xref: /openbmc/linux/net/ipv4/af_inet.c (revision 2572f00d)
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/inet_common.h>
115 #include <net/ip_tunnels.h>
116 #include <net/xfrm.h>
117 #include <net/net_namespace.h>
118 #include <net/secure_seq.h>
119 #ifdef CONFIG_IP_MROUTE
120 #include <linux/mroute.h>
121 #endif
122 #include <net/l3mdev.h>
123 
124 
125 /* The inetsw table contains everything that inet_create needs to
126  * build a new socket.
127  */
128 static struct list_head inetsw[SOCK_MAX];
129 static DEFINE_SPINLOCK(inetsw_lock);
130 
131 /* New destruction routine */
132 
133 void inet_sock_destruct(struct sock *sk)
134 {
135 	struct inet_sock *inet = inet_sk(sk);
136 
137 	__skb_queue_purge(&sk->sk_receive_queue);
138 	__skb_queue_purge(&sk->sk_error_queue);
139 
140 	sk_mem_reclaim(sk);
141 
142 	if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
143 		pr_err("Attempt to release TCP socket in state %d %p\n",
144 		       sk->sk_state, sk);
145 		return;
146 	}
147 	if (!sock_flag(sk, SOCK_DEAD)) {
148 		pr_err("Attempt to release alive inet socket %p\n", sk);
149 		return;
150 	}
151 
152 	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
153 	WARN_ON(atomic_read(&sk->sk_wmem_alloc));
154 	WARN_ON(sk->sk_wmem_queued);
155 	WARN_ON(sk->sk_forward_alloc);
156 
157 	kfree(rcu_dereference_protected(inet->inet_opt, 1));
158 	dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
159 	dst_release(sk->sk_rx_dst);
160 	sk_refcnt_debug_dec(sk);
161 }
162 EXPORT_SYMBOL(inet_sock_destruct);
163 
164 /*
165  *	The routines beyond this point handle the behaviour of an AF_INET
166  *	socket object. Mostly it punts to the subprotocols of IP to do
167  *	the work.
168  */
169 
170 /*
171  *	Automatically bind an unbound socket.
172  */
173 
174 static int inet_autobind(struct sock *sk)
175 {
176 	struct inet_sock *inet;
177 	/* We may need to bind the socket. */
178 	lock_sock(sk);
179 	inet = inet_sk(sk);
180 	if (!inet->inet_num) {
181 		if (sk->sk_prot->get_port(sk, 0)) {
182 			release_sock(sk);
183 			return -EAGAIN;
184 		}
185 		inet->inet_sport = htons(inet->inet_num);
186 	}
187 	release_sock(sk);
188 	return 0;
189 }
190 
191 /*
192  *	Move a socket into listening state.
193  */
194 int inet_listen(struct socket *sock, int backlog)
195 {
196 	struct sock *sk = sock->sk;
197 	unsigned char old_state;
198 	int err;
199 
200 	lock_sock(sk);
201 
202 	err = -EINVAL;
203 	if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
204 		goto out;
205 
206 	old_state = sk->sk_state;
207 	if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
208 		goto out;
209 
210 	/* Really, if the socket is already in listen state
211 	 * we can only allow the backlog to be adjusted.
212 	 */
213 	if (old_state != TCP_LISTEN) {
214 		/* Check special setups for testing purpose to enable TFO w/o
215 		 * requiring TCP_FASTOPEN sockopt.
216 		 * Note that only TCP sockets (SOCK_STREAM) will reach here.
217 		 * Also fastopenq may already been allocated because this
218 		 * socket was in TCP_LISTEN state previously but was
219 		 * shutdown() (rather than close()).
220 		 */
221 		if ((sysctl_tcp_fastopen & TFO_SERVER_ENABLE) != 0 &&
222 		    !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
223 			if ((sysctl_tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) != 0)
224 				fastopen_queue_tune(sk, backlog);
225 			else if ((sysctl_tcp_fastopen &
226 				  TFO_SERVER_WO_SOCKOPT2) != 0)
227 				fastopen_queue_tune(sk,
228 				    ((uint)sysctl_tcp_fastopen) >> 16);
229 
230 			tcp_fastopen_init_key_once(true);
231 		}
232 		err = inet_csk_listen_start(sk, backlog);
233 		if (err)
234 			goto out;
235 	}
236 	sk->sk_max_ack_backlog = backlog;
237 	err = 0;
238 
239 out:
240 	release_sock(sk);
241 	return err;
242 }
243 EXPORT_SYMBOL(inet_listen);
244 
245 /*
246  *	Create an inet socket.
247  */
248 
249 static int inet_create(struct net *net, struct socket *sock, int protocol,
250 		       int kern)
251 {
252 	struct sock *sk;
253 	struct inet_protosw *answer;
254 	struct inet_sock *inet;
255 	struct proto *answer_prot;
256 	unsigned char answer_flags;
257 	int try_loading_module = 0;
258 	int err;
259 
260 	if (protocol < 0 || protocol >= IPPROTO_MAX)
261 		return -EINVAL;
262 
263 	sock->state = SS_UNCONNECTED;
264 
265 	/* Look for the requested type/protocol pair. */
266 lookup_protocol:
267 	err = -ESOCKTNOSUPPORT;
268 	rcu_read_lock();
269 	list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
270 
271 		err = 0;
272 		/* Check the non-wild match. */
273 		if (protocol == answer->protocol) {
274 			if (protocol != IPPROTO_IP)
275 				break;
276 		} else {
277 			/* Check for the two wild cases. */
278 			if (IPPROTO_IP == protocol) {
279 				protocol = answer->protocol;
280 				break;
281 			}
282 			if (IPPROTO_IP == answer->protocol)
283 				break;
284 		}
285 		err = -EPROTONOSUPPORT;
286 	}
287 
288 	if (unlikely(err)) {
289 		if (try_loading_module < 2) {
290 			rcu_read_unlock();
291 			/*
292 			 * Be more specific, e.g. net-pf-2-proto-132-type-1
293 			 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
294 			 */
295 			if (++try_loading_module == 1)
296 				request_module("net-pf-%d-proto-%d-type-%d",
297 					       PF_INET, protocol, sock->type);
298 			/*
299 			 * Fall back to generic, e.g. net-pf-2-proto-132
300 			 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
301 			 */
302 			else
303 				request_module("net-pf-%d-proto-%d",
304 					       PF_INET, protocol);
305 			goto lookup_protocol;
306 		} else
307 			goto out_rcu_unlock;
308 	}
309 
310 	err = -EPERM;
311 	if (sock->type == SOCK_RAW && !kern &&
312 	    !ns_capable(net->user_ns, CAP_NET_RAW))
313 		goto out_rcu_unlock;
314 
315 	sock->ops = answer->ops;
316 	answer_prot = answer->prot;
317 	answer_flags = answer->flags;
318 	rcu_read_unlock();
319 
320 	WARN_ON(!answer_prot->slab);
321 
322 	err = -ENOBUFS;
323 	sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
324 	if (!sk)
325 		goto out;
326 
327 	err = 0;
328 	if (INET_PROTOSW_REUSE & answer_flags)
329 		sk->sk_reuse = SK_CAN_REUSE;
330 
331 	inet = inet_sk(sk);
332 	inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
333 
334 	inet->nodefrag = 0;
335 
336 	if (SOCK_RAW == sock->type) {
337 		inet->inet_num = protocol;
338 		if (IPPROTO_RAW == protocol)
339 			inet->hdrincl = 1;
340 	}
341 
342 	if (net->ipv4.sysctl_ip_no_pmtu_disc)
343 		inet->pmtudisc = IP_PMTUDISC_DONT;
344 	else
345 		inet->pmtudisc = IP_PMTUDISC_WANT;
346 
347 	inet->inet_id = 0;
348 
349 	sock_init_data(sock, sk);
350 
351 	sk->sk_destruct	   = inet_sock_destruct;
352 	sk->sk_protocol	   = protocol;
353 	sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
354 
355 	inet->uc_ttl	= -1;
356 	inet->mc_loop	= 1;
357 	inet->mc_ttl	= 1;
358 	inet->mc_all	= 1;
359 	inet->mc_index	= 0;
360 	inet->mc_list	= NULL;
361 	inet->rcv_tos	= 0;
362 
363 	sk_refcnt_debug_inc(sk);
364 
365 	if (inet->inet_num) {
366 		/* It assumes that any protocol which allows
367 		 * the user to assign a number at socket
368 		 * creation time automatically
369 		 * shares.
370 		 */
371 		inet->inet_sport = htons(inet->inet_num);
372 		/* Add to protocol hash chains. */
373 		sk->sk_prot->hash(sk);
374 	}
375 
376 	if (sk->sk_prot->init) {
377 		err = sk->sk_prot->init(sk);
378 		if (err)
379 			sk_common_release(sk);
380 	}
381 out:
382 	return err;
383 out_rcu_unlock:
384 	rcu_read_unlock();
385 	goto out;
386 }
387 
388 
389 /*
390  *	The peer socket should always be NULL (or else). When we call this
391  *	function we are destroying the object and from then on nobody
392  *	should refer to it.
393  */
394 int inet_release(struct socket *sock)
395 {
396 	struct sock *sk = sock->sk;
397 
398 	if (sk) {
399 		long timeout;
400 
401 		/* Applications forget to leave groups before exiting */
402 		ip_mc_drop_socket(sk);
403 
404 		/* If linger is set, we don't return until the close
405 		 * is complete.  Otherwise we return immediately. The
406 		 * actually closing is done the same either way.
407 		 *
408 		 * If the close is due to the process exiting, we never
409 		 * linger..
410 		 */
411 		timeout = 0;
412 		if (sock_flag(sk, SOCK_LINGER) &&
413 		    !(current->flags & PF_EXITING))
414 			timeout = sk->sk_lingertime;
415 		sock->sk = NULL;
416 		sk->sk_prot->close(sk, timeout);
417 	}
418 	return 0;
419 }
420 EXPORT_SYMBOL(inet_release);
421 
422 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
423 {
424 	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
425 	struct sock *sk = sock->sk;
426 	struct inet_sock *inet = inet_sk(sk);
427 	struct net *net = sock_net(sk);
428 	unsigned short snum;
429 	int chk_addr_ret;
430 	u32 tb_id = RT_TABLE_LOCAL;
431 	int err;
432 
433 	/* If the socket has its own bind function then use it. (RAW) */
434 	if (sk->sk_prot->bind) {
435 		err = sk->sk_prot->bind(sk, uaddr, addr_len);
436 		goto out;
437 	}
438 	err = -EINVAL;
439 	if (addr_len < sizeof(struct sockaddr_in))
440 		goto out;
441 
442 	if (addr->sin_family != AF_INET) {
443 		/* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
444 		 * only if s_addr is INADDR_ANY.
445 		 */
446 		err = -EAFNOSUPPORT;
447 		if (addr->sin_family != AF_UNSPEC ||
448 		    addr->sin_addr.s_addr != htonl(INADDR_ANY))
449 			goto out;
450 	}
451 
452 	tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
453 	chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
454 
455 	/* Not specified by any standard per-se, however it breaks too
456 	 * many applications when removed.  It is unfortunate since
457 	 * allowing applications to make a non-local bind solves
458 	 * several problems with systems using dynamic addressing.
459 	 * (ie. your servers still start up even if your ISDN link
460 	 *  is temporarily down)
461 	 */
462 	err = -EADDRNOTAVAIL;
463 	if (!net->ipv4.sysctl_ip_nonlocal_bind &&
464 	    !(inet->freebind || inet->transparent) &&
465 	    addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
466 	    chk_addr_ret != RTN_LOCAL &&
467 	    chk_addr_ret != RTN_MULTICAST &&
468 	    chk_addr_ret != RTN_BROADCAST)
469 		goto out;
470 
471 	snum = ntohs(addr->sin_port);
472 	err = -EACCES;
473 	if (snum && snum < PROT_SOCK &&
474 	    !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
475 		goto out;
476 
477 	/*      We keep a pair of addresses. rcv_saddr is the one
478 	 *      used by hash lookups, and saddr is used for transmit.
479 	 *
480 	 *      In the BSD API these are the same except where it
481 	 *      would be illegal to use them (multicast/broadcast) in
482 	 *      which case the sending device address is used.
483 	 */
484 	lock_sock(sk);
485 
486 	/* Check these errors (active socket, double bind). */
487 	err = -EINVAL;
488 	if (sk->sk_state != TCP_CLOSE || inet->inet_num)
489 		goto out_release_sock;
490 
491 	inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
492 	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
493 		inet->inet_saddr = 0;  /* Use device */
494 
495 	/* Make sure we are allowed to bind here. */
496 	if ((snum || !inet->bind_address_no_port) &&
497 	    sk->sk_prot->get_port(sk, snum)) {
498 		inet->inet_saddr = inet->inet_rcv_saddr = 0;
499 		err = -EADDRINUSE;
500 		goto out_release_sock;
501 	}
502 
503 	if (inet->inet_rcv_saddr)
504 		sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
505 	if (snum)
506 		sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
507 	inet->inet_sport = htons(inet->inet_num);
508 	inet->inet_daddr = 0;
509 	inet->inet_dport = 0;
510 	sk_dst_reset(sk);
511 	err = 0;
512 out_release_sock:
513 	release_sock(sk);
514 out:
515 	return err;
516 }
517 EXPORT_SYMBOL(inet_bind);
518 
519 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
520 		       int addr_len, int flags)
521 {
522 	struct sock *sk = sock->sk;
523 
524 	if (addr_len < sizeof(uaddr->sa_family))
525 		return -EINVAL;
526 	if (uaddr->sa_family == AF_UNSPEC)
527 		return sk->sk_prot->disconnect(sk, flags);
528 
529 	if (!inet_sk(sk)->inet_num && inet_autobind(sk))
530 		return -EAGAIN;
531 	return sk->sk_prot->connect(sk, uaddr, addr_len);
532 }
533 EXPORT_SYMBOL(inet_dgram_connect);
534 
535 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
536 {
537 	DEFINE_WAIT(wait);
538 
539 	prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
540 	sk->sk_write_pending += writebias;
541 
542 	/* Basic assumption: if someone sets sk->sk_err, he _must_
543 	 * change state of the socket from TCP_SYN_*.
544 	 * Connect() does not allow to get error notifications
545 	 * without closing the socket.
546 	 */
547 	while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
548 		release_sock(sk);
549 		timeo = schedule_timeout(timeo);
550 		lock_sock(sk);
551 		if (signal_pending(current) || !timeo)
552 			break;
553 		prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
554 	}
555 	finish_wait(sk_sleep(sk), &wait);
556 	sk->sk_write_pending -= writebias;
557 	return timeo;
558 }
559 
560 /*
561  *	Connect to a remote host. There is regrettably still a little
562  *	TCP 'magic' in here.
563  */
564 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
565 			  int addr_len, int flags)
566 {
567 	struct sock *sk = sock->sk;
568 	int err;
569 	long timeo;
570 
571 	if (addr_len < sizeof(uaddr->sa_family))
572 		return -EINVAL;
573 
574 	if (uaddr->sa_family == AF_UNSPEC) {
575 		err = sk->sk_prot->disconnect(sk, flags);
576 		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
577 		goto out;
578 	}
579 
580 	switch (sock->state) {
581 	default:
582 		err = -EINVAL;
583 		goto out;
584 	case SS_CONNECTED:
585 		err = -EISCONN;
586 		goto out;
587 	case SS_CONNECTING:
588 		err = -EALREADY;
589 		/* Fall out of switch with err, set for this state */
590 		break;
591 	case SS_UNCONNECTED:
592 		err = -EISCONN;
593 		if (sk->sk_state != TCP_CLOSE)
594 			goto out;
595 
596 		err = sk->sk_prot->connect(sk, uaddr, addr_len);
597 		if (err < 0)
598 			goto out;
599 
600 		sock->state = SS_CONNECTING;
601 
602 		/* Just entered SS_CONNECTING state; the only
603 		 * difference is that return value in non-blocking
604 		 * case is EINPROGRESS, rather than EALREADY.
605 		 */
606 		err = -EINPROGRESS;
607 		break;
608 	}
609 
610 	timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
611 
612 	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
613 		int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
614 				tcp_sk(sk)->fastopen_req &&
615 				tcp_sk(sk)->fastopen_req->data ? 1 : 0;
616 
617 		/* Error code is set above */
618 		if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
619 			goto out;
620 
621 		err = sock_intr_errno(timeo);
622 		if (signal_pending(current))
623 			goto out;
624 	}
625 
626 	/* Connection was closed by RST, timeout, ICMP error
627 	 * or another process disconnected us.
628 	 */
629 	if (sk->sk_state == TCP_CLOSE)
630 		goto sock_error;
631 
632 	/* sk->sk_err may be not zero now, if RECVERR was ordered by user
633 	 * and error was received after socket entered established state.
634 	 * Hence, it is handled normally after connect() return successfully.
635 	 */
636 
637 	sock->state = SS_CONNECTED;
638 	err = 0;
639 out:
640 	return err;
641 
642 sock_error:
643 	err = sock_error(sk) ? : -ECONNABORTED;
644 	sock->state = SS_UNCONNECTED;
645 	if (sk->sk_prot->disconnect(sk, flags))
646 		sock->state = SS_DISCONNECTING;
647 	goto out;
648 }
649 EXPORT_SYMBOL(__inet_stream_connect);
650 
651 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
652 			int addr_len, int flags)
653 {
654 	int err;
655 
656 	lock_sock(sock->sk);
657 	err = __inet_stream_connect(sock, uaddr, addr_len, flags);
658 	release_sock(sock->sk);
659 	return err;
660 }
661 EXPORT_SYMBOL(inet_stream_connect);
662 
663 /*
664  *	Accept a pending connection. The TCP layer now gives BSD semantics.
665  */
666 
667 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
668 {
669 	struct sock *sk1 = sock->sk;
670 	int err = -EINVAL;
671 	struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
672 
673 	if (!sk2)
674 		goto do_err;
675 
676 	lock_sock(sk2);
677 
678 	sock_rps_record_flow(sk2);
679 	WARN_ON(!((1 << sk2->sk_state) &
680 		  (TCPF_ESTABLISHED | TCPF_SYN_RECV |
681 		  TCPF_CLOSE_WAIT | TCPF_CLOSE)));
682 
683 	sock_graft(sk2, newsock);
684 
685 	newsock->state = SS_CONNECTED;
686 	err = 0;
687 	release_sock(sk2);
688 do_err:
689 	return err;
690 }
691 EXPORT_SYMBOL(inet_accept);
692 
693 
694 /*
695  *	This does both peername and sockname.
696  */
697 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
698 			int *uaddr_len, int peer)
699 {
700 	struct sock *sk		= sock->sk;
701 	struct inet_sock *inet	= inet_sk(sk);
702 	DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
703 
704 	sin->sin_family = AF_INET;
705 	if (peer) {
706 		if (!inet->inet_dport ||
707 		    (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
708 		     peer == 1))
709 			return -ENOTCONN;
710 		sin->sin_port = inet->inet_dport;
711 		sin->sin_addr.s_addr = inet->inet_daddr;
712 	} else {
713 		__be32 addr = inet->inet_rcv_saddr;
714 		if (!addr)
715 			addr = inet->inet_saddr;
716 		sin->sin_port = inet->inet_sport;
717 		sin->sin_addr.s_addr = addr;
718 	}
719 	memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
720 	*uaddr_len = sizeof(*sin);
721 	return 0;
722 }
723 EXPORT_SYMBOL(inet_getname);
724 
725 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
726 {
727 	struct sock *sk = sock->sk;
728 
729 	sock_rps_record_flow(sk);
730 
731 	/* We may need to bind the socket. */
732 	if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
733 	    inet_autobind(sk))
734 		return -EAGAIN;
735 
736 	return sk->sk_prot->sendmsg(sk, msg, size);
737 }
738 EXPORT_SYMBOL(inet_sendmsg);
739 
740 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
741 		      size_t size, int flags)
742 {
743 	struct sock *sk = sock->sk;
744 
745 	sock_rps_record_flow(sk);
746 
747 	/* We may need to bind the socket. */
748 	if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
749 	    inet_autobind(sk))
750 		return -EAGAIN;
751 
752 	if (sk->sk_prot->sendpage)
753 		return sk->sk_prot->sendpage(sk, page, offset, size, flags);
754 	return sock_no_sendpage(sock, page, offset, size, flags);
755 }
756 EXPORT_SYMBOL(inet_sendpage);
757 
758 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
759 		 int flags)
760 {
761 	struct sock *sk = sock->sk;
762 	int addr_len = 0;
763 	int err;
764 
765 	sock_rps_record_flow(sk);
766 
767 	err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
768 				   flags & ~MSG_DONTWAIT, &addr_len);
769 	if (err >= 0)
770 		msg->msg_namelen = addr_len;
771 	return err;
772 }
773 EXPORT_SYMBOL(inet_recvmsg);
774 
775 int inet_shutdown(struct socket *sock, int how)
776 {
777 	struct sock *sk = sock->sk;
778 	int err = 0;
779 
780 	/* This should really check to make sure
781 	 * the socket is a TCP socket. (WHY AC...)
782 	 */
783 	how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
784 		       1->2 bit 2 snds.
785 		       2->3 */
786 	if ((how & ~SHUTDOWN_MASK) || !how)	/* MAXINT->0 */
787 		return -EINVAL;
788 
789 	lock_sock(sk);
790 	if (sock->state == SS_CONNECTING) {
791 		if ((1 << sk->sk_state) &
792 		    (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
793 			sock->state = SS_DISCONNECTING;
794 		else
795 			sock->state = SS_CONNECTED;
796 	}
797 
798 	switch (sk->sk_state) {
799 	case TCP_CLOSE:
800 		err = -ENOTCONN;
801 		/* Hack to wake up other listeners, who can poll for
802 		   POLLHUP, even on eg. unconnected UDP sockets -- RR */
803 	default:
804 		sk->sk_shutdown |= how;
805 		if (sk->sk_prot->shutdown)
806 			sk->sk_prot->shutdown(sk, how);
807 		break;
808 
809 	/* Remaining two branches are temporary solution for missing
810 	 * close() in multithreaded environment. It is _not_ a good idea,
811 	 * but we have no choice until close() is repaired at VFS level.
812 	 */
813 	case TCP_LISTEN:
814 		if (!(how & RCV_SHUTDOWN))
815 			break;
816 		/* Fall through */
817 	case TCP_SYN_SENT:
818 		err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
819 		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
820 		break;
821 	}
822 
823 	/* Wake up anyone sleeping in poll. */
824 	sk->sk_state_change(sk);
825 	release_sock(sk);
826 	return err;
827 }
828 EXPORT_SYMBOL(inet_shutdown);
829 
830 /*
831  *	ioctl() calls you can issue on an INET socket. Most of these are
832  *	device configuration and stuff and very rarely used. Some ioctls
833  *	pass on to the socket itself.
834  *
835  *	NOTE: I like the idea of a module for the config stuff. ie ifconfig
836  *	loads the devconfigure module does its configuring and unloads it.
837  *	There's a good 20K of config code hanging around the kernel.
838  */
839 
840 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
841 {
842 	struct sock *sk = sock->sk;
843 	int err = 0;
844 	struct net *net = sock_net(sk);
845 
846 	switch (cmd) {
847 	case SIOCGSTAMP:
848 		err = sock_get_timestamp(sk, (struct timeval __user *)arg);
849 		break;
850 	case SIOCGSTAMPNS:
851 		err = sock_get_timestampns(sk, (struct timespec __user *)arg);
852 		break;
853 	case SIOCADDRT:
854 	case SIOCDELRT:
855 	case SIOCRTMSG:
856 		err = ip_rt_ioctl(net, cmd, (void __user *)arg);
857 		break;
858 	case SIOCDARP:
859 	case SIOCGARP:
860 	case SIOCSARP:
861 		err = arp_ioctl(net, cmd, (void __user *)arg);
862 		break;
863 	case SIOCGIFADDR:
864 	case SIOCSIFADDR:
865 	case SIOCGIFBRDADDR:
866 	case SIOCSIFBRDADDR:
867 	case SIOCGIFNETMASK:
868 	case SIOCSIFNETMASK:
869 	case SIOCGIFDSTADDR:
870 	case SIOCSIFDSTADDR:
871 	case SIOCSIFPFLAGS:
872 	case SIOCGIFPFLAGS:
873 	case SIOCSIFFLAGS:
874 		err = devinet_ioctl(net, cmd, (void __user *)arg);
875 		break;
876 	default:
877 		if (sk->sk_prot->ioctl)
878 			err = sk->sk_prot->ioctl(sk, cmd, arg);
879 		else
880 			err = -ENOIOCTLCMD;
881 		break;
882 	}
883 	return err;
884 }
885 EXPORT_SYMBOL(inet_ioctl);
886 
887 #ifdef CONFIG_COMPAT
888 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
889 {
890 	struct sock *sk = sock->sk;
891 	int err = -ENOIOCTLCMD;
892 
893 	if (sk->sk_prot->compat_ioctl)
894 		err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
895 
896 	return err;
897 }
898 #endif
899 
900 const struct proto_ops inet_stream_ops = {
901 	.family		   = PF_INET,
902 	.owner		   = THIS_MODULE,
903 	.release	   = inet_release,
904 	.bind		   = inet_bind,
905 	.connect	   = inet_stream_connect,
906 	.socketpair	   = sock_no_socketpair,
907 	.accept		   = inet_accept,
908 	.getname	   = inet_getname,
909 	.poll		   = tcp_poll,
910 	.ioctl		   = inet_ioctl,
911 	.listen		   = inet_listen,
912 	.shutdown	   = inet_shutdown,
913 	.setsockopt	   = sock_common_setsockopt,
914 	.getsockopt	   = sock_common_getsockopt,
915 	.sendmsg	   = inet_sendmsg,
916 	.recvmsg	   = inet_recvmsg,
917 	.mmap		   = sock_no_mmap,
918 	.sendpage	   = inet_sendpage,
919 	.splice_read	   = tcp_splice_read,
920 #ifdef CONFIG_COMPAT
921 	.compat_setsockopt = compat_sock_common_setsockopt,
922 	.compat_getsockopt = compat_sock_common_getsockopt,
923 	.compat_ioctl	   = inet_compat_ioctl,
924 #endif
925 };
926 EXPORT_SYMBOL(inet_stream_ops);
927 
928 const struct proto_ops inet_dgram_ops = {
929 	.family		   = PF_INET,
930 	.owner		   = THIS_MODULE,
931 	.release	   = inet_release,
932 	.bind		   = inet_bind,
933 	.connect	   = inet_dgram_connect,
934 	.socketpair	   = sock_no_socketpair,
935 	.accept		   = sock_no_accept,
936 	.getname	   = inet_getname,
937 	.poll		   = udp_poll,
938 	.ioctl		   = inet_ioctl,
939 	.listen		   = sock_no_listen,
940 	.shutdown	   = inet_shutdown,
941 	.setsockopt	   = sock_common_setsockopt,
942 	.getsockopt	   = sock_common_getsockopt,
943 	.sendmsg	   = inet_sendmsg,
944 	.recvmsg	   = inet_recvmsg,
945 	.mmap		   = sock_no_mmap,
946 	.sendpage	   = inet_sendpage,
947 #ifdef CONFIG_COMPAT
948 	.compat_setsockopt = compat_sock_common_setsockopt,
949 	.compat_getsockopt = compat_sock_common_getsockopt,
950 	.compat_ioctl	   = inet_compat_ioctl,
951 #endif
952 };
953 EXPORT_SYMBOL(inet_dgram_ops);
954 
955 /*
956  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
957  * udp_poll
958  */
959 static const struct proto_ops inet_sockraw_ops = {
960 	.family		   = PF_INET,
961 	.owner		   = THIS_MODULE,
962 	.release	   = inet_release,
963 	.bind		   = inet_bind,
964 	.connect	   = inet_dgram_connect,
965 	.socketpair	   = sock_no_socketpair,
966 	.accept		   = sock_no_accept,
967 	.getname	   = inet_getname,
968 	.poll		   = datagram_poll,
969 	.ioctl		   = inet_ioctl,
970 	.listen		   = sock_no_listen,
971 	.shutdown	   = inet_shutdown,
972 	.setsockopt	   = sock_common_setsockopt,
973 	.getsockopt	   = sock_common_getsockopt,
974 	.sendmsg	   = inet_sendmsg,
975 	.recvmsg	   = inet_recvmsg,
976 	.mmap		   = sock_no_mmap,
977 	.sendpage	   = inet_sendpage,
978 #ifdef CONFIG_COMPAT
979 	.compat_setsockopt = compat_sock_common_setsockopt,
980 	.compat_getsockopt = compat_sock_common_getsockopt,
981 	.compat_ioctl	   = inet_compat_ioctl,
982 #endif
983 };
984 
985 static const struct net_proto_family inet_family_ops = {
986 	.family = PF_INET,
987 	.create = inet_create,
988 	.owner	= THIS_MODULE,
989 };
990 
991 /* Upon startup we insert all the elements in inetsw_array[] into
992  * the linked list inetsw.
993  */
994 static struct inet_protosw inetsw_array[] =
995 {
996 	{
997 		.type =       SOCK_STREAM,
998 		.protocol =   IPPROTO_TCP,
999 		.prot =       &tcp_prot,
1000 		.ops =        &inet_stream_ops,
1001 		.flags =      INET_PROTOSW_PERMANENT |
1002 			      INET_PROTOSW_ICSK,
1003 	},
1004 
1005 	{
1006 		.type =       SOCK_DGRAM,
1007 		.protocol =   IPPROTO_UDP,
1008 		.prot =       &udp_prot,
1009 		.ops =        &inet_dgram_ops,
1010 		.flags =      INET_PROTOSW_PERMANENT,
1011        },
1012 
1013        {
1014 		.type =       SOCK_DGRAM,
1015 		.protocol =   IPPROTO_ICMP,
1016 		.prot =       &ping_prot,
1017 		.ops =        &inet_dgram_ops,
1018 		.flags =      INET_PROTOSW_REUSE,
1019        },
1020 
1021        {
1022 	       .type =       SOCK_RAW,
1023 	       .protocol =   IPPROTO_IP,	/* wild card */
1024 	       .prot =       &raw_prot,
1025 	       .ops =        &inet_sockraw_ops,
1026 	       .flags =      INET_PROTOSW_REUSE,
1027        }
1028 };
1029 
1030 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1031 
1032 void inet_register_protosw(struct inet_protosw *p)
1033 {
1034 	struct list_head *lh;
1035 	struct inet_protosw *answer;
1036 	int protocol = p->protocol;
1037 	struct list_head *last_perm;
1038 
1039 	spin_lock_bh(&inetsw_lock);
1040 
1041 	if (p->type >= SOCK_MAX)
1042 		goto out_illegal;
1043 
1044 	/* If we are trying to override a permanent protocol, bail. */
1045 	last_perm = &inetsw[p->type];
1046 	list_for_each(lh, &inetsw[p->type]) {
1047 		answer = list_entry(lh, struct inet_protosw, list);
1048 		/* Check only the non-wild match. */
1049 		if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1050 			break;
1051 		if (protocol == answer->protocol)
1052 			goto out_permanent;
1053 		last_perm = lh;
1054 	}
1055 
1056 	/* Add the new entry after the last permanent entry if any, so that
1057 	 * the new entry does not override a permanent entry when matched with
1058 	 * a wild-card protocol. But it is allowed to override any existing
1059 	 * non-permanent entry.  This means that when we remove this entry, the
1060 	 * system automatically returns to the old behavior.
1061 	 */
1062 	list_add_rcu(&p->list, last_perm);
1063 out:
1064 	spin_unlock_bh(&inetsw_lock);
1065 
1066 	return;
1067 
1068 out_permanent:
1069 	pr_err("Attempt to override permanent protocol %d\n", protocol);
1070 	goto out;
1071 
1072 out_illegal:
1073 	pr_err("Ignoring attempt to register invalid socket type %d\n",
1074 	       p->type);
1075 	goto out;
1076 }
1077 EXPORT_SYMBOL(inet_register_protosw);
1078 
1079 void inet_unregister_protosw(struct inet_protosw *p)
1080 {
1081 	if (INET_PROTOSW_PERMANENT & p->flags) {
1082 		pr_err("Attempt to unregister permanent protocol %d\n",
1083 		       p->protocol);
1084 	} else {
1085 		spin_lock_bh(&inetsw_lock);
1086 		list_del_rcu(&p->list);
1087 		spin_unlock_bh(&inetsw_lock);
1088 
1089 		synchronize_net();
1090 	}
1091 }
1092 EXPORT_SYMBOL(inet_unregister_protosw);
1093 
1094 /*
1095  *      Shall we try to damage output packets if routing dev changes?
1096  */
1097 
1098 int sysctl_ip_dynaddr __read_mostly;
1099 
1100 static int inet_sk_reselect_saddr(struct sock *sk)
1101 {
1102 	struct inet_sock *inet = inet_sk(sk);
1103 	__be32 old_saddr = inet->inet_saddr;
1104 	__be32 daddr = inet->inet_daddr;
1105 	struct flowi4 *fl4;
1106 	struct rtable *rt;
1107 	__be32 new_saddr;
1108 	struct ip_options_rcu *inet_opt;
1109 
1110 	inet_opt = rcu_dereference_protected(inet->inet_opt,
1111 					     sock_owned_by_user(sk));
1112 	if (inet_opt && inet_opt->opt.srr)
1113 		daddr = inet_opt->opt.faddr;
1114 
1115 	/* Query new route. */
1116 	fl4 = &inet->cork.fl.u.ip4;
1117 	rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1118 			      sk->sk_bound_dev_if, sk->sk_protocol,
1119 			      inet->inet_sport, inet->inet_dport, sk);
1120 	if (IS_ERR(rt))
1121 		return PTR_ERR(rt);
1122 
1123 	sk_setup_caps(sk, &rt->dst);
1124 
1125 	new_saddr = fl4->saddr;
1126 
1127 	if (new_saddr == old_saddr)
1128 		return 0;
1129 
1130 	if (sysctl_ip_dynaddr > 1) {
1131 		pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1132 			__func__, &old_saddr, &new_saddr);
1133 	}
1134 
1135 	inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1136 
1137 	/*
1138 	 * XXX The only one ugly spot where we need to
1139 	 * XXX really change the sockets identity after
1140 	 * XXX it has entered the hashes. -DaveM
1141 	 *
1142 	 * Besides that, it does not check for connection
1143 	 * uniqueness. Wait for troubles.
1144 	 */
1145 	__sk_prot_rehash(sk);
1146 	return 0;
1147 }
1148 
1149 int inet_sk_rebuild_header(struct sock *sk)
1150 {
1151 	struct inet_sock *inet = inet_sk(sk);
1152 	struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1153 	__be32 daddr;
1154 	struct ip_options_rcu *inet_opt;
1155 	struct flowi4 *fl4;
1156 	int err;
1157 
1158 	/* Route is OK, nothing to do. */
1159 	if (rt)
1160 		return 0;
1161 
1162 	/* Reroute. */
1163 	rcu_read_lock();
1164 	inet_opt = rcu_dereference(inet->inet_opt);
1165 	daddr = inet->inet_daddr;
1166 	if (inet_opt && inet_opt->opt.srr)
1167 		daddr = inet_opt->opt.faddr;
1168 	rcu_read_unlock();
1169 	fl4 = &inet->cork.fl.u.ip4;
1170 	rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1171 				   inet->inet_dport, inet->inet_sport,
1172 				   sk->sk_protocol, RT_CONN_FLAGS(sk),
1173 				   sk->sk_bound_dev_if);
1174 	if (!IS_ERR(rt)) {
1175 		err = 0;
1176 		sk_setup_caps(sk, &rt->dst);
1177 	} else {
1178 		err = PTR_ERR(rt);
1179 
1180 		/* Routing failed... */
1181 		sk->sk_route_caps = 0;
1182 		/*
1183 		 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1184 		 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1185 		 */
1186 		if (!sysctl_ip_dynaddr ||
1187 		    sk->sk_state != TCP_SYN_SENT ||
1188 		    (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1189 		    (err = inet_sk_reselect_saddr(sk)) != 0)
1190 			sk->sk_err_soft = -err;
1191 	}
1192 
1193 	return err;
1194 }
1195 EXPORT_SYMBOL(inet_sk_rebuild_header);
1196 
1197 static struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1198 					netdev_features_t features)
1199 {
1200 	struct sk_buff *segs = ERR_PTR(-EINVAL);
1201 	const struct net_offload *ops;
1202 	unsigned int offset = 0;
1203 	bool udpfrag, encap;
1204 	struct iphdr *iph;
1205 	int proto;
1206 	int nhoff;
1207 	int ihl;
1208 	int id;
1209 
1210 	if (unlikely(skb_shinfo(skb)->gso_type &
1211 		     ~(SKB_GSO_TCPV4 |
1212 		       SKB_GSO_UDP |
1213 		       SKB_GSO_DODGY |
1214 		       SKB_GSO_TCP_ECN |
1215 		       SKB_GSO_GRE |
1216 		       SKB_GSO_GRE_CSUM |
1217 		       SKB_GSO_IPIP |
1218 		       SKB_GSO_SIT |
1219 		       SKB_GSO_TCPV6 |
1220 		       SKB_GSO_UDP_TUNNEL |
1221 		       SKB_GSO_UDP_TUNNEL_CSUM |
1222 		       SKB_GSO_TUNNEL_REMCSUM |
1223 		       0)))
1224 		goto out;
1225 
1226 	skb_reset_network_header(skb);
1227 	nhoff = skb_network_header(skb) - skb_mac_header(skb);
1228 	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1229 		goto out;
1230 
1231 	iph = ip_hdr(skb);
1232 	ihl = iph->ihl * 4;
1233 	if (ihl < sizeof(*iph))
1234 		goto out;
1235 
1236 	id = ntohs(iph->id);
1237 	proto = iph->protocol;
1238 
1239 	/* Warning: after this point, iph might be no longer valid */
1240 	if (unlikely(!pskb_may_pull(skb, ihl)))
1241 		goto out;
1242 	__skb_pull(skb, ihl);
1243 
1244 	encap = SKB_GSO_CB(skb)->encap_level > 0;
1245 	if (encap)
1246 		features &= skb->dev->hw_enc_features;
1247 	SKB_GSO_CB(skb)->encap_level += ihl;
1248 
1249 	skb_reset_transport_header(skb);
1250 
1251 	segs = ERR_PTR(-EPROTONOSUPPORT);
1252 
1253 	if (skb->encapsulation &&
1254 	    skb_shinfo(skb)->gso_type & (SKB_GSO_SIT|SKB_GSO_IPIP))
1255 		udpfrag = proto == IPPROTO_UDP && encap;
1256 	else
1257 		udpfrag = proto == IPPROTO_UDP && !skb->encapsulation;
1258 
1259 	ops = rcu_dereference(inet_offloads[proto]);
1260 	if (likely(ops && ops->callbacks.gso_segment))
1261 		segs = ops->callbacks.gso_segment(skb, features);
1262 
1263 	if (IS_ERR_OR_NULL(segs))
1264 		goto out;
1265 
1266 	skb = segs;
1267 	do {
1268 		iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1269 		if (udpfrag) {
1270 			iph->id = htons(id);
1271 			iph->frag_off = htons(offset >> 3);
1272 			if (skb->next)
1273 				iph->frag_off |= htons(IP_MF);
1274 			offset += skb->len - nhoff - ihl;
1275 		} else {
1276 			iph->id = htons(id++);
1277 		}
1278 		iph->tot_len = htons(skb->len - nhoff);
1279 		ip_send_check(iph);
1280 		if (encap)
1281 			skb_reset_inner_headers(skb);
1282 		skb->network_header = (u8 *)iph - skb->head;
1283 	} while ((skb = skb->next));
1284 
1285 out:
1286 	return segs;
1287 }
1288 
1289 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1290 					 struct sk_buff *skb)
1291 {
1292 	const struct net_offload *ops;
1293 	struct sk_buff **pp = NULL;
1294 	struct sk_buff *p;
1295 	const struct iphdr *iph;
1296 	unsigned int hlen;
1297 	unsigned int off;
1298 	unsigned int id;
1299 	int flush = 1;
1300 	int proto;
1301 
1302 	off = skb_gro_offset(skb);
1303 	hlen = off + sizeof(*iph);
1304 	iph = skb_gro_header_fast(skb, off);
1305 	if (skb_gro_header_hard(skb, hlen)) {
1306 		iph = skb_gro_header_slow(skb, hlen, off);
1307 		if (unlikely(!iph))
1308 			goto out;
1309 	}
1310 
1311 	proto = iph->protocol;
1312 
1313 	rcu_read_lock();
1314 	ops = rcu_dereference(inet_offloads[proto]);
1315 	if (!ops || !ops->callbacks.gro_receive)
1316 		goto out_unlock;
1317 
1318 	if (*(u8 *)iph != 0x45)
1319 		goto out_unlock;
1320 
1321 	if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1322 		goto out_unlock;
1323 
1324 	id = ntohl(*(__be32 *)&iph->id);
1325 	flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1326 	id >>= 16;
1327 
1328 	for (p = *head; p; p = p->next) {
1329 		struct iphdr *iph2;
1330 
1331 		if (!NAPI_GRO_CB(p)->same_flow)
1332 			continue;
1333 
1334 		iph2 = (struct iphdr *)(p->data + off);
1335 		/* The above works because, with the exception of the top
1336 		 * (inner most) layer, we only aggregate pkts with the same
1337 		 * hdr length so all the hdrs we'll need to verify will start
1338 		 * at the same offset.
1339 		 */
1340 		if ((iph->protocol ^ iph2->protocol) |
1341 		    ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1342 		    ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1343 			NAPI_GRO_CB(p)->same_flow = 0;
1344 			continue;
1345 		}
1346 
1347 		/* All fields must match except length and checksum. */
1348 		NAPI_GRO_CB(p)->flush |=
1349 			(iph->ttl ^ iph2->ttl) |
1350 			(iph->tos ^ iph2->tos) |
1351 			((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1352 
1353 		/* Save the IP ID check to be included later when we get to
1354 		 * the transport layer so only the inner most IP ID is checked.
1355 		 * This is because some GSO/TSO implementations do not
1356 		 * correctly increment the IP ID for the outer hdrs.
1357 		 */
1358 		NAPI_GRO_CB(p)->flush_id =
1359 			    ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1360 		NAPI_GRO_CB(p)->flush |= flush;
1361 	}
1362 
1363 	NAPI_GRO_CB(skb)->flush |= flush;
1364 	skb_set_network_header(skb, off);
1365 	/* The above will be needed by the transport layer if there is one
1366 	 * immediately following this IP hdr.
1367 	 */
1368 
1369 	/* Note : No need to call skb_gro_postpull_rcsum() here,
1370 	 * as we already checked checksum over ipv4 header was 0
1371 	 */
1372 	skb_gro_pull(skb, sizeof(*iph));
1373 	skb_set_transport_header(skb, skb_gro_offset(skb));
1374 
1375 	pp = ops->callbacks.gro_receive(head, skb);
1376 
1377 out_unlock:
1378 	rcu_read_unlock();
1379 
1380 out:
1381 	NAPI_GRO_CB(skb)->flush |= flush;
1382 
1383 	return pp;
1384 }
1385 
1386 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1387 {
1388 	if (sk->sk_family == AF_INET)
1389 		return ip_recv_error(sk, msg, len, addr_len);
1390 #if IS_ENABLED(CONFIG_IPV6)
1391 	if (sk->sk_family == AF_INET6)
1392 		return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1393 #endif
1394 	return -EINVAL;
1395 }
1396 
1397 static int inet_gro_complete(struct sk_buff *skb, int nhoff)
1398 {
1399 	__be16 newlen = htons(skb->len - nhoff);
1400 	struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1401 	const struct net_offload *ops;
1402 	int proto = iph->protocol;
1403 	int err = -ENOSYS;
1404 
1405 	if (skb->encapsulation)
1406 		skb_set_inner_network_header(skb, nhoff);
1407 
1408 	csum_replace2(&iph->check, iph->tot_len, newlen);
1409 	iph->tot_len = newlen;
1410 
1411 	rcu_read_lock();
1412 	ops = rcu_dereference(inet_offloads[proto]);
1413 	if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1414 		goto out_unlock;
1415 
1416 	/* Only need to add sizeof(*iph) to get to the next hdr below
1417 	 * because any hdr with option will have been flushed in
1418 	 * inet_gro_receive().
1419 	 */
1420 	err = ops->callbacks.gro_complete(skb, nhoff + sizeof(*iph));
1421 
1422 out_unlock:
1423 	rcu_read_unlock();
1424 
1425 	return err;
1426 }
1427 
1428 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1429 			 unsigned short type, unsigned char protocol,
1430 			 struct net *net)
1431 {
1432 	struct socket *sock;
1433 	int rc = sock_create_kern(net, family, type, protocol, &sock);
1434 
1435 	if (rc == 0) {
1436 		*sk = sock->sk;
1437 		(*sk)->sk_allocation = GFP_ATOMIC;
1438 		/*
1439 		 * Unhash it so that IP input processing does not even see it,
1440 		 * we do not wish this socket to see incoming packets.
1441 		 */
1442 		(*sk)->sk_prot->unhash(*sk);
1443 	}
1444 	return rc;
1445 }
1446 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1447 
1448 u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt)
1449 {
1450 	return  *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt);
1451 }
1452 EXPORT_SYMBOL_GPL(snmp_get_cpu_field);
1453 
1454 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1455 {
1456 	unsigned long res = 0;
1457 	int i;
1458 
1459 	for_each_possible_cpu(i)
1460 		res += snmp_get_cpu_field(mib, i, offt);
1461 	return res;
1462 }
1463 EXPORT_SYMBOL_GPL(snmp_fold_field);
1464 
1465 #if BITS_PER_LONG==32
1466 
1467 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1468 			 size_t syncp_offset)
1469 {
1470 	void *bhptr;
1471 	struct u64_stats_sync *syncp;
1472 	u64 v;
1473 	unsigned int start;
1474 
1475 	bhptr = per_cpu_ptr(mib, cpu);
1476 	syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1477 	do {
1478 		start = u64_stats_fetch_begin_irq(syncp);
1479 		v = *(((u64 *)bhptr) + offt);
1480 	} while (u64_stats_fetch_retry_irq(syncp, start));
1481 
1482 	return v;
1483 }
1484 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1485 
1486 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1487 {
1488 	u64 res = 0;
1489 	int cpu;
1490 
1491 	for_each_possible_cpu(cpu) {
1492 		res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1493 	}
1494 	return res;
1495 }
1496 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1497 #endif
1498 
1499 #ifdef CONFIG_IP_MULTICAST
1500 static const struct net_protocol igmp_protocol = {
1501 	.handler =	igmp_rcv,
1502 	.netns_ok =	1,
1503 };
1504 #endif
1505 
1506 static const struct net_protocol tcp_protocol = {
1507 	.early_demux	=	tcp_v4_early_demux,
1508 	.handler	=	tcp_v4_rcv,
1509 	.err_handler	=	tcp_v4_err,
1510 	.no_policy	=	1,
1511 	.netns_ok	=	1,
1512 	.icmp_strict_tag_validation = 1,
1513 };
1514 
1515 static const struct net_protocol udp_protocol = {
1516 	.early_demux =	udp_v4_early_demux,
1517 	.handler =	udp_rcv,
1518 	.err_handler =	udp_err,
1519 	.no_policy =	1,
1520 	.netns_ok =	1,
1521 };
1522 
1523 static const struct net_protocol icmp_protocol = {
1524 	.handler =	icmp_rcv,
1525 	.err_handler =	icmp_err,
1526 	.no_policy =	1,
1527 	.netns_ok =	1,
1528 };
1529 
1530 static __net_init int ipv4_mib_init_net(struct net *net)
1531 {
1532 	int i;
1533 
1534 	net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1535 	if (!net->mib.tcp_statistics)
1536 		goto err_tcp_mib;
1537 	net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1538 	if (!net->mib.ip_statistics)
1539 		goto err_ip_mib;
1540 
1541 	for_each_possible_cpu(i) {
1542 		struct ipstats_mib *af_inet_stats;
1543 		af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1544 		u64_stats_init(&af_inet_stats->syncp);
1545 	}
1546 
1547 	net->mib.net_statistics = alloc_percpu(struct linux_mib);
1548 	if (!net->mib.net_statistics)
1549 		goto err_net_mib;
1550 	net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1551 	if (!net->mib.udp_statistics)
1552 		goto err_udp_mib;
1553 	net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1554 	if (!net->mib.udplite_statistics)
1555 		goto err_udplite_mib;
1556 	net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1557 	if (!net->mib.icmp_statistics)
1558 		goto err_icmp_mib;
1559 	net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1560 					      GFP_KERNEL);
1561 	if (!net->mib.icmpmsg_statistics)
1562 		goto err_icmpmsg_mib;
1563 
1564 	tcp_mib_init(net);
1565 	return 0;
1566 
1567 err_icmpmsg_mib:
1568 	free_percpu(net->mib.icmp_statistics);
1569 err_icmp_mib:
1570 	free_percpu(net->mib.udplite_statistics);
1571 err_udplite_mib:
1572 	free_percpu(net->mib.udp_statistics);
1573 err_udp_mib:
1574 	free_percpu(net->mib.net_statistics);
1575 err_net_mib:
1576 	free_percpu(net->mib.ip_statistics);
1577 err_ip_mib:
1578 	free_percpu(net->mib.tcp_statistics);
1579 err_tcp_mib:
1580 	return -ENOMEM;
1581 }
1582 
1583 static __net_exit void ipv4_mib_exit_net(struct net *net)
1584 {
1585 	kfree(net->mib.icmpmsg_statistics);
1586 	free_percpu(net->mib.icmp_statistics);
1587 	free_percpu(net->mib.udplite_statistics);
1588 	free_percpu(net->mib.udp_statistics);
1589 	free_percpu(net->mib.net_statistics);
1590 	free_percpu(net->mib.ip_statistics);
1591 	free_percpu(net->mib.tcp_statistics);
1592 }
1593 
1594 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1595 	.init = ipv4_mib_init_net,
1596 	.exit = ipv4_mib_exit_net,
1597 };
1598 
1599 static int __init init_ipv4_mibs(void)
1600 {
1601 	return register_pernet_subsys(&ipv4_mib_ops);
1602 }
1603 
1604 static __net_init int inet_init_net(struct net *net)
1605 {
1606 	/*
1607 	 * Set defaults for local port range
1608 	 */
1609 	seqlock_init(&net->ipv4.ip_local_ports.lock);
1610 	net->ipv4.ip_local_ports.range[0] =  32768;
1611 	net->ipv4.ip_local_ports.range[1] =  60999;
1612 
1613 	seqlock_init(&net->ipv4.ping_group_range.lock);
1614 	/*
1615 	 * Sane defaults - nobody may create ping sockets.
1616 	 * Boot scripts should set this to distro-specific group.
1617 	 */
1618 	net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1619 	net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1620 	return 0;
1621 }
1622 
1623 static __net_exit void inet_exit_net(struct net *net)
1624 {
1625 }
1626 
1627 static __net_initdata struct pernet_operations af_inet_ops = {
1628 	.init = inet_init_net,
1629 	.exit = inet_exit_net,
1630 };
1631 
1632 static int __init init_inet_pernet_ops(void)
1633 {
1634 	return register_pernet_subsys(&af_inet_ops);
1635 }
1636 
1637 static int ipv4_proc_init(void);
1638 
1639 /*
1640  *	IP protocol layer initialiser
1641  */
1642 
1643 static struct packet_offload ip_packet_offload __read_mostly = {
1644 	.type = cpu_to_be16(ETH_P_IP),
1645 	.callbacks = {
1646 		.gso_segment = inet_gso_segment,
1647 		.gro_receive = inet_gro_receive,
1648 		.gro_complete = inet_gro_complete,
1649 	},
1650 };
1651 
1652 static const struct net_offload ipip_offload = {
1653 	.callbacks = {
1654 		.gso_segment	= inet_gso_segment,
1655 		.gro_receive	= inet_gro_receive,
1656 		.gro_complete	= inet_gro_complete,
1657 	},
1658 };
1659 
1660 static int __init ipv4_offload_init(void)
1661 {
1662 	/*
1663 	 * Add offloads
1664 	 */
1665 	if (udpv4_offload_init() < 0)
1666 		pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1667 	if (tcpv4_offload_init() < 0)
1668 		pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1669 
1670 	dev_add_offload(&ip_packet_offload);
1671 	inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1672 	return 0;
1673 }
1674 
1675 fs_initcall(ipv4_offload_init);
1676 
1677 static struct packet_type ip_packet_type __read_mostly = {
1678 	.type = cpu_to_be16(ETH_P_IP),
1679 	.func = ip_rcv,
1680 };
1681 
1682 static int __init inet_init(void)
1683 {
1684 	struct inet_protosw *q;
1685 	struct list_head *r;
1686 	int rc = -EINVAL;
1687 
1688 	sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1689 
1690 	rc = proto_register(&tcp_prot, 1);
1691 	if (rc)
1692 		goto out;
1693 
1694 	rc = proto_register(&udp_prot, 1);
1695 	if (rc)
1696 		goto out_unregister_tcp_proto;
1697 
1698 	rc = proto_register(&raw_prot, 1);
1699 	if (rc)
1700 		goto out_unregister_udp_proto;
1701 
1702 	rc = proto_register(&ping_prot, 1);
1703 	if (rc)
1704 		goto out_unregister_raw_proto;
1705 
1706 	/*
1707 	 *	Tell SOCKET that we are alive...
1708 	 */
1709 
1710 	(void)sock_register(&inet_family_ops);
1711 
1712 #ifdef CONFIG_SYSCTL
1713 	ip_static_sysctl_init();
1714 #endif
1715 
1716 	/*
1717 	 *	Add all the base protocols.
1718 	 */
1719 
1720 	if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1721 		pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1722 	if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1723 		pr_crit("%s: Cannot add UDP protocol\n", __func__);
1724 	if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1725 		pr_crit("%s: Cannot add TCP protocol\n", __func__);
1726 #ifdef CONFIG_IP_MULTICAST
1727 	if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1728 		pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1729 #endif
1730 
1731 	/* Register the socket-side information for inet_create. */
1732 	for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1733 		INIT_LIST_HEAD(r);
1734 
1735 	for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1736 		inet_register_protosw(q);
1737 
1738 	/*
1739 	 *	Set the ARP module up
1740 	 */
1741 
1742 	arp_init();
1743 
1744 	/*
1745 	 *	Set the IP module up
1746 	 */
1747 
1748 	ip_init();
1749 
1750 	tcp_v4_init();
1751 
1752 	/* Setup TCP slab cache for open requests. */
1753 	tcp_init();
1754 
1755 	/* Setup UDP memory threshold */
1756 	udp_init();
1757 
1758 	/* Add UDP-Lite (RFC 3828) */
1759 	udplite4_register();
1760 
1761 	ping_init();
1762 
1763 	/*
1764 	 *	Set the ICMP layer up
1765 	 */
1766 
1767 	if (icmp_init() < 0)
1768 		panic("Failed to create the ICMP control socket.\n");
1769 
1770 	/*
1771 	 *	Initialise the multicast router
1772 	 */
1773 #if defined(CONFIG_IP_MROUTE)
1774 	if (ip_mr_init())
1775 		pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1776 #endif
1777 
1778 	if (init_inet_pernet_ops())
1779 		pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
1780 	/*
1781 	 *	Initialise per-cpu ipv4 mibs
1782 	 */
1783 
1784 	if (init_ipv4_mibs())
1785 		pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1786 
1787 	ipv4_proc_init();
1788 
1789 	ipfrag_init();
1790 
1791 	dev_add_pack(&ip_packet_type);
1792 
1793 	ip_tunnel_core_init();
1794 
1795 	rc = 0;
1796 out:
1797 	return rc;
1798 out_unregister_raw_proto:
1799 	proto_unregister(&raw_prot);
1800 out_unregister_udp_proto:
1801 	proto_unregister(&udp_prot);
1802 out_unregister_tcp_proto:
1803 	proto_unregister(&tcp_prot);
1804 	goto out;
1805 }
1806 
1807 fs_initcall(inet_init);
1808 
1809 /* ------------------------------------------------------------------------ */
1810 
1811 #ifdef CONFIG_PROC_FS
1812 static int __init ipv4_proc_init(void)
1813 {
1814 	int rc = 0;
1815 
1816 	if (raw_proc_init())
1817 		goto out_raw;
1818 	if (tcp4_proc_init())
1819 		goto out_tcp;
1820 	if (udp4_proc_init())
1821 		goto out_udp;
1822 	if (ping_proc_init())
1823 		goto out_ping;
1824 	if (ip_misc_proc_init())
1825 		goto out_misc;
1826 out:
1827 	return rc;
1828 out_misc:
1829 	ping_proc_exit();
1830 out_ping:
1831 	udp4_proc_exit();
1832 out_udp:
1833 	tcp4_proc_exit();
1834 out_tcp:
1835 	raw_proc_exit();
1836 out_raw:
1837 	rc = -ENOMEM;
1838 	goto out;
1839 }
1840 
1841 #else /* CONFIG_PROC_FS */
1842 static int __init ipv4_proc_init(void)
1843 {
1844 	return 0;
1845 }
1846 #endif /* CONFIG_PROC_FS */
1847 
1848 MODULE_ALIAS_NETPROTO(PF_INET);
1849 
1850