xref: /linux/net/ipv4/af_inet.c (revision 65aa371ea52a92dd10826a2ea74bd2c395ee90a8)
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/tcp.h>
103 #include <net/udp.h>
104 #include <net/udplite.h>
105 #include <net/ping.h>
106 #include <linux/skbuff.h>
107 #include <net/sock.h>
108 #include <net/raw.h>
109 #include <net/icmp.h>
110 #include <net/inet_common.h>
111 #include <net/ip_tunnels.h>
112 #include <net/xfrm.h>
113 #include <net/net_namespace.h>
114 #include <net/secure_seq.h>
115 #ifdef CONFIG_IP_MROUTE
116 #include <linux/mroute.h>
117 #endif
118 #include <net/l3mdev.h>
119 #include <net/compat.h>
120 
121 #include <trace/events/sock.h>
122 
123 /* The inetsw table contains everything that inet_create needs to
124  * build a new socket.
125  */
126 static struct list_head inetsw[SOCK_MAX];
127 static DEFINE_SPINLOCK(inetsw_lock);
128 
129 /* New destruction routine */
130 
131 void inet_sock_destruct(struct sock *sk)
132 {
133 	struct inet_sock *inet = inet_sk(sk);
134 
135 	__skb_queue_purge(&sk->sk_receive_queue);
136 	__skb_queue_purge(&sk->sk_error_queue);
137 
138 	sk_mem_reclaim_final(sk);
139 
140 	if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
141 		pr_err("Attempt to release TCP socket in state %d %p\n",
142 		       sk->sk_state, sk);
143 		return;
144 	}
145 	if (!sock_flag(sk, SOCK_DEAD)) {
146 		pr_err("Attempt to release alive inet socket %p\n", sk);
147 		return;
148 	}
149 
150 	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
151 	WARN_ON(refcount_read(&sk->sk_wmem_alloc));
152 	WARN_ON(sk->sk_wmem_queued);
153 	WARN_ON(sk->sk_forward_alloc);
154 
155 	kfree(rcu_dereference_protected(inet->inet_opt, 1));
156 	dst_release(rcu_dereference_protected(sk->sk_dst_cache, 1));
157 	dst_release(sk->sk_rx_dst);
158 	sk_refcnt_debug_dec(sk);
159 }
160 EXPORT_SYMBOL(inet_sock_destruct);
161 
162 /*
163  *	The routines beyond this point handle the behaviour of an AF_INET
164  *	socket object. Mostly it punts to the subprotocols of IP to do
165  *	the work.
166  */
167 
168 /*
169  *	Automatically bind an unbound socket.
170  */
171 
172 static int inet_autobind(struct sock *sk)
173 {
174 	struct inet_sock *inet;
175 	/* We may need to bind the socket. */
176 	lock_sock(sk);
177 	inet = inet_sk(sk);
178 	if (!inet->inet_num) {
179 		if (sk->sk_prot->get_port(sk, 0)) {
180 			release_sock(sk);
181 			return -EAGAIN;
182 		}
183 		inet->inet_sport = htons(inet->inet_num);
184 	}
185 	release_sock(sk);
186 	return 0;
187 }
188 
189 /*
190  *	Move a socket into listening state.
191  */
192 int inet_listen(struct socket *sock, int backlog)
193 {
194 	struct sock *sk = sock->sk;
195 	unsigned char old_state;
196 	int err, tcp_fastopen;
197 
198 	lock_sock(sk);
199 
200 	err = -EINVAL;
201 	if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
202 		goto out;
203 
204 	old_state = sk->sk_state;
205 	if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
206 		goto out;
207 
208 	WRITE_ONCE(sk->sk_max_ack_backlog, backlog);
209 	/* Really, if the socket is already in listen state
210 	 * we can only allow the backlog to be adjusted.
211 	 */
212 	if (old_state != TCP_LISTEN) {
213 		/* Enable TFO w/o requiring TCP_FASTOPEN socket option.
214 		 * Note that only TCP sockets (SOCK_STREAM) will reach here.
215 		 * Also fastopen backlog may already been set via the option
216 		 * because the socket was in TCP_LISTEN state previously but
217 		 * was shutdown() rather than close().
218 		 */
219 		tcp_fastopen = sock_net(sk)->ipv4.sysctl_tcp_fastopen;
220 		if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
221 		    (tcp_fastopen & TFO_SERVER_ENABLE) &&
222 		    !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
223 			fastopen_queue_tune(sk, backlog);
224 			tcp_fastopen_init_key_once(sock_net(sk));
225 		}
226 
227 		err = inet_csk_listen_start(sk, backlog);
228 		if (err)
229 			goto out;
230 		tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL);
231 	}
232 	err = 0;
233 
234 out:
235 	release_sock(sk);
236 	return err;
237 }
238 EXPORT_SYMBOL(inet_listen);
239 
240 /*
241  *	Create an inet socket.
242  */
243 
244 static int inet_create(struct net *net, struct socket *sock, int protocol,
245 		       int kern)
246 {
247 	struct sock *sk;
248 	struct inet_protosw *answer;
249 	struct inet_sock *inet;
250 	struct proto *answer_prot;
251 	unsigned char answer_flags;
252 	int try_loading_module = 0;
253 	int err;
254 
255 	if (protocol < 0 || protocol >= IPPROTO_MAX)
256 		return -EINVAL;
257 
258 	sock->state = SS_UNCONNECTED;
259 
260 	/* Look for the requested type/protocol pair. */
261 lookup_protocol:
262 	err = -ESOCKTNOSUPPORT;
263 	rcu_read_lock();
264 	list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
265 
266 		err = 0;
267 		/* Check the non-wild match. */
268 		if (protocol == answer->protocol) {
269 			if (protocol != IPPROTO_IP)
270 				break;
271 		} else {
272 			/* Check for the two wild cases. */
273 			if (IPPROTO_IP == protocol) {
274 				protocol = answer->protocol;
275 				break;
276 			}
277 			if (IPPROTO_IP == answer->protocol)
278 				break;
279 		}
280 		err = -EPROTONOSUPPORT;
281 	}
282 
283 	if (unlikely(err)) {
284 		if (try_loading_module < 2) {
285 			rcu_read_unlock();
286 			/*
287 			 * Be more specific, e.g. net-pf-2-proto-132-type-1
288 			 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
289 			 */
290 			if (++try_loading_module == 1)
291 				request_module("net-pf-%d-proto-%d-type-%d",
292 					       PF_INET, protocol, sock->type);
293 			/*
294 			 * Fall back to generic, e.g. net-pf-2-proto-132
295 			 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
296 			 */
297 			else
298 				request_module("net-pf-%d-proto-%d",
299 					       PF_INET, protocol);
300 			goto lookup_protocol;
301 		} else
302 			goto out_rcu_unlock;
303 	}
304 
305 	err = -EPERM;
306 	if (sock->type == SOCK_RAW && !kern &&
307 	    !ns_capable(net->user_ns, CAP_NET_RAW))
308 		goto out_rcu_unlock;
309 
310 	sock->ops = answer->ops;
311 	answer_prot = answer->prot;
312 	answer_flags = answer->flags;
313 	rcu_read_unlock();
314 
315 	WARN_ON(!answer_prot->slab);
316 
317 	err = -ENOMEM;
318 	sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
319 	if (!sk)
320 		goto out;
321 
322 	err = 0;
323 	if (INET_PROTOSW_REUSE & answer_flags)
324 		sk->sk_reuse = SK_CAN_REUSE;
325 
326 	inet = inet_sk(sk);
327 	inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
328 
329 	inet->nodefrag = 0;
330 
331 	if (SOCK_RAW == sock->type) {
332 		inet->inet_num = protocol;
333 		if (IPPROTO_RAW == protocol)
334 			inet->hdrincl = 1;
335 	}
336 
337 	if (net->ipv4.sysctl_ip_no_pmtu_disc)
338 		inet->pmtudisc = IP_PMTUDISC_DONT;
339 	else
340 		inet->pmtudisc = IP_PMTUDISC_WANT;
341 
342 	inet->inet_id = 0;
343 
344 	sock_init_data(sock, sk);
345 
346 	sk->sk_destruct	   = inet_sock_destruct;
347 	sk->sk_protocol	   = protocol;
348 	sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
349 
350 	inet->uc_ttl	= -1;
351 	inet->mc_loop	= 1;
352 	inet->mc_ttl	= 1;
353 	inet->mc_all	= 1;
354 	inet->mc_index	= 0;
355 	inet->mc_list	= NULL;
356 	inet->rcv_tos	= 0;
357 
358 	sk_refcnt_debug_inc(sk);
359 
360 	if (inet->inet_num) {
361 		/* It assumes that any protocol which allows
362 		 * the user to assign a number at socket
363 		 * creation time automatically
364 		 * shares.
365 		 */
366 		inet->inet_sport = htons(inet->inet_num);
367 		/* Add to protocol hash chains. */
368 		err = sk->sk_prot->hash(sk);
369 		if (err) {
370 			sk_common_release(sk);
371 			goto out;
372 		}
373 	}
374 
375 	if (sk->sk_prot->init) {
376 		err = sk->sk_prot->init(sk);
377 		if (err) {
378 			sk_common_release(sk);
379 			goto out;
380 		}
381 	}
382 
383 	if (!kern) {
384 		err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
385 		if (err) {
386 			sk_common_release(sk);
387 			goto out;
388 		}
389 	}
390 out:
391 	return err;
392 out_rcu_unlock:
393 	rcu_read_unlock();
394 	goto out;
395 }
396 
397 
398 /*
399  *	The peer socket should always be NULL (or else). When we call this
400  *	function we are destroying the object and from then on nobody
401  *	should refer to it.
402  */
403 int inet_release(struct socket *sock)
404 {
405 	struct sock *sk = sock->sk;
406 
407 	if (sk) {
408 		long timeout;
409 
410 		if (!sk->sk_kern_sock)
411 			BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk);
412 
413 		/* Applications forget to leave groups before exiting */
414 		ip_mc_drop_socket(sk);
415 
416 		/* If linger is set, we don't return until the close
417 		 * is complete.  Otherwise we return immediately. The
418 		 * actually closing is done the same either way.
419 		 *
420 		 * If the close is due to the process exiting, we never
421 		 * linger..
422 		 */
423 		timeout = 0;
424 		if (sock_flag(sk, SOCK_LINGER) &&
425 		    !(current->flags & PF_EXITING))
426 			timeout = sk->sk_lingertime;
427 		sk->sk_prot->close(sk, timeout);
428 		sock->sk = NULL;
429 	}
430 	return 0;
431 }
432 EXPORT_SYMBOL(inet_release);
433 
434 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
435 {
436 	struct sock *sk = sock->sk;
437 	u32 flags = BIND_WITH_LOCK;
438 	int err;
439 
440 	/* If the socket has its own bind function then use it. (RAW) */
441 	if (sk->sk_prot->bind) {
442 		return sk->sk_prot->bind(sk, uaddr, addr_len);
443 	}
444 	if (addr_len < sizeof(struct sockaddr_in))
445 		return -EINVAL;
446 
447 	/* BPF prog is run before any checks are done so that if the prog
448 	 * changes context in a wrong way it will be caught.
449 	 */
450 	err = BPF_CGROUP_RUN_PROG_INET_BIND_LOCK(sk, uaddr,
451 						 CGROUP_INET4_BIND, &flags);
452 	if (err)
453 		return err;
454 
455 	return __inet_bind(sk, uaddr, addr_len, flags);
456 }
457 EXPORT_SYMBOL(inet_bind);
458 
459 int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
460 		u32 flags)
461 {
462 	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
463 	struct inet_sock *inet = inet_sk(sk);
464 	struct net *net = sock_net(sk);
465 	unsigned short snum;
466 	int chk_addr_ret;
467 	u32 tb_id = RT_TABLE_LOCAL;
468 	int err;
469 
470 	if (addr->sin_family != AF_INET) {
471 		/* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
472 		 * only if s_addr is INADDR_ANY.
473 		 */
474 		err = -EAFNOSUPPORT;
475 		if (addr->sin_family != AF_UNSPEC ||
476 		    addr->sin_addr.s_addr != htonl(INADDR_ANY))
477 			goto out;
478 	}
479 
480 	tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
481 	chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
482 
483 	/* Not specified by any standard per-se, however it breaks too
484 	 * many applications when removed.  It is unfortunate since
485 	 * allowing applications to make a non-local bind solves
486 	 * several problems with systems using dynamic addressing.
487 	 * (ie. your servers still start up even if your ISDN link
488 	 *  is temporarily down)
489 	 */
490 	err = -EADDRNOTAVAIL;
491 	if (!inet_can_nonlocal_bind(net, inet) &&
492 	    addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
493 	    chk_addr_ret != RTN_LOCAL &&
494 	    chk_addr_ret != RTN_MULTICAST &&
495 	    chk_addr_ret != RTN_BROADCAST)
496 		goto out;
497 
498 	snum = ntohs(addr->sin_port);
499 	err = -EACCES;
500 	if (!(flags & BIND_NO_CAP_NET_BIND_SERVICE) &&
501 	    snum && inet_port_requires_bind_service(net, snum) &&
502 	    !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
503 		goto out;
504 
505 	/*      We keep a pair of addresses. rcv_saddr is the one
506 	 *      used by hash lookups, and saddr is used for transmit.
507 	 *
508 	 *      In the BSD API these are the same except where it
509 	 *      would be illegal to use them (multicast/broadcast) in
510 	 *      which case the sending device address is used.
511 	 */
512 	if (flags & BIND_WITH_LOCK)
513 		lock_sock(sk);
514 
515 	/* Check these errors (active socket, double bind). */
516 	err = -EINVAL;
517 	if (sk->sk_state != TCP_CLOSE || inet->inet_num)
518 		goto out_release_sock;
519 
520 	inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
521 	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
522 		inet->inet_saddr = 0;  /* Use device */
523 
524 	/* Make sure we are allowed to bind here. */
525 	if (snum || !(inet->bind_address_no_port ||
526 		      (flags & BIND_FORCE_ADDRESS_NO_PORT))) {
527 		if (sk->sk_prot->get_port(sk, snum)) {
528 			inet->inet_saddr = inet->inet_rcv_saddr = 0;
529 			err = -EADDRINUSE;
530 			goto out_release_sock;
531 		}
532 		if (!(flags & BIND_FROM_BPF)) {
533 			err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
534 			if (err) {
535 				inet->inet_saddr = inet->inet_rcv_saddr = 0;
536 				goto out_release_sock;
537 			}
538 		}
539 	}
540 
541 	if (inet->inet_rcv_saddr)
542 		sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
543 	if (snum)
544 		sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
545 	inet->inet_sport = htons(inet->inet_num);
546 	inet->inet_daddr = 0;
547 	inet->inet_dport = 0;
548 	sk_dst_reset(sk);
549 	err = 0;
550 out_release_sock:
551 	if (flags & BIND_WITH_LOCK)
552 		release_sock(sk);
553 out:
554 	return err;
555 }
556 
557 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
558 		       int addr_len, int flags)
559 {
560 	struct sock *sk = sock->sk;
561 	int err;
562 
563 	if (addr_len < sizeof(uaddr->sa_family))
564 		return -EINVAL;
565 	if (uaddr->sa_family == AF_UNSPEC)
566 		return sk->sk_prot->disconnect(sk, flags);
567 
568 	if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
569 		err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
570 		if (err)
571 			return err;
572 	}
573 
574 	if (data_race(!inet_sk(sk)->inet_num) && inet_autobind(sk))
575 		return -EAGAIN;
576 	return sk->sk_prot->connect(sk, uaddr, addr_len);
577 }
578 EXPORT_SYMBOL(inet_dgram_connect);
579 
580 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
581 {
582 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
583 
584 	add_wait_queue(sk_sleep(sk), &wait);
585 	sk->sk_write_pending += writebias;
586 
587 	/* Basic assumption: if someone sets sk->sk_err, he _must_
588 	 * change state of the socket from TCP_SYN_*.
589 	 * Connect() does not allow to get error notifications
590 	 * without closing the socket.
591 	 */
592 	while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
593 		release_sock(sk);
594 		timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
595 		lock_sock(sk);
596 		if (signal_pending(current) || !timeo)
597 			break;
598 	}
599 	remove_wait_queue(sk_sleep(sk), &wait);
600 	sk->sk_write_pending -= writebias;
601 	return timeo;
602 }
603 
604 /*
605  *	Connect to a remote host. There is regrettably still a little
606  *	TCP 'magic' in here.
607  */
608 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
609 			  int addr_len, int flags, int is_sendmsg)
610 {
611 	struct sock *sk = sock->sk;
612 	int err;
613 	long timeo;
614 
615 	/*
616 	 * uaddr can be NULL and addr_len can be 0 if:
617 	 * sk is a TCP fastopen active socket and
618 	 * TCP_FASTOPEN_CONNECT sockopt is set and
619 	 * we already have a valid cookie for this socket.
620 	 * In this case, user can call write() after connect().
621 	 * write() will invoke tcp_sendmsg_fastopen() which calls
622 	 * __inet_stream_connect().
623 	 */
624 	if (uaddr) {
625 		if (addr_len < sizeof(uaddr->sa_family))
626 			return -EINVAL;
627 
628 		if (uaddr->sa_family == AF_UNSPEC) {
629 			err = sk->sk_prot->disconnect(sk, flags);
630 			sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
631 			goto out;
632 		}
633 	}
634 
635 	switch (sock->state) {
636 	default:
637 		err = -EINVAL;
638 		goto out;
639 	case SS_CONNECTED:
640 		err = -EISCONN;
641 		goto out;
642 	case SS_CONNECTING:
643 		if (inet_sk(sk)->defer_connect)
644 			err = is_sendmsg ? -EINPROGRESS : -EISCONN;
645 		else
646 			err = -EALREADY;
647 		/* Fall out of switch with err, set for this state */
648 		break;
649 	case SS_UNCONNECTED:
650 		err = -EISCONN;
651 		if (sk->sk_state != TCP_CLOSE)
652 			goto out;
653 
654 		if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
655 			err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
656 			if (err)
657 				goto out;
658 		}
659 
660 		err = sk->sk_prot->connect(sk, uaddr, addr_len);
661 		if (err < 0)
662 			goto out;
663 
664 		sock->state = SS_CONNECTING;
665 
666 		if (!err && inet_sk(sk)->defer_connect)
667 			goto out;
668 
669 		/* Just entered SS_CONNECTING state; the only
670 		 * difference is that return value in non-blocking
671 		 * case is EINPROGRESS, rather than EALREADY.
672 		 */
673 		err = -EINPROGRESS;
674 		break;
675 	}
676 
677 	timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
678 
679 	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
680 		int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
681 				tcp_sk(sk)->fastopen_req &&
682 				tcp_sk(sk)->fastopen_req->data ? 1 : 0;
683 
684 		/* Error code is set above */
685 		if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
686 			goto out;
687 
688 		err = sock_intr_errno(timeo);
689 		if (signal_pending(current))
690 			goto out;
691 	}
692 
693 	/* Connection was closed by RST, timeout, ICMP error
694 	 * or another process disconnected us.
695 	 */
696 	if (sk->sk_state == TCP_CLOSE)
697 		goto sock_error;
698 
699 	/* sk->sk_err may be not zero now, if RECVERR was ordered by user
700 	 * and error was received after socket entered established state.
701 	 * Hence, it is handled normally after connect() return successfully.
702 	 */
703 
704 	sock->state = SS_CONNECTED;
705 	err = 0;
706 out:
707 	return err;
708 
709 sock_error:
710 	err = sock_error(sk) ? : -ECONNABORTED;
711 	sock->state = SS_UNCONNECTED;
712 	if (sk->sk_prot->disconnect(sk, flags))
713 		sock->state = SS_DISCONNECTING;
714 	goto out;
715 }
716 EXPORT_SYMBOL(__inet_stream_connect);
717 
718 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
719 			int addr_len, int flags)
720 {
721 	int err;
722 
723 	lock_sock(sock->sk);
724 	err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
725 	release_sock(sock->sk);
726 	return err;
727 }
728 EXPORT_SYMBOL(inet_stream_connect);
729 
730 /*
731  *	Accept a pending connection. The TCP layer now gives BSD semantics.
732  */
733 
734 int inet_accept(struct socket *sock, struct socket *newsock, int flags,
735 		bool kern)
736 {
737 	struct sock *sk1 = sock->sk;
738 	int err = -EINVAL;
739 	struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err, kern);
740 
741 	if (!sk2)
742 		goto do_err;
743 
744 	lock_sock(sk2);
745 
746 	sock_rps_record_flow(sk2);
747 	WARN_ON(!((1 << sk2->sk_state) &
748 		  (TCPF_ESTABLISHED | TCPF_SYN_RECV |
749 		  TCPF_CLOSE_WAIT | TCPF_CLOSE)));
750 
751 	sock_graft(sk2, newsock);
752 
753 	newsock->state = SS_CONNECTED;
754 	err = 0;
755 	release_sock(sk2);
756 do_err:
757 	return err;
758 }
759 EXPORT_SYMBOL(inet_accept);
760 
761 /*
762  *	This does both peername and sockname.
763  */
764 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
765 		 int peer)
766 {
767 	struct sock *sk		= sock->sk;
768 	struct inet_sock *inet	= inet_sk(sk);
769 	DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
770 
771 	sin->sin_family = AF_INET;
772 	if (peer) {
773 		if (!inet->inet_dport ||
774 		    (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
775 		     peer == 1))
776 			return -ENOTCONN;
777 		sin->sin_port = inet->inet_dport;
778 		sin->sin_addr.s_addr = inet->inet_daddr;
779 		BPF_CGROUP_RUN_SA_PROG_LOCK(sk, (struct sockaddr *)sin,
780 					    CGROUP_INET4_GETPEERNAME,
781 					    NULL);
782 	} else {
783 		__be32 addr = inet->inet_rcv_saddr;
784 		if (!addr)
785 			addr = inet->inet_saddr;
786 		sin->sin_port = inet->inet_sport;
787 		sin->sin_addr.s_addr = addr;
788 		BPF_CGROUP_RUN_SA_PROG_LOCK(sk, (struct sockaddr *)sin,
789 					    CGROUP_INET4_GETSOCKNAME,
790 					    NULL);
791 	}
792 	memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
793 	return sizeof(*sin);
794 }
795 EXPORT_SYMBOL(inet_getname);
796 
797 int inet_send_prepare(struct sock *sk)
798 {
799 	sock_rps_record_flow(sk);
800 
801 	/* We may need to bind the socket. */
802 	if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind &&
803 	    inet_autobind(sk))
804 		return -EAGAIN;
805 
806 	return 0;
807 }
808 EXPORT_SYMBOL_GPL(inet_send_prepare);
809 
810 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
811 {
812 	struct sock *sk = sock->sk;
813 
814 	if (unlikely(inet_send_prepare(sk)))
815 		return -EAGAIN;
816 
817 	return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
818 			       sk, msg, size);
819 }
820 EXPORT_SYMBOL(inet_sendmsg);
821 
822 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
823 		      size_t size, int flags)
824 {
825 	struct sock *sk = sock->sk;
826 
827 	if (unlikely(inet_send_prepare(sk)))
828 		return -EAGAIN;
829 
830 	if (sk->sk_prot->sendpage)
831 		return sk->sk_prot->sendpage(sk, page, offset, size, flags);
832 	return sock_no_sendpage(sock, page, offset, size, flags);
833 }
834 EXPORT_SYMBOL(inet_sendpage);
835 
836 INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
837 					  size_t, int, int, int *));
838 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
839 		 int flags)
840 {
841 	struct sock *sk = sock->sk;
842 	int addr_len = 0;
843 	int err;
844 
845 	if (likely(!(flags & MSG_ERRQUEUE)))
846 		sock_rps_record_flow(sk);
847 
848 	err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
849 			      sk, msg, size, flags & MSG_DONTWAIT,
850 			      flags & ~MSG_DONTWAIT, &addr_len);
851 	if (err >= 0)
852 		msg->msg_namelen = addr_len;
853 	return err;
854 }
855 EXPORT_SYMBOL(inet_recvmsg);
856 
857 int inet_shutdown(struct socket *sock, int how)
858 {
859 	struct sock *sk = sock->sk;
860 	int err = 0;
861 
862 	/* This should really check to make sure
863 	 * the socket is a TCP socket. (WHY AC...)
864 	 */
865 	how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
866 		       1->2 bit 2 snds.
867 		       2->3 */
868 	if ((how & ~SHUTDOWN_MASK) || !how)	/* MAXINT->0 */
869 		return -EINVAL;
870 
871 	lock_sock(sk);
872 	if (sock->state == SS_CONNECTING) {
873 		if ((1 << sk->sk_state) &
874 		    (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
875 			sock->state = SS_DISCONNECTING;
876 		else
877 			sock->state = SS_CONNECTED;
878 	}
879 
880 	switch (sk->sk_state) {
881 	case TCP_CLOSE:
882 		err = -ENOTCONN;
883 		/* Hack to wake up other listeners, who can poll for
884 		   EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
885 		fallthrough;
886 	default:
887 		sk->sk_shutdown |= how;
888 		if (sk->sk_prot->shutdown)
889 			sk->sk_prot->shutdown(sk, how);
890 		break;
891 
892 	/* Remaining two branches are temporary solution for missing
893 	 * close() in multithreaded environment. It is _not_ a good idea,
894 	 * but we have no choice until close() is repaired at VFS level.
895 	 */
896 	case TCP_LISTEN:
897 		if (!(how & RCV_SHUTDOWN))
898 			break;
899 		fallthrough;
900 	case TCP_SYN_SENT:
901 		err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
902 		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
903 		break;
904 	}
905 
906 	/* Wake up anyone sleeping in poll. */
907 	sk->sk_state_change(sk);
908 	release_sock(sk);
909 	return err;
910 }
911 EXPORT_SYMBOL(inet_shutdown);
912 
913 /*
914  *	ioctl() calls you can issue on an INET socket. Most of these are
915  *	device configuration and stuff and very rarely used. Some ioctls
916  *	pass on to the socket itself.
917  *
918  *	NOTE: I like the idea of a module for the config stuff. ie ifconfig
919  *	loads the devconfigure module does its configuring and unloads it.
920  *	There's a good 20K of config code hanging around the kernel.
921  */
922 
923 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
924 {
925 	struct sock *sk = sock->sk;
926 	int err = 0;
927 	struct net *net = sock_net(sk);
928 	void __user *p = (void __user *)arg;
929 	struct ifreq ifr;
930 	struct rtentry rt;
931 
932 	switch (cmd) {
933 	case SIOCADDRT:
934 	case SIOCDELRT:
935 		if (copy_from_user(&rt, p, sizeof(struct rtentry)))
936 			return -EFAULT;
937 		err = ip_rt_ioctl(net, cmd, &rt);
938 		break;
939 	case SIOCRTMSG:
940 		err = -EINVAL;
941 		break;
942 	case SIOCDARP:
943 	case SIOCGARP:
944 	case SIOCSARP:
945 		err = arp_ioctl(net, cmd, (void __user *)arg);
946 		break;
947 	case SIOCGIFADDR:
948 	case SIOCGIFBRDADDR:
949 	case SIOCGIFNETMASK:
950 	case SIOCGIFDSTADDR:
951 	case SIOCGIFPFLAGS:
952 		if (get_user_ifreq(&ifr, NULL, p))
953 			return -EFAULT;
954 		err = devinet_ioctl(net, cmd, &ifr);
955 		if (!err && put_user_ifreq(&ifr, p))
956 			err = -EFAULT;
957 		break;
958 
959 	case SIOCSIFADDR:
960 	case SIOCSIFBRDADDR:
961 	case SIOCSIFNETMASK:
962 	case SIOCSIFDSTADDR:
963 	case SIOCSIFPFLAGS:
964 	case SIOCSIFFLAGS:
965 		if (get_user_ifreq(&ifr, NULL, p))
966 			return -EFAULT;
967 		err = devinet_ioctl(net, cmd, &ifr);
968 		break;
969 	default:
970 		if (sk->sk_prot->ioctl)
971 			err = sk->sk_prot->ioctl(sk, cmd, arg);
972 		else
973 			err = -ENOIOCTLCMD;
974 		break;
975 	}
976 	return err;
977 }
978 EXPORT_SYMBOL(inet_ioctl);
979 
980 #ifdef CONFIG_COMPAT
981 static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd,
982 		struct compat_rtentry __user *ur)
983 {
984 	compat_uptr_t rtdev;
985 	struct rtentry rt;
986 
987 	if (copy_from_user(&rt.rt_dst, &ur->rt_dst,
988 			3 * sizeof(struct sockaddr)) ||
989 	    get_user(rt.rt_flags, &ur->rt_flags) ||
990 	    get_user(rt.rt_metric, &ur->rt_metric) ||
991 	    get_user(rt.rt_mtu, &ur->rt_mtu) ||
992 	    get_user(rt.rt_window, &ur->rt_window) ||
993 	    get_user(rt.rt_irtt, &ur->rt_irtt) ||
994 	    get_user(rtdev, &ur->rt_dev))
995 		return -EFAULT;
996 
997 	rt.rt_dev = compat_ptr(rtdev);
998 	return ip_rt_ioctl(sock_net(sk), cmd, &rt);
999 }
1000 
1001 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1002 {
1003 	void __user *argp = compat_ptr(arg);
1004 	struct sock *sk = sock->sk;
1005 
1006 	switch (cmd) {
1007 	case SIOCADDRT:
1008 	case SIOCDELRT:
1009 		return inet_compat_routing_ioctl(sk, cmd, argp);
1010 	default:
1011 		if (!sk->sk_prot->compat_ioctl)
1012 			return -ENOIOCTLCMD;
1013 		return sk->sk_prot->compat_ioctl(sk, cmd, arg);
1014 	}
1015 }
1016 #endif /* CONFIG_COMPAT */
1017 
1018 const struct proto_ops inet_stream_ops = {
1019 	.family		   = PF_INET,
1020 	.owner		   = THIS_MODULE,
1021 	.release	   = inet_release,
1022 	.bind		   = inet_bind,
1023 	.connect	   = inet_stream_connect,
1024 	.socketpair	   = sock_no_socketpair,
1025 	.accept		   = inet_accept,
1026 	.getname	   = inet_getname,
1027 	.poll		   = tcp_poll,
1028 	.ioctl		   = inet_ioctl,
1029 	.gettstamp	   = sock_gettstamp,
1030 	.listen		   = inet_listen,
1031 	.shutdown	   = inet_shutdown,
1032 	.setsockopt	   = sock_common_setsockopt,
1033 	.getsockopt	   = sock_common_getsockopt,
1034 	.sendmsg	   = inet_sendmsg,
1035 	.recvmsg	   = inet_recvmsg,
1036 #ifdef CONFIG_MMU
1037 	.mmap		   = tcp_mmap,
1038 #endif
1039 	.sendpage	   = inet_sendpage,
1040 	.splice_read	   = tcp_splice_read,
1041 	.read_sock	   = tcp_read_sock,
1042 	.sendmsg_locked    = tcp_sendmsg_locked,
1043 	.sendpage_locked   = tcp_sendpage_locked,
1044 	.peek_len	   = tcp_peek_len,
1045 #ifdef CONFIG_COMPAT
1046 	.compat_ioctl	   = inet_compat_ioctl,
1047 #endif
1048 	.set_rcvlowat	   = tcp_set_rcvlowat,
1049 };
1050 EXPORT_SYMBOL(inet_stream_ops);
1051 
1052 const struct proto_ops inet_dgram_ops = {
1053 	.family		   = PF_INET,
1054 	.owner		   = THIS_MODULE,
1055 	.release	   = inet_release,
1056 	.bind		   = inet_bind,
1057 	.connect	   = inet_dgram_connect,
1058 	.socketpair	   = sock_no_socketpair,
1059 	.accept		   = sock_no_accept,
1060 	.getname	   = inet_getname,
1061 	.poll		   = udp_poll,
1062 	.ioctl		   = inet_ioctl,
1063 	.gettstamp	   = sock_gettstamp,
1064 	.listen		   = sock_no_listen,
1065 	.shutdown	   = inet_shutdown,
1066 	.setsockopt	   = sock_common_setsockopt,
1067 	.getsockopt	   = sock_common_getsockopt,
1068 	.sendmsg	   = inet_sendmsg,
1069 	.read_sock	   = udp_read_sock,
1070 	.recvmsg	   = inet_recvmsg,
1071 	.mmap		   = sock_no_mmap,
1072 	.sendpage	   = inet_sendpage,
1073 	.set_peek_off	   = sk_set_peek_off,
1074 #ifdef CONFIG_COMPAT
1075 	.compat_ioctl	   = inet_compat_ioctl,
1076 #endif
1077 };
1078 EXPORT_SYMBOL(inet_dgram_ops);
1079 
1080 /*
1081  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1082  * udp_poll
1083  */
1084 static const struct proto_ops inet_sockraw_ops = {
1085 	.family		   = PF_INET,
1086 	.owner		   = THIS_MODULE,
1087 	.release	   = inet_release,
1088 	.bind		   = inet_bind,
1089 	.connect	   = inet_dgram_connect,
1090 	.socketpair	   = sock_no_socketpair,
1091 	.accept		   = sock_no_accept,
1092 	.getname	   = inet_getname,
1093 	.poll		   = datagram_poll,
1094 	.ioctl		   = inet_ioctl,
1095 	.gettstamp	   = sock_gettstamp,
1096 	.listen		   = sock_no_listen,
1097 	.shutdown	   = inet_shutdown,
1098 	.setsockopt	   = sock_common_setsockopt,
1099 	.getsockopt	   = sock_common_getsockopt,
1100 	.sendmsg	   = inet_sendmsg,
1101 	.recvmsg	   = inet_recvmsg,
1102 	.mmap		   = sock_no_mmap,
1103 	.sendpage	   = inet_sendpage,
1104 #ifdef CONFIG_COMPAT
1105 	.compat_ioctl	   = inet_compat_ioctl,
1106 #endif
1107 };
1108 
1109 static const struct net_proto_family inet_family_ops = {
1110 	.family = PF_INET,
1111 	.create = inet_create,
1112 	.owner	= THIS_MODULE,
1113 };
1114 
1115 /* Upon startup we insert all the elements in inetsw_array[] into
1116  * the linked list inetsw.
1117  */
1118 static struct inet_protosw inetsw_array[] =
1119 {
1120 	{
1121 		.type =       SOCK_STREAM,
1122 		.protocol =   IPPROTO_TCP,
1123 		.prot =       &tcp_prot,
1124 		.ops =        &inet_stream_ops,
1125 		.flags =      INET_PROTOSW_PERMANENT |
1126 			      INET_PROTOSW_ICSK,
1127 	},
1128 
1129 	{
1130 		.type =       SOCK_DGRAM,
1131 		.protocol =   IPPROTO_UDP,
1132 		.prot =       &udp_prot,
1133 		.ops =        &inet_dgram_ops,
1134 		.flags =      INET_PROTOSW_PERMANENT,
1135        },
1136 
1137        {
1138 		.type =       SOCK_DGRAM,
1139 		.protocol =   IPPROTO_ICMP,
1140 		.prot =       &ping_prot,
1141 		.ops =        &inet_sockraw_ops,
1142 		.flags =      INET_PROTOSW_REUSE,
1143        },
1144 
1145        {
1146 	       .type =       SOCK_RAW,
1147 	       .protocol =   IPPROTO_IP,	/* wild card */
1148 	       .prot =       &raw_prot,
1149 	       .ops =        &inet_sockraw_ops,
1150 	       .flags =      INET_PROTOSW_REUSE,
1151        }
1152 };
1153 
1154 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1155 
1156 void inet_register_protosw(struct inet_protosw *p)
1157 {
1158 	struct list_head *lh;
1159 	struct inet_protosw *answer;
1160 	int protocol = p->protocol;
1161 	struct list_head *last_perm;
1162 
1163 	spin_lock_bh(&inetsw_lock);
1164 
1165 	if (p->type >= SOCK_MAX)
1166 		goto out_illegal;
1167 
1168 	/* If we are trying to override a permanent protocol, bail. */
1169 	last_perm = &inetsw[p->type];
1170 	list_for_each(lh, &inetsw[p->type]) {
1171 		answer = list_entry(lh, struct inet_protosw, list);
1172 		/* Check only the non-wild match. */
1173 		if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1174 			break;
1175 		if (protocol == answer->protocol)
1176 			goto out_permanent;
1177 		last_perm = lh;
1178 	}
1179 
1180 	/* Add the new entry after the last permanent entry if any, so that
1181 	 * the new entry does not override a permanent entry when matched with
1182 	 * a wild-card protocol. But it is allowed to override any existing
1183 	 * non-permanent entry.  This means that when we remove this entry, the
1184 	 * system automatically returns to the old behavior.
1185 	 */
1186 	list_add_rcu(&p->list, last_perm);
1187 out:
1188 	spin_unlock_bh(&inetsw_lock);
1189 
1190 	return;
1191 
1192 out_permanent:
1193 	pr_err("Attempt to override permanent protocol %d\n", protocol);
1194 	goto out;
1195 
1196 out_illegal:
1197 	pr_err("Ignoring attempt to register invalid socket type %d\n",
1198 	       p->type);
1199 	goto out;
1200 }
1201 EXPORT_SYMBOL(inet_register_protosw);
1202 
1203 void inet_unregister_protosw(struct inet_protosw *p)
1204 {
1205 	if (INET_PROTOSW_PERMANENT & p->flags) {
1206 		pr_err("Attempt to unregister permanent protocol %d\n",
1207 		       p->protocol);
1208 	} else {
1209 		spin_lock_bh(&inetsw_lock);
1210 		list_del_rcu(&p->list);
1211 		spin_unlock_bh(&inetsw_lock);
1212 
1213 		synchronize_net();
1214 	}
1215 }
1216 EXPORT_SYMBOL(inet_unregister_protosw);
1217 
1218 static int inet_sk_reselect_saddr(struct sock *sk)
1219 {
1220 	struct inet_sock *inet = inet_sk(sk);
1221 	__be32 old_saddr = inet->inet_saddr;
1222 	__be32 daddr = inet->inet_daddr;
1223 	struct flowi4 *fl4;
1224 	struct rtable *rt;
1225 	__be32 new_saddr;
1226 	struct ip_options_rcu *inet_opt;
1227 
1228 	inet_opt = rcu_dereference_protected(inet->inet_opt,
1229 					     lockdep_sock_is_held(sk));
1230 	if (inet_opt && inet_opt->opt.srr)
1231 		daddr = inet_opt->opt.faddr;
1232 
1233 	/* Query new route. */
1234 	fl4 = &inet->cork.fl.u.ip4;
1235 	rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1236 			      sk->sk_bound_dev_if, sk->sk_protocol,
1237 			      inet->inet_sport, inet->inet_dport, sk);
1238 	if (IS_ERR(rt))
1239 		return PTR_ERR(rt);
1240 
1241 	sk_setup_caps(sk, &rt->dst);
1242 
1243 	new_saddr = fl4->saddr;
1244 
1245 	if (new_saddr == old_saddr)
1246 		return 0;
1247 
1248 	if (sock_net(sk)->ipv4.sysctl_ip_dynaddr > 1) {
1249 		pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1250 			__func__, &old_saddr, &new_saddr);
1251 	}
1252 
1253 	inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1254 
1255 	/*
1256 	 * XXX The only one ugly spot where we need to
1257 	 * XXX really change the sockets identity after
1258 	 * XXX it has entered the hashes. -DaveM
1259 	 *
1260 	 * Besides that, it does not check for connection
1261 	 * uniqueness. Wait for troubles.
1262 	 */
1263 	return __sk_prot_rehash(sk);
1264 }
1265 
1266 int inet_sk_rebuild_header(struct sock *sk)
1267 {
1268 	struct inet_sock *inet = inet_sk(sk);
1269 	struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1270 	__be32 daddr;
1271 	struct ip_options_rcu *inet_opt;
1272 	struct flowi4 *fl4;
1273 	int err;
1274 
1275 	/* Route is OK, nothing to do. */
1276 	if (rt)
1277 		return 0;
1278 
1279 	/* Reroute. */
1280 	rcu_read_lock();
1281 	inet_opt = rcu_dereference(inet->inet_opt);
1282 	daddr = inet->inet_daddr;
1283 	if (inet_opt && inet_opt->opt.srr)
1284 		daddr = inet_opt->opt.faddr;
1285 	rcu_read_unlock();
1286 	fl4 = &inet->cork.fl.u.ip4;
1287 	rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1288 				   inet->inet_dport, inet->inet_sport,
1289 				   sk->sk_protocol, RT_CONN_FLAGS(sk),
1290 				   sk->sk_bound_dev_if);
1291 	if (!IS_ERR(rt)) {
1292 		err = 0;
1293 		sk_setup_caps(sk, &rt->dst);
1294 	} else {
1295 		err = PTR_ERR(rt);
1296 
1297 		/* Routing failed... */
1298 		sk->sk_route_caps = 0;
1299 		/*
1300 		 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1301 		 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1302 		 */
1303 		if (!sock_net(sk)->ipv4.sysctl_ip_dynaddr ||
1304 		    sk->sk_state != TCP_SYN_SENT ||
1305 		    (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1306 		    (err = inet_sk_reselect_saddr(sk)) != 0)
1307 			sk->sk_err_soft = -err;
1308 	}
1309 
1310 	return err;
1311 }
1312 EXPORT_SYMBOL(inet_sk_rebuild_header);
1313 
1314 void inet_sk_set_state(struct sock *sk, int state)
1315 {
1316 	trace_inet_sock_set_state(sk, sk->sk_state, state);
1317 	sk->sk_state = state;
1318 }
1319 EXPORT_SYMBOL(inet_sk_set_state);
1320 
1321 void inet_sk_state_store(struct sock *sk, int newstate)
1322 {
1323 	trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1324 	smp_store_release(&sk->sk_state, newstate);
1325 }
1326 
1327 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1328 				 netdev_features_t features)
1329 {
1330 	bool udpfrag = false, fixedid = false, gso_partial, encap;
1331 	struct sk_buff *segs = ERR_PTR(-EINVAL);
1332 	const struct net_offload *ops;
1333 	unsigned int offset = 0;
1334 	struct iphdr *iph;
1335 	int proto, tot_len;
1336 	int nhoff;
1337 	int ihl;
1338 	int id;
1339 
1340 	skb_reset_network_header(skb);
1341 	nhoff = skb_network_header(skb) - skb_mac_header(skb);
1342 	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1343 		goto out;
1344 
1345 	iph = ip_hdr(skb);
1346 	ihl = iph->ihl * 4;
1347 	if (ihl < sizeof(*iph))
1348 		goto out;
1349 
1350 	id = ntohs(iph->id);
1351 	proto = iph->protocol;
1352 
1353 	/* Warning: after this point, iph might be no longer valid */
1354 	if (unlikely(!pskb_may_pull(skb, ihl)))
1355 		goto out;
1356 	__skb_pull(skb, ihl);
1357 
1358 	encap = SKB_GSO_CB(skb)->encap_level > 0;
1359 	if (encap)
1360 		features &= skb->dev->hw_enc_features;
1361 	SKB_GSO_CB(skb)->encap_level += ihl;
1362 
1363 	skb_reset_transport_header(skb);
1364 
1365 	segs = ERR_PTR(-EPROTONOSUPPORT);
1366 
1367 	if (!skb->encapsulation || encap) {
1368 		udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1369 		fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1370 
1371 		/* fixed ID is invalid if DF bit is not set */
1372 		if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1373 			goto out;
1374 	}
1375 
1376 	ops = rcu_dereference(inet_offloads[proto]);
1377 	if (likely(ops && ops->callbacks.gso_segment))
1378 		segs = ops->callbacks.gso_segment(skb, features);
1379 
1380 	if (IS_ERR_OR_NULL(segs))
1381 		goto out;
1382 
1383 	gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1384 
1385 	skb = segs;
1386 	do {
1387 		iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1388 		if (udpfrag) {
1389 			iph->frag_off = htons(offset >> 3);
1390 			if (skb->next)
1391 				iph->frag_off |= htons(IP_MF);
1392 			offset += skb->len - nhoff - ihl;
1393 			tot_len = skb->len - nhoff;
1394 		} else if (skb_is_gso(skb)) {
1395 			if (!fixedid) {
1396 				iph->id = htons(id);
1397 				id += skb_shinfo(skb)->gso_segs;
1398 			}
1399 
1400 			if (gso_partial)
1401 				tot_len = skb_shinfo(skb)->gso_size +
1402 					  SKB_GSO_CB(skb)->data_offset +
1403 					  skb->head - (unsigned char *)iph;
1404 			else
1405 				tot_len = skb->len - nhoff;
1406 		} else {
1407 			if (!fixedid)
1408 				iph->id = htons(id++);
1409 			tot_len = skb->len - nhoff;
1410 		}
1411 		iph->tot_len = htons(tot_len);
1412 		ip_send_check(iph);
1413 		if (encap)
1414 			skb_reset_inner_headers(skb);
1415 		skb->network_header = (u8 *)iph - skb->head;
1416 		skb_reset_mac_len(skb);
1417 	} while ((skb = skb->next));
1418 
1419 out:
1420 	return segs;
1421 }
1422 
1423 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1424 					netdev_features_t features)
1425 {
1426 	if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1427 		return ERR_PTR(-EINVAL);
1428 
1429 	return inet_gso_segment(skb, features);
1430 }
1431 
1432 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1433 {
1434 	const struct net_offload *ops;
1435 	struct sk_buff *pp = NULL;
1436 	const struct iphdr *iph;
1437 	struct sk_buff *p;
1438 	unsigned int hlen;
1439 	unsigned int off;
1440 	unsigned int id;
1441 	int flush = 1;
1442 	int proto;
1443 
1444 	off = skb_gro_offset(skb);
1445 	hlen = off + sizeof(*iph);
1446 	iph = skb_gro_header_fast(skb, off);
1447 	if (skb_gro_header_hard(skb, hlen)) {
1448 		iph = skb_gro_header_slow(skb, hlen, off);
1449 		if (unlikely(!iph))
1450 			goto out;
1451 	}
1452 
1453 	proto = iph->protocol;
1454 
1455 	rcu_read_lock();
1456 	ops = rcu_dereference(inet_offloads[proto]);
1457 	if (!ops || !ops->callbacks.gro_receive)
1458 		goto out_unlock;
1459 
1460 	if (*(u8 *)iph != 0x45)
1461 		goto out_unlock;
1462 
1463 	if (ip_is_fragment(iph))
1464 		goto out_unlock;
1465 
1466 	if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1467 		goto out_unlock;
1468 
1469 	id = ntohl(*(__be32 *)&iph->id);
1470 	flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1471 	id >>= 16;
1472 
1473 	list_for_each_entry(p, head, list) {
1474 		struct iphdr *iph2;
1475 		u16 flush_id;
1476 
1477 		if (!NAPI_GRO_CB(p)->same_flow)
1478 			continue;
1479 
1480 		iph2 = (struct iphdr *)(p->data + off);
1481 		/* The above works because, with the exception of the top
1482 		 * (inner most) layer, we only aggregate pkts with the same
1483 		 * hdr length so all the hdrs we'll need to verify will start
1484 		 * at the same offset.
1485 		 */
1486 		if ((iph->protocol ^ iph2->protocol) |
1487 		    ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1488 		    ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1489 			NAPI_GRO_CB(p)->same_flow = 0;
1490 			continue;
1491 		}
1492 
1493 		/* All fields must match except length and checksum. */
1494 		NAPI_GRO_CB(p)->flush |=
1495 			(iph->ttl ^ iph2->ttl) |
1496 			(iph->tos ^ iph2->tos) |
1497 			((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1498 
1499 		NAPI_GRO_CB(p)->flush |= flush;
1500 
1501 		/* We need to store of the IP ID check to be included later
1502 		 * when we can verify that this packet does in fact belong
1503 		 * to a given flow.
1504 		 */
1505 		flush_id = (u16)(id - ntohs(iph2->id));
1506 
1507 		/* This bit of code makes it much easier for us to identify
1508 		 * the cases where we are doing atomic vs non-atomic IP ID
1509 		 * checks.  Specifically an atomic check can return IP ID
1510 		 * values 0 - 0xFFFF, while a non-atomic check can only
1511 		 * return 0 or 0xFFFF.
1512 		 */
1513 		if (!NAPI_GRO_CB(p)->is_atomic ||
1514 		    !(iph->frag_off & htons(IP_DF))) {
1515 			flush_id ^= NAPI_GRO_CB(p)->count;
1516 			flush_id = flush_id ? 0xFFFF : 0;
1517 		}
1518 
1519 		/* If the previous IP ID value was based on an atomic
1520 		 * datagram we can overwrite the value and ignore it.
1521 		 */
1522 		if (NAPI_GRO_CB(skb)->is_atomic)
1523 			NAPI_GRO_CB(p)->flush_id = flush_id;
1524 		else
1525 			NAPI_GRO_CB(p)->flush_id |= flush_id;
1526 	}
1527 
1528 	NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1529 	NAPI_GRO_CB(skb)->flush |= flush;
1530 	skb_set_network_header(skb, off);
1531 	/* The above will be needed by the transport layer if there is one
1532 	 * immediately following this IP hdr.
1533 	 */
1534 
1535 	/* Note : No need to call skb_gro_postpull_rcsum() here,
1536 	 * as we already checked checksum over ipv4 header was 0
1537 	 */
1538 	skb_gro_pull(skb, sizeof(*iph));
1539 	skb_set_transport_header(skb, skb_gro_offset(skb));
1540 
1541 	pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1542 				       ops->callbacks.gro_receive, head, skb);
1543 
1544 out_unlock:
1545 	rcu_read_unlock();
1546 
1547 out:
1548 	skb_gro_flush_final(skb, pp, flush);
1549 
1550 	return pp;
1551 }
1552 
1553 static struct sk_buff *ipip_gro_receive(struct list_head *head,
1554 					struct sk_buff *skb)
1555 {
1556 	if (NAPI_GRO_CB(skb)->encap_mark) {
1557 		NAPI_GRO_CB(skb)->flush = 1;
1558 		return NULL;
1559 	}
1560 
1561 	NAPI_GRO_CB(skb)->encap_mark = 1;
1562 
1563 	return inet_gro_receive(head, skb);
1564 }
1565 
1566 #define SECONDS_PER_DAY	86400
1567 
1568 /* inet_current_timestamp - Return IP network timestamp
1569  *
1570  * Return milliseconds since midnight in network byte order.
1571  */
1572 __be32 inet_current_timestamp(void)
1573 {
1574 	u32 secs;
1575 	u32 msecs;
1576 	struct timespec64 ts;
1577 
1578 	ktime_get_real_ts64(&ts);
1579 
1580 	/* Get secs since midnight. */
1581 	(void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1582 	/* Convert to msecs. */
1583 	msecs = secs * MSEC_PER_SEC;
1584 	/* Convert nsec to msec. */
1585 	msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1586 
1587 	/* Convert to network byte order. */
1588 	return htonl(msecs);
1589 }
1590 EXPORT_SYMBOL(inet_current_timestamp);
1591 
1592 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1593 {
1594 	if (sk->sk_family == AF_INET)
1595 		return ip_recv_error(sk, msg, len, addr_len);
1596 #if IS_ENABLED(CONFIG_IPV6)
1597 	if (sk->sk_family == AF_INET6)
1598 		return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1599 #endif
1600 	return -EINVAL;
1601 }
1602 
1603 int inet_gro_complete(struct sk_buff *skb, int nhoff)
1604 {
1605 	__be16 newlen = htons(skb->len - nhoff);
1606 	struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1607 	const struct net_offload *ops;
1608 	int proto = iph->protocol;
1609 	int err = -ENOSYS;
1610 
1611 	if (skb->encapsulation) {
1612 		skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1613 		skb_set_inner_network_header(skb, nhoff);
1614 	}
1615 
1616 	csum_replace2(&iph->check, iph->tot_len, newlen);
1617 	iph->tot_len = newlen;
1618 
1619 	rcu_read_lock();
1620 	ops = rcu_dereference(inet_offloads[proto]);
1621 	if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1622 		goto out_unlock;
1623 
1624 	/* Only need to add sizeof(*iph) to get to the next hdr below
1625 	 * because any hdr with option will have been flushed in
1626 	 * inet_gro_receive().
1627 	 */
1628 	err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1629 			      tcp4_gro_complete, udp4_gro_complete,
1630 			      skb, nhoff + sizeof(*iph));
1631 
1632 out_unlock:
1633 	rcu_read_unlock();
1634 
1635 	return err;
1636 }
1637 
1638 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1639 {
1640 	skb->encapsulation = 1;
1641 	skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1642 	return inet_gro_complete(skb, nhoff);
1643 }
1644 
1645 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1646 			 unsigned short type, unsigned char protocol,
1647 			 struct net *net)
1648 {
1649 	struct socket *sock;
1650 	int rc = sock_create_kern(net, family, type, protocol, &sock);
1651 
1652 	if (rc == 0) {
1653 		*sk = sock->sk;
1654 		(*sk)->sk_allocation = GFP_ATOMIC;
1655 		/*
1656 		 * Unhash it so that IP input processing does not even see it,
1657 		 * we do not wish this socket to see incoming packets.
1658 		 */
1659 		(*sk)->sk_prot->unhash(*sk);
1660 	}
1661 	return rc;
1662 }
1663 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1664 
1665 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1666 {
1667 	unsigned long res = 0;
1668 	int i;
1669 
1670 	for_each_possible_cpu(i)
1671 		res += snmp_get_cpu_field(mib, i, offt);
1672 	return res;
1673 }
1674 EXPORT_SYMBOL_GPL(snmp_fold_field);
1675 
1676 #if BITS_PER_LONG==32
1677 
1678 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1679 			 size_t syncp_offset)
1680 {
1681 	void *bhptr;
1682 	struct u64_stats_sync *syncp;
1683 	u64 v;
1684 	unsigned int start;
1685 
1686 	bhptr = per_cpu_ptr(mib, cpu);
1687 	syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1688 	do {
1689 		start = u64_stats_fetch_begin_irq(syncp);
1690 		v = *(((u64 *)bhptr) + offt);
1691 	} while (u64_stats_fetch_retry_irq(syncp, start));
1692 
1693 	return v;
1694 }
1695 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1696 
1697 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1698 {
1699 	u64 res = 0;
1700 	int cpu;
1701 
1702 	for_each_possible_cpu(cpu) {
1703 		res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1704 	}
1705 	return res;
1706 }
1707 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1708 #endif
1709 
1710 #ifdef CONFIG_IP_MULTICAST
1711 static const struct net_protocol igmp_protocol = {
1712 	.handler =	igmp_rcv,
1713 };
1714 #endif
1715 
1716 /* thinking of making this const? Don't.
1717  * early_demux can change based on sysctl.
1718  */
1719 static struct net_protocol tcp_protocol = {
1720 	.early_demux	=	tcp_v4_early_demux,
1721 	.early_demux_handler =  tcp_v4_early_demux,
1722 	.handler	=	tcp_v4_rcv,
1723 	.err_handler	=	tcp_v4_err,
1724 	.no_policy	=	1,
1725 	.icmp_strict_tag_validation = 1,
1726 };
1727 
1728 /* thinking of making this const? Don't.
1729  * early_demux can change based on sysctl.
1730  */
1731 static struct net_protocol udp_protocol = {
1732 	.early_demux =	udp_v4_early_demux,
1733 	.early_demux_handler =	udp_v4_early_demux,
1734 	.handler =	udp_rcv,
1735 	.err_handler =	udp_err,
1736 	.no_policy =	1,
1737 };
1738 
1739 static const struct net_protocol icmp_protocol = {
1740 	.handler =	icmp_rcv,
1741 	.err_handler =	icmp_err,
1742 	.no_policy =	1,
1743 };
1744 
1745 static __net_init int ipv4_mib_init_net(struct net *net)
1746 {
1747 	int i;
1748 
1749 	net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1750 	if (!net->mib.tcp_statistics)
1751 		goto err_tcp_mib;
1752 	net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1753 	if (!net->mib.ip_statistics)
1754 		goto err_ip_mib;
1755 
1756 	for_each_possible_cpu(i) {
1757 		struct ipstats_mib *af_inet_stats;
1758 		af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1759 		u64_stats_init(&af_inet_stats->syncp);
1760 	}
1761 
1762 	net->mib.net_statistics = alloc_percpu(struct linux_mib);
1763 	if (!net->mib.net_statistics)
1764 		goto err_net_mib;
1765 	net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1766 	if (!net->mib.udp_statistics)
1767 		goto err_udp_mib;
1768 	net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1769 	if (!net->mib.udplite_statistics)
1770 		goto err_udplite_mib;
1771 	net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1772 	if (!net->mib.icmp_statistics)
1773 		goto err_icmp_mib;
1774 	net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1775 					      GFP_KERNEL);
1776 	if (!net->mib.icmpmsg_statistics)
1777 		goto err_icmpmsg_mib;
1778 
1779 	tcp_mib_init(net);
1780 	return 0;
1781 
1782 err_icmpmsg_mib:
1783 	free_percpu(net->mib.icmp_statistics);
1784 err_icmp_mib:
1785 	free_percpu(net->mib.udplite_statistics);
1786 err_udplite_mib:
1787 	free_percpu(net->mib.udp_statistics);
1788 err_udp_mib:
1789 	free_percpu(net->mib.net_statistics);
1790 err_net_mib:
1791 	free_percpu(net->mib.ip_statistics);
1792 err_ip_mib:
1793 	free_percpu(net->mib.tcp_statistics);
1794 err_tcp_mib:
1795 	return -ENOMEM;
1796 }
1797 
1798 static __net_exit void ipv4_mib_exit_net(struct net *net)
1799 {
1800 	kfree(net->mib.icmpmsg_statistics);
1801 	free_percpu(net->mib.icmp_statistics);
1802 	free_percpu(net->mib.udplite_statistics);
1803 	free_percpu(net->mib.udp_statistics);
1804 	free_percpu(net->mib.net_statistics);
1805 	free_percpu(net->mib.ip_statistics);
1806 	free_percpu(net->mib.tcp_statistics);
1807 #ifdef CONFIG_MPTCP
1808 	/* allocated on demand, see mptcp_init_sock() */
1809 	free_percpu(net->mib.mptcp_statistics);
1810 #endif
1811 }
1812 
1813 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1814 	.init = ipv4_mib_init_net,
1815 	.exit = ipv4_mib_exit_net,
1816 };
1817 
1818 static int __init init_ipv4_mibs(void)
1819 {
1820 	return register_pernet_subsys(&ipv4_mib_ops);
1821 }
1822 
1823 static __net_init int inet_init_net(struct net *net)
1824 {
1825 	/*
1826 	 * Set defaults for local port range
1827 	 */
1828 	seqlock_init(&net->ipv4.ip_local_ports.lock);
1829 	net->ipv4.ip_local_ports.range[0] =  32768;
1830 	net->ipv4.ip_local_ports.range[1] =  60999;
1831 
1832 	seqlock_init(&net->ipv4.ping_group_range.lock);
1833 	/*
1834 	 * Sane defaults - nobody may create ping sockets.
1835 	 * Boot scripts should set this to distro-specific group.
1836 	 */
1837 	net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1838 	net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1839 
1840 	/* Default values for sysctl-controlled parameters.
1841 	 * We set them here, in case sysctl is not compiled.
1842 	 */
1843 	net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1844 	net->ipv4.sysctl_ip_fwd_update_priority = 1;
1845 	net->ipv4.sysctl_ip_dynaddr = 0;
1846 	net->ipv4.sysctl_ip_early_demux = 1;
1847 	net->ipv4.sysctl_udp_early_demux = 1;
1848 	net->ipv4.sysctl_tcp_early_demux = 1;
1849 	net->ipv4.sysctl_nexthop_compat_mode = 1;
1850 #ifdef CONFIG_SYSCTL
1851 	net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1852 #endif
1853 
1854 	/* Some igmp sysctl, whose values are always used */
1855 	net->ipv4.sysctl_igmp_max_memberships = 20;
1856 	net->ipv4.sysctl_igmp_max_msf = 10;
1857 	/* IGMP reports for link-local multicast groups are enabled by default */
1858 	net->ipv4.sysctl_igmp_llm_reports = 1;
1859 	net->ipv4.sysctl_igmp_qrv = 2;
1860 
1861 	net->ipv4.sysctl_fib_notify_on_flag_change = 0;
1862 
1863 	return 0;
1864 }
1865 
1866 static __net_initdata struct pernet_operations af_inet_ops = {
1867 	.init = inet_init_net,
1868 };
1869 
1870 static int __init init_inet_pernet_ops(void)
1871 {
1872 	return register_pernet_subsys(&af_inet_ops);
1873 }
1874 
1875 static int ipv4_proc_init(void);
1876 
1877 /*
1878  *	IP protocol layer initialiser
1879  */
1880 
1881 static struct packet_offload ip_packet_offload __read_mostly = {
1882 	.type = cpu_to_be16(ETH_P_IP),
1883 	.callbacks = {
1884 		.gso_segment = inet_gso_segment,
1885 		.gro_receive = inet_gro_receive,
1886 		.gro_complete = inet_gro_complete,
1887 	},
1888 };
1889 
1890 static const struct net_offload ipip_offload = {
1891 	.callbacks = {
1892 		.gso_segment	= ipip_gso_segment,
1893 		.gro_receive	= ipip_gro_receive,
1894 		.gro_complete	= ipip_gro_complete,
1895 	},
1896 };
1897 
1898 static int __init ipip_offload_init(void)
1899 {
1900 	return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1901 }
1902 
1903 static int __init ipv4_offload_init(void)
1904 {
1905 	/*
1906 	 * Add offloads
1907 	 */
1908 	if (udpv4_offload_init() < 0)
1909 		pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1910 	if (tcpv4_offload_init() < 0)
1911 		pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1912 	if (ipip_offload_init() < 0)
1913 		pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1914 
1915 	dev_add_offload(&ip_packet_offload);
1916 	return 0;
1917 }
1918 
1919 fs_initcall(ipv4_offload_init);
1920 
1921 static struct packet_type ip_packet_type __read_mostly = {
1922 	.type = cpu_to_be16(ETH_P_IP),
1923 	.func = ip_rcv,
1924 	.list_func = ip_list_rcv,
1925 };
1926 
1927 static int __init inet_init(void)
1928 {
1929 	struct inet_protosw *q;
1930 	struct list_head *r;
1931 	int rc;
1932 
1933 	sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1934 
1935 	rc = proto_register(&tcp_prot, 1);
1936 	if (rc)
1937 		goto out;
1938 
1939 	rc = proto_register(&udp_prot, 1);
1940 	if (rc)
1941 		goto out_unregister_tcp_proto;
1942 
1943 	rc = proto_register(&raw_prot, 1);
1944 	if (rc)
1945 		goto out_unregister_udp_proto;
1946 
1947 	rc = proto_register(&ping_prot, 1);
1948 	if (rc)
1949 		goto out_unregister_raw_proto;
1950 
1951 	/*
1952 	 *	Tell SOCKET that we are alive...
1953 	 */
1954 
1955 	(void)sock_register(&inet_family_ops);
1956 
1957 #ifdef CONFIG_SYSCTL
1958 	ip_static_sysctl_init();
1959 #endif
1960 
1961 	/*
1962 	 *	Add all the base protocols.
1963 	 */
1964 
1965 	if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1966 		pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1967 	if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1968 		pr_crit("%s: Cannot add UDP protocol\n", __func__);
1969 	if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1970 		pr_crit("%s: Cannot add TCP protocol\n", __func__);
1971 #ifdef CONFIG_IP_MULTICAST
1972 	if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1973 		pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1974 #endif
1975 
1976 	/* Register the socket-side information for inet_create. */
1977 	for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1978 		INIT_LIST_HEAD(r);
1979 
1980 	for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1981 		inet_register_protosw(q);
1982 
1983 	/*
1984 	 *	Set the ARP module up
1985 	 */
1986 
1987 	arp_init();
1988 
1989 	/*
1990 	 *	Set the IP module up
1991 	 */
1992 
1993 	ip_init();
1994 
1995 	/* Setup TCP slab cache for open requests. */
1996 	tcp_init();
1997 
1998 	/* Setup UDP memory threshold */
1999 	udp_init();
2000 
2001 	/* Add UDP-Lite (RFC 3828) */
2002 	udplite4_register();
2003 
2004 	raw_init();
2005 
2006 	ping_init();
2007 
2008 	/*
2009 	 *	Set the ICMP layer up
2010 	 */
2011 
2012 	if (icmp_init() < 0)
2013 		panic("Failed to create the ICMP control socket.\n");
2014 
2015 	/*
2016 	 *	Initialise the multicast router
2017 	 */
2018 #if defined(CONFIG_IP_MROUTE)
2019 	if (ip_mr_init())
2020 		pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2021 #endif
2022 
2023 	if (init_inet_pernet_ops())
2024 		pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2025 	/*
2026 	 *	Initialise per-cpu ipv4 mibs
2027 	 */
2028 
2029 	if (init_ipv4_mibs())
2030 		pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
2031 
2032 	ipv4_proc_init();
2033 
2034 	ipfrag_init();
2035 
2036 	dev_add_pack(&ip_packet_type);
2037 
2038 	ip_tunnel_core_init();
2039 
2040 	rc = 0;
2041 out:
2042 	return rc;
2043 out_unregister_raw_proto:
2044 	proto_unregister(&raw_prot);
2045 out_unregister_udp_proto:
2046 	proto_unregister(&udp_prot);
2047 out_unregister_tcp_proto:
2048 	proto_unregister(&tcp_prot);
2049 	goto out;
2050 }
2051 
2052 fs_initcall(inet_init);
2053 
2054 /* ------------------------------------------------------------------------ */
2055 
2056 #ifdef CONFIG_PROC_FS
2057 static int __init ipv4_proc_init(void)
2058 {
2059 	int rc = 0;
2060 
2061 	if (raw_proc_init())
2062 		goto out_raw;
2063 	if (tcp4_proc_init())
2064 		goto out_tcp;
2065 	if (udp4_proc_init())
2066 		goto out_udp;
2067 	if (ping_proc_init())
2068 		goto out_ping;
2069 	if (ip_misc_proc_init())
2070 		goto out_misc;
2071 out:
2072 	return rc;
2073 out_misc:
2074 	ping_proc_exit();
2075 out_ping:
2076 	udp4_proc_exit();
2077 out_udp:
2078 	tcp4_proc_exit();
2079 out_tcp:
2080 	raw_proc_exit();
2081 out_raw:
2082 	rc = -ENOMEM;
2083 	goto out;
2084 }
2085 
2086 #else /* CONFIG_PROC_FS */
2087 static int __init ipv4_proc_init(void)
2088 {
2089 	return 0;
2090 }
2091 #endif /* CONFIG_PROC_FS */
2092