xref: /linux/net/ipv4/af_inet.c (revision 74f1af95820fc2ee580a775a3a17c416db30b38c)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		PF_INET protocol family socket handler.
8  *
9  * Authors:	Ross Biro
10  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11  *		Florian La Roche, <flla@stud.uni-sb.de>
12  *		Alan Cox, <A.Cox@swansea.ac.uk>
13  *
14  * Changes (see also sock.c)
15  *
16  *		piggy,
17  *		Karl Knutson	:	Socket protocol table
18  *		A.N.Kuznetsov	:	Socket death error in accept().
19  *		John Richardson :	Fix non blocking error in connect()
20  *					so sockets that fail to connect
21  *					don't return -EINPROGRESS.
22  *		Alan Cox	:	Asynchronous I/O support
23  *		Alan Cox	:	Keep correct socket pointer on sock
24  *					structures
25  *					when accept() ed
26  *		Alan Cox	:	Semantics of SO_LINGER aren't state
27  *					moved to close when you look carefully.
28  *					With this fixed and the accept bug fixed
29  *					some RPC stuff seems happier.
30  *		Niibe Yutaka	:	4.4BSD style write async I/O
31  *		Alan Cox,
32  *		Tony Gale 	:	Fixed reuse semantics.
33  *		Alan Cox	:	bind() shouldn't abort existing but dead
34  *					sockets. Stops FTP netin:.. I hope.
35  *		Alan Cox	:	bind() works correctly for RAW sockets.
36  *					Note that FreeBSD at least was broken
37  *					in this respect so be careful with
38  *					compatibility tests...
39  *		Alan Cox	:	routing cache support
40  *		Alan Cox	:	memzero the socket structure for
41  *					compactness.
42  *		Matt Day	:	nonblock connect error handler
43  *		Alan Cox	:	Allow large numbers of pending sockets
44  *					(eg for big web sites), but only if
45  *					specifically application requested.
46  *		Alan Cox	:	New buffering throughout IP. Used
47  *					dumbly.
48  *		Alan Cox	:	New buffering now used smartly.
49  *		Alan Cox	:	BSD rather than common sense
50  *					interpretation of listen.
51  *		Germano Caronni	:	Assorted small races.
52  *		Alan Cox	:	sendmsg/recvmsg basic support.
53  *		Alan Cox	:	Only sendmsg/recvmsg now supported.
54  *		Alan Cox	:	Locked down bind (see security list).
55  *		Alan Cox	:	Loosened bind a little.
56  *		Mike McLagan	:	ADD/DEL DLCI Ioctls
57  *	Willy Konynenberg	:	Transparent proxying support.
58  *		David S. Miller	:	New socket lookup architecture.
59  *					Some other random speedups.
60  *		Cyrus Durgin	:	Cleaned up file for kmod hacks.
61  *		Andi Kleen	:	Fix inet_stream_connect TCP race.
62  */
63 
64 #define pr_fmt(fmt) "IPv4: " fmt
65 
66 #include <linux/err.h>
67 #include <linux/errno.h>
68 #include <linux/types.h>
69 #include <linux/socket.h>
70 #include <linux/in.h>
71 #include <linux/kernel.h>
72 #include <linux/kmod.h>
73 #include <linux/sched.h>
74 #include <linux/timer.h>
75 #include <linux/string.h>
76 #include <linux/sockios.h>
77 #include <linux/net.h>
78 #include <linux/capability.h>
79 #include <linux/fcntl.h>
80 #include <linux/mm.h>
81 #include <linux/interrupt.h>
82 #include <linux/stat.h>
83 #include <linux/init.h>
84 #include <linux/poll.h>
85 #include <linux/netfilter_ipv4.h>
86 #include <linux/random.h>
87 #include <linux/slab.h>
88 
89 #include <linux/uaccess.h>
90 
91 #include <linux/inet.h>
92 #include <linux/igmp.h>
93 #include <linux/inetdevice.h>
94 #include <linux/netdevice.h>
95 #include <net/checksum.h>
96 #include <net/ip.h>
97 #include <net/protocol.h>
98 #include <net/arp.h>
99 #include <net/route.h>
100 #include <net/ip_fib.h>
101 #include <net/inet_connection_sock.h>
102 #include <net/gro.h>
103 #include <net/gso.h>
104 #include <net/tcp.h>
105 #include <net/udp.h>
106 #include <net/udplite.h>
107 #include <net/ping.h>
108 #include <linux/skbuff.h>
109 #include <net/sock.h>
110 #include <net/raw.h>
111 #include <net/icmp.h>
112 #include <net/inet_common.h>
113 #include <net/ip_tunnels.h>
114 #include <net/xfrm.h>
115 #include <net/net_namespace.h>
116 #include <net/secure_seq.h>
117 #ifdef CONFIG_IP_MROUTE
118 #include <linux/mroute.h>
119 #endif
120 #include <net/l3mdev.h>
121 #include <net/compat.h>
122 #include <net/rps.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_final(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_ONCE(atomic_read(&sk->sk_rmem_alloc));
154 	WARN_ON_ONCE(refcount_read(&sk->sk_wmem_alloc));
155 	WARN_ON_ONCE(sk->sk_wmem_queued);
156 	WARN_ON_ONCE(sk->sk_forward_alloc);
157 
158 	kfree(rcu_dereference_protected(inet->inet_opt, 1));
159 	dst_release(rcu_dereference_protected(sk->sk_dst_cache, 1));
160 	dst_release(rcu_dereference_protected(sk->sk_rx_dst, 1));
161 }
162 EXPORT_SYMBOL(inet_sock_destruct);
163 
164 /*
165  *	The routines beyond this point handle the behaviour of an AF_INET
166  *	socket object. Mostly it punts to the subprotocols of IP to do
167  *	the work.
168  */
169 
170 /*
171  *	Automatically bind an unbound socket.
172  */
173 
174 static int inet_autobind(struct sock *sk)
175 {
176 	struct inet_sock *inet;
177 	/* We may need to bind the socket. */
178 	lock_sock(sk);
179 	inet = inet_sk(sk);
180 	if (!inet->inet_num) {
181 		if (sk->sk_prot->get_port(sk, 0)) {
182 			release_sock(sk);
183 			return -EAGAIN;
184 		}
185 		inet->inet_sport = htons(inet->inet_num);
186 	}
187 	release_sock(sk);
188 	return 0;
189 }
190 
191 int __inet_listen_sk(struct sock *sk, int backlog)
192 {
193 	unsigned char old_state = sk->sk_state;
194 	int err, tcp_fastopen;
195 
196 	if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
197 		return -EINVAL;
198 
199 	WRITE_ONCE(sk->sk_max_ack_backlog, backlog);
200 	/* Really, if the socket is already in listen state
201 	 * we can only allow the backlog to be adjusted.
202 	 */
203 	if (old_state != TCP_LISTEN) {
204 		/* Enable TFO w/o requiring TCP_FASTOPEN socket option.
205 		 * Note that only TCP sockets (SOCK_STREAM) will reach here.
206 		 * Also fastopen backlog may already been set via the option
207 		 * because the socket was in TCP_LISTEN state previously but
208 		 * was shutdown() rather than close().
209 		 */
210 		tcp_fastopen = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fastopen);
211 		if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
212 		    (tcp_fastopen & TFO_SERVER_ENABLE) &&
213 		    !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
214 			fastopen_queue_tune(sk, backlog);
215 			tcp_fastopen_init_key_once(sock_net(sk));
216 		}
217 
218 		err = inet_csk_listen_start(sk);
219 		if (err)
220 			return err;
221 
222 		tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL);
223 	}
224 	return 0;
225 }
226 
227 /*
228  *	Move a socket into listening state.
229  */
230 int inet_listen(struct socket *sock, int backlog)
231 {
232 	struct sock *sk = sock->sk;
233 	int err = -EINVAL;
234 
235 	lock_sock(sk);
236 
237 	if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
238 		goto out;
239 
240 	err = __inet_listen_sk(sk, backlog);
241 
242 out:
243 	release_sock(sk);
244 	return err;
245 }
246 EXPORT_SYMBOL(inet_listen);
247 
248 /*
249  *	Create an inet socket.
250  */
251 
252 static int inet_create(struct net *net, struct socket *sock, int protocol,
253 		       int kern)
254 {
255 	struct sock *sk;
256 	struct inet_protosw *answer;
257 	struct inet_sock *inet;
258 	struct proto *answer_prot;
259 	unsigned char answer_flags;
260 	int try_loading_module = 0;
261 	int err;
262 
263 	if (protocol < 0 || protocol >= IPPROTO_MAX)
264 		return -EINVAL;
265 
266 	sock->state = SS_UNCONNECTED;
267 
268 	/* Look for the requested type/protocol pair. */
269 lookup_protocol:
270 	err = -ESOCKTNOSUPPORT;
271 	rcu_read_lock();
272 	list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
273 
274 		err = 0;
275 		/* Check the non-wild match. */
276 		if (protocol == answer->protocol) {
277 			if (protocol != IPPROTO_IP)
278 				break;
279 		} else {
280 			/* Check for the two wild cases. */
281 			if (IPPROTO_IP == protocol) {
282 				protocol = answer->protocol;
283 				break;
284 			}
285 			if (IPPROTO_IP == answer->protocol)
286 				break;
287 		}
288 		err = -EPROTONOSUPPORT;
289 	}
290 
291 	if (unlikely(err)) {
292 		if (try_loading_module < 2) {
293 			rcu_read_unlock();
294 			/*
295 			 * Be more specific, e.g. net-pf-2-proto-132-type-1
296 			 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
297 			 */
298 			if (++try_loading_module == 1)
299 				request_module("net-pf-%d-proto-%d-type-%d",
300 					       PF_INET, protocol, sock->type);
301 			/*
302 			 * Fall back to generic, e.g. net-pf-2-proto-132
303 			 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
304 			 */
305 			else
306 				request_module("net-pf-%d-proto-%d",
307 					       PF_INET, protocol);
308 			goto lookup_protocol;
309 		} else
310 			goto out_rcu_unlock;
311 	}
312 
313 	err = -EPERM;
314 	if (sock->type == SOCK_RAW && !kern &&
315 	    !ns_capable(net->user_ns, CAP_NET_RAW))
316 		goto out_rcu_unlock;
317 
318 	sock->ops = answer->ops;
319 	answer_prot = answer->prot;
320 	answer_flags = answer->flags;
321 	rcu_read_unlock();
322 
323 	WARN_ON(!answer_prot->slab);
324 
325 	err = -ENOMEM;
326 	sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
327 	if (!sk)
328 		goto out;
329 
330 	err = 0;
331 	if (INET_PROTOSW_REUSE & answer_flags)
332 		sk->sk_reuse = SK_CAN_REUSE;
333 
334 	if (INET_PROTOSW_ICSK & answer_flags)
335 		inet_init_csk_locks(sk);
336 
337 	inet = inet_sk(sk);
338 	inet_assign_bit(IS_ICSK, sk, INET_PROTOSW_ICSK & answer_flags);
339 
340 	inet_clear_bit(NODEFRAG, sk);
341 
342 	if (SOCK_RAW == sock->type) {
343 		inet->inet_num = protocol;
344 		if (IPPROTO_RAW == protocol)
345 			inet_set_bit(HDRINCL, sk);
346 	}
347 
348 	if (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc))
349 		inet->pmtudisc = IP_PMTUDISC_DONT;
350 	else
351 		inet->pmtudisc = IP_PMTUDISC_WANT;
352 
353 	atomic_set(&inet->inet_id, 0);
354 
355 	sock_init_data(sock, sk);
356 
357 	sk->sk_destruct	   = inet_sock_destruct;
358 	sk->sk_protocol	   = protocol;
359 	sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
360 	sk->sk_txrehash = READ_ONCE(net->core.sysctl_txrehash);
361 
362 	inet->uc_ttl	= -1;
363 	inet_set_bit(MC_LOOP, sk);
364 	inet->mc_ttl	= 1;
365 	inet_set_bit(MC_ALL, sk);
366 	inet->mc_index	= 0;
367 	inet->mc_list	= NULL;
368 	inet->rcv_tos	= 0;
369 
370 	if (inet->inet_num) {
371 		/* It assumes that any protocol which allows
372 		 * the user to assign a number at socket
373 		 * creation time automatically
374 		 * shares.
375 		 */
376 		inet->inet_sport = htons(inet->inet_num);
377 		/* Add to protocol hash chains. */
378 		err = sk->sk_prot->hash(sk);
379 		if (err)
380 			goto out_sk_release;
381 	}
382 
383 	if (sk->sk_prot->init) {
384 		err = sk->sk_prot->init(sk);
385 		if (err)
386 			goto out_sk_release;
387 	}
388 
389 	if (!kern) {
390 		err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
391 		if (err)
392 			goto out_sk_release;
393 	}
394 out:
395 	return err;
396 out_rcu_unlock:
397 	rcu_read_unlock();
398 	goto out;
399 out_sk_release:
400 	sk_common_release(sk);
401 	sock->sk = NULL;
402 	goto out;
403 }
404 
405 
406 /*
407  *	The peer socket should always be NULL (or else). When we call this
408  *	function we are destroying the object and from then on nobody
409  *	should refer to it.
410  */
411 int inet_release(struct socket *sock)
412 {
413 	struct sock *sk = sock->sk;
414 
415 	if (sk) {
416 		long timeout;
417 
418 		if (!sk->sk_kern_sock)
419 			BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk);
420 
421 		/* Applications forget to leave groups before exiting */
422 		ip_mc_drop_socket(sk);
423 
424 		/* If linger is set, we don't return until the close
425 		 * is complete.  Otherwise we return immediately. The
426 		 * actually closing is done the same either way.
427 		 *
428 		 * If the close is due to the process exiting, we never
429 		 * linger..
430 		 */
431 		timeout = 0;
432 		if (sock_flag(sk, SOCK_LINGER) &&
433 		    !(current->flags & PF_EXITING))
434 			timeout = sk->sk_lingertime;
435 		sk->sk_prot->close(sk, timeout);
436 		sock->sk = NULL;
437 	}
438 	return 0;
439 }
440 EXPORT_SYMBOL(inet_release);
441 
442 int inet_bind_sk(struct sock *sk, struct sockaddr *uaddr, int addr_len)
443 {
444 	u32 flags = BIND_WITH_LOCK;
445 	int err;
446 
447 	/* If the socket has its own bind function then use it. (RAW) */
448 	if (sk->sk_prot->bind) {
449 		return sk->sk_prot->bind(sk, uaddr, addr_len);
450 	}
451 	if (addr_len < sizeof(struct sockaddr_in))
452 		return -EINVAL;
453 
454 	/* BPF prog is run before any checks are done so that if the prog
455 	 * changes context in a wrong way it will be caught.
456 	 */
457 	err = BPF_CGROUP_RUN_PROG_INET_BIND_LOCK(sk, uaddr, &addr_len,
458 						 CGROUP_INET4_BIND, &flags);
459 	if (err)
460 		return err;
461 
462 	return __inet_bind(sk, uaddr, addr_len, flags);
463 }
464 
465 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
466 {
467 	return inet_bind_sk(sock->sk, uaddr, addr_len);
468 }
469 EXPORT_SYMBOL(inet_bind);
470 
471 int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
472 		u32 flags)
473 {
474 	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
475 	struct inet_sock *inet = inet_sk(sk);
476 	struct net *net = sock_net(sk);
477 	unsigned short snum;
478 	int chk_addr_ret;
479 	u32 tb_id = RT_TABLE_LOCAL;
480 	int err;
481 
482 	if (addr->sin_family != AF_INET) {
483 		/* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
484 		 * only if s_addr is INADDR_ANY.
485 		 */
486 		err = -EAFNOSUPPORT;
487 		if (addr->sin_family != AF_UNSPEC ||
488 		    addr->sin_addr.s_addr != htonl(INADDR_ANY))
489 			goto out;
490 	}
491 
492 	tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
493 	chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
494 
495 	/* Not specified by any standard per-se, however it breaks too
496 	 * many applications when removed.  It is unfortunate since
497 	 * allowing applications to make a non-local bind solves
498 	 * several problems with systems using dynamic addressing.
499 	 * (ie. your servers still start up even if your ISDN link
500 	 *  is temporarily down)
501 	 */
502 	err = -EADDRNOTAVAIL;
503 	if (!inet_addr_valid_or_nonlocal(net, inet, addr->sin_addr.s_addr,
504 	                                 chk_addr_ret))
505 		goto out;
506 
507 	snum = ntohs(addr->sin_port);
508 	err = -EACCES;
509 	if (!(flags & BIND_NO_CAP_NET_BIND_SERVICE) &&
510 	    snum && inet_port_requires_bind_service(net, snum) &&
511 	    !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
512 		goto out;
513 
514 	/*      We keep a pair of addresses. rcv_saddr is the one
515 	 *      used by hash lookups, and saddr is used for transmit.
516 	 *
517 	 *      In the BSD API these are the same except where it
518 	 *      would be illegal to use them (multicast/broadcast) in
519 	 *      which case the sending device address is used.
520 	 */
521 	if (flags & BIND_WITH_LOCK)
522 		lock_sock(sk);
523 
524 	/* Check these errors (active socket, double bind). */
525 	err = -EINVAL;
526 	if (sk->sk_state != TCP_CLOSE || inet->inet_num)
527 		goto out_release_sock;
528 
529 	inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
530 	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
531 		inet->inet_saddr = 0;  /* Use device */
532 
533 	/* Make sure we are allowed to bind here. */
534 	if (snum || !(inet_test_bit(BIND_ADDRESS_NO_PORT, sk) ||
535 		      (flags & BIND_FORCE_ADDRESS_NO_PORT))) {
536 		err = sk->sk_prot->get_port(sk, snum);
537 		if (err) {
538 			inet->inet_saddr = inet->inet_rcv_saddr = 0;
539 			goto out_release_sock;
540 		}
541 		if (!(flags & BIND_FROM_BPF)) {
542 			err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
543 			if (err) {
544 				inet->inet_saddr = inet->inet_rcv_saddr = 0;
545 				if (sk->sk_prot->put_port)
546 					sk->sk_prot->put_port(sk);
547 				goto out_release_sock;
548 			}
549 		}
550 	}
551 
552 	if (inet->inet_rcv_saddr)
553 		sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
554 	if (snum)
555 		sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
556 	inet->inet_sport = htons(inet->inet_num);
557 	inet->inet_daddr = 0;
558 	inet->inet_dport = 0;
559 	sk_dst_reset(sk);
560 	err = 0;
561 out_release_sock:
562 	if (flags & BIND_WITH_LOCK)
563 		release_sock(sk);
564 out:
565 	return err;
566 }
567 
568 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
569 		       int addr_len, int flags)
570 {
571 	struct sock *sk = sock->sk;
572 	const struct proto *prot;
573 	int err;
574 
575 	if (addr_len < sizeof(uaddr->sa_family))
576 		return -EINVAL;
577 
578 	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
579 	prot = READ_ONCE(sk->sk_prot);
580 
581 	if (uaddr->sa_family == AF_UNSPEC)
582 		return prot->disconnect(sk, flags);
583 
584 	if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
585 		err = prot->pre_connect(sk, uaddr, addr_len);
586 		if (err)
587 			return err;
588 	}
589 
590 	if (data_race(!inet_sk(sk)->inet_num) && inet_autobind(sk))
591 		return -EAGAIN;
592 	return prot->connect(sk, uaddr, addr_len);
593 }
594 EXPORT_SYMBOL(inet_dgram_connect);
595 
596 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
597 {
598 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
599 
600 	add_wait_queue(sk_sleep(sk), &wait);
601 	sk->sk_write_pending += writebias;
602 
603 	/* Basic assumption: if someone sets sk->sk_err, he _must_
604 	 * change state of the socket from TCP_SYN_*.
605 	 * Connect() does not allow to get error notifications
606 	 * without closing the socket.
607 	 */
608 	while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
609 		release_sock(sk);
610 		timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
611 		lock_sock(sk);
612 		if (signal_pending(current) || !timeo)
613 			break;
614 	}
615 	remove_wait_queue(sk_sleep(sk), &wait);
616 	sk->sk_write_pending -= writebias;
617 	return timeo;
618 }
619 
620 /*
621  *	Connect to a remote host. There is regrettably still a little
622  *	TCP 'magic' in here.
623  */
624 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
625 			  int addr_len, int flags, int is_sendmsg)
626 {
627 	struct sock *sk = sock->sk;
628 	int err;
629 	long timeo;
630 
631 	/*
632 	 * uaddr can be NULL and addr_len can be 0 if:
633 	 * sk is a TCP fastopen active socket and
634 	 * TCP_FASTOPEN_CONNECT sockopt is set and
635 	 * we already have a valid cookie for this socket.
636 	 * In this case, user can call write() after connect().
637 	 * write() will invoke tcp_sendmsg_fastopen() which calls
638 	 * __inet_stream_connect().
639 	 */
640 	if (uaddr) {
641 		if (addr_len < sizeof(uaddr->sa_family))
642 			return -EINVAL;
643 
644 		if (uaddr->sa_family == AF_UNSPEC) {
645 			sk->sk_disconnects++;
646 			err = sk->sk_prot->disconnect(sk, flags);
647 			sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
648 			goto out;
649 		}
650 	}
651 
652 	switch (sock->state) {
653 	default:
654 		err = -EINVAL;
655 		goto out;
656 	case SS_CONNECTED:
657 		err = -EISCONN;
658 		goto out;
659 	case SS_CONNECTING:
660 		if (inet_test_bit(DEFER_CONNECT, sk))
661 			err = is_sendmsg ? -EINPROGRESS : -EISCONN;
662 		else
663 			err = -EALREADY;
664 		/* Fall out of switch with err, set for this state */
665 		break;
666 	case SS_UNCONNECTED:
667 		err = -EISCONN;
668 		if (sk->sk_state != TCP_CLOSE)
669 			goto out;
670 
671 		if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
672 			err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
673 			if (err)
674 				goto out;
675 		}
676 
677 		err = sk->sk_prot->connect(sk, uaddr, addr_len);
678 		if (err < 0)
679 			goto out;
680 
681 		sock->state = SS_CONNECTING;
682 
683 		if (!err && inet_test_bit(DEFER_CONNECT, sk))
684 			goto out;
685 
686 		/* Just entered SS_CONNECTING state; the only
687 		 * difference is that return value in non-blocking
688 		 * case is EINPROGRESS, rather than EALREADY.
689 		 */
690 		err = -EINPROGRESS;
691 		break;
692 	}
693 
694 	timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
695 
696 	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
697 		int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
698 				tcp_sk(sk)->fastopen_req &&
699 				tcp_sk(sk)->fastopen_req->data ? 1 : 0;
700 		int dis = sk->sk_disconnects;
701 
702 		/* Error code is set above */
703 		if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
704 			goto out;
705 
706 		err = sock_intr_errno(timeo);
707 		if (signal_pending(current))
708 			goto out;
709 
710 		if (dis != sk->sk_disconnects) {
711 			err = -EPIPE;
712 			goto out;
713 		}
714 	}
715 
716 	/* Connection was closed by RST, timeout, ICMP error
717 	 * or another process disconnected us.
718 	 */
719 	if (sk->sk_state == TCP_CLOSE)
720 		goto sock_error;
721 
722 	/* sk->sk_err may be not zero now, if RECVERR was ordered by user
723 	 * and error was received after socket entered established state.
724 	 * Hence, it is handled normally after connect() return successfully.
725 	 */
726 
727 	sock->state = SS_CONNECTED;
728 	err = 0;
729 out:
730 	return err;
731 
732 sock_error:
733 	err = sock_error(sk) ? : -ECONNABORTED;
734 	sock->state = SS_UNCONNECTED;
735 	sk->sk_disconnects++;
736 	if (sk->sk_prot->disconnect(sk, flags))
737 		sock->state = SS_DISCONNECTING;
738 	goto out;
739 }
740 EXPORT_SYMBOL(__inet_stream_connect);
741 
742 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
743 			int addr_len, int flags)
744 {
745 	int err;
746 
747 	lock_sock(sock->sk);
748 	err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
749 	release_sock(sock->sk);
750 	return err;
751 }
752 EXPORT_SYMBOL(inet_stream_connect);
753 
754 void __inet_accept(struct socket *sock, struct socket *newsock, struct sock *newsk)
755 {
756 	sock_rps_record_flow(newsk);
757 	WARN_ON(!((1 << newsk->sk_state) &
758 		  (TCPF_ESTABLISHED | TCPF_SYN_RECV |
759 		   TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2 |
760 		   TCPF_CLOSING | TCPF_CLOSE_WAIT |
761 		   TCPF_CLOSE)));
762 
763 	if (test_bit(SOCK_SUPPORT_ZC, &sock->flags))
764 		set_bit(SOCK_SUPPORT_ZC, &newsock->flags);
765 	sock_graft(newsk, newsock);
766 
767 	newsock->state = SS_CONNECTED;
768 }
769 
770 /*
771  *	Accept a pending connection. The TCP layer now gives BSD semantics.
772  */
773 
774 int inet_accept(struct socket *sock, struct socket *newsock,
775 		struct proto_accept_arg *arg)
776 {
777 	struct sock *sk1 = sock->sk, *sk2;
778 
779 	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
780 	arg->err = -EINVAL;
781 	sk2 = READ_ONCE(sk1->sk_prot)->accept(sk1, arg);
782 	if (!sk2)
783 		return arg->err;
784 
785 	lock_sock(sk2);
786 	__inet_accept(sock, newsock, sk2);
787 	release_sock(sk2);
788 	return 0;
789 }
790 EXPORT_SYMBOL(inet_accept);
791 
792 /*
793  *	This does both peername and sockname.
794  */
795 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
796 		 int peer)
797 {
798 	struct sock *sk		= sock->sk;
799 	struct inet_sock *inet	= inet_sk(sk);
800 	DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
801 	int sin_addr_len = sizeof(*sin);
802 
803 	sin->sin_family = AF_INET;
804 	lock_sock(sk);
805 	if (peer) {
806 		if (!inet->inet_dport ||
807 		    (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
808 		     peer == 1)) {
809 			release_sock(sk);
810 			return -ENOTCONN;
811 		}
812 		sin->sin_port = inet->inet_dport;
813 		sin->sin_addr.s_addr = inet->inet_daddr;
814 		BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin, &sin_addr_len,
815 				       CGROUP_INET4_GETPEERNAME);
816 	} else {
817 		__be32 addr = inet->inet_rcv_saddr;
818 		if (!addr)
819 			addr = inet->inet_saddr;
820 		sin->sin_port = inet->inet_sport;
821 		sin->sin_addr.s_addr = addr;
822 		BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin, &sin_addr_len,
823 				       CGROUP_INET4_GETSOCKNAME);
824 	}
825 	release_sock(sk);
826 	memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
827 	return sin_addr_len;
828 }
829 EXPORT_SYMBOL(inet_getname);
830 
831 int inet_send_prepare(struct sock *sk)
832 {
833 	sock_rps_record_flow(sk);
834 
835 	/* We may need to bind the socket. */
836 	if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind &&
837 	    inet_autobind(sk))
838 		return -EAGAIN;
839 
840 	return 0;
841 }
842 EXPORT_SYMBOL_GPL(inet_send_prepare);
843 
844 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
845 {
846 	struct sock *sk = sock->sk;
847 
848 	if (unlikely(inet_send_prepare(sk)))
849 		return -EAGAIN;
850 
851 	return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
852 			       sk, msg, size);
853 }
854 EXPORT_SYMBOL(inet_sendmsg);
855 
856 void inet_splice_eof(struct socket *sock)
857 {
858 	const struct proto *prot;
859 	struct sock *sk = sock->sk;
860 
861 	if (unlikely(inet_send_prepare(sk)))
862 		return;
863 
864 	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
865 	prot = READ_ONCE(sk->sk_prot);
866 	if (prot->splice_eof)
867 		prot->splice_eof(sock);
868 }
869 EXPORT_SYMBOL_GPL(inet_splice_eof);
870 
871 INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
872 					  size_t, int, int *));
873 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
874 		 int flags)
875 {
876 	struct sock *sk = sock->sk;
877 	int addr_len = 0;
878 	int err;
879 
880 	if (likely(!(flags & MSG_ERRQUEUE)))
881 		sock_rps_record_flow(sk);
882 
883 	err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
884 			      sk, msg, size, flags, &addr_len);
885 	if (err >= 0)
886 		msg->msg_namelen = addr_len;
887 	return err;
888 }
889 EXPORT_SYMBOL(inet_recvmsg);
890 
891 int inet_shutdown(struct socket *sock, int how)
892 {
893 	struct sock *sk = sock->sk;
894 	int err = 0;
895 
896 	/* This should really check to make sure
897 	 * the socket is a TCP socket. (WHY AC...)
898 	 */
899 	how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
900 		       1->2 bit 2 snds.
901 		       2->3 */
902 	if ((how & ~SHUTDOWN_MASK) || !how)	/* MAXINT->0 */
903 		return -EINVAL;
904 
905 	lock_sock(sk);
906 	if (sock->state == SS_CONNECTING) {
907 		if ((1 << sk->sk_state) &
908 		    (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
909 			sock->state = SS_DISCONNECTING;
910 		else
911 			sock->state = SS_CONNECTED;
912 	}
913 
914 	switch (sk->sk_state) {
915 	case TCP_CLOSE:
916 		err = -ENOTCONN;
917 		/* Hack to wake up other listeners, who can poll for
918 		   EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
919 		fallthrough;
920 	default:
921 		WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | how);
922 		if (sk->sk_prot->shutdown)
923 			sk->sk_prot->shutdown(sk, how);
924 		break;
925 
926 	/* Remaining two branches are temporary solution for missing
927 	 * close() in multithreaded environment. It is _not_ a good idea,
928 	 * but we have no choice until close() is repaired at VFS level.
929 	 */
930 	case TCP_LISTEN:
931 		if (!(how & RCV_SHUTDOWN))
932 			break;
933 		fallthrough;
934 	case TCP_SYN_SENT:
935 		err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
936 		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
937 		break;
938 	}
939 
940 	/* Wake up anyone sleeping in poll. */
941 	sk->sk_state_change(sk);
942 	release_sock(sk);
943 	return err;
944 }
945 EXPORT_SYMBOL(inet_shutdown);
946 
947 /*
948  *	ioctl() calls you can issue on an INET socket. Most of these are
949  *	device configuration and stuff and very rarely used. Some ioctls
950  *	pass on to the socket itself.
951  *
952  *	NOTE: I like the idea of a module for the config stuff. ie ifconfig
953  *	loads the devconfigure module does its configuring and unloads it.
954  *	There's a good 20K of config code hanging around the kernel.
955  */
956 
957 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
958 {
959 	struct sock *sk = sock->sk;
960 	int err = 0;
961 	struct net *net = sock_net(sk);
962 	void __user *p = (void __user *)arg;
963 	struct ifreq ifr;
964 	struct rtentry rt;
965 
966 	switch (cmd) {
967 	case SIOCADDRT:
968 	case SIOCDELRT:
969 		if (copy_from_user(&rt, p, sizeof(struct rtentry)))
970 			return -EFAULT;
971 		err = ip_rt_ioctl(net, cmd, &rt);
972 		break;
973 	case SIOCRTMSG:
974 		err = -EINVAL;
975 		break;
976 	case SIOCDARP:
977 	case SIOCGARP:
978 	case SIOCSARP:
979 		err = arp_ioctl(net, cmd, (void __user *)arg);
980 		break;
981 	case SIOCGIFADDR:
982 	case SIOCGIFBRDADDR:
983 	case SIOCGIFNETMASK:
984 	case SIOCGIFDSTADDR:
985 	case SIOCGIFPFLAGS:
986 		if (get_user_ifreq(&ifr, NULL, p))
987 			return -EFAULT;
988 		err = devinet_ioctl(net, cmd, &ifr);
989 		if (!err && put_user_ifreq(&ifr, p))
990 			err = -EFAULT;
991 		break;
992 
993 	case SIOCSIFADDR:
994 	case SIOCSIFBRDADDR:
995 	case SIOCSIFNETMASK:
996 	case SIOCSIFDSTADDR:
997 	case SIOCSIFPFLAGS:
998 	case SIOCSIFFLAGS:
999 		if (get_user_ifreq(&ifr, NULL, p))
1000 			return -EFAULT;
1001 		err = devinet_ioctl(net, cmd, &ifr);
1002 		break;
1003 	default:
1004 		if (sk->sk_prot->ioctl)
1005 			err = sk_ioctl(sk, cmd, (void __user *)arg);
1006 		else
1007 			err = -ENOIOCTLCMD;
1008 		break;
1009 	}
1010 	return err;
1011 }
1012 EXPORT_SYMBOL(inet_ioctl);
1013 
1014 #ifdef CONFIG_COMPAT
1015 static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd,
1016 		struct compat_rtentry __user *ur)
1017 {
1018 	compat_uptr_t rtdev;
1019 	struct rtentry rt;
1020 
1021 	if (copy_from_user(&rt.rt_dst, &ur->rt_dst,
1022 			3 * sizeof(struct sockaddr)) ||
1023 	    get_user(rt.rt_flags, &ur->rt_flags) ||
1024 	    get_user(rt.rt_metric, &ur->rt_metric) ||
1025 	    get_user(rt.rt_mtu, &ur->rt_mtu) ||
1026 	    get_user(rt.rt_window, &ur->rt_window) ||
1027 	    get_user(rt.rt_irtt, &ur->rt_irtt) ||
1028 	    get_user(rtdev, &ur->rt_dev))
1029 		return -EFAULT;
1030 
1031 	rt.rt_dev = compat_ptr(rtdev);
1032 	return ip_rt_ioctl(sock_net(sk), cmd, &rt);
1033 }
1034 
1035 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1036 {
1037 	void __user *argp = compat_ptr(arg);
1038 	struct sock *sk = sock->sk;
1039 
1040 	switch (cmd) {
1041 	case SIOCADDRT:
1042 	case SIOCDELRT:
1043 		return inet_compat_routing_ioctl(sk, cmd, argp);
1044 	default:
1045 		if (!sk->sk_prot->compat_ioctl)
1046 			return -ENOIOCTLCMD;
1047 		return sk->sk_prot->compat_ioctl(sk, cmd, arg);
1048 	}
1049 }
1050 #endif /* CONFIG_COMPAT */
1051 
1052 const struct proto_ops inet_stream_ops = {
1053 	.family		   = PF_INET,
1054 	.owner		   = THIS_MODULE,
1055 	.release	   = inet_release,
1056 	.bind		   = inet_bind,
1057 	.connect	   = inet_stream_connect,
1058 	.socketpair	   = sock_no_socketpair,
1059 	.accept		   = inet_accept,
1060 	.getname	   = inet_getname,
1061 	.poll		   = tcp_poll,
1062 	.ioctl		   = inet_ioctl,
1063 	.gettstamp	   = sock_gettstamp,
1064 	.listen		   = inet_listen,
1065 	.shutdown	   = inet_shutdown,
1066 	.setsockopt	   = sock_common_setsockopt,
1067 	.getsockopt	   = sock_common_getsockopt,
1068 	.sendmsg	   = inet_sendmsg,
1069 	.recvmsg	   = inet_recvmsg,
1070 #ifdef CONFIG_MMU
1071 	.mmap		   = tcp_mmap,
1072 #endif
1073 	.splice_eof	   = inet_splice_eof,
1074 	.splice_read	   = tcp_splice_read,
1075 	.set_peek_off      = sk_set_peek_off,
1076 	.read_sock	   = tcp_read_sock,
1077 	.read_skb	   = tcp_read_skb,
1078 	.sendmsg_locked    = tcp_sendmsg_locked,
1079 	.peek_len	   = tcp_peek_len,
1080 #ifdef CONFIG_COMPAT
1081 	.compat_ioctl	   = inet_compat_ioctl,
1082 #endif
1083 	.set_rcvlowat	   = tcp_set_rcvlowat,
1084 };
1085 EXPORT_SYMBOL(inet_stream_ops);
1086 
1087 const struct proto_ops inet_dgram_ops = {
1088 	.family		   = PF_INET,
1089 	.owner		   = THIS_MODULE,
1090 	.release	   = inet_release,
1091 	.bind		   = inet_bind,
1092 	.connect	   = inet_dgram_connect,
1093 	.socketpair	   = sock_no_socketpair,
1094 	.accept		   = sock_no_accept,
1095 	.getname	   = inet_getname,
1096 	.poll		   = udp_poll,
1097 	.ioctl		   = inet_ioctl,
1098 	.gettstamp	   = sock_gettstamp,
1099 	.listen		   = sock_no_listen,
1100 	.shutdown	   = inet_shutdown,
1101 	.setsockopt	   = sock_common_setsockopt,
1102 	.getsockopt	   = sock_common_getsockopt,
1103 	.sendmsg	   = inet_sendmsg,
1104 	.read_skb	   = udp_read_skb,
1105 	.recvmsg	   = inet_recvmsg,
1106 	.mmap		   = sock_no_mmap,
1107 	.splice_eof	   = inet_splice_eof,
1108 	.set_peek_off	   = udp_set_peek_off,
1109 #ifdef CONFIG_COMPAT
1110 	.compat_ioctl	   = inet_compat_ioctl,
1111 #endif
1112 };
1113 EXPORT_SYMBOL(inet_dgram_ops);
1114 
1115 /*
1116  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1117  * udp_poll
1118  */
1119 static const struct proto_ops inet_sockraw_ops = {
1120 	.family		   = PF_INET,
1121 	.owner		   = THIS_MODULE,
1122 	.release	   = inet_release,
1123 	.bind		   = inet_bind,
1124 	.connect	   = inet_dgram_connect,
1125 	.socketpair	   = sock_no_socketpair,
1126 	.accept		   = sock_no_accept,
1127 	.getname	   = inet_getname,
1128 	.poll		   = datagram_poll,
1129 	.ioctl		   = inet_ioctl,
1130 	.gettstamp	   = sock_gettstamp,
1131 	.listen		   = sock_no_listen,
1132 	.shutdown	   = inet_shutdown,
1133 	.setsockopt	   = sock_common_setsockopt,
1134 	.getsockopt	   = sock_common_getsockopt,
1135 	.sendmsg	   = inet_sendmsg,
1136 	.recvmsg	   = inet_recvmsg,
1137 	.mmap		   = sock_no_mmap,
1138 	.splice_eof	   = inet_splice_eof,
1139 #ifdef CONFIG_COMPAT
1140 	.compat_ioctl	   = inet_compat_ioctl,
1141 #endif
1142 };
1143 
1144 static const struct net_proto_family inet_family_ops = {
1145 	.family = PF_INET,
1146 	.create = inet_create,
1147 	.owner	= THIS_MODULE,
1148 };
1149 
1150 /* Upon startup we insert all the elements in inetsw_array[] into
1151  * the linked list inetsw.
1152  */
1153 static struct inet_protosw inetsw_array[] =
1154 {
1155 	{
1156 		.type =       SOCK_STREAM,
1157 		.protocol =   IPPROTO_TCP,
1158 		.prot =       &tcp_prot,
1159 		.ops =        &inet_stream_ops,
1160 		.flags =      INET_PROTOSW_PERMANENT |
1161 			      INET_PROTOSW_ICSK,
1162 	},
1163 
1164 	{
1165 		.type =       SOCK_DGRAM,
1166 		.protocol =   IPPROTO_UDP,
1167 		.prot =       &udp_prot,
1168 		.ops =        &inet_dgram_ops,
1169 		.flags =      INET_PROTOSW_PERMANENT,
1170        },
1171 
1172        {
1173 		.type =       SOCK_DGRAM,
1174 		.protocol =   IPPROTO_ICMP,
1175 		.prot =       &ping_prot,
1176 		.ops =        &inet_sockraw_ops,
1177 		.flags =      INET_PROTOSW_REUSE,
1178        },
1179 
1180        {
1181 	       .type =       SOCK_RAW,
1182 	       .protocol =   IPPROTO_IP,	/* wild card */
1183 	       .prot =       &raw_prot,
1184 	       .ops =        &inet_sockraw_ops,
1185 	       .flags =      INET_PROTOSW_REUSE,
1186        }
1187 };
1188 
1189 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1190 
1191 void inet_register_protosw(struct inet_protosw *p)
1192 {
1193 	struct list_head *lh;
1194 	struct inet_protosw *answer;
1195 	int protocol = p->protocol;
1196 	struct list_head *last_perm;
1197 
1198 	spin_lock_bh(&inetsw_lock);
1199 
1200 	if (p->type >= SOCK_MAX)
1201 		goto out_illegal;
1202 
1203 	/* If we are trying to override a permanent protocol, bail. */
1204 	last_perm = &inetsw[p->type];
1205 	list_for_each(lh, &inetsw[p->type]) {
1206 		answer = list_entry(lh, struct inet_protosw, list);
1207 		/* Check only the non-wild match. */
1208 		if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1209 			break;
1210 		if (protocol == answer->protocol)
1211 			goto out_permanent;
1212 		last_perm = lh;
1213 	}
1214 
1215 	/* Add the new entry after the last permanent entry if any, so that
1216 	 * the new entry does not override a permanent entry when matched with
1217 	 * a wild-card protocol. But it is allowed to override any existing
1218 	 * non-permanent entry.  This means that when we remove this entry, the
1219 	 * system automatically returns to the old behavior.
1220 	 */
1221 	list_add_rcu(&p->list, last_perm);
1222 out:
1223 	spin_unlock_bh(&inetsw_lock);
1224 
1225 	return;
1226 
1227 out_permanent:
1228 	pr_err("Attempt to override permanent protocol %d\n", protocol);
1229 	goto out;
1230 
1231 out_illegal:
1232 	pr_err("Ignoring attempt to register invalid socket type %d\n",
1233 	       p->type);
1234 	goto out;
1235 }
1236 EXPORT_SYMBOL(inet_register_protosw);
1237 
1238 void inet_unregister_protosw(struct inet_protosw *p)
1239 {
1240 	if (INET_PROTOSW_PERMANENT & p->flags) {
1241 		pr_err("Attempt to unregister permanent protocol %d\n",
1242 		       p->protocol);
1243 	} else {
1244 		spin_lock_bh(&inetsw_lock);
1245 		list_del_rcu(&p->list);
1246 		spin_unlock_bh(&inetsw_lock);
1247 
1248 		synchronize_net();
1249 	}
1250 }
1251 EXPORT_SYMBOL(inet_unregister_protosw);
1252 
1253 static int inet_sk_reselect_saddr(struct sock *sk)
1254 {
1255 	struct inet_sock *inet = inet_sk(sk);
1256 	__be32 old_saddr = inet->inet_saddr;
1257 	__be32 daddr = inet->inet_daddr;
1258 	struct flowi4 *fl4;
1259 	struct rtable *rt;
1260 	__be32 new_saddr;
1261 	struct ip_options_rcu *inet_opt;
1262 	int err;
1263 
1264 	inet_opt = rcu_dereference_protected(inet->inet_opt,
1265 					     lockdep_sock_is_held(sk));
1266 	if (inet_opt && inet_opt->opt.srr)
1267 		daddr = inet_opt->opt.faddr;
1268 
1269 	/* Query new route. */
1270 	fl4 = &inet->cork.fl.u.ip4;
1271 	rt = ip_route_connect(fl4, daddr, 0, sk->sk_bound_dev_if,
1272 			      sk->sk_protocol, inet->inet_sport,
1273 			      inet->inet_dport, sk);
1274 	if (IS_ERR(rt))
1275 		return PTR_ERR(rt);
1276 
1277 	new_saddr = fl4->saddr;
1278 
1279 	if (new_saddr == old_saddr) {
1280 		sk_setup_caps(sk, &rt->dst);
1281 		return 0;
1282 	}
1283 
1284 	err = inet_bhash2_update_saddr(sk, &new_saddr, AF_INET);
1285 	if (err) {
1286 		ip_rt_put(rt);
1287 		return err;
1288 	}
1289 
1290 	sk_setup_caps(sk, &rt->dst);
1291 
1292 	if (READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) > 1) {
1293 		pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1294 			__func__, &old_saddr, &new_saddr);
1295 	}
1296 
1297 	/*
1298 	 * XXX The only one ugly spot where we need to
1299 	 * XXX really change the sockets identity after
1300 	 * XXX it has entered the hashes. -DaveM
1301 	 *
1302 	 * Besides that, it does not check for connection
1303 	 * uniqueness. Wait for troubles.
1304 	 */
1305 	return __sk_prot_rehash(sk);
1306 }
1307 
1308 int inet_sk_rebuild_header(struct sock *sk)
1309 {
1310 	struct rtable *rt = dst_rtable(__sk_dst_check(sk, 0));
1311 	struct inet_sock *inet = inet_sk(sk);
1312 	struct flowi4 *fl4;
1313 	int err;
1314 
1315 	/* Route is OK, nothing to do. */
1316 	if (rt)
1317 		return 0;
1318 
1319 	/* Reroute. */
1320 	fl4 = &inet->cork.fl.u.ip4;
1321 	inet_sk_init_flowi4(inet, fl4);
1322 	rt = ip_route_output_flow(sock_net(sk), fl4, sk);
1323 	if (!IS_ERR(rt)) {
1324 		err = 0;
1325 		sk_setup_caps(sk, &rt->dst);
1326 	} else {
1327 		err = PTR_ERR(rt);
1328 
1329 		/* Routing failed... */
1330 		sk->sk_route_caps = 0;
1331 
1332 		if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) ||
1333 		    sk->sk_state != TCP_SYN_SENT ||
1334 		    (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1335 		    (err = inet_sk_reselect_saddr(sk)) != 0)
1336 			WRITE_ONCE(sk->sk_err_soft, -err);
1337 	}
1338 
1339 	return err;
1340 }
1341 EXPORT_SYMBOL(inet_sk_rebuild_header);
1342 
1343 void inet_sk_set_state(struct sock *sk, int state)
1344 {
1345 	trace_inet_sock_set_state(sk, sk->sk_state, state);
1346 	sk->sk_state = state;
1347 }
1348 EXPORT_SYMBOL(inet_sk_set_state);
1349 
1350 void inet_sk_state_store(struct sock *sk, int newstate)
1351 {
1352 	trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1353 	smp_store_release(&sk->sk_state, newstate);
1354 }
1355 
1356 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1357 				 netdev_features_t features)
1358 {
1359 	bool udpfrag = false, fixedid = false, gso_partial, encap;
1360 	struct sk_buff *segs = ERR_PTR(-EINVAL);
1361 	const struct net_offload *ops;
1362 	unsigned int offset = 0;
1363 	struct iphdr *iph;
1364 	int proto, tot_len;
1365 	int nhoff;
1366 	int ihl;
1367 	int id;
1368 
1369 	skb_reset_network_header(skb);
1370 	nhoff = skb_network_header(skb) - skb_mac_header(skb);
1371 	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1372 		goto out;
1373 
1374 	iph = ip_hdr(skb);
1375 	ihl = iph->ihl * 4;
1376 	if (ihl < sizeof(*iph))
1377 		goto out;
1378 
1379 	id = ntohs(iph->id);
1380 	proto = iph->protocol;
1381 
1382 	/* Warning: after this point, iph might be no longer valid */
1383 	if (unlikely(!pskb_may_pull(skb, ihl)))
1384 		goto out;
1385 	__skb_pull(skb, ihl);
1386 
1387 	encap = SKB_GSO_CB(skb)->encap_level > 0;
1388 	if (encap)
1389 		features &= skb->dev->hw_enc_features;
1390 	SKB_GSO_CB(skb)->encap_level += ihl;
1391 
1392 	skb_reset_transport_header(skb);
1393 
1394 	segs = ERR_PTR(-EPROTONOSUPPORT);
1395 
1396 	if (!skb->encapsulation || encap) {
1397 		udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1398 		fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1399 
1400 		/* fixed ID is invalid if DF bit is not set */
1401 		if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1402 			goto out;
1403 	}
1404 
1405 	ops = rcu_dereference(inet_offloads[proto]);
1406 	if (likely(ops && ops->callbacks.gso_segment)) {
1407 		segs = ops->callbacks.gso_segment(skb, features);
1408 		if (!segs)
1409 			skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
1410 	}
1411 
1412 	if (IS_ERR_OR_NULL(segs))
1413 		goto out;
1414 
1415 	gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1416 
1417 	skb = segs;
1418 	do {
1419 		iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1420 		if (udpfrag) {
1421 			iph->frag_off = htons(offset >> 3);
1422 			if (skb->next)
1423 				iph->frag_off |= htons(IP_MF);
1424 			offset += skb->len - nhoff - ihl;
1425 			tot_len = skb->len - nhoff;
1426 		} else if (skb_is_gso(skb)) {
1427 			if (!fixedid) {
1428 				iph->id = htons(id);
1429 				id += skb_shinfo(skb)->gso_segs;
1430 			}
1431 
1432 			if (gso_partial)
1433 				tot_len = skb_shinfo(skb)->gso_size +
1434 					  SKB_GSO_CB(skb)->data_offset +
1435 					  skb->head - (unsigned char *)iph;
1436 			else
1437 				tot_len = skb->len - nhoff;
1438 		} else {
1439 			if (!fixedid)
1440 				iph->id = htons(id++);
1441 			tot_len = skb->len - nhoff;
1442 		}
1443 		iph->tot_len = htons(tot_len);
1444 		ip_send_check(iph);
1445 		if (encap)
1446 			skb_reset_inner_headers(skb);
1447 		skb->network_header = (u8 *)iph - skb->head;
1448 		skb_reset_mac_len(skb);
1449 	} while ((skb = skb->next));
1450 
1451 out:
1452 	return segs;
1453 }
1454 
1455 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1456 					netdev_features_t features)
1457 {
1458 	if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1459 		return ERR_PTR(-EINVAL);
1460 
1461 	return inet_gso_segment(skb, features);
1462 }
1463 
1464 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1465 {
1466 	const struct net_offload *ops;
1467 	struct sk_buff *pp = NULL;
1468 	const struct iphdr *iph;
1469 	struct sk_buff *p;
1470 	unsigned int hlen;
1471 	unsigned int off;
1472 	int flush = 1;
1473 	int proto;
1474 
1475 	off = skb_gro_offset(skb);
1476 	hlen = off + sizeof(*iph);
1477 	iph = skb_gro_header(skb, hlen, off);
1478 	if (unlikely(!iph))
1479 		goto out;
1480 
1481 	proto = iph->protocol;
1482 
1483 	ops = rcu_dereference(inet_offloads[proto]);
1484 	if (!ops || !ops->callbacks.gro_receive)
1485 		goto out;
1486 
1487 	if (*(u8 *)iph != 0x45)
1488 		goto out;
1489 
1490 	if (ip_is_fragment(iph))
1491 		goto out;
1492 
1493 	if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1494 		goto out;
1495 
1496 	NAPI_GRO_CB(skb)->proto = proto;
1497 	flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (ntohl(*(__be32 *)&iph->id) & ~IP_DF));
1498 
1499 	list_for_each_entry(p, head, list) {
1500 		struct iphdr *iph2;
1501 
1502 		if (!NAPI_GRO_CB(p)->same_flow)
1503 			continue;
1504 
1505 		iph2 = (struct iphdr *)(p->data + off);
1506 		/* The above works because, with the exception of the top
1507 		 * (inner most) layer, we only aggregate pkts with the same
1508 		 * hdr length so all the hdrs we'll need to verify will start
1509 		 * at the same offset.
1510 		 */
1511 		if ((iph->protocol ^ iph2->protocol) |
1512 		    ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1513 		    ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1514 			NAPI_GRO_CB(p)->same_flow = 0;
1515 			continue;
1516 		}
1517 	}
1518 
1519 	NAPI_GRO_CB(skb)->flush |= flush;
1520 	NAPI_GRO_CB(skb)->network_offsets[NAPI_GRO_CB(skb)->encap_mark] = off;
1521 
1522 	/* Note : No need to call skb_gro_postpull_rcsum() here,
1523 	 * as we already checked checksum over ipv4 header was 0
1524 	 */
1525 	skb_gro_pull(skb, sizeof(*iph));
1526 	skb_set_transport_header(skb, skb_gro_offset(skb));
1527 
1528 	pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1529 				       ops->callbacks.gro_receive, head, skb);
1530 
1531 out:
1532 	skb_gro_flush_final(skb, pp, flush);
1533 
1534 	return pp;
1535 }
1536 
1537 static struct sk_buff *ipip_gro_receive(struct list_head *head,
1538 					struct sk_buff *skb)
1539 {
1540 	if (NAPI_GRO_CB(skb)->encap_mark) {
1541 		NAPI_GRO_CB(skb)->flush = 1;
1542 		return NULL;
1543 	}
1544 
1545 	NAPI_GRO_CB(skb)->encap_mark = 1;
1546 
1547 	return inet_gro_receive(head, skb);
1548 }
1549 
1550 #define SECONDS_PER_DAY	86400
1551 
1552 /* inet_current_timestamp - Return IP network timestamp
1553  *
1554  * Return milliseconds since midnight in network byte order.
1555  */
1556 __be32 inet_current_timestamp(void)
1557 {
1558 	u32 secs;
1559 	u32 msecs;
1560 	struct timespec64 ts;
1561 
1562 	ktime_get_real_ts64(&ts);
1563 
1564 	/* Get secs since midnight. */
1565 	(void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1566 	/* Convert to msecs. */
1567 	msecs = secs * MSEC_PER_SEC;
1568 	/* Convert nsec to msec. */
1569 	msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1570 
1571 	/* Convert to network byte order. */
1572 	return htonl(msecs);
1573 }
1574 EXPORT_SYMBOL(inet_current_timestamp);
1575 
1576 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1577 {
1578 	unsigned int family = READ_ONCE(sk->sk_family);
1579 
1580 	if (family == AF_INET)
1581 		return ip_recv_error(sk, msg, len, addr_len);
1582 #if IS_ENABLED(CONFIG_IPV6)
1583 	if (family == AF_INET6)
1584 		return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1585 #endif
1586 	return -EINVAL;
1587 }
1588 EXPORT_SYMBOL(inet_recv_error);
1589 
1590 int inet_gro_complete(struct sk_buff *skb, int nhoff)
1591 {
1592 	struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1593 	const struct net_offload *ops;
1594 	__be16 totlen = iph->tot_len;
1595 	int proto = iph->protocol;
1596 	int err = -ENOSYS;
1597 
1598 	if (skb->encapsulation) {
1599 		skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1600 		skb_set_inner_network_header(skb, nhoff);
1601 	}
1602 
1603 	iph_set_totlen(iph, skb->len - nhoff);
1604 	csum_replace2(&iph->check, totlen, iph->tot_len);
1605 
1606 	ops = rcu_dereference(inet_offloads[proto]);
1607 	if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1608 		goto out;
1609 
1610 	/* Only need to add sizeof(*iph) to get to the next hdr below
1611 	 * because any hdr with option will have been flushed in
1612 	 * inet_gro_receive().
1613 	 */
1614 	err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1615 			      tcp4_gro_complete, udp4_gro_complete,
1616 			      skb, nhoff + sizeof(*iph));
1617 
1618 out:
1619 	return err;
1620 }
1621 
1622 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1623 {
1624 	skb->encapsulation = 1;
1625 	skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1626 	return inet_gro_complete(skb, nhoff);
1627 }
1628 
1629 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1630 			 unsigned short type, unsigned char protocol,
1631 			 struct net *net)
1632 {
1633 	struct socket *sock;
1634 	int rc = sock_create_kern(net, family, type, protocol, &sock);
1635 
1636 	if (rc == 0) {
1637 		*sk = sock->sk;
1638 		(*sk)->sk_allocation = GFP_ATOMIC;
1639 		(*sk)->sk_use_task_frag = false;
1640 		/*
1641 		 * Unhash it so that IP input processing does not even see it,
1642 		 * we do not wish this socket to see incoming packets.
1643 		 */
1644 		(*sk)->sk_prot->unhash(*sk);
1645 	}
1646 	return rc;
1647 }
1648 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1649 
1650 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1651 {
1652 	unsigned long res = 0;
1653 	int i;
1654 
1655 	for_each_possible_cpu(i)
1656 		res += snmp_get_cpu_field(mib, i, offt);
1657 	return res;
1658 }
1659 EXPORT_SYMBOL_GPL(snmp_fold_field);
1660 
1661 #if BITS_PER_LONG==32
1662 
1663 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1664 			 size_t syncp_offset)
1665 {
1666 	void *bhptr;
1667 	struct u64_stats_sync *syncp;
1668 	u64 v;
1669 	unsigned int start;
1670 
1671 	bhptr = per_cpu_ptr(mib, cpu);
1672 	syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1673 	do {
1674 		start = u64_stats_fetch_begin(syncp);
1675 		v = *(((u64 *)bhptr) + offt);
1676 	} while (u64_stats_fetch_retry(syncp, start));
1677 
1678 	return v;
1679 }
1680 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1681 
1682 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1683 {
1684 	u64 res = 0;
1685 	int cpu;
1686 
1687 	for_each_possible_cpu(cpu) {
1688 		res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1689 	}
1690 	return res;
1691 }
1692 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1693 #endif
1694 
1695 #ifdef CONFIG_IP_MULTICAST
1696 static const struct net_protocol igmp_protocol = {
1697 	.handler =	igmp_rcv,
1698 };
1699 #endif
1700 
1701 static const struct net_protocol icmp_protocol = {
1702 	.handler =	icmp_rcv,
1703 	.err_handler =	icmp_err,
1704 	.no_policy =	1,
1705 };
1706 
1707 static __net_init int ipv4_mib_init_net(struct net *net)
1708 {
1709 	int i;
1710 
1711 	net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1712 	if (!net->mib.tcp_statistics)
1713 		goto err_tcp_mib;
1714 	net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1715 	if (!net->mib.ip_statistics)
1716 		goto err_ip_mib;
1717 
1718 	for_each_possible_cpu(i) {
1719 		struct ipstats_mib *af_inet_stats;
1720 		af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1721 		u64_stats_init(&af_inet_stats->syncp);
1722 	}
1723 
1724 	net->mib.net_statistics = alloc_percpu(struct linux_mib);
1725 	if (!net->mib.net_statistics)
1726 		goto err_net_mib;
1727 	net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1728 	if (!net->mib.udp_statistics)
1729 		goto err_udp_mib;
1730 	net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1731 	if (!net->mib.udplite_statistics)
1732 		goto err_udplite_mib;
1733 	net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1734 	if (!net->mib.icmp_statistics)
1735 		goto err_icmp_mib;
1736 	net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1737 					      GFP_KERNEL);
1738 	if (!net->mib.icmpmsg_statistics)
1739 		goto err_icmpmsg_mib;
1740 
1741 	tcp_mib_init(net);
1742 	return 0;
1743 
1744 err_icmpmsg_mib:
1745 	free_percpu(net->mib.icmp_statistics);
1746 err_icmp_mib:
1747 	free_percpu(net->mib.udplite_statistics);
1748 err_udplite_mib:
1749 	free_percpu(net->mib.udp_statistics);
1750 err_udp_mib:
1751 	free_percpu(net->mib.net_statistics);
1752 err_net_mib:
1753 	free_percpu(net->mib.ip_statistics);
1754 err_ip_mib:
1755 	free_percpu(net->mib.tcp_statistics);
1756 err_tcp_mib:
1757 	return -ENOMEM;
1758 }
1759 
1760 static __net_exit void ipv4_mib_exit_net(struct net *net)
1761 {
1762 	kfree(net->mib.icmpmsg_statistics);
1763 	free_percpu(net->mib.icmp_statistics);
1764 	free_percpu(net->mib.udplite_statistics);
1765 	free_percpu(net->mib.udp_statistics);
1766 	free_percpu(net->mib.net_statistics);
1767 	free_percpu(net->mib.ip_statistics);
1768 	free_percpu(net->mib.tcp_statistics);
1769 #ifdef CONFIG_MPTCP
1770 	/* allocated on demand, see mptcp_init_sock() */
1771 	free_percpu(net->mib.mptcp_statistics);
1772 #endif
1773 }
1774 
1775 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1776 	.init = ipv4_mib_init_net,
1777 	.exit = ipv4_mib_exit_net,
1778 };
1779 
1780 static int __init init_ipv4_mibs(void)
1781 {
1782 	return register_pernet_subsys(&ipv4_mib_ops);
1783 }
1784 
1785 static __net_init int inet_init_net(struct net *net)
1786 {
1787 	/*
1788 	 * Set defaults for local port range
1789 	 */
1790 	net->ipv4.ip_local_ports.range = 60999u << 16 | 32768u;
1791 
1792 	seqlock_init(&net->ipv4.ping_group_range.lock);
1793 	/*
1794 	 * Sane defaults - nobody may create ping sockets.
1795 	 * Boot scripts should set this to distro-specific group.
1796 	 */
1797 	net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1798 	net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1799 
1800 	/* Default values for sysctl-controlled parameters.
1801 	 * We set them here, in case sysctl is not compiled.
1802 	 */
1803 	net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1804 	net->ipv4.sysctl_ip_fwd_update_priority = 1;
1805 	net->ipv4.sysctl_ip_dynaddr = 0;
1806 	net->ipv4.sysctl_ip_early_demux = 1;
1807 	net->ipv4.sysctl_udp_early_demux = 1;
1808 	net->ipv4.sysctl_tcp_early_demux = 1;
1809 	net->ipv4.sysctl_nexthop_compat_mode = 1;
1810 #ifdef CONFIG_SYSCTL
1811 	net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1812 #endif
1813 
1814 	/* Some igmp sysctl, whose values are always used */
1815 	net->ipv4.sysctl_igmp_max_memberships = 20;
1816 	net->ipv4.sysctl_igmp_max_msf = 10;
1817 	/* IGMP reports for link-local multicast groups are enabled by default */
1818 	net->ipv4.sysctl_igmp_llm_reports = 1;
1819 	net->ipv4.sysctl_igmp_qrv = 2;
1820 
1821 	net->ipv4.sysctl_fib_notify_on_flag_change = 0;
1822 
1823 	return 0;
1824 }
1825 
1826 static __net_initdata struct pernet_operations af_inet_ops = {
1827 	.init = inet_init_net,
1828 };
1829 
1830 static int __init init_inet_pernet_ops(void)
1831 {
1832 	return register_pernet_subsys(&af_inet_ops);
1833 }
1834 
1835 static int ipv4_proc_init(void);
1836 
1837 /*
1838  *	IP protocol layer initialiser
1839  */
1840 
1841 
1842 static const struct net_offload ipip_offload = {
1843 	.callbacks = {
1844 		.gso_segment	= ipip_gso_segment,
1845 		.gro_receive	= ipip_gro_receive,
1846 		.gro_complete	= ipip_gro_complete,
1847 	},
1848 };
1849 
1850 static int __init ipip_offload_init(void)
1851 {
1852 	return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1853 }
1854 
1855 static int __init ipv4_offload_init(void)
1856 {
1857 	/*
1858 	 * Add offloads
1859 	 */
1860 	if (udpv4_offload_init() < 0)
1861 		pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1862 	if (tcpv4_offload_init() < 0)
1863 		pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1864 	if (ipip_offload_init() < 0)
1865 		pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1866 
1867 	net_hotdata.ip_packet_offload = (struct packet_offload) {
1868 		.type = cpu_to_be16(ETH_P_IP),
1869 		.callbacks = {
1870 			.gso_segment = inet_gso_segment,
1871 			.gro_receive = inet_gro_receive,
1872 			.gro_complete = inet_gro_complete,
1873 		},
1874 	};
1875 	dev_add_offload(&net_hotdata.ip_packet_offload);
1876 	return 0;
1877 }
1878 
1879 fs_initcall(ipv4_offload_init);
1880 
1881 static struct packet_type ip_packet_type __read_mostly = {
1882 	.type = cpu_to_be16(ETH_P_IP),
1883 	.func = ip_rcv,
1884 	.list_func = ip_list_rcv,
1885 };
1886 
1887 static int __init inet_init(void)
1888 {
1889 	struct inet_protosw *q;
1890 	struct list_head *r;
1891 	int rc;
1892 
1893 	sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1894 
1895 	raw_hashinfo_init(&raw_v4_hashinfo);
1896 
1897 	rc = proto_register(&tcp_prot, 1);
1898 	if (rc)
1899 		goto out;
1900 
1901 	rc = proto_register(&udp_prot, 1);
1902 	if (rc)
1903 		goto out_unregister_tcp_proto;
1904 
1905 	rc = proto_register(&raw_prot, 1);
1906 	if (rc)
1907 		goto out_unregister_udp_proto;
1908 
1909 	rc = proto_register(&ping_prot, 1);
1910 	if (rc)
1911 		goto out_unregister_raw_proto;
1912 
1913 	/*
1914 	 *	Tell SOCKET that we are alive...
1915 	 */
1916 
1917 	(void)sock_register(&inet_family_ops);
1918 
1919 #ifdef CONFIG_SYSCTL
1920 	ip_static_sysctl_init();
1921 #endif
1922 
1923 	/*
1924 	 *	Add all the base protocols.
1925 	 */
1926 
1927 	if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1928 		pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1929 
1930 	net_hotdata.udp_protocol = (struct net_protocol) {
1931 		.handler =	udp_rcv,
1932 		.err_handler =	udp_err,
1933 		.no_policy =	1,
1934 	};
1935 	if (inet_add_protocol(&net_hotdata.udp_protocol, IPPROTO_UDP) < 0)
1936 		pr_crit("%s: Cannot add UDP protocol\n", __func__);
1937 
1938 	net_hotdata.tcp_protocol = (struct net_protocol) {
1939 		.handler	=	tcp_v4_rcv,
1940 		.err_handler	=	tcp_v4_err,
1941 		.no_policy	=	1,
1942 		.icmp_strict_tag_validation = 1,
1943 	};
1944 	if (inet_add_protocol(&net_hotdata.tcp_protocol, IPPROTO_TCP) < 0)
1945 		pr_crit("%s: Cannot add TCP protocol\n", __func__);
1946 #ifdef CONFIG_IP_MULTICAST
1947 	if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1948 		pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1949 #endif
1950 
1951 	/* Register the socket-side information for inet_create. */
1952 	for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1953 		INIT_LIST_HEAD(r);
1954 
1955 	for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1956 		inet_register_protosw(q);
1957 
1958 	/*
1959 	 *	Set the ARP module up
1960 	 */
1961 
1962 	arp_init();
1963 
1964 	/*
1965 	 *	Set the IP module up
1966 	 */
1967 
1968 	ip_init();
1969 
1970 	/* Initialise per-cpu ipv4 mibs */
1971 	if (init_ipv4_mibs())
1972 		panic("%s: Cannot init ipv4 mibs\n", __func__);
1973 
1974 	/* Setup TCP slab cache for open requests. */
1975 	tcp_init();
1976 
1977 	/* Setup UDP memory threshold */
1978 	udp_init();
1979 
1980 	/* Add UDP-Lite (RFC 3828) */
1981 	udplite4_register();
1982 
1983 	raw_init();
1984 
1985 	ping_init();
1986 
1987 	/*
1988 	 *	Set the ICMP layer up
1989 	 */
1990 
1991 	if (icmp_init() < 0)
1992 		panic("Failed to create the ICMP control socket.\n");
1993 
1994 	/*
1995 	 *	Initialise the multicast router
1996 	 */
1997 #if defined(CONFIG_IP_MROUTE)
1998 	if (ip_mr_init())
1999 		pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2000 #endif
2001 
2002 	if (init_inet_pernet_ops())
2003 		pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2004 
2005 	ipv4_proc_init();
2006 
2007 	ipfrag_init();
2008 
2009 	dev_add_pack(&ip_packet_type);
2010 
2011 	ip_tunnel_core_init();
2012 
2013 	rc = 0;
2014 out:
2015 	return rc;
2016 out_unregister_raw_proto:
2017 	proto_unregister(&raw_prot);
2018 out_unregister_udp_proto:
2019 	proto_unregister(&udp_prot);
2020 out_unregister_tcp_proto:
2021 	proto_unregister(&tcp_prot);
2022 	goto out;
2023 }
2024 
2025 fs_initcall(inet_init);
2026 
2027 /* ------------------------------------------------------------------------ */
2028 
2029 #ifdef CONFIG_PROC_FS
2030 static int __init ipv4_proc_init(void)
2031 {
2032 	int rc = 0;
2033 
2034 	if (raw_proc_init())
2035 		goto out_raw;
2036 	if (tcp4_proc_init())
2037 		goto out_tcp;
2038 	if (udp4_proc_init())
2039 		goto out_udp;
2040 	if (ping_proc_init())
2041 		goto out_ping;
2042 	if (ip_misc_proc_init())
2043 		goto out_misc;
2044 out:
2045 	return rc;
2046 out_misc:
2047 	ping_proc_exit();
2048 out_ping:
2049 	udp4_proc_exit();
2050 out_udp:
2051 	tcp4_proc_exit();
2052 out_tcp:
2053 	raw_proc_exit();
2054 out_raw:
2055 	rc = -ENOMEM;
2056 	goto out;
2057 }
2058 
2059 #else /* CONFIG_PROC_FS */
2060 static int __init ipv4_proc_init(void)
2061 {
2062 	return 0;
2063 }
2064 #endif /* CONFIG_PROC_FS */
2065