xref: /openbmc/linux/net/ipv4/af_inet.c (revision 1edd0337)
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/tcp.h>
104 #include <net/udp.h>
105 #include <net/udplite.h>
106 #include <net/ping.h>
107 #include <linux/skbuff.h>
108 #include <net/sock.h>
109 #include <net/raw.h>
110 #include <net/icmp.h>
111 #include <net/inet_common.h>
112 #include <net/ip_tunnels.h>
113 #include <net/xfrm.h>
114 #include <net/net_namespace.h>
115 #include <net/secure_seq.h>
116 #ifdef CONFIG_IP_MROUTE
117 #include <linux/mroute.h>
118 #endif
119 #include <net/l3mdev.h>
120 #include <net/compat.h>
121 
122 #include <trace/events/sock.h>
123 
124 /* The inetsw table contains everything that inet_create needs to
125  * build a new socket.
126  */
127 static struct list_head inetsw[SOCK_MAX];
128 static DEFINE_SPINLOCK(inetsw_lock);
129 
130 /* New destruction routine */
131 
132 void inet_sock_destruct(struct sock *sk)
133 {
134 	struct inet_sock *inet = inet_sk(sk);
135 
136 	__skb_queue_purge(&sk->sk_receive_queue);
137 	__skb_queue_purge(&sk->sk_error_queue);
138 
139 	sk_mem_reclaim_final(sk);
140 
141 	if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
142 		pr_err("Attempt to release TCP socket in state %d %p\n",
143 		       sk->sk_state, sk);
144 		return;
145 	}
146 	if (!sock_flag(sk, SOCK_DEAD)) {
147 		pr_err("Attempt to release alive inet socket %p\n", sk);
148 		return;
149 	}
150 
151 	WARN_ON_ONCE(atomic_read(&sk->sk_rmem_alloc));
152 	WARN_ON_ONCE(refcount_read(&sk->sk_wmem_alloc));
153 	WARN_ON_ONCE(sk->sk_wmem_queued);
154 	WARN_ON_ONCE(sk_forward_alloc_get(sk));
155 
156 	kfree(rcu_dereference_protected(inet->inet_opt, 1));
157 	dst_release(rcu_dereference_protected(sk->sk_dst_cache, 1));
158 	dst_release(rcu_dereference_protected(sk->sk_rx_dst, 1));
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 = READ_ONCE(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);
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 (READ_ONCE(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 	sk->sk_txrehash = READ_ONCE(net->core.sysctl_txrehash);
350 
351 	inet->uc_ttl	= -1;
352 	inet->mc_loop	= 1;
353 	inet->mc_ttl	= 1;
354 	inet->mc_all	= 1;
355 	inet->mc_index	= 0;
356 	inet->mc_list	= NULL;
357 	inet->rcv_tos	= 0;
358 
359 	if (inet->inet_num) {
360 		/* It assumes that any protocol which allows
361 		 * the user to assign a number at socket
362 		 * creation time automatically
363 		 * shares.
364 		 */
365 		inet->inet_sport = htons(inet->inet_num);
366 		/* Add to protocol hash chains. */
367 		err = sk->sk_prot->hash(sk);
368 		if (err) {
369 			sk_common_release(sk);
370 			goto out;
371 		}
372 	}
373 
374 	if (sk->sk_prot->init) {
375 		err = sk->sk_prot->init(sk);
376 		if (err) {
377 			sk_common_release(sk);
378 			goto out;
379 		}
380 	}
381 
382 	if (!kern) {
383 		err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
384 		if (err) {
385 			sk_common_release(sk);
386 			goto out;
387 		}
388 	}
389 out:
390 	return err;
391 out_rcu_unlock:
392 	rcu_read_unlock();
393 	goto out;
394 }
395 
396 
397 /*
398  *	The peer socket should always be NULL (or else). When we call this
399  *	function we are destroying the object and from then on nobody
400  *	should refer to it.
401  */
402 int inet_release(struct socket *sock)
403 {
404 	struct sock *sk = sock->sk;
405 
406 	if (sk) {
407 		long timeout;
408 
409 		if (!sk->sk_kern_sock)
410 			BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk);
411 
412 		/* Applications forget to leave groups before exiting */
413 		ip_mc_drop_socket(sk);
414 
415 		/* If linger is set, we don't return until the close
416 		 * is complete.  Otherwise we return immediately. The
417 		 * actually closing is done the same either way.
418 		 *
419 		 * If the close is due to the process exiting, we never
420 		 * linger..
421 		 */
422 		timeout = 0;
423 		if (sock_flag(sk, SOCK_LINGER) &&
424 		    !(current->flags & PF_EXITING))
425 			timeout = sk->sk_lingertime;
426 		sk->sk_prot->close(sk, timeout);
427 		sock->sk = NULL;
428 	}
429 	return 0;
430 }
431 EXPORT_SYMBOL(inet_release);
432 
433 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
434 {
435 	struct sock *sk = sock->sk;
436 	u32 flags = BIND_WITH_LOCK;
437 	int err;
438 
439 	/* If the socket has its own bind function then use it. (RAW) */
440 	if (sk->sk_prot->bind) {
441 		return sk->sk_prot->bind(sk, uaddr, addr_len);
442 	}
443 	if (addr_len < sizeof(struct sockaddr_in))
444 		return -EINVAL;
445 
446 	/* BPF prog is run before any checks are done so that if the prog
447 	 * changes context in a wrong way it will be caught.
448 	 */
449 	err = BPF_CGROUP_RUN_PROG_INET_BIND_LOCK(sk, uaddr,
450 						 CGROUP_INET4_BIND, &flags);
451 	if (err)
452 		return err;
453 
454 	return __inet_bind(sk, uaddr, addr_len, flags);
455 }
456 EXPORT_SYMBOL(inet_bind);
457 
458 int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
459 		u32 flags)
460 {
461 	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
462 	struct inet_sock *inet = inet_sk(sk);
463 	struct net *net = sock_net(sk);
464 	unsigned short snum;
465 	int chk_addr_ret;
466 	u32 tb_id = RT_TABLE_LOCAL;
467 	int err;
468 
469 	if (addr->sin_family != AF_INET) {
470 		/* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
471 		 * only if s_addr is INADDR_ANY.
472 		 */
473 		err = -EAFNOSUPPORT;
474 		if (addr->sin_family != AF_UNSPEC ||
475 		    addr->sin_addr.s_addr != htonl(INADDR_ANY))
476 			goto out;
477 	}
478 
479 	tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
480 	chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
481 
482 	/* Not specified by any standard per-se, however it breaks too
483 	 * many applications when removed.  It is unfortunate since
484 	 * allowing applications to make a non-local bind solves
485 	 * several problems with systems using dynamic addressing.
486 	 * (ie. your servers still start up even if your ISDN link
487 	 *  is temporarily down)
488 	 */
489 	err = -EADDRNOTAVAIL;
490 	if (!inet_addr_valid_or_nonlocal(net, inet, addr->sin_addr.s_addr,
491 	                                 chk_addr_ret))
492 		goto out;
493 
494 	snum = ntohs(addr->sin_port);
495 	err = -EACCES;
496 	if (!(flags & BIND_NO_CAP_NET_BIND_SERVICE) &&
497 	    snum && inet_port_requires_bind_service(net, snum) &&
498 	    !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
499 		goto out;
500 
501 	/*      We keep a pair of addresses. rcv_saddr is the one
502 	 *      used by hash lookups, and saddr is used for transmit.
503 	 *
504 	 *      In the BSD API these are the same except where it
505 	 *      would be illegal to use them (multicast/broadcast) in
506 	 *      which case the sending device address is used.
507 	 */
508 	if (flags & BIND_WITH_LOCK)
509 		lock_sock(sk);
510 
511 	/* Check these errors (active socket, double bind). */
512 	err = -EINVAL;
513 	if (sk->sk_state != TCP_CLOSE || inet->inet_num)
514 		goto out_release_sock;
515 
516 	inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
517 	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
518 		inet->inet_saddr = 0;  /* Use device */
519 
520 	/* Make sure we are allowed to bind here. */
521 	if (snum || !(inet->bind_address_no_port ||
522 		      (flags & BIND_FORCE_ADDRESS_NO_PORT))) {
523 		err = sk->sk_prot->get_port(sk, snum);
524 		if (err) {
525 			inet->inet_saddr = inet->inet_rcv_saddr = 0;
526 			goto out_release_sock;
527 		}
528 		if (!(flags & BIND_FROM_BPF)) {
529 			err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
530 			if (err) {
531 				inet->inet_saddr = inet->inet_rcv_saddr = 0;
532 				if (sk->sk_prot->put_port)
533 					sk->sk_prot->put_port(sk);
534 				goto out_release_sock;
535 			}
536 		}
537 	}
538 
539 	if (inet->inet_rcv_saddr)
540 		sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
541 	if (snum)
542 		sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
543 	inet->inet_sport = htons(inet->inet_num);
544 	inet->inet_daddr = 0;
545 	inet->inet_dport = 0;
546 	sk_dst_reset(sk);
547 	err = 0;
548 out_release_sock:
549 	if (flags & BIND_WITH_LOCK)
550 		release_sock(sk);
551 out:
552 	return err;
553 }
554 
555 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
556 		       int addr_len, int flags)
557 {
558 	struct sock *sk = sock->sk;
559 	const struct proto *prot;
560 	int err;
561 
562 	if (addr_len < sizeof(uaddr->sa_family))
563 		return -EINVAL;
564 
565 	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
566 	prot = READ_ONCE(sk->sk_prot);
567 
568 	if (uaddr->sa_family == AF_UNSPEC)
569 		return prot->disconnect(sk, flags);
570 
571 	if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
572 		err = prot->pre_connect(sk, uaddr, addr_len);
573 		if (err)
574 			return err;
575 	}
576 
577 	if (data_race(!inet_sk(sk)->inet_num) && inet_autobind(sk))
578 		return -EAGAIN;
579 	return prot->connect(sk, uaddr, addr_len);
580 }
581 EXPORT_SYMBOL(inet_dgram_connect);
582 
583 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
584 {
585 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
586 
587 	add_wait_queue(sk_sleep(sk), &wait);
588 	sk->sk_write_pending += writebias;
589 
590 	/* Basic assumption: if someone sets sk->sk_err, he _must_
591 	 * change state of the socket from TCP_SYN_*.
592 	 * Connect() does not allow to get error notifications
593 	 * without closing the socket.
594 	 */
595 	while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
596 		release_sock(sk);
597 		timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
598 		lock_sock(sk);
599 		if (signal_pending(current) || !timeo)
600 			break;
601 	}
602 	remove_wait_queue(sk_sleep(sk), &wait);
603 	sk->sk_write_pending -= writebias;
604 	return timeo;
605 }
606 
607 /*
608  *	Connect to a remote host. There is regrettably still a little
609  *	TCP 'magic' in here.
610  */
611 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
612 			  int addr_len, int flags, int is_sendmsg)
613 {
614 	struct sock *sk = sock->sk;
615 	int err;
616 	long timeo;
617 
618 	/*
619 	 * uaddr can be NULL and addr_len can be 0 if:
620 	 * sk is a TCP fastopen active socket and
621 	 * TCP_FASTOPEN_CONNECT sockopt is set and
622 	 * we already have a valid cookie for this socket.
623 	 * In this case, user can call write() after connect().
624 	 * write() will invoke tcp_sendmsg_fastopen() which calls
625 	 * __inet_stream_connect().
626 	 */
627 	if (uaddr) {
628 		if (addr_len < sizeof(uaddr->sa_family))
629 			return -EINVAL;
630 
631 		if (uaddr->sa_family == AF_UNSPEC) {
632 			err = sk->sk_prot->disconnect(sk, flags);
633 			sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
634 			goto out;
635 		}
636 	}
637 
638 	switch (sock->state) {
639 	default:
640 		err = -EINVAL;
641 		goto out;
642 	case SS_CONNECTED:
643 		err = -EISCONN;
644 		goto out;
645 	case SS_CONNECTING:
646 		if (inet_sk(sk)->defer_connect)
647 			err = is_sendmsg ? -EINPROGRESS : -EISCONN;
648 		else
649 			err = -EALREADY;
650 		/* Fall out of switch with err, set for this state */
651 		break;
652 	case SS_UNCONNECTED:
653 		err = -EISCONN;
654 		if (sk->sk_state != TCP_CLOSE)
655 			goto out;
656 
657 		if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
658 			err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
659 			if (err)
660 				goto out;
661 		}
662 
663 		err = sk->sk_prot->connect(sk, uaddr, addr_len);
664 		if (err < 0)
665 			goto out;
666 
667 		sock->state = SS_CONNECTING;
668 
669 		if (!err && inet_sk(sk)->defer_connect)
670 			goto out;
671 
672 		/* Just entered SS_CONNECTING state; the only
673 		 * difference is that return value in non-blocking
674 		 * case is EINPROGRESS, rather than EALREADY.
675 		 */
676 		err = -EINPROGRESS;
677 		break;
678 	}
679 
680 	timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
681 
682 	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
683 		int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
684 				tcp_sk(sk)->fastopen_req &&
685 				tcp_sk(sk)->fastopen_req->data ? 1 : 0;
686 
687 		/* Error code is set above */
688 		if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
689 			goto out;
690 
691 		err = sock_intr_errno(timeo);
692 		if (signal_pending(current))
693 			goto out;
694 	}
695 
696 	/* Connection was closed by RST, timeout, ICMP error
697 	 * or another process disconnected us.
698 	 */
699 	if (sk->sk_state == TCP_CLOSE)
700 		goto sock_error;
701 
702 	/* sk->sk_err may be not zero now, if RECVERR was ordered by user
703 	 * and error was received after socket entered established state.
704 	 * Hence, it is handled normally after connect() return successfully.
705 	 */
706 
707 	sock->state = SS_CONNECTED;
708 	err = 0;
709 out:
710 	return err;
711 
712 sock_error:
713 	err = sock_error(sk) ? : -ECONNABORTED;
714 	sock->state = SS_UNCONNECTED;
715 	if (sk->sk_prot->disconnect(sk, flags))
716 		sock->state = SS_DISCONNECTING;
717 	goto out;
718 }
719 EXPORT_SYMBOL(__inet_stream_connect);
720 
721 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
722 			int addr_len, int flags)
723 {
724 	int err;
725 
726 	lock_sock(sock->sk);
727 	err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
728 	release_sock(sock->sk);
729 	return err;
730 }
731 EXPORT_SYMBOL(inet_stream_connect);
732 
733 /*
734  *	Accept a pending connection. The TCP layer now gives BSD semantics.
735  */
736 
737 int inet_accept(struct socket *sock, struct socket *newsock, int flags,
738 		bool kern)
739 {
740 	struct sock *sk1 = sock->sk, *sk2;
741 	int err = -EINVAL;
742 
743 	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
744 	sk2 = READ_ONCE(sk1->sk_prot)->accept(sk1, flags, &err, kern);
745 	if (!sk2)
746 		goto do_err;
747 
748 	lock_sock(sk2);
749 
750 	sock_rps_record_flow(sk2);
751 	WARN_ON(!((1 << sk2->sk_state) &
752 		  (TCPF_ESTABLISHED | TCPF_SYN_RECV |
753 		  TCPF_CLOSE_WAIT | TCPF_CLOSE)));
754 
755 	if (test_bit(SOCK_SUPPORT_ZC, &sock->flags))
756 		set_bit(SOCK_SUPPORT_ZC, &newsock->flags);
757 	sock_graft(sk2, newsock);
758 
759 	newsock->state = SS_CONNECTED;
760 	err = 0;
761 	release_sock(sk2);
762 do_err:
763 	return err;
764 }
765 EXPORT_SYMBOL(inet_accept);
766 
767 /*
768  *	This does both peername and sockname.
769  */
770 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
771 		 int peer)
772 {
773 	struct sock *sk		= sock->sk;
774 	struct inet_sock *inet	= inet_sk(sk);
775 	DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
776 
777 	sin->sin_family = AF_INET;
778 	lock_sock(sk);
779 	if (peer) {
780 		if (!inet->inet_dport ||
781 		    (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
782 		     peer == 1)) {
783 			release_sock(sk);
784 			return -ENOTCONN;
785 		}
786 		sin->sin_port = inet->inet_dport;
787 		sin->sin_addr.s_addr = inet->inet_daddr;
788 		BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin,
789 				       CGROUP_INET4_GETPEERNAME);
790 	} else {
791 		__be32 addr = inet->inet_rcv_saddr;
792 		if (!addr)
793 			addr = inet->inet_saddr;
794 		sin->sin_port = inet->inet_sport;
795 		sin->sin_addr.s_addr = addr;
796 		BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin,
797 				       CGROUP_INET4_GETSOCKNAME);
798 	}
799 	release_sock(sk);
800 	memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
801 	return sizeof(*sin);
802 }
803 EXPORT_SYMBOL(inet_getname);
804 
805 int inet_send_prepare(struct sock *sk)
806 {
807 	sock_rps_record_flow(sk);
808 
809 	/* We may need to bind the socket. */
810 	if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind &&
811 	    inet_autobind(sk))
812 		return -EAGAIN;
813 
814 	return 0;
815 }
816 EXPORT_SYMBOL_GPL(inet_send_prepare);
817 
818 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
819 {
820 	struct sock *sk = sock->sk;
821 
822 	if (unlikely(inet_send_prepare(sk)))
823 		return -EAGAIN;
824 
825 	return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
826 			       sk, msg, size);
827 }
828 EXPORT_SYMBOL(inet_sendmsg);
829 
830 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
831 		      size_t size, int flags)
832 {
833 	struct sock *sk = sock->sk;
834 	const struct proto *prot;
835 
836 	if (unlikely(inet_send_prepare(sk)))
837 		return -EAGAIN;
838 
839 	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
840 	prot = READ_ONCE(sk->sk_prot);
841 	if (prot->sendpage)
842 		return prot->sendpage(sk, page, offset, size, flags);
843 	return sock_no_sendpage(sock, page, offset, size, flags);
844 }
845 EXPORT_SYMBOL(inet_sendpage);
846 
847 INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
848 					  size_t, int, int *));
849 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
850 		 int flags)
851 {
852 	struct sock *sk = sock->sk;
853 	int addr_len = 0;
854 	int err;
855 
856 	if (likely(!(flags & MSG_ERRQUEUE)))
857 		sock_rps_record_flow(sk);
858 
859 	err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
860 			      sk, msg, size, flags, &addr_len);
861 	if (err >= 0)
862 		msg->msg_namelen = addr_len;
863 	return err;
864 }
865 EXPORT_SYMBOL(inet_recvmsg);
866 
867 int inet_shutdown(struct socket *sock, int how)
868 {
869 	struct sock *sk = sock->sk;
870 	int err = 0;
871 
872 	/* This should really check to make sure
873 	 * the socket is a TCP socket. (WHY AC...)
874 	 */
875 	how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
876 		       1->2 bit 2 snds.
877 		       2->3 */
878 	if ((how & ~SHUTDOWN_MASK) || !how)	/* MAXINT->0 */
879 		return -EINVAL;
880 
881 	lock_sock(sk);
882 	if (sock->state == SS_CONNECTING) {
883 		if ((1 << sk->sk_state) &
884 		    (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
885 			sock->state = SS_DISCONNECTING;
886 		else
887 			sock->state = SS_CONNECTED;
888 	}
889 
890 	switch (sk->sk_state) {
891 	case TCP_CLOSE:
892 		err = -ENOTCONN;
893 		/* Hack to wake up other listeners, who can poll for
894 		   EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
895 		fallthrough;
896 	default:
897 		sk->sk_shutdown |= how;
898 		if (sk->sk_prot->shutdown)
899 			sk->sk_prot->shutdown(sk, how);
900 		break;
901 
902 	/* Remaining two branches are temporary solution for missing
903 	 * close() in multithreaded environment. It is _not_ a good idea,
904 	 * but we have no choice until close() is repaired at VFS level.
905 	 */
906 	case TCP_LISTEN:
907 		if (!(how & RCV_SHUTDOWN))
908 			break;
909 		fallthrough;
910 	case TCP_SYN_SENT:
911 		err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
912 		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
913 		break;
914 	}
915 
916 	/* Wake up anyone sleeping in poll. */
917 	sk->sk_state_change(sk);
918 	release_sock(sk);
919 	return err;
920 }
921 EXPORT_SYMBOL(inet_shutdown);
922 
923 /*
924  *	ioctl() calls you can issue on an INET socket. Most of these are
925  *	device configuration and stuff and very rarely used. Some ioctls
926  *	pass on to the socket itself.
927  *
928  *	NOTE: I like the idea of a module for the config stuff. ie ifconfig
929  *	loads the devconfigure module does its configuring and unloads it.
930  *	There's a good 20K of config code hanging around the kernel.
931  */
932 
933 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
934 {
935 	struct sock *sk = sock->sk;
936 	int err = 0;
937 	struct net *net = sock_net(sk);
938 	void __user *p = (void __user *)arg;
939 	struct ifreq ifr;
940 	struct rtentry rt;
941 
942 	switch (cmd) {
943 	case SIOCADDRT:
944 	case SIOCDELRT:
945 		if (copy_from_user(&rt, p, sizeof(struct rtentry)))
946 			return -EFAULT;
947 		err = ip_rt_ioctl(net, cmd, &rt);
948 		break;
949 	case SIOCRTMSG:
950 		err = -EINVAL;
951 		break;
952 	case SIOCDARP:
953 	case SIOCGARP:
954 	case SIOCSARP:
955 		err = arp_ioctl(net, cmd, (void __user *)arg);
956 		break;
957 	case SIOCGIFADDR:
958 	case SIOCGIFBRDADDR:
959 	case SIOCGIFNETMASK:
960 	case SIOCGIFDSTADDR:
961 	case SIOCGIFPFLAGS:
962 		if (get_user_ifreq(&ifr, NULL, p))
963 			return -EFAULT;
964 		err = devinet_ioctl(net, cmd, &ifr);
965 		if (!err && put_user_ifreq(&ifr, p))
966 			err = -EFAULT;
967 		break;
968 
969 	case SIOCSIFADDR:
970 	case SIOCSIFBRDADDR:
971 	case SIOCSIFNETMASK:
972 	case SIOCSIFDSTADDR:
973 	case SIOCSIFPFLAGS:
974 	case SIOCSIFFLAGS:
975 		if (get_user_ifreq(&ifr, NULL, p))
976 			return -EFAULT;
977 		err = devinet_ioctl(net, cmd, &ifr);
978 		break;
979 	default:
980 		if (sk->sk_prot->ioctl)
981 			err = sk->sk_prot->ioctl(sk, cmd, arg);
982 		else
983 			err = -ENOIOCTLCMD;
984 		break;
985 	}
986 	return err;
987 }
988 EXPORT_SYMBOL(inet_ioctl);
989 
990 #ifdef CONFIG_COMPAT
991 static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd,
992 		struct compat_rtentry __user *ur)
993 {
994 	compat_uptr_t rtdev;
995 	struct rtentry rt;
996 
997 	if (copy_from_user(&rt.rt_dst, &ur->rt_dst,
998 			3 * sizeof(struct sockaddr)) ||
999 	    get_user(rt.rt_flags, &ur->rt_flags) ||
1000 	    get_user(rt.rt_metric, &ur->rt_metric) ||
1001 	    get_user(rt.rt_mtu, &ur->rt_mtu) ||
1002 	    get_user(rt.rt_window, &ur->rt_window) ||
1003 	    get_user(rt.rt_irtt, &ur->rt_irtt) ||
1004 	    get_user(rtdev, &ur->rt_dev))
1005 		return -EFAULT;
1006 
1007 	rt.rt_dev = compat_ptr(rtdev);
1008 	return ip_rt_ioctl(sock_net(sk), cmd, &rt);
1009 }
1010 
1011 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1012 {
1013 	void __user *argp = compat_ptr(arg);
1014 	struct sock *sk = sock->sk;
1015 
1016 	switch (cmd) {
1017 	case SIOCADDRT:
1018 	case SIOCDELRT:
1019 		return inet_compat_routing_ioctl(sk, cmd, argp);
1020 	default:
1021 		if (!sk->sk_prot->compat_ioctl)
1022 			return -ENOIOCTLCMD;
1023 		return sk->sk_prot->compat_ioctl(sk, cmd, arg);
1024 	}
1025 }
1026 #endif /* CONFIG_COMPAT */
1027 
1028 const struct proto_ops inet_stream_ops = {
1029 	.family		   = PF_INET,
1030 	.owner		   = THIS_MODULE,
1031 	.release	   = inet_release,
1032 	.bind		   = inet_bind,
1033 	.connect	   = inet_stream_connect,
1034 	.socketpair	   = sock_no_socketpair,
1035 	.accept		   = inet_accept,
1036 	.getname	   = inet_getname,
1037 	.poll		   = tcp_poll,
1038 	.ioctl		   = inet_ioctl,
1039 	.gettstamp	   = sock_gettstamp,
1040 	.listen		   = inet_listen,
1041 	.shutdown	   = inet_shutdown,
1042 	.setsockopt	   = sock_common_setsockopt,
1043 	.getsockopt	   = sock_common_getsockopt,
1044 	.sendmsg	   = inet_sendmsg,
1045 	.recvmsg	   = inet_recvmsg,
1046 #ifdef CONFIG_MMU
1047 	.mmap		   = tcp_mmap,
1048 #endif
1049 	.sendpage	   = inet_sendpage,
1050 	.splice_read	   = tcp_splice_read,
1051 	.read_sock	   = tcp_read_sock,
1052 	.read_skb	   = tcp_read_skb,
1053 	.sendmsg_locked    = tcp_sendmsg_locked,
1054 	.sendpage_locked   = tcp_sendpage_locked,
1055 	.peek_len	   = tcp_peek_len,
1056 #ifdef CONFIG_COMPAT
1057 	.compat_ioctl	   = inet_compat_ioctl,
1058 #endif
1059 	.set_rcvlowat	   = tcp_set_rcvlowat,
1060 };
1061 EXPORT_SYMBOL(inet_stream_ops);
1062 
1063 const struct proto_ops inet_dgram_ops = {
1064 	.family		   = PF_INET,
1065 	.owner		   = THIS_MODULE,
1066 	.release	   = inet_release,
1067 	.bind		   = inet_bind,
1068 	.connect	   = inet_dgram_connect,
1069 	.socketpair	   = sock_no_socketpair,
1070 	.accept		   = sock_no_accept,
1071 	.getname	   = inet_getname,
1072 	.poll		   = udp_poll,
1073 	.ioctl		   = inet_ioctl,
1074 	.gettstamp	   = sock_gettstamp,
1075 	.listen		   = sock_no_listen,
1076 	.shutdown	   = inet_shutdown,
1077 	.setsockopt	   = sock_common_setsockopt,
1078 	.getsockopt	   = sock_common_getsockopt,
1079 	.sendmsg	   = inet_sendmsg,
1080 	.read_skb	   = udp_read_skb,
1081 	.recvmsg	   = inet_recvmsg,
1082 	.mmap		   = sock_no_mmap,
1083 	.sendpage	   = inet_sendpage,
1084 	.set_peek_off	   = sk_set_peek_off,
1085 #ifdef CONFIG_COMPAT
1086 	.compat_ioctl	   = inet_compat_ioctl,
1087 #endif
1088 };
1089 EXPORT_SYMBOL(inet_dgram_ops);
1090 
1091 /*
1092  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1093  * udp_poll
1094  */
1095 static const struct proto_ops inet_sockraw_ops = {
1096 	.family		   = PF_INET,
1097 	.owner		   = THIS_MODULE,
1098 	.release	   = inet_release,
1099 	.bind		   = inet_bind,
1100 	.connect	   = inet_dgram_connect,
1101 	.socketpair	   = sock_no_socketpair,
1102 	.accept		   = sock_no_accept,
1103 	.getname	   = inet_getname,
1104 	.poll		   = datagram_poll,
1105 	.ioctl		   = inet_ioctl,
1106 	.gettstamp	   = sock_gettstamp,
1107 	.listen		   = sock_no_listen,
1108 	.shutdown	   = inet_shutdown,
1109 	.setsockopt	   = sock_common_setsockopt,
1110 	.getsockopt	   = sock_common_getsockopt,
1111 	.sendmsg	   = inet_sendmsg,
1112 	.recvmsg	   = inet_recvmsg,
1113 	.mmap		   = sock_no_mmap,
1114 	.sendpage	   = inet_sendpage,
1115 #ifdef CONFIG_COMPAT
1116 	.compat_ioctl	   = inet_compat_ioctl,
1117 #endif
1118 };
1119 
1120 static const struct net_proto_family inet_family_ops = {
1121 	.family = PF_INET,
1122 	.create = inet_create,
1123 	.owner	= THIS_MODULE,
1124 };
1125 
1126 /* Upon startup we insert all the elements in inetsw_array[] into
1127  * the linked list inetsw.
1128  */
1129 static struct inet_protosw inetsw_array[] =
1130 {
1131 	{
1132 		.type =       SOCK_STREAM,
1133 		.protocol =   IPPROTO_TCP,
1134 		.prot =       &tcp_prot,
1135 		.ops =        &inet_stream_ops,
1136 		.flags =      INET_PROTOSW_PERMANENT |
1137 			      INET_PROTOSW_ICSK,
1138 	},
1139 
1140 	{
1141 		.type =       SOCK_DGRAM,
1142 		.protocol =   IPPROTO_UDP,
1143 		.prot =       &udp_prot,
1144 		.ops =        &inet_dgram_ops,
1145 		.flags =      INET_PROTOSW_PERMANENT,
1146        },
1147 
1148        {
1149 		.type =       SOCK_DGRAM,
1150 		.protocol =   IPPROTO_ICMP,
1151 		.prot =       &ping_prot,
1152 		.ops =        &inet_sockraw_ops,
1153 		.flags =      INET_PROTOSW_REUSE,
1154        },
1155 
1156        {
1157 	       .type =       SOCK_RAW,
1158 	       .protocol =   IPPROTO_IP,	/* wild card */
1159 	       .prot =       &raw_prot,
1160 	       .ops =        &inet_sockraw_ops,
1161 	       .flags =      INET_PROTOSW_REUSE,
1162        }
1163 };
1164 
1165 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1166 
1167 void inet_register_protosw(struct inet_protosw *p)
1168 {
1169 	struct list_head *lh;
1170 	struct inet_protosw *answer;
1171 	int protocol = p->protocol;
1172 	struct list_head *last_perm;
1173 
1174 	spin_lock_bh(&inetsw_lock);
1175 
1176 	if (p->type >= SOCK_MAX)
1177 		goto out_illegal;
1178 
1179 	/* If we are trying to override a permanent protocol, bail. */
1180 	last_perm = &inetsw[p->type];
1181 	list_for_each(lh, &inetsw[p->type]) {
1182 		answer = list_entry(lh, struct inet_protosw, list);
1183 		/* Check only the non-wild match. */
1184 		if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1185 			break;
1186 		if (protocol == answer->protocol)
1187 			goto out_permanent;
1188 		last_perm = lh;
1189 	}
1190 
1191 	/* Add the new entry after the last permanent entry if any, so that
1192 	 * the new entry does not override a permanent entry when matched with
1193 	 * a wild-card protocol. But it is allowed to override any existing
1194 	 * non-permanent entry.  This means that when we remove this entry, the
1195 	 * system automatically returns to the old behavior.
1196 	 */
1197 	list_add_rcu(&p->list, last_perm);
1198 out:
1199 	spin_unlock_bh(&inetsw_lock);
1200 
1201 	return;
1202 
1203 out_permanent:
1204 	pr_err("Attempt to override permanent protocol %d\n", protocol);
1205 	goto out;
1206 
1207 out_illegal:
1208 	pr_err("Ignoring attempt to register invalid socket type %d\n",
1209 	       p->type);
1210 	goto out;
1211 }
1212 EXPORT_SYMBOL(inet_register_protosw);
1213 
1214 void inet_unregister_protosw(struct inet_protosw *p)
1215 {
1216 	if (INET_PROTOSW_PERMANENT & p->flags) {
1217 		pr_err("Attempt to unregister permanent protocol %d\n",
1218 		       p->protocol);
1219 	} else {
1220 		spin_lock_bh(&inetsw_lock);
1221 		list_del_rcu(&p->list);
1222 		spin_unlock_bh(&inetsw_lock);
1223 
1224 		synchronize_net();
1225 	}
1226 }
1227 EXPORT_SYMBOL(inet_unregister_protosw);
1228 
1229 static int inet_sk_reselect_saddr(struct sock *sk)
1230 {
1231 	struct inet_sock *inet = inet_sk(sk);
1232 	__be32 old_saddr = inet->inet_saddr;
1233 	__be32 daddr = inet->inet_daddr;
1234 	struct flowi4 *fl4;
1235 	struct rtable *rt;
1236 	__be32 new_saddr;
1237 	struct ip_options_rcu *inet_opt;
1238 	int err;
1239 
1240 	inet_opt = rcu_dereference_protected(inet->inet_opt,
1241 					     lockdep_sock_is_held(sk));
1242 	if (inet_opt && inet_opt->opt.srr)
1243 		daddr = inet_opt->opt.faddr;
1244 
1245 	/* Query new route. */
1246 	fl4 = &inet->cork.fl.u.ip4;
1247 	rt = ip_route_connect(fl4, daddr, 0, sk->sk_bound_dev_if,
1248 			      sk->sk_protocol, inet->inet_sport,
1249 			      inet->inet_dport, sk);
1250 	if (IS_ERR(rt))
1251 		return PTR_ERR(rt);
1252 
1253 	new_saddr = fl4->saddr;
1254 
1255 	if (new_saddr == old_saddr) {
1256 		sk_setup_caps(sk, &rt->dst);
1257 		return 0;
1258 	}
1259 
1260 	err = inet_bhash2_update_saddr(sk, &new_saddr, AF_INET);
1261 	if (err) {
1262 		ip_rt_put(rt);
1263 		return err;
1264 	}
1265 
1266 	sk_setup_caps(sk, &rt->dst);
1267 
1268 	if (READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) > 1) {
1269 		pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1270 			__func__, &old_saddr, &new_saddr);
1271 	}
1272 
1273 	/*
1274 	 * XXX The only one ugly spot where we need to
1275 	 * XXX really change the sockets identity after
1276 	 * XXX it has entered the hashes. -DaveM
1277 	 *
1278 	 * Besides that, it does not check for connection
1279 	 * uniqueness. Wait for troubles.
1280 	 */
1281 	return __sk_prot_rehash(sk);
1282 }
1283 
1284 int inet_sk_rebuild_header(struct sock *sk)
1285 {
1286 	struct inet_sock *inet = inet_sk(sk);
1287 	struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1288 	__be32 daddr;
1289 	struct ip_options_rcu *inet_opt;
1290 	struct flowi4 *fl4;
1291 	int err;
1292 
1293 	/* Route is OK, nothing to do. */
1294 	if (rt)
1295 		return 0;
1296 
1297 	/* Reroute. */
1298 	rcu_read_lock();
1299 	inet_opt = rcu_dereference(inet->inet_opt);
1300 	daddr = inet->inet_daddr;
1301 	if (inet_opt && inet_opt->opt.srr)
1302 		daddr = inet_opt->opt.faddr;
1303 	rcu_read_unlock();
1304 	fl4 = &inet->cork.fl.u.ip4;
1305 	rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1306 				   inet->inet_dport, inet->inet_sport,
1307 				   sk->sk_protocol, RT_CONN_FLAGS(sk),
1308 				   sk->sk_bound_dev_if);
1309 	if (!IS_ERR(rt)) {
1310 		err = 0;
1311 		sk_setup_caps(sk, &rt->dst);
1312 	} else {
1313 		err = PTR_ERR(rt);
1314 
1315 		/* Routing failed... */
1316 		sk->sk_route_caps = 0;
1317 		/*
1318 		 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1319 		 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1320 		 */
1321 		if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) ||
1322 		    sk->sk_state != TCP_SYN_SENT ||
1323 		    (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1324 		    (err = inet_sk_reselect_saddr(sk)) != 0)
1325 			sk->sk_err_soft = -err;
1326 	}
1327 
1328 	return err;
1329 }
1330 EXPORT_SYMBOL(inet_sk_rebuild_header);
1331 
1332 void inet_sk_set_state(struct sock *sk, int state)
1333 {
1334 	trace_inet_sock_set_state(sk, sk->sk_state, state);
1335 	sk->sk_state = state;
1336 }
1337 EXPORT_SYMBOL(inet_sk_set_state);
1338 
1339 void inet_sk_state_store(struct sock *sk, int newstate)
1340 {
1341 	trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1342 	smp_store_release(&sk->sk_state, newstate);
1343 }
1344 
1345 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1346 				 netdev_features_t features)
1347 {
1348 	bool udpfrag = false, fixedid = false, gso_partial, encap;
1349 	struct sk_buff *segs = ERR_PTR(-EINVAL);
1350 	const struct net_offload *ops;
1351 	unsigned int offset = 0;
1352 	struct iphdr *iph;
1353 	int proto, tot_len;
1354 	int nhoff;
1355 	int ihl;
1356 	int id;
1357 
1358 	skb_reset_network_header(skb);
1359 	nhoff = skb_network_header(skb) - skb_mac_header(skb);
1360 	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1361 		goto out;
1362 
1363 	iph = ip_hdr(skb);
1364 	ihl = iph->ihl * 4;
1365 	if (ihl < sizeof(*iph))
1366 		goto out;
1367 
1368 	id = ntohs(iph->id);
1369 	proto = iph->protocol;
1370 
1371 	/* Warning: after this point, iph might be no longer valid */
1372 	if (unlikely(!pskb_may_pull(skb, ihl)))
1373 		goto out;
1374 	__skb_pull(skb, ihl);
1375 
1376 	encap = SKB_GSO_CB(skb)->encap_level > 0;
1377 	if (encap)
1378 		features &= skb->dev->hw_enc_features;
1379 	SKB_GSO_CB(skb)->encap_level += ihl;
1380 
1381 	skb_reset_transport_header(skb);
1382 
1383 	segs = ERR_PTR(-EPROTONOSUPPORT);
1384 
1385 	if (!skb->encapsulation || encap) {
1386 		udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1387 		fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1388 
1389 		/* fixed ID is invalid if DF bit is not set */
1390 		if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1391 			goto out;
1392 	}
1393 
1394 	ops = rcu_dereference(inet_offloads[proto]);
1395 	if (likely(ops && ops->callbacks.gso_segment)) {
1396 		segs = ops->callbacks.gso_segment(skb, features);
1397 		if (!segs)
1398 			skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
1399 	}
1400 
1401 	if (IS_ERR_OR_NULL(segs))
1402 		goto out;
1403 
1404 	gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1405 
1406 	skb = segs;
1407 	do {
1408 		iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1409 		if (udpfrag) {
1410 			iph->frag_off = htons(offset >> 3);
1411 			if (skb->next)
1412 				iph->frag_off |= htons(IP_MF);
1413 			offset += skb->len - nhoff - ihl;
1414 			tot_len = skb->len - nhoff;
1415 		} else if (skb_is_gso(skb)) {
1416 			if (!fixedid) {
1417 				iph->id = htons(id);
1418 				id += skb_shinfo(skb)->gso_segs;
1419 			}
1420 
1421 			if (gso_partial)
1422 				tot_len = skb_shinfo(skb)->gso_size +
1423 					  SKB_GSO_CB(skb)->data_offset +
1424 					  skb->head - (unsigned char *)iph;
1425 			else
1426 				tot_len = skb->len - nhoff;
1427 		} else {
1428 			if (!fixedid)
1429 				iph->id = htons(id++);
1430 			tot_len = skb->len - nhoff;
1431 		}
1432 		iph->tot_len = htons(tot_len);
1433 		ip_send_check(iph);
1434 		if (encap)
1435 			skb_reset_inner_headers(skb);
1436 		skb->network_header = (u8 *)iph - skb->head;
1437 		skb_reset_mac_len(skb);
1438 	} while ((skb = skb->next));
1439 
1440 out:
1441 	return segs;
1442 }
1443 
1444 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1445 					netdev_features_t features)
1446 {
1447 	if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1448 		return ERR_PTR(-EINVAL);
1449 
1450 	return inet_gso_segment(skb, features);
1451 }
1452 
1453 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1454 {
1455 	const struct net_offload *ops;
1456 	struct sk_buff *pp = NULL;
1457 	const struct iphdr *iph;
1458 	struct sk_buff *p;
1459 	unsigned int hlen;
1460 	unsigned int off;
1461 	unsigned int id;
1462 	int flush = 1;
1463 	int proto;
1464 
1465 	off = skb_gro_offset(skb);
1466 	hlen = off + sizeof(*iph);
1467 	iph = skb_gro_header(skb, hlen, off);
1468 	if (unlikely(!iph))
1469 		goto out;
1470 
1471 	proto = iph->protocol;
1472 
1473 	ops = rcu_dereference(inet_offloads[proto]);
1474 	if (!ops || !ops->callbacks.gro_receive)
1475 		goto out;
1476 
1477 	if (*(u8 *)iph != 0x45)
1478 		goto out;
1479 
1480 	if (ip_is_fragment(iph))
1481 		goto out;
1482 
1483 	if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1484 		goto out;
1485 
1486 	NAPI_GRO_CB(skb)->proto = proto;
1487 	id = ntohl(*(__be32 *)&iph->id);
1488 	flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1489 	id >>= 16;
1490 
1491 	list_for_each_entry(p, head, list) {
1492 		struct iphdr *iph2;
1493 		u16 flush_id;
1494 
1495 		if (!NAPI_GRO_CB(p)->same_flow)
1496 			continue;
1497 
1498 		iph2 = (struct iphdr *)(p->data + off);
1499 		/* The above works because, with the exception of the top
1500 		 * (inner most) layer, we only aggregate pkts with the same
1501 		 * hdr length so all the hdrs we'll need to verify will start
1502 		 * at the same offset.
1503 		 */
1504 		if ((iph->protocol ^ iph2->protocol) |
1505 		    ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1506 		    ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1507 			NAPI_GRO_CB(p)->same_flow = 0;
1508 			continue;
1509 		}
1510 
1511 		/* All fields must match except length and checksum. */
1512 		NAPI_GRO_CB(p)->flush |=
1513 			(iph->ttl ^ iph2->ttl) |
1514 			(iph->tos ^ iph2->tos) |
1515 			((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1516 
1517 		NAPI_GRO_CB(p)->flush |= flush;
1518 
1519 		/* We need to store of the IP ID check to be included later
1520 		 * when we can verify that this packet does in fact belong
1521 		 * to a given flow.
1522 		 */
1523 		flush_id = (u16)(id - ntohs(iph2->id));
1524 
1525 		/* This bit of code makes it much easier for us to identify
1526 		 * the cases where we are doing atomic vs non-atomic IP ID
1527 		 * checks.  Specifically an atomic check can return IP ID
1528 		 * values 0 - 0xFFFF, while a non-atomic check can only
1529 		 * return 0 or 0xFFFF.
1530 		 */
1531 		if (!NAPI_GRO_CB(p)->is_atomic ||
1532 		    !(iph->frag_off & htons(IP_DF))) {
1533 			flush_id ^= NAPI_GRO_CB(p)->count;
1534 			flush_id = flush_id ? 0xFFFF : 0;
1535 		}
1536 
1537 		/* If the previous IP ID value was based on an atomic
1538 		 * datagram we can overwrite the value and ignore it.
1539 		 */
1540 		if (NAPI_GRO_CB(skb)->is_atomic)
1541 			NAPI_GRO_CB(p)->flush_id = flush_id;
1542 		else
1543 			NAPI_GRO_CB(p)->flush_id |= flush_id;
1544 	}
1545 
1546 	NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1547 	NAPI_GRO_CB(skb)->flush |= flush;
1548 	skb_set_network_header(skb, off);
1549 	/* The above will be needed by the transport layer if there is one
1550 	 * immediately following this IP hdr.
1551 	 */
1552 
1553 	/* Note : No need to call skb_gro_postpull_rcsum() here,
1554 	 * as we already checked checksum over ipv4 header was 0
1555 	 */
1556 	skb_gro_pull(skb, sizeof(*iph));
1557 	skb_set_transport_header(skb, skb_gro_offset(skb));
1558 
1559 	pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1560 				       ops->callbacks.gro_receive, head, skb);
1561 
1562 out:
1563 	skb_gro_flush_final(skb, pp, flush);
1564 
1565 	return pp;
1566 }
1567 
1568 static struct sk_buff *ipip_gro_receive(struct list_head *head,
1569 					struct sk_buff *skb)
1570 {
1571 	if (NAPI_GRO_CB(skb)->encap_mark) {
1572 		NAPI_GRO_CB(skb)->flush = 1;
1573 		return NULL;
1574 	}
1575 
1576 	NAPI_GRO_CB(skb)->encap_mark = 1;
1577 
1578 	return inet_gro_receive(head, skb);
1579 }
1580 
1581 #define SECONDS_PER_DAY	86400
1582 
1583 /* inet_current_timestamp - Return IP network timestamp
1584  *
1585  * Return milliseconds since midnight in network byte order.
1586  */
1587 __be32 inet_current_timestamp(void)
1588 {
1589 	u32 secs;
1590 	u32 msecs;
1591 	struct timespec64 ts;
1592 
1593 	ktime_get_real_ts64(&ts);
1594 
1595 	/* Get secs since midnight. */
1596 	(void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1597 	/* Convert to msecs. */
1598 	msecs = secs * MSEC_PER_SEC;
1599 	/* Convert nsec to msec. */
1600 	msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1601 
1602 	/* Convert to network byte order. */
1603 	return htonl(msecs);
1604 }
1605 EXPORT_SYMBOL(inet_current_timestamp);
1606 
1607 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1608 {
1609 	if (sk->sk_family == AF_INET)
1610 		return ip_recv_error(sk, msg, len, addr_len);
1611 #if IS_ENABLED(CONFIG_IPV6)
1612 	if (sk->sk_family == AF_INET6)
1613 		return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1614 #endif
1615 	return -EINVAL;
1616 }
1617 
1618 int inet_gro_complete(struct sk_buff *skb, int nhoff)
1619 {
1620 	struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1621 	const struct net_offload *ops;
1622 	__be16 totlen = iph->tot_len;
1623 	int proto = iph->protocol;
1624 	int err = -ENOSYS;
1625 
1626 	if (skb->encapsulation) {
1627 		skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1628 		skb_set_inner_network_header(skb, nhoff);
1629 	}
1630 
1631 	iph_set_totlen(iph, skb->len - nhoff);
1632 	csum_replace2(&iph->check, totlen, iph->tot_len);
1633 
1634 	ops = rcu_dereference(inet_offloads[proto]);
1635 	if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1636 		goto out;
1637 
1638 	/* Only need to add sizeof(*iph) to get to the next hdr below
1639 	 * because any hdr with option will have been flushed in
1640 	 * inet_gro_receive().
1641 	 */
1642 	err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1643 			      tcp4_gro_complete, udp4_gro_complete,
1644 			      skb, nhoff + sizeof(*iph));
1645 
1646 out:
1647 	return err;
1648 }
1649 
1650 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1651 {
1652 	skb->encapsulation = 1;
1653 	skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1654 	return inet_gro_complete(skb, nhoff);
1655 }
1656 
1657 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1658 			 unsigned short type, unsigned char protocol,
1659 			 struct net *net)
1660 {
1661 	struct socket *sock;
1662 	int rc = sock_create_kern(net, family, type, protocol, &sock);
1663 
1664 	if (rc == 0) {
1665 		*sk = sock->sk;
1666 		(*sk)->sk_allocation = GFP_ATOMIC;
1667 		(*sk)->sk_use_task_frag = false;
1668 		/*
1669 		 * Unhash it so that IP input processing does not even see it,
1670 		 * we do not wish this socket to see incoming packets.
1671 		 */
1672 		(*sk)->sk_prot->unhash(*sk);
1673 	}
1674 	return rc;
1675 }
1676 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1677 
1678 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1679 {
1680 	unsigned long res = 0;
1681 	int i;
1682 
1683 	for_each_possible_cpu(i)
1684 		res += snmp_get_cpu_field(mib, i, offt);
1685 	return res;
1686 }
1687 EXPORT_SYMBOL_GPL(snmp_fold_field);
1688 
1689 #if BITS_PER_LONG==32
1690 
1691 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1692 			 size_t syncp_offset)
1693 {
1694 	void *bhptr;
1695 	struct u64_stats_sync *syncp;
1696 	u64 v;
1697 	unsigned int start;
1698 
1699 	bhptr = per_cpu_ptr(mib, cpu);
1700 	syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1701 	do {
1702 		start = u64_stats_fetch_begin(syncp);
1703 		v = *(((u64 *)bhptr) + offt);
1704 	} while (u64_stats_fetch_retry(syncp, start));
1705 
1706 	return v;
1707 }
1708 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1709 
1710 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1711 {
1712 	u64 res = 0;
1713 	int cpu;
1714 
1715 	for_each_possible_cpu(cpu) {
1716 		res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1717 	}
1718 	return res;
1719 }
1720 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1721 #endif
1722 
1723 #ifdef CONFIG_IP_MULTICAST
1724 static const struct net_protocol igmp_protocol = {
1725 	.handler =	igmp_rcv,
1726 };
1727 #endif
1728 
1729 static const struct net_protocol tcp_protocol = {
1730 	.handler	=	tcp_v4_rcv,
1731 	.err_handler	=	tcp_v4_err,
1732 	.no_policy	=	1,
1733 	.icmp_strict_tag_validation = 1,
1734 };
1735 
1736 static const struct net_protocol udp_protocol = {
1737 	.handler =	udp_rcv,
1738 	.err_handler =	udp_err,
1739 	.no_policy =	1,
1740 };
1741 
1742 static const struct net_protocol icmp_protocol = {
1743 	.handler =	icmp_rcv,
1744 	.err_handler =	icmp_err,
1745 	.no_policy =	1,
1746 };
1747 
1748 static __net_init int ipv4_mib_init_net(struct net *net)
1749 {
1750 	int i;
1751 
1752 	net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1753 	if (!net->mib.tcp_statistics)
1754 		goto err_tcp_mib;
1755 	net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1756 	if (!net->mib.ip_statistics)
1757 		goto err_ip_mib;
1758 
1759 	for_each_possible_cpu(i) {
1760 		struct ipstats_mib *af_inet_stats;
1761 		af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1762 		u64_stats_init(&af_inet_stats->syncp);
1763 	}
1764 
1765 	net->mib.net_statistics = alloc_percpu(struct linux_mib);
1766 	if (!net->mib.net_statistics)
1767 		goto err_net_mib;
1768 	net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1769 	if (!net->mib.udp_statistics)
1770 		goto err_udp_mib;
1771 	net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1772 	if (!net->mib.udplite_statistics)
1773 		goto err_udplite_mib;
1774 	net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1775 	if (!net->mib.icmp_statistics)
1776 		goto err_icmp_mib;
1777 	net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1778 					      GFP_KERNEL);
1779 	if (!net->mib.icmpmsg_statistics)
1780 		goto err_icmpmsg_mib;
1781 
1782 	tcp_mib_init(net);
1783 	return 0;
1784 
1785 err_icmpmsg_mib:
1786 	free_percpu(net->mib.icmp_statistics);
1787 err_icmp_mib:
1788 	free_percpu(net->mib.udplite_statistics);
1789 err_udplite_mib:
1790 	free_percpu(net->mib.udp_statistics);
1791 err_udp_mib:
1792 	free_percpu(net->mib.net_statistics);
1793 err_net_mib:
1794 	free_percpu(net->mib.ip_statistics);
1795 err_ip_mib:
1796 	free_percpu(net->mib.tcp_statistics);
1797 err_tcp_mib:
1798 	return -ENOMEM;
1799 }
1800 
1801 static __net_exit void ipv4_mib_exit_net(struct net *net)
1802 {
1803 	kfree(net->mib.icmpmsg_statistics);
1804 	free_percpu(net->mib.icmp_statistics);
1805 	free_percpu(net->mib.udplite_statistics);
1806 	free_percpu(net->mib.udp_statistics);
1807 	free_percpu(net->mib.net_statistics);
1808 	free_percpu(net->mib.ip_statistics);
1809 	free_percpu(net->mib.tcp_statistics);
1810 #ifdef CONFIG_MPTCP
1811 	/* allocated on demand, see mptcp_init_sock() */
1812 	free_percpu(net->mib.mptcp_statistics);
1813 #endif
1814 }
1815 
1816 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1817 	.init = ipv4_mib_init_net,
1818 	.exit = ipv4_mib_exit_net,
1819 };
1820 
1821 static int __init init_ipv4_mibs(void)
1822 {
1823 	return register_pernet_subsys(&ipv4_mib_ops);
1824 }
1825 
1826 static __net_init int inet_init_net(struct net *net)
1827 {
1828 	/*
1829 	 * Set defaults for local port range
1830 	 */
1831 	seqlock_init(&net->ipv4.ip_local_ports.lock);
1832 	net->ipv4.ip_local_ports.range[0] =  32768;
1833 	net->ipv4.ip_local_ports.range[1] =  60999;
1834 
1835 	seqlock_init(&net->ipv4.ping_group_range.lock);
1836 	/*
1837 	 * Sane defaults - nobody may create ping sockets.
1838 	 * Boot scripts should set this to distro-specific group.
1839 	 */
1840 	net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1841 	net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1842 
1843 	/* Default values for sysctl-controlled parameters.
1844 	 * We set them here, in case sysctl is not compiled.
1845 	 */
1846 	net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1847 	net->ipv4.sysctl_ip_fwd_update_priority = 1;
1848 	net->ipv4.sysctl_ip_dynaddr = 0;
1849 	net->ipv4.sysctl_ip_early_demux = 1;
1850 	net->ipv4.sysctl_udp_early_demux = 1;
1851 	net->ipv4.sysctl_tcp_early_demux = 1;
1852 	net->ipv4.sysctl_nexthop_compat_mode = 1;
1853 #ifdef CONFIG_SYSCTL
1854 	net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1855 #endif
1856 
1857 	/* Some igmp sysctl, whose values are always used */
1858 	net->ipv4.sysctl_igmp_max_memberships = 20;
1859 	net->ipv4.sysctl_igmp_max_msf = 10;
1860 	/* IGMP reports for link-local multicast groups are enabled by default */
1861 	net->ipv4.sysctl_igmp_llm_reports = 1;
1862 	net->ipv4.sysctl_igmp_qrv = 2;
1863 
1864 	net->ipv4.sysctl_fib_notify_on_flag_change = 0;
1865 
1866 	return 0;
1867 }
1868 
1869 static __net_initdata struct pernet_operations af_inet_ops = {
1870 	.init = inet_init_net,
1871 };
1872 
1873 static int __init init_inet_pernet_ops(void)
1874 {
1875 	return register_pernet_subsys(&af_inet_ops);
1876 }
1877 
1878 static int ipv4_proc_init(void);
1879 
1880 /*
1881  *	IP protocol layer initialiser
1882  */
1883 
1884 static struct packet_offload ip_packet_offload __read_mostly = {
1885 	.type = cpu_to_be16(ETH_P_IP),
1886 	.callbacks = {
1887 		.gso_segment = inet_gso_segment,
1888 		.gro_receive = inet_gro_receive,
1889 		.gro_complete = inet_gro_complete,
1890 	},
1891 };
1892 
1893 static const struct net_offload ipip_offload = {
1894 	.callbacks = {
1895 		.gso_segment	= ipip_gso_segment,
1896 		.gro_receive	= ipip_gro_receive,
1897 		.gro_complete	= ipip_gro_complete,
1898 	},
1899 };
1900 
1901 static int __init ipip_offload_init(void)
1902 {
1903 	return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1904 }
1905 
1906 static int __init ipv4_offload_init(void)
1907 {
1908 	/*
1909 	 * Add offloads
1910 	 */
1911 	if (udpv4_offload_init() < 0)
1912 		pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1913 	if (tcpv4_offload_init() < 0)
1914 		pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1915 	if (ipip_offload_init() < 0)
1916 		pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1917 
1918 	dev_add_offload(&ip_packet_offload);
1919 	return 0;
1920 }
1921 
1922 fs_initcall(ipv4_offload_init);
1923 
1924 static struct packet_type ip_packet_type __read_mostly = {
1925 	.type = cpu_to_be16(ETH_P_IP),
1926 	.func = ip_rcv,
1927 	.list_func = ip_list_rcv,
1928 };
1929 
1930 static int __init inet_init(void)
1931 {
1932 	struct inet_protosw *q;
1933 	struct list_head *r;
1934 	int rc;
1935 
1936 	sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1937 
1938 	raw_hashinfo_init(&raw_v4_hashinfo);
1939 
1940 	rc = proto_register(&tcp_prot, 1);
1941 	if (rc)
1942 		goto out;
1943 
1944 	rc = proto_register(&udp_prot, 1);
1945 	if (rc)
1946 		goto out_unregister_tcp_proto;
1947 
1948 	rc = proto_register(&raw_prot, 1);
1949 	if (rc)
1950 		goto out_unregister_udp_proto;
1951 
1952 	rc = proto_register(&ping_prot, 1);
1953 	if (rc)
1954 		goto out_unregister_raw_proto;
1955 
1956 	/*
1957 	 *	Tell SOCKET that we are alive...
1958 	 */
1959 
1960 	(void)sock_register(&inet_family_ops);
1961 
1962 #ifdef CONFIG_SYSCTL
1963 	ip_static_sysctl_init();
1964 #endif
1965 
1966 	/*
1967 	 *	Add all the base protocols.
1968 	 */
1969 
1970 	if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1971 		pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1972 	if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1973 		pr_crit("%s: Cannot add UDP protocol\n", __func__);
1974 	if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1975 		pr_crit("%s: Cannot add TCP protocol\n", __func__);
1976 #ifdef CONFIG_IP_MULTICAST
1977 	if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1978 		pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1979 #endif
1980 
1981 	/* Register the socket-side information for inet_create. */
1982 	for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1983 		INIT_LIST_HEAD(r);
1984 
1985 	for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1986 		inet_register_protosw(q);
1987 
1988 	/*
1989 	 *	Set the ARP module up
1990 	 */
1991 
1992 	arp_init();
1993 
1994 	/*
1995 	 *	Set the IP module up
1996 	 */
1997 
1998 	ip_init();
1999 
2000 	/* Initialise per-cpu ipv4 mibs */
2001 	if (init_ipv4_mibs())
2002 		panic("%s: Cannot init ipv4 mibs\n", __func__);
2003 
2004 	/* Setup TCP slab cache for open requests. */
2005 	tcp_init();
2006 
2007 	/* Setup UDP memory threshold */
2008 	udp_init();
2009 
2010 	/* Add UDP-Lite (RFC 3828) */
2011 	udplite4_register();
2012 
2013 	raw_init();
2014 
2015 	ping_init();
2016 
2017 	/*
2018 	 *	Set the ICMP layer up
2019 	 */
2020 
2021 	if (icmp_init() < 0)
2022 		panic("Failed to create the ICMP control socket.\n");
2023 
2024 	/*
2025 	 *	Initialise the multicast router
2026 	 */
2027 #if defined(CONFIG_IP_MROUTE)
2028 	if (ip_mr_init())
2029 		pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2030 #endif
2031 
2032 	if (init_inet_pernet_ops())
2033 		pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2034 
2035 	ipv4_proc_init();
2036 
2037 	ipfrag_init();
2038 
2039 	dev_add_pack(&ip_packet_type);
2040 
2041 	ip_tunnel_core_init();
2042 
2043 	rc = 0;
2044 out:
2045 	return rc;
2046 out_unregister_raw_proto:
2047 	proto_unregister(&raw_prot);
2048 out_unregister_udp_proto:
2049 	proto_unregister(&udp_prot);
2050 out_unregister_tcp_proto:
2051 	proto_unregister(&tcp_prot);
2052 	goto out;
2053 }
2054 
2055 fs_initcall(inet_init);
2056 
2057 /* ------------------------------------------------------------------------ */
2058 
2059 #ifdef CONFIG_PROC_FS
2060 static int __init ipv4_proc_init(void)
2061 {
2062 	int rc = 0;
2063 
2064 	if (raw_proc_init())
2065 		goto out_raw;
2066 	if (tcp4_proc_init())
2067 		goto out_tcp;
2068 	if (udp4_proc_init())
2069 		goto out_udp;
2070 	if (ping_proc_init())
2071 		goto out_ping;
2072 	if (ip_misc_proc_init())
2073 		goto out_misc;
2074 out:
2075 	return rc;
2076 out_misc:
2077 	ping_proc_exit();
2078 out_ping:
2079 	udp4_proc_exit();
2080 out_udp:
2081 	tcp4_proc_exit();
2082 out_tcp:
2083 	raw_proc_exit();
2084 out_raw:
2085 	rc = -ENOMEM;
2086 	goto out;
2087 }
2088 
2089 #else /* CONFIG_PROC_FS */
2090 static int __init ipv4_proc_init(void)
2091 {
2092 	return 0;
2093 }
2094 #endif /* CONFIG_PROC_FS */
2095