xref: /linux/net/ipv4/af_inet.c (revision ef9226cd56b718c79184a3466d32984a51cb449c)
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_forward_alloc_get(sk));
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 			sk_common_release(sk);
381 			goto out;
382 		}
383 	}
384 
385 	if (sk->sk_prot->init) {
386 		err = sk->sk_prot->init(sk);
387 		if (err) {
388 			sk_common_release(sk);
389 			goto out;
390 		}
391 	}
392 
393 	if (!kern) {
394 		err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
395 		if (err) {
396 			sk_common_release(sk);
397 			goto out;
398 		}
399 	}
400 out:
401 	return err;
402 out_rcu_unlock:
403 	rcu_read_unlock();
404 	goto out;
405 }
406 
407 
408 /*
409  *	The peer socket should always be NULL (or else). When we call this
410  *	function we are destroying the object and from then on nobody
411  *	should refer to it.
412  */
413 int inet_release(struct socket *sock)
414 {
415 	struct sock *sk = sock->sk;
416 
417 	if (sk) {
418 		long timeout;
419 
420 		if (!sk->sk_kern_sock)
421 			BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk);
422 
423 		/* Applications forget to leave groups before exiting */
424 		ip_mc_drop_socket(sk);
425 
426 		/* If linger is set, we don't return until the close
427 		 * is complete.  Otherwise we return immediately. The
428 		 * actually closing is done the same either way.
429 		 *
430 		 * If the close is due to the process exiting, we never
431 		 * linger..
432 		 */
433 		timeout = 0;
434 		if (sock_flag(sk, SOCK_LINGER) &&
435 		    !(current->flags & PF_EXITING))
436 			timeout = sk->sk_lingertime;
437 		sk->sk_prot->close(sk, timeout);
438 		sock->sk = NULL;
439 	}
440 	return 0;
441 }
442 EXPORT_SYMBOL(inet_release);
443 
444 int inet_bind_sk(struct sock *sk, struct sockaddr *uaddr, int addr_len)
445 {
446 	u32 flags = BIND_WITH_LOCK;
447 	int err;
448 
449 	/* If the socket has its own bind function then use it. (RAW) */
450 	if (sk->sk_prot->bind) {
451 		return sk->sk_prot->bind(sk, uaddr, addr_len);
452 	}
453 	if (addr_len < sizeof(struct sockaddr_in))
454 		return -EINVAL;
455 
456 	/* BPF prog is run before any checks are done so that if the prog
457 	 * changes context in a wrong way it will be caught.
458 	 */
459 	err = BPF_CGROUP_RUN_PROG_INET_BIND_LOCK(sk, uaddr, &addr_len,
460 						 CGROUP_INET4_BIND, &flags);
461 	if (err)
462 		return err;
463 
464 	return __inet_bind(sk, uaddr, addr_len, flags);
465 }
466 
467 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
468 {
469 	return inet_bind_sk(sock->sk, uaddr, addr_len);
470 }
471 EXPORT_SYMBOL(inet_bind);
472 
473 int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
474 		u32 flags)
475 {
476 	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
477 	struct inet_sock *inet = inet_sk(sk);
478 	struct net *net = sock_net(sk);
479 	unsigned short snum;
480 	int chk_addr_ret;
481 	u32 tb_id = RT_TABLE_LOCAL;
482 	int err;
483 
484 	if (addr->sin_family != AF_INET) {
485 		/* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
486 		 * only if s_addr is INADDR_ANY.
487 		 */
488 		err = -EAFNOSUPPORT;
489 		if (addr->sin_family != AF_UNSPEC ||
490 		    addr->sin_addr.s_addr != htonl(INADDR_ANY))
491 			goto out;
492 	}
493 
494 	tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
495 	chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
496 
497 	/* Not specified by any standard per-se, however it breaks too
498 	 * many applications when removed.  It is unfortunate since
499 	 * allowing applications to make a non-local bind solves
500 	 * several problems with systems using dynamic addressing.
501 	 * (ie. your servers still start up even if your ISDN link
502 	 *  is temporarily down)
503 	 */
504 	err = -EADDRNOTAVAIL;
505 	if (!inet_addr_valid_or_nonlocal(net, inet, addr->sin_addr.s_addr,
506 	                                 chk_addr_ret))
507 		goto out;
508 
509 	snum = ntohs(addr->sin_port);
510 	err = -EACCES;
511 	if (!(flags & BIND_NO_CAP_NET_BIND_SERVICE) &&
512 	    snum && inet_port_requires_bind_service(net, snum) &&
513 	    !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
514 		goto out;
515 
516 	/*      We keep a pair of addresses. rcv_saddr is the one
517 	 *      used by hash lookups, and saddr is used for transmit.
518 	 *
519 	 *      In the BSD API these are the same except where it
520 	 *      would be illegal to use them (multicast/broadcast) in
521 	 *      which case the sending device address is used.
522 	 */
523 	if (flags & BIND_WITH_LOCK)
524 		lock_sock(sk);
525 
526 	/* Check these errors (active socket, double bind). */
527 	err = -EINVAL;
528 	if (sk->sk_state != TCP_CLOSE || inet->inet_num)
529 		goto out_release_sock;
530 
531 	inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
532 	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
533 		inet->inet_saddr = 0;  /* Use device */
534 
535 	/* Make sure we are allowed to bind here. */
536 	if (snum || !(inet_test_bit(BIND_ADDRESS_NO_PORT, sk) ||
537 		      (flags & BIND_FORCE_ADDRESS_NO_PORT))) {
538 		err = sk->sk_prot->get_port(sk, snum);
539 		if (err) {
540 			inet->inet_saddr = inet->inet_rcv_saddr = 0;
541 			goto out_release_sock;
542 		}
543 		if (!(flags & BIND_FROM_BPF)) {
544 			err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
545 			if (err) {
546 				inet->inet_saddr = inet->inet_rcv_saddr = 0;
547 				if (sk->sk_prot->put_port)
548 					sk->sk_prot->put_port(sk);
549 				goto out_release_sock;
550 			}
551 		}
552 	}
553 
554 	if (inet->inet_rcv_saddr)
555 		sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
556 	if (snum)
557 		sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
558 	inet->inet_sport = htons(inet->inet_num);
559 	inet->inet_daddr = 0;
560 	inet->inet_dport = 0;
561 	sk_dst_reset(sk);
562 	err = 0;
563 out_release_sock:
564 	if (flags & BIND_WITH_LOCK)
565 		release_sock(sk);
566 out:
567 	return err;
568 }
569 
570 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
571 		       int addr_len, int flags)
572 {
573 	struct sock *sk = sock->sk;
574 	const struct proto *prot;
575 	int err;
576 
577 	if (addr_len < sizeof(uaddr->sa_family))
578 		return -EINVAL;
579 
580 	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
581 	prot = READ_ONCE(sk->sk_prot);
582 
583 	if (uaddr->sa_family == AF_UNSPEC)
584 		return prot->disconnect(sk, flags);
585 
586 	if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
587 		err = prot->pre_connect(sk, uaddr, addr_len);
588 		if (err)
589 			return err;
590 	}
591 
592 	if (data_race(!inet_sk(sk)->inet_num) && inet_autobind(sk))
593 		return -EAGAIN;
594 	return prot->connect(sk, uaddr, addr_len);
595 }
596 EXPORT_SYMBOL(inet_dgram_connect);
597 
598 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
599 {
600 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
601 
602 	add_wait_queue(sk_sleep(sk), &wait);
603 	sk->sk_write_pending += writebias;
604 
605 	/* Basic assumption: if someone sets sk->sk_err, he _must_
606 	 * change state of the socket from TCP_SYN_*.
607 	 * Connect() does not allow to get error notifications
608 	 * without closing the socket.
609 	 */
610 	while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
611 		release_sock(sk);
612 		timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
613 		lock_sock(sk);
614 		if (signal_pending(current) || !timeo)
615 			break;
616 	}
617 	remove_wait_queue(sk_sleep(sk), &wait);
618 	sk->sk_write_pending -= writebias;
619 	return timeo;
620 }
621 
622 /*
623  *	Connect to a remote host. There is regrettably still a little
624  *	TCP 'magic' in here.
625  */
626 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
627 			  int addr_len, int flags, int is_sendmsg)
628 {
629 	struct sock *sk = sock->sk;
630 	int err;
631 	long timeo;
632 
633 	/*
634 	 * uaddr can be NULL and addr_len can be 0 if:
635 	 * sk is a TCP fastopen active socket and
636 	 * TCP_FASTOPEN_CONNECT sockopt is set and
637 	 * we already have a valid cookie for this socket.
638 	 * In this case, user can call write() after connect().
639 	 * write() will invoke tcp_sendmsg_fastopen() which calls
640 	 * __inet_stream_connect().
641 	 */
642 	if (uaddr) {
643 		if (addr_len < sizeof(uaddr->sa_family))
644 			return -EINVAL;
645 
646 		if (uaddr->sa_family == AF_UNSPEC) {
647 			sk->sk_disconnects++;
648 			err = sk->sk_prot->disconnect(sk, flags);
649 			sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
650 			goto out;
651 		}
652 	}
653 
654 	switch (sock->state) {
655 	default:
656 		err = -EINVAL;
657 		goto out;
658 	case SS_CONNECTED:
659 		err = -EISCONN;
660 		goto out;
661 	case SS_CONNECTING:
662 		if (inet_test_bit(DEFER_CONNECT, sk))
663 			err = is_sendmsg ? -EINPROGRESS : -EISCONN;
664 		else
665 			err = -EALREADY;
666 		/* Fall out of switch with err, set for this state */
667 		break;
668 	case SS_UNCONNECTED:
669 		err = -EISCONN;
670 		if (sk->sk_state != TCP_CLOSE)
671 			goto out;
672 
673 		if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
674 			err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
675 			if (err)
676 				goto out;
677 		}
678 
679 		err = sk->sk_prot->connect(sk, uaddr, addr_len);
680 		if (err < 0)
681 			goto out;
682 
683 		sock->state = SS_CONNECTING;
684 
685 		if (!err && inet_test_bit(DEFER_CONNECT, sk))
686 			goto out;
687 
688 		/* Just entered SS_CONNECTING state; the only
689 		 * difference is that return value in non-blocking
690 		 * case is EINPROGRESS, rather than EALREADY.
691 		 */
692 		err = -EINPROGRESS;
693 		break;
694 	}
695 
696 	timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
697 
698 	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
699 		int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
700 				tcp_sk(sk)->fastopen_req &&
701 				tcp_sk(sk)->fastopen_req->data ? 1 : 0;
702 		int dis = sk->sk_disconnects;
703 
704 		/* Error code is set above */
705 		if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
706 			goto out;
707 
708 		err = sock_intr_errno(timeo);
709 		if (signal_pending(current))
710 			goto out;
711 
712 		if (dis != sk->sk_disconnects) {
713 			err = -EPIPE;
714 			goto out;
715 		}
716 	}
717 
718 	/* Connection was closed by RST, timeout, ICMP error
719 	 * or another process disconnected us.
720 	 */
721 	if (sk->sk_state == TCP_CLOSE)
722 		goto sock_error;
723 
724 	/* sk->sk_err may be not zero now, if RECVERR was ordered by user
725 	 * and error was received after socket entered established state.
726 	 * Hence, it is handled normally after connect() return successfully.
727 	 */
728 
729 	sock->state = SS_CONNECTED;
730 	err = 0;
731 out:
732 	return err;
733 
734 sock_error:
735 	err = sock_error(sk) ? : -ECONNABORTED;
736 	sock->state = SS_UNCONNECTED;
737 	sk->sk_disconnects++;
738 	if (sk->sk_prot->disconnect(sk, flags))
739 		sock->state = SS_DISCONNECTING;
740 	goto out;
741 }
742 EXPORT_SYMBOL(__inet_stream_connect);
743 
744 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
745 			int addr_len, int flags)
746 {
747 	int err;
748 
749 	lock_sock(sock->sk);
750 	err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
751 	release_sock(sock->sk);
752 	return err;
753 }
754 EXPORT_SYMBOL(inet_stream_connect);
755 
756 void __inet_accept(struct socket *sock, struct socket *newsock, struct sock *newsk)
757 {
758 	sock_rps_record_flow(newsk);
759 	WARN_ON(!((1 << newsk->sk_state) &
760 		  (TCPF_ESTABLISHED | TCPF_SYN_RECV |
761 		  TCPF_CLOSE_WAIT | 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, int flags,
775 		bool kern)
776 {
777 	struct sock *sk1 = sock->sk, *sk2;
778 	int err = -EINVAL;
779 
780 	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
781 	sk2 = READ_ONCE(sk1->sk_prot)->accept(sk1, flags, &err, kern);
782 	if (!sk2)
783 		return 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 inet_sock *inet = inet_sk(sk);
1311 	struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1312 	__be32 daddr;
1313 	struct ip_options_rcu *inet_opt;
1314 	struct flowi4 *fl4;
1315 	int err;
1316 
1317 	/* Route is OK, nothing to do. */
1318 	if (rt)
1319 		return 0;
1320 
1321 	/* Reroute. */
1322 	rcu_read_lock();
1323 	inet_opt = rcu_dereference(inet->inet_opt);
1324 	daddr = inet->inet_daddr;
1325 	if (inet_opt && inet_opt->opt.srr)
1326 		daddr = inet_opt->opt.faddr;
1327 	rcu_read_unlock();
1328 	fl4 = &inet->cork.fl.u.ip4;
1329 	rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1330 				   inet->inet_dport, inet->inet_sport,
1331 				   sk->sk_protocol, ip_sock_rt_tos(sk),
1332 				   sk->sk_bound_dev_if);
1333 	if (!IS_ERR(rt)) {
1334 		err = 0;
1335 		sk_setup_caps(sk, &rt->dst);
1336 	} else {
1337 		err = PTR_ERR(rt);
1338 
1339 		/* Routing failed... */
1340 		sk->sk_route_caps = 0;
1341 		/*
1342 		 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1343 		 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1344 		 */
1345 		if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) ||
1346 		    sk->sk_state != TCP_SYN_SENT ||
1347 		    (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1348 		    (err = inet_sk_reselect_saddr(sk)) != 0)
1349 			WRITE_ONCE(sk->sk_err_soft, -err);
1350 	}
1351 
1352 	return err;
1353 }
1354 EXPORT_SYMBOL(inet_sk_rebuild_header);
1355 
1356 void inet_sk_set_state(struct sock *sk, int state)
1357 {
1358 	trace_inet_sock_set_state(sk, sk->sk_state, state);
1359 	sk->sk_state = state;
1360 }
1361 EXPORT_SYMBOL(inet_sk_set_state);
1362 
1363 void inet_sk_state_store(struct sock *sk, int newstate)
1364 {
1365 	trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1366 	smp_store_release(&sk->sk_state, newstate);
1367 }
1368 
1369 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1370 				 netdev_features_t features)
1371 {
1372 	bool udpfrag = false, fixedid = false, gso_partial, encap;
1373 	struct sk_buff *segs = ERR_PTR(-EINVAL);
1374 	const struct net_offload *ops;
1375 	unsigned int offset = 0;
1376 	struct iphdr *iph;
1377 	int proto, tot_len;
1378 	int nhoff;
1379 	int ihl;
1380 	int id;
1381 
1382 	skb_reset_network_header(skb);
1383 	nhoff = skb_network_header(skb) - skb_mac_header(skb);
1384 	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1385 		goto out;
1386 
1387 	iph = ip_hdr(skb);
1388 	ihl = iph->ihl * 4;
1389 	if (ihl < sizeof(*iph))
1390 		goto out;
1391 
1392 	id = ntohs(iph->id);
1393 	proto = iph->protocol;
1394 
1395 	/* Warning: after this point, iph might be no longer valid */
1396 	if (unlikely(!pskb_may_pull(skb, ihl)))
1397 		goto out;
1398 	__skb_pull(skb, ihl);
1399 
1400 	encap = SKB_GSO_CB(skb)->encap_level > 0;
1401 	if (encap)
1402 		features &= skb->dev->hw_enc_features;
1403 	SKB_GSO_CB(skb)->encap_level += ihl;
1404 
1405 	skb_reset_transport_header(skb);
1406 
1407 	segs = ERR_PTR(-EPROTONOSUPPORT);
1408 
1409 	if (!skb->encapsulation || encap) {
1410 		udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1411 		fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1412 
1413 		/* fixed ID is invalid if DF bit is not set */
1414 		if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1415 			goto out;
1416 	}
1417 
1418 	ops = rcu_dereference(inet_offloads[proto]);
1419 	if (likely(ops && ops->callbacks.gso_segment)) {
1420 		segs = ops->callbacks.gso_segment(skb, features);
1421 		if (!segs)
1422 			skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
1423 	}
1424 
1425 	if (IS_ERR_OR_NULL(segs))
1426 		goto out;
1427 
1428 	gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1429 
1430 	skb = segs;
1431 	do {
1432 		iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1433 		if (udpfrag) {
1434 			iph->frag_off = htons(offset >> 3);
1435 			if (skb->next)
1436 				iph->frag_off |= htons(IP_MF);
1437 			offset += skb->len - nhoff - ihl;
1438 			tot_len = skb->len - nhoff;
1439 		} else if (skb_is_gso(skb)) {
1440 			if (!fixedid) {
1441 				iph->id = htons(id);
1442 				id += skb_shinfo(skb)->gso_segs;
1443 			}
1444 
1445 			if (gso_partial)
1446 				tot_len = skb_shinfo(skb)->gso_size +
1447 					  SKB_GSO_CB(skb)->data_offset +
1448 					  skb->head - (unsigned char *)iph;
1449 			else
1450 				tot_len = skb->len - nhoff;
1451 		} else {
1452 			if (!fixedid)
1453 				iph->id = htons(id++);
1454 			tot_len = skb->len - nhoff;
1455 		}
1456 		iph->tot_len = htons(tot_len);
1457 		ip_send_check(iph);
1458 		if (encap)
1459 			skb_reset_inner_headers(skb);
1460 		skb->network_header = (u8 *)iph - skb->head;
1461 		skb_reset_mac_len(skb);
1462 	} while ((skb = skb->next));
1463 
1464 out:
1465 	return segs;
1466 }
1467 
1468 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1469 					netdev_features_t features)
1470 {
1471 	if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1472 		return ERR_PTR(-EINVAL);
1473 
1474 	return inet_gso_segment(skb, features);
1475 }
1476 
1477 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1478 {
1479 	const struct net_offload *ops;
1480 	struct sk_buff *pp = NULL;
1481 	const struct iphdr *iph;
1482 	struct sk_buff *p;
1483 	unsigned int hlen;
1484 	unsigned int off;
1485 	unsigned int id;
1486 	int flush = 1;
1487 	int proto;
1488 
1489 	off = skb_gro_offset(skb);
1490 	hlen = off + sizeof(*iph);
1491 	iph = skb_gro_header(skb, hlen, off);
1492 	if (unlikely(!iph))
1493 		goto out;
1494 
1495 	proto = iph->protocol;
1496 
1497 	ops = rcu_dereference(inet_offloads[proto]);
1498 	if (!ops || !ops->callbacks.gro_receive)
1499 		goto out;
1500 
1501 	if (*(u8 *)iph != 0x45)
1502 		goto out;
1503 
1504 	if (ip_is_fragment(iph))
1505 		goto out;
1506 
1507 	if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1508 		goto out;
1509 
1510 	NAPI_GRO_CB(skb)->proto = proto;
1511 	id = ntohl(*(__be32 *)&iph->id);
1512 	flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1513 	id >>= 16;
1514 
1515 	list_for_each_entry(p, head, list) {
1516 		struct iphdr *iph2;
1517 		u16 flush_id;
1518 
1519 		if (!NAPI_GRO_CB(p)->same_flow)
1520 			continue;
1521 
1522 		iph2 = (struct iphdr *)(p->data + off);
1523 		/* The above works because, with the exception of the top
1524 		 * (inner most) layer, we only aggregate pkts with the same
1525 		 * hdr length so all the hdrs we'll need to verify will start
1526 		 * at the same offset.
1527 		 */
1528 		if ((iph->protocol ^ iph2->protocol) |
1529 		    ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1530 		    ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1531 			NAPI_GRO_CB(p)->same_flow = 0;
1532 			continue;
1533 		}
1534 
1535 		/* All fields must match except length and checksum. */
1536 		NAPI_GRO_CB(p)->flush |=
1537 			(iph->ttl ^ iph2->ttl) |
1538 			(iph->tos ^ iph2->tos) |
1539 			((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1540 
1541 		NAPI_GRO_CB(p)->flush |= flush;
1542 
1543 		/* We need to store of the IP ID check to be included later
1544 		 * when we can verify that this packet does in fact belong
1545 		 * to a given flow.
1546 		 */
1547 		flush_id = (u16)(id - ntohs(iph2->id));
1548 
1549 		/* This bit of code makes it much easier for us to identify
1550 		 * the cases where we are doing atomic vs non-atomic IP ID
1551 		 * checks.  Specifically an atomic check can return IP ID
1552 		 * values 0 - 0xFFFF, while a non-atomic check can only
1553 		 * return 0 or 0xFFFF.
1554 		 */
1555 		if (!NAPI_GRO_CB(p)->is_atomic ||
1556 		    !(iph->frag_off & htons(IP_DF))) {
1557 			flush_id ^= NAPI_GRO_CB(p)->count;
1558 			flush_id = flush_id ? 0xFFFF : 0;
1559 		}
1560 
1561 		/* If the previous IP ID value was based on an atomic
1562 		 * datagram we can overwrite the value and ignore it.
1563 		 */
1564 		if (NAPI_GRO_CB(skb)->is_atomic)
1565 			NAPI_GRO_CB(p)->flush_id = flush_id;
1566 		else
1567 			NAPI_GRO_CB(p)->flush_id |= flush_id;
1568 	}
1569 
1570 	NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1571 	NAPI_GRO_CB(skb)->flush |= flush;
1572 	skb_set_network_header(skb, off);
1573 	/* The above will be needed by the transport layer if there is one
1574 	 * immediately following this IP hdr.
1575 	 */
1576 
1577 	/* Note : No need to call skb_gro_postpull_rcsum() here,
1578 	 * as we already checked checksum over ipv4 header was 0
1579 	 */
1580 	skb_gro_pull(skb, sizeof(*iph));
1581 	skb_set_transport_header(skb, skb_gro_offset(skb));
1582 
1583 	pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1584 				       ops->callbacks.gro_receive, head, skb);
1585 
1586 out:
1587 	skb_gro_flush_final(skb, pp, flush);
1588 
1589 	return pp;
1590 }
1591 
1592 static struct sk_buff *ipip_gro_receive(struct list_head *head,
1593 					struct sk_buff *skb)
1594 {
1595 	if (NAPI_GRO_CB(skb)->encap_mark) {
1596 		NAPI_GRO_CB(skb)->flush = 1;
1597 		return NULL;
1598 	}
1599 
1600 	NAPI_GRO_CB(skb)->encap_mark = 1;
1601 
1602 	return inet_gro_receive(head, skb);
1603 }
1604 
1605 #define SECONDS_PER_DAY	86400
1606 
1607 /* inet_current_timestamp - Return IP network timestamp
1608  *
1609  * Return milliseconds since midnight in network byte order.
1610  */
1611 __be32 inet_current_timestamp(void)
1612 {
1613 	u32 secs;
1614 	u32 msecs;
1615 	struct timespec64 ts;
1616 
1617 	ktime_get_real_ts64(&ts);
1618 
1619 	/* Get secs since midnight. */
1620 	(void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1621 	/* Convert to msecs. */
1622 	msecs = secs * MSEC_PER_SEC;
1623 	/* Convert nsec to msec. */
1624 	msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1625 
1626 	/* Convert to network byte order. */
1627 	return htonl(msecs);
1628 }
1629 EXPORT_SYMBOL(inet_current_timestamp);
1630 
1631 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1632 {
1633 	unsigned int family = READ_ONCE(sk->sk_family);
1634 
1635 	if (family == AF_INET)
1636 		return ip_recv_error(sk, msg, len, addr_len);
1637 #if IS_ENABLED(CONFIG_IPV6)
1638 	if (family == AF_INET6)
1639 		return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1640 #endif
1641 	return -EINVAL;
1642 }
1643 EXPORT_SYMBOL(inet_recv_error);
1644 
1645 int inet_gro_complete(struct sk_buff *skb, int nhoff)
1646 {
1647 	struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1648 	const struct net_offload *ops;
1649 	__be16 totlen = iph->tot_len;
1650 	int proto = iph->protocol;
1651 	int err = -ENOSYS;
1652 
1653 	if (skb->encapsulation) {
1654 		skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1655 		skb_set_inner_network_header(skb, nhoff);
1656 	}
1657 
1658 	iph_set_totlen(iph, skb->len - nhoff);
1659 	csum_replace2(&iph->check, totlen, iph->tot_len);
1660 
1661 	ops = rcu_dereference(inet_offloads[proto]);
1662 	if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1663 		goto out;
1664 
1665 	/* Only need to add sizeof(*iph) to get to the next hdr below
1666 	 * because any hdr with option will have been flushed in
1667 	 * inet_gro_receive().
1668 	 */
1669 	err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1670 			      tcp4_gro_complete, udp4_gro_complete,
1671 			      skb, nhoff + sizeof(*iph));
1672 
1673 out:
1674 	return err;
1675 }
1676 
1677 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1678 {
1679 	skb->encapsulation = 1;
1680 	skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1681 	return inet_gro_complete(skb, nhoff);
1682 }
1683 
1684 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1685 			 unsigned short type, unsigned char protocol,
1686 			 struct net *net)
1687 {
1688 	struct socket *sock;
1689 	int rc = sock_create_kern(net, family, type, protocol, &sock);
1690 
1691 	if (rc == 0) {
1692 		*sk = sock->sk;
1693 		(*sk)->sk_allocation = GFP_ATOMIC;
1694 		(*sk)->sk_use_task_frag = false;
1695 		/*
1696 		 * Unhash it so that IP input processing does not even see it,
1697 		 * we do not wish this socket to see incoming packets.
1698 		 */
1699 		(*sk)->sk_prot->unhash(*sk);
1700 	}
1701 	return rc;
1702 }
1703 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1704 
1705 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1706 {
1707 	unsigned long res = 0;
1708 	int i;
1709 
1710 	for_each_possible_cpu(i)
1711 		res += snmp_get_cpu_field(mib, i, offt);
1712 	return res;
1713 }
1714 EXPORT_SYMBOL_GPL(snmp_fold_field);
1715 
1716 #if BITS_PER_LONG==32
1717 
1718 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1719 			 size_t syncp_offset)
1720 {
1721 	void *bhptr;
1722 	struct u64_stats_sync *syncp;
1723 	u64 v;
1724 	unsigned int start;
1725 
1726 	bhptr = per_cpu_ptr(mib, cpu);
1727 	syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1728 	do {
1729 		start = u64_stats_fetch_begin(syncp);
1730 		v = *(((u64 *)bhptr) + offt);
1731 	} while (u64_stats_fetch_retry(syncp, start));
1732 
1733 	return v;
1734 }
1735 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1736 
1737 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1738 {
1739 	u64 res = 0;
1740 	int cpu;
1741 
1742 	for_each_possible_cpu(cpu) {
1743 		res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1744 	}
1745 	return res;
1746 }
1747 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1748 #endif
1749 
1750 #ifdef CONFIG_IP_MULTICAST
1751 static const struct net_protocol igmp_protocol = {
1752 	.handler =	igmp_rcv,
1753 };
1754 #endif
1755 
1756 static const struct net_protocol icmp_protocol = {
1757 	.handler =	icmp_rcv,
1758 	.err_handler =	icmp_err,
1759 	.no_policy =	1,
1760 };
1761 
1762 static __net_init int ipv4_mib_init_net(struct net *net)
1763 {
1764 	int i;
1765 
1766 	net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1767 	if (!net->mib.tcp_statistics)
1768 		goto err_tcp_mib;
1769 	net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1770 	if (!net->mib.ip_statistics)
1771 		goto err_ip_mib;
1772 
1773 	for_each_possible_cpu(i) {
1774 		struct ipstats_mib *af_inet_stats;
1775 		af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1776 		u64_stats_init(&af_inet_stats->syncp);
1777 	}
1778 
1779 	net->mib.net_statistics = alloc_percpu(struct linux_mib);
1780 	if (!net->mib.net_statistics)
1781 		goto err_net_mib;
1782 	net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1783 	if (!net->mib.udp_statistics)
1784 		goto err_udp_mib;
1785 	net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1786 	if (!net->mib.udplite_statistics)
1787 		goto err_udplite_mib;
1788 	net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1789 	if (!net->mib.icmp_statistics)
1790 		goto err_icmp_mib;
1791 	net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1792 					      GFP_KERNEL);
1793 	if (!net->mib.icmpmsg_statistics)
1794 		goto err_icmpmsg_mib;
1795 
1796 	tcp_mib_init(net);
1797 	return 0;
1798 
1799 err_icmpmsg_mib:
1800 	free_percpu(net->mib.icmp_statistics);
1801 err_icmp_mib:
1802 	free_percpu(net->mib.udplite_statistics);
1803 err_udplite_mib:
1804 	free_percpu(net->mib.udp_statistics);
1805 err_udp_mib:
1806 	free_percpu(net->mib.net_statistics);
1807 err_net_mib:
1808 	free_percpu(net->mib.ip_statistics);
1809 err_ip_mib:
1810 	free_percpu(net->mib.tcp_statistics);
1811 err_tcp_mib:
1812 	return -ENOMEM;
1813 }
1814 
1815 static __net_exit void ipv4_mib_exit_net(struct net *net)
1816 {
1817 	kfree(net->mib.icmpmsg_statistics);
1818 	free_percpu(net->mib.icmp_statistics);
1819 	free_percpu(net->mib.udplite_statistics);
1820 	free_percpu(net->mib.udp_statistics);
1821 	free_percpu(net->mib.net_statistics);
1822 	free_percpu(net->mib.ip_statistics);
1823 	free_percpu(net->mib.tcp_statistics);
1824 #ifdef CONFIG_MPTCP
1825 	/* allocated on demand, see mptcp_init_sock() */
1826 	free_percpu(net->mib.mptcp_statistics);
1827 #endif
1828 }
1829 
1830 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1831 	.init = ipv4_mib_init_net,
1832 	.exit = ipv4_mib_exit_net,
1833 };
1834 
1835 static int __init init_ipv4_mibs(void)
1836 {
1837 	return register_pernet_subsys(&ipv4_mib_ops);
1838 }
1839 
1840 static __net_init int inet_init_net(struct net *net)
1841 {
1842 	/*
1843 	 * Set defaults for local port range
1844 	 */
1845 	net->ipv4.ip_local_ports.range = 60999u << 16 | 32768u;
1846 
1847 	seqlock_init(&net->ipv4.ping_group_range.lock);
1848 	/*
1849 	 * Sane defaults - nobody may create ping sockets.
1850 	 * Boot scripts should set this to distro-specific group.
1851 	 */
1852 	net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1853 	net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1854 
1855 	/* Default values for sysctl-controlled parameters.
1856 	 * We set them here, in case sysctl is not compiled.
1857 	 */
1858 	net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1859 	net->ipv4.sysctl_ip_fwd_update_priority = 1;
1860 	net->ipv4.sysctl_ip_dynaddr = 0;
1861 	net->ipv4.sysctl_ip_early_demux = 1;
1862 	net->ipv4.sysctl_udp_early_demux = 1;
1863 	net->ipv4.sysctl_tcp_early_demux = 1;
1864 	net->ipv4.sysctl_nexthop_compat_mode = 1;
1865 #ifdef CONFIG_SYSCTL
1866 	net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1867 #endif
1868 
1869 	/* Some igmp sysctl, whose values are always used */
1870 	net->ipv4.sysctl_igmp_max_memberships = 20;
1871 	net->ipv4.sysctl_igmp_max_msf = 10;
1872 	/* IGMP reports for link-local multicast groups are enabled by default */
1873 	net->ipv4.sysctl_igmp_llm_reports = 1;
1874 	net->ipv4.sysctl_igmp_qrv = 2;
1875 
1876 	net->ipv4.sysctl_fib_notify_on_flag_change = 0;
1877 
1878 	return 0;
1879 }
1880 
1881 static __net_initdata struct pernet_operations af_inet_ops = {
1882 	.init = inet_init_net,
1883 };
1884 
1885 static int __init init_inet_pernet_ops(void)
1886 {
1887 	return register_pernet_subsys(&af_inet_ops);
1888 }
1889 
1890 static int ipv4_proc_init(void);
1891 
1892 /*
1893  *	IP protocol layer initialiser
1894  */
1895 
1896 
1897 static const struct net_offload ipip_offload = {
1898 	.callbacks = {
1899 		.gso_segment	= ipip_gso_segment,
1900 		.gro_receive	= ipip_gro_receive,
1901 		.gro_complete	= ipip_gro_complete,
1902 	},
1903 };
1904 
1905 static int __init ipip_offload_init(void)
1906 {
1907 	return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1908 }
1909 
1910 static int __init ipv4_offload_init(void)
1911 {
1912 	/*
1913 	 * Add offloads
1914 	 */
1915 	if (udpv4_offload_init() < 0)
1916 		pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1917 	if (tcpv4_offload_init() < 0)
1918 		pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1919 	if (ipip_offload_init() < 0)
1920 		pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1921 
1922 	net_hotdata.ip_packet_offload = (struct packet_offload) {
1923 		.type = cpu_to_be16(ETH_P_IP),
1924 		.callbacks = {
1925 			.gso_segment = inet_gso_segment,
1926 			.gro_receive = inet_gro_receive,
1927 			.gro_complete = inet_gro_complete,
1928 		},
1929 	};
1930 	dev_add_offload(&net_hotdata.ip_packet_offload);
1931 	return 0;
1932 }
1933 
1934 fs_initcall(ipv4_offload_init);
1935 
1936 static struct packet_type ip_packet_type __read_mostly = {
1937 	.type = cpu_to_be16(ETH_P_IP),
1938 	.func = ip_rcv,
1939 	.list_func = ip_list_rcv,
1940 };
1941 
1942 static int __init inet_init(void)
1943 {
1944 	struct inet_protosw *q;
1945 	struct list_head *r;
1946 	int rc;
1947 
1948 	sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1949 
1950 	raw_hashinfo_init(&raw_v4_hashinfo);
1951 
1952 	rc = proto_register(&tcp_prot, 1);
1953 	if (rc)
1954 		goto out;
1955 
1956 	rc = proto_register(&udp_prot, 1);
1957 	if (rc)
1958 		goto out_unregister_tcp_proto;
1959 
1960 	rc = proto_register(&raw_prot, 1);
1961 	if (rc)
1962 		goto out_unregister_udp_proto;
1963 
1964 	rc = proto_register(&ping_prot, 1);
1965 	if (rc)
1966 		goto out_unregister_raw_proto;
1967 
1968 	/*
1969 	 *	Tell SOCKET that we are alive...
1970 	 */
1971 
1972 	(void)sock_register(&inet_family_ops);
1973 
1974 #ifdef CONFIG_SYSCTL
1975 	ip_static_sysctl_init();
1976 #endif
1977 
1978 	/*
1979 	 *	Add all the base protocols.
1980 	 */
1981 
1982 	if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1983 		pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1984 
1985 	net_hotdata.udp_protocol = (struct net_protocol) {
1986 		.handler =	udp_rcv,
1987 		.err_handler =	udp_err,
1988 		.no_policy =	1,
1989 	};
1990 	if (inet_add_protocol(&net_hotdata.udp_protocol, IPPROTO_UDP) < 0)
1991 		pr_crit("%s: Cannot add UDP protocol\n", __func__);
1992 
1993 	net_hotdata.tcp_protocol = (struct net_protocol) {
1994 		.handler	=	tcp_v4_rcv,
1995 		.err_handler	=	tcp_v4_err,
1996 		.no_policy	=	1,
1997 		.icmp_strict_tag_validation = 1,
1998 	};
1999 	if (inet_add_protocol(&net_hotdata.tcp_protocol, IPPROTO_TCP) < 0)
2000 		pr_crit("%s: Cannot add TCP protocol\n", __func__);
2001 #ifdef CONFIG_IP_MULTICAST
2002 	if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
2003 		pr_crit("%s: Cannot add IGMP protocol\n", __func__);
2004 #endif
2005 
2006 	/* Register the socket-side information for inet_create. */
2007 	for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
2008 		INIT_LIST_HEAD(r);
2009 
2010 	for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
2011 		inet_register_protosw(q);
2012 
2013 	/*
2014 	 *	Set the ARP module up
2015 	 */
2016 
2017 	arp_init();
2018 
2019 	/*
2020 	 *	Set the IP module up
2021 	 */
2022 
2023 	ip_init();
2024 
2025 	/* Initialise per-cpu ipv4 mibs */
2026 	if (init_ipv4_mibs())
2027 		panic("%s: Cannot init ipv4 mibs\n", __func__);
2028 
2029 	/* Setup TCP slab cache for open requests. */
2030 	tcp_init();
2031 
2032 	/* Setup UDP memory threshold */
2033 	udp_init();
2034 
2035 	/* Add UDP-Lite (RFC 3828) */
2036 	udplite4_register();
2037 
2038 	raw_init();
2039 
2040 	ping_init();
2041 
2042 	/*
2043 	 *	Set the ICMP layer up
2044 	 */
2045 
2046 	if (icmp_init() < 0)
2047 		panic("Failed to create the ICMP control socket.\n");
2048 
2049 	/*
2050 	 *	Initialise the multicast router
2051 	 */
2052 #if defined(CONFIG_IP_MROUTE)
2053 	if (ip_mr_init())
2054 		pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2055 #endif
2056 
2057 	if (init_inet_pernet_ops())
2058 		pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2059 
2060 	ipv4_proc_init();
2061 
2062 	ipfrag_init();
2063 
2064 	dev_add_pack(&ip_packet_type);
2065 
2066 	ip_tunnel_core_init();
2067 
2068 	rc = 0;
2069 out:
2070 	return rc;
2071 out_unregister_raw_proto:
2072 	proto_unregister(&raw_prot);
2073 out_unregister_udp_proto:
2074 	proto_unregister(&udp_prot);
2075 out_unregister_tcp_proto:
2076 	proto_unregister(&tcp_prot);
2077 	goto out;
2078 }
2079 
2080 fs_initcall(inet_init);
2081 
2082 /* ------------------------------------------------------------------------ */
2083 
2084 #ifdef CONFIG_PROC_FS
2085 static int __init ipv4_proc_init(void)
2086 {
2087 	int rc = 0;
2088 
2089 	if (raw_proc_init())
2090 		goto out_raw;
2091 	if (tcp4_proc_init())
2092 		goto out_tcp;
2093 	if (udp4_proc_init())
2094 		goto out_udp;
2095 	if (ping_proc_init())
2096 		goto out_ping;
2097 	if (ip_misc_proc_init())
2098 		goto out_misc;
2099 out:
2100 	return rc;
2101 out_misc:
2102 	ping_proc_exit();
2103 out_ping:
2104 	udp4_proc_exit();
2105 out_udp:
2106 	tcp4_proc_exit();
2107 out_tcp:
2108 	raw_proc_exit();
2109 out_raw:
2110 	rc = -ENOMEM;
2111 	goto out;
2112 }
2113 
2114 #else /* CONFIG_PROC_FS */
2115 static int __init ipv4_proc_init(void)
2116 {
2117 	return 0;
2118 }
2119 #endif /* CONFIG_PROC_FS */
2120