xref: /openbmc/linux/net/ipv4/af_inet.c (revision 0c7beb2d)
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 	if (sk->sk_rx_skb_cache) {
140 		__kfree_skb(sk->sk_rx_skb_cache);
141 		sk->sk_rx_skb_cache = NULL;
142 	}
143 	__skb_queue_purge(&sk->sk_error_queue);
144 
145 	sk_mem_reclaim(sk);
146 
147 	if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
148 		pr_err("Attempt to release TCP socket in state %d %p\n",
149 		       sk->sk_state, sk);
150 		return;
151 	}
152 	if (!sock_flag(sk, SOCK_DEAD)) {
153 		pr_err("Attempt to release alive inet socket %p\n", sk);
154 		return;
155 	}
156 
157 	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
158 	WARN_ON(refcount_read(&sk->sk_wmem_alloc));
159 	WARN_ON(sk->sk_wmem_queued);
160 	WARN_ON(sk->sk_forward_alloc);
161 
162 	kfree(rcu_dereference_protected(inet->inet_opt, 1));
163 	dst_release(rcu_dereference_protected(sk->sk_dst_cache, 1));
164 	dst_release(sk->sk_rx_dst);
165 	sk_refcnt_debug_dec(sk);
166 }
167 EXPORT_SYMBOL(inet_sock_destruct);
168 
169 /*
170  *	The routines beyond this point handle the behaviour of an AF_INET
171  *	socket object. Mostly it punts to the subprotocols of IP to do
172  *	the work.
173  */
174 
175 /*
176  *	Automatically bind an unbound socket.
177  */
178 
179 static int inet_autobind(struct sock *sk)
180 {
181 	struct inet_sock *inet;
182 	/* We may need to bind the socket. */
183 	lock_sock(sk);
184 	inet = inet_sk(sk);
185 	if (!inet->inet_num) {
186 		if (sk->sk_prot->get_port(sk, 0)) {
187 			release_sock(sk);
188 			return -EAGAIN;
189 		}
190 		inet->inet_sport = htons(inet->inet_num);
191 	}
192 	release_sock(sk);
193 	return 0;
194 }
195 
196 /*
197  *	Move a socket into listening state.
198  */
199 int inet_listen(struct socket *sock, int backlog)
200 {
201 	struct sock *sk = sock->sk;
202 	unsigned char old_state;
203 	int err, tcp_fastopen;
204 
205 	lock_sock(sk);
206 
207 	err = -EINVAL;
208 	if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
209 		goto out;
210 
211 	old_state = sk->sk_state;
212 	if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
213 		goto out;
214 
215 	sk->sk_max_ack_backlog = backlog;
216 	/* Really, if the socket is already in listen state
217 	 * we can only allow the backlog to be adjusted.
218 	 */
219 	if (old_state != TCP_LISTEN) {
220 		/* Enable TFO w/o requiring TCP_FASTOPEN socket option.
221 		 * Note that only TCP sockets (SOCK_STREAM) will reach here.
222 		 * Also fastopen backlog may already been set via the option
223 		 * because the socket was in TCP_LISTEN state previously but
224 		 * was shutdown() rather than close().
225 		 */
226 		tcp_fastopen = sock_net(sk)->ipv4.sysctl_tcp_fastopen;
227 		if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
228 		    (tcp_fastopen & TFO_SERVER_ENABLE) &&
229 		    !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
230 			fastopen_queue_tune(sk, backlog);
231 			tcp_fastopen_init_key_once(sock_net(sk));
232 		}
233 
234 		err = inet_csk_listen_start(sk, backlog);
235 		if (err)
236 			goto out;
237 		tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL);
238 	}
239 	err = 0;
240 
241 out:
242 	release_sock(sk);
243 	return err;
244 }
245 EXPORT_SYMBOL(inet_listen);
246 
247 /*
248  *	Create an inet socket.
249  */
250 
251 static int inet_create(struct net *net, struct socket *sock, int protocol,
252 		       int kern)
253 {
254 	struct sock *sk;
255 	struct inet_protosw *answer;
256 	struct inet_sock *inet;
257 	struct proto *answer_prot;
258 	unsigned char answer_flags;
259 	int try_loading_module = 0;
260 	int err;
261 
262 	if (protocol < 0 || protocol >= IPPROTO_MAX)
263 		return -EINVAL;
264 
265 	sock->state = SS_UNCONNECTED;
266 
267 	/* Look for the requested type/protocol pair. */
268 lookup_protocol:
269 	err = -ESOCKTNOSUPPORT;
270 	rcu_read_lock();
271 	list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
272 
273 		err = 0;
274 		/* Check the non-wild match. */
275 		if (protocol == answer->protocol) {
276 			if (protocol != IPPROTO_IP)
277 				break;
278 		} else {
279 			/* Check for the two wild cases. */
280 			if (IPPROTO_IP == protocol) {
281 				protocol = answer->protocol;
282 				break;
283 			}
284 			if (IPPROTO_IP == answer->protocol)
285 				break;
286 		}
287 		err = -EPROTONOSUPPORT;
288 	}
289 
290 	if (unlikely(err)) {
291 		if (try_loading_module < 2) {
292 			rcu_read_unlock();
293 			/*
294 			 * Be more specific, e.g. net-pf-2-proto-132-type-1
295 			 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
296 			 */
297 			if (++try_loading_module == 1)
298 				request_module("net-pf-%d-proto-%d-type-%d",
299 					       PF_INET, protocol, sock->type);
300 			/*
301 			 * Fall back to generic, e.g. net-pf-2-proto-132
302 			 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
303 			 */
304 			else
305 				request_module("net-pf-%d-proto-%d",
306 					       PF_INET, protocol);
307 			goto lookup_protocol;
308 		} else
309 			goto out_rcu_unlock;
310 	}
311 
312 	err = -EPERM;
313 	if (sock->type == SOCK_RAW && !kern &&
314 	    !ns_capable(net->user_ns, CAP_NET_RAW))
315 		goto out_rcu_unlock;
316 
317 	sock->ops = answer->ops;
318 	answer_prot = answer->prot;
319 	answer_flags = answer->flags;
320 	rcu_read_unlock();
321 
322 	WARN_ON(!answer_prot->slab);
323 
324 	err = -ENOBUFS;
325 	sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
326 	if (!sk)
327 		goto out;
328 
329 	err = 0;
330 	if (INET_PROTOSW_REUSE & answer_flags)
331 		sk->sk_reuse = SK_CAN_REUSE;
332 
333 	inet = inet_sk(sk);
334 	inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
335 
336 	inet->nodefrag = 0;
337 
338 	if (SOCK_RAW == sock->type) {
339 		inet->inet_num = protocol;
340 		if (IPPROTO_RAW == protocol)
341 			inet->hdrincl = 1;
342 	}
343 
344 	if (net->ipv4.sysctl_ip_no_pmtu_disc)
345 		inet->pmtudisc = IP_PMTUDISC_DONT;
346 	else
347 		inet->pmtudisc = IP_PMTUDISC_WANT;
348 
349 	inet->inet_id = 0;
350 
351 	sock_init_data(sock, sk);
352 
353 	sk->sk_destruct	   = inet_sock_destruct;
354 	sk->sk_protocol	   = protocol;
355 	sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
356 
357 	inet->uc_ttl	= -1;
358 	inet->mc_loop	= 1;
359 	inet->mc_ttl	= 1;
360 	inet->mc_all	= 1;
361 	inet->mc_index	= 0;
362 	inet->mc_list	= NULL;
363 	inet->rcv_tos	= 0;
364 
365 	sk_refcnt_debug_inc(sk);
366 
367 	if (inet->inet_num) {
368 		/* It assumes that any protocol which allows
369 		 * the user to assign a number at socket
370 		 * creation time automatically
371 		 * shares.
372 		 */
373 		inet->inet_sport = htons(inet->inet_num);
374 		/* Add to protocol hash chains. */
375 		err = sk->sk_prot->hash(sk);
376 		if (err) {
377 			sk_common_release(sk);
378 			goto out;
379 		}
380 	}
381 
382 	if (sk->sk_prot->init) {
383 		err = sk->sk_prot->init(sk);
384 		if (err) {
385 			sk_common_release(sk);
386 			goto out;
387 		}
388 	}
389 
390 	if (!kern) {
391 		err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
392 		if (err) {
393 			sk_common_release(sk);
394 			goto out;
395 		}
396 	}
397 out:
398 	return err;
399 out_rcu_unlock:
400 	rcu_read_unlock();
401 	goto out;
402 }
403 
404 
405 /*
406  *	The peer socket should always be NULL (or else). When we call this
407  *	function we are destroying the object and from then on nobody
408  *	should refer to it.
409  */
410 int inet_release(struct socket *sock)
411 {
412 	struct sock *sk = sock->sk;
413 
414 	if (sk) {
415 		long timeout;
416 
417 		/* Applications forget to leave groups before exiting */
418 		ip_mc_drop_socket(sk);
419 
420 		/* If linger is set, we don't return until the close
421 		 * is complete.  Otherwise we return immediately. The
422 		 * actually closing is done the same either way.
423 		 *
424 		 * If the close is due to the process exiting, we never
425 		 * linger..
426 		 */
427 		timeout = 0;
428 		if (sock_flag(sk, SOCK_LINGER) &&
429 		    !(current->flags & PF_EXITING))
430 			timeout = sk->sk_lingertime;
431 		sock->sk = NULL;
432 		sk->sk_prot->close(sk, timeout);
433 	}
434 	return 0;
435 }
436 EXPORT_SYMBOL(inet_release);
437 
438 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
439 {
440 	struct sock *sk = sock->sk;
441 	int err;
442 
443 	/* If the socket has its own bind function then use it. (RAW) */
444 	if (sk->sk_prot->bind) {
445 		return sk->sk_prot->bind(sk, uaddr, addr_len);
446 	}
447 	if (addr_len < sizeof(struct sockaddr_in))
448 		return -EINVAL;
449 
450 	/* BPF prog is run before any checks are done so that if the prog
451 	 * changes context in a wrong way it will be caught.
452 	 */
453 	err = BPF_CGROUP_RUN_PROG_INET4_BIND(sk, uaddr);
454 	if (err)
455 		return err;
456 
457 	return __inet_bind(sk, uaddr, addr_len, false, true);
458 }
459 EXPORT_SYMBOL(inet_bind);
460 
461 int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
462 		bool force_bind_address_no_port, bool with_lock)
463 {
464 	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
465 	struct inet_sock *inet = inet_sk(sk);
466 	struct net *net = sock_net(sk);
467 	unsigned short snum;
468 	int chk_addr_ret;
469 	u32 tb_id = RT_TABLE_LOCAL;
470 	int err;
471 
472 	if (addr->sin_family != AF_INET) {
473 		/* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
474 		 * only if s_addr is INADDR_ANY.
475 		 */
476 		err = -EAFNOSUPPORT;
477 		if (addr->sin_family != AF_UNSPEC ||
478 		    addr->sin_addr.s_addr != htonl(INADDR_ANY))
479 			goto out;
480 	}
481 
482 	tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
483 	chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
484 
485 	/* Not specified by any standard per-se, however it breaks too
486 	 * many applications when removed.  It is unfortunate since
487 	 * allowing applications to make a non-local bind solves
488 	 * several problems with systems using dynamic addressing.
489 	 * (ie. your servers still start up even if your ISDN link
490 	 *  is temporarily down)
491 	 */
492 	err = -EADDRNOTAVAIL;
493 	if (!inet_can_nonlocal_bind(net, inet) &&
494 	    addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
495 	    chk_addr_ret != RTN_LOCAL &&
496 	    chk_addr_ret != RTN_MULTICAST &&
497 	    chk_addr_ret != RTN_BROADCAST)
498 		goto out;
499 
500 	snum = ntohs(addr->sin_port);
501 	err = -EACCES;
502 	if (snum && snum < inet_prot_sock(net) &&
503 	    !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
504 		goto out;
505 
506 	/*      We keep a pair of addresses. rcv_saddr is the one
507 	 *      used by hash lookups, and saddr is used for transmit.
508 	 *
509 	 *      In the BSD API these are the same except where it
510 	 *      would be illegal to use them (multicast/broadcast) in
511 	 *      which case the sending device address is used.
512 	 */
513 	if (with_lock)
514 		lock_sock(sk);
515 
516 	/* Check these errors (active socket, double bind). */
517 	err = -EINVAL;
518 	if (sk->sk_state != TCP_CLOSE || inet->inet_num)
519 		goto out_release_sock;
520 
521 	inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
522 	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
523 		inet->inet_saddr = 0;  /* Use device */
524 
525 	/* Make sure we are allowed to bind here. */
526 	if (snum || !(inet->bind_address_no_port ||
527 		      force_bind_address_no_port)) {
528 		if (sk->sk_prot->get_port(sk, snum)) {
529 			inet->inet_saddr = inet->inet_rcv_saddr = 0;
530 			err = -EADDRINUSE;
531 			goto out_release_sock;
532 		}
533 		err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
534 		if (err) {
535 			inet->inet_saddr = inet->inet_rcv_saddr = 0;
536 			goto out_release_sock;
537 		}
538 	}
539 
540 	if (inet->inet_rcv_saddr)
541 		sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
542 	if (snum)
543 		sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
544 	inet->inet_sport = htons(inet->inet_num);
545 	inet->inet_daddr = 0;
546 	inet->inet_dport = 0;
547 	sk_dst_reset(sk);
548 	err = 0;
549 out_release_sock:
550 	if (with_lock)
551 		release_sock(sk);
552 out:
553 	return err;
554 }
555 
556 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
557 		       int addr_len, int flags)
558 {
559 	struct sock *sk = sock->sk;
560 	int err;
561 
562 	if (addr_len < sizeof(uaddr->sa_family))
563 		return -EINVAL;
564 	if (uaddr->sa_family == AF_UNSPEC)
565 		return sk->sk_prot->disconnect(sk, flags);
566 
567 	if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
568 		err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
569 		if (err)
570 			return err;
571 	}
572 
573 	if (!inet_sk(sk)->inet_num && inet_autobind(sk))
574 		return -EAGAIN;
575 	return sk->sk_prot->connect(sk, uaddr, addr_len);
576 }
577 EXPORT_SYMBOL(inet_dgram_connect);
578 
579 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
580 {
581 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
582 
583 	add_wait_queue(sk_sleep(sk), &wait);
584 	sk->sk_write_pending += writebias;
585 
586 	/* Basic assumption: if someone sets sk->sk_err, he _must_
587 	 * change state of the socket from TCP_SYN_*.
588 	 * Connect() does not allow to get error notifications
589 	 * without closing the socket.
590 	 */
591 	while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
592 		release_sock(sk);
593 		timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
594 		lock_sock(sk);
595 		if (signal_pending(current) || !timeo)
596 			break;
597 	}
598 	remove_wait_queue(sk_sleep(sk), &wait);
599 	sk->sk_write_pending -= writebias;
600 	return timeo;
601 }
602 
603 /*
604  *	Connect to a remote host. There is regrettably still a little
605  *	TCP 'magic' in here.
606  */
607 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
608 			  int addr_len, int flags, int is_sendmsg)
609 {
610 	struct sock *sk = sock->sk;
611 	int err;
612 	long timeo;
613 
614 	/*
615 	 * uaddr can be NULL and addr_len can be 0 if:
616 	 * sk is a TCP fastopen active socket and
617 	 * TCP_FASTOPEN_CONNECT sockopt is set and
618 	 * we already have a valid cookie for this socket.
619 	 * In this case, user can call write() after connect().
620 	 * write() will invoke tcp_sendmsg_fastopen() which calls
621 	 * __inet_stream_connect().
622 	 */
623 	if (uaddr) {
624 		if (addr_len < sizeof(uaddr->sa_family))
625 			return -EINVAL;
626 
627 		if (uaddr->sa_family == AF_UNSPEC) {
628 			err = sk->sk_prot->disconnect(sk, flags);
629 			sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
630 			goto out;
631 		}
632 	}
633 
634 	switch (sock->state) {
635 	default:
636 		err = -EINVAL;
637 		goto out;
638 	case SS_CONNECTED:
639 		err = -EISCONN;
640 		goto out;
641 	case SS_CONNECTING:
642 		if (inet_sk(sk)->defer_connect)
643 			err = is_sendmsg ? -EINPROGRESS : -EISCONN;
644 		else
645 			err = -EALREADY;
646 		/* Fall out of switch with err, set for this state */
647 		break;
648 	case SS_UNCONNECTED:
649 		err = -EISCONN;
650 		if (sk->sk_state != TCP_CLOSE)
651 			goto out;
652 
653 		if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
654 			err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
655 			if (err)
656 				goto out;
657 		}
658 
659 		err = sk->sk_prot->connect(sk, uaddr, addr_len);
660 		if (err < 0)
661 			goto out;
662 
663 		sock->state = SS_CONNECTING;
664 
665 		if (!err && inet_sk(sk)->defer_connect)
666 			goto out;
667 
668 		/* Just entered SS_CONNECTING state; the only
669 		 * difference is that return value in non-blocking
670 		 * case is EINPROGRESS, rather than EALREADY.
671 		 */
672 		err = -EINPROGRESS;
673 		break;
674 	}
675 
676 	timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
677 
678 	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
679 		int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
680 				tcp_sk(sk)->fastopen_req &&
681 				tcp_sk(sk)->fastopen_req->data ? 1 : 0;
682 
683 		/* Error code is set above */
684 		if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
685 			goto out;
686 
687 		err = sock_intr_errno(timeo);
688 		if (signal_pending(current))
689 			goto out;
690 	}
691 
692 	/* Connection was closed by RST, timeout, ICMP error
693 	 * or another process disconnected us.
694 	 */
695 	if (sk->sk_state == TCP_CLOSE)
696 		goto sock_error;
697 
698 	/* sk->sk_err may be not zero now, if RECVERR was ordered by user
699 	 * and error was received after socket entered established state.
700 	 * Hence, it is handled normally after connect() return successfully.
701 	 */
702 
703 	sock->state = SS_CONNECTED;
704 	err = 0;
705 out:
706 	return err;
707 
708 sock_error:
709 	err = sock_error(sk) ? : -ECONNABORTED;
710 	sock->state = SS_UNCONNECTED;
711 	if (sk->sk_prot->disconnect(sk, flags))
712 		sock->state = SS_DISCONNECTING;
713 	goto out;
714 }
715 EXPORT_SYMBOL(__inet_stream_connect);
716 
717 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
718 			int addr_len, int flags)
719 {
720 	int err;
721 
722 	lock_sock(sock->sk);
723 	err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
724 	release_sock(sock->sk);
725 	return err;
726 }
727 EXPORT_SYMBOL(inet_stream_connect);
728 
729 /*
730  *	Accept a pending connection. The TCP layer now gives BSD semantics.
731  */
732 
733 int inet_accept(struct socket *sock, struct socket *newsock, int flags,
734 		bool kern)
735 {
736 	struct sock *sk1 = sock->sk;
737 	int err = -EINVAL;
738 	struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err, kern);
739 
740 	if (!sk2)
741 		goto do_err;
742 
743 	lock_sock(sk2);
744 
745 	sock_rps_record_flow(sk2);
746 	WARN_ON(!((1 << sk2->sk_state) &
747 		  (TCPF_ESTABLISHED | TCPF_SYN_RECV |
748 		  TCPF_CLOSE_WAIT | TCPF_CLOSE)));
749 
750 	sock_graft(sk2, newsock);
751 
752 	newsock->state = SS_CONNECTED;
753 	err = 0;
754 	release_sock(sk2);
755 do_err:
756 	return err;
757 }
758 EXPORT_SYMBOL(inet_accept);
759 
760 
761 /*
762  *	This does both peername and sockname.
763  */
764 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
765 			int peer)
766 {
767 	struct sock *sk		= sock->sk;
768 	struct inet_sock *inet	= inet_sk(sk);
769 	DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
770 
771 	sin->sin_family = AF_INET;
772 	if (peer) {
773 		if (!inet->inet_dport ||
774 		    (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
775 		     peer == 1))
776 			return -ENOTCONN;
777 		sin->sin_port = inet->inet_dport;
778 		sin->sin_addr.s_addr = inet->inet_daddr;
779 	} else {
780 		__be32 addr = inet->inet_rcv_saddr;
781 		if (!addr)
782 			addr = inet->inet_saddr;
783 		sin->sin_port = inet->inet_sport;
784 		sin->sin_addr.s_addr = addr;
785 	}
786 	memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
787 	return sizeof(*sin);
788 }
789 EXPORT_SYMBOL(inet_getname);
790 
791 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
792 {
793 	struct sock *sk = sock->sk;
794 
795 	sock_rps_record_flow(sk);
796 
797 	/* We may need to bind the socket. */
798 	if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
799 	    inet_autobind(sk))
800 		return -EAGAIN;
801 
802 	return sk->sk_prot->sendmsg(sk, msg, size);
803 }
804 EXPORT_SYMBOL(inet_sendmsg);
805 
806 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
807 		      size_t size, int flags)
808 {
809 	struct sock *sk = sock->sk;
810 
811 	sock_rps_record_flow(sk);
812 
813 	/* We may need to bind the socket. */
814 	if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
815 	    inet_autobind(sk))
816 		return -EAGAIN;
817 
818 	if (sk->sk_prot->sendpage)
819 		return sk->sk_prot->sendpage(sk, page, offset, size, flags);
820 	return sock_no_sendpage(sock, page, offset, size, flags);
821 }
822 EXPORT_SYMBOL(inet_sendpage);
823 
824 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
825 		 int flags)
826 {
827 	struct sock *sk = sock->sk;
828 	int addr_len = 0;
829 	int err;
830 
831 	if (likely(!(flags & MSG_ERRQUEUE)))
832 		sock_rps_record_flow(sk);
833 
834 	err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
835 				   flags & ~MSG_DONTWAIT, &addr_len);
836 	if (err >= 0)
837 		msg->msg_namelen = addr_len;
838 	return err;
839 }
840 EXPORT_SYMBOL(inet_recvmsg);
841 
842 int inet_shutdown(struct socket *sock, int how)
843 {
844 	struct sock *sk = sock->sk;
845 	int err = 0;
846 
847 	/* This should really check to make sure
848 	 * the socket is a TCP socket. (WHY AC...)
849 	 */
850 	how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
851 		       1->2 bit 2 snds.
852 		       2->3 */
853 	if ((how & ~SHUTDOWN_MASK) || !how)	/* MAXINT->0 */
854 		return -EINVAL;
855 
856 	lock_sock(sk);
857 	if (sock->state == SS_CONNECTING) {
858 		if ((1 << sk->sk_state) &
859 		    (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
860 			sock->state = SS_DISCONNECTING;
861 		else
862 			sock->state = SS_CONNECTED;
863 	}
864 
865 	switch (sk->sk_state) {
866 	case TCP_CLOSE:
867 		err = -ENOTCONN;
868 		/* Hack to wake up other listeners, who can poll for
869 		   EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
870 		/* fall through */
871 	default:
872 		sk->sk_shutdown |= how;
873 		if (sk->sk_prot->shutdown)
874 			sk->sk_prot->shutdown(sk, how);
875 		break;
876 
877 	/* Remaining two branches are temporary solution for missing
878 	 * close() in multithreaded environment. It is _not_ a good idea,
879 	 * but we have no choice until close() is repaired at VFS level.
880 	 */
881 	case TCP_LISTEN:
882 		if (!(how & RCV_SHUTDOWN))
883 			break;
884 		/* fall through */
885 	case TCP_SYN_SENT:
886 		err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
887 		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
888 		break;
889 	}
890 
891 	/* Wake up anyone sleeping in poll. */
892 	sk->sk_state_change(sk);
893 	release_sock(sk);
894 	return err;
895 }
896 EXPORT_SYMBOL(inet_shutdown);
897 
898 /*
899  *	ioctl() calls you can issue on an INET socket. Most of these are
900  *	device configuration and stuff and very rarely used. Some ioctls
901  *	pass on to the socket itself.
902  *
903  *	NOTE: I like the idea of a module for the config stuff. ie ifconfig
904  *	loads the devconfigure module does its configuring and unloads it.
905  *	There's a good 20K of config code hanging around the kernel.
906  */
907 
908 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
909 {
910 	struct sock *sk = sock->sk;
911 	int err = 0;
912 	struct net *net = sock_net(sk);
913 	void __user *p = (void __user *)arg;
914 	struct ifreq ifr;
915 	struct rtentry rt;
916 
917 	switch (cmd) {
918 	case SIOCGSTAMP:
919 		err = sock_get_timestamp(sk, (struct timeval __user *)arg);
920 		break;
921 	case SIOCGSTAMPNS:
922 		err = sock_get_timestampns(sk, (struct timespec __user *)arg);
923 		break;
924 	case SIOCADDRT:
925 	case SIOCDELRT:
926 		if (copy_from_user(&rt, p, sizeof(struct rtentry)))
927 			return -EFAULT;
928 		err = ip_rt_ioctl(net, cmd, &rt);
929 		break;
930 	case SIOCRTMSG:
931 		err = -EINVAL;
932 		break;
933 	case SIOCDARP:
934 	case SIOCGARP:
935 	case SIOCSARP:
936 		err = arp_ioctl(net, cmd, (void __user *)arg);
937 		break;
938 	case SIOCGIFADDR:
939 	case SIOCGIFBRDADDR:
940 	case SIOCGIFNETMASK:
941 	case SIOCGIFDSTADDR:
942 	case SIOCGIFPFLAGS:
943 		if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
944 			return -EFAULT;
945 		err = devinet_ioctl(net, cmd, &ifr);
946 		if (!err && copy_to_user(p, &ifr, sizeof(struct ifreq)))
947 			err = -EFAULT;
948 		break;
949 
950 	case SIOCSIFADDR:
951 	case SIOCSIFBRDADDR:
952 	case SIOCSIFNETMASK:
953 	case SIOCSIFDSTADDR:
954 	case SIOCSIFPFLAGS:
955 	case SIOCSIFFLAGS:
956 		if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
957 			return -EFAULT;
958 		err = devinet_ioctl(net, cmd, &ifr);
959 		break;
960 	default:
961 		if (sk->sk_prot->ioctl)
962 			err = sk->sk_prot->ioctl(sk, cmd, arg);
963 		else
964 			err = -ENOIOCTLCMD;
965 		break;
966 	}
967 	return err;
968 }
969 EXPORT_SYMBOL(inet_ioctl);
970 
971 #ifdef CONFIG_COMPAT
972 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
973 {
974 	struct sock *sk = sock->sk;
975 	int err = -ENOIOCTLCMD;
976 
977 	if (sk->sk_prot->compat_ioctl)
978 		err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
979 
980 	return err;
981 }
982 #endif
983 
984 const struct proto_ops inet_stream_ops = {
985 	.family		   = PF_INET,
986 	.owner		   = THIS_MODULE,
987 	.release	   = inet_release,
988 	.bind		   = inet_bind,
989 	.connect	   = inet_stream_connect,
990 	.socketpair	   = sock_no_socketpair,
991 	.accept		   = inet_accept,
992 	.getname	   = inet_getname,
993 	.poll		   = tcp_poll,
994 	.ioctl		   = inet_ioctl,
995 	.listen		   = inet_listen,
996 	.shutdown	   = inet_shutdown,
997 	.setsockopt	   = sock_common_setsockopt,
998 	.getsockopt	   = sock_common_getsockopt,
999 	.sendmsg	   = inet_sendmsg,
1000 	.recvmsg	   = inet_recvmsg,
1001 #ifdef CONFIG_MMU
1002 	.mmap		   = tcp_mmap,
1003 #endif
1004 	.sendpage	   = inet_sendpage,
1005 	.splice_read	   = tcp_splice_read,
1006 	.read_sock	   = tcp_read_sock,
1007 	.sendmsg_locked    = tcp_sendmsg_locked,
1008 	.sendpage_locked   = tcp_sendpage_locked,
1009 	.peek_len	   = tcp_peek_len,
1010 #ifdef CONFIG_COMPAT
1011 	.compat_setsockopt = compat_sock_common_setsockopt,
1012 	.compat_getsockopt = compat_sock_common_getsockopt,
1013 	.compat_ioctl	   = inet_compat_ioctl,
1014 #endif
1015 	.set_rcvlowat	   = tcp_set_rcvlowat,
1016 };
1017 EXPORT_SYMBOL(inet_stream_ops);
1018 
1019 const struct proto_ops inet_dgram_ops = {
1020 	.family		   = PF_INET,
1021 	.owner		   = THIS_MODULE,
1022 	.release	   = inet_release,
1023 	.bind		   = inet_bind,
1024 	.connect	   = inet_dgram_connect,
1025 	.socketpair	   = sock_no_socketpair,
1026 	.accept		   = sock_no_accept,
1027 	.getname	   = inet_getname,
1028 	.poll		   = udp_poll,
1029 	.ioctl		   = inet_ioctl,
1030 	.listen		   = sock_no_listen,
1031 	.shutdown	   = inet_shutdown,
1032 	.setsockopt	   = sock_common_setsockopt,
1033 	.getsockopt	   = sock_common_getsockopt,
1034 	.sendmsg	   = inet_sendmsg,
1035 	.recvmsg	   = inet_recvmsg,
1036 	.mmap		   = sock_no_mmap,
1037 	.sendpage	   = inet_sendpage,
1038 	.set_peek_off	   = sk_set_peek_off,
1039 #ifdef CONFIG_COMPAT
1040 	.compat_setsockopt = compat_sock_common_setsockopt,
1041 	.compat_getsockopt = compat_sock_common_getsockopt,
1042 	.compat_ioctl	   = inet_compat_ioctl,
1043 #endif
1044 };
1045 EXPORT_SYMBOL(inet_dgram_ops);
1046 
1047 /*
1048  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1049  * udp_poll
1050  */
1051 static const struct proto_ops inet_sockraw_ops = {
1052 	.family		   = PF_INET,
1053 	.owner		   = THIS_MODULE,
1054 	.release	   = inet_release,
1055 	.bind		   = inet_bind,
1056 	.connect	   = inet_dgram_connect,
1057 	.socketpair	   = sock_no_socketpair,
1058 	.accept		   = sock_no_accept,
1059 	.getname	   = inet_getname,
1060 	.poll		   = datagram_poll,
1061 	.ioctl		   = inet_ioctl,
1062 	.listen		   = sock_no_listen,
1063 	.shutdown	   = inet_shutdown,
1064 	.setsockopt	   = sock_common_setsockopt,
1065 	.getsockopt	   = sock_common_getsockopt,
1066 	.sendmsg	   = inet_sendmsg,
1067 	.recvmsg	   = inet_recvmsg,
1068 	.mmap		   = sock_no_mmap,
1069 	.sendpage	   = inet_sendpage,
1070 #ifdef CONFIG_COMPAT
1071 	.compat_setsockopt = compat_sock_common_setsockopt,
1072 	.compat_getsockopt = compat_sock_common_getsockopt,
1073 	.compat_ioctl	   = inet_compat_ioctl,
1074 #endif
1075 };
1076 
1077 static const struct net_proto_family inet_family_ops = {
1078 	.family = PF_INET,
1079 	.create = inet_create,
1080 	.owner	= THIS_MODULE,
1081 };
1082 
1083 /* Upon startup we insert all the elements in inetsw_array[] into
1084  * the linked list inetsw.
1085  */
1086 static struct inet_protosw inetsw_array[] =
1087 {
1088 	{
1089 		.type =       SOCK_STREAM,
1090 		.protocol =   IPPROTO_TCP,
1091 		.prot =       &tcp_prot,
1092 		.ops =        &inet_stream_ops,
1093 		.flags =      INET_PROTOSW_PERMANENT |
1094 			      INET_PROTOSW_ICSK,
1095 	},
1096 
1097 	{
1098 		.type =       SOCK_DGRAM,
1099 		.protocol =   IPPROTO_UDP,
1100 		.prot =       &udp_prot,
1101 		.ops =        &inet_dgram_ops,
1102 		.flags =      INET_PROTOSW_PERMANENT,
1103        },
1104 
1105        {
1106 		.type =       SOCK_DGRAM,
1107 		.protocol =   IPPROTO_ICMP,
1108 		.prot =       &ping_prot,
1109 		.ops =        &inet_sockraw_ops,
1110 		.flags =      INET_PROTOSW_REUSE,
1111        },
1112 
1113        {
1114 	       .type =       SOCK_RAW,
1115 	       .protocol =   IPPROTO_IP,	/* wild card */
1116 	       .prot =       &raw_prot,
1117 	       .ops =        &inet_sockraw_ops,
1118 	       .flags =      INET_PROTOSW_REUSE,
1119        }
1120 };
1121 
1122 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1123 
1124 void inet_register_protosw(struct inet_protosw *p)
1125 {
1126 	struct list_head *lh;
1127 	struct inet_protosw *answer;
1128 	int protocol = p->protocol;
1129 	struct list_head *last_perm;
1130 
1131 	spin_lock_bh(&inetsw_lock);
1132 
1133 	if (p->type >= SOCK_MAX)
1134 		goto out_illegal;
1135 
1136 	/* If we are trying to override a permanent protocol, bail. */
1137 	last_perm = &inetsw[p->type];
1138 	list_for_each(lh, &inetsw[p->type]) {
1139 		answer = list_entry(lh, struct inet_protosw, list);
1140 		/* Check only the non-wild match. */
1141 		if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1142 			break;
1143 		if (protocol == answer->protocol)
1144 			goto out_permanent;
1145 		last_perm = lh;
1146 	}
1147 
1148 	/* Add the new entry after the last permanent entry if any, so that
1149 	 * the new entry does not override a permanent entry when matched with
1150 	 * a wild-card protocol. But it is allowed to override any existing
1151 	 * non-permanent entry.  This means that when we remove this entry, the
1152 	 * system automatically returns to the old behavior.
1153 	 */
1154 	list_add_rcu(&p->list, last_perm);
1155 out:
1156 	spin_unlock_bh(&inetsw_lock);
1157 
1158 	return;
1159 
1160 out_permanent:
1161 	pr_err("Attempt to override permanent protocol %d\n", protocol);
1162 	goto out;
1163 
1164 out_illegal:
1165 	pr_err("Ignoring attempt to register invalid socket type %d\n",
1166 	       p->type);
1167 	goto out;
1168 }
1169 EXPORT_SYMBOL(inet_register_protosw);
1170 
1171 void inet_unregister_protosw(struct inet_protosw *p)
1172 {
1173 	if (INET_PROTOSW_PERMANENT & p->flags) {
1174 		pr_err("Attempt to unregister permanent protocol %d\n",
1175 		       p->protocol);
1176 	} else {
1177 		spin_lock_bh(&inetsw_lock);
1178 		list_del_rcu(&p->list);
1179 		spin_unlock_bh(&inetsw_lock);
1180 
1181 		synchronize_net();
1182 	}
1183 }
1184 EXPORT_SYMBOL(inet_unregister_protosw);
1185 
1186 static int inet_sk_reselect_saddr(struct sock *sk)
1187 {
1188 	struct inet_sock *inet = inet_sk(sk);
1189 	__be32 old_saddr = inet->inet_saddr;
1190 	__be32 daddr = inet->inet_daddr;
1191 	struct flowi4 *fl4;
1192 	struct rtable *rt;
1193 	__be32 new_saddr;
1194 	struct ip_options_rcu *inet_opt;
1195 
1196 	inet_opt = rcu_dereference_protected(inet->inet_opt,
1197 					     lockdep_sock_is_held(sk));
1198 	if (inet_opt && inet_opt->opt.srr)
1199 		daddr = inet_opt->opt.faddr;
1200 
1201 	/* Query new route. */
1202 	fl4 = &inet->cork.fl.u.ip4;
1203 	rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1204 			      sk->sk_bound_dev_if, sk->sk_protocol,
1205 			      inet->inet_sport, inet->inet_dport, sk);
1206 	if (IS_ERR(rt))
1207 		return PTR_ERR(rt);
1208 
1209 	sk_setup_caps(sk, &rt->dst);
1210 
1211 	new_saddr = fl4->saddr;
1212 
1213 	if (new_saddr == old_saddr)
1214 		return 0;
1215 
1216 	if (sock_net(sk)->ipv4.sysctl_ip_dynaddr > 1) {
1217 		pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1218 			__func__, &old_saddr, &new_saddr);
1219 	}
1220 
1221 	inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1222 
1223 	/*
1224 	 * XXX The only one ugly spot where we need to
1225 	 * XXX really change the sockets identity after
1226 	 * XXX it has entered the hashes. -DaveM
1227 	 *
1228 	 * Besides that, it does not check for connection
1229 	 * uniqueness. Wait for troubles.
1230 	 */
1231 	return __sk_prot_rehash(sk);
1232 }
1233 
1234 int inet_sk_rebuild_header(struct sock *sk)
1235 {
1236 	struct inet_sock *inet = inet_sk(sk);
1237 	struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1238 	__be32 daddr;
1239 	struct ip_options_rcu *inet_opt;
1240 	struct flowi4 *fl4;
1241 	int err;
1242 
1243 	/* Route is OK, nothing to do. */
1244 	if (rt)
1245 		return 0;
1246 
1247 	/* Reroute. */
1248 	rcu_read_lock();
1249 	inet_opt = rcu_dereference(inet->inet_opt);
1250 	daddr = inet->inet_daddr;
1251 	if (inet_opt && inet_opt->opt.srr)
1252 		daddr = inet_opt->opt.faddr;
1253 	rcu_read_unlock();
1254 	fl4 = &inet->cork.fl.u.ip4;
1255 	rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1256 				   inet->inet_dport, inet->inet_sport,
1257 				   sk->sk_protocol, RT_CONN_FLAGS(sk),
1258 				   sk->sk_bound_dev_if);
1259 	if (!IS_ERR(rt)) {
1260 		err = 0;
1261 		sk_setup_caps(sk, &rt->dst);
1262 	} else {
1263 		err = PTR_ERR(rt);
1264 
1265 		/* Routing failed... */
1266 		sk->sk_route_caps = 0;
1267 		/*
1268 		 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1269 		 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1270 		 */
1271 		if (!sock_net(sk)->ipv4.sysctl_ip_dynaddr ||
1272 		    sk->sk_state != TCP_SYN_SENT ||
1273 		    (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1274 		    (err = inet_sk_reselect_saddr(sk)) != 0)
1275 			sk->sk_err_soft = -err;
1276 	}
1277 
1278 	return err;
1279 }
1280 EXPORT_SYMBOL(inet_sk_rebuild_header);
1281 
1282 void inet_sk_set_state(struct sock *sk, int state)
1283 {
1284 	trace_inet_sock_set_state(sk, sk->sk_state, state);
1285 	sk->sk_state = state;
1286 }
1287 EXPORT_SYMBOL(inet_sk_set_state);
1288 
1289 void inet_sk_state_store(struct sock *sk, int newstate)
1290 {
1291 	trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1292 	smp_store_release(&sk->sk_state, newstate);
1293 }
1294 
1295 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1296 				 netdev_features_t features)
1297 {
1298 	bool udpfrag = false, fixedid = false, gso_partial, encap;
1299 	struct sk_buff *segs = ERR_PTR(-EINVAL);
1300 	const struct net_offload *ops;
1301 	unsigned int offset = 0;
1302 	struct iphdr *iph;
1303 	int proto, tot_len;
1304 	int nhoff;
1305 	int ihl;
1306 	int id;
1307 
1308 	skb_reset_network_header(skb);
1309 	nhoff = skb_network_header(skb) - skb_mac_header(skb);
1310 	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1311 		goto out;
1312 
1313 	iph = ip_hdr(skb);
1314 	ihl = iph->ihl * 4;
1315 	if (ihl < sizeof(*iph))
1316 		goto out;
1317 
1318 	id = ntohs(iph->id);
1319 	proto = iph->protocol;
1320 
1321 	/* Warning: after this point, iph might be no longer valid */
1322 	if (unlikely(!pskb_may_pull(skb, ihl)))
1323 		goto out;
1324 	__skb_pull(skb, ihl);
1325 
1326 	encap = SKB_GSO_CB(skb)->encap_level > 0;
1327 	if (encap)
1328 		features &= skb->dev->hw_enc_features;
1329 	SKB_GSO_CB(skb)->encap_level += ihl;
1330 
1331 	skb_reset_transport_header(skb);
1332 
1333 	segs = ERR_PTR(-EPROTONOSUPPORT);
1334 
1335 	if (!skb->encapsulation || encap) {
1336 		udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1337 		fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1338 
1339 		/* fixed ID is invalid if DF bit is not set */
1340 		if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1341 			goto out;
1342 	}
1343 
1344 	ops = rcu_dereference(inet_offloads[proto]);
1345 	if (likely(ops && ops->callbacks.gso_segment))
1346 		segs = ops->callbacks.gso_segment(skb, features);
1347 
1348 	if (IS_ERR_OR_NULL(segs))
1349 		goto out;
1350 
1351 	gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1352 
1353 	skb = segs;
1354 	do {
1355 		iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1356 		if (udpfrag) {
1357 			iph->frag_off = htons(offset >> 3);
1358 			if (skb->next)
1359 				iph->frag_off |= htons(IP_MF);
1360 			offset += skb->len - nhoff - ihl;
1361 			tot_len = skb->len - nhoff;
1362 		} else if (skb_is_gso(skb)) {
1363 			if (!fixedid) {
1364 				iph->id = htons(id);
1365 				id += skb_shinfo(skb)->gso_segs;
1366 			}
1367 
1368 			if (gso_partial)
1369 				tot_len = skb_shinfo(skb)->gso_size +
1370 					  SKB_GSO_CB(skb)->data_offset +
1371 					  skb->head - (unsigned char *)iph;
1372 			else
1373 				tot_len = skb->len - nhoff;
1374 		} else {
1375 			if (!fixedid)
1376 				iph->id = htons(id++);
1377 			tot_len = skb->len - nhoff;
1378 		}
1379 		iph->tot_len = htons(tot_len);
1380 		ip_send_check(iph);
1381 		if (encap)
1382 			skb_reset_inner_headers(skb);
1383 		skb->network_header = (u8 *)iph - skb->head;
1384 		skb_reset_mac_len(skb);
1385 	} while ((skb = skb->next));
1386 
1387 out:
1388 	return segs;
1389 }
1390 EXPORT_SYMBOL(inet_gso_segment);
1391 
1392 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1393 					netdev_features_t features)
1394 {
1395 	if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1396 		return ERR_PTR(-EINVAL);
1397 
1398 	return inet_gso_segment(skb, features);
1399 }
1400 
1401 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *,
1402 							   struct sk_buff *));
1403 INDIRECT_CALLABLE_DECLARE(struct sk_buff *udp4_gro_receive(struct list_head *,
1404 							   struct sk_buff *));
1405 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1406 {
1407 	const struct net_offload *ops;
1408 	struct sk_buff *pp = NULL;
1409 	const struct iphdr *iph;
1410 	struct sk_buff *p;
1411 	unsigned int hlen;
1412 	unsigned int off;
1413 	unsigned int id;
1414 	int flush = 1;
1415 	int proto;
1416 
1417 	off = skb_gro_offset(skb);
1418 	hlen = off + sizeof(*iph);
1419 	iph = skb_gro_header_fast(skb, off);
1420 	if (skb_gro_header_hard(skb, hlen)) {
1421 		iph = skb_gro_header_slow(skb, hlen, off);
1422 		if (unlikely(!iph))
1423 			goto out;
1424 	}
1425 
1426 	proto = iph->protocol;
1427 
1428 	rcu_read_lock();
1429 	ops = rcu_dereference(inet_offloads[proto]);
1430 	if (!ops || !ops->callbacks.gro_receive)
1431 		goto out_unlock;
1432 
1433 	if (*(u8 *)iph != 0x45)
1434 		goto out_unlock;
1435 
1436 	if (ip_is_fragment(iph))
1437 		goto out_unlock;
1438 
1439 	if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1440 		goto out_unlock;
1441 
1442 	id = ntohl(*(__be32 *)&iph->id);
1443 	flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1444 	id >>= 16;
1445 
1446 	list_for_each_entry(p, head, list) {
1447 		struct iphdr *iph2;
1448 		u16 flush_id;
1449 
1450 		if (!NAPI_GRO_CB(p)->same_flow)
1451 			continue;
1452 
1453 		iph2 = (struct iphdr *)(p->data + off);
1454 		/* The above works because, with the exception of the top
1455 		 * (inner most) layer, we only aggregate pkts with the same
1456 		 * hdr length so all the hdrs we'll need to verify will start
1457 		 * at the same offset.
1458 		 */
1459 		if ((iph->protocol ^ iph2->protocol) |
1460 		    ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1461 		    ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1462 			NAPI_GRO_CB(p)->same_flow = 0;
1463 			continue;
1464 		}
1465 
1466 		/* All fields must match except length and checksum. */
1467 		NAPI_GRO_CB(p)->flush |=
1468 			(iph->ttl ^ iph2->ttl) |
1469 			(iph->tos ^ iph2->tos) |
1470 			((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1471 
1472 		NAPI_GRO_CB(p)->flush |= flush;
1473 
1474 		/* We need to store of the IP ID check to be included later
1475 		 * when we can verify that this packet does in fact belong
1476 		 * to a given flow.
1477 		 */
1478 		flush_id = (u16)(id - ntohs(iph2->id));
1479 
1480 		/* This bit of code makes it much easier for us to identify
1481 		 * the cases where we are doing atomic vs non-atomic IP ID
1482 		 * checks.  Specifically an atomic check can return IP ID
1483 		 * values 0 - 0xFFFF, while a non-atomic check can only
1484 		 * return 0 or 0xFFFF.
1485 		 */
1486 		if (!NAPI_GRO_CB(p)->is_atomic ||
1487 		    !(iph->frag_off & htons(IP_DF))) {
1488 			flush_id ^= NAPI_GRO_CB(p)->count;
1489 			flush_id = flush_id ? 0xFFFF : 0;
1490 		}
1491 
1492 		/* If the previous IP ID value was based on an atomic
1493 		 * datagram we can overwrite the value and ignore it.
1494 		 */
1495 		if (NAPI_GRO_CB(skb)->is_atomic)
1496 			NAPI_GRO_CB(p)->flush_id = flush_id;
1497 		else
1498 			NAPI_GRO_CB(p)->flush_id |= flush_id;
1499 	}
1500 
1501 	NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1502 	NAPI_GRO_CB(skb)->flush |= flush;
1503 	skb_set_network_header(skb, off);
1504 	/* The above will be needed by the transport layer if there is one
1505 	 * immediately following this IP hdr.
1506 	 */
1507 
1508 	/* Note : No need to call skb_gro_postpull_rcsum() here,
1509 	 * as we already checked checksum over ipv4 header was 0
1510 	 */
1511 	skb_gro_pull(skb, sizeof(*iph));
1512 	skb_set_transport_header(skb, skb_gro_offset(skb));
1513 
1514 	pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1515 				       ops->callbacks.gro_receive, head, skb);
1516 
1517 out_unlock:
1518 	rcu_read_unlock();
1519 
1520 out:
1521 	skb_gro_flush_final(skb, pp, flush);
1522 
1523 	return pp;
1524 }
1525 EXPORT_SYMBOL(inet_gro_receive);
1526 
1527 static struct sk_buff *ipip_gro_receive(struct list_head *head,
1528 					struct sk_buff *skb)
1529 {
1530 	if (NAPI_GRO_CB(skb)->encap_mark) {
1531 		NAPI_GRO_CB(skb)->flush = 1;
1532 		return NULL;
1533 	}
1534 
1535 	NAPI_GRO_CB(skb)->encap_mark = 1;
1536 
1537 	return inet_gro_receive(head, skb);
1538 }
1539 
1540 #define SECONDS_PER_DAY	86400
1541 
1542 /* inet_current_timestamp - Return IP network timestamp
1543  *
1544  * Return milliseconds since midnight in network byte order.
1545  */
1546 __be32 inet_current_timestamp(void)
1547 {
1548 	u32 secs;
1549 	u32 msecs;
1550 	struct timespec64 ts;
1551 
1552 	ktime_get_real_ts64(&ts);
1553 
1554 	/* Get secs since midnight. */
1555 	(void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1556 	/* Convert to msecs. */
1557 	msecs = secs * MSEC_PER_SEC;
1558 	/* Convert nsec to msec. */
1559 	msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1560 
1561 	/* Convert to network byte order. */
1562 	return htonl(msecs);
1563 }
1564 EXPORT_SYMBOL(inet_current_timestamp);
1565 
1566 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1567 {
1568 	if (sk->sk_family == AF_INET)
1569 		return ip_recv_error(sk, msg, len, addr_len);
1570 #if IS_ENABLED(CONFIG_IPV6)
1571 	if (sk->sk_family == AF_INET6)
1572 		return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1573 #endif
1574 	return -EINVAL;
1575 }
1576 
1577 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *, int));
1578 INDIRECT_CALLABLE_DECLARE(int udp4_gro_complete(struct sk_buff *, int));
1579 int inet_gro_complete(struct sk_buff *skb, int nhoff)
1580 {
1581 	__be16 newlen = htons(skb->len - nhoff);
1582 	struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1583 	const struct net_offload *ops;
1584 	int proto = iph->protocol;
1585 	int err = -ENOSYS;
1586 
1587 	if (skb->encapsulation) {
1588 		skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1589 		skb_set_inner_network_header(skb, nhoff);
1590 	}
1591 
1592 	csum_replace2(&iph->check, iph->tot_len, newlen);
1593 	iph->tot_len = newlen;
1594 
1595 	rcu_read_lock();
1596 	ops = rcu_dereference(inet_offloads[proto]);
1597 	if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1598 		goto out_unlock;
1599 
1600 	/* Only need to add sizeof(*iph) to get to the next hdr below
1601 	 * because any hdr with option will have been flushed in
1602 	 * inet_gro_receive().
1603 	 */
1604 	err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1605 			      tcp4_gro_complete, udp4_gro_complete,
1606 			      skb, nhoff + sizeof(*iph));
1607 
1608 out_unlock:
1609 	rcu_read_unlock();
1610 
1611 	return err;
1612 }
1613 EXPORT_SYMBOL(inet_gro_complete);
1614 
1615 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1616 {
1617 	skb->encapsulation = 1;
1618 	skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1619 	return inet_gro_complete(skb, nhoff);
1620 }
1621 
1622 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1623 			 unsigned short type, unsigned char protocol,
1624 			 struct net *net)
1625 {
1626 	struct socket *sock;
1627 	int rc = sock_create_kern(net, family, type, protocol, &sock);
1628 
1629 	if (rc == 0) {
1630 		*sk = sock->sk;
1631 		(*sk)->sk_allocation = GFP_ATOMIC;
1632 		/*
1633 		 * Unhash it so that IP input processing does not even see it,
1634 		 * we do not wish this socket to see incoming packets.
1635 		 */
1636 		(*sk)->sk_prot->unhash(*sk);
1637 	}
1638 	return rc;
1639 }
1640 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1641 
1642 u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt)
1643 {
1644 	return  *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt);
1645 }
1646 EXPORT_SYMBOL_GPL(snmp_get_cpu_field);
1647 
1648 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1649 {
1650 	unsigned long res = 0;
1651 	int i;
1652 
1653 	for_each_possible_cpu(i)
1654 		res += snmp_get_cpu_field(mib, i, offt);
1655 	return res;
1656 }
1657 EXPORT_SYMBOL_GPL(snmp_fold_field);
1658 
1659 #if BITS_PER_LONG==32
1660 
1661 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1662 			 size_t syncp_offset)
1663 {
1664 	void *bhptr;
1665 	struct u64_stats_sync *syncp;
1666 	u64 v;
1667 	unsigned int start;
1668 
1669 	bhptr = per_cpu_ptr(mib, cpu);
1670 	syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1671 	do {
1672 		start = u64_stats_fetch_begin_irq(syncp);
1673 		v = *(((u64 *)bhptr) + offt);
1674 	} while (u64_stats_fetch_retry_irq(syncp, start));
1675 
1676 	return v;
1677 }
1678 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1679 
1680 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1681 {
1682 	u64 res = 0;
1683 	int cpu;
1684 
1685 	for_each_possible_cpu(cpu) {
1686 		res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1687 	}
1688 	return res;
1689 }
1690 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1691 #endif
1692 
1693 #ifdef CONFIG_IP_MULTICAST
1694 static const struct net_protocol igmp_protocol = {
1695 	.handler =	igmp_rcv,
1696 	.netns_ok =	1,
1697 };
1698 #endif
1699 
1700 /* thinking of making this const? Don't.
1701  * early_demux can change based on sysctl.
1702  */
1703 static struct net_protocol tcp_protocol = {
1704 	.early_demux	=	tcp_v4_early_demux,
1705 	.early_demux_handler =  tcp_v4_early_demux,
1706 	.handler	=	tcp_v4_rcv,
1707 	.err_handler	=	tcp_v4_err,
1708 	.no_policy	=	1,
1709 	.netns_ok	=	1,
1710 	.icmp_strict_tag_validation = 1,
1711 };
1712 
1713 /* thinking of making this const? Don't.
1714  * early_demux can change based on sysctl.
1715  */
1716 static struct net_protocol udp_protocol = {
1717 	.early_demux =	udp_v4_early_demux,
1718 	.early_demux_handler =	udp_v4_early_demux,
1719 	.handler =	udp_rcv,
1720 	.err_handler =	udp_err,
1721 	.no_policy =	1,
1722 	.netns_ok =	1,
1723 };
1724 
1725 static const struct net_protocol icmp_protocol = {
1726 	.handler =	icmp_rcv,
1727 	.err_handler =	icmp_err,
1728 	.no_policy =	1,
1729 	.netns_ok =	1,
1730 };
1731 
1732 static __net_init int ipv4_mib_init_net(struct net *net)
1733 {
1734 	int i;
1735 
1736 	net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1737 	if (!net->mib.tcp_statistics)
1738 		goto err_tcp_mib;
1739 	net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1740 	if (!net->mib.ip_statistics)
1741 		goto err_ip_mib;
1742 
1743 	for_each_possible_cpu(i) {
1744 		struct ipstats_mib *af_inet_stats;
1745 		af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1746 		u64_stats_init(&af_inet_stats->syncp);
1747 	}
1748 
1749 	net->mib.net_statistics = alloc_percpu(struct linux_mib);
1750 	if (!net->mib.net_statistics)
1751 		goto err_net_mib;
1752 	net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1753 	if (!net->mib.udp_statistics)
1754 		goto err_udp_mib;
1755 	net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1756 	if (!net->mib.udplite_statistics)
1757 		goto err_udplite_mib;
1758 	net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1759 	if (!net->mib.icmp_statistics)
1760 		goto err_icmp_mib;
1761 	net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1762 					      GFP_KERNEL);
1763 	if (!net->mib.icmpmsg_statistics)
1764 		goto err_icmpmsg_mib;
1765 
1766 	tcp_mib_init(net);
1767 	return 0;
1768 
1769 err_icmpmsg_mib:
1770 	free_percpu(net->mib.icmp_statistics);
1771 err_icmp_mib:
1772 	free_percpu(net->mib.udplite_statistics);
1773 err_udplite_mib:
1774 	free_percpu(net->mib.udp_statistics);
1775 err_udp_mib:
1776 	free_percpu(net->mib.net_statistics);
1777 err_net_mib:
1778 	free_percpu(net->mib.ip_statistics);
1779 err_ip_mib:
1780 	free_percpu(net->mib.tcp_statistics);
1781 err_tcp_mib:
1782 	return -ENOMEM;
1783 }
1784 
1785 static __net_exit void ipv4_mib_exit_net(struct net *net)
1786 {
1787 	kfree(net->mib.icmpmsg_statistics);
1788 	free_percpu(net->mib.icmp_statistics);
1789 	free_percpu(net->mib.udplite_statistics);
1790 	free_percpu(net->mib.udp_statistics);
1791 	free_percpu(net->mib.net_statistics);
1792 	free_percpu(net->mib.ip_statistics);
1793 	free_percpu(net->mib.tcp_statistics);
1794 }
1795 
1796 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1797 	.init = ipv4_mib_init_net,
1798 	.exit = ipv4_mib_exit_net,
1799 };
1800 
1801 static int __init init_ipv4_mibs(void)
1802 {
1803 	return register_pernet_subsys(&ipv4_mib_ops);
1804 }
1805 
1806 static __net_init int inet_init_net(struct net *net)
1807 {
1808 	/*
1809 	 * Set defaults for local port range
1810 	 */
1811 	seqlock_init(&net->ipv4.ip_local_ports.lock);
1812 	net->ipv4.ip_local_ports.range[0] =  32768;
1813 	net->ipv4.ip_local_ports.range[1] =  60999;
1814 
1815 	seqlock_init(&net->ipv4.ping_group_range.lock);
1816 	/*
1817 	 * Sane defaults - nobody may create ping sockets.
1818 	 * Boot scripts should set this to distro-specific group.
1819 	 */
1820 	net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1821 	net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1822 
1823 	/* Default values for sysctl-controlled parameters.
1824 	 * We set them here, in case sysctl is not compiled.
1825 	 */
1826 	net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1827 	net->ipv4.sysctl_ip_fwd_update_priority = 1;
1828 	net->ipv4.sysctl_ip_dynaddr = 0;
1829 	net->ipv4.sysctl_ip_early_demux = 1;
1830 	net->ipv4.sysctl_udp_early_demux = 1;
1831 	net->ipv4.sysctl_tcp_early_demux = 1;
1832 #ifdef CONFIG_SYSCTL
1833 	net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1834 #endif
1835 
1836 	/* Some igmp sysctl, whose values are always used */
1837 	net->ipv4.sysctl_igmp_max_memberships = 20;
1838 	net->ipv4.sysctl_igmp_max_msf = 10;
1839 	/* IGMP reports for link-local multicast groups are enabled by default */
1840 	net->ipv4.sysctl_igmp_llm_reports = 1;
1841 	net->ipv4.sysctl_igmp_qrv = 2;
1842 
1843 	return 0;
1844 }
1845 
1846 static __net_exit void inet_exit_net(struct net *net)
1847 {
1848 }
1849 
1850 static __net_initdata struct pernet_operations af_inet_ops = {
1851 	.init = inet_init_net,
1852 	.exit = inet_exit_net,
1853 };
1854 
1855 static int __init init_inet_pernet_ops(void)
1856 {
1857 	return register_pernet_subsys(&af_inet_ops);
1858 }
1859 
1860 static int ipv4_proc_init(void);
1861 
1862 /*
1863  *	IP protocol layer initialiser
1864  */
1865 
1866 static struct packet_offload ip_packet_offload __read_mostly = {
1867 	.type = cpu_to_be16(ETH_P_IP),
1868 	.callbacks = {
1869 		.gso_segment = inet_gso_segment,
1870 		.gro_receive = inet_gro_receive,
1871 		.gro_complete = inet_gro_complete,
1872 	},
1873 };
1874 
1875 static const struct net_offload ipip_offload = {
1876 	.callbacks = {
1877 		.gso_segment	= ipip_gso_segment,
1878 		.gro_receive	= ipip_gro_receive,
1879 		.gro_complete	= ipip_gro_complete,
1880 	},
1881 };
1882 
1883 static int __init ipip_offload_init(void)
1884 {
1885 	return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1886 }
1887 
1888 static int __init ipv4_offload_init(void)
1889 {
1890 	/*
1891 	 * Add offloads
1892 	 */
1893 	if (udpv4_offload_init() < 0)
1894 		pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1895 	if (tcpv4_offload_init() < 0)
1896 		pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1897 	if (ipip_offload_init() < 0)
1898 		pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1899 
1900 	dev_add_offload(&ip_packet_offload);
1901 	return 0;
1902 }
1903 
1904 fs_initcall(ipv4_offload_init);
1905 
1906 static struct packet_type ip_packet_type __read_mostly = {
1907 	.type = cpu_to_be16(ETH_P_IP),
1908 	.func = ip_rcv,
1909 	.list_func = ip_list_rcv,
1910 };
1911 
1912 static int __init inet_init(void)
1913 {
1914 	struct inet_protosw *q;
1915 	struct list_head *r;
1916 	int rc = -EINVAL;
1917 
1918 	sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1919 
1920 	rc = proto_register(&tcp_prot, 1);
1921 	if (rc)
1922 		goto out;
1923 
1924 	rc = proto_register(&udp_prot, 1);
1925 	if (rc)
1926 		goto out_unregister_tcp_proto;
1927 
1928 	rc = proto_register(&raw_prot, 1);
1929 	if (rc)
1930 		goto out_unregister_udp_proto;
1931 
1932 	rc = proto_register(&ping_prot, 1);
1933 	if (rc)
1934 		goto out_unregister_raw_proto;
1935 
1936 	/*
1937 	 *	Tell SOCKET that we are alive...
1938 	 */
1939 
1940 	(void)sock_register(&inet_family_ops);
1941 
1942 #ifdef CONFIG_SYSCTL
1943 	ip_static_sysctl_init();
1944 #endif
1945 
1946 	/*
1947 	 *	Add all the base protocols.
1948 	 */
1949 
1950 	if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1951 		pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1952 	if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1953 		pr_crit("%s: Cannot add UDP protocol\n", __func__);
1954 	if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1955 		pr_crit("%s: Cannot add TCP protocol\n", __func__);
1956 #ifdef CONFIG_IP_MULTICAST
1957 	if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1958 		pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1959 #endif
1960 
1961 	/* Register the socket-side information for inet_create. */
1962 	for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1963 		INIT_LIST_HEAD(r);
1964 
1965 	for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1966 		inet_register_protosw(q);
1967 
1968 	/*
1969 	 *	Set the ARP module up
1970 	 */
1971 
1972 	arp_init();
1973 
1974 	/*
1975 	 *	Set the IP module up
1976 	 */
1977 
1978 	ip_init();
1979 
1980 	/* Setup TCP slab cache for open requests. */
1981 	tcp_init();
1982 
1983 	/* Setup UDP memory threshold */
1984 	udp_init();
1985 
1986 	/* Add UDP-Lite (RFC 3828) */
1987 	udplite4_register();
1988 
1989 	raw_init();
1990 
1991 	ping_init();
1992 
1993 	/*
1994 	 *	Set the ICMP layer up
1995 	 */
1996 
1997 	if (icmp_init() < 0)
1998 		panic("Failed to create the ICMP control socket.\n");
1999 
2000 	/*
2001 	 *	Initialise the multicast router
2002 	 */
2003 #if defined(CONFIG_IP_MROUTE)
2004 	if (ip_mr_init())
2005 		pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2006 #endif
2007 
2008 	if (init_inet_pernet_ops())
2009 		pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2010 	/*
2011 	 *	Initialise per-cpu ipv4 mibs
2012 	 */
2013 
2014 	if (init_ipv4_mibs())
2015 		pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
2016 
2017 	ipv4_proc_init();
2018 
2019 	ipfrag_init();
2020 
2021 	dev_add_pack(&ip_packet_type);
2022 
2023 	ip_tunnel_core_init();
2024 
2025 	rc = 0;
2026 out:
2027 	return rc;
2028 out_unregister_raw_proto:
2029 	proto_unregister(&raw_prot);
2030 out_unregister_udp_proto:
2031 	proto_unregister(&udp_prot);
2032 out_unregister_tcp_proto:
2033 	proto_unregister(&tcp_prot);
2034 	goto out;
2035 }
2036 
2037 fs_initcall(inet_init);
2038 
2039 /* ------------------------------------------------------------------------ */
2040 
2041 #ifdef CONFIG_PROC_FS
2042 static int __init ipv4_proc_init(void)
2043 {
2044 	int rc = 0;
2045 
2046 	if (raw_proc_init())
2047 		goto out_raw;
2048 	if (tcp4_proc_init())
2049 		goto out_tcp;
2050 	if (udp4_proc_init())
2051 		goto out_udp;
2052 	if (ping_proc_init())
2053 		goto out_ping;
2054 	if (ip_misc_proc_init())
2055 		goto out_misc;
2056 out:
2057 	return rc;
2058 out_misc:
2059 	ping_proc_exit();
2060 out_ping:
2061 	udp4_proc_exit();
2062 out_udp:
2063 	tcp4_proc_exit();
2064 out_tcp:
2065 	raw_proc_exit();
2066 out_raw:
2067 	rc = -ENOMEM;
2068 	goto out;
2069 }
2070 
2071 #else /* CONFIG_PROC_FS */
2072 static int __init ipv4_proc_init(void)
2073 {
2074 	return 0;
2075 }
2076 #endif /* CONFIG_PROC_FS */
2077