xref: /openbmc/linux/net/ipv4/af_inet.c (revision 05bcf503)
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 		/* Check special setups for testing purpose to enable TFO w/o
216 		 * requiring TCP_FASTOPEN sockopt.
217 		 * Note that only TCP sockets (SOCK_STREAM) will reach here.
218 		 * Also fastopenq may already been allocated because this
219 		 * socket was in TCP_LISTEN state previously but was
220 		 * shutdown() (rather than close()).
221 		 */
222 		if ((sysctl_tcp_fastopen & TFO_SERVER_ENABLE) != 0 &&
223 		    inet_csk(sk)->icsk_accept_queue.fastopenq == NULL) {
224 			if ((sysctl_tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) != 0)
225 				err = fastopen_init_queue(sk, backlog);
226 			else if ((sysctl_tcp_fastopen &
227 				  TFO_SERVER_WO_SOCKOPT2) != 0)
228 				err = fastopen_init_queue(sk,
229 				    ((uint)sysctl_tcp_fastopen) >> 16);
230 			else
231 				err = 0;
232 			if (err)
233 				goto out;
234 		}
235 		err = inet_csk_listen_start(sk, backlog);
236 		if (err)
237 			goto out;
238 	}
239 	sk->sk_max_ack_backlog = backlog;
240 	err = 0;
241 
242 out:
243 	release_sock(sk);
244 	return err;
245 }
246 EXPORT_SYMBOL(inet_listen);
247 
248 u32 inet_ehash_secret __read_mostly;
249 EXPORT_SYMBOL(inet_ehash_secret);
250 
251 /*
252  * inet_ehash_secret must be set exactly once
253  */
254 void build_ehash_secret(void)
255 {
256 	u32 rnd;
257 
258 	do {
259 		get_random_bytes(&rnd, sizeof(rnd));
260 	} while (rnd == 0);
261 
262 	cmpxchg(&inet_ehash_secret, 0, rnd);
263 }
264 EXPORT_SYMBOL(build_ehash_secret);
265 
266 static inline int inet_netns_ok(struct net *net, __u8 protocol)
267 {
268 	const struct net_protocol *ipprot;
269 
270 	if (net_eq(net, &init_net))
271 		return 1;
272 
273 	ipprot = rcu_dereference(inet_protos[protocol]);
274 	if (ipprot == NULL) {
275 		/* raw IP is OK */
276 		return 1;
277 	}
278 	return ipprot->netns_ok;
279 }
280 
281 /*
282  *	Create an inet socket.
283  */
284 
285 static int inet_create(struct net *net, struct socket *sock, int protocol,
286 		       int kern)
287 {
288 	struct sock *sk;
289 	struct inet_protosw *answer;
290 	struct inet_sock *inet;
291 	struct proto *answer_prot;
292 	unsigned char answer_flags;
293 	char answer_no_check;
294 	int try_loading_module = 0;
295 	int err;
296 
297 	if (unlikely(!inet_ehash_secret))
298 		if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
299 			build_ehash_secret();
300 
301 	sock->state = SS_UNCONNECTED;
302 
303 	/* Look for the requested type/protocol pair. */
304 lookup_protocol:
305 	err = -ESOCKTNOSUPPORT;
306 	rcu_read_lock();
307 	list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
308 
309 		err = 0;
310 		/* Check the non-wild match. */
311 		if (protocol == answer->protocol) {
312 			if (protocol != IPPROTO_IP)
313 				break;
314 		} else {
315 			/* Check for the two wild cases. */
316 			if (IPPROTO_IP == protocol) {
317 				protocol = answer->protocol;
318 				break;
319 			}
320 			if (IPPROTO_IP == answer->protocol)
321 				break;
322 		}
323 		err = -EPROTONOSUPPORT;
324 	}
325 
326 	if (unlikely(err)) {
327 		if (try_loading_module < 2) {
328 			rcu_read_unlock();
329 			/*
330 			 * Be more specific, e.g. net-pf-2-proto-132-type-1
331 			 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
332 			 */
333 			if (++try_loading_module == 1)
334 				request_module("net-pf-%d-proto-%d-type-%d",
335 					       PF_INET, protocol, sock->type);
336 			/*
337 			 * Fall back to generic, e.g. net-pf-2-proto-132
338 			 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
339 			 */
340 			else
341 				request_module("net-pf-%d-proto-%d",
342 					       PF_INET, protocol);
343 			goto lookup_protocol;
344 		} else
345 			goto out_rcu_unlock;
346 	}
347 
348 	err = -EPERM;
349 	if (sock->type == SOCK_RAW && !kern && !capable(CAP_NET_RAW))
350 		goto out_rcu_unlock;
351 
352 	err = -EAFNOSUPPORT;
353 	if (!inet_netns_ok(net, protocol))
354 		goto out_rcu_unlock;
355 
356 	sock->ops = answer->ops;
357 	answer_prot = answer->prot;
358 	answer_no_check = answer->no_check;
359 	answer_flags = answer->flags;
360 	rcu_read_unlock();
361 
362 	WARN_ON(answer_prot->slab == NULL);
363 
364 	err = -ENOBUFS;
365 	sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
366 	if (sk == NULL)
367 		goto out;
368 
369 	err = 0;
370 	sk->sk_no_check = answer_no_check;
371 	if (INET_PROTOSW_REUSE & answer_flags)
372 		sk->sk_reuse = SK_CAN_REUSE;
373 
374 	inet = inet_sk(sk);
375 	inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
376 
377 	inet->nodefrag = 0;
378 
379 	if (SOCK_RAW == sock->type) {
380 		inet->inet_num = protocol;
381 		if (IPPROTO_RAW == protocol)
382 			inet->hdrincl = 1;
383 	}
384 
385 	if (ipv4_config.no_pmtu_disc)
386 		inet->pmtudisc = IP_PMTUDISC_DONT;
387 	else
388 		inet->pmtudisc = IP_PMTUDISC_WANT;
389 
390 	inet->inet_id = 0;
391 
392 	sock_init_data(sock, sk);
393 
394 	sk->sk_destruct	   = inet_sock_destruct;
395 	sk->sk_protocol	   = protocol;
396 	sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
397 
398 	inet->uc_ttl	= -1;
399 	inet->mc_loop	= 1;
400 	inet->mc_ttl	= 1;
401 	inet->mc_all	= 1;
402 	inet->mc_index	= 0;
403 	inet->mc_list	= NULL;
404 	inet->rcv_tos	= 0;
405 
406 	sk_refcnt_debug_inc(sk);
407 
408 	if (inet->inet_num) {
409 		/* It assumes that any protocol which allows
410 		 * the user to assign a number at socket
411 		 * creation time automatically
412 		 * shares.
413 		 */
414 		inet->inet_sport = htons(inet->inet_num);
415 		/* Add to protocol hash chains. */
416 		sk->sk_prot->hash(sk);
417 	}
418 
419 	if (sk->sk_prot->init) {
420 		err = sk->sk_prot->init(sk);
421 		if (err)
422 			sk_common_release(sk);
423 	}
424 out:
425 	return err;
426 out_rcu_unlock:
427 	rcu_read_unlock();
428 	goto out;
429 }
430 
431 
432 /*
433  *	The peer socket should always be NULL (or else). When we call this
434  *	function we are destroying the object and from then on nobody
435  *	should refer to it.
436  */
437 int inet_release(struct socket *sock)
438 {
439 	struct sock *sk = sock->sk;
440 
441 	if (sk) {
442 		long timeout;
443 
444 		sock_rps_reset_flow(sk);
445 
446 		/* Applications forget to leave groups before exiting */
447 		ip_mc_drop_socket(sk);
448 
449 		/* If linger is set, we don't return until the close
450 		 * is complete.  Otherwise we return immediately. The
451 		 * actually closing is done the same either way.
452 		 *
453 		 * If the close is due to the process exiting, we never
454 		 * linger..
455 		 */
456 		timeout = 0;
457 		if (sock_flag(sk, SOCK_LINGER) &&
458 		    !(current->flags & PF_EXITING))
459 			timeout = sk->sk_lingertime;
460 		sock->sk = NULL;
461 		sk->sk_prot->close(sk, timeout);
462 	}
463 	return 0;
464 }
465 EXPORT_SYMBOL(inet_release);
466 
467 /* It is off by default, see below. */
468 int sysctl_ip_nonlocal_bind __read_mostly;
469 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
470 
471 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
472 {
473 	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
474 	struct sock *sk = sock->sk;
475 	struct inet_sock *inet = inet_sk(sk);
476 	unsigned short snum;
477 	int chk_addr_ret;
478 	int err;
479 
480 	/* If the socket has its own bind function then use it. (RAW) */
481 	if (sk->sk_prot->bind) {
482 		err = sk->sk_prot->bind(sk, uaddr, addr_len);
483 		goto out;
484 	}
485 	err = -EINVAL;
486 	if (addr_len < sizeof(struct sockaddr_in))
487 		goto out;
488 
489 	if (addr->sin_family != AF_INET) {
490 		/* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
491 		 * only if s_addr is INADDR_ANY.
492 		 */
493 		err = -EAFNOSUPPORT;
494 		if (addr->sin_family != AF_UNSPEC ||
495 		    addr->sin_addr.s_addr != htonl(INADDR_ANY))
496 			goto out;
497 	}
498 
499 	chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr);
500 
501 	/* Not specified by any standard per-se, however it breaks too
502 	 * many applications when removed.  It is unfortunate since
503 	 * allowing applications to make a non-local bind solves
504 	 * several problems with systems using dynamic addressing.
505 	 * (ie. your servers still start up even if your ISDN link
506 	 *  is temporarily down)
507 	 */
508 	err = -EADDRNOTAVAIL;
509 	if (!sysctl_ip_nonlocal_bind &&
510 	    !(inet->freebind || inet->transparent) &&
511 	    addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
512 	    chk_addr_ret != RTN_LOCAL &&
513 	    chk_addr_ret != RTN_MULTICAST &&
514 	    chk_addr_ret != RTN_BROADCAST)
515 		goto out;
516 
517 	snum = ntohs(addr->sin_port);
518 	err = -EACCES;
519 	if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
520 		goto out;
521 
522 	/*      We keep a pair of addresses. rcv_saddr is the one
523 	 *      used by hash lookups, and saddr is used for transmit.
524 	 *
525 	 *      In the BSD API these are the same except where it
526 	 *      would be illegal to use them (multicast/broadcast) in
527 	 *      which case the sending device address is used.
528 	 */
529 	lock_sock(sk);
530 
531 	/* Check these errors (active socket, double bind). */
532 	err = -EINVAL;
533 	if (sk->sk_state != TCP_CLOSE || inet->inet_num)
534 		goto out_release_sock;
535 
536 	inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
537 	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
538 		inet->inet_saddr = 0;  /* Use device */
539 
540 	/* Make sure we are allowed to bind here. */
541 	if (sk->sk_prot->get_port(sk, snum)) {
542 		inet->inet_saddr = inet->inet_rcv_saddr = 0;
543 		err = -EADDRINUSE;
544 		goto out_release_sock;
545 	}
546 
547 	if (inet->inet_rcv_saddr)
548 		sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
549 	if (snum)
550 		sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
551 	inet->inet_sport = htons(inet->inet_num);
552 	inet->inet_daddr = 0;
553 	inet->inet_dport = 0;
554 	sk_dst_reset(sk);
555 	err = 0;
556 out_release_sock:
557 	release_sock(sk);
558 out:
559 	return err;
560 }
561 EXPORT_SYMBOL(inet_bind);
562 
563 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
564 		       int addr_len, int flags)
565 {
566 	struct sock *sk = sock->sk;
567 
568 	if (addr_len < sizeof(uaddr->sa_family))
569 		return -EINVAL;
570 	if (uaddr->sa_family == AF_UNSPEC)
571 		return sk->sk_prot->disconnect(sk, flags);
572 
573 	if (!inet_sk(sk)->inet_num && inet_autobind(sk))
574 		return -EAGAIN;
575 	return sk->sk_prot->connect(sk, uaddr, addr_len);
576 }
577 EXPORT_SYMBOL(inet_dgram_connect);
578 
579 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
580 {
581 	DEFINE_WAIT(wait);
582 
583 	prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
584 	sk->sk_write_pending += writebias;
585 
586 	/* Basic assumption: if someone sets sk->sk_err, he _must_
587 	 * change state of the socket from TCP_SYN_*.
588 	 * Connect() does not allow to get error notifications
589 	 * without closing the socket.
590 	 */
591 	while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
592 		release_sock(sk);
593 		timeo = schedule_timeout(timeo);
594 		lock_sock(sk);
595 		if (signal_pending(current) || !timeo)
596 			break;
597 		prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
598 	}
599 	finish_wait(sk_sleep(sk), &wait);
600 	sk->sk_write_pending -= writebias;
601 	return timeo;
602 }
603 
604 /*
605  *	Connect to a remote host. There is regrettably still a little
606  *	TCP 'magic' in here.
607  */
608 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
609 			  int addr_len, int flags)
610 {
611 	struct sock *sk = sock->sk;
612 	int err;
613 	long timeo;
614 
615 	if (addr_len < sizeof(uaddr->sa_family))
616 		return -EINVAL;
617 
618 	if (uaddr->sa_family == AF_UNSPEC) {
619 		err = sk->sk_prot->disconnect(sk, flags);
620 		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
621 		goto out;
622 	}
623 
624 	switch (sock->state) {
625 	default:
626 		err = -EINVAL;
627 		goto out;
628 	case SS_CONNECTED:
629 		err = -EISCONN;
630 		goto out;
631 	case SS_CONNECTING:
632 		err = -EALREADY;
633 		/* Fall out of switch with err, set for this state */
634 		break;
635 	case SS_UNCONNECTED:
636 		err = -EISCONN;
637 		if (sk->sk_state != TCP_CLOSE)
638 			goto out;
639 
640 		err = sk->sk_prot->connect(sk, uaddr, addr_len);
641 		if (err < 0)
642 			goto out;
643 
644 		sock->state = SS_CONNECTING;
645 
646 		/* Just entered SS_CONNECTING state; the only
647 		 * difference is that return value in non-blocking
648 		 * case is EINPROGRESS, rather than EALREADY.
649 		 */
650 		err = -EINPROGRESS;
651 		break;
652 	}
653 
654 	timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
655 
656 	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
657 		int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
658 				tcp_sk(sk)->fastopen_req &&
659 				tcp_sk(sk)->fastopen_req->data ? 1 : 0;
660 
661 		/* Error code is set above */
662 		if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
663 			goto out;
664 
665 		err = sock_intr_errno(timeo);
666 		if (signal_pending(current))
667 			goto out;
668 	}
669 
670 	/* Connection was closed by RST, timeout, ICMP error
671 	 * or another process disconnected us.
672 	 */
673 	if (sk->sk_state == TCP_CLOSE)
674 		goto sock_error;
675 
676 	/* sk->sk_err may be not zero now, if RECVERR was ordered by user
677 	 * and error was received after socket entered established state.
678 	 * Hence, it is handled normally after connect() return successfully.
679 	 */
680 
681 	sock->state = SS_CONNECTED;
682 	err = 0;
683 out:
684 	return err;
685 
686 sock_error:
687 	err = sock_error(sk) ? : -ECONNABORTED;
688 	sock->state = SS_UNCONNECTED;
689 	if (sk->sk_prot->disconnect(sk, flags))
690 		sock->state = SS_DISCONNECTING;
691 	goto out;
692 }
693 EXPORT_SYMBOL(__inet_stream_connect);
694 
695 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
696 			int addr_len, int flags)
697 {
698 	int err;
699 
700 	lock_sock(sock->sk);
701 	err = __inet_stream_connect(sock, uaddr, addr_len, flags);
702 	release_sock(sock->sk);
703 	return err;
704 }
705 EXPORT_SYMBOL(inet_stream_connect);
706 
707 /*
708  *	Accept a pending connection. The TCP layer now gives BSD semantics.
709  */
710 
711 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
712 {
713 	struct sock *sk1 = sock->sk;
714 	int err = -EINVAL;
715 	struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
716 
717 	if (!sk2)
718 		goto do_err;
719 
720 	lock_sock(sk2);
721 
722 	sock_rps_record_flow(sk2);
723 	WARN_ON(!((1 << sk2->sk_state) &
724 		  (TCPF_ESTABLISHED | TCPF_SYN_RECV |
725 		  TCPF_CLOSE_WAIT | TCPF_CLOSE)));
726 
727 	sock_graft(sk2, newsock);
728 
729 	newsock->state = SS_CONNECTED;
730 	err = 0;
731 	release_sock(sk2);
732 do_err:
733 	return err;
734 }
735 EXPORT_SYMBOL(inet_accept);
736 
737 
738 /*
739  *	This does both peername and sockname.
740  */
741 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
742 			int *uaddr_len, int peer)
743 {
744 	struct sock *sk		= sock->sk;
745 	struct inet_sock *inet	= inet_sk(sk);
746 	DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
747 
748 	sin->sin_family = AF_INET;
749 	if (peer) {
750 		if (!inet->inet_dport ||
751 		    (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
752 		     peer == 1))
753 			return -ENOTCONN;
754 		sin->sin_port = inet->inet_dport;
755 		sin->sin_addr.s_addr = inet->inet_daddr;
756 	} else {
757 		__be32 addr = inet->inet_rcv_saddr;
758 		if (!addr)
759 			addr = inet->inet_saddr;
760 		sin->sin_port = inet->inet_sport;
761 		sin->sin_addr.s_addr = addr;
762 	}
763 	memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
764 	*uaddr_len = sizeof(*sin);
765 	return 0;
766 }
767 EXPORT_SYMBOL(inet_getname);
768 
769 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
770 		 size_t size)
771 {
772 	struct sock *sk = sock->sk;
773 
774 	sock_rps_record_flow(sk);
775 
776 	/* We may need to bind the socket. */
777 	if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
778 	    inet_autobind(sk))
779 		return -EAGAIN;
780 
781 	return sk->sk_prot->sendmsg(iocb, sk, msg, size);
782 }
783 EXPORT_SYMBOL(inet_sendmsg);
784 
785 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
786 		      size_t size, int flags)
787 {
788 	struct sock *sk = sock->sk;
789 
790 	sock_rps_record_flow(sk);
791 
792 	/* We may need to bind the socket. */
793 	if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
794 	    inet_autobind(sk))
795 		return -EAGAIN;
796 
797 	if (sk->sk_prot->sendpage)
798 		return sk->sk_prot->sendpage(sk, page, offset, size, flags);
799 	return sock_no_sendpage(sock, page, offset, size, flags);
800 }
801 EXPORT_SYMBOL(inet_sendpage);
802 
803 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
804 		 size_t size, int flags)
805 {
806 	struct sock *sk = sock->sk;
807 	int addr_len = 0;
808 	int err;
809 
810 	sock_rps_record_flow(sk);
811 
812 	err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
813 				   flags & ~MSG_DONTWAIT, &addr_len);
814 	if (err >= 0)
815 		msg->msg_namelen = addr_len;
816 	return err;
817 }
818 EXPORT_SYMBOL(inet_recvmsg);
819 
820 int inet_shutdown(struct socket *sock, int how)
821 {
822 	struct sock *sk = sock->sk;
823 	int err = 0;
824 
825 	/* This should really check to make sure
826 	 * the socket is a TCP socket. (WHY AC...)
827 	 */
828 	how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
829 		       1->2 bit 2 snds.
830 		       2->3 */
831 	if ((how & ~SHUTDOWN_MASK) || !how)	/* MAXINT->0 */
832 		return -EINVAL;
833 
834 	lock_sock(sk);
835 	if (sock->state == SS_CONNECTING) {
836 		if ((1 << sk->sk_state) &
837 		    (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
838 			sock->state = SS_DISCONNECTING;
839 		else
840 			sock->state = SS_CONNECTED;
841 	}
842 
843 	switch (sk->sk_state) {
844 	case TCP_CLOSE:
845 		err = -ENOTCONN;
846 		/* Hack to wake up other listeners, who can poll for
847 		   POLLHUP, even on eg. unconnected UDP sockets -- RR */
848 	default:
849 		sk->sk_shutdown |= how;
850 		if (sk->sk_prot->shutdown)
851 			sk->sk_prot->shutdown(sk, how);
852 		break;
853 
854 	/* Remaining two branches are temporary solution for missing
855 	 * close() in multithreaded environment. It is _not_ a good idea,
856 	 * but we have no choice until close() is repaired at VFS level.
857 	 */
858 	case TCP_LISTEN:
859 		if (!(how & RCV_SHUTDOWN))
860 			break;
861 		/* Fall through */
862 	case TCP_SYN_SENT:
863 		err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
864 		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
865 		break;
866 	}
867 
868 	/* Wake up anyone sleeping in poll. */
869 	sk->sk_state_change(sk);
870 	release_sock(sk);
871 	return err;
872 }
873 EXPORT_SYMBOL(inet_shutdown);
874 
875 /*
876  *	ioctl() calls you can issue on an INET socket. Most of these are
877  *	device configuration and stuff and very rarely used. Some ioctls
878  *	pass on to the socket itself.
879  *
880  *	NOTE: I like the idea of a module for the config stuff. ie ifconfig
881  *	loads the devconfigure module does its configuring and unloads it.
882  *	There's a good 20K of config code hanging around the kernel.
883  */
884 
885 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
886 {
887 	struct sock *sk = sock->sk;
888 	int err = 0;
889 	struct net *net = sock_net(sk);
890 
891 	switch (cmd) {
892 	case SIOCGSTAMP:
893 		err = sock_get_timestamp(sk, (struct timeval __user *)arg);
894 		break;
895 	case SIOCGSTAMPNS:
896 		err = sock_get_timestampns(sk, (struct timespec __user *)arg);
897 		break;
898 	case SIOCADDRT:
899 	case SIOCDELRT:
900 	case SIOCRTMSG:
901 		err = ip_rt_ioctl(net, cmd, (void __user *)arg);
902 		break;
903 	case SIOCDARP:
904 	case SIOCGARP:
905 	case SIOCSARP:
906 		err = arp_ioctl(net, cmd, (void __user *)arg);
907 		break;
908 	case SIOCGIFADDR:
909 	case SIOCSIFADDR:
910 	case SIOCGIFBRDADDR:
911 	case SIOCSIFBRDADDR:
912 	case SIOCGIFNETMASK:
913 	case SIOCSIFNETMASK:
914 	case SIOCGIFDSTADDR:
915 	case SIOCSIFDSTADDR:
916 	case SIOCSIFPFLAGS:
917 	case SIOCGIFPFLAGS:
918 	case SIOCSIFFLAGS:
919 		err = devinet_ioctl(net, cmd, (void __user *)arg);
920 		break;
921 	default:
922 		if (sk->sk_prot->ioctl)
923 			err = sk->sk_prot->ioctl(sk, cmd, arg);
924 		else
925 			err = -ENOIOCTLCMD;
926 		break;
927 	}
928 	return err;
929 }
930 EXPORT_SYMBOL(inet_ioctl);
931 
932 #ifdef CONFIG_COMPAT
933 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
934 {
935 	struct sock *sk = sock->sk;
936 	int err = -ENOIOCTLCMD;
937 
938 	if (sk->sk_prot->compat_ioctl)
939 		err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
940 
941 	return err;
942 }
943 #endif
944 
945 const struct proto_ops inet_stream_ops = {
946 	.family		   = PF_INET,
947 	.owner		   = THIS_MODULE,
948 	.release	   = inet_release,
949 	.bind		   = inet_bind,
950 	.connect	   = inet_stream_connect,
951 	.socketpair	   = sock_no_socketpair,
952 	.accept		   = inet_accept,
953 	.getname	   = inet_getname,
954 	.poll		   = tcp_poll,
955 	.ioctl		   = inet_ioctl,
956 	.listen		   = inet_listen,
957 	.shutdown	   = inet_shutdown,
958 	.setsockopt	   = sock_common_setsockopt,
959 	.getsockopt	   = sock_common_getsockopt,
960 	.sendmsg	   = inet_sendmsg,
961 	.recvmsg	   = inet_recvmsg,
962 	.mmap		   = sock_no_mmap,
963 	.sendpage	   = inet_sendpage,
964 	.splice_read	   = tcp_splice_read,
965 #ifdef CONFIG_COMPAT
966 	.compat_setsockopt = compat_sock_common_setsockopt,
967 	.compat_getsockopt = compat_sock_common_getsockopt,
968 	.compat_ioctl	   = inet_compat_ioctl,
969 #endif
970 };
971 EXPORT_SYMBOL(inet_stream_ops);
972 
973 const struct proto_ops inet_dgram_ops = {
974 	.family		   = PF_INET,
975 	.owner		   = THIS_MODULE,
976 	.release	   = inet_release,
977 	.bind		   = inet_bind,
978 	.connect	   = inet_dgram_connect,
979 	.socketpair	   = sock_no_socketpair,
980 	.accept		   = sock_no_accept,
981 	.getname	   = inet_getname,
982 	.poll		   = udp_poll,
983 	.ioctl		   = inet_ioctl,
984 	.listen		   = sock_no_listen,
985 	.shutdown	   = inet_shutdown,
986 	.setsockopt	   = sock_common_setsockopt,
987 	.getsockopt	   = sock_common_getsockopt,
988 	.sendmsg	   = inet_sendmsg,
989 	.recvmsg	   = inet_recvmsg,
990 	.mmap		   = sock_no_mmap,
991 	.sendpage	   = inet_sendpage,
992 #ifdef CONFIG_COMPAT
993 	.compat_setsockopt = compat_sock_common_setsockopt,
994 	.compat_getsockopt = compat_sock_common_getsockopt,
995 	.compat_ioctl	   = inet_compat_ioctl,
996 #endif
997 };
998 EXPORT_SYMBOL(inet_dgram_ops);
999 
1000 /*
1001  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1002  * udp_poll
1003  */
1004 static const struct proto_ops inet_sockraw_ops = {
1005 	.family		   = PF_INET,
1006 	.owner		   = THIS_MODULE,
1007 	.release	   = inet_release,
1008 	.bind		   = inet_bind,
1009 	.connect	   = inet_dgram_connect,
1010 	.socketpair	   = sock_no_socketpair,
1011 	.accept		   = sock_no_accept,
1012 	.getname	   = inet_getname,
1013 	.poll		   = datagram_poll,
1014 	.ioctl		   = inet_ioctl,
1015 	.listen		   = sock_no_listen,
1016 	.shutdown	   = inet_shutdown,
1017 	.setsockopt	   = sock_common_setsockopt,
1018 	.getsockopt	   = sock_common_getsockopt,
1019 	.sendmsg	   = inet_sendmsg,
1020 	.recvmsg	   = inet_recvmsg,
1021 	.mmap		   = sock_no_mmap,
1022 	.sendpage	   = inet_sendpage,
1023 #ifdef CONFIG_COMPAT
1024 	.compat_setsockopt = compat_sock_common_setsockopt,
1025 	.compat_getsockopt = compat_sock_common_getsockopt,
1026 	.compat_ioctl	   = inet_compat_ioctl,
1027 #endif
1028 };
1029 
1030 static const struct net_proto_family inet_family_ops = {
1031 	.family = PF_INET,
1032 	.create = inet_create,
1033 	.owner	= THIS_MODULE,
1034 };
1035 
1036 /* Upon startup we insert all the elements in inetsw_array[] into
1037  * the linked list inetsw.
1038  */
1039 static struct inet_protosw inetsw_array[] =
1040 {
1041 	{
1042 		.type =       SOCK_STREAM,
1043 		.protocol =   IPPROTO_TCP,
1044 		.prot =       &tcp_prot,
1045 		.ops =        &inet_stream_ops,
1046 		.no_check =   0,
1047 		.flags =      INET_PROTOSW_PERMANENT |
1048 			      INET_PROTOSW_ICSK,
1049 	},
1050 
1051 	{
1052 		.type =       SOCK_DGRAM,
1053 		.protocol =   IPPROTO_UDP,
1054 		.prot =       &udp_prot,
1055 		.ops =        &inet_dgram_ops,
1056 		.no_check =   UDP_CSUM_DEFAULT,
1057 		.flags =      INET_PROTOSW_PERMANENT,
1058        },
1059 
1060        {
1061 		.type =       SOCK_DGRAM,
1062 		.protocol =   IPPROTO_ICMP,
1063 		.prot =       &ping_prot,
1064 		.ops =        &inet_dgram_ops,
1065 		.no_check =   UDP_CSUM_DEFAULT,
1066 		.flags =      INET_PROTOSW_REUSE,
1067        },
1068 
1069        {
1070 	       .type =       SOCK_RAW,
1071 	       .protocol =   IPPROTO_IP,	/* wild card */
1072 	       .prot =       &raw_prot,
1073 	       .ops =        &inet_sockraw_ops,
1074 	       .no_check =   UDP_CSUM_DEFAULT,
1075 	       .flags =      INET_PROTOSW_REUSE,
1076        }
1077 };
1078 
1079 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1080 
1081 void inet_register_protosw(struct inet_protosw *p)
1082 {
1083 	struct list_head *lh;
1084 	struct inet_protosw *answer;
1085 	int protocol = p->protocol;
1086 	struct list_head *last_perm;
1087 
1088 	spin_lock_bh(&inetsw_lock);
1089 
1090 	if (p->type >= SOCK_MAX)
1091 		goto out_illegal;
1092 
1093 	/* If we are trying to override a permanent protocol, bail. */
1094 	answer = NULL;
1095 	last_perm = &inetsw[p->type];
1096 	list_for_each(lh, &inetsw[p->type]) {
1097 		answer = list_entry(lh, struct inet_protosw, list);
1098 
1099 		/* Check only the non-wild match. */
1100 		if (INET_PROTOSW_PERMANENT & answer->flags) {
1101 			if (protocol == answer->protocol)
1102 				break;
1103 			last_perm = lh;
1104 		}
1105 
1106 		answer = NULL;
1107 	}
1108 	if (answer)
1109 		goto out_permanent;
1110 
1111 	/* Add the new entry after the last permanent entry if any, so that
1112 	 * the new entry does not override a permanent entry when matched with
1113 	 * a wild-card protocol. But it is allowed to override any existing
1114 	 * non-permanent entry.  This means that when we remove this entry, the
1115 	 * system automatically returns to the old behavior.
1116 	 */
1117 	list_add_rcu(&p->list, last_perm);
1118 out:
1119 	spin_unlock_bh(&inetsw_lock);
1120 
1121 	return;
1122 
1123 out_permanent:
1124 	pr_err("Attempt to override permanent protocol %d\n", protocol);
1125 	goto out;
1126 
1127 out_illegal:
1128 	pr_err("Ignoring attempt to register invalid socket type %d\n",
1129 	       p->type);
1130 	goto out;
1131 }
1132 EXPORT_SYMBOL(inet_register_protosw);
1133 
1134 void inet_unregister_protosw(struct inet_protosw *p)
1135 {
1136 	if (INET_PROTOSW_PERMANENT & p->flags) {
1137 		pr_err("Attempt to unregister permanent protocol %d\n",
1138 		       p->protocol);
1139 	} else {
1140 		spin_lock_bh(&inetsw_lock);
1141 		list_del_rcu(&p->list);
1142 		spin_unlock_bh(&inetsw_lock);
1143 
1144 		synchronize_net();
1145 	}
1146 }
1147 EXPORT_SYMBOL(inet_unregister_protosw);
1148 
1149 /*
1150  *      Shall we try to damage output packets if routing dev changes?
1151  */
1152 
1153 int sysctl_ip_dynaddr __read_mostly;
1154 
1155 static int inet_sk_reselect_saddr(struct sock *sk)
1156 {
1157 	struct inet_sock *inet = inet_sk(sk);
1158 	__be32 old_saddr = inet->inet_saddr;
1159 	__be32 daddr = inet->inet_daddr;
1160 	struct flowi4 *fl4;
1161 	struct rtable *rt;
1162 	__be32 new_saddr;
1163 	struct ip_options_rcu *inet_opt;
1164 
1165 	inet_opt = rcu_dereference_protected(inet->inet_opt,
1166 					     sock_owned_by_user(sk));
1167 	if (inet_opt && inet_opt->opt.srr)
1168 		daddr = inet_opt->opt.faddr;
1169 
1170 	/* Query new route. */
1171 	fl4 = &inet->cork.fl.u.ip4;
1172 	rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1173 			      sk->sk_bound_dev_if, sk->sk_protocol,
1174 			      inet->inet_sport, inet->inet_dport, sk, false);
1175 	if (IS_ERR(rt))
1176 		return PTR_ERR(rt);
1177 
1178 	sk_setup_caps(sk, &rt->dst);
1179 
1180 	new_saddr = fl4->saddr;
1181 
1182 	if (new_saddr == old_saddr)
1183 		return 0;
1184 
1185 	if (sysctl_ip_dynaddr > 1) {
1186 		pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1187 			__func__, &old_saddr, &new_saddr);
1188 	}
1189 
1190 	inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1191 
1192 	/*
1193 	 * XXX The only one ugly spot where we need to
1194 	 * XXX really change the sockets identity after
1195 	 * XXX it has entered the hashes. -DaveM
1196 	 *
1197 	 * Besides that, it does not check for connection
1198 	 * uniqueness. Wait for troubles.
1199 	 */
1200 	__sk_prot_rehash(sk);
1201 	return 0;
1202 }
1203 
1204 int inet_sk_rebuild_header(struct sock *sk)
1205 {
1206 	struct inet_sock *inet = inet_sk(sk);
1207 	struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1208 	__be32 daddr;
1209 	struct ip_options_rcu *inet_opt;
1210 	struct flowi4 *fl4;
1211 	int err;
1212 
1213 	/* Route is OK, nothing to do. */
1214 	if (rt)
1215 		return 0;
1216 
1217 	/* Reroute. */
1218 	rcu_read_lock();
1219 	inet_opt = rcu_dereference(inet->inet_opt);
1220 	daddr = inet->inet_daddr;
1221 	if (inet_opt && inet_opt->opt.srr)
1222 		daddr = inet_opt->opt.faddr;
1223 	rcu_read_unlock();
1224 	fl4 = &inet->cork.fl.u.ip4;
1225 	rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1226 				   inet->inet_dport, inet->inet_sport,
1227 				   sk->sk_protocol, RT_CONN_FLAGS(sk),
1228 				   sk->sk_bound_dev_if);
1229 	if (!IS_ERR(rt)) {
1230 		err = 0;
1231 		sk_setup_caps(sk, &rt->dst);
1232 	} else {
1233 		err = PTR_ERR(rt);
1234 
1235 		/* Routing failed... */
1236 		sk->sk_route_caps = 0;
1237 		/*
1238 		 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1239 		 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1240 		 */
1241 		if (!sysctl_ip_dynaddr ||
1242 		    sk->sk_state != TCP_SYN_SENT ||
1243 		    (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1244 		    (err = inet_sk_reselect_saddr(sk)) != 0)
1245 			sk->sk_err_soft = -err;
1246 	}
1247 
1248 	return err;
1249 }
1250 EXPORT_SYMBOL(inet_sk_rebuild_header);
1251 
1252 static int inet_gso_send_check(struct sk_buff *skb)
1253 {
1254 	const struct net_protocol *ops;
1255 	const struct iphdr *iph;
1256 	int proto;
1257 	int ihl;
1258 	int err = -EINVAL;
1259 
1260 	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1261 		goto out;
1262 
1263 	iph = ip_hdr(skb);
1264 	ihl = iph->ihl * 4;
1265 	if (ihl < sizeof(*iph))
1266 		goto out;
1267 
1268 	if (unlikely(!pskb_may_pull(skb, ihl)))
1269 		goto out;
1270 
1271 	__skb_pull(skb, ihl);
1272 	skb_reset_transport_header(skb);
1273 	iph = ip_hdr(skb);
1274 	proto = iph->protocol;
1275 	err = -EPROTONOSUPPORT;
1276 
1277 	rcu_read_lock();
1278 	ops = rcu_dereference(inet_protos[proto]);
1279 	if (likely(ops && ops->gso_send_check))
1280 		err = ops->gso_send_check(skb);
1281 	rcu_read_unlock();
1282 
1283 out:
1284 	return err;
1285 }
1286 
1287 static struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1288 	netdev_features_t features)
1289 {
1290 	struct sk_buff *segs = ERR_PTR(-EINVAL);
1291 	const struct net_protocol *ops;
1292 	struct iphdr *iph;
1293 	int proto;
1294 	int ihl;
1295 	int id;
1296 	unsigned int offset = 0;
1297 
1298 	if (!(features & NETIF_F_V4_CSUM))
1299 		features &= ~NETIF_F_SG;
1300 
1301 	if (unlikely(skb_shinfo(skb)->gso_type &
1302 		     ~(SKB_GSO_TCPV4 |
1303 		       SKB_GSO_UDP |
1304 		       SKB_GSO_DODGY |
1305 		       SKB_GSO_TCP_ECN |
1306 		       0)))
1307 		goto out;
1308 
1309 	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1310 		goto out;
1311 
1312 	iph = ip_hdr(skb);
1313 	ihl = iph->ihl * 4;
1314 	if (ihl < sizeof(*iph))
1315 		goto out;
1316 
1317 	if (unlikely(!pskb_may_pull(skb, ihl)))
1318 		goto out;
1319 
1320 	__skb_pull(skb, ihl);
1321 	skb_reset_transport_header(skb);
1322 	iph = ip_hdr(skb);
1323 	id = ntohs(iph->id);
1324 	proto = iph->protocol;
1325 	segs = ERR_PTR(-EPROTONOSUPPORT);
1326 
1327 	rcu_read_lock();
1328 	ops = rcu_dereference(inet_protos[proto]);
1329 	if (likely(ops && ops->gso_segment))
1330 		segs = ops->gso_segment(skb, features);
1331 	rcu_read_unlock();
1332 
1333 	if (!segs || IS_ERR(segs))
1334 		goto out;
1335 
1336 	skb = segs;
1337 	do {
1338 		iph = ip_hdr(skb);
1339 		if (proto == IPPROTO_UDP) {
1340 			iph->id = htons(id);
1341 			iph->frag_off = htons(offset >> 3);
1342 			if (skb->next != NULL)
1343 				iph->frag_off |= htons(IP_MF);
1344 			offset += (skb->len - skb->mac_len - iph->ihl * 4);
1345 		} else
1346 			iph->id = htons(id++);
1347 		iph->tot_len = htons(skb->len - skb->mac_len);
1348 		iph->check = 0;
1349 		iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1350 	} while ((skb = skb->next));
1351 
1352 out:
1353 	return segs;
1354 }
1355 
1356 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1357 					 struct sk_buff *skb)
1358 {
1359 	const struct net_protocol *ops;
1360 	struct sk_buff **pp = NULL;
1361 	struct sk_buff *p;
1362 	const struct iphdr *iph;
1363 	unsigned int hlen;
1364 	unsigned int off;
1365 	unsigned int id;
1366 	int flush = 1;
1367 	int proto;
1368 
1369 	off = skb_gro_offset(skb);
1370 	hlen = off + sizeof(*iph);
1371 	iph = skb_gro_header_fast(skb, off);
1372 	if (skb_gro_header_hard(skb, hlen)) {
1373 		iph = skb_gro_header_slow(skb, hlen, off);
1374 		if (unlikely(!iph))
1375 			goto out;
1376 	}
1377 
1378 	proto = iph->protocol;
1379 
1380 	rcu_read_lock();
1381 	ops = rcu_dereference(inet_protos[proto]);
1382 	if (!ops || !ops->gro_receive)
1383 		goto out_unlock;
1384 
1385 	if (*(u8 *)iph != 0x45)
1386 		goto out_unlock;
1387 
1388 	if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1389 		goto out_unlock;
1390 
1391 	id = ntohl(*(__be32 *)&iph->id);
1392 	flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF));
1393 	id >>= 16;
1394 
1395 	for (p = *head; p; p = p->next) {
1396 		struct iphdr *iph2;
1397 
1398 		if (!NAPI_GRO_CB(p)->same_flow)
1399 			continue;
1400 
1401 		iph2 = ip_hdr(p);
1402 
1403 		if ((iph->protocol ^ iph2->protocol) |
1404 		    ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1405 		    ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1406 			NAPI_GRO_CB(p)->same_flow = 0;
1407 			continue;
1408 		}
1409 
1410 		/* All fields must match except length and checksum. */
1411 		NAPI_GRO_CB(p)->flush |=
1412 			(iph->ttl ^ iph2->ttl) |
1413 			(iph->tos ^ iph2->tos) |
1414 			((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1415 
1416 		NAPI_GRO_CB(p)->flush |= flush;
1417 	}
1418 
1419 	NAPI_GRO_CB(skb)->flush |= flush;
1420 	skb_gro_pull(skb, sizeof(*iph));
1421 	skb_set_transport_header(skb, skb_gro_offset(skb));
1422 
1423 	pp = ops->gro_receive(head, skb);
1424 
1425 out_unlock:
1426 	rcu_read_unlock();
1427 
1428 out:
1429 	NAPI_GRO_CB(skb)->flush |= flush;
1430 
1431 	return pp;
1432 }
1433 
1434 static int inet_gro_complete(struct sk_buff *skb)
1435 {
1436 	__be16 newlen = htons(skb->len - skb_network_offset(skb));
1437 	struct iphdr *iph = ip_hdr(skb);
1438 	const struct net_protocol *ops;
1439 	int proto = iph->protocol;
1440 	int err = -ENOSYS;
1441 
1442 	csum_replace2(&iph->check, iph->tot_len, newlen);
1443 	iph->tot_len = newlen;
1444 
1445 	rcu_read_lock();
1446 	ops = rcu_dereference(inet_protos[proto]);
1447 	if (WARN_ON(!ops || !ops->gro_complete))
1448 		goto out_unlock;
1449 
1450 	err = ops->gro_complete(skb);
1451 
1452 out_unlock:
1453 	rcu_read_unlock();
1454 
1455 	return err;
1456 }
1457 
1458 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1459 			 unsigned short type, unsigned char protocol,
1460 			 struct net *net)
1461 {
1462 	struct socket *sock;
1463 	int rc = sock_create_kern(family, type, protocol, &sock);
1464 
1465 	if (rc == 0) {
1466 		*sk = sock->sk;
1467 		(*sk)->sk_allocation = GFP_ATOMIC;
1468 		/*
1469 		 * Unhash it so that IP input processing does not even see it,
1470 		 * we do not wish this socket to see incoming packets.
1471 		 */
1472 		(*sk)->sk_prot->unhash(*sk);
1473 
1474 		sk_change_net(*sk, net);
1475 	}
1476 	return rc;
1477 }
1478 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1479 
1480 unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1481 {
1482 	unsigned long res = 0;
1483 	int i, j;
1484 
1485 	for_each_possible_cpu(i) {
1486 		for (j = 0; j < SNMP_ARRAY_SZ; j++)
1487 			res += *(((unsigned long *) per_cpu_ptr(mib[j], i)) + offt);
1488 	}
1489 	return res;
1490 }
1491 EXPORT_SYMBOL_GPL(snmp_fold_field);
1492 
1493 #if BITS_PER_LONG==32
1494 
1495 u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset)
1496 {
1497 	u64 res = 0;
1498 	int cpu;
1499 
1500 	for_each_possible_cpu(cpu) {
1501 		void *bhptr;
1502 		struct u64_stats_sync *syncp;
1503 		u64 v;
1504 		unsigned int start;
1505 
1506 		bhptr = per_cpu_ptr(mib[0], cpu);
1507 		syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1508 		do {
1509 			start = u64_stats_fetch_begin_bh(syncp);
1510 			v = *(((u64 *) bhptr) + offt);
1511 		} while (u64_stats_fetch_retry_bh(syncp, start));
1512 
1513 		res += v;
1514 	}
1515 	return res;
1516 }
1517 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1518 #endif
1519 
1520 int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align)
1521 {
1522 	BUG_ON(ptr == NULL);
1523 	ptr[0] = __alloc_percpu(mibsize, align);
1524 	if (!ptr[0])
1525 		return -ENOMEM;
1526 #if SNMP_ARRAY_SZ == 2
1527 	ptr[1] = __alloc_percpu(mibsize, align);
1528 	if (!ptr[1]) {
1529 		free_percpu(ptr[0]);
1530 		ptr[0] = NULL;
1531 		return -ENOMEM;
1532 	}
1533 #endif
1534 	return 0;
1535 }
1536 EXPORT_SYMBOL_GPL(snmp_mib_init);
1537 
1538 void snmp_mib_free(void __percpu *ptr[SNMP_ARRAY_SZ])
1539 {
1540 	int i;
1541 
1542 	BUG_ON(ptr == NULL);
1543 	for (i = 0; i < SNMP_ARRAY_SZ; i++) {
1544 		free_percpu(ptr[i]);
1545 		ptr[i] = NULL;
1546 	}
1547 }
1548 EXPORT_SYMBOL_GPL(snmp_mib_free);
1549 
1550 #ifdef CONFIG_IP_MULTICAST
1551 static const struct net_protocol igmp_protocol = {
1552 	.handler =	igmp_rcv,
1553 	.netns_ok =	1,
1554 };
1555 #endif
1556 
1557 static const struct net_protocol tcp_protocol = {
1558 	.early_demux	=	tcp_v4_early_demux,
1559 	.handler	=	tcp_v4_rcv,
1560 	.err_handler	=	tcp_v4_err,
1561 	.gso_send_check	=	tcp_v4_gso_send_check,
1562 	.gso_segment	=	tcp_tso_segment,
1563 	.gro_receive	=	tcp4_gro_receive,
1564 	.gro_complete	=	tcp4_gro_complete,
1565 	.no_policy	=	1,
1566 	.netns_ok	=	1,
1567 };
1568 
1569 static const struct net_protocol udp_protocol = {
1570 	.handler =	udp_rcv,
1571 	.err_handler =	udp_err,
1572 	.gso_send_check = udp4_ufo_send_check,
1573 	.gso_segment = udp4_ufo_fragment,
1574 	.no_policy =	1,
1575 	.netns_ok =	1,
1576 };
1577 
1578 static const struct net_protocol icmp_protocol = {
1579 	.handler =	icmp_rcv,
1580 	.err_handler =	ping_err,
1581 	.no_policy =	1,
1582 	.netns_ok =	1,
1583 };
1584 
1585 static __net_init int ipv4_mib_init_net(struct net *net)
1586 {
1587 	if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1588 			  sizeof(struct tcp_mib),
1589 			  __alignof__(struct tcp_mib)) < 0)
1590 		goto err_tcp_mib;
1591 	if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1592 			  sizeof(struct ipstats_mib),
1593 			  __alignof__(struct ipstats_mib)) < 0)
1594 		goto err_ip_mib;
1595 	if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1596 			  sizeof(struct linux_mib),
1597 			  __alignof__(struct linux_mib)) < 0)
1598 		goto err_net_mib;
1599 	if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1600 			  sizeof(struct udp_mib),
1601 			  __alignof__(struct udp_mib)) < 0)
1602 		goto err_udp_mib;
1603 	if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1604 			  sizeof(struct udp_mib),
1605 			  __alignof__(struct udp_mib)) < 0)
1606 		goto err_udplite_mib;
1607 	if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1608 			  sizeof(struct icmp_mib),
1609 			  __alignof__(struct icmp_mib)) < 0)
1610 		goto err_icmp_mib;
1611 	net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1612 					      GFP_KERNEL);
1613 	if (!net->mib.icmpmsg_statistics)
1614 		goto err_icmpmsg_mib;
1615 
1616 	tcp_mib_init(net);
1617 	return 0;
1618 
1619 err_icmpmsg_mib:
1620 	snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1621 err_icmp_mib:
1622 	snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1623 err_udplite_mib:
1624 	snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1625 err_udp_mib:
1626 	snmp_mib_free((void __percpu **)net->mib.net_statistics);
1627 err_net_mib:
1628 	snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1629 err_ip_mib:
1630 	snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1631 err_tcp_mib:
1632 	return -ENOMEM;
1633 }
1634 
1635 static __net_exit void ipv4_mib_exit_net(struct net *net)
1636 {
1637 	kfree(net->mib.icmpmsg_statistics);
1638 	snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1639 	snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1640 	snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1641 	snmp_mib_free((void __percpu **)net->mib.net_statistics);
1642 	snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1643 	snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1644 }
1645 
1646 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1647 	.init = ipv4_mib_init_net,
1648 	.exit = ipv4_mib_exit_net,
1649 };
1650 
1651 static int __init init_ipv4_mibs(void)
1652 {
1653 	return register_pernet_subsys(&ipv4_mib_ops);
1654 }
1655 
1656 static int ipv4_proc_init(void);
1657 
1658 /*
1659  *	IP protocol layer initialiser
1660  */
1661 
1662 static struct packet_type ip_packet_type __read_mostly = {
1663 	.type = cpu_to_be16(ETH_P_IP),
1664 	.func = ip_rcv,
1665 	.gso_send_check = inet_gso_send_check,
1666 	.gso_segment = inet_gso_segment,
1667 	.gro_receive = inet_gro_receive,
1668 	.gro_complete = inet_gro_complete,
1669 };
1670 
1671 static int __init inet_init(void)
1672 {
1673 	struct sk_buff *dummy_skb;
1674 	struct inet_protosw *q;
1675 	struct list_head *r;
1676 	int rc = -EINVAL;
1677 
1678 	BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1679 
1680 	sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL);
1681 	if (!sysctl_local_reserved_ports)
1682 		goto out;
1683 
1684 	rc = proto_register(&tcp_prot, 1);
1685 	if (rc)
1686 		goto out_free_reserved_ports;
1687 
1688 	rc = proto_register(&udp_prot, 1);
1689 	if (rc)
1690 		goto out_unregister_tcp_proto;
1691 
1692 	rc = proto_register(&raw_prot, 1);
1693 	if (rc)
1694 		goto out_unregister_udp_proto;
1695 
1696 	rc = proto_register(&ping_prot, 1);
1697 	if (rc)
1698 		goto out_unregister_raw_proto;
1699 
1700 	/*
1701 	 *	Tell SOCKET that we are alive...
1702 	 */
1703 
1704 	(void)sock_register(&inet_family_ops);
1705 
1706 #ifdef CONFIG_SYSCTL
1707 	ip_static_sysctl_init();
1708 #endif
1709 
1710 	tcp_prot.sysctl_mem = init_net.ipv4.sysctl_tcp_mem;
1711 
1712 	/*
1713 	 *	Add all the base protocols.
1714 	 */
1715 
1716 	if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1717 		pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1718 	if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1719 		pr_crit("%s: Cannot add UDP protocol\n", __func__);
1720 	if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1721 		pr_crit("%s: Cannot add TCP protocol\n", __func__);
1722 #ifdef CONFIG_IP_MULTICAST
1723 	if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1724 		pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1725 #endif
1726 
1727 	/* Register the socket-side information for inet_create. */
1728 	for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1729 		INIT_LIST_HEAD(r);
1730 
1731 	for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1732 		inet_register_protosw(q);
1733 
1734 	/*
1735 	 *	Set the ARP module up
1736 	 */
1737 
1738 	arp_init();
1739 
1740 	/*
1741 	 *	Set the IP module up
1742 	 */
1743 
1744 	ip_init();
1745 
1746 	tcp_v4_init();
1747 
1748 	/* Setup TCP slab cache for open requests. */
1749 	tcp_init();
1750 
1751 	/* Setup UDP memory threshold */
1752 	udp_init();
1753 
1754 	/* Add UDP-Lite (RFC 3828) */
1755 	udplite4_register();
1756 
1757 	ping_init();
1758 
1759 	/*
1760 	 *	Set the ICMP layer up
1761 	 */
1762 
1763 	if (icmp_init() < 0)
1764 		panic("Failed to create the ICMP control socket.\n");
1765 
1766 	/*
1767 	 *	Initialise the multicast router
1768 	 */
1769 #if defined(CONFIG_IP_MROUTE)
1770 	if (ip_mr_init())
1771 		pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1772 #endif
1773 	/*
1774 	 *	Initialise per-cpu ipv4 mibs
1775 	 */
1776 
1777 	if (init_ipv4_mibs())
1778 		pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1779 
1780 	ipv4_proc_init();
1781 
1782 	ipfrag_init();
1783 
1784 	dev_add_pack(&ip_packet_type);
1785 
1786 	rc = 0;
1787 out:
1788 	return rc;
1789 out_unregister_raw_proto:
1790 	proto_unregister(&raw_prot);
1791 out_unregister_udp_proto:
1792 	proto_unregister(&udp_prot);
1793 out_unregister_tcp_proto:
1794 	proto_unregister(&tcp_prot);
1795 out_free_reserved_ports:
1796 	kfree(sysctl_local_reserved_ports);
1797 	goto out;
1798 }
1799 
1800 fs_initcall(inet_init);
1801 
1802 /* ------------------------------------------------------------------------ */
1803 
1804 #ifdef CONFIG_PROC_FS
1805 static int __init ipv4_proc_init(void)
1806 {
1807 	int rc = 0;
1808 
1809 	if (raw_proc_init())
1810 		goto out_raw;
1811 	if (tcp4_proc_init())
1812 		goto out_tcp;
1813 	if (udp4_proc_init())
1814 		goto out_udp;
1815 	if (ping_proc_init())
1816 		goto out_ping;
1817 	if (ip_misc_proc_init())
1818 		goto out_misc;
1819 out:
1820 	return rc;
1821 out_misc:
1822 	ping_proc_exit();
1823 out_ping:
1824 	udp4_proc_exit();
1825 out_udp:
1826 	tcp4_proc_exit();
1827 out_tcp:
1828 	raw_proc_exit();
1829 out_raw:
1830 	rc = -ENOMEM;
1831 	goto out;
1832 }
1833 
1834 #else /* CONFIG_PROC_FS */
1835 static int __init ipv4_proc_init(void)
1836 {
1837 	return 0;
1838 }
1839 #endif /* CONFIG_PROC_FS */
1840 
1841 MODULE_ALIAS_NETPROTO(PF_INET);
1842 
1843