xref: /linux/net/ipv4/af_inet.c (revision 77a6401a8722be20ea8db98ac900c93ccc7068ff)
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/psp.h>
106 #include <net/udp.h>
107 #include <net/udplite.h>
108 #include <net/ping.h>
109 #include <linux/skbuff.h>
110 #include <net/sock.h>
111 #include <net/raw.h>
112 #include <net/icmp.h>
113 #include <net/inet_common.h>
114 #include <net/ip_tunnels.h>
115 #include <net/xfrm.h>
116 #include <net/net_namespace.h>
117 #include <net/secure_seq.h>
118 #ifdef CONFIG_IP_MROUTE
119 #include <linux/mroute.h>
120 #endif
121 #include <net/l3mdev.h>
122 #include <net/compat.h>
123 #include <net/rps.h>
124 
125 #include <trace/events/sock.h>
126 
127 /* The inetsw table contains everything that inet_create needs to
128  * build a new socket.
129  */
130 static struct list_head inetsw[SOCK_MAX];
131 static DEFINE_SPINLOCK(inetsw_lock);
132 
133 /* New destruction routine */
134 
135 void inet_sock_destruct(struct sock *sk)
136 {
137 	struct inet_sock *inet = inet_sk(sk);
138 
139 	__skb_queue_purge(&sk->sk_receive_queue);
140 	__skb_queue_purge(&sk->sk_error_queue);
141 
142 	sk_mem_reclaim_final(sk);
143 
144 	if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
145 		pr_err("Attempt to release TCP socket in state %d %p\n",
146 		       sk->sk_state, sk);
147 		return;
148 	}
149 	if (!sock_flag(sk, SOCK_DEAD)) {
150 		pr_err("Attempt to release alive inet socket %p\n", sk);
151 		return;
152 	}
153 
154 	WARN_ON_ONCE(atomic_read(&sk->sk_rmem_alloc));
155 	WARN_ON_ONCE(refcount_read(&sk->sk_wmem_alloc));
156 	WARN_ON_ONCE(sk->sk_wmem_queued);
157 	WARN_ON_ONCE(sk->sk_forward_alloc);
158 
159 	kfree(rcu_dereference_protected(inet->inet_opt, 1));
160 	dst_release(rcu_dereference_protected(sk->sk_dst_cache, 1));
161 	dst_release(rcu_dereference_protected(sk->sk_rx_dst, 1));
162 	psp_sk_assoc_free(sk);
163 }
164 EXPORT_SYMBOL(inet_sock_destruct);
165 
166 /*
167  *	The routines beyond this point handle the behaviour of an AF_INET
168  *	socket object. Mostly it punts to the subprotocols of IP to do
169  *	the work.
170  */
171 
172 /*
173  *	Automatically bind an unbound socket.
174  */
175 
176 static int inet_autobind(struct sock *sk)
177 {
178 	struct inet_sock *inet;
179 	/* We may need to bind the socket. */
180 	lock_sock(sk);
181 	inet = inet_sk(sk);
182 	if (!inet->inet_num) {
183 		if (sk->sk_prot->get_port(sk, 0)) {
184 			release_sock(sk);
185 			return -EAGAIN;
186 		}
187 		inet->inet_sport = htons(inet->inet_num);
188 	}
189 	release_sock(sk);
190 	return 0;
191 }
192 
193 int __inet_listen_sk(struct sock *sk, int backlog)
194 {
195 	unsigned char old_state = sk->sk_state;
196 	int err, tcp_fastopen;
197 
198 	if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
199 		return -EINVAL;
200 
201 	WRITE_ONCE(sk->sk_max_ack_backlog, backlog);
202 	/* Really, if the socket is already in listen state
203 	 * we can only allow the backlog to be adjusted.
204 	 */
205 	if (old_state != TCP_LISTEN) {
206 		/* Enable TFO w/o requiring TCP_FASTOPEN socket option.
207 		 * Note that only TCP sockets (SOCK_STREAM) will reach here.
208 		 * Also fastopen backlog may already been set via the option
209 		 * because the socket was in TCP_LISTEN state previously but
210 		 * was shutdown() rather than close().
211 		 */
212 		tcp_fastopen = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fastopen);
213 		if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
214 		    (tcp_fastopen & TFO_SERVER_ENABLE) &&
215 		    !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
216 			fastopen_queue_tune(sk, backlog);
217 			tcp_fastopen_init_key_once(sock_net(sk));
218 		}
219 
220 		err = inet_csk_listen_start(sk);
221 		if (err)
222 			return err;
223 
224 		tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL);
225 	}
226 	return 0;
227 }
228 
229 /*
230  *	Move a socket into listening state.
231  */
232 int inet_listen(struct socket *sock, int backlog)
233 {
234 	struct sock *sk = sock->sk;
235 	int err = -EINVAL;
236 
237 	lock_sock(sk);
238 
239 	if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
240 		goto out;
241 
242 	err = __inet_listen_sk(sk, backlog);
243 
244 out:
245 	release_sock(sk);
246 	return err;
247 }
248 EXPORT_SYMBOL(inet_listen);
249 
250 /*
251  *	Create an inet socket.
252  */
253 
254 static int inet_create(struct net *net, struct socket *sock, int protocol,
255 		       int kern)
256 {
257 	struct sock *sk;
258 	struct inet_protosw *answer;
259 	struct inet_sock *inet;
260 	struct proto *answer_prot;
261 	unsigned char answer_flags;
262 	int try_loading_module = 0;
263 	int err;
264 
265 	if (protocol < 0 || protocol >= IPPROTO_MAX)
266 		return -EINVAL;
267 
268 	sock->state = SS_UNCONNECTED;
269 
270 	/* Look for the requested type/protocol pair. */
271 lookup_protocol:
272 	err = -ESOCKTNOSUPPORT;
273 	rcu_read_lock();
274 	list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
275 
276 		err = 0;
277 		/* Check the non-wild match. */
278 		if (protocol == answer->protocol) {
279 			if (protocol != IPPROTO_IP)
280 				break;
281 		} else {
282 			/* Check for the two wild cases. */
283 			if (IPPROTO_IP == protocol) {
284 				protocol = answer->protocol;
285 				break;
286 			}
287 			if (IPPROTO_IP == answer->protocol)
288 				break;
289 		}
290 		err = -EPROTONOSUPPORT;
291 	}
292 
293 	if (unlikely(err)) {
294 		if (try_loading_module < 2) {
295 			rcu_read_unlock();
296 			/*
297 			 * Be more specific, e.g. net-pf-2-proto-132-type-1
298 			 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
299 			 */
300 			if (++try_loading_module == 1)
301 				request_module("net-pf-%d-proto-%d-type-%d",
302 					       PF_INET, protocol, sock->type);
303 			/*
304 			 * Fall back to generic, e.g. net-pf-2-proto-132
305 			 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
306 			 */
307 			else
308 				request_module("net-pf-%d-proto-%d",
309 					       PF_INET, protocol);
310 			goto lookup_protocol;
311 		} else
312 			goto out_rcu_unlock;
313 	}
314 
315 	err = -EPERM;
316 	if (sock->type == SOCK_RAW && !kern &&
317 	    !ns_capable(net->user_ns, CAP_NET_RAW))
318 		goto out_rcu_unlock;
319 
320 	sock->ops = answer->ops;
321 	answer_prot = answer->prot;
322 	answer_flags = answer->flags;
323 	rcu_read_unlock();
324 
325 	WARN_ON(!answer_prot->slab);
326 
327 	err = -ENOMEM;
328 	sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
329 	if (!sk)
330 		goto out;
331 
332 	err = 0;
333 	if (INET_PROTOSW_REUSE & answer_flags)
334 		sk->sk_reuse = SK_CAN_REUSE;
335 
336 	if (INET_PROTOSW_ICSK & answer_flags)
337 		inet_init_csk_locks(sk);
338 
339 	inet = inet_sk(sk);
340 	inet_assign_bit(IS_ICSK, sk, INET_PROTOSW_ICSK & answer_flags);
341 
342 	inet_clear_bit(NODEFRAG, sk);
343 
344 	if (SOCK_RAW == sock->type) {
345 		inet->inet_num = protocol;
346 		if (IPPROTO_RAW == protocol)
347 			inet_set_bit(HDRINCL, sk);
348 	}
349 
350 	if (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc))
351 		inet->pmtudisc = IP_PMTUDISC_DONT;
352 	else
353 		inet->pmtudisc = IP_PMTUDISC_WANT;
354 
355 	atomic_set(&inet->inet_id, 0);
356 
357 	sock_init_data(sock, sk);
358 
359 	sk->sk_destruct	   = inet_sock_destruct;
360 	sk->sk_protocol	   = protocol;
361 	sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
362 	sk->sk_txrehash = READ_ONCE(net->core.sysctl_txrehash);
363 
364 	inet->uc_ttl	= -1;
365 	inet_set_bit(MC_LOOP, sk);
366 	inet->mc_ttl	= 1;
367 	inet_set_bit(MC_ALL, sk);
368 	inet->mc_index	= 0;
369 	inet->mc_list	= NULL;
370 	inet->rcv_tos	= 0;
371 
372 	if (inet->inet_num) {
373 		/* It assumes that any protocol which allows
374 		 * the user to assign a number at socket
375 		 * creation time automatically
376 		 * shares.
377 		 */
378 		inet->inet_sport = htons(inet->inet_num);
379 		/* Add to protocol hash chains. */
380 		err = sk->sk_prot->hash(sk);
381 		if (err)
382 			goto out_sk_release;
383 	}
384 
385 	if (sk->sk_prot->init) {
386 		err = sk->sk_prot->init(sk);
387 		if (err)
388 			goto out_sk_release;
389 	}
390 
391 	if (!kern) {
392 		err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
393 		if (err)
394 			goto out_sk_release;
395 	}
396 out:
397 	return err;
398 out_rcu_unlock:
399 	rcu_read_unlock();
400 	goto out;
401 out_sk_release:
402 	sk_common_release(sk);
403 	sock->sk = NULL;
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_unsized *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_unsized *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_unsized *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_unsized *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_unsized *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_unsized *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 	if (mem_cgroup_sockets_enabled) {
759 		mem_cgroup_sk_alloc(newsk);
760 		__sk_charge(newsk, GFP_KERNEL);
761 	}
762 
763 	sock_rps_record_flow(newsk);
764 	WARN_ON(!((1 << newsk->sk_state) &
765 		  (TCPF_ESTABLISHED | TCPF_SYN_RECV |
766 		   TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2 |
767 		   TCPF_CLOSING | TCPF_CLOSE_WAIT |
768 		   TCPF_CLOSE)));
769 
770 	if (test_bit(SOCK_SUPPORT_ZC, &sock->flags))
771 		set_bit(SOCK_SUPPORT_ZC, &newsock->flags);
772 	sock_graft(newsk, newsock);
773 
774 	newsock->state = SS_CONNECTED;
775 }
776 EXPORT_SYMBOL_GPL(__inet_accept);
777 
778 /*
779  *	Accept a pending connection. The TCP layer now gives BSD semantics.
780  */
781 
782 int inet_accept(struct socket *sock, struct socket *newsock,
783 		struct proto_accept_arg *arg)
784 {
785 	struct sock *sk1 = sock->sk, *sk2;
786 
787 	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
788 	arg->err = -EINVAL;
789 	sk2 = READ_ONCE(sk1->sk_prot)->accept(sk1, arg);
790 	if (!sk2)
791 		return arg->err;
792 
793 	lock_sock(sk2);
794 	__inet_accept(sock, newsock, sk2);
795 	release_sock(sk2);
796 	return 0;
797 }
798 EXPORT_SYMBOL(inet_accept);
799 
800 /*
801  *	This does both peername and sockname.
802  */
803 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
804 		 int peer)
805 {
806 	struct sock *sk		= sock->sk;
807 	struct inet_sock *inet	= inet_sk(sk);
808 	DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
809 	int sin_addr_len = sizeof(*sin);
810 
811 	sin->sin_family = AF_INET;
812 	lock_sock(sk);
813 	if (peer) {
814 		if (!inet->inet_dport ||
815 		    (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
816 		     peer == 1)) {
817 			release_sock(sk);
818 			return -ENOTCONN;
819 		}
820 		sin->sin_port = inet->inet_dport;
821 		sin->sin_addr.s_addr = inet->inet_daddr;
822 		BPF_CGROUP_RUN_SA_PROG(sk, sin, &sin_addr_len,
823 				       CGROUP_INET4_GETPEERNAME);
824 	} else {
825 		__be32 addr = inet->inet_rcv_saddr;
826 		if (!addr)
827 			addr = inet->inet_saddr;
828 		sin->sin_port = inet->inet_sport;
829 		sin->sin_addr.s_addr = addr;
830 		BPF_CGROUP_RUN_SA_PROG(sk, sin, &sin_addr_len,
831 				       CGROUP_INET4_GETSOCKNAME);
832 	}
833 	release_sock(sk);
834 	memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
835 	return sin_addr_len;
836 }
837 EXPORT_SYMBOL(inet_getname);
838 
839 int inet_send_prepare(struct sock *sk)
840 {
841 	sock_rps_record_flow(sk);
842 
843 	/* We may need to bind the socket. */
844 	if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind &&
845 	    inet_autobind(sk))
846 		return -EAGAIN;
847 
848 	return 0;
849 }
850 EXPORT_SYMBOL_GPL(inet_send_prepare);
851 
852 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
853 {
854 	struct sock *sk = sock->sk;
855 	const struct proto *prot;
856 
857 	if (unlikely(inet_send_prepare(sk)))
858 		return -EAGAIN;
859 
860 	prot = READ_ONCE(sk->sk_prot);
861 	return INDIRECT_CALL_2(prot->sendmsg, tcp_sendmsg, udp_sendmsg,
862 			       sk, msg, size);
863 }
864 EXPORT_SYMBOL(inet_sendmsg);
865 
866 void inet_splice_eof(struct socket *sock)
867 {
868 	const struct proto *prot;
869 	struct sock *sk = sock->sk;
870 
871 	if (unlikely(inet_send_prepare(sk)))
872 		return;
873 
874 	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
875 	prot = READ_ONCE(sk->sk_prot);
876 	if (prot->splice_eof)
877 		prot->splice_eof(sock);
878 }
879 EXPORT_SYMBOL_GPL(inet_splice_eof);
880 
881 INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
882 					  size_t, int));
883 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
884 		 int flags)
885 {
886 	struct sock *sk = sock->sk;
887 	const struct proto *prot;
888 
889 	if (likely(!(flags & MSG_ERRQUEUE)))
890 		sock_rps_record_flow(sk);
891 
892 	prot = READ_ONCE(sk->sk_prot);
893 	return INDIRECT_CALL_2(prot->recvmsg, tcp_recvmsg, udp_recvmsg,
894 			       sk, msg, size, flags);
895 }
896 EXPORT_SYMBOL(inet_recvmsg);
897 
898 int inet_shutdown(struct socket *sock, int how)
899 {
900 	struct sock *sk = sock->sk;
901 	int err = 0;
902 
903 	/* This should really check to make sure
904 	 * the socket is a TCP socket. (WHY AC...)
905 	 */
906 	how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
907 		       1->2 bit 2 snds.
908 		       2->3 */
909 	if ((how & ~SHUTDOWN_MASK) || !how)	/* MAXINT->0 */
910 		return -EINVAL;
911 
912 	lock_sock(sk);
913 	if (sock->state == SS_CONNECTING) {
914 		if ((1 << sk->sk_state) &
915 		    (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
916 			sock->state = SS_DISCONNECTING;
917 		else
918 			sock->state = SS_CONNECTED;
919 	}
920 
921 	switch (sk->sk_state) {
922 	case TCP_CLOSE:
923 		err = -ENOTCONN;
924 		/* Hack to wake up other listeners, who can poll for
925 		   EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
926 		fallthrough;
927 	default:
928 		WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | how);
929 		if (sk->sk_prot->shutdown)
930 			sk->sk_prot->shutdown(sk, how);
931 		break;
932 
933 	/* Remaining two branches are temporary solution for missing
934 	 * close() in multithreaded environment. It is _not_ a good idea,
935 	 * but we have no choice until close() is repaired at VFS level.
936 	 */
937 	case TCP_LISTEN:
938 		if (!(how & RCV_SHUTDOWN))
939 			break;
940 		fallthrough;
941 	case TCP_SYN_SENT:
942 		err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
943 		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
944 		break;
945 	}
946 
947 	/* Wake up anyone sleeping in poll. */
948 	sk->sk_state_change(sk);
949 	release_sock(sk);
950 	return err;
951 }
952 EXPORT_SYMBOL(inet_shutdown);
953 
954 /*
955  *	ioctl() calls you can issue on an INET socket. Most of these are
956  *	device configuration and stuff and very rarely used. Some ioctls
957  *	pass on to the socket itself.
958  *
959  *	NOTE: I like the idea of a module for the config stuff. ie ifconfig
960  *	loads the devconfigure module does its configuring and unloads it.
961  *	There's a good 20K of config code hanging around the kernel.
962  */
963 
964 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
965 {
966 	struct sock *sk = sock->sk;
967 	int err = 0;
968 	struct net *net = sock_net(sk);
969 	void __user *p = (void __user *)arg;
970 	struct ifreq ifr;
971 	struct rtentry rt;
972 
973 	switch (cmd) {
974 	case SIOCADDRT:
975 	case SIOCDELRT:
976 		if (copy_from_user(&rt, p, sizeof(struct rtentry)))
977 			return -EFAULT;
978 		err = ip_rt_ioctl(net, cmd, &rt);
979 		break;
980 	case SIOCRTMSG:
981 		err = -EINVAL;
982 		break;
983 	case SIOCDARP:
984 	case SIOCGARP:
985 	case SIOCSARP:
986 		err = arp_ioctl(net, cmd, (void __user *)arg);
987 		break;
988 	case SIOCGIFADDR:
989 	case SIOCGIFBRDADDR:
990 	case SIOCGIFNETMASK:
991 	case SIOCGIFDSTADDR:
992 	case SIOCGIFPFLAGS:
993 		if (get_user_ifreq(&ifr, NULL, p))
994 			return -EFAULT;
995 		err = devinet_ioctl(net, cmd, &ifr);
996 		if (!err && put_user_ifreq(&ifr, p))
997 			err = -EFAULT;
998 		break;
999 
1000 	case SIOCSIFADDR:
1001 	case SIOCSIFBRDADDR:
1002 	case SIOCSIFNETMASK:
1003 	case SIOCSIFDSTADDR:
1004 	case SIOCSIFPFLAGS:
1005 	case SIOCSIFFLAGS:
1006 		if (get_user_ifreq(&ifr, NULL, p))
1007 			return -EFAULT;
1008 		err = devinet_ioctl(net, cmd, &ifr);
1009 		break;
1010 	default:
1011 		if (sk->sk_prot->ioctl)
1012 			err = sk_ioctl(sk, cmd, (void __user *)arg);
1013 		else
1014 			err = -ENOIOCTLCMD;
1015 		break;
1016 	}
1017 	return err;
1018 }
1019 EXPORT_SYMBOL(inet_ioctl);
1020 
1021 #ifdef CONFIG_COMPAT
1022 static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd,
1023 		struct compat_rtentry __user *ur)
1024 {
1025 	compat_uptr_t rtdev;
1026 	struct rtentry rt;
1027 
1028 	if (copy_from_user(&rt.rt_dst, &ur->rt_dst,
1029 			3 * sizeof(struct sockaddr)) ||
1030 	    get_user(rt.rt_flags, &ur->rt_flags) ||
1031 	    get_user(rt.rt_metric, &ur->rt_metric) ||
1032 	    get_user(rt.rt_mtu, &ur->rt_mtu) ||
1033 	    get_user(rt.rt_window, &ur->rt_window) ||
1034 	    get_user(rt.rt_irtt, &ur->rt_irtt) ||
1035 	    get_user(rtdev, &ur->rt_dev))
1036 		return -EFAULT;
1037 
1038 	rt.rt_dev = compat_ptr(rtdev);
1039 	return ip_rt_ioctl(sock_net(sk), cmd, &rt);
1040 }
1041 
1042 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1043 {
1044 	void __user *argp = compat_ptr(arg);
1045 	struct sock *sk = sock->sk;
1046 
1047 	switch (cmd) {
1048 	case SIOCADDRT:
1049 	case SIOCDELRT:
1050 		return inet_compat_routing_ioctl(sk, cmd, argp);
1051 	default:
1052 		if (!sk->sk_prot->compat_ioctl)
1053 			return -ENOIOCTLCMD;
1054 		return sk->sk_prot->compat_ioctl(sk, cmd, arg);
1055 	}
1056 }
1057 #endif /* CONFIG_COMPAT */
1058 
1059 const struct proto_ops inet_stream_ops = {
1060 	.family		   = PF_INET,
1061 	.owner		   = THIS_MODULE,
1062 	.release	   = inet_release,
1063 	.bind		   = inet_bind,
1064 	.connect	   = inet_stream_connect,
1065 	.socketpair	   = sock_no_socketpair,
1066 	.accept		   = inet_accept,
1067 	.getname	   = inet_getname,
1068 	.poll		   = tcp_poll,
1069 	.ioctl		   = inet_ioctl,
1070 	.gettstamp	   = sock_gettstamp,
1071 	.listen		   = inet_listen,
1072 	.shutdown	   = inet_shutdown,
1073 	.setsockopt	   = sock_common_setsockopt,
1074 	.getsockopt	   = sock_common_getsockopt,
1075 	.sendmsg	   = inet_sendmsg,
1076 	.recvmsg	   = inet_recvmsg,
1077 #ifdef CONFIG_MMU
1078 	.mmap		   = tcp_mmap,
1079 #endif
1080 	.splice_eof	   = inet_splice_eof,
1081 	.splice_read	   = tcp_splice_read,
1082 	.set_peek_off      = sk_set_peek_off,
1083 	.read_sock	   = tcp_read_sock,
1084 	.read_skb	   = tcp_read_skb,
1085 	.sendmsg_locked    = tcp_sendmsg_locked,
1086 	.peek_len	   = tcp_peek_len,
1087 #ifdef CONFIG_COMPAT
1088 	.compat_ioctl	   = inet_compat_ioctl,
1089 #endif
1090 	.set_rcvlowat	   = tcp_set_rcvlowat,
1091 };
1092 EXPORT_SYMBOL(inet_stream_ops);
1093 
1094 const struct proto_ops inet_dgram_ops = {
1095 	.family		   = PF_INET,
1096 	.owner		   = THIS_MODULE,
1097 	.release	   = inet_release,
1098 	.bind		   = inet_bind,
1099 	.connect	   = inet_dgram_connect,
1100 	.socketpair	   = sock_no_socketpair,
1101 	.accept		   = sock_no_accept,
1102 	.getname	   = inet_getname,
1103 	.poll		   = udp_poll,
1104 	.ioctl		   = inet_ioctl,
1105 	.gettstamp	   = sock_gettstamp,
1106 	.listen		   = sock_no_listen,
1107 	.shutdown	   = inet_shutdown,
1108 	.setsockopt	   = sock_common_setsockopt,
1109 	.getsockopt	   = sock_common_getsockopt,
1110 	.sendmsg	   = inet_sendmsg,
1111 	.read_skb	   = udp_read_skb,
1112 	.recvmsg	   = inet_recvmsg,
1113 	.mmap		   = sock_no_mmap,
1114 	.splice_eof	   = inet_splice_eof,
1115 	.set_peek_off	   = udp_set_peek_off,
1116 #ifdef CONFIG_COMPAT
1117 	.compat_ioctl	   = inet_compat_ioctl,
1118 #endif
1119 };
1120 EXPORT_SYMBOL(inet_dgram_ops);
1121 
1122 /*
1123  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1124  * udp_poll
1125  */
1126 static const struct proto_ops inet_sockraw_ops = {
1127 	.family		   = PF_INET,
1128 	.owner		   = THIS_MODULE,
1129 	.release	   = inet_release,
1130 	.bind		   = inet_bind,
1131 	.connect	   = inet_dgram_connect,
1132 	.socketpair	   = sock_no_socketpair,
1133 	.accept		   = sock_no_accept,
1134 	.getname	   = inet_getname,
1135 	.poll		   = datagram_poll,
1136 	.ioctl		   = inet_ioctl,
1137 	.gettstamp	   = sock_gettstamp,
1138 	.listen		   = sock_no_listen,
1139 	.shutdown	   = inet_shutdown,
1140 	.setsockopt	   = sock_common_setsockopt,
1141 	.getsockopt	   = sock_common_getsockopt,
1142 	.sendmsg	   = inet_sendmsg,
1143 	.recvmsg	   = inet_recvmsg,
1144 	.mmap		   = sock_no_mmap,
1145 	.splice_eof	   = inet_splice_eof,
1146 #ifdef CONFIG_COMPAT
1147 	.compat_ioctl	   = inet_compat_ioctl,
1148 #endif
1149 };
1150 
1151 static const struct net_proto_family inet_family_ops = {
1152 	.family = PF_INET,
1153 	.create = inet_create,
1154 	.owner	= THIS_MODULE,
1155 };
1156 
1157 /* Upon startup we insert all the elements in inetsw_array[] into
1158  * the linked list inetsw.
1159  */
1160 static struct inet_protosw inetsw_array[] =
1161 {
1162 	{
1163 		.type =       SOCK_STREAM,
1164 		.protocol =   IPPROTO_TCP,
1165 		.prot =       &tcp_prot,
1166 		.ops =        &inet_stream_ops,
1167 		.flags =      INET_PROTOSW_PERMANENT |
1168 			      INET_PROTOSW_ICSK,
1169 	},
1170 
1171 	{
1172 		.type =       SOCK_DGRAM,
1173 		.protocol =   IPPROTO_UDP,
1174 		.prot =       &udp_prot,
1175 		.ops =        &inet_dgram_ops,
1176 		.flags =      INET_PROTOSW_PERMANENT,
1177        },
1178 
1179        {
1180 		.type =       SOCK_DGRAM,
1181 		.protocol =   IPPROTO_ICMP,
1182 		.prot =       &ping_prot,
1183 		.ops =        &inet_sockraw_ops,
1184 		.flags =      INET_PROTOSW_REUSE,
1185        },
1186 
1187        {
1188 	       .type =       SOCK_RAW,
1189 	       .protocol =   IPPROTO_IP,	/* wild card */
1190 	       .prot =       &raw_prot,
1191 	       .ops =        &inet_sockraw_ops,
1192 	       .flags =      INET_PROTOSW_REUSE,
1193        }
1194 };
1195 
1196 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1197 
1198 void inet_register_protosw(struct inet_protosw *p)
1199 {
1200 	struct list_head *lh;
1201 	struct inet_protosw *answer;
1202 	int protocol = p->protocol;
1203 	struct list_head *last_perm;
1204 
1205 	spin_lock_bh(&inetsw_lock);
1206 
1207 	if (p->type >= SOCK_MAX)
1208 		goto out_illegal;
1209 
1210 	/* If we are trying to override a permanent protocol, bail. */
1211 	last_perm = &inetsw[p->type];
1212 	list_for_each(lh, &inetsw[p->type]) {
1213 		answer = list_entry(lh, struct inet_protosw, list);
1214 		/* Check only the non-wild match. */
1215 		if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1216 			break;
1217 		if (protocol == answer->protocol)
1218 			goto out_permanent;
1219 		last_perm = lh;
1220 	}
1221 
1222 	/* Add the new entry after the last permanent entry if any, so that
1223 	 * the new entry does not override a permanent entry when matched with
1224 	 * a wild-card protocol. But it is allowed to override any existing
1225 	 * non-permanent entry.  This means that when we remove this entry, the
1226 	 * system automatically returns to the old behavior.
1227 	 */
1228 	list_add_rcu(&p->list, last_perm);
1229 out:
1230 	spin_unlock_bh(&inetsw_lock);
1231 
1232 	return;
1233 
1234 out_permanent:
1235 	pr_err("Attempt to override permanent protocol %d\n", protocol);
1236 	goto out;
1237 
1238 out_illegal:
1239 	pr_err("Ignoring attempt to register invalid socket type %d\n",
1240 	       p->type);
1241 	goto out;
1242 }
1243 EXPORT_SYMBOL(inet_register_protosw);
1244 
1245 void inet_unregister_protosw(struct inet_protosw *p)
1246 {
1247 	if (INET_PROTOSW_PERMANENT & p->flags) {
1248 		pr_err("Attempt to unregister permanent protocol %d\n",
1249 		       p->protocol);
1250 	} else {
1251 		spin_lock_bh(&inetsw_lock);
1252 		list_del_rcu(&p->list);
1253 		spin_unlock_bh(&inetsw_lock);
1254 
1255 		synchronize_net();
1256 	}
1257 }
1258 EXPORT_SYMBOL(inet_unregister_protosw);
1259 
1260 static int inet_sk_reselect_saddr(struct sock *sk)
1261 {
1262 	struct inet_sock *inet = inet_sk(sk);
1263 	__be32 old_saddr = inet->inet_saddr;
1264 	__be32 daddr = inet->inet_daddr;
1265 	struct flowi4 *fl4;
1266 	struct rtable *rt;
1267 	__be32 new_saddr;
1268 	struct ip_options_rcu *inet_opt;
1269 	int err;
1270 
1271 	inet_opt = rcu_dereference_protected(inet->inet_opt,
1272 					     lockdep_sock_is_held(sk));
1273 	if (inet_opt && inet_opt->opt.srr)
1274 		daddr = inet_opt->opt.faddr;
1275 
1276 	/* Query new route. */
1277 	fl4 = &inet->cork.fl.u.ip4;
1278 	rt = ip_route_connect(fl4, daddr, 0, sk->sk_bound_dev_if,
1279 			      sk->sk_protocol, inet->inet_sport,
1280 			      inet->inet_dport, sk);
1281 	if (IS_ERR(rt))
1282 		return PTR_ERR(rt);
1283 
1284 	new_saddr = fl4->saddr;
1285 
1286 	if (new_saddr == old_saddr) {
1287 		sk_setup_caps(sk, &rt->dst);
1288 		return 0;
1289 	}
1290 
1291 	err = inet_bhash2_update_saddr(sk, &new_saddr, AF_INET);
1292 	if (err) {
1293 		ip_rt_put(rt);
1294 		return err;
1295 	}
1296 
1297 	sk_setup_caps(sk, &rt->dst);
1298 
1299 	if (READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) > 1) {
1300 		pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1301 			__func__, &old_saddr, &new_saddr);
1302 	}
1303 
1304 	/*
1305 	 * XXX The only one ugly spot where we need to
1306 	 * XXX really change the sockets identity after
1307 	 * XXX it has entered the hashes. -DaveM
1308 	 *
1309 	 * Besides that, it does not check for connection
1310 	 * uniqueness. Wait for troubles.
1311 	 */
1312 	return __sk_prot_rehash(sk);
1313 }
1314 
1315 int inet_sk_rebuild_header(struct sock *sk)
1316 {
1317 	struct rtable *rt = dst_rtable(__sk_dst_check(sk, 0));
1318 	struct inet_sock *inet = inet_sk(sk);
1319 	struct flowi4 *fl4;
1320 	int err;
1321 
1322 	/* Route is OK, nothing to do. */
1323 	if (rt)
1324 		return 0;
1325 
1326 	/* Reroute. */
1327 	fl4 = &inet->cork.fl.u.ip4;
1328 	inet_sk_init_flowi4(inet, fl4);
1329 	rt = ip_route_output_flow(sock_net(sk), fl4, sk);
1330 	if (!IS_ERR(rt)) {
1331 		err = 0;
1332 		sk_setup_caps(sk, &rt->dst);
1333 	} else {
1334 		err = PTR_ERR(rt);
1335 
1336 		/* Routing failed... */
1337 		sk->sk_route_caps = 0;
1338 
1339 		if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) ||
1340 		    sk->sk_state != TCP_SYN_SENT ||
1341 		    (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1342 		    (err = inet_sk_reselect_saddr(sk)) != 0)
1343 			WRITE_ONCE(sk->sk_err_soft, -err);
1344 	}
1345 
1346 	return err;
1347 }
1348 EXPORT_SYMBOL(inet_sk_rebuild_header);
1349 
1350 void inet_sk_set_state(struct sock *sk, int state)
1351 {
1352 	trace_inet_sock_set_state(sk, sk->sk_state, state);
1353 	sk->sk_state = state;
1354 }
1355 EXPORT_SYMBOL(inet_sk_set_state);
1356 
1357 void inet_sk_state_store(struct sock *sk, int newstate)
1358 {
1359 	trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1360 	smp_store_release(&sk->sk_state, newstate);
1361 }
1362 
1363 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1364 				 netdev_features_t features)
1365 {
1366 	bool udpfrag = false, fixedid = false, gso_partial, encap;
1367 	struct sk_buff *segs = ERR_PTR(-EINVAL);
1368 	const struct net_offload *ops;
1369 	unsigned int offset = 0;
1370 	struct iphdr *iph;
1371 	int proto, tot_len;
1372 	int nhoff;
1373 	int ihl;
1374 	int id;
1375 
1376 	skb_reset_network_header(skb);
1377 	nhoff = skb_network_header(skb) - skb_mac_header(skb);
1378 	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1379 		goto out;
1380 
1381 	iph = ip_hdr(skb);
1382 	ihl = iph->ihl * 4;
1383 	if (ihl < sizeof(*iph))
1384 		goto out;
1385 
1386 	id = ntohs(iph->id);
1387 	proto = iph->protocol;
1388 
1389 	/* Warning: after this point, iph might be no longer valid */
1390 	if (unlikely(!pskb_may_pull(skb, ihl)))
1391 		goto out;
1392 	__skb_pull(skb, ihl);
1393 
1394 	encap = SKB_GSO_CB(skb)->encap_level > 0;
1395 	if (encap)
1396 		features &= skb->dev->hw_enc_features;
1397 	SKB_GSO_CB(skb)->encap_level += ihl;
1398 
1399 	skb_reset_transport_header(skb);
1400 
1401 	segs = ERR_PTR(-EPROTONOSUPPORT);
1402 
1403 	fixedid = !!(skb_shinfo(skb)->gso_type & (SKB_GSO_TCP_FIXEDID << encap));
1404 
1405 	if (!skb->encapsulation || encap)
1406 		udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1407 
1408 	ops = rcu_dereference(inet_offloads[proto]);
1409 	if (likely(ops && ops->callbacks.gso_segment)) {
1410 		segs = ops->callbacks.gso_segment(skb, features);
1411 		if (!segs)
1412 			skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
1413 	}
1414 
1415 	if (IS_ERR_OR_NULL(segs))
1416 		goto out;
1417 
1418 	gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1419 
1420 	skb = segs;
1421 	do {
1422 		iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1423 		if (udpfrag) {
1424 			iph->frag_off = htons(offset >> 3);
1425 			if (skb->next)
1426 				iph->frag_off |= htons(IP_MF);
1427 			offset += skb->len - nhoff - ihl;
1428 			tot_len = skb->len - nhoff;
1429 		} else if (skb_is_gso(skb)) {
1430 			if (!fixedid) {
1431 				iph->id = htons(id);
1432 				id += skb_shinfo(skb)->gso_segs;
1433 			}
1434 
1435 			if (gso_partial)
1436 				tot_len = skb_shinfo(skb)->gso_size +
1437 					  SKB_GSO_CB(skb)->data_offset +
1438 					  skb->head - (unsigned char *)iph;
1439 			else
1440 				tot_len = skb->len - nhoff;
1441 		} else {
1442 			if (!fixedid)
1443 				iph->id = htons(id++);
1444 			tot_len = skb->len - nhoff;
1445 		}
1446 		iph->tot_len = htons(tot_len);
1447 		ip_send_check(iph);
1448 		if (encap)
1449 			skb_reset_inner_headers(skb);
1450 		skb->network_header = (u8 *)iph - skb->head;
1451 		skb_reset_mac_len(skb);
1452 	} while ((skb = skb->next));
1453 
1454 out:
1455 	return segs;
1456 }
1457 
1458 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1459 					netdev_features_t features)
1460 {
1461 	if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1462 		return ERR_PTR(-EINVAL);
1463 
1464 	return inet_gso_segment(skb, features);
1465 }
1466 
1467 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1468 {
1469 	const struct net_offload *ops;
1470 	struct sk_buff *pp = NULL;
1471 	const struct iphdr *iph;
1472 	struct sk_buff *p;
1473 	unsigned int hlen;
1474 	unsigned int off;
1475 	int flush = 1;
1476 	int proto;
1477 
1478 	off = skb_gro_offset(skb);
1479 	hlen = off + sizeof(*iph);
1480 	iph = skb_gro_header(skb, hlen, off);
1481 	if (unlikely(!iph))
1482 		goto out;
1483 
1484 	proto = iph->protocol;
1485 
1486 	ops = rcu_dereference(inet_offloads[proto]);
1487 	if (!ops || !ops->callbacks.gro_receive)
1488 		goto out;
1489 
1490 	if (*(u8 *)iph != 0x45)
1491 		goto out;
1492 
1493 	if (ip_is_fragment(iph))
1494 		goto out;
1495 
1496 	if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1497 		goto out;
1498 
1499 	NAPI_GRO_CB(skb)->proto = proto;
1500 	flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (ntohl(*(__be32 *)&iph->id) & ~IP_DF));
1501 
1502 	list_for_each_entry(p, head, list) {
1503 		struct iphdr *iph2;
1504 
1505 		if (!NAPI_GRO_CB(p)->same_flow)
1506 			continue;
1507 
1508 		iph2 = (struct iphdr *)(p->data + off);
1509 		/* The above works because, with the exception of the top
1510 		 * (inner most) layer, we only aggregate pkts with the same
1511 		 * hdr length so all the hdrs we'll need to verify will start
1512 		 * at the same offset.
1513 		 */
1514 		if ((iph->protocol ^ iph2->protocol) |
1515 		    ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1516 		    ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1517 			NAPI_GRO_CB(p)->same_flow = 0;
1518 			continue;
1519 		}
1520 	}
1521 
1522 	NAPI_GRO_CB(skb)->flush |= flush;
1523 	NAPI_GRO_CB(skb)->network_offsets[NAPI_GRO_CB(skb)->encap_mark] = off;
1524 
1525 	/* Note : No need to call skb_gro_postpull_rcsum() here,
1526 	 * as we already checked checksum over ipv4 header was 0
1527 	 */
1528 	skb_gro_pull(skb, sizeof(*iph));
1529 	skb_set_transport_header(skb, skb_gro_offset(skb));
1530 
1531 	pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1532 				       ops->callbacks.gro_receive, head, skb);
1533 
1534 out:
1535 	skb_gro_flush_final(skb, pp, flush);
1536 
1537 	return pp;
1538 }
1539 
1540 static struct sk_buff *ipip_gro_receive(struct list_head *head,
1541 					struct sk_buff *skb)
1542 {
1543 	if (NAPI_GRO_CB(skb)->encap_mark) {
1544 		NAPI_GRO_CB(skb)->flush = 1;
1545 		return NULL;
1546 	}
1547 
1548 	NAPI_GRO_CB(skb)->encap_mark = 1;
1549 
1550 	return inet_gro_receive(head, skb);
1551 }
1552 
1553 #define SECONDS_PER_DAY	86400
1554 
1555 /* inet_current_timestamp - Return IP network timestamp
1556  *
1557  * Return milliseconds since midnight in network byte order.
1558  */
1559 __be32 inet_current_timestamp(void)
1560 {
1561 	u32 secs;
1562 	u32 msecs;
1563 	struct timespec64 ts;
1564 
1565 	ktime_get_real_ts64(&ts);
1566 
1567 	/* Get secs since midnight. */
1568 	(void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1569 	/* Convert to msecs. */
1570 	msecs = secs * MSEC_PER_SEC;
1571 	/* Convert nsec to msec. */
1572 	msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1573 
1574 	/* Convert to network byte order. */
1575 	return htonl(msecs);
1576 }
1577 EXPORT_SYMBOL(inet_current_timestamp);
1578 
1579 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len)
1580 {
1581 	unsigned int family = READ_ONCE(sk->sk_family);
1582 
1583 	if (family == AF_INET)
1584 		return ip_recv_error(sk, msg, len);
1585 #if IS_ENABLED(CONFIG_IPV6)
1586 	if (family == AF_INET6)
1587 		return pingv6_ops.ipv6_recv_error(sk, msg, len);
1588 #endif
1589 	return -EINVAL;
1590 }
1591 EXPORT_SYMBOL(inet_recv_error);
1592 
1593 int inet_gro_complete(struct sk_buff *skb, int nhoff)
1594 {
1595 	struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1596 	const struct net_offload *ops;
1597 	__be16 totlen = iph->tot_len;
1598 	int proto = iph->protocol;
1599 	int err = -ENOSYS;
1600 
1601 	if (skb->encapsulation) {
1602 		skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1603 		skb_set_inner_network_header(skb, nhoff);
1604 	}
1605 
1606 	iph_set_totlen(iph, skb->len - nhoff);
1607 	csum_replace2(&iph->check, totlen, iph->tot_len);
1608 
1609 	ops = rcu_dereference(inet_offloads[proto]);
1610 	if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1611 		goto out;
1612 
1613 	/* Only need to add sizeof(*iph) to get to the next hdr below
1614 	 * because any hdr with option will have been flushed in
1615 	 * inet_gro_receive().
1616 	 */
1617 	err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1618 			      tcp4_gro_complete, udp4_gro_complete,
1619 			      skb, nhoff + sizeof(*iph));
1620 
1621 out:
1622 	return err;
1623 }
1624 
1625 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1626 {
1627 	skb->encapsulation = 1;
1628 	skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1629 	return inet_gro_complete(skb, nhoff);
1630 }
1631 
1632 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1633 			 unsigned short type, unsigned char protocol,
1634 			 struct net *net)
1635 {
1636 	struct socket *sock;
1637 	int rc = sock_create_kern(net, family, type, protocol, &sock);
1638 
1639 	if (rc == 0) {
1640 		*sk = sock->sk;
1641 		(*sk)->sk_allocation = GFP_ATOMIC;
1642 		(*sk)->sk_use_task_frag = false;
1643 		/*
1644 		 * Unhash it so that IP input processing does not even see it,
1645 		 * we do not wish this socket to see incoming packets.
1646 		 */
1647 		(*sk)->sk_prot->unhash(*sk);
1648 	}
1649 	return rc;
1650 }
1651 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1652 
1653 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1654 {
1655 	unsigned long res = 0;
1656 	int i;
1657 
1658 	for_each_possible_cpu(i)
1659 		res += snmp_get_cpu_field(mib, i, offt);
1660 	return res;
1661 }
1662 EXPORT_SYMBOL_GPL(snmp_fold_field);
1663 
1664 #if BITS_PER_LONG==32
1665 
1666 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1667 			 size_t syncp_offset)
1668 {
1669 	void *bhptr;
1670 	struct u64_stats_sync *syncp;
1671 	u64 v;
1672 	unsigned int start;
1673 
1674 	bhptr = per_cpu_ptr(mib, cpu);
1675 	syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1676 	do {
1677 		start = u64_stats_fetch_begin(syncp);
1678 		v = *(((u64 *)bhptr) + offt);
1679 	} while (u64_stats_fetch_retry(syncp, start));
1680 
1681 	return v;
1682 }
1683 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1684 
1685 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1686 {
1687 	u64 res = 0;
1688 	int cpu;
1689 
1690 	for_each_possible_cpu(cpu) {
1691 		res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1692 	}
1693 	return res;
1694 }
1695 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1696 #endif
1697 
1698 #ifdef CONFIG_IP_MULTICAST
1699 static const struct net_protocol igmp_protocol = {
1700 	.handler =	igmp_rcv,
1701 };
1702 #endif
1703 
1704 static const struct net_protocol icmp_protocol = {
1705 	.handler =	icmp_rcv,
1706 	.err_handler =	icmp_err,
1707 	.no_policy =	1,
1708 };
1709 
1710 static __net_init int ipv4_mib_init_net(struct net *net)
1711 {
1712 	int i;
1713 
1714 	net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1715 	if (!net->mib.tcp_statistics)
1716 		goto err_tcp_mib;
1717 	net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1718 	if (!net->mib.ip_statistics)
1719 		goto err_ip_mib;
1720 
1721 	for_each_possible_cpu(i) {
1722 		struct ipstats_mib *af_inet_stats;
1723 		af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1724 		u64_stats_init(&af_inet_stats->syncp);
1725 	}
1726 
1727 	net->mib.net_statistics = alloc_percpu(struct linux_mib);
1728 	if (!net->mib.net_statistics)
1729 		goto err_net_mib;
1730 	net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1731 	if (!net->mib.udp_statistics)
1732 		goto err_udp_mib;
1733 	net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1734 	if (!net->mib.udplite_statistics)
1735 		goto err_udplite_mib;
1736 	net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1737 	if (!net->mib.icmp_statistics)
1738 		goto err_icmp_mib;
1739 	net->mib.icmpmsg_statistics = kzalloc_obj(struct icmpmsg_mib);
1740 	if (!net->mib.icmpmsg_statistics)
1741 		goto err_icmpmsg_mib;
1742 
1743 	tcp_mib_init(net);
1744 	return 0;
1745 
1746 err_icmpmsg_mib:
1747 	free_percpu(net->mib.icmp_statistics);
1748 err_icmp_mib:
1749 	free_percpu(net->mib.udplite_statistics);
1750 err_udplite_mib:
1751 	free_percpu(net->mib.udp_statistics);
1752 err_udp_mib:
1753 	free_percpu(net->mib.net_statistics);
1754 err_net_mib:
1755 	free_percpu(net->mib.ip_statistics);
1756 err_ip_mib:
1757 	free_percpu(net->mib.tcp_statistics);
1758 err_tcp_mib:
1759 	return -ENOMEM;
1760 }
1761 
1762 static __net_exit void ipv4_mib_exit_net(struct net *net)
1763 {
1764 	kfree(net->mib.icmpmsg_statistics);
1765 	free_percpu(net->mib.icmp_statistics);
1766 	free_percpu(net->mib.udplite_statistics);
1767 	free_percpu(net->mib.udp_statistics);
1768 	free_percpu(net->mib.net_statistics);
1769 	free_percpu(net->mib.ip_statistics);
1770 	free_percpu(net->mib.tcp_statistics);
1771 #ifdef CONFIG_MPTCP
1772 	/* allocated on demand, see mptcp_init_sock() */
1773 	free_percpu(net->mib.mptcp_statistics);
1774 #endif
1775 }
1776 
1777 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1778 	.init = ipv4_mib_init_net,
1779 	.exit = ipv4_mib_exit_net,
1780 };
1781 
1782 static int __init init_ipv4_mibs(void)
1783 {
1784 	return register_pernet_subsys(&ipv4_mib_ops);
1785 }
1786 
1787 static __net_init int inet_init_net(struct net *net)
1788 {
1789 	/*
1790 	 * Set defaults for local port range
1791 	 */
1792 	net->ipv4.ip_local_ports.range = 60999u << 16 | 32768u;
1793 
1794 	seqlock_init(&net->ipv4.ping_group_range.lock);
1795 	/*
1796 	 * Sane defaults - nobody may create ping sockets.
1797 	 * Boot scripts should set this to distro-specific group.
1798 	 */
1799 	net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1800 	net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1801 
1802 	/* Default values for sysctl-controlled parameters.
1803 	 * We set them here, in case sysctl is not compiled.
1804 	 */
1805 	net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1806 	net->ipv4.sysctl_ip_fwd_update_priority = 1;
1807 	net->ipv4.sysctl_ip_dynaddr = 0;
1808 	net->ipv4.sysctl_ip_early_demux = 1;
1809 	net->ipv4.sysctl_udp_early_demux = 1;
1810 	net->ipv4.sysctl_tcp_early_demux = 1;
1811 	net->ipv4.sysctl_nexthop_compat_mode = 1;
1812 #ifdef CONFIG_SYSCTL
1813 	net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1814 #endif
1815 
1816 	/* Some igmp sysctl, whose values are always used */
1817 	net->ipv4.sysctl_igmp_max_memberships = 20;
1818 	net->ipv4.sysctl_igmp_max_msf = 10;
1819 	/* IGMP reports for link-local multicast groups are enabled by default */
1820 	net->ipv4.sysctl_igmp_llm_reports = 1;
1821 	net->ipv4.sysctl_igmp_qrv = 2;
1822 
1823 	net->ipv4.sysctl_fib_notify_on_flag_change = 0;
1824 
1825 	return 0;
1826 }
1827 
1828 static __net_initdata struct pernet_operations af_inet_ops = {
1829 	.init = inet_init_net,
1830 };
1831 
1832 static int __init init_inet_pernet_ops(void)
1833 {
1834 	return register_pernet_subsys(&af_inet_ops);
1835 }
1836 
1837 static int ipv4_proc_init(void);
1838 
1839 /*
1840  *	IP protocol layer initialiser
1841  */
1842 
1843 
1844 static const struct net_offload ipip_offload = {
1845 	.callbacks = {
1846 		.gso_segment	= ipip_gso_segment,
1847 		.gro_receive	= ipip_gro_receive,
1848 		.gro_complete	= ipip_gro_complete,
1849 	},
1850 };
1851 
1852 static int __init ipip_offload_init(void)
1853 {
1854 	return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1855 }
1856 
1857 static int __init ipv4_offload_init(void)
1858 {
1859 	/*
1860 	 * Add offloads
1861 	 */
1862 	if (udpv4_offload_init() < 0)
1863 		pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1864 	if (tcpv4_offload_init() < 0)
1865 		pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1866 	if (ipip_offload_init() < 0)
1867 		pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1868 
1869 	net_hotdata.ip_packet_offload = (struct packet_offload) {
1870 		.type = cpu_to_be16(ETH_P_IP),
1871 		.callbacks = {
1872 			.gso_segment = inet_gso_segment,
1873 			.gro_receive = inet_gro_receive,
1874 			.gro_complete = inet_gro_complete,
1875 		},
1876 	};
1877 	dev_add_offload(&net_hotdata.ip_packet_offload);
1878 	return 0;
1879 }
1880 
1881 fs_initcall(ipv4_offload_init);
1882 
1883 static struct packet_type ip_packet_type __read_mostly = {
1884 	.type = cpu_to_be16(ETH_P_IP),
1885 	.func = ip_rcv,
1886 	.list_func = ip_list_rcv,
1887 };
1888 
1889 static int __init inet_init(void)
1890 {
1891 	struct inet_protosw *q;
1892 	struct list_head *r;
1893 	int rc;
1894 
1895 	sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1896 
1897 	raw_hashinfo_init(&raw_v4_hashinfo);
1898 
1899 	rc = proto_register(&tcp_prot, 1);
1900 	if (rc)
1901 		goto out;
1902 
1903 	rc = proto_register(&udp_prot, 1);
1904 	if (rc)
1905 		goto out_unregister_tcp_proto;
1906 
1907 	rc = proto_register(&raw_prot, 1);
1908 	if (rc)
1909 		goto out_unregister_udp_proto;
1910 
1911 	rc = proto_register(&ping_prot, 1);
1912 	if (rc)
1913 		goto out_unregister_raw_proto;
1914 
1915 	/*
1916 	 *	Tell SOCKET that we are alive...
1917 	 */
1918 
1919 	(void)sock_register(&inet_family_ops);
1920 
1921 #ifdef CONFIG_SYSCTL
1922 	ip_static_sysctl_init();
1923 #endif
1924 
1925 	/*
1926 	 *	Add all the base protocols.
1927 	 */
1928 
1929 	if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1930 		pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1931 
1932 	net_hotdata.udp_protocol = (struct net_protocol) {
1933 		.handler =	udp_rcv,
1934 		.err_handler =	udp_err,
1935 		.no_policy =	1,
1936 	};
1937 	if (inet_add_protocol(&net_hotdata.udp_protocol, IPPROTO_UDP) < 0)
1938 		pr_crit("%s: Cannot add UDP protocol\n", __func__);
1939 
1940 	net_hotdata.tcp_protocol = (struct net_protocol) {
1941 		.handler	=	tcp_v4_rcv,
1942 		.err_handler	=	tcp_v4_err,
1943 		.no_policy	=	1,
1944 		.icmp_strict_tag_validation = 1,
1945 	};
1946 	if (inet_add_protocol(&net_hotdata.tcp_protocol, IPPROTO_TCP) < 0)
1947 		pr_crit("%s: Cannot add TCP protocol\n", __func__);
1948 #ifdef CONFIG_IP_MULTICAST
1949 	if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1950 		pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1951 #endif
1952 
1953 	/* Register the socket-side information for inet_create. */
1954 	for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1955 		INIT_LIST_HEAD(r);
1956 
1957 	for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1958 		inet_register_protosw(q);
1959 
1960 	/*
1961 	 *	Set the ARP module up
1962 	 */
1963 
1964 	arp_init();
1965 
1966 	/*
1967 	 *	Set the IP module up
1968 	 */
1969 
1970 	ip_init();
1971 
1972 	/* Initialise per-cpu ipv4 mibs */
1973 	if (init_ipv4_mibs())
1974 		panic("%s: Cannot init ipv4 mibs\n", __func__);
1975 
1976 	/* Setup TCP slab cache for open requests. */
1977 	tcp_init();
1978 
1979 	/* Setup UDP memory threshold */
1980 	udp_init();
1981 
1982 	/* Add UDP-Lite (RFC 3828) */
1983 	udplite4_register();
1984 
1985 	raw_init();
1986 
1987 	ping_init();
1988 
1989 	/*
1990 	 *	Set the ICMP layer up
1991 	 */
1992 
1993 	if (icmp_init() < 0)
1994 		panic("Failed to create the ICMP control socket.\n");
1995 
1996 	/*
1997 	 *	Initialise the multicast router
1998 	 */
1999 #if defined(CONFIG_IP_MROUTE)
2000 	if (ip_mr_init())
2001 		pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2002 #endif
2003 
2004 	if (init_inet_pernet_ops())
2005 		pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2006 
2007 	ipv4_proc_init();
2008 
2009 	ipfrag_init();
2010 
2011 	dev_add_pack(&ip_packet_type);
2012 
2013 	ip_tunnel_core_init();
2014 
2015 	rc = 0;
2016 out:
2017 	return rc;
2018 out_unregister_raw_proto:
2019 	proto_unregister(&raw_prot);
2020 out_unregister_udp_proto:
2021 	proto_unregister(&udp_prot);
2022 out_unregister_tcp_proto:
2023 	proto_unregister(&tcp_prot);
2024 	goto out;
2025 }
2026 
2027 fs_initcall(inet_init);
2028 
2029 /* ------------------------------------------------------------------------ */
2030 
2031 #ifdef CONFIG_PROC_FS
2032 static int __init ipv4_proc_init(void)
2033 {
2034 	int rc = 0;
2035 
2036 	if (raw_proc_init())
2037 		goto out_raw;
2038 	if (tcp4_proc_init())
2039 		goto out_tcp;
2040 	if (udp4_proc_init())
2041 		goto out_udp;
2042 	if (ping_proc_init())
2043 		goto out_ping;
2044 	if (ip_misc_proc_init())
2045 		goto out_misc;
2046 out:
2047 	return rc;
2048 out_misc:
2049 	ping_proc_exit();
2050 out_ping:
2051 	udp4_proc_exit();
2052 out_udp:
2053 	tcp4_proc_exit();
2054 out_tcp:
2055 	raw_proc_exit();
2056 out_raw:
2057 	rc = -ENOMEM;
2058 	goto out;
2059 }
2060 
2061 #else /* CONFIG_PROC_FS */
2062 static int __init ipv4_proc_init(void)
2063 {
2064 	return 0;
2065 }
2066 #endif /* CONFIG_PROC_FS */
2067