto;
1674 
1675 	rc = proto_register(&ping_prot, 1);
1676 	if (rc)
1677 		goto out_unregister_raw_proto;
1678 
1679 	/*
1680 	 *	Tell SOCKET that we are alive...
1681 	 */
1682 
1683 	(void)sock_register(&inet_family_ops);
1684 
1685 #ifdef CONFIG_SYSCTL
1686 	ip_static_sysctl_init();
1687 #endif
1688 
1689 	tcp_prot.sysctl_mem = init_net.ipv4.sysctl_tcp_mem;
1690 
1691 	/*
1692 	 *	Add all the base protocols.
1693 	 */
1694 
1695 	if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1696 		pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1697 	if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1698 		pr_crit("%s: Cannot add UDP protocol\n", __func__);
1699 	if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1700 		pr_crit("%s: Cannot add TCP protocol\n", __func__);
1701 #ifdef CONFIG_IP_MULTICAST
1702 	if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1703 		pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1704 #endif
1705 
1706 	/* Register the socket-side information for inet_create. */
1707 	for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1708 		INIT_LIST_HEAD(r);
1709 
1710 	for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1711 		inet_register_protosw(q);
1712 
1713 	/*
1714 	 *	Set the ARP module up
1715 	 */
1716 
1717 	arp_init();
1718 
1719 	/*
1720 	 *	Set the IP module up
1721 	 */
1722 
1723 	ip_init();
1724 
1725 	tcp_v4_init();
1726 
1727 	/* Setup TCP slab cache for open requests. */
1728 	tcp_init();
1729 
1730 	/* Setup UDP memory threshold */
1731 	udp_init();
1732 
1733 	/* Add UDP-Lite (RFC 3828) */
1734 	udplite4_register();
1735 
1736 	ping_init();
1737 
1738 	/*
1739 	 *	Set the ICMP layer up
1740 	 */
1741 
1742 	if (icmp_init() < 0)
1743 		panic("Failed to create the ICMP control socket.\n");
1744 
1745 	/*
1746 	 *	Initialise the multicast router
1747 	 */
1748 #if defined(CONFIG_IP_MROUTE)
1749 	if (ip_mr_init())
1750 		pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1751 #endif
1752 	/*
1753 	 *	Initialise per-cpu ipv4 mibs
1754 	 */
1755 
1756 	if (init_ipv4_mibs())
1757 		pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1758 
1759 	ipv4_proc_init();
1760 
1761 	ipfrag_init();
1762 
1763 	dev_add_pack(&ip_packet_type);
1764 
1765 	rc = 0;
1766 out:
1767 	return rc;
1768 out_unregister_raw_proto:
1769 	proto_unregister(&raw_prot);
1770 out_unregister_udp_proto:
1771 	proto_unregister(&udp_prot);
1772 out_unregister_tcp_proto:
1773 	proto_unregister(&tcp_prot);
1774 out_free_reserved_ports:
1775 	kfree(sysctl_local_reserved_ports);
1776 	goto out;
1777 }
1778 
1779 fs_initcall(inet_init);
1780 
1781 /* ------------------------------------------------------------------------ */
1782 
1783 #ifdef CONFIG_PROC_FS
1784 static int __init ipv4_proc_init(void)
1785 {
1786 	int rc = 0;
1787 
1788 	if (raw_proc_init())
1789 		goto out_raw;
1790 	if (tcp4_proc_init())
1791 		goto out_tcp;
1792 	if (udp4_proc_init())
1793 		goto out_udp;
1794 	if (ping_proc_init())
1795 		goto out_ping;
1796 	if (ip_misc_proc_init())
1797 		goto out_misc;
1798 out:
1799 	return rc;
1800 out_misc:
1801 	ping_proc_exit();
1802 out_ping:
1803 	udp4_proc_exit();
1804 out_udp:
1805 	tcp4_proc_exit();
1806 out_tcp:
1807 	raw_proc_exit();
1808 out_raw:
1809 	rc = -ENOMEM;
1810 	goto out;
1811 }
1812 
1813 #else /* CONFIG_PROC_FS */
1814 static int __init ipv4_proc_init(void)
1815 {
1816 	return 0;
1817 }
1818 #endif /* CONFIG_PROC_FS */
1819 
1820 MODULE_ALIAS_NETPROTO(PF_INET);
1821 
1822