xref: /linux/net/ipv4/af_inet.c (revision 9406b485dea5e25bed7c81cd822747d494cc8bde)
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 
120 #include <trace/events/sock.h>
121 
122 /* The inetsw table contains everything that inet_create needs to
123  * build a new socket.
124  */
125 static struct list_head inetsw[SOCK_MAX];
126 static DEFINE_SPINLOCK(inetsw_lock);
127 
128 /* New destruction routine */
129 
130 void inet_sock_destruct(struct sock *sk)
131 {
132 	struct inet_sock *inet = inet_sk(sk);
133 
134 	__skb_queue_purge(&sk->sk_receive_queue);
135 	if (sk->sk_rx_skb_cache) {
136 		__kfree_skb(sk->sk_rx_skb_cache);
137 		sk->sk_rx_skb_cache = NULL;
138 	}
139 	__skb_queue_purge(&sk->sk_error_queue);
140 
141 	sk_mem_reclaim(sk);
142 
143 	if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
144 		pr_err("Attempt to release TCP socket in state %d %p\n",
145 		       sk->sk_state, sk);
146 		return;
147 	}
148 	if (!sock_flag(sk, SOCK_DEAD)) {
149 		pr_err("Attempt to release alive inet socket %p\n", sk);
150 		return;
151 	}
152 
153 	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
154 	WARN_ON(refcount_read(&sk->sk_wmem_alloc));
155 	WARN_ON(sk->sk_wmem_queued);
156 	WARN_ON(sk->sk_forward_alloc);
157 
158 	kfree(rcu_dereference_protected(inet->inet_opt, 1));
159 	dst_release(rcu_dereference_protected(sk->sk_dst_cache, 1));
160 	dst_release(sk->sk_rx_dst);
161 	sk_refcnt_debug_dec(sk);
162 }
163 EXPORT_SYMBOL(inet_sock_destruct);
164 
165 /*
166  *	The routines beyond this point handle the behaviour of an AF_INET
167  *	socket object. Mostly it punts to the subprotocols of IP to do
168  *	the work.
169  */
170 
171 /*
172  *	Automatically bind an unbound socket.
173  */
174 
175 static int inet_autobind(struct sock *sk)
176 {
177 	struct inet_sock *inet;
178 	/* We may need to bind the socket. */
179 	lock_sock(sk);
180 	inet = inet_sk(sk);
181 	if (!inet->inet_num) {
182 		if (sk->sk_prot->get_port(sk, 0)) {
183 			release_sock(sk);
184 			return -EAGAIN;
185 		}
186 		inet->inet_sport = htons(inet->inet_num);
187 	}
188 	release_sock(sk);
189 	return 0;
190 }
191 
192 /*
193  *	Move a socket into listening state.
194  */
195 int inet_listen(struct socket *sock, int backlog)
196 {
197 	struct sock *sk = sock->sk;
198 	unsigned char old_state;
199 	int err, tcp_fastopen;
200 
201 	lock_sock(sk);
202 
203 	err = -EINVAL;
204 	if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
205 		goto out;
206 
207 	old_state = sk->sk_state;
208 	if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
209 		goto out;
210 
211 	WRITE_ONCE(sk->sk_max_ack_backlog, backlog);
212 	/* Really, if the socket is already in listen state
213 	 * we can only allow the backlog to be adjusted.
214 	 */
215 	if (old_state != TCP_LISTEN) {
216 		/* Enable TFO w/o requiring TCP_FASTOPEN socket option.
217 		 * Note that only TCP sockets (SOCK_STREAM) will reach here.
218 		 * Also fastopen backlog may already been set via the option
219 		 * because the socket was in TCP_LISTEN state previously but
220 		 * was shutdown() rather than close().
221 		 */
222 		tcp_fastopen = sock_net(sk)->ipv4.sysctl_tcp_fastopen;
223 		if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
224 		    (tcp_fastopen & TFO_SERVER_ENABLE) &&
225 		    !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
226 			fastopen_queue_tune(sk, backlog);
227 			tcp_fastopen_init_key_once(sock_net(sk));
228 		}
229 
230 		err = inet_csk_listen_start(sk, backlog);
231 		if (err)
232 			goto out;
233 		tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL);
234 	}
235 	err = 0;
236 
237 out:
238 	release_sock(sk);
239 	return err;
240 }
241 EXPORT_SYMBOL(inet_listen);
242 
243 /*
244  *	Create an inet socket.
245  */
246 
247 static int inet_create(struct net *net, struct socket *sock, int protocol,
248 		       int kern)
249 {
250 	struct sock *sk;
251 	struct inet_protosw *answer;
252 	struct inet_sock *inet;
253 	struct proto *answer_prot;
254 	unsigned char answer_flags;
255 	int try_loading_module = 0;
256 	int err;
257 
258 	if (protocol < 0 || protocol >= IPPROTO_MAX)
259 		return -EINVAL;
260 
261 	sock->state = SS_UNCONNECTED;
262 
263 	/* Look for the requested type/protocol pair. */
264 lookup_protocol:
265 	err = -ESOCKTNOSUPPORT;
266 	rcu_read_lock();
267 	list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
268 
269 		err = 0;
270 		/* Check the non-wild match. */
271 		if (protocol == answer->protocol) {
272 			if (protocol != IPPROTO_IP)
273 				break;
274 		} else {
275 			/* Check for the two wild cases. */
276 			if (IPPROTO_IP == protocol) {
277 				protocol = answer->protocol;
278 				break;
279 			}
280 			if (IPPROTO_IP == answer->protocol)
281 				break;
282 		}
283 		err = -EPROTONOSUPPORT;
284 	}
285 
286 	if (unlikely(err)) {
287 		if (try_loading_module < 2) {
288 			rcu_read_unlock();
289 			/*
290 			 * Be more specific, e.g. net-pf-2-proto-132-type-1
291 			 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
292 			 */
293 			if (++try_loading_module == 1)
294 				request_module("net-pf-%d-proto-%d-type-%d",
295 					       PF_INET, protocol, sock->type);
296 			/*
297 			 * Fall back to generic, e.g. net-pf-2-proto-132
298 			 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
299 			 */
300 			else
301 				request_module("net-pf-%d-proto-%d",
302 					       PF_INET, protocol);
303 			goto lookup_protocol;
304 		} else
305 			goto out_rcu_unlock;
306 	}
307 
308 	err = -EPERM;
309 	if (sock->type == SOCK_RAW && !kern &&
310 	    !ns_capable(net->user_ns, CAP_NET_RAW))
311 		goto out_rcu_unlock;
312 
313 	sock->ops = answer->ops;
314 	answer_prot = answer->prot;
315 	answer_flags = answer->flags;
316 	rcu_read_unlock();
317 
318 	WARN_ON(!answer_prot->slab);
319 
320 	err = -ENOBUFS;
321 	sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
322 	if (!sk)
323 		goto out;
324 
325 	err = 0;
326 	if (INET_PROTOSW_REUSE & answer_flags)
327 		sk->sk_reuse = SK_CAN_REUSE;
328 
329 	inet = inet_sk(sk);
330 	inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
331 
332 	inet->nodefrag = 0;
333 
334 	if (SOCK_RAW == sock->type) {
335 		inet->inet_num = protocol;
336 		if (IPPROTO_RAW == protocol)
337 			inet->hdrincl = 1;
338 	}
339 
340 	if (net->ipv4.sysctl_ip_no_pmtu_disc)
341 		inet->pmtudisc = IP_PMTUDISC_DONT;
342 	else
343 		inet->pmtudisc = IP_PMTUDISC_WANT;
344 
345 	inet->inet_id = 0;
346 
347 	sock_init_data(sock, sk);
348 
349 	sk->sk_destruct	   = inet_sock_destruct;
350 	sk->sk_protocol	   = protocol;
351 	sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
352 
353 	inet->uc_ttl	= -1;
354 	inet->mc_loop	= 1;
355 	inet->mc_ttl	= 1;
356 	inet->mc_all	= 1;
357 	inet->mc_index	= 0;
358 	inet->mc_list	= NULL;
359 	inet->rcv_tos	= 0;
360 
361 	sk_refcnt_debug_inc(sk);
362 
363 	if (inet->inet_num) {
364 		/* It assumes that any protocol which allows
365 		 * the user to assign a number at socket
366 		 * creation time automatically
367 		 * shares.
368 		 */
369 		inet->inet_sport = htons(inet->inet_num);
370 		/* Add to protocol hash chains. */
371 		err = sk->sk_prot->hash(sk);
372 		if (err) {
373 			sk_common_release(sk);
374 			goto out;
375 		}
376 	}
377 
378 	if (sk->sk_prot->init) {
379 		err = sk->sk_prot->init(sk);
380 		if (err) {
381 			sk_common_release(sk);
382 			goto out;
383 		}
384 	}
385 
386 	if (!kern) {
387 		err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
388 		if (err) {
389 			sk_common_release(sk);
390 			goto out;
391 		}
392 	}
393 out:
394 	return err;
395 out_rcu_unlock:
396 	rcu_read_unlock();
397 	goto out;
398 }
399 
400 
401 /*
402  *	The peer socket should always be NULL (or else). When we call this
403  *	function we are destroying the object and from then on nobody
404  *	should refer to it.
405  */
406 int inet_release(struct socket *sock)
407 {
408 	struct sock *sk = sock->sk;
409 
410 	if (sk) {
411 		long timeout;
412 
413 		/* Applications forget to leave groups before exiting */
414 		ip_mc_drop_socket(sk);
415 
416 		/* If linger is set, we don't return until the close
417 		 * is complete.  Otherwise we return immediately. The
418 		 * actually closing is done the same either way.
419 		 *
420 		 * If the close is due to the process exiting, we never
421 		 * linger..
422 		 */
423 		timeout = 0;
424 		if (sock_flag(sk, SOCK_LINGER) &&
425 		    !(current->flags & PF_EXITING))
426 			timeout = sk->sk_lingertime;
427 		sk->sk_prot->close(sk, timeout);
428 		sock->sk = NULL;
429 	}
430 	return 0;
431 }
432 EXPORT_SYMBOL(inet_release);
433 
434 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
435 {
436 	struct sock *sk = sock->sk;
437 	int err;
438 
439 	/* If the socket has its own bind function then use it. (RAW) */
440 	if (sk->sk_prot->bind) {
441 		return sk->sk_prot->bind(sk, uaddr, addr_len);
442 	}
443 	if (addr_len < sizeof(struct sockaddr_in))
444 		return -EINVAL;
445 
446 	/* BPF prog is run before any checks are done so that if the prog
447 	 * changes context in a wrong way it will be caught.
448 	 */
449 	err = BPF_CGROUP_RUN_PROG_INET4_BIND(sk, uaddr);
450 	if (err)
451 		return err;
452 
453 	return __inet_bind(sk, uaddr, addr_len, BIND_WITH_LOCK);
454 }
455 EXPORT_SYMBOL(inet_bind);
456 
457 int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
458 		u32 flags)
459 {
460 	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
461 	struct inet_sock *inet = inet_sk(sk);
462 	struct net *net = sock_net(sk);
463 	unsigned short snum;
464 	int chk_addr_ret;
465 	u32 tb_id = RT_TABLE_LOCAL;
466 	int err;
467 
468 	if (addr->sin_family != AF_INET) {
469 		/* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
470 		 * only if s_addr is INADDR_ANY.
471 		 */
472 		err = -EAFNOSUPPORT;
473 		if (addr->sin_family != AF_UNSPEC ||
474 		    addr->sin_addr.s_addr != htonl(INADDR_ANY))
475 			goto out;
476 	}
477 
478 	tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
479 	chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
480 
481 	/* Not specified by any standard per-se, however it breaks too
482 	 * many applications when removed.  It is unfortunate since
483 	 * allowing applications to make a non-local bind solves
484 	 * several problems with systems using dynamic addressing.
485 	 * (ie. your servers still start up even if your ISDN link
486 	 *  is temporarily down)
487 	 */
488 	err = -EADDRNOTAVAIL;
489 	if (!inet_can_nonlocal_bind(net, inet) &&
490 	    addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
491 	    chk_addr_ret != RTN_LOCAL &&
492 	    chk_addr_ret != RTN_MULTICAST &&
493 	    chk_addr_ret != RTN_BROADCAST)
494 		goto out;
495 
496 	snum = ntohs(addr->sin_port);
497 	err = -EACCES;
498 	if (snum && inet_port_requires_bind_service(net, snum) &&
499 	    !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
500 		goto out;
501 
502 	/*      We keep a pair of addresses. rcv_saddr is the one
503 	 *      used by hash lookups, and saddr is used for transmit.
504 	 *
505 	 *      In the BSD API these are the same except where it
506 	 *      would be illegal to use them (multicast/broadcast) in
507 	 *      which case the sending device address is used.
508 	 */
509 	if (flags & BIND_WITH_LOCK)
510 		lock_sock(sk);
511 
512 	/* Check these errors (active socket, double bind). */
513 	err = -EINVAL;
514 	if (sk->sk_state != TCP_CLOSE || inet->inet_num)
515 		goto out_release_sock;
516 
517 	inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
518 	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
519 		inet->inet_saddr = 0;  /* Use device */
520 
521 	/* Make sure we are allowed to bind here. */
522 	if (snum || !(inet->bind_address_no_port ||
523 		      (flags & BIND_FORCE_ADDRESS_NO_PORT))) {
524 		if (sk->sk_prot->get_port(sk, snum)) {
525 			inet->inet_saddr = inet->inet_rcv_saddr = 0;
526 			err = -EADDRINUSE;
527 			goto out_release_sock;
528 		}
529 		if (!(flags & BIND_FROM_BPF)) {
530 			err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
531 			if (err) {
532 				inet->inet_saddr = inet->inet_rcv_saddr = 0;
533 				goto out_release_sock;
534 			}
535 		}
536 	}
537 
538 	if (inet->inet_rcv_saddr)
539 		sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
540 	if (snum)
541 		sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
542 	inet->inet_sport = htons(inet->inet_num);
543 	inet->inet_daddr = 0;
544 	inet->inet_dport = 0;
545 	sk_dst_reset(sk);
546 	err = 0;
547 out_release_sock:
548 	if (flags & BIND_WITH_LOCK)
549 		release_sock(sk);
550 out:
551 	return err;
552 }
553 
554 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
555 		       int addr_len, int flags)
556 {
557 	struct sock *sk = sock->sk;
558 	int err;
559 
560 	if (addr_len < sizeof(uaddr->sa_family))
561 		return -EINVAL;
562 	if (uaddr->sa_family == AF_UNSPEC)
563 		return sk->sk_prot->disconnect(sk, flags);
564 
565 	if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
566 		err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
567 		if (err)
568 			return err;
569 	}
570 
571 	if (!inet_sk(sk)->inet_num && inet_autobind(sk))
572 		return -EAGAIN;
573 	return sk->sk_prot->connect(sk, uaddr, addr_len);
574 }
575 EXPORT_SYMBOL(inet_dgram_connect);
576 
577 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
578 {
579 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
580 
581 	add_wait_queue(sk_sleep(sk), &wait);
582 	sk->sk_write_pending += writebias;
583 
584 	/* Basic assumption: if someone sets sk->sk_err, he _must_
585 	 * change state of the socket from TCP_SYN_*.
586 	 * Connect() does not allow to get error notifications
587 	 * without closing the socket.
588 	 */
589 	while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
590 		release_sock(sk);
591 		timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
592 		lock_sock(sk);
593 		if (signal_pending(current) || !timeo)
594 			break;
595 	}
596 	remove_wait_queue(sk_sleep(sk), &wait);
597 	sk->sk_write_pending -= writebias;
598 	return timeo;
599 }
600 
601 /*
602  *	Connect to a remote host. There is regrettably still a little
603  *	TCP 'magic' in here.
604  */
605 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
606 			  int addr_len, int flags, int is_sendmsg)
607 {
608 	struct sock *sk = sock->sk;
609 	int err;
610 	long timeo;
611 
612 	/*
613 	 * uaddr can be NULL and addr_len can be 0 if:
614 	 * sk is a TCP fastopen active socket and
615 	 * TCP_FASTOPEN_CONNECT sockopt is set and
616 	 * we already have a valid cookie for this socket.
617 	 * In this case, user can call write() after connect().
618 	 * write() will invoke tcp_sendmsg_fastopen() which calls
619 	 * __inet_stream_connect().
620 	 */
621 	if (uaddr) {
622 		if (addr_len < sizeof(uaddr->sa_family))
623 			return -EINVAL;
624 
625 		if (uaddr->sa_family == AF_UNSPEC) {
626 			err = sk->sk_prot->disconnect(sk, flags);
627 			sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
628 			goto out;
629 		}
630 	}
631 
632 	switch (sock->state) {
633 	default:
634 		err = -EINVAL;
635 		goto out;
636 	case SS_CONNECTED:
637 		err = -EISCONN;
638 		goto out;
639 	case SS_CONNECTING:
640 		if (inet_sk(sk)->defer_connect)
641 			err = is_sendmsg ? -EINPROGRESS : -EISCONN;
642 		else
643 			err = -EALREADY;
644 		/* Fall out of switch with err, set for this state */
645 		break;
646 	case SS_UNCONNECTED:
647 		err = -EISCONN;
648 		if (sk->sk_state != TCP_CLOSE)
649 			goto out;
650 
651 		if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
652 			err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
653 			if (err)
654 				goto out;
655 		}
656 
657 		err = sk->sk_prot->connect(sk, uaddr, addr_len);
658 		if (err < 0)
659 			goto out;
660 
661 		sock->state = SS_CONNECTING;
662 
663 		if (!err && inet_sk(sk)->defer_connect)
664 			goto out;
665 
666 		/* Just entered SS_CONNECTING state; the only
667 		 * difference is that return value in non-blocking
668 		 * case is EINPROGRESS, rather than EALREADY.
669 		 */
670 		err = -EINPROGRESS;
671 		break;
672 	}
673 
674 	timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
675 
676 	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
677 		int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
678 				tcp_sk(sk)->fastopen_req &&
679 				tcp_sk(sk)->fastopen_req->data ? 1 : 0;
680 
681 		/* Error code is set above */
682 		if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
683 			goto out;
684 
685 		err = sock_intr_errno(timeo);
686 		if (signal_pending(current))
687 			goto out;
688 	}
689 
690 	/* Connection was closed by RST, timeout, ICMP error
691 	 * or another process disconnected us.
692 	 */
693 	if (sk->sk_state == TCP_CLOSE)
694 		goto sock_error;
695 
696 	/* sk->sk_err may be not zero now, if RECVERR was ordered by user
697 	 * and error was received after socket entered established state.
698 	 * Hence, it is handled normally after connect() return successfully.
699 	 */
700 
701 	sock->state = SS_CONNECTED;
702 	err = 0;
703 out:
704 	return err;
705 
706 sock_error:
707 	err = sock_error(sk) ? : -ECONNABORTED;
708 	sock->state = SS_UNCONNECTED;
709 	if (sk->sk_prot->disconnect(sk, flags))
710 		sock->state = SS_DISCONNECTING;
711 	goto out;
712 }
713 EXPORT_SYMBOL(__inet_stream_connect);
714 
715 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
716 			int addr_len, int flags)
717 {
718 	int err;
719 
720 	lock_sock(sock->sk);
721 	err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
722 	release_sock(sock->sk);
723 	return err;
724 }
725 EXPORT_SYMBOL(inet_stream_connect);
726 
727 /*
728  *	Accept a pending connection. The TCP layer now gives BSD semantics.
729  */
730 
731 int inet_accept(struct socket *sock, struct socket *newsock, int flags,
732 		bool kern)
733 {
734 	struct sock *sk1 = sock->sk;
735 	int err = -EINVAL;
736 	struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err, kern);
737 
738 	if (!sk2)
739 		goto do_err;
740 
741 	lock_sock(sk2);
742 
743 	sock_rps_record_flow(sk2);
744 	WARN_ON(!((1 << sk2->sk_state) &
745 		  (TCPF_ESTABLISHED | TCPF_SYN_RECV |
746 		  TCPF_CLOSE_WAIT | TCPF_CLOSE)));
747 
748 	sock_graft(sk2, newsock);
749 
750 	newsock->state = SS_CONNECTED;
751 	err = 0;
752 	release_sock(sk2);
753 do_err:
754 	return err;
755 }
756 EXPORT_SYMBOL(inet_accept);
757 
758 
759 /*
760  *	This does both peername and sockname.
761  */
762 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
763 			int peer)
764 {
765 	struct sock *sk		= sock->sk;
766 	struct inet_sock *inet	= inet_sk(sk);
767 	DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
768 
769 	sin->sin_family = AF_INET;
770 	if (peer) {
771 		if (!inet->inet_dport ||
772 		    (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
773 		     peer == 1))
774 			return -ENOTCONN;
775 		sin->sin_port = inet->inet_dport;
776 		sin->sin_addr.s_addr = inet->inet_daddr;
777 	} else {
778 		__be32 addr = inet->inet_rcv_saddr;
779 		if (!addr)
780 			addr = inet->inet_saddr;
781 		sin->sin_port = inet->inet_sport;
782 		sin->sin_addr.s_addr = addr;
783 	}
784 	memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
785 	return sizeof(*sin);
786 }
787 EXPORT_SYMBOL(inet_getname);
788 
789 int inet_send_prepare(struct sock *sk)
790 {
791 	sock_rps_record_flow(sk);
792 
793 	/* We may need to bind the socket. */
794 	if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
795 	    inet_autobind(sk))
796 		return -EAGAIN;
797 
798 	return 0;
799 }
800 EXPORT_SYMBOL_GPL(inet_send_prepare);
801 
802 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
803 {
804 	struct sock *sk = sock->sk;
805 
806 	if (unlikely(inet_send_prepare(sk)))
807 		return -EAGAIN;
808 
809 	return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
810 			       sk, msg, size);
811 }
812 EXPORT_SYMBOL(inet_sendmsg);
813 
814 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
815 		      size_t size, int flags)
816 {
817 	struct sock *sk = sock->sk;
818 
819 	if (unlikely(inet_send_prepare(sk)))
820 		return -EAGAIN;
821 
822 	if (sk->sk_prot->sendpage)
823 		return sk->sk_prot->sendpage(sk, page, offset, size, flags);
824 	return sock_no_sendpage(sock, page, offset, size, flags);
825 }
826 EXPORT_SYMBOL(inet_sendpage);
827 
828 INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
829 					  size_t, int, int, int *));
830 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
831 		 int flags)
832 {
833 	struct sock *sk = sock->sk;
834 	int addr_len = 0;
835 	int err;
836 
837 	if (likely(!(flags & MSG_ERRQUEUE)))
838 		sock_rps_record_flow(sk);
839 
840 	err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
841 			      sk, msg, size, flags & MSG_DONTWAIT,
842 			      flags & ~MSG_DONTWAIT, &addr_len);
843 	if (err >= 0)
844 		msg->msg_namelen = addr_len;
845 	return err;
846 }
847 EXPORT_SYMBOL(inet_recvmsg);
848 
849 int inet_shutdown(struct socket *sock, int how)
850 {
851 	struct sock *sk = sock->sk;
852 	int err = 0;
853 
854 	/* This should really check to make sure
855 	 * the socket is a TCP socket. (WHY AC...)
856 	 */
857 	how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
858 		       1->2 bit 2 snds.
859 		       2->3 */
860 	if ((how & ~SHUTDOWN_MASK) || !how)	/* MAXINT->0 */
861 		return -EINVAL;
862 
863 	lock_sock(sk);
864 	if (sock->state == SS_CONNECTING) {
865 		if ((1 << sk->sk_state) &
866 		    (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
867 			sock->state = SS_DISCONNECTING;
868 		else
869 			sock->state = SS_CONNECTED;
870 	}
871 
872 	switch (sk->sk_state) {
873 	case TCP_CLOSE:
874 		err = -ENOTCONN;
875 		/* Hack to wake up other listeners, who can poll for
876 		   EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
877 		fallthrough;
878 	default:
879 		sk->sk_shutdown |= how;
880 		if (sk->sk_prot->shutdown)
881 			sk->sk_prot->shutdown(sk, how);
882 		break;
883 
884 	/* Remaining two branches are temporary solution for missing
885 	 * close() in multithreaded environment. It is _not_ a good idea,
886 	 * but we have no choice until close() is repaired at VFS level.
887 	 */
888 	case TCP_LISTEN:
889 		if (!(how & RCV_SHUTDOWN))
890 			break;
891 		fallthrough;
892 	case TCP_SYN_SENT:
893 		err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
894 		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
895 		break;
896 	}
897 
898 	/* Wake up anyone sleeping in poll. */
899 	sk->sk_state_change(sk);
900 	release_sock(sk);
901 	return err;
902 }
903 EXPORT_SYMBOL(inet_shutdown);
904 
905 /*
906  *	ioctl() calls you can issue on an INET socket. Most of these are
907  *	device configuration and stuff and very rarely used. Some ioctls
908  *	pass on to the socket itself.
909  *
910  *	NOTE: I like the idea of a module for the config stuff. ie ifconfig
911  *	loads the devconfigure module does its configuring and unloads it.
912  *	There's a good 20K of config code hanging around the kernel.
913  */
914 
915 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
916 {
917 	struct sock *sk = sock->sk;
918 	int err = 0;
919 	struct net *net = sock_net(sk);
920 	void __user *p = (void __user *)arg;
921 	struct ifreq ifr;
922 	struct rtentry rt;
923 
924 	switch (cmd) {
925 	case SIOCADDRT:
926 	case SIOCDELRT:
927 		if (copy_from_user(&rt, p, sizeof(struct rtentry)))
928 			return -EFAULT;
929 		err = ip_rt_ioctl(net, cmd, &rt);
930 		break;
931 	case SIOCRTMSG:
932 		err = -EINVAL;
933 		break;
934 	case SIOCDARP:
935 	case SIOCGARP:
936 	case SIOCSARP:
937 		err = arp_ioctl(net, cmd, (void __user *)arg);
938 		break;
939 	case SIOCGIFADDR:
940 	case SIOCGIFBRDADDR:
941 	case SIOCGIFNETMASK:
942 	case SIOCGIFDSTADDR:
943 	case SIOCGIFPFLAGS:
944 		if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
945 			return -EFAULT;
946 		err = devinet_ioctl(net, cmd, &ifr);
947 		if (!err && copy_to_user(p, &ifr, sizeof(struct ifreq)))
948 			err = -EFAULT;
949 		break;
950 
951 	case SIOCSIFADDR:
952 	case SIOCSIFBRDADDR:
953 	case SIOCSIFNETMASK:
954 	case SIOCSIFDSTADDR:
955 	case SIOCSIFPFLAGS:
956 	case SIOCSIFFLAGS:
957 		if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
958 			return -EFAULT;
959 		err = devinet_ioctl(net, cmd, &ifr);
960 		break;
961 	default:
962 		if (sk->sk_prot->ioctl)
963 			err = sk->sk_prot->ioctl(sk, cmd, arg);
964 		else
965 			err = -ENOIOCTLCMD;
966 		break;
967 	}
968 	return err;
969 }
970 EXPORT_SYMBOL(inet_ioctl);
971 
972 #ifdef CONFIG_COMPAT
973 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
974 {
975 	struct sock *sk = sock->sk;
976 	int err = -ENOIOCTLCMD;
977 
978 	if (sk->sk_prot->compat_ioctl)
979 		err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
980 
981 	return err;
982 }
983 #endif
984 
985 const struct proto_ops inet_stream_ops = {
986 	.family		   = PF_INET,
987 	.owner		   = THIS_MODULE,
988 	.release	   = inet_release,
989 	.bind		   = inet_bind,
990 	.connect	   = inet_stream_connect,
991 	.socketpair	   = sock_no_socketpair,
992 	.accept		   = inet_accept,
993 	.getname	   = inet_getname,
994 	.poll		   = tcp_poll,
995 	.ioctl		   = inet_ioctl,
996 	.gettstamp	   = sock_gettstamp,
997 	.listen		   = inet_listen,
998 	.shutdown	   = inet_shutdown,
999 	.setsockopt	   = sock_common_setsockopt,
1000 	.getsockopt	   = sock_common_getsockopt,
1001 	.sendmsg	   = inet_sendmsg,
1002 	.recvmsg	   = inet_recvmsg,
1003 #ifdef CONFIG_MMU
1004 	.mmap		   = tcp_mmap,
1005 #endif
1006 	.sendpage	   = inet_sendpage,
1007 	.splice_read	   = tcp_splice_read,
1008 	.read_sock	   = tcp_read_sock,
1009 	.sendmsg_locked    = tcp_sendmsg_locked,
1010 	.sendpage_locked   = tcp_sendpage_locked,
1011 	.peek_len	   = tcp_peek_len,
1012 #ifdef CONFIG_COMPAT
1013 	.compat_setsockopt = compat_sock_common_setsockopt,
1014 	.compat_getsockopt = compat_sock_common_getsockopt,
1015 	.compat_ioctl	   = inet_compat_ioctl,
1016 #endif
1017 	.set_rcvlowat	   = tcp_set_rcvlowat,
1018 };
1019 EXPORT_SYMBOL(inet_stream_ops);
1020 
1021 const struct proto_ops inet_dgram_ops = {
1022 	.family		   = PF_INET,
1023 	.owner		   = THIS_MODULE,
1024 	.release	   = inet_release,
1025 	.bind		   = inet_bind,
1026 	.connect	   = inet_dgram_connect,
1027 	.socketpair	   = sock_no_socketpair,
1028 	.accept		   = sock_no_accept,
1029 	.getname	   = inet_getname,
1030 	.poll		   = udp_poll,
1031 	.ioctl		   = inet_ioctl,
1032 	.gettstamp	   = sock_gettstamp,
1033 	.listen		   = sock_no_listen,
1034 	.shutdown	   = inet_shutdown,
1035 	.setsockopt	   = sock_common_setsockopt,
1036 	.getsockopt	   = sock_common_getsockopt,
1037 	.sendmsg	   = inet_sendmsg,
1038 	.recvmsg	   = inet_recvmsg,
1039 	.mmap		   = sock_no_mmap,
1040 	.sendpage	   = inet_sendpage,
1041 	.set_peek_off	   = sk_set_peek_off,
1042 #ifdef CONFIG_COMPAT
1043 	.compat_setsockopt = compat_sock_common_setsockopt,
1044 	.compat_getsockopt = compat_sock_common_getsockopt,
1045 	.compat_ioctl	   = inet_compat_ioctl,
1046 #endif
1047 };
1048 EXPORT_SYMBOL(inet_dgram_ops);
1049 
1050 /*
1051  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1052  * udp_poll
1053  */
1054 static const struct proto_ops inet_sockraw_ops = {
1055 	.family		   = PF_INET,
1056 	.owner		   = THIS_MODULE,
1057 	.release	   = inet_release,
1058 	.bind		   = inet_bind,
1059 	.connect	   = inet_dgram_connect,
1060 	.socketpair	   = sock_no_socketpair,
1061 	.accept		   = sock_no_accept,
1062 	.getname	   = inet_getname,
1063 	.poll		   = datagram_poll,
1064 	.ioctl		   = inet_ioctl,
1065 	.gettstamp	   = sock_gettstamp,
1066 	.listen		   = sock_no_listen,
1067 	.shutdown	   = inet_shutdown,
1068 	.setsockopt	   = sock_common_setsockopt,
1069 	.getsockopt	   = sock_common_getsockopt,
1070 	.sendmsg	   = inet_sendmsg,
1071 	.recvmsg	   = inet_recvmsg,
1072 	.mmap		   = sock_no_mmap,
1073 	.sendpage	   = inet_sendpage,
1074 #ifdef CONFIG_COMPAT
1075 	.compat_setsockopt = compat_sock_common_setsockopt,
1076 	.compat_getsockopt = compat_sock_common_getsockopt,
1077 	.compat_ioctl	   = inet_compat_ioctl,
1078 #endif
1079 };
1080 
1081 static const struct net_proto_family inet_family_ops = {
1082 	.family = PF_INET,
1083 	.create = inet_create,
1084 	.owner	= THIS_MODULE,
1085 };
1086 
1087 /* Upon startup we insert all the elements in inetsw_array[] into
1088  * the linked list inetsw.
1089  */
1090 static struct inet_protosw inetsw_array[] =
1091 {
1092 	{
1093 		.type =       SOCK_STREAM,
1094 		.protocol =   IPPROTO_TCP,
1095 		.prot =       &tcp_prot,
1096 		.ops =        &inet_stream_ops,
1097 		.flags =      INET_PROTOSW_PERMANENT |
1098 			      INET_PROTOSW_ICSK,
1099 	},
1100 
1101 	{
1102 		.type =       SOCK_DGRAM,
1103 		.protocol =   IPPROTO_UDP,
1104 		.prot =       &udp_prot,
1105 		.ops =        &inet_dgram_ops,
1106 		.flags =      INET_PROTOSW_PERMANENT,
1107        },
1108 
1109        {
1110 		.type =       SOCK_DGRAM,
1111 		.protocol =   IPPROTO_ICMP,
1112 		.prot =       &ping_prot,
1113 		.ops =        &inet_sockraw_ops,
1114 		.flags =      INET_PROTOSW_REUSE,
1115        },
1116 
1117        {
1118 	       .type =       SOCK_RAW,
1119 	       .protocol =   IPPROTO_IP,	/* wild card */
1120 	       .prot =       &raw_prot,
1121 	       .ops =        &inet_sockraw_ops,
1122 	       .flags =      INET_PROTOSW_REUSE,
1123        }
1124 };
1125 
1126 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1127 
1128 void inet_register_protosw(struct inet_protosw *p)
1129 {
1130 	struct list_head *lh;
1131 	struct inet_protosw *answer;
1132 	int protocol = p->protocol;
1133 	struct list_head *last_perm;
1134 
1135 	spin_lock_bh(&inetsw_lock);
1136 
1137 	if (p->type >= SOCK_MAX)
1138 		goto out_illegal;
1139 
1140 	/* If we are trying to override a permanent protocol, bail. */
1141 	last_perm = &inetsw[p->type];
1142 	list_for_each(lh, &inetsw[p->type]) {
1143 		answer = list_entry(lh, struct inet_protosw, list);
1144 		/* Check only the non-wild match. */
1145 		if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1146 			break;
1147 		if (protocol == answer->protocol)
1148 			goto out_permanent;
1149 		last_perm = lh;
1150 	}
1151 
1152 	/* Add the new entry after the last permanent entry if any, so that
1153 	 * the new entry does not override a permanent entry when matched with
1154 	 * a wild-card protocol. But it is allowed to override any existing
1155 	 * non-permanent entry.  This means that when we remove this entry, the
1156 	 * system automatically returns to the old behavior.
1157 	 */
1158 	list_add_rcu(&p->list, last_perm);
1159 out:
1160 	spin_unlock_bh(&inetsw_lock);
1161 
1162 	return;
1163 
1164 out_permanent:
1165 	pr_err("Attempt to override permanent protocol %d\n", protocol);
1166 	goto out;
1167 
1168 out_illegal:
1169 	pr_err("Ignoring attempt to register invalid socket type %d\n",
1170 	       p->type);
1171 	goto out;
1172 }
1173 EXPORT_SYMBOL(inet_register_protosw);
1174 
1175 void inet_unregister_protosw(struct inet_protosw *p)
1176 {
1177 	if (INET_PROTOSW_PERMANENT & p->flags) {
1178 		pr_err("Attempt to unregister permanent protocol %d\n",
1179 		       p->protocol);
1180 	} else {
1181 		spin_lock_bh(&inetsw_lock);
1182 		list_del_rcu(&p->list);
1183 		spin_unlock_bh(&inetsw_lock);
1184 
1185 		synchronize_net();
1186 	}
1187 }
1188 EXPORT_SYMBOL(inet_unregister_protosw);
1189 
1190 static int inet_sk_reselect_saddr(struct sock *sk)
1191 {
1192 	struct inet_sock *inet = inet_sk(sk);
1193 	__be32 old_saddr = inet->inet_saddr;
1194 	__be32 daddr = inet->inet_daddr;
1195 	struct flowi4 *fl4;
1196 	struct rtable *rt;
1197 	__be32 new_saddr;
1198 	struct ip_options_rcu *inet_opt;
1199 
1200 	inet_opt = rcu_dereference_protected(inet->inet_opt,
1201 					     lockdep_sock_is_held(sk));
1202 	if (inet_opt && inet_opt->opt.srr)
1203 		daddr = inet_opt->opt.faddr;
1204 
1205 	/* Query new route. */
1206 	fl4 = &inet->cork.fl.u.ip4;
1207 	rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1208 			      sk->sk_bound_dev_if, sk->sk_protocol,
1209 			      inet->inet_sport, inet->inet_dport, sk);
1210 	if (IS_ERR(rt))
1211 		return PTR_ERR(rt);
1212 
1213 	sk_setup_caps(sk, &rt->dst);
1214 
1215 	new_saddr = fl4->saddr;
1216 
1217 	if (new_saddr == old_saddr)
1218 		return 0;
1219 
1220 	if (sock_net(sk)->ipv4.sysctl_ip_dynaddr > 1) {
1221 		pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1222 			__func__, &old_saddr, &new_saddr);
1223 	}
1224 
1225 	inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1226 
1227 	/*
1228 	 * XXX The only one ugly spot where we need to
1229 	 * XXX really change the sockets identity after
1230 	 * XXX it has entered the hashes. -DaveM
1231 	 *
1232 	 * Besides that, it does not check for connection
1233 	 * uniqueness. Wait for troubles.
1234 	 */
1235 	return __sk_prot_rehash(sk);
1236 }
1237 
1238 int inet_sk_rebuild_header(struct sock *sk)
1239 {
1240 	struct inet_sock *inet = inet_sk(sk);
1241 	struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1242 	__be32 daddr;
1243 	struct ip_options_rcu *inet_opt;
1244 	struct flowi4 *fl4;
1245 	int err;
1246 
1247 	/* Route is OK, nothing to do. */
1248 	if (rt)
1249 		return 0;
1250 
1251 	/* Reroute. */
1252 	rcu_read_lock();
1253 	inet_opt = rcu_dereference(inet->inet_opt);
1254 	daddr = inet->inet_daddr;
1255 	if (inet_opt && inet_opt->opt.srr)
1256 		daddr = inet_opt->opt.faddr;
1257 	rcu_read_unlock();
1258 	fl4 = &inet->cork.fl.u.ip4;
1259 	rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1260 				   inet->inet_dport, inet->inet_sport,
1261 				   sk->sk_protocol, RT_CONN_FLAGS(sk),
1262 				   sk->sk_bound_dev_if);
1263 	if (!IS_ERR(rt)) {
1264 		err = 0;
1265 		sk_setup_caps(sk, &rt->dst);
1266 	} else {
1267 		err = PTR_ERR(rt);
1268 
1269 		/* Routing failed... */
1270 		sk->sk_route_caps = 0;
1271 		/*
1272 		 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1273 		 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1274 		 */
1275 		if (!sock_net(sk)->ipv4.sysctl_ip_dynaddr ||
1276 		    sk->sk_state != TCP_SYN_SENT ||
1277 		    (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1278 		    (err = inet_sk_reselect_saddr(sk)) != 0)
1279 			sk->sk_err_soft = -err;
1280 	}
1281 
1282 	return err;
1283 }
1284 EXPORT_SYMBOL(inet_sk_rebuild_header);
1285 
1286 void inet_sk_set_state(struct sock *sk, int state)
1287 {
1288 	trace_inet_sock_set_state(sk, sk->sk_state, state);
1289 	sk->sk_state = state;
1290 }
1291 EXPORT_SYMBOL(inet_sk_set_state);
1292 
1293 void inet_sk_state_store(struct sock *sk, int newstate)
1294 {
1295 	trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1296 	smp_store_release(&sk->sk_state, newstate);
1297 }
1298 
1299 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1300 				 netdev_features_t features)
1301 {
1302 	bool udpfrag = false, fixedid = false, gso_partial, encap;
1303 	struct sk_buff *segs = ERR_PTR(-EINVAL);
1304 	const struct net_offload *ops;
1305 	unsigned int offset = 0;
1306 	struct iphdr *iph;
1307 	int proto, tot_len;
1308 	int nhoff;
1309 	int ihl;
1310 	int id;
1311 
1312 	skb_reset_network_header(skb);
1313 	nhoff = skb_network_header(skb) - skb_mac_header(skb);
1314 	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1315 		goto out;
1316 
1317 	iph = ip_hdr(skb);
1318 	ihl = iph->ihl * 4;
1319 	if (ihl < sizeof(*iph))
1320 		goto out;
1321 
1322 	id = ntohs(iph->id);
1323 	proto = iph->protocol;
1324 
1325 	/* Warning: after this point, iph might be no longer valid */
1326 	if (unlikely(!pskb_may_pull(skb, ihl)))
1327 		goto out;
1328 	__skb_pull(skb, ihl);
1329 
1330 	encap = SKB_GSO_CB(skb)->encap_level > 0;
1331 	if (encap)
1332 		features &= skb->dev->hw_enc_features;
1333 	SKB_GSO_CB(skb)->encap_level += ihl;
1334 
1335 	skb_reset_transport_header(skb);
1336 
1337 	segs = ERR_PTR(-EPROTONOSUPPORT);
1338 
1339 	if (!skb->encapsulation || encap) {
1340 		udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1341 		fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1342 
1343 		/* fixed ID is invalid if DF bit is not set */
1344 		if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1345 			goto out;
1346 	}
1347 
1348 	ops = rcu_dereference(inet_offloads[proto]);
1349 	if (likely(ops && ops->callbacks.gso_segment))
1350 		segs = ops->callbacks.gso_segment(skb, features);
1351 
1352 	if (IS_ERR_OR_NULL(segs))
1353 		goto out;
1354 
1355 	gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1356 
1357 	skb = segs;
1358 	do {
1359 		iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1360 		if (udpfrag) {
1361 			iph->frag_off = htons(offset >> 3);
1362 			if (skb->next)
1363 				iph->frag_off |= htons(IP_MF);
1364 			offset += skb->len - nhoff - ihl;
1365 			tot_len = skb->len - nhoff;
1366 		} else if (skb_is_gso(skb)) {
1367 			if (!fixedid) {
1368 				iph->id = htons(id);
1369 				id += skb_shinfo(skb)->gso_segs;
1370 			}
1371 
1372 			if (gso_partial)
1373 				tot_len = skb_shinfo(skb)->gso_size +
1374 					  SKB_GSO_CB(skb)->data_offset +
1375 					  skb->head - (unsigned char *)iph;
1376 			else
1377 				tot_len = skb->len - nhoff;
1378 		} else {
1379 			if (!fixedid)
1380 				iph->id = htons(id++);
1381 			tot_len = skb->len - nhoff;
1382 		}
1383 		iph->tot_len = htons(tot_len);
1384 		ip_send_check(iph);
1385 		if (encap)
1386 			skb_reset_inner_headers(skb);
1387 		skb->network_header = (u8 *)iph - skb->head;
1388 		skb_reset_mac_len(skb);
1389 	} while ((skb = skb->next));
1390 
1391 out:
1392 	return segs;
1393 }
1394 EXPORT_SYMBOL(inet_gso_segment);
1395 
1396 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1397 					netdev_features_t features)
1398 {
1399 	if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1400 		return ERR_PTR(-EINVAL);
1401 
1402 	return inet_gso_segment(skb, features);
1403 }
1404 
1405 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *,
1406 							   struct sk_buff *));
1407 INDIRECT_CALLABLE_DECLARE(struct sk_buff *udp4_gro_receive(struct list_head *,
1408 							   struct sk_buff *));
1409 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1410 {
1411 	const struct net_offload *ops;
1412 	struct sk_buff *pp = NULL;
1413 	const struct iphdr *iph;
1414 	struct sk_buff *p;
1415 	unsigned int hlen;
1416 	unsigned int off;
1417 	unsigned int id;
1418 	int flush = 1;
1419 	int proto;
1420 
1421 	off = skb_gro_offset(skb);
1422 	hlen = off + sizeof(*iph);
1423 	iph = skb_gro_header_fast(skb, off);
1424 	if (skb_gro_header_hard(skb, hlen)) {
1425 		iph = skb_gro_header_slow(skb, hlen, off);
1426 		if (unlikely(!iph))
1427 			goto out;
1428 	}
1429 
1430 	proto = iph->protocol;
1431 
1432 	rcu_read_lock();
1433 	ops = rcu_dereference(inet_offloads[proto]);
1434 	if (!ops || !ops->callbacks.gro_receive)
1435 		goto out_unlock;
1436 
1437 	if (*(u8 *)iph != 0x45)
1438 		goto out_unlock;
1439 
1440 	if (ip_is_fragment(iph))
1441 		goto out_unlock;
1442 
1443 	if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1444 		goto out_unlock;
1445 
1446 	id = ntohl(*(__be32 *)&iph->id);
1447 	flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1448 	id >>= 16;
1449 
1450 	list_for_each_entry(p, head, list) {
1451 		struct iphdr *iph2;
1452 		u16 flush_id;
1453 
1454 		if (!NAPI_GRO_CB(p)->same_flow)
1455 			continue;
1456 
1457 		iph2 = (struct iphdr *)(p->data + off);
1458 		/* The above works because, with the exception of the top
1459 		 * (inner most) layer, we only aggregate pkts with the same
1460 		 * hdr length so all the hdrs we'll need to verify will start
1461 		 * at the same offset.
1462 		 */
1463 		if ((iph->protocol ^ iph2->protocol) |
1464 		    ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1465 		    ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1466 			NAPI_GRO_CB(p)->same_flow = 0;
1467 			continue;
1468 		}
1469 
1470 		/* All fields must match except length and checksum. */
1471 		NAPI_GRO_CB(p)->flush |=
1472 			(iph->ttl ^ iph2->ttl) |
1473 			(iph->tos ^ iph2->tos) |
1474 			((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1475 
1476 		NAPI_GRO_CB(p)->flush |= flush;
1477 
1478 		/* We need to store of the IP ID check to be included later
1479 		 * when we can verify that this packet does in fact belong
1480 		 * to a given flow.
1481 		 */
1482 		flush_id = (u16)(id - ntohs(iph2->id));
1483 
1484 		/* This bit of code makes it much easier for us to identify
1485 		 * the cases where we are doing atomic vs non-atomic IP ID
1486 		 * checks.  Specifically an atomic check can return IP ID
1487 		 * values 0 - 0xFFFF, while a non-atomic check can only
1488 		 * return 0 or 0xFFFF.
1489 		 */
1490 		if (!NAPI_GRO_CB(p)->is_atomic ||
1491 		    !(iph->frag_off & htons(IP_DF))) {
1492 			flush_id ^= NAPI_GRO_CB(p)->count;
1493 			flush_id = flush_id ? 0xFFFF : 0;
1494 		}
1495 
1496 		/* If the previous IP ID value was based on an atomic
1497 		 * datagram we can overwrite the value and ignore it.
1498 		 */
1499 		if (NAPI_GRO_CB(skb)->is_atomic)
1500 			NAPI_GRO_CB(p)->flush_id = flush_id;
1501 		else
1502 			NAPI_GRO_CB(p)->flush_id |= flush_id;
1503 	}
1504 
1505 	NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1506 	NAPI_GRO_CB(skb)->flush |= flush;
1507 	skb_set_network_header(skb, off);
1508 	/* The above will be needed by the transport layer if there is one
1509 	 * immediately following this IP hdr.
1510 	 */
1511 
1512 	/* Note : No need to call skb_gro_postpull_rcsum() here,
1513 	 * as we already checked checksum over ipv4 header was 0
1514 	 */
1515 	skb_gro_pull(skb, sizeof(*iph));
1516 	skb_set_transport_header(skb, skb_gro_offset(skb));
1517 
1518 	pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1519 				       ops->callbacks.gro_receive, head, skb);
1520 
1521 out_unlock:
1522 	rcu_read_unlock();
1523 
1524 out:
1525 	skb_gro_flush_final(skb, pp, flush);
1526 
1527 	return pp;
1528 }
1529 EXPORT_SYMBOL(inet_gro_receive);
1530 
1531 static struct sk_buff *ipip_gro_receive(struct list_head *head,
1532 					struct sk_buff *skb)
1533 {
1534 	if (NAPI_GRO_CB(skb)->encap_mark) {
1535 		NAPI_GRO_CB(skb)->flush = 1;
1536 		return NULL;
1537 	}
1538 
1539 	NAPI_GRO_CB(skb)->encap_mark = 1;
1540 
1541 	return inet_gro_receive(head, skb);
1542 }
1543 
1544 #define SECONDS_PER_DAY	86400
1545 
1546 /* inet_current_timestamp - Return IP network timestamp
1547  *
1548  * Return milliseconds since midnight in network byte order.
1549  */
1550 __be32 inet_current_timestamp(void)
1551 {
1552 	u32 secs;
1553 	u32 msecs;
1554 	struct timespec64 ts;
1555 
1556 	ktime_get_real_ts64(&ts);
1557 
1558 	/* Get secs since midnight. */
1559 	(void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1560 	/* Convert to msecs. */
1561 	msecs = secs * MSEC_PER_SEC;
1562 	/* Convert nsec to msec. */
1563 	msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1564 
1565 	/* Convert to network byte order. */
1566 	return htonl(msecs);
1567 }
1568 EXPORT_SYMBOL(inet_current_timestamp);
1569 
1570 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1571 {
1572 	if (sk->sk_family == AF_INET)
1573 		return ip_recv_error(sk, msg, len, addr_len);
1574 #if IS_ENABLED(CONFIG_IPV6)
1575 	if (sk->sk_family == AF_INET6)
1576 		return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1577 #endif
1578 	return -EINVAL;
1579 }
1580 
1581 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *, int));
1582 INDIRECT_CALLABLE_DECLARE(int udp4_gro_complete(struct sk_buff *, int));
1583 int inet_gro_complete(struct sk_buff *skb, int nhoff)
1584 {
1585 	__be16 newlen = htons(skb->len - nhoff);
1586 	struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1587 	const struct net_offload *ops;
1588 	int proto = iph->protocol;
1589 	int err = -ENOSYS;
1590 
1591 	if (skb->encapsulation) {
1592 		skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1593 		skb_set_inner_network_header(skb, nhoff);
1594 	}
1595 
1596 	csum_replace2(&iph->check, iph->tot_len, newlen);
1597 	iph->tot_len = newlen;
1598 
1599 	rcu_read_lock();
1600 	ops = rcu_dereference(inet_offloads[proto]);
1601 	if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1602 		goto out_unlock;
1603 
1604 	/* Only need to add sizeof(*iph) to get to the next hdr below
1605 	 * because any hdr with option will have been flushed in
1606 	 * inet_gro_receive().
1607 	 */
1608 	err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1609 			      tcp4_gro_complete, udp4_gro_complete,
1610 			      skb, nhoff + sizeof(*iph));
1611 
1612 out_unlock:
1613 	rcu_read_unlock();
1614 
1615 	return err;
1616 }
1617 EXPORT_SYMBOL(inet_gro_complete);
1618 
1619 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1620 {
1621 	skb->encapsulation = 1;
1622 	skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1623 	return inet_gro_complete(skb, nhoff);
1624 }
1625 
1626 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1627 			 unsigned short type, unsigned char protocol,
1628 			 struct net *net)
1629 {
1630 	struct socket *sock;
1631 	int rc = sock_create_kern(net, family, type, protocol, &sock);
1632 
1633 	if (rc == 0) {
1634 		*sk = sock->sk;
1635 		(*sk)->sk_allocation = GFP_ATOMIC;
1636 		/*
1637 		 * Unhash it so that IP input processing does not even see it,
1638 		 * we do not wish this socket to see incoming packets.
1639 		 */
1640 		(*sk)->sk_prot->unhash(*sk);
1641 	}
1642 	return rc;
1643 }
1644 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1645 
1646 u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt)
1647 {
1648 	return  *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt);
1649 }
1650 EXPORT_SYMBOL_GPL(snmp_get_cpu_field);
1651 
1652 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1653 {
1654 	unsigned long res = 0;
1655 	int i;
1656 
1657 	for_each_possible_cpu(i)
1658 		res += snmp_get_cpu_field(mib, i, offt);
1659 	return res;
1660 }
1661 EXPORT_SYMBOL_GPL(snmp_fold_field);
1662 
1663 #if BITS_PER_LONG==32
1664 
1665 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1666 			 size_t syncp_offset)
1667 {
1668 	void *bhptr;
1669 	struct u64_stats_sync *syncp;
1670 	u64 v;
1671 	unsigned int start;
1672 
1673 	bhptr = per_cpu_ptr(mib, cpu);
1674 	syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1675 	do {
1676 		start = u64_stats_fetch_begin_irq(syncp);
1677 		v = *(((u64 *)bhptr) + offt);
1678 	} while (u64_stats_fetch_retry_irq(syncp, start));
1679 
1680 	return v;
1681 }
1682 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1683 
1684 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1685 {
1686 	u64 res = 0;
1687 	int cpu;
1688 
1689 	for_each_possible_cpu(cpu) {
1690 		res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1691 	}
1692 	return res;
1693 }
1694 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1695 #endif
1696 
1697 #ifdef CONFIG_IP_MULTICAST
1698 static const struct net_protocol igmp_protocol = {
1699 	.handler =	igmp_rcv,
1700 	.netns_ok =	1,
1701 };
1702 #endif
1703 
1704 /* thinking of making this const? Don't.
1705  * early_demux can change based on sysctl.
1706  */
1707 static struct net_protocol tcp_protocol = {
1708 	.early_demux	=	tcp_v4_early_demux,
1709 	.early_demux_handler =  tcp_v4_early_demux,
1710 	.handler	=	tcp_v4_rcv,
1711 	.err_handler	=	tcp_v4_err,
1712 	.no_policy	=	1,
1713 	.netns_ok	=	1,
1714 	.icmp_strict_tag_validation = 1,
1715 };
1716 
1717 /* thinking of making this const? Don't.
1718  * early_demux can change based on sysctl.
1719  */
1720 static struct net_protocol udp_protocol = {
1721 	.early_demux =	udp_v4_early_demux,
1722 	.early_demux_handler =	udp_v4_early_demux,
1723 	.handler =	udp_rcv,
1724 	.err_handler =	udp_err,
1725 	.no_policy =	1,
1726 	.netns_ok =	1,
1727 };
1728 
1729 static const struct net_protocol icmp_protocol = {
1730 	.handler =	icmp_rcv,
1731 	.err_handler =	icmp_err,
1732 	.no_policy =	1,
1733 	.netns_ok =	1,
1734 };
1735 
1736 static __net_init int ipv4_mib_init_net(struct net *net)
1737 {
1738 	int i;
1739 
1740 	net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1741 	if (!net->mib.tcp_statistics)
1742 		goto err_tcp_mib;
1743 	net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1744 	if (!net->mib.ip_statistics)
1745 		goto err_ip_mib;
1746 
1747 	for_each_possible_cpu(i) {
1748 		struct ipstats_mib *af_inet_stats;
1749 		af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1750 		u64_stats_init(&af_inet_stats->syncp);
1751 	}
1752 
1753 	net->mib.net_statistics = alloc_percpu(struct linux_mib);
1754 	if (!net->mib.net_statistics)
1755 		goto err_net_mib;
1756 	net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1757 	if (!net->mib.udp_statistics)
1758 		goto err_udp_mib;
1759 	net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1760 	if (!net->mib.udplite_statistics)
1761 		goto err_udplite_mib;
1762 	net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1763 	if (!net->mib.icmp_statistics)
1764 		goto err_icmp_mib;
1765 	net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1766 					      GFP_KERNEL);
1767 	if (!net->mib.icmpmsg_statistics)
1768 		goto err_icmpmsg_mib;
1769 
1770 	tcp_mib_init(net);
1771 	return 0;
1772 
1773 err_icmpmsg_mib:
1774 	free_percpu(net->mib.icmp_statistics);
1775 err_icmp_mib:
1776 	free_percpu(net->mib.udplite_statistics);
1777 err_udplite_mib:
1778 	free_percpu(net->mib.udp_statistics);
1779 err_udp_mib:
1780 	free_percpu(net->mib.net_statistics);
1781 err_net_mib:
1782 	free_percpu(net->mib.ip_statistics);
1783 err_ip_mib:
1784 	free_percpu(net->mib.tcp_statistics);
1785 err_tcp_mib:
1786 	return -ENOMEM;
1787 }
1788 
1789 static __net_exit void ipv4_mib_exit_net(struct net *net)
1790 {
1791 	kfree(net->mib.icmpmsg_statistics);
1792 	free_percpu(net->mib.icmp_statistics);
1793 	free_percpu(net->mib.udplite_statistics);
1794 	free_percpu(net->mib.udp_statistics);
1795 	free_percpu(net->mib.net_statistics);
1796 	free_percpu(net->mib.ip_statistics);
1797 	free_percpu(net->mib.tcp_statistics);
1798 #ifdef CONFIG_MPTCP
1799 	/* allocated on demand, see mptcp_init_sock() */
1800 	free_percpu(net->mib.mptcp_statistics);
1801 #endif
1802 }
1803 
1804 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1805 	.init = ipv4_mib_init_net,
1806 	.exit = ipv4_mib_exit_net,
1807 };
1808 
1809 static int __init init_ipv4_mibs(void)
1810 {
1811 	return register_pernet_subsys(&ipv4_mib_ops);
1812 }
1813 
1814 static __net_init int inet_init_net(struct net *net)
1815 {
1816 	/*
1817 	 * Set defaults for local port range
1818 	 */
1819 	seqlock_init(&net->ipv4.ip_local_ports.lock);
1820 	net->ipv4.ip_local_ports.range[0] =  32768;
1821 	net->ipv4.ip_local_ports.range[1] =  60999;
1822 
1823 	seqlock_init(&net->ipv4.ping_group_range.lock);
1824 	/*
1825 	 * Sane defaults - nobody may create ping sockets.
1826 	 * Boot scripts should set this to distro-specific group.
1827 	 */
1828 	net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1829 	net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1830 
1831 	/* Default values for sysctl-controlled parameters.
1832 	 * We set them here, in case sysctl is not compiled.
1833 	 */
1834 	net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1835 	net->ipv4.sysctl_ip_fwd_update_priority = 1;
1836 	net->ipv4.sysctl_ip_dynaddr = 0;
1837 	net->ipv4.sysctl_ip_early_demux = 1;
1838 	net->ipv4.sysctl_udp_early_demux = 1;
1839 	net->ipv4.sysctl_tcp_early_demux = 1;
1840 	net->ipv4.sysctl_nexthop_compat_mode = 1;
1841 #ifdef CONFIG_SYSCTL
1842 	net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1843 #endif
1844 
1845 	/* Some igmp sysctl, whose values are always used */
1846 	net->ipv4.sysctl_igmp_max_memberships = 20;
1847 	net->ipv4.sysctl_igmp_max_msf = 10;
1848 	/* IGMP reports for link-local multicast groups are enabled by default */
1849 	net->ipv4.sysctl_igmp_llm_reports = 1;
1850 	net->ipv4.sysctl_igmp_qrv = 2;
1851 
1852 	return 0;
1853 }
1854 
1855 static __net_initdata struct pernet_operations af_inet_ops = {
1856 	.init = inet_init_net,
1857 };
1858 
1859 static int __init init_inet_pernet_ops(void)
1860 {
1861 	return register_pernet_subsys(&af_inet_ops);
1862 }
1863 
1864 static int ipv4_proc_init(void);
1865 
1866 /*
1867  *	IP protocol layer initialiser
1868  */
1869 
1870 static struct packet_offload ip_packet_offload __read_mostly = {
1871 	.type = cpu_to_be16(ETH_P_IP),
1872 	.callbacks = {
1873 		.gso_segment = inet_gso_segment,
1874 		.gro_receive = inet_gro_receive,
1875 		.gro_complete = inet_gro_complete,
1876 	},
1877 };
1878 
1879 static const struct net_offload ipip_offload = {
1880 	.callbacks = {
1881 		.gso_segment	= ipip_gso_segment,
1882 		.gro_receive	= ipip_gro_receive,
1883 		.gro_complete	= ipip_gro_complete,
1884 	},
1885 };
1886 
1887 static int __init ipip_offload_init(void)
1888 {
1889 	return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1890 }
1891 
1892 static int __init ipv4_offload_init(void)
1893 {
1894 	/*
1895 	 * Add offloads
1896 	 */
1897 	if (udpv4_offload_init() < 0)
1898 		pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1899 	if (tcpv4_offload_init() < 0)
1900 		pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1901 	if (ipip_offload_init() < 0)
1902 		pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1903 
1904 	dev_add_offload(&ip_packet_offload);
1905 	return 0;
1906 }
1907 
1908 fs_initcall(ipv4_offload_init);
1909 
1910 static struct packet_type ip_packet_type __read_mostly = {
1911 	.type = cpu_to_be16(ETH_P_IP),
1912 	.func = ip_rcv,
1913 	.list_func = ip_list_rcv,
1914 };
1915 
1916 static int __init inet_init(void)
1917 {
1918 	struct inet_protosw *q;
1919 	struct list_head *r;
1920 	int rc;
1921 
1922 	sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1923 
1924 	rc = proto_register(&tcp_prot, 1);
1925 	if (rc)
1926 		goto out;
1927 
1928 	rc = proto_register(&udp_prot, 1);
1929 	if (rc)
1930 		goto out_unregister_tcp_proto;
1931 
1932 	rc = proto_register(&raw_prot, 1);
1933 	if (rc)
1934 		goto out_unregister_udp_proto;
1935 
1936 	rc = proto_register(&ping_prot, 1);
1937 	if (rc)
1938 		goto out_unregister_raw_proto;
1939 
1940 	/*
1941 	 *	Tell SOCKET that we are alive...
1942 	 */
1943 
1944 	(void)sock_register(&inet_family_ops);
1945 
1946 #ifdef CONFIG_SYSCTL
1947 	ip_static_sysctl_init();
1948 #endif
1949 
1950 	/*
1951 	 *	Add all the base protocols.
1952 	 */
1953 
1954 	if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1955 		pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1956 	if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1957 		pr_crit("%s: Cannot add UDP protocol\n", __func__);
1958 	if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1959 		pr_crit("%s: Cannot add TCP protocol\n", __func__);
1960 #ifdef CONFIG_IP_MULTICAST
1961 	if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1962 		pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1963 #endif
1964 
1965 	/* Register the socket-side information for inet_create. */
1966 	for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1967 		INIT_LIST_HEAD(r);
1968 
1969 	for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1970 		inet_register_protosw(q);
1971 
1972 	/*
1973 	 *	Set the ARP module up
1974 	 */
1975 
1976 	arp_init();
1977 
1978 	/*
1979 	 *	Set the IP module up
1980 	 */
1981 
1982 	ip_init();
1983 
1984 	/* Setup TCP slab cache for open requests. */
1985 	tcp_init();
1986 
1987 	/* Setup UDP memory threshold */
1988 	udp_init();
1989 
1990 	/* Add UDP-Lite (RFC 3828) */
1991 	udplite4_register();
1992 
1993 	raw_init();
1994 
1995 	ping_init();
1996 
1997 	/*
1998 	 *	Set the ICMP layer up
1999 	 */
2000 
2001 	if (icmp_init() < 0)
2002 		panic("Failed to create the ICMP control socket.\n");
2003 
2004 	/*
2005 	 *	Initialise the multicast router
2006 	 */
2007 #if defined(CONFIG_IP_MROUTE)
2008 	if (ip_mr_init())
2009 		pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2010 #endif
2011 
2012 	if (init_inet_pernet_ops())
2013 		pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2014 	/*
2015 	 *	Initialise per-cpu ipv4 mibs
2016 	 */
2017 
2018 	if (init_ipv4_mibs())
2019 		pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
2020 
2021 	ipv4_proc_init();
2022 
2023 	ipfrag_init();
2024 
2025 	dev_add_pack(&ip_packet_type);
2026 
2027 	ip_tunnel_core_init();
2028 
2029 	rc = 0;
2030 out:
2031 	return rc;
2032 out_unregister_raw_proto:
2033 	proto_unregister(&raw_prot);
2034 out_unregister_udp_proto:
2035 	proto_unregister(&udp_prot);
2036 out_unregister_tcp_proto:
2037 	proto_unregister(&tcp_prot);
2038 	goto out;
2039 }
2040 
2041 fs_initcall(inet_init);
2042 
2043 /* ------------------------------------------------------------------------ */
2044 
2045 #ifdef CONFIG_PROC_FS
2046 static int __init ipv4_proc_init(void)
2047 {
2048 	int rc = 0;
2049 
2050 	if (raw_proc_init())
2051 		goto out_raw;
2052 	if (tcp4_proc_init())
2053 		goto out_tcp;
2054 	if (udp4_proc_init())
2055 		goto out_udp;
2056 	if (ping_proc_init())
2057 		goto out_ping;
2058 	if (ip_misc_proc_init())
2059 		goto out_misc;
2060 out:
2061 	return rc;
2062 out_misc:
2063 	ping_proc_exit();
2064 out_ping:
2065 	udp4_proc_exit();
2066 out_udp:
2067 	tcp4_proc_exit();
2068 out_tcp:
2069 	raw_proc_exit();
2070 out_raw:
2071 	rc = -ENOMEM;
2072 	goto out;
2073 }
2074 
2075 #else /* CONFIG_PROC_FS */
2076 static int __init ipv4_proc_init(void)
2077 {
2078 	return 0;
2079 }
2080 #endif /* CONFIG_PROC_FS */
2081