xref: /linux/net/ipv4/af_inet.c (revision 6ebe6dbd6886af07b102aca42e44edbee94a22d9)
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/kmod.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 <linux/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 #include <trace/events/sock.h>
125 
126 /* The inetsw table contains everything that inet_create needs to
127  * build a new socket.
128  */
129 static struct list_head inetsw[SOCK_MAX];
130 static DEFINE_SPINLOCK(inetsw_lock);
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(refcount_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, tcp_fastopen;
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 		/* Enable TFO w/o requiring TCP_FASTOPEN socket option.
216 		 * Note that only TCP sockets (SOCK_STREAM) will reach here.
217 		 * Also fastopen backlog may already been set via the option
218 		 * because the socket was in TCP_LISTEN state previously but
219 		 * was shutdown() rather than close().
220 		 */
221 		tcp_fastopen = sock_net(sk)->ipv4.sysctl_tcp_fastopen;
222 		if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
223 		    (tcp_fastopen & TFO_SERVER_ENABLE) &&
224 		    !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
225 			fastopen_queue_tune(sk, backlog);
226 			tcp_fastopen_init_key_once(sock_net(sk));
227 		}
228 
229 		err = inet_csk_listen_start(sk, backlog);
230 		if (err)
231 			goto out;
232 	}
233 	sk->sk_max_ack_backlog = backlog;
234 	err = 0;
235 
236 out:
237 	release_sock(sk);
238 	return err;
239 }
240 EXPORT_SYMBOL(inet_listen);
241 
242 /*
243  *	Create an inet socket.
244  */
245 
246 static int inet_create(struct net *net, struct socket *sock, int protocol,
247 		       int kern)
248 {
249 	struct sock *sk;
250 	struct inet_protosw *answer;
251 	struct inet_sock *inet;
252 	struct proto *answer_prot;
253 	unsigned char answer_flags;
254 	int try_loading_module = 0;
255 	int err;
256 
257 	if (protocol < 0 || protocol >= IPPROTO_MAX)
258 		return -EINVAL;
259 
260 	sock->state = SS_UNCONNECTED;
261 
262 	/* Look for the requested type/protocol pair. */
263 lookup_protocol:
264 	err = -ESOCKTNOSUPPORT;
265 	rcu_read_lock();
266 	list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
267 
268 		err = 0;
269 		/* Check the non-wild match. */
270 		if (protocol == answer->protocol) {
271 			if (protocol != IPPROTO_IP)
272 				break;
273 		} else {
274 			/* Check for the two wild cases. */
275 			if (IPPROTO_IP == protocol) {
276 				protocol = answer->protocol;
277 				break;
278 			}
279 			if (IPPROTO_IP == answer->protocol)
280 				break;
281 		}
282 		err = -EPROTONOSUPPORT;
283 	}
284 
285 	if (unlikely(err)) {
286 		if (try_loading_module < 2) {
287 			rcu_read_unlock();
288 			/*
289 			 * Be more specific, e.g. net-pf-2-proto-132-type-1
290 			 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
291 			 */
292 			if (++try_loading_module == 1)
293 				request_module("net-pf-%d-proto-%d-type-%d",
294 					       PF_INET, protocol, sock->type);
295 			/*
296 			 * Fall back to generic, e.g. net-pf-2-proto-132
297 			 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
298 			 */
299 			else
300 				request_module("net-pf-%d-proto-%d",
301 					       PF_INET, protocol);
302 			goto lookup_protocol;
303 		} else
304 			goto out_rcu_unlock;
305 	}
306 
307 	err = -EPERM;
308 	if (sock->type == SOCK_RAW && !kern &&
309 	    !ns_capable(net->user_ns, CAP_NET_RAW))
310 		goto out_rcu_unlock;
311 
312 	sock->ops = answer->ops;
313 	answer_prot = answer->prot;
314 	answer_flags = answer->flags;
315 	rcu_read_unlock();
316 
317 	WARN_ON(!answer_prot->slab);
318 
319 	err = -ENOBUFS;
320 	sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
321 	if (!sk)
322 		goto out;
323 
324 	err = 0;
325 	if (INET_PROTOSW_REUSE & answer_flags)
326 		sk->sk_reuse = SK_CAN_REUSE;
327 
328 	inet = inet_sk(sk);
329 	inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
330 
331 	inet->nodefrag = 0;
332 
333 	if (SOCK_RAW == sock->type) {
334 		inet->inet_num = protocol;
335 		if (IPPROTO_RAW == protocol)
336 			inet->hdrincl = 1;
337 	}
338 
339 	if (net->ipv4.sysctl_ip_no_pmtu_disc)
340 		inet->pmtudisc = IP_PMTUDISC_DONT;
341 	else
342 		inet->pmtudisc = IP_PMTUDISC_WANT;
343 
344 	inet->inet_id = 0;
345 
346 	sock_init_data(sock, sk);
347 
348 	sk->sk_destruct	   = inet_sock_destruct;
349 	sk->sk_protocol	   = protocol;
350 	sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
351 
352 	inet->uc_ttl	= -1;
353 	inet->mc_loop	= 1;
354 	inet->mc_ttl	= 1;
355 	inet->mc_all	= 1;
356 	inet->mc_index	= 0;
357 	inet->mc_list	= NULL;
358 	inet->rcv_tos	= 0;
359 
360 	sk_refcnt_debug_inc(sk);
361 
362 	if (inet->inet_num) {
363 		/* It assumes that any protocol which allows
364 		 * the user to assign a number at socket
365 		 * creation time automatically
366 		 * shares.
367 		 */
368 		inet->inet_sport = htons(inet->inet_num);
369 		/* Add to protocol hash chains. */
370 		err = sk->sk_prot->hash(sk);
371 		if (err) {
372 			sk_common_release(sk);
373 			goto out;
374 		}
375 	}
376 
377 	if (sk->sk_prot->init) {
378 		err = sk->sk_prot->init(sk);
379 		if (err) {
380 			sk_common_release(sk);
381 			goto out;
382 		}
383 	}
384 
385 	if (!kern) {
386 		err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
387 		if (err) {
388 			sk_common_release(sk);
389 			goto out;
390 		}
391 	}
392 out:
393 	return err;
394 out_rcu_unlock:
395 	rcu_read_unlock();
396 	goto out;
397 }
398 
399 
400 /*
401  *	The peer socket should always be NULL (or else). When we call this
402  *	function we are destroying the object and from then on nobody
403  *	should refer to it.
404  */
405 int inet_release(struct socket *sock)
406 {
407 	struct sock *sk = sock->sk;
408 
409 	if (sk) {
410 		long timeout;
411 
412 		/* Applications forget to leave groups before exiting */
413 		ip_mc_drop_socket(sk);
414 
415 		/* If linger is set, we don't return until the close
416 		 * is complete.  Otherwise we return immediately. The
417 		 * actually closing is done the same either way.
418 		 *
419 		 * If the close is due to the process exiting, we never
420 		 * linger..
421 		 */
422 		timeout = 0;
423 		if (sock_flag(sk, SOCK_LINGER) &&
424 		    !(current->flags & PF_EXITING))
425 			timeout = sk->sk_lingertime;
426 		sock->sk = NULL;
427 		sk->sk_prot->close(sk, timeout);
428 	}
429 	return 0;
430 }
431 EXPORT_SYMBOL(inet_release);
432 
433 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
434 {
435 	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
436 	struct sock *sk = sock->sk;
437 	struct inet_sock *inet = inet_sk(sk);
438 	struct net *net = sock_net(sk);
439 	unsigned short snum;
440 	int chk_addr_ret;
441 	u32 tb_id = RT_TABLE_LOCAL;
442 	int err;
443 
444 	/* If the socket has its own bind function then use it. (RAW) */
445 	if (sk->sk_prot->bind) {
446 		err = sk->sk_prot->bind(sk, uaddr, addr_len);
447 		goto out;
448 	}
449 	err = -EINVAL;
450 	if (addr_len < sizeof(struct sockaddr_in))
451 		goto out;
452 
453 	if (addr->sin_family != AF_INET) {
454 		/* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
455 		 * only if s_addr is INADDR_ANY.
456 		 */
457 		err = -EAFNOSUPPORT;
458 		if (addr->sin_family != AF_UNSPEC ||
459 		    addr->sin_addr.s_addr != htonl(INADDR_ANY))
460 			goto out;
461 	}
462 
463 	tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
464 	chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
465 
466 	/* Not specified by any standard per-se, however it breaks too
467 	 * many applications when removed.  It is unfortunate since
468 	 * allowing applications to make a non-local bind solves
469 	 * several problems with systems using dynamic addressing.
470 	 * (ie. your servers still start up even if your ISDN link
471 	 *  is temporarily down)
472 	 */
473 	err = -EADDRNOTAVAIL;
474 	if (!net->ipv4.sysctl_ip_nonlocal_bind &&
475 	    !(inet->freebind || inet->transparent) &&
476 	    addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
477 	    chk_addr_ret != RTN_LOCAL &&
478 	    chk_addr_ret != RTN_MULTICAST &&
479 	    chk_addr_ret != RTN_BROADCAST)
480 		goto out;
481 
482 	snum = ntohs(addr->sin_port);
483 	err = -EACCES;
484 	if (snum && snum < inet_prot_sock(net) &&
485 	    !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
486 		goto out;
487 
488 	/*      We keep a pair of addresses. rcv_saddr is the one
489 	 *      used by hash lookups, and saddr is used for transmit.
490 	 *
491 	 *      In the BSD API these are the same except where it
492 	 *      would be illegal to use them (multicast/broadcast) in
493 	 *      which case the sending device address is used.
494 	 */
495 	lock_sock(sk);
496 
497 	/* Check these errors (active socket, double bind). */
498 	err = -EINVAL;
499 	if (sk->sk_state != TCP_CLOSE || inet->inet_num)
500 		goto out_release_sock;
501 
502 	inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
503 	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
504 		inet->inet_saddr = 0;  /* Use device */
505 
506 	/* Make sure we are allowed to bind here. */
507 	if ((snum || !inet->bind_address_no_port) &&
508 	    sk->sk_prot->get_port(sk, snum)) {
509 		inet->inet_saddr = inet->inet_rcv_saddr = 0;
510 		err = -EADDRINUSE;
511 		goto out_release_sock;
512 	}
513 
514 	if (inet->inet_rcv_saddr)
515 		sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
516 	if (snum)
517 		sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
518 	inet->inet_sport = htons(inet->inet_num);
519 	inet->inet_daddr = 0;
520 	inet->inet_dport = 0;
521 	sk_dst_reset(sk);
522 	err = 0;
523 out_release_sock:
524 	release_sock(sk);
525 out:
526 	return err;
527 }
528 EXPORT_SYMBOL(inet_bind);
529 
530 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
531 		       int addr_len, int flags)
532 {
533 	struct sock *sk = sock->sk;
534 
535 	if (addr_len < sizeof(uaddr->sa_family))
536 		return -EINVAL;
537 	if (uaddr->sa_family == AF_UNSPEC)
538 		return sk->sk_prot->disconnect(sk, flags);
539 
540 	if (!inet_sk(sk)->inet_num && inet_autobind(sk))
541 		return -EAGAIN;
542 	return sk->sk_prot->connect(sk, uaddr, addr_len);
543 }
544 EXPORT_SYMBOL(inet_dgram_connect);
545 
546 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
547 {
548 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
549 
550 	add_wait_queue(sk_sleep(sk), &wait);
551 	sk->sk_write_pending += writebias;
552 
553 	/* Basic assumption: if someone sets sk->sk_err, he _must_
554 	 * change state of the socket from TCP_SYN_*.
555 	 * Connect() does not allow to get error notifications
556 	 * without closing the socket.
557 	 */
558 	while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
559 		release_sock(sk);
560 		timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
561 		lock_sock(sk);
562 		if (signal_pending(current) || !timeo)
563 			break;
564 	}
565 	remove_wait_queue(sk_sleep(sk), &wait);
566 	sk->sk_write_pending -= writebias;
567 	return timeo;
568 }
569 
570 /*
571  *	Connect to a remote host. There is regrettably still a little
572  *	TCP 'magic' in here.
573  */
574 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
575 			  int addr_len, int flags, int is_sendmsg)
576 {
577 	struct sock *sk = sock->sk;
578 	int err;
579 	long timeo;
580 
581 	/*
582 	 * uaddr can be NULL and addr_len can be 0 if:
583 	 * sk is a TCP fastopen active socket and
584 	 * TCP_FASTOPEN_CONNECT sockopt is set and
585 	 * we already have a valid cookie for this socket.
586 	 * In this case, user can call write() after connect().
587 	 * write() will invoke tcp_sendmsg_fastopen() which calls
588 	 * __inet_stream_connect().
589 	 */
590 	if (uaddr) {
591 		if (addr_len < sizeof(uaddr->sa_family))
592 			return -EINVAL;
593 
594 		if (uaddr->sa_family == AF_UNSPEC) {
595 			err = sk->sk_prot->disconnect(sk, flags);
596 			sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
597 			goto out;
598 		}
599 	}
600 
601 	switch (sock->state) {
602 	default:
603 		err = -EINVAL;
604 		goto out;
605 	case SS_CONNECTED:
606 		err = -EISCONN;
607 		goto out;
608 	case SS_CONNECTING:
609 		if (inet_sk(sk)->defer_connect)
610 			err = is_sendmsg ? -EINPROGRESS : -EISCONN;
611 		else
612 			err = -EALREADY;
613 		/* Fall out of switch with err, set for this state */
614 		break;
615 	case SS_UNCONNECTED:
616 		err = -EISCONN;
617 		if (sk->sk_state != TCP_CLOSE)
618 			goto out;
619 
620 		err = sk->sk_prot->connect(sk, uaddr, addr_len);
621 		if (err < 0)
622 			goto out;
623 
624 		sock->state = SS_CONNECTING;
625 
626 		if (!err && inet_sk(sk)->defer_connect)
627 			goto out;
628 
629 		/* Just entered SS_CONNECTING state; the only
630 		 * difference is that return value in non-blocking
631 		 * case is EINPROGRESS, rather than EALREADY.
632 		 */
633 		err = -EINPROGRESS;
634 		break;
635 	}
636 
637 	timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
638 
639 	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
640 		int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
641 				tcp_sk(sk)->fastopen_req &&
642 				tcp_sk(sk)->fastopen_req->data ? 1 : 0;
643 
644 		/* Error code is set above */
645 		if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
646 			goto out;
647 
648 		err = sock_intr_errno(timeo);
649 		if (signal_pending(current))
650 			goto out;
651 	}
652 
653 	/* Connection was closed by RST, timeout, ICMP error
654 	 * or another process disconnected us.
655 	 */
656 	if (sk->sk_state == TCP_CLOSE)
657 		goto sock_error;
658 
659 	/* sk->sk_err may be not zero now, if RECVERR was ordered by user
660 	 * and error was received after socket entered established state.
661 	 * Hence, it is handled normally after connect() return successfully.
662 	 */
663 
664 	sock->state = SS_CONNECTED;
665 	err = 0;
666 out:
667 	return err;
668 
669 sock_error:
670 	err = sock_error(sk) ? : -ECONNABORTED;
671 	sock->state = SS_UNCONNECTED;
672 	if (sk->sk_prot->disconnect(sk, flags))
673 		sock->state = SS_DISCONNECTING;
674 	goto out;
675 }
676 EXPORT_SYMBOL(__inet_stream_connect);
677 
678 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
679 			int addr_len, int flags)
680 {
681 	int err;
682 
683 	lock_sock(sock->sk);
684 	err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
685 	release_sock(sock->sk);
686 	return err;
687 }
688 EXPORT_SYMBOL(inet_stream_connect);
689 
690 /*
691  *	Accept a pending connection. The TCP layer now gives BSD semantics.
692  */
693 
694 int inet_accept(struct socket *sock, struct socket *newsock, int flags,
695 		bool kern)
696 {
697 	struct sock *sk1 = sock->sk;
698 	int err = -EINVAL;
699 	struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err, kern);
700 
701 	if (!sk2)
702 		goto do_err;
703 
704 	lock_sock(sk2);
705 
706 	sock_rps_record_flow(sk2);
707 	WARN_ON(!((1 << sk2->sk_state) &
708 		  (TCPF_ESTABLISHED | TCPF_SYN_RECV |
709 		  TCPF_CLOSE_WAIT | TCPF_CLOSE)));
710 
711 	sock_graft(sk2, newsock);
712 
713 	newsock->state = SS_CONNECTED;
714 	err = 0;
715 	release_sock(sk2);
716 do_err:
717 	return err;
718 }
719 EXPORT_SYMBOL(inet_accept);
720 
721 
722 /*
723  *	This does both peername and sockname.
724  */
725 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
726 			int *uaddr_len, int peer)
727 {
728 	struct sock *sk		= sock->sk;
729 	struct inet_sock *inet	= inet_sk(sk);
730 	DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
731 
732 	sin->sin_family = AF_INET;
733 	if (peer) {
734 		if (!inet->inet_dport ||
735 		    (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
736 		     peer == 1))
737 			return -ENOTCONN;
738 		sin->sin_port = inet->inet_dport;
739 		sin->sin_addr.s_addr = inet->inet_daddr;
740 	} else {
741 		__be32 addr = inet->inet_rcv_saddr;
742 		if (!addr)
743 			addr = inet->inet_saddr;
744 		sin->sin_port = inet->inet_sport;
745 		sin->sin_addr.s_addr = addr;
746 	}
747 	memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
748 	*uaddr_len = sizeof(*sin);
749 	return 0;
750 }
751 EXPORT_SYMBOL(inet_getname);
752 
753 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
754 {
755 	struct sock *sk = sock->sk;
756 
757 	sock_rps_record_flow(sk);
758 
759 	/* We may need to bind the socket. */
760 	if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
761 	    inet_autobind(sk))
762 		return -EAGAIN;
763 
764 	return sk->sk_prot->sendmsg(sk, msg, size);
765 }
766 EXPORT_SYMBOL(inet_sendmsg);
767 
768 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
769 		      size_t size, int flags)
770 {
771 	struct sock *sk = sock->sk;
772 
773 	sock_rps_record_flow(sk);
774 
775 	/* We may need to bind the socket. */
776 	if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
777 	    inet_autobind(sk))
778 		return -EAGAIN;
779 
780 	if (sk->sk_prot->sendpage)
781 		return sk->sk_prot->sendpage(sk, page, offset, size, flags);
782 	return sock_no_sendpage(sock, page, offset, size, flags);
783 }
784 EXPORT_SYMBOL(inet_sendpage);
785 
786 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
787 		 int flags)
788 {
789 	struct sock *sk = sock->sk;
790 	int addr_len = 0;
791 	int err;
792 
793 	sock_rps_record_flow(sk);
794 
795 	err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
796 				   flags & ~MSG_DONTWAIT, &addr_len);
797 	if (err >= 0)
798 		msg->msg_namelen = addr_len;
799 	return err;
800 }
801 EXPORT_SYMBOL(inet_recvmsg);
802 
803 int inet_shutdown(struct socket *sock, int how)
804 {
805 	struct sock *sk = sock->sk;
806 	int err = 0;
807 
808 	/* This should really check to make sure
809 	 * the socket is a TCP socket. (WHY AC...)
810 	 */
811 	how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
812 		       1->2 bit 2 snds.
813 		       2->3 */
814 	if ((how & ~SHUTDOWN_MASK) || !how)	/* MAXINT->0 */
815 		return -EINVAL;
816 
817 	lock_sock(sk);
818 	if (sock->state == SS_CONNECTING) {
819 		if ((1 << sk->sk_state) &
820 		    (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
821 			sock->state = SS_DISCONNECTING;
822 		else
823 			sock->state = SS_CONNECTED;
824 	}
825 
826 	switch (sk->sk_state) {
827 	case TCP_CLOSE:
828 		err = -ENOTCONN;
829 		/* Hack to wake up other listeners, who can poll for
830 		   POLLHUP, even on eg. unconnected UDP sockets -- RR */
831 		/* fall through */
832 	default:
833 		sk->sk_shutdown |= how;
834 		if (sk->sk_prot->shutdown)
835 			sk->sk_prot->shutdown(sk, how);
836 		break;
837 
838 	/* Remaining two branches are temporary solution for missing
839 	 * close() in multithreaded environment. It is _not_ a good idea,
840 	 * but we have no choice until close() is repaired at VFS level.
841 	 */
842 	case TCP_LISTEN:
843 		if (!(how & RCV_SHUTDOWN))
844 			break;
845 		/* fall through */
846 	case TCP_SYN_SENT:
847 		err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
848 		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
849 		break;
850 	}
851 
852 	/* Wake up anyone sleeping in poll. */
853 	sk->sk_state_change(sk);
854 	release_sock(sk);
855 	return err;
856 }
857 EXPORT_SYMBOL(inet_shutdown);
858 
859 /*
860  *	ioctl() calls you can issue on an INET socket. Most of these are
861  *	device configuration and stuff and very rarely used. Some ioctls
862  *	pass on to the socket itself.
863  *
864  *	NOTE: I like the idea of a module for the config stuff. ie ifconfig
865  *	loads the devconfigure module does its configuring and unloads it.
866  *	There's a good 20K of config code hanging around the kernel.
867  */
868 
869 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
870 {
871 	struct sock *sk = sock->sk;
872 	int err = 0;
873 	struct net *net = sock_net(sk);
874 
875 	switch (cmd) {
876 	case SIOCGSTAMP:
877 		err = sock_get_timestamp(sk, (struct timeval __user *)arg);
878 		break;
879 	case SIOCGSTAMPNS:
880 		err = sock_get_timestampns(sk, (struct timespec __user *)arg);
881 		break;
882 	case SIOCADDRT:
883 	case SIOCDELRT:
884 	case SIOCRTMSG:
885 		err = ip_rt_ioctl(net, cmd, (void __user *)arg);
886 		break;
887 	case SIOCDARP:
888 	case SIOCGARP:
889 	case SIOCSARP:
890 		err = arp_ioctl(net, cmd, (void __user *)arg);
891 		break;
892 	case SIOCGIFADDR:
893 	case SIOCSIFADDR:
894 	case SIOCGIFBRDADDR:
895 	case SIOCSIFBRDADDR:
896 	case SIOCGIFNETMASK:
897 	case SIOCSIFNETMASK:
898 	case SIOCGIFDSTADDR:
899 	case SIOCSIFDSTADDR:
900 	case SIOCSIFPFLAGS:
901 	case SIOCGIFPFLAGS:
902 	case SIOCSIFFLAGS:
903 		err = devinet_ioctl(net, cmd, (void __user *)arg);
904 		break;
905 	default:
906 		if (sk->sk_prot->ioctl)
907 			err = sk->sk_prot->ioctl(sk, cmd, arg);
908 		else
909 			err = -ENOIOCTLCMD;
910 		break;
911 	}
912 	return err;
913 }
914 EXPORT_SYMBOL(inet_ioctl);
915 
916 #ifdef CONFIG_COMPAT
917 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
918 {
919 	struct sock *sk = sock->sk;
920 	int err = -ENOIOCTLCMD;
921 
922 	if (sk->sk_prot->compat_ioctl)
923 		err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
924 
925 	return err;
926 }
927 #endif
928 
929 const struct proto_ops inet_stream_ops = {
930 	.family		   = PF_INET,
931 	.owner		   = THIS_MODULE,
932 	.release	   = inet_release,
933 	.bind		   = inet_bind,
934 	.connect	   = inet_stream_connect,
935 	.socketpair	   = sock_no_socketpair,
936 	.accept		   = inet_accept,
937 	.getname	   = inet_getname,
938 	.poll		   = tcp_poll,
939 	.ioctl		   = inet_ioctl,
940 	.listen		   = inet_listen,
941 	.shutdown	   = inet_shutdown,
942 	.setsockopt	   = sock_common_setsockopt,
943 	.getsockopt	   = sock_common_getsockopt,
944 	.sendmsg	   = inet_sendmsg,
945 	.recvmsg	   = inet_recvmsg,
946 	.mmap		   = sock_no_mmap,
947 	.sendpage	   = inet_sendpage,
948 	.splice_read	   = tcp_splice_read,
949 	.read_sock	   = tcp_read_sock,
950 	.sendmsg_locked    = tcp_sendmsg_locked,
951 	.sendpage_locked   = tcp_sendpage_locked,
952 	.peek_len	   = tcp_peek_len,
953 #ifdef CONFIG_COMPAT
954 	.compat_setsockopt = compat_sock_common_setsockopt,
955 	.compat_getsockopt = compat_sock_common_getsockopt,
956 	.compat_ioctl	   = inet_compat_ioctl,
957 #endif
958 };
959 EXPORT_SYMBOL(inet_stream_ops);
960 
961 const struct proto_ops inet_dgram_ops = {
962 	.family		   = PF_INET,
963 	.owner		   = THIS_MODULE,
964 	.release	   = inet_release,
965 	.bind		   = inet_bind,
966 	.connect	   = inet_dgram_connect,
967 	.socketpair	   = sock_no_socketpair,
968 	.accept		   = sock_no_accept,
969 	.getname	   = inet_getname,
970 	.poll		   = udp_poll,
971 	.ioctl		   = inet_ioctl,
972 	.listen		   = sock_no_listen,
973 	.shutdown	   = inet_shutdown,
974 	.setsockopt	   = sock_common_setsockopt,
975 	.getsockopt	   = sock_common_getsockopt,
976 	.sendmsg	   = inet_sendmsg,
977 	.recvmsg	   = inet_recvmsg,
978 	.mmap		   = sock_no_mmap,
979 	.sendpage	   = inet_sendpage,
980 	.set_peek_off	   = sk_set_peek_off,
981 #ifdef CONFIG_COMPAT
982 	.compat_setsockopt = compat_sock_common_setsockopt,
983 	.compat_getsockopt = compat_sock_common_getsockopt,
984 	.compat_ioctl	   = inet_compat_ioctl,
985 #endif
986 };
987 EXPORT_SYMBOL(inet_dgram_ops);
988 
989 /*
990  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
991  * udp_poll
992  */
993 static const struct proto_ops inet_sockraw_ops = {
994 	.family		   = PF_INET,
995 	.owner		   = THIS_MODULE,
996 	.release	   = inet_release,
997 	.bind		   = inet_bind,
998 	.connect	   = inet_dgram_connect,
999 	.socketpair	   = sock_no_socketpair,
1000 	.accept		   = sock_no_accept,
1001 	.getname	   = inet_getname,
1002 	.poll		   = datagram_poll,
1003 	.ioctl		   = inet_ioctl,
1004 	.listen		   = sock_no_listen,
1005 	.shutdown	   = inet_shutdown,
1006 	.setsockopt	   = sock_common_setsockopt,
1007 	.getsockopt	   = sock_common_getsockopt,
1008 	.sendmsg	   = inet_sendmsg,
1009 	.recvmsg	   = inet_recvmsg,
1010 	.mmap		   = sock_no_mmap,
1011 	.sendpage	   = inet_sendpage,
1012 #ifdef CONFIG_COMPAT
1013 	.compat_setsockopt = compat_sock_common_setsockopt,
1014 	.compat_getsockopt = compat_sock_common_getsockopt,
1015 	.compat_ioctl	   = inet_compat_ioctl,
1016 #endif
1017 };
1018 
1019 static const struct net_proto_family inet_family_ops = {
1020 	.family = PF_INET,
1021 	.create = inet_create,
1022 	.owner	= THIS_MODULE,
1023 };
1024 
1025 /* Upon startup we insert all the elements in inetsw_array[] into
1026  * the linked list inetsw.
1027  */
1028 static struct inet_protosw inetsw_array[] =
1029 {
1030 	{
1031 		.type =       SOCK_STREAM,
1032 		.protocol =   IPPROTO_TCP,
1033 		.prot =       &tcp_prot,
1034 		.ops =        &inet_stream_ops,
1035 		.flags =      INET_PROTOSW_PERMANENT |
1036 			      INET_PROTOSW_ICSK,
1037 	},
1038 
1039 	{
1040 		.type =       SOCK_DGRAM,
1041 		.protocol =   IPPROTO_UDP,
1042 		.prot =       &udp_prot,
1043 		.ops =        &inet_dgram_ops,
1044 		.flags =      INET_PROTOSW_PERMANENT,
1045        },
1046 
1047        {
1048 		.type =       SOCK_DGRAM,
1049 		.protocol =   IPPROTO_ICMP,
1050 		.prot =       &ping_prot,
1051 		.ops =        &inet_sockraw_ops,
1052 		.flags =      INET_PROTOSW_REUSE,
1053        },
1054 
1055        {
1056 	       .type =       SOCK_RAW,
1057 	       .protocol =   IPPROTO_IP,	/* wild card */
1058 	       .prot =       &raw_prot,
1059 	       .ops =        &inet_sockraw_ops,
1060 	       .flags =      INET_PROTOSW_REUSE,
1061        }
1062 };
1063 
1064 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1065 
1066 void inet_register_protosw(struct inet_protosw *p)
1067 {
1068 	struct list_head *lh;
1069 	struct inet_protosw *answer;
1070 	int protocol = p->protocol;
1071 	struct list_head *last_perm;
1072 
1073 	spin_lock_bh(&inetsw_lock);
1074 
1075 	if (p->type >= SOCK_MAX)
1076 		goto out_illegal;
1077 
1078 	/* If we are trying to override a permanent protocol, bail. */
1079 	last_perm = &inetsw[p->type];
1080 	list_for_each(lh, &inetsw[p->type]) {
1081 		answer = list_entry(lh, struct inet_protosw, list);
1082 		/* Check only the non-wild match. */
1083 		if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1084 			break;
1085 		if (protocol == answer->protocol)
1086 			goto out_permanent;
1087 		last_perm = lh;
1088 	}
1089 
1090 	/* Add the new entry after the last permanent entry if any, so that
1091 	 * the new entry does not override a permanent entry when matched with
1092 	 * a wild-card protocol. But it is allowed to override any existing
1093 	 * non-permanent entry.  This means that when we remove this entry, the
1094 	 * system automatically returns to the old behavior.
1095 	 */
1096 	list_add_rcu(&p->list, last_perm);
1097 out:
1098 	spin_unlock_bh(&inetsw_lock);
1099 
1100 	return;
1101 
1102 out_permanent:
1103 	pr_err("Attempt to override permanent protocol %d\n", protocol);
1104 	goto out;
1105 
1106 out_illegal:
1107 	pr_err("Ignoring attempt to register invalid socket type %d\n",
1108 	       p->type);
1109 	goto out;
1110 }
1111 EXPORT_SYMBOL(inet_register_protosw);
1112 
1113 void inet_unregister_protosw(struct inet_protosw *p)
1114 {
1115 	if (INET_PROTOSW_PERMANENT & p->flags) {
1116 		pr_err("Attempt to unregister permanent protocol %d\n",
1117 		       p->protocol);
1118 	} else {
1119 		spin_lock_bh(&inetsw_lock);
1120 		list_del_rcu(&p->list);
1121 		spin_unlock_bh(&inetsw_lock);
1122 
1123 		synchronize_net();
1124 	}
1125 }
1126 EXPORT_SYMBOL(inet_unregister_protosw);
1127 
1128 static int inet_sk_reselect_saddr(struct sock *sk)
1129 {
1130 	struct inet_sock *inet = inet_sk(sk);
1131 	__be32 old_saddr = inet->inet_saddr;
1132 	__be32 daddr = inet->inet_daddr;
1133 	struct flowi4 *fl4;
1134 	struct rtable *rt;
1135 	__be32 new_saddr;
1136 	struct ip_options_rcu *inet_opt;
1137 
1138 	inet_opt = rcu_dereference_protected(inet->inet_opt,
1139 					     lockdep_sock_is_held(sk));
1140 	if (inet_opt && inet_opt->opt.srr)
1141 		daddr = inet_opt->opt.faddr;
1142 
1143 	/* Query new route. */
1144 	fl4 = &inet->cork.fl.u.ip4;
1145 	rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1146 			      sk->sk_bound_dev_if, sk->sk_protocol,
1147 			      inet->inet_sport, inet->inet_dport, sk);
1148 	if (IS_ERR(rt))
1149 		return PTR_ERR(rt);
1150 
1151 	sk_setup_caps(sk, &rt->dst);
1152 
1153 	new_saddr = fl4->saddr;
1154 
1155 	if (new_saddr == old_saddr)
1156 		return 0;
1157 
1158 	if (sock_net(sk)->ipv4.sysctl_ip_dynaddr > 1) {
1159 		pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1160 			__func__, &old_saddr, &new_saddr);
1161 	}
1162 
1163 	inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1164 
1165 	/*
1166 	 * XXX The only one ugly spot where we need to
1167 	 * XXX really change the sockets identity after
1168 	 * XXX it has entered the hashes. -DaveM
1169 	 *
1170 	 * Besides that, it does not check for connection
1171 	 * uniqueness. Wait for troubles.
1172 	 */
1173 	return __sk_prot_rehash(sk);
1174 }
1175 
1176 int inet_sk_rebuild_header(struct sock *sk)
1177 {
1178 	struct inet_sock *inet = inet_sk(sk);
1179 	struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1180 	__be32 daddr;
1181 	struct ip_options_rcu *inet_opt;
1182 	struct flowi4 *fl4;
1183 	int err;
1184 
1185 	/* Route is OK, nothing to do. */
1186 	if (rt)
1187 		return 0;
1188 
1189 	/* Reroute. */
1190 	rcu_read_lock();
1191 	inet_opt = rcu_dereference(inet->inet_opt);
1192 	daddr = inet->inet_daddr;
1193 	if (inet_opt && inet_opt->opt.srr)
1194 		daddr = inet_opt->opt.faddr;
1195 	rcu_read_unlock();
1196 	fl4 = &inet->cork.fl.u.ip4;
1197 	rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1198 				   inet->inet_dport, inet->inet_sport,
1199 				   sk->sk_protocol, RT_CONN_FLAGS(sk),
1200 				   sk->sk_bound_dev_if);
1201 	if (!IS_ERR(rt)) {
1202 		err = 0;
1203 		sk_setup_caps(sk, &rt->dst);
1204 	} else {
1205 		err = PTR_ERR(rt);
1206 
1207 		/* Routing failed... */
1208 		sk->sk_route_caps = 0;
1209 		/*
1210 		 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1211 		 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1212 		 */
1213 		if (!sock_net(sk)->ipv4.sysctl_ip_dynaddr ||
1214 		    sk->sk_state != TCP_SYN_SENT ||
1215 		    (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1216 		    (err = inet_sk_reselect_saddr(sk)) != 0)
1217 			sk->sk_err_soft = -err;
1218 	}
1219 
1220 	return err;
1221 }
1222 EXPORT_SYMBOL(inet_sk_rebuild_header);
1223 
1224 void inet_sk_set_state(struct sock *sk, int state)
1225 {
1226 	trace_inet_sock_set_state(sk, sk->sk_state, state);
1227 	sk->sk_state = state;
1228 }
1229 EXPORT_SYMBOL(inet_sk_set_state);
1230 
1231 void inet_sk_state_store(struct sock *sk, int newstate)
1232 {
1233 	trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1234 	smp_store_release(&sk->sk_state, newstate);
1235 }
1236 
1237 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1238 				 netdev_features_t features)
1239 {
1240 	bool udpfrag = false, fixedid = false, gso_partial, encap;
1241 	struct sk_buff *segs = ERR_PTR(-EINVAL);
1242 	const struct net_offload *ops;
1243 	unsigned int offset = 0;
1244 	struct iphdr *iph;
1245 	int proto, tot_len;
1246 	int nhoff;
1247 	int ihl;
1248 	int id;
1249 
1250 	skb_reset_network_header(skb);
1251 	nhoff = skb_network_header(skb) - skb_mac_header(skb);
1252 	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1253 		goto out;
1254 
1255 	iph = ip_hdr(skb);
1256 	ihl = iph->ihl * 4;
1257 	if (ihl < sizeof(*iph))
1258 		goto out;
1259 
1260 	id = ntohs(iph->id);
1261 	proto = iph->protocol;
1262 
1263 	/* Warning: after this point, iph might be no longer valid */
1264 	if (unlikely(!pskb_may_pull(skb, ihl)))
1265 		goto out;
1266 	__skb_pull(skb, ihl);
1267 
1268 	encap = SKB_GSO_CB(skb)->encap_level > 0;
1269 	if (encap)
1270 		features &= skb->dev->hw_enc_features;
1271 	SKB_GSO_CB(skb)->encap_level += ihl;
1272 
1273 	skb_reset_transport_header(skb);
1274 
1275 	segs = ERR_PTR(-EPROTONOSUPPORT);
1276 
1277 	if (!skb->encapsulation || encap) {
1278 		udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1279 		fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1280 
1281 		/* fixed ID is invalid if DF bit is not set */
1282 		if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1283 			goto out;
1284 	}
1285 
1286 	ops = rcu_dereference(inet_offloads[proto]);
1287 	if (likely(ops && ops->callbacks.gso_segment))
1288 		segs = ops->callbacks.gso_segment(skb, features);
1289 
1290 	if (IS_ERR_OR_NULL(segs))
1291 		goto out;
1292 
1293 	gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1294 
1295 	skb = segs;
1296 	do {
1297 		iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1298 		if (udpfrag) {
1299 			iph->frag_off = htons(offset >> 3);
1300 			if (skb->next)
1301 				iph->frag_off |= htons(IP_MF);
1302 			offset += skb->len - nhoff - ihl;
1303 			tot_len = skb->len - nhoff;
1304 		} else if (skb_is_gso(skb)) {
1305 			if (!fixedid) {
1306 				iph->id = htons(id);
1307 				id += skb_shinfo(skb)->gso_segs;
1308 			}
1309 
1310 			if (gso_partial)
1311 				tot_len = skb_shinfo(skb)->gso_size +
1312 					  SKB_GSO_CB(skb)->data_offset +
1313 					  skb->head - (unsigned char *)iph;
1314 			else
1315 				tot_len = skb->len - nhoff;
1316 		} else {
1317 			if (!fixedid)
1318 				iph->id = htons(id++);
1319 			tot_len = skb->len - nhoff;
1320 		}
1321 		iph->tot_len = htons(tot_len);
1322 		ip_send_check(iph);
1323 		if (encap)
1324 			skb_reset_inner_headers(skb);
1325 		skb->network_header = (u8 *)iph - skb->head;
1326 	} while ((skb = skb->next));
1327 
1328 out:
1329 	return segs;
1330 }
1331 EXPORT_SYMBOL(inet_gso_segment);
1332 
1333 struct sk_buff **inet_gro_receive(struct sk_buff **head, struct sk_buff *skb)
1334 {
1335 	const struct net_offload *ops;
1336 	struct sk_buff **pp = NULL;
1337 	struct sk_buff *p;
1338 	const struct iphdr *iph;
1339 	unsigned int hlen;
1340 	unsigned int off;
1341 	unsigned int id;
1342 	int flush = 1;
1343 	int proto;
1344 
1345 	off = skb_gro_offset(skb);
1346 	hlen = off + sizeof(*iph);
1347 	iph = skb_gro_header_fast(skb, off);
1348 	if (skb_gro_header_hard(skb, hlen)) {
1349 		iph = skb_gro_header_slow(skb, hlen, off);
1350 		if (unlikely(!iph))
1351 			goto out;
1352 	}
1353 
1354 	proto = iph->protocol;
1355 
1356 	rcu_read_lock();
1357 	ops = rcu_dereference(inet_offloads[proto]);
1358 	if (!ops || !ops->callbacks.gro_receive)
1359 		goto out_unlock;
1360 
1361 	if (*(u8 *)iph != 0x45)
1362 		goto out_unlock;
1363 
1364 	if (ip_is_fragment(iph))
1365 		goto out_unlock;
1366 
1367 	if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1368 		goto out_unlock;
1369 
1370 	id = ntohl(*(__be32 *)&iph->id);
1371 	flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1372 	id >>= 16;
1373 
1374 	for (p = *head; p; p = p->next) {
1375 		struct iphdr *iph2;
1376 		u16 flush_id;
1377 
1378 		if (!NAPI_GRO_CB(p)->same_flow)
1379 			continue;
1380 
1381 		iph2 = (struct iphdr *)(p->data + off);
1382 		/* The above works because, with the exception of the top
1383 		 * (inner most) layer, we only aggregate pkts with the same
1384 		 * hdr length so all the hdrs we'll need to verify will start
1385 		 * at the same offset.
1386 		 */
1387 		if ((iph->protocol ^ iph2->protocol) |
1388 		    ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1389 		    ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1390 			NAPI_GRO_CB(p)->same_flow = 0;
1391 			continue;
1392 		}
1393 
1394 		/* All fields must match except length and checksum. */
1395 		NAPI_GRO_CB(p)->flush |=
1396 			(iph->ttl ^ iph2->ttl) |
1397 			(iph->tos ^ iph2->tos) |
1398 			((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1399 
1400 		NAPI_GRO_CB(p)->flush |= flush;
1401 
1402 		/* We need to store of the IP ID check to be included later
1403 		 * when we can verify that this packet does in fact belong
1404 		 * to a given flow.
1405 		 */
1406 		flush_id = (u16)(id - ntohs(iph2->id));
1407 
1408 		/* This bit of code makes it much easier for us to identify
1409 		 * the cases where we are doing atomic vs non-atomic IP ID
1410 		 * checks.  Specifically an atomic check can return IP ID
1411 		 * values 0 - 0xFFFF, while a non-atomic check can only
1412 		 * return 0 or 0xFFFF.
1413 		 */
1414 		if (!NAPI_GRO_CB(p)->is_atomic ||
1415 		    !(iph->frag_off & htons(IP_DF))) {
1416 			flush_id ^= NAPI_GRO_CB(p)->count;
1417 			flush_id = flush_id ? 0xFFFF : 0;
1418 		}
1419 
1420 		/* If the previous IP ID value was based on an atomic
1421 		 * datagram we can overwrite the value and ignore it.
1422 		 */
1423 		if (NAPI_GRO_CB(skb)->is_atomic)
1424 			NAPI_GRO_CB(p)->flush_id = flush_id;
1425 		else
1426 			NAPI_GRO_CB(p)->flush_id |= flush_id;
1427 	}
1428 
1429 	NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1430 	NAPI_GRO_CB(skb)->flush |= flush;
1431 	skb_set_network_header(skb, off);
1432 	/* The above will be needed by the transport layer if there is one
1433 	 * immediately following this IP hdr.
1434 	 */
1435 
1436 	/* Note : No need to call skb_gro_postpull_rcsum() here,
1437 	 * as we already checked checksum over ipv4 header was 0
1438 	 */
1439 	skb_gro_pull(skb, sizeof(*iph));
1440 	skb_set_transport_header(skb, skb_gro_offset(skb));
1441 
1442 	pp = call_gro_receive(ops->callbacks.gro_receive, head, skb);
1443 
1444 out_unlock:
1445 	rcu_read_unlock();
1446 
1447 out:
1448 	skb_gro_flush_final(skb, pp, flush);
1449 
1450 	return pp;
1451 }
1452 EXPORT_SYMBOL(inet_gro_receive);
1453 
1454 static struct sk_buff **ipip_gro_receive(struct sk_buff **head,
1455 					 struct sk_buff *skb)
1456 {
1457 	if (NAPI_GRO_CB(skb)->encap_mark) {
1458 		NAPI_GRO_CB(skb)->flush = 1;
1459 		return NULL;
1460 	}
1461 
1462 	NAPI_GRO_CB(skb)->encap_mark = 1;
1463 
1464 	return inet_gro_receive(head, skb);
1465 }
1466 
1467 #define SECONDS_PER_DAY	86400
1468 
1469 /* inet_current_timestamp - Return IP network timestamp
1470  *
1471  * Return milliseconds since midnight in network byte order.
1472  */
1473 __be32 inet_current_timestamp(void)
1474 {
1475 	u32 secs;
1476 	u32 msecs;
1477 	struct timespec64 ts;
1478 
1479 	ktime_get_real_ts64(&ts);
1480 
1481 	/* Get secs since midnight. */
1482 	(void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1483 	/* Convert to msecs. */
1484 	msecs = secs * MSEC_PER_SEC;
1485 	/* Convert nsec to msec. */
1486 	msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1487 
1488 	/* Convert to network byte order. */
1489 	return htonl(msecs);
1490 }
1491 EXPORT_SYMBOL(inet_current_timestamp);
1492 
1493 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1494 {
1495 	if (sk->sk_family == AF_INET)
1496 		return ip_recv_error(sk, msg, len, addr_len);
1497 #if IS_ENABLED(CONFIG_IPV6)
1498 	if (sk->sk_family == AF_INET6)
1499 		return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1500 #endif
1501 	return -EINVAL;
1502 }
1503 
1504 int inet_gro_complete(struct sk_buff *skb, int nhoff)
1505 {
1506 	__be16 newlen = htons(skb->len - nhoff);
1507 	struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1508 	const struct net_offload *ops;
1509 	int proto = iph->protocol;
1510 	int err = -ENOSYS;
1511 
1512 	if (skb->encapsulation) {
1513 		skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1514 		skb_set_inner_network_header(skb, nhoff);
1515 	}
1516 
1517 	csum_replace2(&iph->check, iph->tot_len, newlen);
1518 	iph->tot_len = newlen;
1519 
1520 	rcu_read_lock();
1521 	ops = rcu_dereference(inet_offloads[proto]);
1522 	if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1523 		goto out_unlock;
1524 
1525 	/* Only need to add sizeof(*iph) to get to the next hdr below
1526 	 * because any hdr with option will have been flushed in
1527 	 * inet_gro_receive().
1528 	 */
1529 	err = ops->callbacks.gro_complete(skb, nhoff + sizeof(*iph));
1530 
1531 out_unlock:
1532 	rcu_read_unlock();
1533 
1534 	return err;
1535 }
1536 EXPORT_SYMBOL(inet_gro_complete);
1537 
1538 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1539 {
1540 	skb->encapsulation = 1;
1541 	skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1542 	return inet_gro_complete(skb, nhoff);
1543 }
1544 
1545 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1546 			 unsigned short type, unsigned char protocol,
1547 			 struct net *net)
1548 {
1549 	struct socket *sock;
1550 	int rc = sock_create_kern(net, family, type, protocol, &sock);
1551 
1552 	if (rc == 0) {
1553 		*sk = sock->sk;
1554 		(*sk)->sk_allocation = GFP_ATOMIC;
1555 		/*
1556 		 * Unhash it so that IP input processing does not even see it,
1557 		 * we do not wish this socket to see incoming packets.
1558 		 */
1559 		(*sk)->sk_prot->unhash(*sk);
1560 	}
1561 	return rc;
1562 }
1563 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1564 
1565 u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt)
1566 {
1567 	return  *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt);
1568 }
1569 EXPORT_SYMBOL_GPL(snmp_get_cpu_field);
1570 
1571 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1572 {
1573 	unsigned long res = 0;
1574 	int i;
1575 
1576 	for_each_possible_cpu(i)
1577 		res += snmp_get_cpu_field(mib, i, offt);
1578 	return res;
1579 }
1580 EXPORT_SYMBOL_GPL(snmp_fold_field);
1581 
1582 #if BITS_PER_LONG==32
1583 
1584 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1585 			 size_t syncp_offset)
1586 {
1587 	void *bhptr;
1588 	struct u64_stats_sync *syncp;
1589 	u64 v;
1590 	unsigned int start;
1591 
1592 	bhptr = per_cpu_ptr(mib, cpu);
1593 	syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1594 	do {
1595 		start = u64_stats_fetch_begin_irq(syncp);
1596 		v = *(((u64 *)bhptr) + offt);
1597 	} while (u64_stats_fetch_retry_irq(syncp, start));
1598 
1599 	return v;
1600 }
1601 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1602 
1603 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1604 {
1605 	u64 res = 0;
1606 	int cpu;
1607 
1608 	for_each_possible_cpu(cpu) {
1609 		res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1610 	}
1611 	return res;
1612 }
1613 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1614 #endif
1615 
1616 #ifdef CONFIG_IP_MULTICAST
1617 static const struct net_protocol igmp_protocol = {
1618 	.handler =	igmp_rcv,
1619 	.netns_ok =	1,
1620 };
1621 #endif
1622 
1623 /* thinking of making this const? Don't.
1624  * early_demux can change based on sysctl.
1625  */
1626 static struct net_protocol tcp_protocol = {
1627 	.early_demux	=	tcp_v4_early_demux,
1628 	.early_demux_handler =  tcp_v4_early_demux,
1629 	.handler	=	tcp_v4_rcv,
1630 	.err_handler	=	tcp_v4_err,
1631 	.no_policy	=	1,
1632 	.netns_ok	=	1,
1633 	.icmp_strict_tag_validation = 1,
1634 };
1635 
1636 /* thinking of making this const? Don't.
1637  * early_demux can change based on sysctl.
1638  */
1639 static struct net_protocol udp_protocol = {
1640 	.early_demux =	udp_v4_early_demux,
1641 	.early_demux_handler =	udp_v4_early_demux,
1642 	.handler =	udp_rcv,
1643 	.err_handler =	udp_err,
1644 	.no_policy =	1,
1645 	.netns_ok =	1,
1646 };
1647 
1648 static const struct net_protocol icmp_protocol = {
1649 	.handler =	icmp_rcv,
1650 	.err_handler =	icmp_err,
1651 	.no_policy =	1,
1652 	.netns_ok =	1,
1653 };
1654 
1655 static __net_init int ipv4_mib_init_net(struct net *net)
1656 {
1657 	int i;
1658 
1659 	net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1660 	if (!net->mib.tcp_statistics)
1661 		goto err_tcp_mib;
1662 	net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1663 	if (!net->mib.ip_statistics)
1664 		goto err_ip_mib;
1665 
1666 	for_each_possible_cpu(i) {
1667 		struct ipstats_mib *af_inet_stats;
1668 		af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1669 		u64_stats_init(&af_inet_stats->syncp);
1670 	}
1671 
1672 	net->mib.net_statistics = alloc_percpu(struct linux_mib);
1673 	if (!net->mib.net_statistics)
1674 		goto err_net_mib;
1675 	net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1676 	if (!net->mib.udp_statistics)
1677 		goto err_udp_mib;
1678 	net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1679 	if (!net->mib.udplite_statistics)
1680 		goto err_udplite_mib;
1681 	net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1682 	if (!net->mib.icmp_statistics)
1683 		goto err_icmp_mib;
1684 	net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1685 					      GFP_KERNEL);
1686 	if (!net->mib.icmpmsg_statistics)
1687 		goto err_icmpmsg_mib;
1688 
1689 	tcp_mib_init(net);
1690 	return 0;
1691 
1692 err_icmpmsg_mib:
1693 	free_percpu(net->mib.icmp_statistics);
1694 err_icmp_mib:
1695 	free_percpu(net->mib.udplite_statistics);
1696 err_udplite_mib:
1697 	free_percpu(net->mib.udp_statistics);
1698 err_udp_mib:
1699 	free_percpu(net->mib.net_statistics);
1700 err_net_mib:
1701 	free_percpu(net->mib.ip_statistics);
1702 err_ip_mib:
1703 	free_percpu(net->mib.tcp_statistics);
1704 err_tcp_mib:
1705 	return -ENOMEM;
1706 }
1707 
1708 static __net_exit void ipv4_mib_exit_net(struct net *net)
1709 {
1710 	kfree(net->mib.icmpmsg_statistics);
1711 	free_percpu(net->mib.icmp_statistics);
1712 	free_percpu(net->mib.udplite_statistics);
1713 	free_percpu(net->mib.udp_statistics);
1714 	free_percpu(net->mib.net_statistics);
1715 	free_percpu(net->mib.ip_statistics);
1716 	free_percpu(net->mib.tcp_statistics);
1717 }
1718 
1719 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1720 	.init = ipv4_mib_init_net,
1721 	.exit = ipv4_mib_exit_net,
1722 };
1723 
1724 static int __init init_ipv4_mibs(void)
1725 {
1726 	return register_pernet_subsys(&ipv4_mib_ops);
1727 }
1728 
1729 static __net_init int inet_init_net(struct net *net)
1730 {
1731 	/*
1732 	 * Set defaults for local port range
1733 	 */
1734 	seqlock_init(&net->ipv4.ip_local_ports.lock);
1735 	net->ipv4.ip_local_ports.range[0] =  32768;
1736 	net->ipv4.ip_local_ports.range[1] =  60999;
1737 
1738 	seqlock_init(&net->ipv4.ping_group_range.lock);
1739 	/*
1740 	 * Sane defaults - nobody may create ping sockets.
1741 	 * Boot scripts should set this to distro-specific group.
1742 	 */
1743 	net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1744 	net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1745 
1746 	/* Default values for sysctl-controlled parameters.
1747 	 * We set them here, in case sysctl is not compiled.
1748 	 */
1749 	net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1750 	net->ipv4.sysctl_ip_dynaddr = 0;
1751 	net->ipv4.sysctl_ip_early_demux = 1;
1752 	net->ipv4.sysctl_udp_early_demux = 1;
1753 	net->ipv4.sysctl_tcp_early_demux = 1;
1754 #ifdef CONFIG_SYSCTL
1755 	net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1756 #endif
1757 
1758 	/* Some igmp sysctl, whose values are always used */
1759 	net->ipv4.sysctl_igmp_max_memberships = 20;
1760 	net->ipv4.sysctl_igmp_max_msf = 10;
1761 	/* IGMP reports for link-local multicast groups are enabled by default */
1762 	net->ipv4.sysctl_igmp_llm_reports = 1;
1763 	net->ipv4.sysctl_igmp_qrv = 2;
1764 
1765 	return 0;
1766 }
1767 
1768 static __net_exit void inet_exit_net(struct net *net)
1769 {
1770 }
1771 
1772 static __net_initdata struct pernet_operations af_inet_ops = {
1773 	.init = inet_init_net,
1774 	.exit = inet_exit_net,
1775 };
1776 
1777 static int __init init_inet_pernet_ops(void)
1778 {
1779 	return register_pernet_subsys(&af_inet_ops);
1780 }
1781 
1782 static int ipv4_proc_init(void);
1783 
1784 /*
1785  *	IP protocol layer initialiser
1786  */
1787 
1788 static struct packet_offload ip_packet_offload __read_mostly = {
1789 	.type = cpu_to_be16(ETH_P_IP),
1790 	.callbacks = {
1791 		.gso_segment = inet_gso_segment,
1792 		.gro_receive = inet_gro_receive,
1793 		.gro_complete = inet_gro_complete,
1794 	},
1795 };
1796 
1797 static const struct net_offload ipip_offload = {
1798 	.callbacks = {
1799 		.gso_segment	= inet_gso_segment,
1800 		.gro_receive	= ipip_gro_receive,
1801 		.gro_complete	= ipip_gro_complete,
1802 	},
1803 };
1804 
1805 static int __init ipip_offload_init(void)
1806 {
1807 	return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1808 }
1809 
1810 static int __init ipv4_offload_init(void)
1811 {
1812 	/*
1813 	 * Add offloads
1814 	 */
1815 	if (udpv4_offload_init() < 0)
1816 		pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1817 	if (tcpv4_offload_init() < 0)
1818 		pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1819 	if (ipip_offload_init() < 0)
1820 		pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1821 
1822 	dev_add_offload(&ip_packet_offload);
1823 	return 0;
1824 }
1825 
1826 fs_initcall(ipv4_offload_init);
1827 
1828 static struct packet_type ip_packet_type __read_mostly = {
1829 	.type = cpu_to_be16(ETH_P_IP),
1830 	.func = ip_rcv,
1831 };
1832 
1833 static int __init inet_init(void)
1834 {
1835 	struct inet_protosw *q;
1836 	struct list_head *r;
1837 	int rc = -EINVAL;
1838 
1839 	sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1840 
1841 	rc = proto_register(&tcp_prot, 1);
1842 	if (rc)
1843 		goto out;
1844 
1845 	rc = proto_register(&udp_prot, 1);
1846 	if (rc)
1847 		goto out_unregister_tcp_proto;
1848 
1849 	rc = proto_register(&raw_prot, 1);
1850 	if (rc)
1851 		goto out_unregister_udp_proto;
1852 
1853 	rc = proto_register(&ping_prot, 1);
1854 	if (rc)
1855 		goto out_unregister_raw_proto;
1856 
1857 	/*
1858 	 *	Tell SOCKET that we are alive...
1859 	 */
1860 
1861 	(void)sock_register(&inet_family_ops);
1862 
1863 #ifdef CONFIG_SYSCTL
1864 	ip_static_sysctl_init();
1865 #endif
1866 
1867 	/*
1868 	 *	Add all the base protocols.
1869 	 */
1870 
1871 	if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1872 		pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1873 	if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1874 		pr_crit("%s: Cannot add UDP protocol\n", __func__);
1875 	if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1876 		pr_crit("%s: Cannot add TCP protocol\n", __func__);
1877 #ifdef CONFIG_IP_MULTICAST
1878 	if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1879 		pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1880 #endif
1881 
1882 	/* Register the socket-side information for inet_create. */
1883 	for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1884 		INIT_LIST_HEAD(r);
1885 
1886 	for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1887 		inet_register_protosw(q);
1888 
1889 	/*
1890 	 *	Set the ARP module up
1891 	 */
1892 
1893 	arp_init();
1894 
1895 	/*
1896 	 *	Set the IP module up
1897 	 */
1898 
1899 	ip_init();
1900 
1901 	/* Setup TCP slab cache for open requests. */
1902 	tcp_init();
1903 
1904 	/* Setup UDP memory threshold */
1905 	udp_init();
1906 
1907 	/* Add UDP-Lite (RFC 3828) */
1908 	udplite4_register();
1909 
1910 	ping_init();
1911 
1912 	/*
1913 	 *	Set the ICMP layer up
1914 	 */
1915 
1916 	if (icmp_init() < 0)
1917 		panic("Failed to create the ICMP control socket.\n");
1918 
1919 	/*
1920 	 *	Initialise the multicast router
1921 	 */
1922 #if defined(CONFIG_IP_MROUTE)
1923 	if (ip_mr_init())
1924 		pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1925 #endif
1926 
1927 	if (init_inet_pernet_ops())
1928 		pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
1929 	/*
1930 	 *	Initialise per-cpu ipv4 mibs
1931 	 */
1932 
1933 	if (init_ipv4_mibs())
1934 		pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1935 
1936 	ipv4_proc_init();
1937 
1938 	ipfrag_init();
1939 
1940 	dev_add_pack(&ip_packet_type);
1941 
1942 	ip_tunnel_core_init();
1943 
1944 	rc = 0;
1945 out:
1946 	return rc;
1947 out_unregister_raw_proto:
1948 	proto_unregister(&raw_prot);
1949 out_unregister_udp_proto:
1950 	proto_unregister(&udp_prot);
1951 out_unregister_tcp_proto:
1952 	proto_unregister(&tcp_prot);
1953 	goto out;
1954 }
1955 
1956 fs_initcall(inet_init);
1957 
1958 /* ------------------------------------------------------------------------ */
1959 
1960 #ifdef CONFIG_PROC_FS
1961 static int __init ipv4_proc_init(void)
1962 {
1963 	int rc = 0;
1964 
1965 	if (raw_proc_init())
1966 		goto out_raw;
1967 	if (tcp4_proc_init())
1968 		goto out_tcp;
1969 	if (udp4_proc_init())
1970 		goto out_udp;
1971 	if (ping_proc_init())
1972 		goto out_ping;
1973 	if (ip_misc_proc_init())
1974 		goto out_misc;
1975 out:
1976 	return rc;
1977 out_misc:
1978 	ping_proc_exit();
1979 out_ping:
1980 	udp4_proc_exit();
1981 out_udp:
1982 	tcp4_proc_exit();
1983 out_tcp:
1984 	raw_proc_exit();
1985 out_raw:
1986 	rc = -ENOMEM;
1987 	goto out;
1988 }
1989 
1990 #else /* CONFIG_PROC_FS */
1991 static int __init ipv4_proc_init(void)
1992 {
1993 	return 0;
1994 }
1995 #endif /* CONFIG_PROC_FS */
1996