xref: /linux/net/ipv4/af_inet.c (revision 1bc80d673087e5704adbb3ee8e4b785c14899cce)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		PF_INET protocol family socket handler.
8  *
9  * Authors:	Ross Biro
10  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11  *		Florian La Roche, <flla@stud.uni-sb.de>
12  *		Alan Cox, <A.Cox@swansea.ac.uk>
13  *
14  * Changes (see also sock.c)
15  *
16  *		piggy,
17  *		Karl Knutson	:	Socket protocol table
18  *		A.N.Kuznetsov	:	Socket death error in accept().
19  *		John Richardson :	Fix non blocking error in connect()
20  *					so sockets that fail to connect
21  *					don't return -EINPROGRESS.
22  *		Alan Cox	:	Asynchronous I/O support
23  *		Alan Cox	:	Keep correct socket pointer on sock
24  *					structures
25  *					when accept() ed
26  *		Alan Cox	:	Semantics of SO_LINGER aren't state
27  *					moved to close when you look carefully.
28  *					With this fixed and the accept bug fixed
29  *					some RPC stuff seems happier.
30  *		Niibe Yutaka	:	4.4BSD style write async I/O
31  *		Alan Cox,
32  *		Tony Gale 	:	Fixed reuse semantics.
33  *		Alan Cox	:	bind() shouldn't abort existing but dead
34  *					sockets. Stops FTP netin:.. I hope.
35  *		Alan Cox	:	bind() works correctly for RAW sockets.
36  *					Note that FreeBSD at least was broken
37  *					in this respect so be careful with
38  *					compatibility tests...
39  *		Alan Cox	:	routing cache support
40  *		Alan Cox	:	memzero the socket structure for
41  *					compactness.
42  *		Matt Day	:	nonblock connect error handler
43  *		Alan Cox	:	Allow large numbers of pending sockets
44  *					(eg for big web sites), but only if
45  *					specifically application requested.
46  *		Alan Cox	:	New buffering throughout IP. Used
47  *					dumbly.
48  *		Alan Cox	:	New buffering now used smartly.
49  *		Alan Cox	:	BSD rather than common sense
50  *					interpretation of listen.
51  *		Germano Caronni	:	Assorted small races.
52  *		Alan Cox	:	sendmsg/recvmsg basic support.
53  *		Alan Cox	:	Only sendmsg/recvmsg now supported.
54  *		Alan Cox	:	Locked down bind (see security list).
55  *		Alan Cox	:	Loosened bind a little.
56  *		Mike McLagan	:	ADD/DEL DLCI Ioctls
57  *	Willy Konynenberg	:	Transparent proxying support.
58  *		David S. Miller	:	New socket lookup architecture.
59  *					Some other random speedups.
60  *		Cyrus Durgin	:	Cleaned up file for kmod hacks.
61  *		Andi Kleen	:	Fix inet_stream_connect TCP race.
62  */
63 
64 #define pr_fmt(fmt) "IPv4: " fmt
65 
66 #include <linux/err.h>
67 #include <linux/errno.h>
68 #include <linux/types.h>
69 #include <linux/socket.h>
70 #include <linux/in.h>
71 #include <linux/kernel.h>
72 #include <linux/kmod.h>
73 #include <linux/sched.h>
74 #include <linux/timer.h>
75 #include <linux/string.h>
76 #include <linux/sockios.h>
77 #include <linux/net.h>
78 #include <linux/capability.h>
79 #include <linux/fcntl.h>
80 #include <linux/mm.h>
81 #include <linux/interrupt.h>
82 #include <linux/stat.h>
83 #include <linux/init.h>
84 #include <linux/poll.h>
85 #include <linux/netfilter_ipv4.h>
86 #include <linux/random.h>
87 #include <linux/slab.h>
88 
89 #include <linux/uaccess.h>
90 
91 #include <linux/inet.h>
92 #include <linux/igmp.h>
93 #include <linux/inetdevice.h>
94 #include <linux/netdevice.h>
95 #include <net/checksum.h>
96 #include <net/ip.h>
97 #include <net/protocol.h>
98 #include <net/arp.h>
99 #include <net/route.h>
100 #include <net/ip_fib.h>
101 #include <net/inet_connection_sock.h>
102 #include <net/gro.h>
103 #include <net/gso.h>
104 #include <net/tcp.h>
105 #include <net/psp.h>
106 #include <net/udp.h>
107 #include <net/udplite.h>
108 #include <net/ping.h>
109 #include <linux/skbuff.h>
110 #include <net/sock.h>
111 #include <net/raw.h>
112 #include <net/icmp.h>
113 #include <net/inet_common.h>
114 #include <net/ip_tunnels.h>
115 #include <net/xfrm.h>
116 #include <net/net_namespace.h>
117 #include <net/secure_seq.h>
118 #ifdef CONFIG_IP_MROUTE
119 #include <linux/mroute.h>
120 #endif
121 #include <net/l3mdev.h>
122 #include <net/compat.h>
123 #include <net/rps.h>
124 
125 #include <trace/events/sock.h>
126 
127 /* The inetsw table contains everything that inet_create needs to
128  * build a new socket.
129  */
130 static struct list_head inetsw[SOCK_MAX];
131 static DEFINE_SPINLOCK(inetsw_lock);
132 
133 /* New destruction routine */
134 
135 void inet_sock_destruct(struct sock *sk)
136 {
137 	struct inet_sock *inet = inet_sk(sk);
138 
139 	__skb_queue_purge(&sk->sk_receive_queue);
140 	__skb_queue_purge(&sk->sk_error_queue);
141 
142 	sk_mem_reclaim_final(sk);
143 
144 	if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
145 		pr_err("Attempt to release TCP socket in state %d %p\n",
146 		       sk->sk_state, sk);
147 		return;
148 	}
149 	if (!sock_flag(sk, SOCK_DEAD)) {
150 		pr_err("Attempt to release alive inet socket %p\n", sk);
151 		return;
152 	}
153 
154 	WARN_ON_ONCE(atomic_read(&sk->sk_rmem_alloc));
155 	WARN_ON_ONCE(refcount_read(&sk->sk_wmem_alloc));
156 	WARN_ON_ONCE(sk->sk_wmem_queued);
157 	WARN_ON_ONCE(sk->sk_forward_alloc);
158 
159 	kfree(rcu_dereference_protected(inet->inet_opt, 1));
160 	dst_release(rcu_dereference_protected(sk->sk_dst_cache, 1));
161 	dst_release(rcu_dereference_protected(sk->sk_rx_dst, 1));
162 	psp_sk_assoc_free(sk);
163 }
164 EXPORT_SYMBOL(inet_sock_destruct);
165 
166 /*
167  *	The routines beyond this point handle the behaviour of an AF_INET
168  *	socket object. Mostly it punts to the subprotocols of IP to do
169  *	the work.
170  */
171 
172 /*
173  *	Automatically bind an unbound socket.
174  */
175 
176 static int inet_autobind(struct sock *sk)
177 {
178 	struct inet_sock *inet;
179 	/* We may need to bind the socket. */
180 	lock_sock(sk);
181 	inet = inet_sk(sk);
182 	if (!inet->inet_num) {
183 		if (sk->sk_prot->get_port(sk, 0)) {
184 			release_sock(sk);
185 			return -EAGAIN;
186 		}
187 		inet->inet_sport = htons(inet->inet_num);
188 	}
189 	release_sock(sk);
190 	return 0;
191 }
192 
193 int __inet_listen_sk(struct sock *sk, int backlog)
194 {
195 	unsigned char old_state = sk->sk_state;
196 	int err, tcp_fastopen;
197 
198 	if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
199 		return -EINVAL;
200 
201 	WRITE_ONCE(sk->sk_max_ack_backlog, backlog);
202 	/* Really, if the socket is already in listen state
203 	 * we can only allow the backlog to be adjusted.
204 	 */
205 	if (old_state != TCP_LISTEN) {
206 		/* Enable TFO w/o requiring TCP_FASTOPEN socket option.
207 		 * Note that only TCP sockets (SOCK_STREAM) will reach here.
208 		 * Also fastopen backlog may already been set via the option
209 		 * because the socket was in TCP_LISTEN state previously but
210 		 * was shutdown() rather than close().
211 		 */
212 		tcp_fastopen = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fastopen);
213 		if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
214 		    (tcp_fastopen & TFO_SERVER_ENABLE) &&
215 		    !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
216 			fastopen_queue_tune(sk, backlog);
217 			tcp_fastopen_init_key_once(sock_net(sk));
218 		}
219 
220 		err = inet_csk_listen_start(sk);
221 		if (err)
222 			return err;
223 
224 		tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL);
225 	}
226 	return 0;
227 }
228 
229 /*
230  *	Move a socket into listening state.
231  */
232 int inet_listen(struct socket *sock, int backlog)
233 {
234 	struct sock *sk = sock->sk;
235 	int err = -EINVAL;
236 
237 	lock_sock(sk);
238 
239 	if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
240 		goto out;
241 
242 	err = __inet_listen_sk(sk, backlog);
243 
244 out:
245 	release_sock(sk);
246 	return err;
247 }
248 EXPORT_SYMBOL(inet_listen);
249 
250 /*
251  *	Create an inet socket.
252  */
253 
254 static int inet_create(struct net *net, struct socket *sock, int protocol,
255 		       int kern)
256 {
257 	struct sock *sk;
258 	struct inet_protosw *answer;
259 	struct inet_sock *inet;
260 	struct proto *answer_prot;
261 	unsigned char answer_flags;
262 	int try_loading_module = 0;
263 	int err;
264 
265 	if (protocol < 0 || protocol >= IPPROTO_MAX)
266 		return -EINVAL;
267 
268 	sock->state = SS_UNCONNECTED;
269 
270 	/* Look for the requested type/protocol pair. */
271 lookup_protocol:
272 	err = -ESOCKTNOSUPPORT;
273 	rcu_read_lock();
274 	list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
275 
276 		err = 0;
277 		/* Check the non-wild match. */
278 		if (protocol == answer->protocol) {
279 			if (protocol != IPPROTO_IP)
280 				break;
281 		} else {
282 			/* Check for the two wild cases. */
283 			if (IPPROTO_IP == protocol) {
284 				protocol = answer->protocol;
285 				break;
286 			}
287 			if (IPPROTO_IP == answer->protocol)
288 				break;
289 		}
290 		err = -EPROTONOSUPPORT;
291 	}
292 
293 	if (unlikely(err)) {
294 		if (try_loading_module < 2) {
295 			rcu_read_unlock();
296 			/*
297 			 * Be more specific, e.g. net-pf-2-proto-132-type-1
298 			 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
299 			 */
300 			if (++try_loading_module == 1)
301 				request_module("net-pf-%d-proto-%d-type-%d",
302 					       PF_INET, protocol, sock->type);
303 			/*
304 			 * Fall back to generic, e.g. net-pf-2-proto-132
305 			 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
306 			 */
307 			else
308 				request_module("net-pf-%d-proto-%d",
309 					       PF_INET, protocol);
310 			goto lookup_protocol;
311 		} else
312 			goto out_rcu_unlock;
313 	}
314 
315 	err = -EPERM;
316 	if (sock->type == SOCK_RAW && !kern &&
317 	    !ns_capable(net->user_ns, CAP_NET_RAW))
318 		goto out_rcu_unlock;
319 
320 	sock->ops = answer->ops;
321 	answer_prot = answer->prot;
322 	answer_flags = answer->flags;
323 	rcu_read_unlock();
324 
325 	WARN_ON(!answer_prot->slab);
326 
327 	err = -ENOMEM;
328 	sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
329 	if (!sk)
330 		goto out;
331 
332 	err = 0;
333 	if (INET_PROTOSW_REUSE & answer_flags)
334 		sk->sk_reuse = SK_CAN_REUSE;
335 
336 	if (INET_PROTOSW_ICSK & answer_flags)
337 		inet_init_csk_locks(sk);
338 
339 	inet = inet_sk(sk);
340 	inet_assign_bit(IS_ICSK, sk, INET_PROTOSW_ICSK & answer_flags);
341 
342 	inet_clear_bit(NODEFRAG, sk);
343 
344 	if (SOCK_RAW == sock->type) {
345 		inet->inet_num = protocol;
346 		if (IPPROTO_RAW == protocol)
347 			inet_set_bit(HDRINCL, sk);
348 	}
349 
350 	if (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc))
351 		inet->pmtudisc = IP_PMTUDISC_DONT;
352 	else
353 		inet->pmtudisc = IP_PMTUDISC_WANT;
354 
355 	atomic_set(&inet->inet_id, 0);
356 
357 	sock_init_data(sock, sk);
358 
359 	sk->sk_destruct	   = inet_sock_destruct;
360 	sk->sk_protocol	   = protocol;
361 	sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
362 	sk->sk_txrehash = READ_ONCE(net->core.sysctl_txrehash);
363 
364 	inet->uc_ttl	= -1;
365 	inet_set_bit(MC_LOOP, sk);
366 	inet->mc_ttl	= 1;
367 	inet_set_bit(MC_ALL, sk);
368 	inet->mc_index	= 0;
369 	inet->mc_list	= NULL;
370 	inet->rcv_tos	= 0;
371 
372 	if (inet->inet_num) {
373 		/* It assumes that any protocol which allows
374 		 * the user to assign a number at socket
375 		 * creation time automatically
376 		 * shares.
377 		 */
378 		inet->inet_sport = htons(inet->inet_num);
379 		/* Add to protocol hash chains. */
380 		err = sk->sk_prot->hash(sk);
381 		if (err)
382 			goto out_sk_release;
383 	}
384 
385 	if (sk->sk_prot->init) {
386 		err = sk->sk_prot->init(sk);
387 		if (err)
388 			goto out_sk_release;
389 	}
390 
391 	if (!kern) {
392 		err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
393 		if (err)
394 			goto out_sk_release;
395 	}
396 out:
397 	return err;
398 out_rcu_unlock:
399 	rcu_read_unlock();
400 	goto out;
401 out_sk_release:
402 	sk_common_release(sk);
403 	sock->sk = NULL;
404 	goto out;
405 }
406 
407 
408 /*
409  *	The peer socket should always be NULL (or else). When we call this
410  *	function we are destroying the object and from then on nobody
411  *	should refer to it.
412  */
413 int inet_release(struct socket *sock)
414 {
415 	struct sock *sk = sock->sk;
416 
417 	if (sk) {
418 		long timeout;
419 
420 		if (!sk->sk_kern_sock)
421 			BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk);
422 
423 		/* Applications forget to leave groups before exiting */
424 		ip_mc_drop_socket(sk);
425 
426 		/* If linger is set, we don't return until the close
427 		 * is complete.  Otherwise we return immediately. The
428 		 * actually closing is done the same either way.
429 		 *
430 		 * If the close is due to the process exiting, we never
431 		 * linger..
432 		 */
433 		timeout = 0;
434 		if (sock_flag(sk, SOCK_LINGER) &&
435 		    !(current->flags & PF_EXITING))
436 			timeout = sk->sk_lingertime;
437 		sk->sk_prot->close(sk, timeout);
438 		sock->sk = NULL;
439 	}
440 	return 0;
441 }
442 EXPORT_SYMBOL(inet_release);
443 
444 int inet_bind_sk(struct sock *sk, struct sockaddr *uaddr, int addr_len)
445 {
446 	u32 flags = BIND_WITH_LOCK;
447 	int err;
448 
449 	/* If the socket has its own bind function then use it. (RAW) */
450 	if (sk->sk_prot->bind) {
451 		return sk->sk_prot->bind(sk, uaddr, addr_len);
452 	}
453 	if (addr_len < sizeof(struct sockaddr_in))
454 		return -EINVAL;
455 
456 	/* BPF prog is run before any checks are done so that if the prog
457 	 * changes context in a wrong way it will be caught.
458 	 */
459 	err = BPF_CGROUP_RUN_PROG_INET_BIND_LOCK(sk, uaddr, &addr_len,
460 						 CGROUP_INET4_BIND, &flags);
461 	if (err)
462 		return err;
463 
464 	return __inet_bind(sk, uaddr, addr_len, flags);
465 }
466 
467 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
468 {
469 	return inet_bind_sk(sock->sk, uaddr, addr_len);
470 }
471 EXPORT_SYMBOL(inet_bind);
472 
473 int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
474 		u32 flags)
475 {
476 	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
477 	struct inet_sock *inet = inet_sk(sk);
478 	struct net *net = sock_net(sk);
479 	unsigned short snum;
480 	int chk_addr_ret;
481 	u32 tb_id = RT_TABLE_LOCAL;
482 	int err;
483 
484 	if (addr->sin_family != AF_INET) {
485 		/* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
486 		 * only if s_addr is INADDR_ANY.
487 		 */
488 		err = -EAFNOSUPPORT;
489 		if (addr->sin_family != AF_UNSPEC ||
490 		    addr->sin_addr.s_addr != htonl(INADDR_ANY))
491 			goto out;
492 	}
493 
494 	tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
495 	chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
496 
497 	/* Not specified by any standard per-se, however it breaks too
498 	 * many applications when removed.  It is unfortunate since
499 	 * allowing applications to make a non-local bind solves
500 	 * several problems with systems using dynamic addressing.
501 	 * (ie. your servers still start up even if your ISDN link
502 	 *  is temporarily down)
503 	 */
504 	err = -EADDRNOTAVAIL;
505 	if (!inet_addr_valid_or_nonlocal(net, inet, addr->sin_addr.s_addr,
506 	                                 chk_addr_ret))
507 		goto out;
508 
509 	snum = ntohs(addr->sin_port);
510 	err = -EACCES;
511 	if (!(flags & BIND_NO_CAP_NET_BIND_SERVICE) &&
512 	    snum && inet_port_requires_bind_service(net, snum) &&
513 	    !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
514 		goto out;
515 
516 	/*      We keep a pair of addresses. rcv_saddr is the one
517 	 *      used by hash lookups, and saddr is used for transmit.
518 	 *
519 	 *      In the BSD API these are the same except where it
520 	 *      would be illegal to use them (multicast/broadcast) in
521 	 *      which case the sending device address is used.
522 	 */
523 	if (flags & BIND_WITH_LOCK)
524 		lock_sock(sk);
525 
526 	/* Check these errors (active socket, double bind). */
527 	err = -EINVAL;
528 	if (sk->sk_state != TCP_CLOSE || inet->inet_num)
529 		goto out_release_sock;
530 
531 	inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
532 	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
533 		inet->inet_saddr = 0;  /* Use device */
534 
535 	/* Make sure we are allowed to bind here. */
536 	if (snum || !(inet_test_bit(BIND_ADDRESS_NO_PORT, sk) ||
537 		      (flags & BIND_FORCE_ADDRESS_NO_PORT))) {
538 		err = sk->sk_prot->get_port(sk, snum);
539 		if (err) {
540 			inet->inet_saddr = inet->inet_rcv_saddr = 0;
541 			goto out_release_sock;
542 		}
543 		if (!(flags & BIND_FROM_BPF)) {
544 			err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
545 			if (err) {
546 				inet->inet_saddr = inet->inet_rcv_saddr = 0;
547 				if (sk->sk_prot->put_port)
548 					sk->sk_prot->put_port(sk);
549 				goto out_release_sock;
550 			}
551 		}
552 	}
553 
554 	if (inet->inet_rcv_saddr)
555 		sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
556 	if (snum)
557 		sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
558 	inet->inet_sport = htons(inet->inet_num);
559 	inet->inet_daddr = 0;
560 	inet->inet_dport = 0;
561 	sk_dst_reset(sk);
562 	err = 0;
563 out_release_sock:
564 	if (flags & BIND_WITH_LOCK)
565 		release_sock(sk);
566 out:
567 	return err;
568 }
569 
570 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
571 		       int addr_len, int flags)
572 {
573 	struct sock *sk = sock->sk;
574 	const struct proto *prot;
575 	int err;
576 
577 	if (addr_len < sizeof(uaddr->sa_family))
578 		return -EINVAL;
579 
580 	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
581 	prot = READ_ONCE(sk->sk_prot);
582 
583 	if (uaddr->sa_family == AF_UNSPEC)
584 		return prot->disconnect(sk, flags);
585 
586 	if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
587 		err = prot->pre_connect(sk, uaddr, addr_len);
588 		if (err)
589 			return err;
590 	}
591 
592 	if (data_race(!inet_sk(sk)->inet_num) && inet_autobind(sk))
593 		return -EAGAIN;
594 	return prot->connect(sk, uaddr, addr_len);
595 }
596 EXPORT_SYMBOL(inet_dgram_connect);
597 
598 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
599 {
600 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
601 
602 	add_wait_queue(sk_sleep(sk), &wait);
603 	sk->sk_write_pending += writebias;
604 
605 	/* Basic assumption: if someone sets sk->sk_err, he _must_
606 	 * change state of the socket from TCP_SYN_*.
607 	 * Connect() does not allow to get error notifications
608 	 * without closing the socket.
609 	 */
610 	while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
611 		release_sock(sk);
612 		timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
613 		lock_sock(sk);
614 		if (signal_pending(current) || !timeo)
615 			break;
616 	}
617 	remove_wait_queue(sk_sleep(sk), &wait);
618 	sk->sk_write_pending -= writebias;
619 	return timeo;
620 }
621 
622 /*
623  *	Connect to a remote host. There is regrettably still a little
624  *	TCP 'magic' in here.
625  */
626 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
627 			  int addr_len, int flags, int is_sendmsg)
628 {
629 	struct sock *sk = sock->sk;
630 	int err;
631 	long timeo;
632 
633 	/*
634 	 * uaddr can be NULL and addr_len can be 0 if:
635 	 * sk is a TCP fastopen active socket and
636 	 * TCP_FASTOPEN_CONNECT sockopt is set and
637 	 * we already have a valid cookie for this socket.
638 	 * In this case, user can call write() after connect().
639 	 * write() will invoke tcp_sendmsg_fastopen() which calls
640 	 * __inet_stream_connect().
641 	 */
642 	if (uaddr) {
643 		if (addr_len < sizeof(uaddr->sa_family))
644 			return -EINVAL;
645 
646 		if (uaddr->sa_family == AF_UNSPEC) {
647 			sk->sk_disconnects++;
648 			err = sk->sk_prot->disconnect(sk, flags);
649 			sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
650 			goto out;
651 		}
652 	}
653 
654 	switch (sock->state) {
655 	default:
656 		err = -EINVAL;
657 		goto out;
658 	case SS_CONNECTED:
659 		err = -EISCONN;
660 		goto out;
661 	case SS_CONNECTING:
662 		if (inet_test_bit(DEFER_CONNECT, sk))
663 			err = is_sendmsg ? -EINPROGRESS : -EISCONN;
664 		else
665 			err = -EALREADY;
666 		/* Fall out of switch with err, set for this state */
667 		break;
668 	case SS_UNCONNECTED:
669 		err = -EISCONN;
670 		if (sk->sk_state != TCP_CLOSE)
671 			goto out;
672 
673 		if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
674 			err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
675 			if (err)
676 				goto out;
677 		}
678 
679 		err = sk->sk_prot->connect(sk, uaddr, addr_len);
680 		if (err < 0)
681 			goto out;
682 
683 		sock->state = SS_CONNECTING;
684 
685 		if (!err && inet_test_bit(DEFER_CONNECT, sk))
686 			goto out;
687 
688 		/* Just entered SS_CONNECTING state; the only
689 		 * difference is that return value in non-blocking
690 		 * case is EINPROGRESS, rather than EALREADY.
691 		 */
692 		err = -EINPROGRESS;
693 		break;
694 	}
695 
696 	timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
697 
698 	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
699 		int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
700 				tcp_sk(sk)->fastopen_req &&
701 				tcp_sk(sk)->fastopen_req->data ? 1 : 0;
702 		int dis = sk->sk_disconnects;
703 
704 		/* Error code is set above */
705 		if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
706 			goto out;
707 
708 		err = sock_intr_errno(timeo);
709 		if (signal_pending(current))
710 			goto out;
711 
712 		if (dis != sk->sk_disconnects) {
713 			err = -EPIPE;
714 			goto out;
715 		}
716 	}
717 
718 	/* Connection was closed by RST, timeout, ICMP error
719 	 * or another process disconnected us.
720 	 */
721 	if (sk->sk_state == TCP_CLOSE)
722 		goto sock_error;
723 
724 	/* sk->sk_err may be not zero now, if RECVERR was ordered by user
725 	 * and error was received after socket entered established state.
726 	 * Hence, it is handled normally after connect() return successfully.
727 	 */
728 
729 	sock->state = SS_CONNECTED;
730 	err = 0;
731 out:
732 	return err;
733 
734 sock_error:
735 	err = sock_error(sk) ? : -ECONNABORTED;
736 	sock->state = SS_UNCONNECTED;
737 	sk->sk_disconnects++;
738 	if (sk->sk_prot->disconnect(sk, flags))
739 		sock->state = SS_DISCONNECTING;
740 	goto out;
741 }
742 EXPORT_SYMBOL(__inet_stream_connect);
743 
744 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
745 			int addr_len, int flags)
746 {
747 	int err;
748 
749 	lock_sock(sock->sk);
750 	err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
751 	release_sock(sock->sk);
752 	return err;
753 }
754 EXPORT_SYMBOL(inet_stream_connect);
755 
756 void __inet_accept(struct socket *sock, struct socket *newsock, struct sock *newsk)
757 {
758 	/* TODO: use sk_clone_lock() in SCTP and remove protocol checks */
759 	if (mem_cgroup_sockets_enabled &&
760 	    (!IS_ENABLED(CONFIG_IP_SCTP) || sk_is_tcp(newsk))) {
761 		gfp_t gfp = GFP_KERNEL | __GFP_NOFAIL;
762 
763 		mem_cgroup_sk_alloc(newsk);
764 
765 		if (mem_cgroup_from_sk(newsk)) {
766 			int amt;
767 
768 			/* The socket has not been accepted yet, no need
769 			 * to look at newsk->sk_wmem_queued.
770 			 */
771 			amt = sk_mem_pages(newsk->sk_forward_alloc +
772 					   atomic_read(&newsk->sk_rmem_alloc));
773 			if (amt)
774 				mem_cgroup_sk_charge(newsk, amt, gfp);
775 		}
776 
777 		kmem_cache_charge(newsk, gfp);
778 	}
779 
780 	sock_rps_record_flow(newsk);
781 	WARN_ON(!((1 << newsk->sk_state) &
782 		  (TCPF_ESTABLISHED | TCPF_SYN_RECV |
783 		   TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2 |
784 		   TCPF_CLOSING | TCPF_CLOSE_WAIT |
785 		   TCPF_CLOSE)));
786 
787 	if (test_bit(SOCK_SUPPORT_ZC, &sock->flags))
788 		set_bit(SOCK_SUPPORT_ZC, &newsock->flags);
789 	sock_graft(newsk, newsock);
790 
791 	newsock->state = SS_CONNECTED;
792 }
793 
794 /*
795  *	Accept a pending connection. The TCP layer now gives BSD semantics.
796  */
797 
798 int inet_accept(struct socket *sock, struct socket *newsock,
799 		struct proto_accept_arg *arg)
800 {
801 	struct sock *sk1 = sock->sk, *sk2;
802 
803 	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
804 	arg->err = -EINVAL;
805 	sk2 = READ_ONCE(sk1->sk_prot)->accept(sk1, arg);
806 	if (!sk2)
807 		return arg->err;
808 
809 	lock_sock(sk2);
810 	__inet_accept(sock, newsock, sk2);
811 	release_sock(sk2);
812 	return 0;
813 }
814 EXPORT_SYMBOL(inet_accept);
815 
816 /*
817  *	This does both peername and sockname.
818  */
819 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
820 		 int peer)
821 {
822 	struct sock *sk		= sock->sk;
823 	struct inet_sock *inet	= inet_sk(sk);
824 	DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
825 	int sin_addr_len = sizeof(*sin);
826 
827 	sin->sin_family = AF_INET;
828 	lock_sock(sk);
829 	if (peer) {
830 		if (!inet->inet_dport ||
831 		    (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
832 		     peer == 1)) {
833 			release_sock(sk);
834 			return -ENOTCONN;
835 		}
836 		sin->sin_port = inet->inet_dport;
837 		sin->sin_addr.s_addr = inet->inet_daddr;
838 		BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin, &sin_addr_len,
839 				       CGROUP_INET4_GETPEERNAME);
840 	} else {
841 		__be32 addr = inet->inet_rcv_saddr;
842 		if (!addr)
843 			addr = inet->inet_saddr;
844 		sin->sin_port = inet->inet_sport;
845 		sin->sin_addr.s_addr = addr;
846 		BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin, &sin_addr_len,
847 				       CGROUP_INET4_GETSOCKNAME);
848 	}
849 	release_sock(sk);
850 	memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
851 	return sin_addr_len;
852 }
853 EXPORT_SYMBOL(inet_getname);
854 
855 int inet_send_prepare(struct sock *sk)
856 {
857 	sock_rps_record_flow(sk);
858 
859 	/* We may need to bind the socket. */
860 	if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind &&
861 	    inet_autobind(sk))
862 		return -EAGAIN;
863 
864 	return 0;
865 }
866 EXPORT_SYMBOL_GPL(inet_send_prepare);
867 
868 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
869 {
870 	struct sock *sk = sock->sk;
871 
872 	if (unlikely(inet_send_prepare(sk)))
873 		return -EAGAIN;
874 
875 	return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
876 			       sk, msg, size);
877 }
878 EXPORT_SYMBOL(inet_sendmsg);
879 
880 void inet_splice_eof(struct socket *sock)
881 {
882 	const struct proto *prot;
883 	struct sock *sk = sock->sk;
884 
885 	if (unlikely(inet_send_prepare(sk)))
886 		return;
887 
888 	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
889 	prot = READ_ONCE(sk->sk_prot);
890 	if (prot->splice_eof)
891 		prot->splice_eof(sock);
892 }
893 EXPORT_SYMBOL_GPL(inet_splice_eof);
894 
895 INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
896 					  size_t, int, int *));
897 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
898 		 int flags)
899 {
900 	struct sock *sk = sock->sk;
901 	int addr_len = 0;
902 	int err;
903 
904 	if (likely(!(flags & MSG_ERRQUEUE)))
905 		sock_rps_record_flow(sk);
906 
907 	err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
908 			      sk, msg, size, flags, &addr_len);
909 	if (err >= 0)
910 		msg->msg_namelen = addr_len;
911 	return err;
912 }
913 EXPORT_SYMBOL(inet_recvmsg);
914 
915 int inet_shutdown(struct socket *sock, int how)
916 {
917 	struct sock *sk = sock->sk;
918 	int err = 0;
919 
920 	/* This should really check to make sure
921 	 * the socket is a TCP socket. (WHY AC...)
922 	 */
923 	how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
924 		       1->2 bit 2 snds.
925 		       2->3 */
926 	if ((how & ~SHUTDOWN_MASK) || !how)	/* MAXINT->0 */
927 		return -EINVAL;
928 
929 	lock_sock(sk);
930 	if (sock->state == SS_CONNECTING) {
931 		if ((1 << sk->sk_state) &
932 		    (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
933 			sock->state = SS_DISCONNECTING;
934 		else
935 			sock->state = SS_CONNECTED;
936 	}
937 
938 	switch (sk->sk_state) {
939 	case TCP_CLOSE:
940 		err = -ENOTCONN;
941 		/* Hack to wake up other listeners, who can poll for
942 		   EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
943 		fallthrough;
944 	default:
945 		WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | how);
946 		if (sk->sk_prot->shutdown)
947 			sk->sk_prot->shutdown(sk, how);
948 		break;
949 
950 	/* Remaining two branches are temporary solution for missing
951 	 * close() in multithreaded environment. It is _not_ a good idea,
952 	 * but we have no choice until close() is repaired at VFS level.
953 	 */
954 	case TCP_LISTEN:
955 		if (!(how & RCV_SHUTDOWN))
956 			break;
957 		fallthrough;
958 	case TCP_SYN_SENT:
959 		err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
960 		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
961 		break;
962 	}
963 
964 	/* Wake up anyone sleeping in poll. */
965 	sk->sk_state_change(sk);
966 	release_sock(sk);
967 	return err;
968 }
969 EXPORT_SYMBOL(inet_shutdown);
970 
971 /*
972  *	ioctl() calls you can issue on an INET socket. Most of these are
973  *	device configuration and stuff and very rarely used. Some ioctls
974  *	pass on to the socket itself.
975  *
976  *	NOTE: I like the idea of a module for the config stuff. ie ifconfig
977  *	loads the devconfigure module does its configuring and unloads it.
978  *	There's a good 20K of config code hanging around the kernel.
979  */
980 
981 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
982 {
983 	struct sock *sk = sock->sk;
984 	int err = 0;
985 	struct net *net = sock_net(sk);
986 	void __user *p = (void __user *)arg;
987 	struct ifreq ifr;
988 	struct rtentry rt;
989 
990 	switch (cmd) {
991 	case SIOCADDRT:
992 	case SIOCDELRT:
993 		if (copy_from_user(&rt, p, sizeof(struct rtentry)))
994 			return -EFAULT;
995 		err = ip_rt_ioctl(net, cmd, &rt);
996 		break;
997 	case SIOCRTMSG:
998 		err = -EINVAL;
999 		break;
1000 	case SIOCDARP:
1001 	case SIOCGARP:
1002 	case SIOCSARP:
1003 		err = arp_ioctl(net, cmd, (void __user *)arg);
1004 		break;
1005 	case SIOCGIFADDR:
1006 	case SIOCGIFBRDADDR:
1007 	case SIOCGIFNETMASK:
1008 	case SIOCGIFDSTADDR:
1009 	case SIOCGIFPFLAGS:
1010 		if (get_user_ifreq(&ifr, NULL, p))
1011 			return -EFAULT;
1012 		err = devinet_ioctl(net, cmd, &ifr);
1013 		if (!err && put_user_ifreq(&ifr, p))
1014 			err = -EFAULT;
1015 		break;
1016 
1017 	case SIOCSIFADDR:
1018 	case SIOCSIFBRDADDR:
1019 	case SIOCSIFNETMASK:
1020 	case SIOCSIFDSTADDR:
1021 	case SIOCSIFPFLAGS:
1022 	case SIOCSIFFLAGS:
1023 		if (get_user_ifreq(&ifr, NULL, p))
1024 			return -EFAULT;
1025 		err = devinet_ioctl(net, cmd, &ifr);
1026 		break;
1027 	default:
1028 		if (sk->sk_prot->ioctl)
1029 			err = sk_ioctl(sk, cmd, (void __user *)arg);
1030 		else
1031 			err = -ENOIOCTLCMD;
1032 		break;
1033 	}
1034 	return err;
1035 }
1036 EXPORT_SYMBOL(inet_ioctl);
1037 
1038 #ifdef CONFIG_COMPAT
1039 static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd,
1040 		struct compat_rtentry __user *ur)
1041 {
1042 	compat_uptr_t rtdev;
1043 	struct rtentry rt;
1044 
1045 	if (copy_from_user(&rt.rt_dst, &ur->rt_dst,
1046 			3 * sizeof(struct sockaddr)) ||
1047 	    get_user(rt.rt_flags, &ur->rt_flags) ||
1048 	    get_user(rt.rt_metric, &ur->rt_metric) ||
1049 	    get_user(rt.rt_mtu, &ur->rt_mtu) ||
1050 	    get_user(rt.rt_window, &ur->rt_window) ||
1051 	    get_user(rt.rt_irtt, &ur->rt_irtt) ||
1052 	    get_user(rtdev, &ur->rt_dev))
1053 		return -EFAULT;
1054 
1055 	rt.rt_dev = compat_ptr(rtdev);
1056 	return ip_rt_ioctl(sock_net(sk), cmd, &rt);
1057 }
1058 
1059 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1060 {
1061 	void __user *argp = compat_ptr(arg);
1062 	struct sock *sk = sock->sk;
1063 
1064 	switch (cmd) {
1065 	case SIOCADDRT:
1066 	case SIOCDELRT:
1067 		return inet_compat_routing_ioctl(sk, cmd, argp);
1068 	default:
1069 		if (!sk->sk_prot->compat_ioctl)
1070 			return -ENOIOCTLCMD;
1071 		return sk->sk_prot->compat_ioctl(sk, cmd, arg);
1072 	}
1073 }
1074 #endif /* CONFIG_COMPAT */
1075 
1076 const struct proto_ops inet_stream_ops = {
1077 	.family		   = PF_INET,
1078 	.owner		   = THIS_MODULE,
1079 	.release	   = inet_release,
1080 	.bind		   = inet_bind,
1081 	.connect	   = inet_stream_connect,
1082 	.socketpair	   = sock_no_socketpair,
1083 	.accept		   = inet_accept,
1084 	.getname	   = inet_getname,
1085 	.poll		   = tcp_poll,
1086 	.ioctl		   = inet_ioctl,
1087 	.gettstamp	   = sock_gettstamp,
1088 	.listen		   = inet_listen,
1089 	.shutdown	   = inet_shutdown,
1090 	.setsockopt	   = sock_common_setsockopt,
1091 	.getsockopt	   = sock_common_getsockopt,
1092 	.sendmsg	   = inet_sendmsg,
1093 	.recvmsg	   = inet_recvmsg,
1094 #ifdef CONFIG_MMU
1095 	.mmap		   = tcp_mmap,
1096 #endif
1097 	.splice_eof	   = inet_splice_eof,
1098 	.splice_read	   = tcp_splice_read,
1099 	.set_peek_off      = sk_set_peek_off,
1100 	.read_sock	   = tcp_read_sock,
1101 	.read_skb	   = tcp_read_skb,
1102 	.sendmsg_locked    = tcp_sendmsg_locked,
1103 	.peek_len	   = tcp_peek_len,
1104 #ifdef CONFIG_COMPAT
1105 	.compat_ioctl	   = inet_compat_ioctl,
1106 #endif
1107 	.set_rcvlowat	   = tcp_set_rcvlowat,
1108 };
1109 EXPORT_SYMBOL(inet_stream_ops);
1110 
1111 const struct proto_ops inet_dgram_ops = {
1112 	.family		   = PF_INET,
1113 	.owner		   = THIS_MODULE,
1114 	.release	   = inet_release,
1115 	.bind		   = inet_bind,
1116 	.connect	   = inet_dgram_connect,
1117 	.socketpair	   = sock_no_socketpair,
1118 	.accept		   = sock_no_accept,
1119 	.getname	   = inet_getname,
1120 	.poll		   = udp_poll,
1121 	.ioctl		   = inet_ioctl,
1122 	.gettstamp	   = sock_gettstamp,
1123 	.listen		   = sock_no_listen,
1124 	.shutdown	   = inet_shutdown,
1125 	.setsockopt	   = sock_common_setsockopt,
1126 	.getsockopt	   = sock_common_getsockopt,
1127 	.sendmsg	   = inet_sendmsg,
1128 	.read_skb	   = udp_read_skb,
1129 	.recvmsg	   = inet_recvmsg,
1130 	.mmap		   = sock_no_mmap,
1131 	.splice_eof	   = inet_splice_eof,
1132 	.set_peek_off	   = udp_set_peek_off,
1133 #ifdef CONFIG_COMPAT
1134 	.compat_ioctl	   = inet_compat_ioctl,
1135 #endif
1136 };
1137 EXPORT_SYMBOL(inet_dgram_ops);
1138 
1139 /*
1140  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1141  * udp_poll
1142  */
1143 static const struct proto_ops inet_sockraw_ops = {
1144 	.family		   = PF_INET,
1145 	.owner		   = THIS_MODULE,
1146 	.release	   = inet_release,
1147 	.bind		   = inet_bind,
1148 	.connect	   = inet_dgram_connect,
1149 	.socketpair	   = sock_no_socketpair,
1150 	.accept		   = sock_no_accept,
1151 	.getname	   = inet_getname,
1152 	.poll		   = datagram_poll,
1153 	.ioctl		   = inet_ioctl,
1154 	.gettstamp	   = sock_gettstamp,
1155 	.listen		   = sock_no_listen,
1156 	.shutdown	   = inet_shutdown,
1157 	.setsockopt	   = sock_common_setsockopt,
1158 	.getsockopt	   = sock_common_getsockopt,
1159 	.sendmsg	   = inet_sendmsg,
1160 	.recvmsg	   = inet_recvmsg,
1161 	.mmap		   = sock_no_mmap,
1162 	.splice_eof	   = inet_splice_eof,
1163 #ifdef CONFIG_COMPAT
1164 	.compat_ioctl	   = inet_compat_ioctl,
1165 #endif
1166 };
1167 
1168 static const struct net_proto_family inet_family_ops = {
1169 	.family = PF_INET,
1170 	.create = inet_create,
1171 	.owner	= THIS_MODULE,
1172 };
1173 
1174 /* Upon startup we insert all the elements in inetsw_array[] into
1175  * the linked list inetsw.
1176  */
1177 static struct inet_protosw inetsw_array[] =
1178 {
1179 	{
1180 		.type =       SOCK_STREAM,
1181 		.protocol =   IPPROTO_TCP,
1182 		.prot =       &tcp_prot,
1183 		.ops =        &inet_stream_ops,
1184 		.flags =      INET_PROTOSW_PERMANENT |
1185 			      INET_PROTOSW_ICSK,
1186 	},
1187 
1188 	{
1189 		.type =       SOCK_DGRAM,
1190 		.protocol =   IPPROTO_UDP,
1191 		.prot =       &udp_prot,
1192 		.ops =        &inet_dgram_ops,
1193 		.flags =      INET_PROTOSW_PERMANENT,
1194        },
1195 
1196        {
1197 		.type =       SOCK_DGRAM,
1198 		.protocol =   IPPROTO_ICMP,
1199 		.prot =       &ping_prot,
1200 		.ops =        &inet_sockraw_ops,
1201 		.flags =      INET_PROTOSW_REUSE,
1202        },
1203 
1204        {
1205 	       .type =       SOCK_RAW,
1206 	       .protocol =   IPPROTO_IP,	/* wild card */
1207 	       .prot =       &raw_prot,
1208 	       .ops =        &inet_sockraw_ops,
1209 	       .flags =      INET_PROTOSW_REUSE,
1210        }
1211 };
1212 
1213 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1214 
1215 void inet_register_protosw(struct inet_protosw *p)
1216 {
1217 	struct list_head *lh;
1218 	struct inet_protosw *answer;
1219 	int protocol = p->protocol;
1220 	struct list_head *last_perm;
1221 
1222 	spin_lock_bh(&inetsw_lock);
1223 
1224 	if (p->type >= SOCK_MAX)
1225 		goto out_illegal;
1226 
1227 	/* If we are trying to override a permanent protocol, bail. */
1228 	last_perm = &inetsw[p->type];
1229 	list_for_each(lh, &inetsw[p->type]) {
1230 		answer = list_entry(lh, struct inet_protosw, list);
1231 		/* Check only the non-wild match. */
1232 		if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1233 			break;
1234 		if (protocol == answer->protocol)
1235 			goto out_permanent;
1236 		last_perm = lh;
1237 	}
1238 
1239 	/* Add the new entry after the last permanent entry if any, so that
1240 	 * the new entry does not override a permanent entry when matched with
1241 	 * a wild-card protocol. But it is allowed to override any existing
1242 	 * non-permanent entry.  This means that when we remove this entry, the
1243 	 * system automatically returns to the old behavior.
1244 	 */
1245 	list_add_rcu(&p->list, last_perm);
1246 out:
1247 	spin_unlock_bh(&inetsw_lock);
1248 
1249 	return;
1250 
1251 out_permanent:
1252 	pr_err("Attempt to override permanent protocol %d\n", protocol);
1253 	goto out;
1254 
1255 out_illegal:
1256 	pr_err("Ignoring attempt to register invalid socket type %d\n",
1257 	       p->type);
1258 	goto out;
1259 }
1260 EXPORT_SYMBOL(inet_register_protosw);
1261 
1262 void inet_unregister_protosw(struct inet_protosw *p)
1263 {
1264 	if (INET_PROTOSW_PERMANENT & p->flags) {
1265 		pr_err("Attempt to unregister permanent protocol %d\n",
1266 		       p->protocol);
1267 	} else {
1268 		spin_lock_bh(&inetsw_lock);
1269 		list_del_rcu(&p->list);
1270 		spin_unlock_bh(&inetsw_lock);
1271 
1272 		synchronize_net();
1273 	}
1274 }
1275 EXPORT_SYMBOL(inet_unregister_protosw);
1276 
1277 static int inet_sk_reselect_saddr(struct sock *sk)
1278 {
1279 	struct inet_sock *inet = inet_sk(sk);
1280 	__be32 old_saddr = inet->inet_saddr;
1281 	__be32 daddr = inet->inet_daddr;
1282 	struct flowi4 *fl4;
1283 	struct rtable *rt;
1284 	__be32 new_saddr;
1285 	struct ip_options_rcu *inet_opt;
1286 	int err;
1287 
1288 	inet_opt = rcu_dereference_protected(inet->inet_opt,
1289 					     lockdep_sock_is_held(sk));
1290 	if (inet_opt && inet_opt->opt.srr)
1291 		daddr = inet_opt->opt.faddr;
1292 
1293 	/* Query new route. */
1294 	fl4 = &inet->cork.fl.u.ip4;
1295 	rt = ip_route_connect(fl4, daddr, 0, sk->sk_bound_dev_if,
1296 			      sk->sk_protocol, inet->inet_sport,
1297 			      inet->inet_dport, sk);
1298 	if (IS_ERR(rt))
1299 		return PTR_ERR(rt);
1300 
1301 	new_saddr = fl4->saddr;
1302 
1303 	if (new_saddr == old_saddr) {
1304 		sk_setup_caps(sk, &rt->dst);
1305 		return 0;
1306 	}
1307 
1308 	err = inet_bhash2_update_saddr(sk, &new_saddr, AF_INET);
1309 	if (err) {
1310 		ip_rt_put(rt);
1311 		return err;
1312 	}
1313 
1314 	sk_setup_caps(sk, &rt->dst);
1315 
1316 	if (READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) > 1) {
1317 		pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1318 			__func__, &old_saddr, &new_saddr);
1319 	}
1320 
1321 	/*
1322 	 * XXX The only one ugly spot where we need to
1323 	 * XXX really change the sockets identity after
1324 	 * XXX it has entered the hashes. -DaveM
1325 	 *
1326 	 * Besides that, it does not check for connection
1327 	 * uniqueness. Wait for troubles.
1328 	 */
1329 	return __sk_prot_rehash(sk);
1330 }
1331 
1332 int inet_sk_rebuild_header(struct sock *sk)
1333 {
1334 	struct rtable *rt = dst_rtable(__sk_dst_check(sk, 0));
1335 	struct inet_sock *inet = inet_sk(sk);
1336 	struct flowi4 *fl4;
1337 	int err;
1338 
1339 	/* Route is OK, nothing to do. */
1340 	if (rt)
1341 		return 0;
1342 
1343 	/* Reroute. */
1344 	fl4 = &inet->cork.fl.u.ip4;
1345 	inet_sk_init_flowi4(inet, fl4);
1346 	rt = ip_route_output_flow(sock_net(sk), fl4, sk);
1347 	if (!IS_ERR(rt)) {
1348 		err = 0;
1349 		sk_setup_caps(sk, &rt->dst);
1350 	} else {
1351 		err = PTR_ERR(rt);
1352 
1353 		/* Routing failed... */
1354 		sk->sk_route_caps = 0;
1355 
1356 		if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) ||
1357 		    sk->sk_state != TCP_SYN_SENT ||
1358 		    (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1359 		    (err = inet_sk_reselect_saddr(sk)) != 0)
1360 			WRITE_ONCE(sk->sk_err_soft, -err);
1361 	}
1362 
1363 	return err;
1364 }
1365 EXPORT_SYMBOL(inet_sk_rebuild_header);
1366 
1367 void inet_sk_set_state(struct sock *sk, int state)
1368 {
1369 	trace_inet_sock_set_state(sk, sk->sk_state, state);
1370 	sk->sk_state = state;
1371 }
1372 EXPORT_SYMBOL(inet_sk_set_state);
1373 
1374 void inet_sk_state_store(struct sock *sk, int newstate)
1375 {
1376 	trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1377 	smp_store_release(&sk->sk_state, newstate);
1378 }
1379 
1380 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1381 				 netdev_features_t features)
1382 {
1383 	bool udpfrag = false, fixedid = false, gso_partial, encap;
1384 	struct sk_buff *segs = ERR_PTR(-EINVAL);
1385 	const struct net_offload *ops;
1386 	unsigned int offset = 0;
1387 	struct iphdr *iph;
1388 	int proto, tot_len;
1389 	int nhoff;
1390 	int ihl;
1391 	int id;
1392 
1393 	skb_reset_network_header(skb);
1394 	nhoff = skb_network_header(skb) - skb_mac_header(skb);
1395 	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1396 		goto out;
1397 
1398 	iph = ip_hdr(skb);
1399 	ihl = iph->ihl * 4;
1400 	if (ihl < sizeof(*iph))
1401 		goto out;
1402 
1403 	id = ntohs(iph->id);
1404 	proto = iph->protocol;
1405 
1406 	/* Warning: after this point, iph might be no longer valid */
1407 	if (unlikely(!pskb_may_pull(skb, ihl)))
1408 		goto out;
1409 	__skb_pull(skb, ihl);
1410 
1411 	encap = SKB_GSO_CB(skb)->encap_level > 0;
1412 	if (encap)
1413 		features &= skb->dev->hw_enc_features;
1414 	SKB_GSO_CB(skb)->encap_level += ihl;
1415 
1416 	skb_reset_transport_header(skb);
1417 
1418 	segs = ERR_PTR(-EPROTONOSUPPORT);
1419 
1420 	fixedid = !!(skb_shinfo(skb)->gso_type & (SKB_GSO_TCP_FIXEDID << encap));
1421 
1422 	if (!skb->encapsulation || encap)
1423 		udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1424 
1425 	ops = rcu_dereference(inet_offloads[proto]);
1426 	if (likely(ops && ops->callbacks.gso_segment)) {
1427 		segs = ops->callbacks.gso_segment(skb, features);
1428 		if (!segs)
1429 			skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
1430 	}
1431 
1432 	if (IS_ERR_OR_NULL(segs))
1433 		goto out;
1434 
1435 	gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1436 
1437 	skb = segs;
1438 	do {
1439 		iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1440 		if (udpfrag) {
1441 			iph->frag_off = htons(offset >> 3);
1442 			if (skb->next)
1443 				iph->frag_off |= htons(IP_MF);
1444 			offset += skb->len - nhoff - ihl;
1445 			tot_len = skb->len - nhoff;
1446 		} else if (skb_is_gso(skb)) {
1447 			if (!fixedid) {
1448 				iph->id = htons(id);
1449 				id += skb_shinfo(skb)->gso_segs;
1450 			}
1451 
1452 			if (gso_partial)
1453 				tot_len = skb_shinfo(skb)->gso_size +
1454 					  SKB_GSO_CB(skb)->data_offset +
1455 					  skb->head - (unsigned char *)iph;
1456 			else
1457 				tot_len = skb->len - nhoff;
1458 		} else {
1459 			if (!fixedid)
1460 				iph->id = htons(id++);
1461 			tot_len = skb->len - nhoff;
1462 		}
1463 		iph->tot_len = htons(tot_len);
1464 		ip_send_check(iph);
1465 		if (encap)
1466 			skb_reset_inner_headers(skb);
1467 		skb->network_header = (u8 *)iph - skb->head;
1468 		skb_reset_mac_len(skb);
1469 	} while ((skb = skb->next));
1470 
1471 out:
1472 	return segs;
1473 }
1474 
1475 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1476 					netdev_features_t features)
1477 {
1478 	if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1479 		return ERR_PTR(-EINVAL);
1480 
1481 	return inet_gso_segment(skb, features);
1482 }
1483 
1484 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1485 {
1486 	const struct net_offload *ops;
1487 	struct sk_buff *pp = NULL;
1488 	const struct iphdr *iph;
1489 	struct sk_buff *p;
1490 	unsigned int hlen;
1491 	unsigned int off;
1492 	int flush = 1;
1493 	int proto;
1494 
1495 	off = skb_gro_offset(skb);
1496 	hlen = off + sizeof(*iph);
1497 	iph = skb_gro_header(skb, hlen, off);
1498 	if (unlikely(!iph))
1499 		goto out;
1500 
1501 	proto = iph->protocol;
1502 
1503 	ops = rcu_dereference(inet_offloads[proto]);
1504 	if (!ops || !ops->callbacks.gro_receive)
1505 		goto out;
1506 
1507 	if (*(u8 *)iph != 0x45)
1508 		goto out;
1509 
1510 	if (ip_is_fragment(iph))
1511 		goto out;
1512 
1513 	if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1514 		goto out;
1515 
1516 	NAPI_GRO_CB(skb)->proto = proto;
1517 	flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (ntohl(*(__be32 *)&iph->id) & ~IP_DF));
1518 
1519 	list_for_each_entry(p, head, list) {
1520 		struct iphdr *iph2;
1521 
1522 		if (!NAPI_GRO_CB(p)->same_flow)
1523 			continue;
1524 
1525 		iph2 = (struct iphdr *)(p->data + off);
1526 		/* The above works because, with the exception of the top
1527 		 * (inner most) layer, we only aggregate pkts with the same
1528 		 * hdr length so all the hdrs we'll need to verify will start
1529 		 * at the same offset.
1530 		 */
1531 		if ((iph->protocol ^ iph2->protocol) |
1532 		    ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1533 		    ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1534 			NAPI_GRO_CB(p)->same_flow = 0;
1535 			continue;
1536 		}
1537 	}
1538 
1539 	NAPI_GRO_CB(skb)->flush |= flush;
1540 	NAPI_GRO_CB(skb)->network_offsets[NAPI_GRO_CB(skb)->encap_mark] = off;
1541 
1542 	/* Note : No need to call skb_gro_postpull_rcsum() here,
1543 	 * as we already checked checksum over ipv4 header was 0
1544 	 */
1545 	skb_gro_pull(skb, sizeof(*iph));
1546 	skb_set_transport_header(skb, skb_gro_offset(skb));
1547 
1548 	pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1549 				       ops->callbacks.gro_receive, head, skb);
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 	unsigned int family = READ_ONCE(sk->sk_family);
1599 
1600 	if (family == AF_INET)
1601 		return ip_recv_error(sk, msg, len, addr_len);
1602 #if IS_ENABLED(CONFIG_IPV6)
1603 	if (family == AF_INET6)
1604 		return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1605 #endif
1606 	return -EINVAL;
1607 }
1608 EXPORT_SYMBOL(inet_recv_error);
1609 
1610 int inet_gro_complete(struct sk_buff *skb, int nhoff)
1611 {
1612 	struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1613 	const struct net_offload *ops;
1614 	__be16 totlen = iph->tot_len;
1615 	int proto = iph->protocol;
1616 	int err = -ENOSYS;
1617 
1618 	if (skb->encapsulation) {
1619 		skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1620 		skb_set_inner_network_header(skb, nhoff);
1621 	}
1622 
1623 	iph_set_totlen(iph, skb->len - nhoff);
1624 	csum_replace2(&iph->check, totlen, iph->tot_len);
1625 
1626 	ops = rcu_dereference(inet_offloads[proto]);
1627 	if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1628 		goto out;
1629 
1630 	/* Only need to add sizeof(*iph) to get to the next hdr below
1631 	 * because any hdr with option will have been flushed in
1632 	 * inet_gro_receive().
1633 	 */
1634 	err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1635 			      tcp4_gro_complete, udp4_gro_complete,
1636 			      skb, nhoff + sizeof(*iph));
1637 
1638 out:
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 		(*sk)->sk_use_task_frag = false;
1660 		/*
1661 		 * Unhash it so that IP input processing does not even see it,
1662 		 * we do not wish this socket to see incoming packets.
1663 		 */
1664 		(*sk)->sk_prot->unhash(*sk);
1665 	}
1666 	return rc;
1667 }
1668 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1669 
1670 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1671 {
1672 	unsigned long res = 0;
1673 	int i;
1674 
1675 	for_each_possible_cpu(i)
1676 		res += snmp_get_cpu_field(mib, i, offt);
1677 	return res;
1678 }
1679 EXPORT_SYMBOL_GPL(snmp_fold_field);
1680 
1681 #if BITS_PER_LONG==32
1682 
1683 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1684 			 size_t syncp_offset)
1685 {
1686 	void *bhptr;
1687 	struct u64_stats_sync *syncp;
1688 	u64 v;
1689 	unsigned int start;
1690 
1691 	bhptr = per_cpu_ptr(mib, cpu);
1692 	syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1693 	do {
1694 		start = u64_stats_fetch_begin(syncp);
1695 		v = *(((u64 *)bhptr) + offt);
1696 	} while (u64_stats_fetch_retry(syncp, start));
1697 
1698 	return v;
1699 }
1700 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1701 
1702 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1703 {
1704 	u64 res = 0;
1705 	int cpu;
1706 
1707 	for_each_possible_cpu(cpu) {
1708 		res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1709 	}
1710 	return res;
1711 }
1712 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1713 #endif
1714 
1715 #ifdef CONFIG_IP_MULTICAST
1716 static const struct net_protocol igmp_protocol = {
1717 	.handler =	igmp_rcv,
1718 };
1719 #endif
1720 
1721 static const struct net_protocol icmp_protocol = {
1722 	.handler =	icmp_rcv,
1723 	.err_handler =	icmp_err,
1724 	.no_policy =	1,
1725 };
1726 
1727 static __net_init int ipv4_mib_init_net(struct net *net)
1728 {
1729 	int i;
1730 
1731 	net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1732 	if (!net->mib.tcp_statistics)
1733 		goto err_tcp_mib;
1734 	net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1735 	if (!net->mib.ip_statistics)
1736 		goto err_ip_mib;
1737 
1738 	for_each_possible_cpu(i) {
1739 		struct ipstats_mib *af_inet_stats;
1740 		af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1741 		u64_stats_init(&af_inet_stats->syncp);
1742 	}
1743 
1744 	net->mib.net_statistics = alloc_percpu(struct linux_mib);
1745 	if (!net->mib.net_statistics)
1746 		goto err_net_mib;
1747 	net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1748 	if (!net->mib.udp_statistics)
1749 		goto err_udp_mib;
1750 	net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1751 	if (!net->mib.udplite_statistics)
1752 		goto err_udplite_mib;
1753 	net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1754 	if (!net->mib.icmp_statistics)
1755 		goto err_icmp_mib;
1756 	net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1757 					      GFP_KERNEL);
1758 	if (!net->mib.icmpmsg_statistics)
1759 		goto err_icmpmsg_mib;
1760 
1761 	tcp_mib_init(net);
1762 	return 0;
1763 
1764 err_icmpmsg_mib:
1765 	free_percpu(net->mib.icmp_statistics);
1766 err_icmp_mib:
1767 	free_percpu(net->mib.udplite_statistics);
1768 err_udplite_mib:
1769 	free_percpu(net->mib.udp_statistics);
1770 err_udp_mib:
1771 	free_percpu(net->mib.net_statistics);
1772 err_net_mib:
1773 	free_percpu(net->mib.ip_statistics);
1774 err_ip_mib:
1775 	free_percpu(net->mib.tcp_statistics);
1776 err_tcp_mib:
1777 	return -ENOMEM;
1778 }
1779 
1780 static __net_exit void ipv4_mib_exit_net(struct net *net)
1781 {
1782 	kfree(net->mib.icmpmsg_statistics);
1783 	free_percpu(net->mib.icmp_statistics);
1784 	free_percpu(net->mib.udplite_statistics);
1785 	free_percpu(net->mib.udp_statistics);
1786 	free_percpu(net->mib.net_statistics);
1787 	free_percpu(net->mib.ip_statistics);
1788 	free_percpu(net->mib.tcp_statistics);
1789 #ifdef CONFIG_MPTCP
1790 	/* allocated on demand, see mptcp_init_sock() */
1791 	free_percpu(net->mib.mptcp_statistics);
1792 #endif
1793 }
1794 
1795 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1796 	.init = ipv4_mib_init_net,
1797 	.exit = ipv4_mib_exit_net,
1798 };
1799 
1800 static int __init init_ipv4_mibs(void)
1801 {
1802 	return register_pernet_subsys(&ipv4_mib_ops);
1803 }
1804 
1805 static __net_init int inet_init_net(struct net *net)
1806 {
1807 	/*
1808 	 * Set defaults for local port range
1809 	 */
1810 	net->ipv4.ip_local_ports.range = 60999u << 16 | 32768u;
1811 
1812 	seqlock_init(&net->ipv4.ping_group_range.lock);
1813 	/*
1814 	 * Sane defaults - nobody may create ping sockets.
1815 	 * Boot scripts should set this to distro-specific group.
1816 	 */
1817 	net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1818 	net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1819 
1820 	/* Default values for sysctl-controlled parameters.
1821 	 * We set them here, in case sysctl is not compiled.
1822 	 */
1823 	net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1824 	net->ipv4.sysctl_ip_fwd_update_priority = 1;
1825 	net->ipv4.sysctl_ip_dynaddr = 0;
1826 	net->ipv4.sysctl_ip_early_demux = 1;
1827 	net->ipv4.sysctl_udp_early_demux = 1;
1828 	net->ipv4.sysctl_tcp_early_demux = 1;
1829 	net->ipv4.sysctl_nexthop_compat_mode = 1;
1830 #ifdef CONFIG_SYSCTL
1831 	net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1832 #endif
1833 
1834 	/* Some igmp sysctl, whose values are always used */
1835 	net->ipv4.sysctl_igmp_max_memberships = 20;
1836 	net->ipv4.sysctl_igmp_max_msf = 10;
1837 	/* IGMP reports for link-local multicast groups are enabled by default */
1838 	net->ipv4.sysctl_igmp_llm_reports = 1;
1839 	net->ipv4.sysctl_igmp_qrv = 2;
1840 
1841 	net->ipv4.sysctl_fib_notify_on_flag_change = 0;
1842 
1843 	return 0;
1844 }
1845 
1846 static __net_initdata struct pernet_operations af_inet_ops = {
1847 	.init = inet_init_net,
1848 };
1849 
1850 static int __init init_inet_pernet_ops(void)
1851 {
1852 	return register_pernet_subsys(&af_inet_ops);
1853 }
1854 
1855 static int ipv4_proc_init(void);
1856 
1857 /*
1858  *	IP protocol layer initialiser
1859  */
1860 
1861 
1862 static const struct net_offload ipip_offload = {
1863 	.callbacks = {
1864 		.gso_segment	= ipip_gso_segment,
1865 		.gro_receive	= ipip_gro_receive,
1866 		.gro_complete	= ipip_gro_complete,
1867 	},
1868 };
1869 
1870 static int __init ipip_offload_init(void)
1871 {
1872 	return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1873 }
1874 
1875 static int __init ipv4_offload_init(void)
1876 {
1877 	/*
1878 	 * Add offloads
1879 	 */
1880 	if (udpv4_offload_init() < 0)
1881 		pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1882 	if (tcpv4_offload_init() < 0)
1883 		pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1884 	if (ipip_offload_init() < 0)
1885 		pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1886 
1887 	net_hotdata.ip_packet_offload = (struct packet_offload) {
1888 		.type = cpu_to_be16(ETH_P_IP),
1889 		.callbacks = {
1890 			.gso_segment = inet_gso_segment,
1891 			.gro_receive = inet_gro_receive,
1892 			.gro_complete = inet_gro_complete,
1893 		},
1894 	};
1895 	dev_add_offload(&net_hotdata.ip_packet_offload);
1896 	return 0;
1897 }
1898 
1899 fs_initcall(ipv4_offload_init);
1900 
1901 static struct packet_type ip_packet_type __read_mostly = {
1902 	.type = cpu_to_be16(ETH_P_IP),
1903 	.func = ip_rcv,
1904 	.list_func = ip_list_rcv,
1905 };
1906 
1907 static int __init inet_init(void)
1908 {
1909 	struct inet_protosw *q;
1910 	struct list_head *r;
1911 	int rc;
1912 
1913 	sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1914 
1915 	raw_hashinfo_init(&raw_v4_hashinfo);
1916 
1917 	rc = proto_register(&tcp_prot, 1);
1918 	if (rc)
1919 		goto out;
1920 
1921 	rc = proto_register(&udp_prot, 1);
1922 	if (rc)
1923 		goto out_unregister_tcp_proto;
1924 
1925 	rc = proto_register(&raw_prot, 1);
1926 	if (rc)
1927 		goto out_unregister_udp_proto;
1928 
1929 	rc = proto_register(&ping_prot, 1);
1930 	if (rc)
1931 		goto out_unregister_raw_proto;
1932 
1933 	/*
1934 	 *	Tell SOCKET that we are alive...
1935 	 */
1936 
1937 	(void)sock_register(&inet_family_ops);
1938 
1939 #ifdef CONFIG_SYSCTL
1940 	ip_static_sysctl_init();
1941 #endif
1942 
1943 	/*
1944 	 *	Add all the base protocols.
1945 	 */
1946 
1947 	if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1948 		pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1949 
1950 	net_hotdata.udp_protocol = (struct net_protocol) {
1951 		.handler =	udp_rcv,
1952 		.err_handler =	udp_err,
1953 		.no_policy =	1,
1954 	};
1955 	if (inet_add_protocol(&net_hotdata.udp_protocol, IPPROTO_UDP) < 0)
1956 		pr_crit("%s: Cannot add UDP protocol\n", __func__);
1957 
1958 	net_hotdata.tcp_protocol = (struct net_protocol) {
1959 		.handler	=	tcp_v4_rcv,
1960 		.err_handler	=	tcp_v4_err,
1961 		.no_policy	=	1,
1962 		.icmp_strict_tag_validation = 1,
1963 	};
1964 	if (inet_add_protocol(&net_hotdata.tcp_protocol, IPPROTO_TCP) < 0)
1965 		pr_crit("%s: Cannot add TCP protocol\n", __func__);
1966 #ifdef CONFIG_IP_MULTICAST
1967 	if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1968 		pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1969 #endif
1970 
1971 	/* Register the socket-side information for inet_create. */
1972 	for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1973 		INIT_LIST_HEAD(r);
1974 
1975 	for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1976 		inet_register_protosw(q);
1977 
1978 	/*
1979 	 *	Set the ARP module up
1980 	 */
1981 
1982 	arp_init();
1983 
1984 	/*
1985 	 *	Set the IP module up
1986 	 */
1987 
1988 	ip_init();
1989 
1990 	/* Initialise per-cpu ipv4 mibs */
1991 	if (init_ipv4_mibs())
1992 		panic("%s: Cannot init ipv4 mibs\n", __func__);
1993 
1994 	/* Setup TCP slab cache for open requests. */
1995 	tcp_init();
1996 
1997 	/* Setup UDP memory threshold */
1998 	udp_init();
1999 
2000 	/* Add UDP-Lite (RFC 3828) */
2001 	udplite4_register();
2002 
2003 	raw_init();
2004 
2005 	ping_init();
2006 
2007 	/*
2008 	 *	Set the ICMP layer up
2009 	 */
2010 
2011 	if (icmp_init() < 0)
2012 		panic("Failed to create the ICMP control socket.\n");
2013 
2014 	/*
2015 	 *	Initialise the multicast router
2016 	 */
2017 #if defined(CONFIG_IP_MROUTE)
2018 	if (ip_mr_init())
2019 		pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2020 #endif
2021 
2022 	if (init_inet_pernet_ops())
2023 		pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2024 
2025 	ipv4_proc_init();
2026 
2027 	ipfrag_init();
2028 
2029 	dev_add_pack(&ip_packet_type);
2030 
2031 	ip_tunnel_core_init();
2032 
2033 	rc = 0;
2034 out:
2035 	return rc;
2036 out_unregister_raw_proto:
2037 	proto_unregister(&raw_prot);
2038 out_unregister_udp_proto:
2039 	proto_unregister(&udp_prot);
2040 out_unregister_tcp_proto:
2041 	proto_unregister(&tcp_prot);
2042 	goto out;
2043 }
2044 
2045 fs_initcall(inet_init);
2046 
2047 /* ------------------------------------------------------------------------ */
2048 
2049 #ifdef CONFIG_PROC_FS
2050 static int __init ipv4_proc_init(void)
2051 {
2052 	int rc = 0;
2053 
2054 	if (raw_proc_init())
2055 		goto out_raw;
2056 	if (tcp4_proc_init())
2057 		goto out_tcp;
2058 	if (udp4_proc_init())
2059 		goto out_udp;
2060 	if (ping_proc_init())
2061 		goto out_ping;
2062 	if (ip_misc_proc_init())
2063 		goto out_misc;
2064 out:
2065 	return rc;
2066 out_misc:
2067 	ping_proc_exit();
2068 out_ping:
2069 	udp4_proc_exit();
2070 out_udp:
2071 	tcp4_proc_exit();
2072 out_tcp:
2073 	raw_proc_exit();
2074 out_raw:
2075 	rc = -ENOMEM;
2076 	goto out;
2077 }
2078 
2079 #else /* CONFIG_PROC_FS */
2080 static int __init ipv4_proc_init(void)
2081 {
2082 	return 0;
2083 }
2084 #endif /* CONFIG_PROC_FS */
2085