xref: /linux/net/ipv4/af_inet.c (revision 08f3e0873ac203449465c2b8473d684e2f9f41d1)
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 	if (sk->sk_rx_skb_cache) {
137 		__kfree_skb(sk->sk_rx_skb_cache);
138 		sk->sk_rx_skb_cache = NULL;
139 	}
140 	__skb_queue_purge(&sk->sk_error_queue);
141 
142 	sk_mem_reclaim(sk);
143 
144 	if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
145 		pr_err("Attempt to release TCP socket in state %d %p\n",
146 		       sk->sk_state, sk);
147 		return;
148 	}
149 	if (!sock_flag(sk, SOCK_DEAD)) {
150 		pr_err("Attempt to release alive inet socket %p\n", sk);
151 		return;
152 	}
153 
154 	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
155 	WARN_ON(refcount_read(&sk->sk_wmem_alloc));
156 	WARN_ON(sk->sk_wmem_queued);
157 	WARN_ON(sk->sk_forward_alloc);
158 
159 	kfree(rcu_dereference_protected(inet->inet_opt, 1));
160 	dst_release(rcu_dereference_protected(sk->sk_dst_cache, 1));
161 	dst_release(sk->sk_rx_dst);
162 	sk_refcnt_debug_dec(sk);
163 }
164 EXPORT_SYMBOL(inet_sock_destruct);
165 
166 /*
167  *	The routines beyond this point handle the behaviour of an AF_INET
168  *	socket object. Mostly it punts to the subprotocols of IP to do
169  *	the work.
170  */
171 
172 /*
173  *	Automatically bind an unbound socket.
174  */
175 
176 static int inet_autobind(struct sock *sk)
177 {
178 	struct inet_sock *inet;
179 	/* We may need to bind the socket. */
180 	lock_sock(sk);
181 	inet = inet_sk(sk);
182 	if (!inet->inet_num) {
183 		if (sk->sk_prot->get_port(sk, 0)) {
184 			release_sock(sk);
185 			return -EAGAIN;
186 		}
187 		inet->inet_sport = htons(inet->inet_num);
188 	}
189 	release_sock(sk);
190 	return 0;
191 }
192 
193 /*
194  *	Move a socket into listening state.
195  */
196 int inet_listen(struct socket *sock, int backlog)
197 {
198 	struct sock *sk = sock->sk;
199 	unsigned char old_state;
200 	int err, tcp_fastopen;
201 
202 	lock_sock(sk);
203 
204 	err = -EINVAL;
205 	if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
206 		goto out;
207 
208 	old_state = sk->sk_state;
209 	if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
210 		goto out;
211 
212 	WRITE_ONCE(sk->sk_max_ack_backlog, backlog);
213 	/* Really, if the socket is already in listen state
214 	 * we can only allow the backlog to be adjusted.
215 	 */
216 	if (old_state != TCP_LISTEN) {
217 		/* Enable TFO w/o requiring TCP_FASTOPEN socket option.
218 		 * Note that only TCP sockets (SOCK_STREAM) will reach here.
219 		 * Also fastopen backlog may already been set via the option
220 		 * because the socket was in TCP_LISTEN state previously but
221 		 * was shutdown() rather than close().
222 		 */
223 		tcp_fastopen = sock_net(sk)->ipv4.sysctl_tcp_fastopen;
224 		if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
225 		    (tcp_fastopen & TFO_SERVER_ENABLE) &&
226 		    !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
227 			fastopen_queue_tune(sk, backlog);
228 			tcp_fastopen_init_key_once(sock_net(sk));
229 		}
230 
231 		err = inet_csk_listen_start(sk, backlog);
232 		if (err)
233 			goto out;
234 		tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL);
235 	}
236 	err = 0;
237 
238 out:
239 	release_sock(sk);
240 	return err;
241 }
242 EXPORT_SYMBOL(inet_listen);
243 
244 /*
245  *	Create an inet socket.
246  */
247 
248 static int inet_create(struct net *net, struct socket *sock, int protocol,
249 		       int kern)
250 {
251 	struct sock *sk;
252 	struct inet_protosw *answer;
253 	struct inet_sock *inet;
254 	struct proto *answer_prot;
255 	unsigned char answer_flags;
256 	int try_loading_module = 0;
257 	int err;
258 
259 	if (protocol < 0 || protocol >= IPPROTO_MAX)
260 		return -EINVAL;
261 
262 	sock->state = SS_UNCONNECTED;
263 
264 	/* Look for the requested type/protocol pair. */
265 lookup_protocol:
266 	err = -ESOCKTNOSUPPORT;
267 	rcu_read_lock();
268 	list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
269 
270 		err = 0;
271 		/* Check the non-wild match. */
272 		if (protocol == answer->protocol) {
273 			if (protocol != IPPROTO_IP)
274 				break;
275 		} else {
276 			/* Check for the two wild cases. */
277 			if (IPPROTO_IP == protocol) {
278 				protocol = answer->protocol;
279 				break;
280 			}
281 			if (IPPROTO_IP == answer->protocol)
282 				break;
283 		}
284 		err = -EPROTONOSUPPORT;
285 	}
286 
287 	if (unlikely(err)) {
288 		if (try_loading_module < 2) {
289 			rcu_read_unlock();
290 			/*
291 			 * Be more specific, e.g. net-pf-2-proto-132-type-1
292 			 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
293 			 */
294 			if (++try_loading_module == 1)
295 				request_module("net-pf-%d-proto-%d-type-%d",
296 					       PF_INET, protocol, sock->type);
297 			/*
298 			 * Fall back to generic, e.g. net-pf-2-proto-132
299 			 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
300 			 */
301 			else
302 				request_module("net-pf-%d-proto-%d",
303 					       PF_INET, protocol);
304 			goto lookup_protocol;
305 		} else
306 			goto out_rcu_unlock;
307 	}
308 
309 	err = -EPERM;
310 	if (sock->type == SOCK_RAW && !kern &&
311 	    !ns_capable(net->user_ns, CAP_NET_RAW))
312 		goto out_rcu_unlock;
313 
314 	sock->ops = answer->ops;
315 	answer_prot = answer->prot;
316 	answer_flags = answer->flags;
317 	rcu_read_unlock();
318 
319 	WARN_ON(!answer_prot->slab);
320 
321 	err = -ENOMEM;
322 	sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
323 	if (!sk)
324 		goto out;
325 
326 	err = 0;
327 	if (INET_PROTOSW_REUSE & answer_flags)
328 		sk->sk_reuse = SK_CAN_REUSE;
329 
330 	inet = inet_sk(sk);
331 	inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
332 
333 	inet->nodefrag = 0;
334 
335 	if (SOCK_RAW == sock->type) {
336 		inet->inet_num = protocol;
337 		if (IPPROTO_RAW == protocol)
338 			inet->hdrincl = 1;
339 	}
340 
341 	if (net->ipv4.sysctl_ip_no_pmtu_disc)
342 		inet->pmtudisc = IP_PMTUDISC_DONT;
343 	else
344 		inet->pmtudisc = IP_PMTUDISC_WANT;
345 
346 	inet->inet_id = 0;
347 
348 	sock_init_data(sock, sk);
349 
350 	sk->sk_destruct	   = inet_sock_destruct;
351 	sk->sk_protocol	   = protocol;
352 	sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
353 
354 	inet->uc_ttl	= -1;
355 	inet->mc_loop	= 1;
356 	inet->mc_ttl	= 1;
357 	inet->mc_all	= 1;
358 	inet->mc_index	= 0;
359 	inet->mc_list	= NULL;
360 	inet->rcv_tos	= 0;
361 
362 	sk_refcnt_debug_inc(sk);
363 
364 	if (inet->inet_num) {
365 		/* It assumes that any protocol which allows
366 		 * the user to assign a number at socket
367 		 * creation time automatically
368 		 * shares.
369 		 */
370 		inet->inet_sport = htons(inet->inet_num);
371 		/* Add to protocol hash chains. */
372 		err = sk->sk_prot->hash(sk);
373 		if (err) {
374 			sk_common_release(sk);
375 			goto out;
376 		}
377 	}
378 
379 	if (sk->sk_prot->init) {
380 		err = sk->sk_prot->init(sk);
381 		if (err) {
382 			sk_common_release(sk);
383 			goto out;
384 		}
385 	}
386 
387 	if (!kern) {
388 		err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
389 		if (err) {
390 			sk_common_release(sk);
391 			goto out;
392 		}
393 	}
394 out:
395 	return err;
396 out_rcu_unlock:
397 	rcu_read_unlock();
398 	goto out;
399 }
400 
401 
402 /*
403  *	The peer socket should always be NULL (or else). When we call this
404  *	function we are destroying the object and from then on nobody
405  *	should refer to it.
406  */
407 int inet_release(struct socket *sock)
408 {
409 	struct sock *sk = sock->sk;
410 
411 	if (sk) {
412 		long timeout;
413 
414 		if (!sk->sk_kern_sock)
415 			BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk);
416 
417 		/* Applications forget to leave groups before exiting */
418 		ip_mc_drop_socket(sk);
419 
420 		/* If linger is set, we don't return until the close
421 		 * is complete.  Otherwise we return immediately. The
422 		 * actually closing is done the same either way.
423 		 *
424 		 * If the close is due to the process exiting, we never
425 		 * linger..
426 		 */
427 		timeout = 0;
428 		if (sock_flag(sk, SOCK_LINGER) &&
429 		    !(current->flags & PF_EXITING))
430 			timeout = sk->sk_lingertime;
431 		sk->sk_prot->close(sk, timeout);
432 		sock->sk = NULL;
433 	}
434 	return 0;
435 }
436 EXPORT_SYMBOL(inet_release);
437 
438 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
439 {
440 	struct sock *sk = sock->sk;
441 	u32 flags = BIND_WITH_LOCK;
442 	int err;
443 
444 	/* If the socket has its own bind function then use it. (RAW) */
445 	if (sk->sk_prot->bind) {
446 		return sk->sk_prot->bind(sk, uaddr, addr_len);
447 	}
448 	if (addr_len < sizeof(struct sockaddr_in))
449 		return -EINVAL;
450 
451 	/* BPF prog is run before any checks are done so that if the prog
452 	 * changes context in a wrong way it will be caught.
453 	 */
454 	err = BPF_CGROUP_RUN_PROG_INET_BIND_LOCK(sk, uaddr,
455 						 CGROUP_INET4_BIND, &flags);
456 	if (err)
457 		return err;
458 
459 	return __inet_bind(sk, uaddr, addr_len, flags);
460 }
461 EXPORT_SYMBOL(inet_bind);
462 
463 int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
464 		u32 flags)
465 {
466 	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
467 	struct inet_sock *inet = inet_sk(sk);
468 	struct net *net = sock_net(sk);
469 	unsigned short snum;
470 	int chk_addr_ret;
471 	u32 tb_id = RT_TABLE_LOCAL;
472 	int err;
473 
474 	if (addr->sin_family != AF_INET) {
475 		/* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
476 		 * only if s_addr is INADDR_ANY.
477 		 */
478 		err = -EAFNOSUPPORT;
479 		if (addr->sin_family != AF_UNSPEC ||
480 		    addr->sin_addr.s_addr != htonl(INADDR_ANY))
481 			goto out;
482 	}
483 
484 	tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
485 	chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
486 
487 	/* Not specified by any standard per-se, however it breaks too
488 	 * many applications when removed.  It is unfortunate since
489 	 * allowing applications to make a non-local bind solves
490 	 * several problems with systems using dynamic addressing.
491 	 * (ie. your servers still start up even if your ISDN link
492 	 *  is temporarily down)
493 	 */
494 	err = -EADDRNOTAVAIL;
495 	if (!inet_can_nonlocal_bind(net, inet) &&
496 	    addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
497 	    chk_addr_ret != RTN_LOCAL &&
498 	    chk_addr_ret != RTN_MULTICAST &&
499 	    chk_addr_ret != RTN_BROADCAST)
500 		goto out;
501 
502 	snum = ntohs(addr->sin_port);
503 	err = -EACCES;
504 	if (!(flags & BIND_NO_CAP_NET_BIND_SERVICE) &&
505 	    snum && inet_port_requires_bind_service(net, snum) &&
506 	    !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
507 		goto out;
508 
509 	/*      We keep a pair of addresses. rcv_saddr is the one
510 	 *      used by hash lookups, and saddr is used for transmit.
511 	 *
512 	 *      In the BSD API these are the same except where it
513 	 *      would be illegal to use them (multicast/broadcast) in
514 	 *      which case the sending device address is used.
515 	 */
516 	if (flags & BIND_WITH_LOCK)
517 		lock_sock(sk);
518 
519 	/* Check these errors (active socket, double bind). */
520 	err = -EINVAL;
521 	if (sk->sk_state != TCP_CLOSE || inet->inet_num)
522 		goto out_release_sock;
523 
524 	inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
525 	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
526 		inet->inet_saddr = 0;  /* Use device */
527 
528 	/* Make sure we are allowed to bind here. */
529 	if (snum || !(inet->bind_address_no_port ||
530 		      (flags & BIND_FORCE_ADDRESS_NO_PORT))) {
531 		if (sk->sk_prot->get_port(sk, snum)) {
532 			inet->inet_saddr = inet->inet_rcv_saddr = 0;
533 			err = -EADDRINUSE;
534 			goto out_release_sock;
535 		}
536 		if (!(flags & BIND_FROM_BPF)) {
537 			err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
538 			if (err) {
539 				inet->inet_saddr = inet->inet_rcv_saddr = 0;
540 				goto out_release_sock;
541 			}
542 		}
543 	}
544 
545 	if (inet->inet_rcv_saddr)
546 		sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
547 	if (snum)
548 		sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
549 	inet->inet_sport = htons(inet->inet_num);
550 	inet->inet_daddr = 0;
551 	inet->inet_dport = 0;
552 	sk_dst_reset(sk);
553 	err = 0;
554 out_release_sock:
555 	if (flags & BIND_WITH_LOCK)
556 		release_sock(sk);
557 out:
558 	return err;
559 }
560 
561 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
562 		       int addr_len, int flags)
563 {
564 	struct sock *sk = sock->sk;
565 	int err;
566 
567 	if (addr_len < sizeof(uaddr->sa_family))
568 		return -EINVAL;
569 	if (uaddr->sa_family == AF_UNSPEC)
570 		return sk->sk_prot->disconnect(sk, flags);
571 
572 	if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
573 		err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
574 		if (err)
575 			return err;
576 	}
577 
578 	if (data_race(!inet_sk(sk)->inet_num) && inet_autobind(sk))
579 		return -EAGAIN;
580 	return sk->sk_prot->connect(sk, uaddr, addr_len);
581 }
582 EXPORT_SYMBOL(inet_dgram_connect);
583 
584 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
585 {
586 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
587 
588 	add_wait_queue(sk_sleep(sk), &wait);
589 	sk->sk_write_pending += writebias;
590 
591 	/* Basic assumption: if someone sets sk->sk_err, he _must_
592 	 * change state of the socket from TCP_SYN_*.
593 	 * Connect() does not allow to get error notifications
594 	 * without closing the socket.
595 	 */
596 	while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
597 		release_sock(sk);
598 		timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
599 		lock_sock(sk);
600 		if (signal_pending(current) || !timeo)
601 			break;
602 	}
603 	remove_wait_queue(sk_sleep(sk), &wait);
604 	sk->sk_write_pending -= writebias;
605 	return timeo;
606 }
607 
608 /*
609  *	Connect to a remote host. There is regrettably still a little
610  *	TCP 'magic' in here.
611  */
612 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
613 			  int addr_len, int flags, int is_sendmsg)
614 {
615 	struct sock *sk = sock->sk;
616 	int err;
617 	long timeo;
618 
619 	/*
620 	 * uaddr can be NULL and addr_len can be 0 if:
621 	 * sk is a TCP fastopen active socket and
622 	 * TCP_FASTOPEN_CONNECT sockopt is set and
623 	 * we already have a valid cookie for this socket.
624 	 * In this case, user can call write() after connect().
625 	 * write() will invoke tcp_sendmsg_fastopen() which calls
626 	 * __inet_stream_connect().
627 	 */
628 	if (uaddr) {
629 		if (addr_len < sizeof(uaddr->sa_family))
630 			return -EINVAL;
631 
632 		if (uaddr->sa_family == AF_UNSPEC) {
633 			err = sk->sk_prot->disconnect(sk, flags);
634 			sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
635 			goto out;
636 		}
637 	}
638 
639 	switch (sock->state) {
640 	default:
641 		err = -EINVAL;
642 		goto out;
643 	case SS_CONNECTED:
644 		err = -EISCONN;
645 		goto out;
646 	case SS_CONNECTING:
647 		if (inet_sk(sk)->defer_connect)
648 			err = is_sendmsg ? -EINPROGRESS : -EISCONN;
649 		else
650 			err = -EALREADY;
651 		/* Fall out of switch with err, set for this state */
652 		break;
653 	case SS_UNCONNECTED:
654 		err = -EISCONN;
655 		if (sk->sk_state != TCP_CLOSE)
656 			goto out;
657 
658 		if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
659 			err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
660 			if (err)
661 				goto out;
662 		}
663 
664 		err = sk->sk_prot->connect(sk, uaddr, addr_len);
665 		if (err < 0)
666 			goto out;
667 
668 		sock->state = SS_CONNECTING;
669 
670 		if (!err && inet_sk(sk)->defer_connect)
671 			goto out;
672 
673 		/* Just entered SS_CONNECTING state; the only
674 		 * difference is that return value in non-blocking
675 		 * case is EINPROGRESS, rather than EALREADY.
676 		 */
677 		err = -EINPROGRESS;
678 		break;
679 	}
680 
681 	timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
682 
683 	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
684 		int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
685 				tcp_sk(sk)->fastopen_req &&
686 				tcp_sk(sk)->fastopen_req->data ? 1 : 0;
687 
688 		/* Error code is set above */
689 		if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
690 			goto out;
691 
692 		err = sock_intr_errno(timeo);
693 		if (signal_pending(current))
694 			goto out;
695 	}
696 
697 	/* Connection was closed by RST, timeout, ICMP error
698 	 * or another process disconnected us.
699 	 */
700 	if (sk->sk_state == TCP_CLOSE)
701 		goto sock_error;
702 
703 	/* sk->sk_err may be not zero now, if RECVERR was ordered by user
704 	 * and error was received after socket entered established state.
705 	 * Hence, it is handled normally after connect() return successfully.
706 	 */
707 
708 	sock->state = SS_CONNECTED;
709 	err = 0;
710 out:
711 	return err;
712 
713 sock_error:
714 	err = sock_error(sk) ? : -ECONNABORTED;
715 	sock->state = SS_UNCONNECTED;
716 	if (sk->sk_prot->disconnect(sk, flags))
717 		sock->state = SS_DISCONNECTING;
718 	goto out;
719 }
720 EXPORT_SYMBOL(__inet_stream_connect);
721 
722 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
723 			int addr_len, int flags)
724 {
725 	int err;
726 
727 	lock_sock(sock->sk);
728 	err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
729 	release_sock(sock->sk);
730 	return err;
731 }
732 EXPORT_SYMBOL(inet_stream_connect);
733 
734 /*
735  *	Accept a pending connection. The TCP layer now gives BSD semantics.
736  */
737 
738 int inet_accept(struct socket *sock, struct socket *newsock, int flags,
739 		bool kern)
740 {
741 	struct sock *sk1 = sock->sk;
742 	int err = -EINVAL;
743 	struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err, kern);
744 
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 	sock_graft(sk2, newsock);
756 
757 	newsock->state = SS_CONNECTED;
758 	err = 0;
759 	release_sock(sk2);
760 do_err:
761 	return err;
762 }
763 EXPORT_SYMBOL(inet_accept);
764 
765 /*
766  *	This does both peername and sockname.
767  */
768 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
769 		 int peer)
770 {
771 	struct sock *sk		= sock->sk;
772 	struct inet_sock *inet	= inet_sk(sk);
773 	DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
774 
775 	sin->sin_family = AF_INET;
776 	if (peer) {
777 		if (!inet->inet_dport ||
778 		    (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
779 		     peer == 1))
780 			return -ENOTCONN;
781 		sin->sin_port = inet->inet_dport;
782 		sin->sin_addr.s_addr = inet->inet_daddr;
783 		BPF_CGROUP_RUN_SA_PROG_LOCK(sk, (struct sockaddr *)sin,
784 					    CGROUP_INET4_GETPEERNAME,
785 					    NULL);
786 	} else {
787 		__be32 addr = inet->inet_rcv_saddr;
788 		if (!addr)
789 			addr = inet->inet_saddr;
790 		sin->sin_port = inet->inet_sport;
791 		sin->sin_addr.s_addr = addr;
792 		BPF_CGROUP_RUN_SA_PROG_LOCK(sk, (struct sockaddr *)sin,
793 					    CGROUP_INET4_GETSOCKNAME,
794 					    NULL);
795 	}
796 	memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
797 	return sizeof(*sin);
798 }
799 EXPORT_SYMBOL(inet_getname);
800 
801 int inet_send_prepare(struct sock *sk)
802 {
803 	sock_rps_record_flow(sk);
804 
805 	/* We may need to bind the socket. */
806 	if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind &&
807 	    inet_autobind(sk))
808 		return -EAGAIN;
809 
810 	return 0;
811 }
812 EXPORT_SYMBOL_GPL(inet_send_prepare);
813 
814 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
815 {
816 	struct sock *sk = sock->sk;
817 
818 	if (unlikely(inet_send_prepare(sk)))
819 		return -EAGAIN;
820 
821 	return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
822 			       sk, msg, size);
823 }
824 EXPORT_SYMBOL(inet_sendmsg);
825 
826 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
827 		      size_t size, int flags)
828 {
829 	struct sock *sk = sock->sk;
830 
831 	if (unlikely(inet_send_prepare(sk)))
832 		return -EAGAIN;
833 
834 	if (sk->sk_prot->sendpage)
835 		return sk->sk_prot->sendpage(sk, page, offset, size, flags);
836 	return sock_no_sendpage(sock, page, offset, size, flags);
837 }
838 EXPORT_SYMBOL(inet_sendpage);
839 
840 INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
841 					  size_t, int, int, int *));
842 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
843 		 int flags)
844 {
845 	struct sock *sk = sock->sk;
846 	int addr_len = 0;
847 	int err;
848 
849 	if (likely(!(flags & MSG_ERRQUEUE)))
850 		sock_rps_record_flow(sk);
851 
852 	err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
853 			      sk, msg, size, flags & MSG_DONTWAIT,
854 			      flags & ~MSG_DONTWAIT, &addr_len);
855 	if (err >= 0)
856 		msg->msg_namelen = addr_len;
857 	return err;
858 }
859 EXPORT_SYMBOL(inet_recvmsg);
860 
861 int inet_shutdown(struct socket *sock, int how)
862 {
863 	struct sock *sk = sock->sk;
864 	int err = 0;
865 
866 	/* This should really check to make sure
867 	 * the socket is a TCP socket. (WHY AC...)
868 	 */
869 	how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
870 		       1->2 bit 2 snds.
871 		       2->3 */
872 	if ((how & ~SHUTDOWN_MASK) || !how)	/* MAXINT->0 */
873 		return -EINVAL;
874 
875 	lock_sock(sk);
876 	if (sock->state == SS_CONNECTING) {
877 		if ((1 << sk->sk_state) &
878 		    (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
879 			sock->state = SS_DISCONNECTING;
880 		else
881 			sock->state = SS_CONNECTED;
882 	}
883 
884 	switch (sk->sk_state) {
885 	case TCP_CLOSE:
886 		err = -ENOTCONN;
887 		/* Hack to wake up other listeners, who can poll for
888 		   EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
889 		fallthrough;
890 	default:
891 		sk->sk_shutdown |= how;
892 		if (sk->sk_prot->shutdown)
893 			sk->sk_prot->shutdown(sk, how);
894 		break;
895 
896 	/* Remaining two branches are temporary solution for missing
897 	 * close() in multithreaded environment. It is _not_ a good idea,
898 	 * but we have no choice until close() is repaired at VFS level.
899 	 */
900 	case TCP_LISTEN:
901 		if (!(how & RCV_SHUTDOWN))
902 			break;
903 		fallthrough;
904 	case TCP_SYN_SENT:
905 		err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
906 		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
907 		break;
908 	}
909 
910 	/* Wake up anyone sleeping in poll. */
911 	sk->sk_state_change(sk);
912 	release_sock(sk);
913 	return err;
914 }
915 EXPORT_SYMBOL(inet_shutdown);
916 
917 /*
918  *	ioctl() calls you can issue on an INET socket. Most of these are
919  *	device configuration and stuff and very rarely used. Some ioctls
920  *	pass on to the socket itself.
921  *
922  *	NOTE: I like the idea of a module for the config stuff. ie ifconfig
923  *	loads the devconfigure module does its configuring and unloads it.
924  *	There's a good 20K of config code hanging around the kernel.
925  */
926 
927 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
928 {
929 	struct sock *sk = sock->sk;
930 	int err = 0;
931 	struct net *net = sock_net(sk);
932 	void __user *p = (void __user *)arg;
933 	struct ifreq ifr;
934 	struct rtentry rt;
935 
936 	switch (cmd) {
937 	case SIOCADDRT:
938 	case SIOCDELRT:
939 		if (copy_from_user(&rt, p, sizeof(struct rtentry)))
940 			return -EFAULT;
941 		err = ip_rt_ioctl(net, cmd, &rt);
942 		break;
943 	case SIOCRTMSG:
944 		err = -EINVAL;
945 		break;
946 	case SIOCDARP:
947 	case SIOCGARP:
948 	case SIOCSARP:
949 		err = arp_ioctl(net, cmd, (void __user *)arg);
950 		break;
951 	case SIOCGIFADDR:
952 	case SIOCGIFBRDADDR:
953 	case SIOCGIFNETMASK:
954 	case SIOCGIFDSTADDR:
955 	case SIOCGIFPFLAGS:
956 		if (get_user_ifreq(&ifr, NULL, p))
957 			return -EFAULT;
958 		err = devinet_ioctl(net, cmd, &ifr);
959 		if (!err && put_user_ifreq(&ifr, p))
960 			err = -EFAULT;
961 		break;
962 
963 	case SIOCSIFADDR:
964 	case SIOCSIFBRDADDR:
965 	case SIOCSIFNETMASK:
966 	case SIOCSIFDSTADDR:
967 	case SIOCSIFPFLAGS:
968 	case SIOCSIFFLAGS:
969 		if (get_user_ifreq(&ifr, NULL, p))
970 			return -EFAULT;
971 		err = devinet_ioctl(net, cmd, &ifr);
972 		break;
973 	default:
974 		if (sk->sk_prot->ioctl)
975 			err = sk->sk_prot->ioctl(sk, cmd, arg);
976 		else
977 			err = -ENOIOCTLCMD;
978 		break;
979 	}
980 	return err;
981 }
982 EXPORT_SYMBOL(inet_ioctl);
983 
984 #ifdef CONFIG_COMPAT
985 static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd,
986 		struct compat_rtentry __user *ur)
987 {
988 	compat_uptr_t rtdev;
989 	struct rtentry rt;
990 
991 	if (copy_from_user(&rt.rt_dst, &ur->rt_dst,
992 			3 * sizeof(struct sockaddr)) ||
993 	    get_user(rt.rt_flags, &ur->rt_flags) ||
994 	    get_user(rt.rt_metric, &ur->rt_metric) ||
995 	    get_user(rt.rt_mtu, &ur->rt_mtu) ||
996 	    get_user(rt.rt_window, &ur->rt_window) ||
997 	    get_user(rt.rt_irtt, &ur->rt_irtt) ||
998 	    get_user(rtdev, &ur->rt_dev))
999 		return -EFAULT;
1000 
1001 	rt.rt_dev = compat_ptr(rtdev);
1002 	return ip_rt_ioctl(sock_net(sk), cmd, &rt);
1003 }
1004 
1005 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1006 {
1007 	void __user *argp = compat_ptr(arg);
1008 	struct sock *sk = sock->sk;
1009 
1010 	switch (cmd) {
1011 	case SIOCADDRT:
1012 	case SIOCDELRT:
1013 		return inet_compat_routing_ioctl(sk, cmd, argp);
1014 	default:
1015 		if (!sk->sk_prot->compat_ioctl)
1016 			return -ENOIOCTLCMD;
1017 		return sk->sk_prot->compat_ioctl(sk, cmd, arg);
1018 	}
1019 }
1020 #endif /* CONFIG_COMPAT */
1021 
1022 const struct proto_ops inet_stream_ops = {
1023 	.family		   = PF_INET,
1024 	.owner		   = THIS_MODULE,
1025 	.release	   = inet_release,
1026 	.bind		   = inet_bind,
1027 	.connect	   = inet_stream_connect,
1028 	.socketpair	   = sock_no_socketpair,
1029 	.accept		   = inet_accept,
1030 	.getname	   = inet_getname,
1031 	.poll		   = tcp_poll,
1032 	.ioctl		   = inet_ioctl,
1033 	.gettstamp	   = sock_gettstamp,
1034 	.listen		   = inet_listen,
1035 	.shutdown	   = inet_shutdown,
1036 	.setsockopt	   = sock_common_setsockopt,
1037 	.getsockopt	   = sock_common_getsockopt,
1038 	.sendmsg	   = inet_sendmsg,
1039 	.recvmsg	   = inet_recvmsg,
1040 #ifdef CONFIG_MMU
1041 	.mmap		   = tcp_mmap,
1042 #endif
1043 	.sendpage	   = inet_sendpage,
1044 	.splice_read	   = tcp_splice_read,
1045 	.read_sock	   = tcp_read_sock,
1046 	.sendmsg_locked    = tcp_sendmsg_locked,
1047 	.sendpage_locked   = tcp_sendpage_locked,
1048 	.peek_len	   = tcp_peek_len,
1049 #ifdef CONFIG_COMPAT
1050 	.compat_ioctl	   = inet_compat_ioctl,
1051 #endif
1052 	.set_rcvlowat	   = tcp_set_rcvlowat,
1053 };
1054 EXPORT_SYMBOL(inet_stream_ops);
1055 
1056 const struct proto_ops inet_dgram_ops = {
1057 	.family		   = PF_INET,
1058 	.owner		   = THIS_MODULE,
1059 	.release	   = inet_release,
1060 	.bind		   = inet_bind,
1061 	.connect	   = inet_dgram_connect,
1062 	.socketpair	   = sock_no_socketpair,
1063 	.accept		   = sock_no_accept,
1064 	.getname	   = inet_getname,
1065 	.poll		   = udp_poll,
1066 	.ioctl		   = inet_ioctl,
1067 	.gettstamp	   = sock_gettstamp,
1068 	.listen		   = sock_no_listen,
1069 	.shutdown	   = inet_shutdown,
1070 	.setsockopt	   = sock_common_setsockopt,
1071 	.getsockopt	   = sock_common_getsockopt,
1072 	.sendmsg	   = inet_sendmsg,
1073 	.read_sock	   = udp_read_sock,
1074 	.recvmsg	   = inet_recvmsg,
1075 	.mmap		   = sock_no_mmap,
1076 	.sendpage	   = inet_sendpage,
1077 	.set_peek_off	   = sk_set_peek_off,
1078 #ifdef CONFIG_COMPAT
1079 	.compat_ioctl	   = inet_compat_ioctl,
1080 #endif
1081 };
1082 EXPORT_SYMBOL(inet_dgram_ops);
1083 
1084 /*
1085  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1086  * udp_poll
1087  */
1088 static const struct proto_ops inet_sockraw_ops = {
1089 	.family		   = PF_INET,
1090 	.owner		   = THIS_MODULE,
1091 	.release	   = inet_release,
1092 	.bind		   = inet_bind,
1093 	.connect	   = inet_dgram_connect,
1094 	.socketpair	   = sock_no_socketpair,
1095 	.accept		   = sock_no_accept,
1096 	.getname	   = inet_getname,
1097 	.poll		   = datagram_poll,
1098 	.ioctl		   = inet_ioctl,
1099 	.gettstamp	   = sock_gettstamp,
1100 	.listen		   = sock_no_listen,
1101 	.shutdown	   = inet_shutdown,
1102 	.setsockopt	   = sock_common_setsockopt,
1103 	.getsockopt	   = sock_common_getsockopt,
1104 	.sendmsg	   = inet_sendmsg,
1105 	.recvmsg	   = inet_recvmsg,
1106 	.mmap		   = sock_no_mmap,
1107 	.sendpage	   = inet_sendpage,
1108 #ifdef CONFIG_COMPAT
1109 	.compat_ioctl	   = inet_compat_ioctl,
1110 #endif
1111 };
1112 
1113 static const struct net_proto_family inet_family_ops = {
1114 	.family = PF_INET,
1115 	.create = inet_create,
1116 	.owner	= THIS_MODULE,
1117 };
1118 
1119 /* Upon startup we insert all the elements in inetsw_array[] into
1120  * the linked list inetsw.
1121  */
1122 static struct inet_protosw inetsw_array[] =
1123 {
1124 	{
1125 		.type =       SOCK_STREAM,
1126 		.protocol =   IPPROTO_TCP,
1127 		.prot =       &tcp_prot,
1128 		.ops =        &inet_stream_ops,
1129 		.flags =      INET_PROTOSW_PERMANENT |
1130 			      INET_PROTOSW_ICSK,
1131 	},
1132 
1133 	{
1134 		.type =       SOCK_DGRAM,
1135 		.protocol =   IPPROTO_UDP,
1136 		.prot =       &udp_prot,
1137 		.ops =        &inet_dgram_ops,
1138 		.flags =      INET_PROTOSW_PERMANENT,
1139        },
1140 
1141        {
1142 		.type =       SOCK_DGRAM,
1143 		.protocol =   IPPROTO_ICMP,
1144 		.prot =       &ping_prot,
1145 		.ops =        &inet_sockraw_ops,
1146 		.flags =      INET_PROTOSW_REUSE,
1147        },
1148 
1149        {
1150 	       .type =       SOCK_RAW,
1151 	       .protocol =   IPPROTO_IP,	/* wild card */
1152 	       .prot =       &raw_prot,
1153 	       .ops =        &inet_sockraw_ops,
1154 	       .flags =      INET_PROTOSW_REUSE,
1155        }
1156 };
1157 
1158 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1159 
1160 void inet_register_protosw(struct inet_protosw *p)
1161 {
1162 	struct list_head *lh;
1163 	struct inet_protosw *answer;
1164 	int protocol = p->protocol;
1165 	struct list_head *last_perm;
1166 
1167 	spin_lock_bh(&inetsw_lock);
1168 
1169 	if (p->type >= SOCK_MAX)
1170 		goto out_illegal;
1171 
1172 	/* If we are trying to override a permanent protocol, bail. */
1173 	last_perm = &inetsw[p->type];
1174 	list_for_each(lh, &inetsw[p->type]) {
1175 		answer = list_entry(lh, struct inet_protosw, list);
1176 		/* Check only the non-wild match. */
1177 		if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1178 			break;
1179 		if (protocol == answer->protocol)
1180 			goto out_permanent;
1181 		last_perm = lh;
1182 	}
1183 
1184 	/* Add the new entry after the last permanent entry if any, so that
1185 	 * the new entry does not override a permanent entry when matched with
1186 	 * a wild-card protocol. But it is allowed to override any existing
1187 	 * non-permanent entry.  This means that when we remove this entry, the
1188 	 * system automatically returns to the old behavior.
1189 	 */
1190 	list_add_rcu(&p->list, last_perm);
1191 out:
1192 	spin_unlock_bh(&inetsw_lock);
1193 
1194 	return;
1195 
1196 out_permanent:
1197 	pr_err("Attempt to override permanent protocol %d\n", protocol);
1198 	goto out;
1199 
1200 out_illegal:
1201 	pr_err("Ignoring attempt to register invalid socket type %d\n",
1202 	       p->type);
1203 	goto out;
1204 }
1205 EXPORT_SYMBOL(inet_register_protosw);
1206 
1207 void inet_unregister_protosw(struct inet_protosw *p)
1208 {
1209 	if (INET_PROTOSW_PERMANENT & p->flags) {
1210 		pr_err("Attempt to unregister permanent protocol %d\n",
1211 		       p->protocol);
1212 	} else {
1213 		spin_lock_bh(&inetsw_lock);
1214 		list_del_rcu(&p->list);
1215 		spin_unlock_bh(&inetsw_lock);
1216 
1217 		synchronize_net();
1218 	}
1219 }
1220 EXPORT_SYMBOL(inet_unregister_protosw);
1221 
1222 static int inet_sk_reselect_saddr(struct sock *sk)
1223 {
1224 	struct inet_sock *inet = inet_sk(sk);
1225 	__be32 old_saddr = inet->inet_saddr;
1226 	__be32 daddr = inet->inet_daddr;
1227 	struct flowi4 *fl4;
1228 	struct rtable *rt;
1229 	__be32 new_saddr;
1230 	struct ip_options_rcu *inet_opt;
1231 
1232 	inet_opt = rcu_dereference_protected(inet->inet_opt,
1233 					     lockdep_sock_is_held(sk));
1234 	if (inet_opt && inet_opt->opt.srr)
1235 		daddr = inet_opt->opt.faddr;
1236 
1237 	/* Query new route. */
1238 	fl4 = &inet->cork.fl.u.ip4;
1239 	rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1240 			      sk->sk_bound_dev_if, sk->sk_protocol,
1241 			      inet->inet_sport, inet->inet_dport, sk);
1242 	if (IS_ERR(rt))
1243 		return PTR_ERR(rt);
1244 
1245 	sk_setup_caps(sk, &rt->dst);
1246 
1247 	new_saddr = fl4->saddr;
1248 
1249 	if (new_saddr == old_saddr)
1250 		return 0;
1251 
1252 	if (sock_net(sk)->ipv4.sysctl_ip_dynaddr > 1) {
1253 		pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1254 			__func__, &old_saddr, &new_saddr);
1255 	}
1256 
1257 	inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1258 
1259 	/*
1260 	 * XXX The only one ugly spot where we need to
1261 	 * XXX really change the sockets identity after
1262 	 * XXX it has entered the hashes. -DaveM
1263 	 *
1264 	 * Besides that, it does not check for connection
1265 	 * uniqueness. Wait for troubles.
1266 	 */
1267 	return __sk_prot_rehash(sk);
1268 }
1269 
1270 int inet_sk_rebuild_header(struct sock *sk)
1271 {
1272 	struct inet_sock *inet = inet_sk(sk);
1273 	struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1274 	__be32 daddr;
1275 	struct ip_options_rcu *inet_opt;
1276 	struct flowi4 *fl4;
1277 	int err;
1278 
1279 	/* Route is OK, nothing to do. */
1280 	if (rt)
1281 		return 0;
1282 
1283 	/* Reroute. */
1284 	rcu_read_lock();
1285 	inet_opt = rcu_dereference(inet->inet_opt);
1286 	daddr = inet->inet_daddr;
1287 	if (inet_opt && inet_opt->opt.srr)
1288 		daddr = inet_opt->opt.faddr;
1289 	rcu_read_unlock();
1290 	fl4 = &inet->cork.fl.u.ip4;
1291 	rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1292 				   inet->inet_dport, inet->inet_sport,
1293 				   sk->sk_protocol, RT_CONN_FLAGS(sk),
1294 				   sk->sk_bound_dev_if);
1295 	if (!IS_ERR(rt)) {
1296 		err = 0;
1297 		sk_setup_caps(sk, &rt->dst);
1298 	} else {
1299 		err = PTR_ERR(rt);
1300 
1301 		/* Routing failed... */
1302 		sk->sk_route_caps = 0;
1303 		/*
1304 		 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1305 		 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1306 		 */
1307 		if (!sock_net(sk)->ipv4.sysctl_ip_dynaddr ||
1308 		    sk->sk_state != TCP_SYN_SENT ||
1309 		    (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1310 		    (err = inet_sk_reselect_saddr(sk)) != 0)
1311 			sk->sk_err_soft = -err;
1312 	}
1313 
1314 	return err;
1315 }
1316 EXPORT_SYMBOL(inet_sk_rebuild_header);
1317 
1318 void inet_sk_set_state(struct sock *sk, int state)
1319 {
1320 	trace_inet_sock_set_state(sk, sk->sk_state, state);
1321 	sk->sk_state = state;
1322 }
1323 EXPORT_SYMBOL(inet_sk_set_state);
1324 
1325 void inet_sk_state_store(struct sock *sk, int newstate)
1326 {
1327 	trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1328 	smp_store_release(&sk->sk_state, newstate);
1329 }
1330 
1331 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1332 				 netdev_features_t features)
1333 {
1334 	bool udpfrag = false, fixedid = false, gso_partial, encap;
1335 	struct sk_buff *segs = ERR_PTR(-EINVAL);
1336 	const struct net_offload *ops;
1337 	unsigned int offset = 0;
1338 	struct iphdr *iph;
1339 	int proto, tot_len;
1340 	int nhoff;
1341 	int ihl;
1342 	int id;
1343 
1344 	skb_reset_network_header(skb);
1345 	nhoff = skb_network_header(skb) - skb_mac_header(skb);
1346 	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1347 		goto out;
1348 
1349 	iph = ip_hdr(skb);
1350 	ihl = iph->ihl * 4;
1351 	if (ihl < sizeof(*iph))
1352 		goto out;
1353 
1354 	id = ntohs(iph->id);
1355 	proto = iph->protocol;
1356 
1357 	/* Warning: after this point, iph might be no longer valid */
1358 	if (unlikely(!pskb_may_pull(skb, ihl)))
1359 		goto out;
1360 	__skb_pull(skb, ihl);
1361 
1362 	encap = SKB_GSO_CB(skb)->encap_level > 0;
1363 	if (encap)
1364 		features &= skb->dev->hw_enc_features;
1365 	SKB_GSO_CB(skb)->encap_level += ihl;
1366 
1367 	skb_reset_transport_header(skb);
1368 
1369 	segs = ERR_PTR(-EPROTONOSUPPORT);
1370 
1371 	if (!skb->encapsulation || encap) {
1372 		udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1373 		fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1374 
1375 		/* fixed ID is invalid if DF bit is not set */
1376 		if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1377 			goto out;
1378 	}
1379 
1380 	ops = rcu_dereference(inet_offloads[proto]);
1381 	if (likely(ops && ops->callbacks.gso_segment))
1382 		segs = ops->callbacks.gso_segment(skb, features);
1383 
1384 	if (IS_ERR_OR_NULL(segs))
1385 		goto out;
1386 
1387 	gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1388 
1389 	skb = segs;
1390 	do {
1391 		iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1392 		if (udpfrag) {
1393 			iph->frag_off = htons(offset >> 3);
1394 			if (skb->next)
1395 				iph->frag_off |= htons(IP_MF);
1396 			offset += skb->len - nhoff - ihl;
1397 			tot_len = skb->len - nhoff;
1398 		} else if (skb_is_gso(skb)) {
1399 			if (!fixedid) {
1400 				iph->id = htons(id);
1401 				id += skb_shinfo(skb)->gso_segs;
1402 			}
1403 
1404 			if (gso_partial)
1405 				tot_len = skb_shinfo(skb)->gso_size +
1406 					  SKB_GSO_CB(skb)->data_offset +
1407 					  skb->head - (unsigned char *)iph;
1408 			else
1409 				tot_len = skb->len - nhoff;
1410 		} else {
1411 			if (!fixedid)
1412 				iph->id = htons(id++);
1413 			tot_len = skb->len - nhoff;
1414 		}
1415 		iph->tot_len = htons(tot_len);
1416 		ip_send_check(iph);
1417 		if (encap)
1418 			skb_reset_inner_headers(skb);
1419 		skb->network_header = (u8 *)iph - skb->head;
1420 		skb_reset_mac_len(skb);
1421 	} while ((skb = skb->next));
1422 
1423 out:
1424 	return segs;
1425 }
1426 
1427 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1428 					netdev_features_t features)
1429 {
1430 	if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1431 		return ERR_PTR(-EINVAL);
1432 
1433 	return inet_gso_segment(skb, features);
1434 }
1435 
1436 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1437 {
1438 	const struct net_offload *ops;
1439 	struct sk_buff *pp = NULL;
1440 	const struct iphdr *iph;
1441 	struct sk_buff *p;
1442 	unsigned int hlen;
1443 	unsigned int off;
1444 	unsigned int id;
1445 	int flush = 1;
1446 	int proto;
1447 
1448 	off = skb_gro_offset(skb);
1449 	hlen = off + sizeof(*iph);
1450 	iph = skb_gro_header_fast(skb, off);
1451 	if (skb_gro_header_hard(skb, hlen)) {
1452 		iph = skb_gro_header_slow(skb, hlen, off);
1453 		if (unlikely(!iph))
1454 			goto out;
1455 	}
1456 
1457 	proto = iph->protocol;
1458 
1459 	rcu_read_lock();
1460 	ops = rcu_dereference(inet_offloads[proto]);
1461 	if (!ops || !ops->callbacks.gro_receive)
1462 		goto out_unlock;
1463 
1464 	if (*(u8 *)iph != 0x45)
1465 		goto out_unlock;
1466 
1467 	if (ip_is_fragment(iph))
1468 		goto out_unlock;
1469 
1470 	if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1471 		goto out_unlock;
1472 
1473 	id = ntohl(*(__be32 *)&iph->id);
1474 	flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1475 	id >>= 16;
1476 
1477 	list_for_each_entry(p, head, list) {
1478 		struct iphdr *iph2;
1479 		u16 flush_id;
1480 
1481 		if (!NAPI_GRO_CB(p)->same_flow)
1482 			continue;
1483 
1484 		iph2 = (struct iphdr *)(p->data + off);
1485 		/* The above works because, with the exception of the top
1486 		 * (inner most) layer, we only aggregate pkts with the same
1487 		 * hdr length so all the hdrs we'll need to verify will start
1488 		 * at the same offset.
1489 		 */
1490 		if ((iph->protocol ^ iph2->protocol) |
1491 		    ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1492 		    ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1493 			NAPI_GRO_CB(p)->same_flow = 0;
1494 			continue;
1495 		}
1496 
1497 		/* All fields must match except length and checksum. */
1498 		NAPI_GRO_CB(p)->flush |=
1499 			(iph->ttl ^ iph2->ttl) |
1500 			(iph->tos ^ iph2->tos) |
1501 			((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1502 
1503 		NAPI_GRO_CB(p)->flush |= flush;
1504 
1505 		/* We need to store of the IP ID check to be included later
1506 		 * when we can verify that this packet does in fact belong
1507 		 * to a given flow.
1508 		 */
1509 		flush_id = (u16)(id - ntohs(iph2->id));
1510 
1511 		/* This bit of code makes it much easier for us to identify
1512 		 * the cases where we are doing atomic vs non-atomic IP ID
1513 		 * checks.  Specifically an atomic check can return IP ID
1514 		 * values 0 - 0xFFFF, while a non-atomic check can only
1515 		 * return 0 or 0xFFFF.
1516 		 */
1517 		if (!NAPI_GRO_CB(p)->is_atomic ||
1518 		    !(iph->frag_off & htons(IP_DF))) {
1519 			flush_id ^= NAPI_GRO_CB(p)->count;
1520 			flush_id = flush_id ? 0xFFFF : 0;
1521 		}
1522 
1523 		/* If the previous IP ID value was based on an atomic
1524 		 * datagram we can overwrite the value and ignore it.
1525 		 */
1526 		if (NAPI_GRO_CB(skb)->is_atomic)
1527 			NAPI_GRO_CB(p)->flush_id = flush_id;
1528 		else
1529 			NAPI_GRO_CB(p)->flush_id |= flush_id;
1530 	}
1531 
1532 	NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1533 	NAPI_GRO_CB(skb)->flush |= flush;
1534 	skb_set_network_header(skb, off);
1535 	/* The above will be needed by the transport layer if there is one
1536 	 * immediately following this IP hdr.
1537 	 */
1538 
1539 	/* Note : No need to call skb_gro_postpull_rcsum() here,
1540 	 * as we already checked checksum over ipv4 header was 0
1541 	 */
1542 	skb_gro_pull(skb, sizeof(*iph));
1543 	skb_set_transport_header(skb, skb_gro_offset(skb));
1544 
1545 	pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1546 				       ops->callbacks.gro_receive, head, skb);
1547 
1548 out_unlock:
1549 	rcu_read_unlock();
1550 
1551 out:
1552 	skb_gro_flush_final(skb, pp, flush);
1553 
1554 	return pp;
1555 }
1556 
1557 static struct sk_buff *ipip_gro_receive(struct list_head *head,
1558 					struct sk_buff *skb)
1559 {
1560 	if (NAPI_GRO_CB(skb)->encap_mark) {
1561 		NAPI_GRO_CB(skb)->flush = 1;
1562 		return NULL;
1563 	}
1564 
1565 	NAPI_GRO_CB(skb)->encap_mark = 1;
1566 
1567 	return inet_gro_receive(head, skb);
1568 }
1569 
1570 #define SECONDS_PER_DAY	86400
1571 
1572 /* inet_current_timestamp - Return IP network timestamp
1573  *
1574  * Return milliseconds since midnight in network byte order.
1575  */
1576 __be32 inet_current_timestamp(void)
1577 {
1578 	u32 secs;
1579 	u32 msecs;
1580 	struct timespec64 ts;
1581 
1582 	ktime_get_real_ts64(&ts);
1583 
1584 	/* Get secs since midnight. */
1585 	(void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1586 	/* Convert to msecs. */
1587 	msecs = secs * MSEC_PER_SEC;
1588 	/* Convert nsec to msec. */
1589 	msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1590 
1591 	/* Convert to network byte order. */
1592 	return htonl(msecs);
1593 }
1594 EXPORT_SYMBOL(inet_current_timestamp);
1595 
1596 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1597 {
1598 	if (sk->sk_family == AF_INET)
1599 		return ip_recv_error(sk, msg, len, addr_len);
1600 #if IS_ENABLED(CONFIG_IPV6)
1601 	if (sk->sk_family == AF_INET6)
1602 		return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1603 #endif
1604 	return -EINVAL;
1605 }
1606 
1607 int inet_gro_complete(struct sk_buff *skb, int nhoff)
1608 {
1609 	__be16 newlen = htons(skb->len - nhoff);
1610 	struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1611 	const struct net_offload *ops;
1612 	int proto = iph->protocol;
1613 	int err = -ENOSYS;
1614 
1615 	if (skb->encapsulation) {
1616 		skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1617 		skb_set_inner_network_header(skb, nhoff);
1618 	}
1619 
1620 	csum_replace2(&iph->check, iph->tot_len, newlen);
1621 	iph->tot_len = newlen;
1622 
1623 	rcu_read_lock();
1624 	ops = rcu_dereference(inet_offloads[proto]);
1625 	if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1626 		goto out_unlock;
1627 
1628 	/* Only need to add sizeof(*iph) to get to the next hdr below
1629 	 * because any hdr with option will have been flushed in
1630 	 * inet_gro_receive().
1631 	 */
1632 	err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1633 			      tcp4_gro_complete, udp4_gro_complete,
1634 			      skb, nhoff + sizeof(*iph));
1635 
1636 out_unlock:
1637 	rcu_read_unlock();
1638 
1639 	return err;
1640 }
1641 
1642 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1643 {
1644 	skb->encapsulation = 1;
1645 	skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1646 	return inet_gro_complete(skb, nhoff);
1647 }
1648 
1649 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1650 			 unsigned short type, unsigned char protocol,
1651 			 struct net *net)
1652 {
1653 	struct socket *sock;
1654 	int rc = sock_create_kern(net, family, type, protocol, &sock);
1655 
1656 	if (rc == 0) {
1657 		*sk = sock->sk;
1658 		(*sk)->sk_allocation = GFP_ATOMIC;
1659 		/*
1660 		 * Unhash it so that IP input processing does not even see it,
1661 		 * we do not wish this socket to see incoming packets.
1662 		 */
1663 		(*sk)->sk_prot->unhash(*sk);
1664 	}
1665 	return rc;
1666 }
1667 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1668 
1669 u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt)
1670 {
1671 	return  *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt);
1672 }
1673 EXPORT_SYMBOL_GPL(snmp_get_cpu_field);
1674 
1675 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1676 {
1677 	unsigned long res = 0;
1678 	int i;
1679 
1680 	for_each_possible_cpu(i)
1681 		res += snmp_get_cpu_field(mib, i, offt);
1682 	return res;
1683 }
1684 EXPORT_SYMBOL_GPL(snmp_fold_field);
1685 
1686 #if BITS_PER_LONG==32
1687 
1688 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1689 			 size_t syncp_offset)
1690 {
1691 	void *bhptr;
1692 	struct u64_stats_sync *syncp;
1693 	u64 v;
1694 	unsigned int start;
1695 
1696 	bhptr = per_cpu_ptr(mib, cpu);
1697 	syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1698 	do {
1699 		start = u64_stats_fetch_begin_irq(syncp);
1700 		v = *(((u64 *)bhptr) + offt);
1701 	} while (u64_stats_fetch_retry_irq(syncp, start));
1702 
1703 	return v;
1704 }
1705 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1706 
1707 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1708 {
1709 	u64 res = 0;
1710 	int cpu;
1711 
1712 	for_each_possible_cpu(cpu) {
1713 		res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1714 	}
1715 	return res;
1716 }
1717 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1718 #endif
1719 
1720 #ifdef CONFIG_IP_MULTICAST
1721 static const struct net_protocol igmp_protocol = {
1722 	.handler =	igmp_rcv,
1723 };
1724 #endif
1725 
1726 /* thinking of making this const? Don't.
1727  * early_demux can change based on sysctl.
1728  */
1729 static struct net_protocol tcp_protocol = {
1730 	.early_demux	=	tcp_v4_early_demux,
1731 	.early_demux_handler =  tcp_v4_early_demux,
1732 	.handler	=	tcp_v4_rcv,
1733 	.err_handler	=	tcp_v4_err,
1734 	.no_policy	=	1,
1735 	.icmp_strict_tag_validation = 1,
1736 };
1737 
1738 /* thinking of making this const? Don't.
1739  * early_demux can change based on sysctl.
1740  */
1741 static struct net_protocol udp_protocol = {
1742 	.early_demux =	udp_v4_early_demux,
1743 	.early_demux_handler =	udp_v4_early_demux,
1744 	.handler =	udp_rcv,
1745 	.err_handler =	udp_err,
1746 	.no_policy =	1,
1747 };
1748 
1749 static const struct net_protocol icmp_protocol = {
1750 	.handler =	icmp_rcv,
1751 	.err_handler =	icmp_err,
1752 	.no_policy =	1,
1753 };
1754 
1755 static __net_init int ipv4_mib_init_net(struct net *net)
1756 {
1757 	int i;
1758 
1759 	net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1760 	if (!net->mib.tcp_statistics)
1761 		goto err_tcp_mib;
1762 	net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1763 	if (!net->mib.ip_statistics)
1764 		goto err_ip_mib;
1765 
1766 	for_each_possible_cpu(i) {
1767 		struct ipstats_mib *af_inet_stats;
1768 		af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1769 		u64_stats_init(&af_inet_stats->syncp);
1770 	}
1771 
1772 	net->mib.net_statistics = alloc_percpu(struct linux_mib);
1773 	if (!net->mib.net_statistics)
1774 		goto err_net_mib;
1775 	net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1776 	if (!net->mib.udp_statistics)
1777 		goto err_udp_mib;
1778 	net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1779 	if (!net->mib.udplite_statistics)
1780 		goto err_udplite_mib;
1781 	net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1782 	if (!net->mib.icmp_statistics)
1783 		goto err_icmp_mib;
1784 	net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1785 					      GFP_KERNEL);
1786 	if (!net->mib.icmpmsg_statistics)
1787 		goto err_icmpmsg_mib;
1788 
1789 	tcp_mib_init(net);
1790 	return 0;
1791 
1792 err_icmpmsg_mib:
1793 	free_percpu(net->mib.icmp_statistics);
1794 err_icmp_mib:
1795 	free_percpu(net->mib.udplite_statistics);
1796 err_udplite_mib:
1797 	free_percpu(net->mib.udp_statistics);
1798 err_udp_mib:
1799 	free_percpu(net->mib.net_statistics);
1800 err_net_mib:
1801 	free_percpu(net->mib.ip_statistics);
1802 err_ip_mib:
1803 	free_percpu(net->mib.tcp_statistics);
1804 err_tcp_mib:
1805 	return -ENOMEM;
1806 }
1807 
1808 static __net_exit void ipv4_mib_exit_net(struct net *net)
1809 {
1810 	kfree(net->mib.icmpmsg_statistics);
1811 	free_percpu(net->mib.icmp_statistics);
1812 	free_percpu(net->mib.udplite_statistics);
1813 	free_percpu(net->mib.udp_statistics);
1814 	free_percpu(net->mib.net_statistics);
1815 	free_percpu(net->mib.ip_statistics);
1816 	free_percpu(net->mib.tcp_statistics);
1817 #ifdef CONFIG_MPTCP
1818 	/* allocated on demand, see mptcp_init_sock() */
1819 	free_percpu(net->mib.mptcp_statistics);
1820 #endif
1821 }
1822 
1823 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1824 	.init = ipv4_mib_init_net,
1825 	.exit = ipv4_mib_exit_net,
1826 };
1827 
1828 static int __init init_ipv4_mibs(void)
1829 {
1830 	return register_pernet_subsys(&ipv4_mib_ops);
1831 }
1832 
1833 static __net_init int inet_init_net(struct net *net)
1834 {
1835 	/*
1836 	 * Set defaults for local port range
1837 	 */
1838 	seqlock_init(&net->ipv4.ip_local_ports.lock);
1839 	net->ipv4.ip_local_ports.range[0] =  32768;
1840 	net->ipv4.ip_local_ports.range[1] =  60999;
1841 
1842 	seqlock_init(&net->ipv4.ping_group_range.lock);
1843 	/*
1844 	 * Sane defaults - nobody may create ping sockets.
1845 	 * Boot scripts should set this to distro-specific group.
1846 	 */
1847 	net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1848 	net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1849 
1850 	/* Default values for sysctl-controlled parameters.
1851 	 * We set them here, in case sysctl is not compiled.
1852 	 */
1853 	net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1854 	net->ipv4.sysctl_ip_fwd_update_priority = 1;
1855 	net->ipv4.sysctl_ip_dynaddr = 0;
1856 	net->ipv4.sysctl_ip_early_demux = 1;
1857 	net->ipv4.sysctl_udp_early_demux = 1;
1858 	net->ipv4.sysctl_tcp_early_demux = 1;
1859 	net->ipv4.sysctl_nexthop_compat_mode = 1;
1860 #ifdef CONFIG_SYSCTL
1861 	net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1862 #endif
1863 
1864 	/* Some igmp sysctl, whose values are always used */
1865 	net->ipv4.sysctl_igmp_max_memberships = 20;
1866 	net->ipv4.sysctl_igmp_max_msf = 10;
1867 	/* IGMP reports for link-local multicast groups are enabled by default */
1868 	net->ipv4.sysctl_igmp_llm_reports = 1;
1869 	net->ipv4.sysctl_igmp_qrv = 2;
1870 
1871 	net->ipv4.sysctl_fib_notify_on_flag_change = 0;
1872 
1873 	return 0;
1874 }
1875 
1876 static __net_initdata struct pernet_operations af_inet_ops = {
1877 	.init = inet_init_net,
1878 };
1879 
1880 static int __init init_inet_pernet_ops(void)
1881 {
1882 	return register_pernet_subsys(&af_inet_ops);
1883 }
1884 
1885 static int ipv4_proc_init(void);
1886 
1887 /*
1888  *	IP protocol layer initialiser
1889  */
1890 
1891 static struct packet_offload ip_packet_offload __read_mostly = {
1892 	.type = cpu_to_be16(ETH_P_IP),
1893 	.callbacks = {
1894 		.gso_segment = inet_gso_segment,
1895 		.gro_receive = inet_gro_receive,
1896 		.gro_complete = inet_gro_complete,
1897 	},
1898 };
1899 
1900 static const struct net_offload ipip_offload = {
1901 	.callbacks = {
1902 		.gso_segment	= ipip_gso_segment,
1903 		.gro_receive	= ipip_gro_receive,
1904 		.gro_complete	= ipip_gro_complete,
1905 	},
1906 };
1907 
1908 static int __init ipip_offload_init(void)
1909 {
1910 	return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1911 }
1912 
1913 static int __init ipv4_offload_init(void)
1914 {
1915 	/*
1916 	 * Add offloads
1917 	 */
1918 	if (udpv4_offload_init() < 0)
1919 		pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1920 	if (tcpv4_offload_init() < 0)
1921 		pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1922 	if (ipip_offload_init() < 0)
1923 		pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1924 
1925 	dev_add_offload(&ip_packet_offload);
1926 	return 0;
1927 }
1928 
1929 fs_initcall(ipv4_offload_init);
1930 
1931 static struct packet_type ip_packet_type __read_mostly = {
1932 	.type = cpu_to_be16(ETH_P_IP),
1933 	.func = ip_rcv,
1934 	.list_func = ip_list_rcv,
1935 };
1936 
1937 static int __init inet_init(void)
1938 {
1939 	struct inet_protosw *q;
1940 	struct list_head *r;
1941 	int rc;
1942 
1943 	sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1944 
1945 	rc = proto_register(&tcp_prot, 1);
1946 	if (rc)
1947 		goto out;
1948 
1949 	rc = proto_register(&udp_prot, 1);
1950 	if (rc)
1951 		goto out_unregister_tcp_proto;
1952 
1953 	rc = proto_register(&raw_prot, 1);
1954 	if (rc)
1955 		goto out_unregister_udp_proto;
1956 
1957 	rc = proto_register(&ping_prot, 1);
1958 	if (rc)
1959 		goto out_unregister_raw_proto;
1960 
1961 	/*
1962 	 *	Tell SOCKET that we are alive...
1963 	 */
1964 
1965 	(void)sock_register(&inet_family_ops);
1966 
1967 #ifdef CONFIG_SYSCTL
1968 	ip_static_sysctl_init();
1969 #endif
1970 
1971 	/*
1972 	 *	Add all the base protocols.
1973 	 */
1974 
1975 	if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1976 		pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1977 	if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1978 		pr_crit("%s: Cannot add UDP protocol\n", __func__);
1979 	if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1980 		pr_crit("%s: Cannot add TCP protocol\n", __func__);
1981 #ifdef CONFIG_IP_MULTICAST
1982 	if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1983 		pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1984 #endif
1985 
1986 	/* Register the socket-side information for inet_create. */
1987 	for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1988 		INIT_LIST_HEAD(r);
1989 
1990 	for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1991 		inet_register_protosw(q);
1992 
1993 	/*
1994 	 *	Set the ARP module up
1995 	 */
1996 
1997 	arp_init();
1998 
1999 	/*
2000 	 *	Set the IP module up
2001 	 */
2002 
2003 	ip_init();
2004 
2005 	/* Setup TCP slab cache for open requests. */
2006 	tcp_init();
2007 
2008 	/* Setup UDP memory threshold */
2009 	udp_init();
2010 
2011 	/* Add UDP-Lite (RFC 3828) */
2012 	udplite4_register();
2013 
2014 	raw_init();
2015 
2016 	ping_init();
2017 
2018 	/*
2019 	 *	Set the ICMP layer up
2020 	 */
2021 
2022 	if (icmp_init() < 0)
2023 		panic("Failed to create the ICMP control socket.\n");
2024 
2025 	/*
2026 	 *	Initialise the multicast router
2027 	 */
2028 #if defined(CONFIG_IP_MROUTE)
2029 	if (ip_mr_init())
2030 		pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2031 #endif
2032 
2033 	if (init_inet_pernet_ops())
2034 		pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2035 	/*
2036 	 *	Initialise per-cpu ipv4 mibs
2037 	 */
2038 
2039 	if (init_ipv4_mibs())
2040 		pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
2041 
2042 	ipv4_proc_init();
2043 
2044 	ipfrag_init();
2045 
2046 	dev_add_pack(&ip_packet_type);
2047 
2048 	ip_tunnel_core_init();
2049 
2050 	rc = 0;
2051 out:
2052 	return rc;
2053 out_unregister_raw_proto:
2054 	proto_unregister(&raw_prot);
2055 out_unregister_udp_proto:
2056 	proto_unregister(&udp_prot);
2057 out_unregister_tcp_proto:
2058 	proto_unregister(&tcp_prot);
2059 	goto out;
2060 }
2061 
2062 fs_initcall(inet_init);
2063 
2064 /* ------------------------------------------------------------------------ */
2065 
2066 #ifdef CONFIG_PROC_FS
2067 static int __init ipv4_proc_init(void)
2068 {
2069 	int rc = 0;
2070 
2071 	if (raw_proc_init())
2072 		goto out_raw;
2073 	if (tcp4_proc_init())
2074 		goto out_tcp;
2075 	if (udp4_proc_init())
2076 		goto out_udp;
2077 	if (ping_proc_init())
2078 		goto out_ping;
2079 	if (ip_misc_proc_init())
2080 		goto out_misc;
2081 out:
2082 	return rc;
2083 out_misc:
2084 	ping_proc_exit();
2085 out_ping:
2086 	udp4_proc_exit();
2087 out_udp:
2088 	tcp4_proc_exit();
2089 out_tcp:
2090 	raw_proc_exit();
2091 out_raw:
2092 	rc = -ENOMEM;
2093 	goto out;
2094 }
2095 
2096 #else /* CONFIG_PROC_FS */
2097 static int __init ipv4_proc_init(void)
2098 {
2099 	return 0;
2100 }
2101 #endif /* CONFIG_PROC_FS */
2102