xref: /openbmc/linux/net/ipv4/af_inet.c (revision 92a2c6b2)
1 /*
2  * INET		An implementation of the TCP/IP protocol suite for the LINUX
3  *		operating system.  INET is implemented using the  BSD Socket
4  *		interface as the means of communication with the user level.
5  *
6  *		PF_INET protocol family socket handler.
7  *
8  * Authors:	Ross Biro
9  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10  *		Florian La Roche, <flla@stud.uni-sb.de>
11  *		Alan Cox, <A.Cox@swansea.ac.uk>
12  *
13  * Changes (see also sock.c)
14  *
15  *		piggy,
16  *		Karl Knutson	:	Socket protocol table
17  *		A.N.Kuznetsov	:	Socket death error in accept().
18  *		John Richardson :	Fix non blocking error in connect()
19  *					so sockets that fail to connect
20  *					don't return -EINPROGRESS.
21  *		Alan Cox	:	Asynchronous I/O support
22  *		Alan Cox	:	Keep correct socket pointer on sock
23  *					structures
24  *					when accept() ed
25  *		Alan Cox	:	Semantics of SO_LINGER aren't state
26  *					moved to close when you look carefully.
27  *					With this fixed and the accept bug fixed
28  *					some RPC stuff seems happier.
29  *		Niibe Yutaka	:	4.4BSD style write async I/O
30  *		Alan Cox,
31  *		Tony Gale 	:	Fixed reuse semantics.
32  *		Alan Cox	:	bind() shouldn't abort existing but dead
33  *					sockets. Stops FTP netin:.. I hope.
34  *		Alan Cox	:	bind() works correctly for RAW sockets.
35  *					Note that FreeBSD at least was broken
36  *					in this respect so be careful with
37  *					compatibility tests...
38  *		Alan Cox	:	routing cache support
39  *		Alan Cox	:	memzero the socket structure for
40  *					compactness.
41  *		Matt Day	:	nonblock connect error handler
42  *		Alan Cox	:	Allow large numbers of pending sockets
43  *					(eg for big web sites), but only if
44  *					specifically application requested.
45  *		Alan Cox	:	New buffering throughout IP. Used
46  *					dumbly.
47  *		Alan Cox	:	New buffering now used smartly.
48  *		Alan Cox	:	BSD rather than common sense
49  *					interpretation of listen.
50  *		Germano Caronni	:	Assorted small races.
51  *		Alan Cox	:	sendmsg/recvmsg basic support.
52  *		Alan Cox	:	Only sendmsg/recvmsg now supported.
53  *		Alan Cox	:	Locked down bind (see security list).
54  *		Alan Cox	:	Loosened bind a little.
55  *		Mike McLagan	:	ADD/DEL DLCI Ioctls
56  *	Willy Konynenberg	:	Transparent proxying support.
57  *		David S. Miller	:	New socket lookup architecture.
58  *					Some other random speedups.
59  *		Cyrus Durgin	:	Cleaned up file for kmod hacks.
60  *		Andi Kleen	:	Fix inet_stream_connect TCP race.
61  *
62  *		This program is free software; you can redistribute it and/or
63  *		modify it under the terms of the GNU General Public License
64  *		as published by the Free Software Foundation; either version
65  *		2 of the License, or (at your option) any later version.
66  */
67 
68 #define pr_fmt(fmt) "IPv4: " fmt
69 
70 #include <linux/err.h>
71 #include <linux/errno.h>
72 #include <linux/types.h>
73 #include <linux/socket.h>
74 #include <linux/in.h>
75 #include <linux/kernel.h>
76 #include <linux/module.h>
77 #include <linux/sched.h>
78 #include <linux/timer.h>
79 #include <linux/string.h>
80 #include <linux/sockios.h>
81 #include <linux/net.h>
82 #include <linux/capability.h>
83 #include <linux/fcntl.h>
84 #include <linux/mm.h>
85 #include <linux/interrupt.h>
86 #include <linux/stat.h>
87 #include <linux/init.h>
88 #include <linux/poll.h>
89 #include <linux/netfilter_ipv4.h>
90 #include <linux/random.h>
91 #include <linux/slab.h>
92 
93 #include <asm/uaccess.h>
94 
95 #include <linux/inet.h>
96 #include <linux/igmp.h>
97 #include <linux/inetdevice.h>
98 #include <linux/netdevice.h>
99 #include <net/checksum.h>
100 #include <net/ip.h>
101 #include <net/protocol.h>
102 #include <net/arp.h>
103 #include <net/route.h>
104 #include <net/ip_fib.h>
105 #include <net/inet_connection_sock.h>
106 #include <net/tcp.h>
107 #include <net/udp.h>
108 #include <net/udplite.h>
109 #include <net/ping.h>
110 #include <linux/skbuff.h>
111 #include <net/sock.h>
112 #include <net/raw.h>
113 #include <net/icmp.h>
114 #include <net/inet_common.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 
122 
123 /* The inetsw table contains everything that inet_create needs to
124  * build a new socket.
125  */
126 static struct list_head inetsw[SOCK_MAX];
127 static DEFINE_SPINLOCK(inetsw_lock);
128 
129 /* New destruction routine */
130 
131 void inet_sock_destruct(struct sock *sk)
132 {
133 	struct inet_sock *inet = inet_sk(sk);
134 
135 	__skb_queue_purge(&sk->sk_receive_queue);
136 	__skb_queue_purge(&sk->sk_error_queue);
137 
138 	sk_mem_reclaim(sk);
139 
140 	if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
141 		pr_err("Attempt to release TCP socket in state %d %p\n",
142 		       sk->sk_state, sk);
143 		return;
144 	}
145 	if (!sock_flag(sk, SOCK_DEAD)) {
146 		pr_err("Attempt to release alive inet socket %p\n", sk);
147 		return;
148 	}
149 
150 	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
151 	WARN_ON(atomic_read(&sk->sk_wmem_alloc));
152 	WARN_ON(sk->sk_wmem_queued);
153 	WARN_ON(sk->sk_forward_alloc);
154 
155 	kfree(rcu_dereference_protected(inet->inet_opt, 1));
156 	dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
157 	dst_release(sk->sk_rx_dst);
158 	sk_refcnt_debug_dec(sk);
159 }
160 EXPORT_SYMBOL(inet_sock_destruct);
161 
162 /*
163  *	The routines beyond this point handle the behaviour of an AF_INET
164  *	socket object. Mostly it punts to the subprotocols of IP to do
165  *	the work.
166  */
167 
168 /*
169  *	Automatically bind an unbound socket.
170  */
171 
172 static int inet_autobind(struct sock *sk)
173 {
174 	struct inet_sock *inet;
175 	/* We may need to bind the socket. */
176 	lock_sock(sk);
177 	inet = inet_sk(sk);
178 	if (!inet->inet_num) {
179 		if (sk->sk_prot->get_port(sk, 0)) {
180 			release_sock(sk);
181 			return -EAGAIN;
182 		}
183 		inet->inet_sport = htons(inet->inet_num);
184 	}
185 	release_sock(sk);
186 	return 0;
187 }
188 
189 /*
190  *	Move a socket into listening state.
191  */
192 int inet_listen(struct socket *sock, int backlog)
193 {
194 	struct sock *sk = sock->sk;
195 	unsigned char old_state;
196 	int err;
197 
198 	lock_sock(sk);
199 
200 	err = -EINVAL;
201 	if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
202 		goto out;
203 
204 	old_state = sk->sk_state;
205 	if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
206 		goto out;
207 
208 	/* Really, if the socket is already in listen state
209 	 * we can only allow the backlog to be adjusted.
210 	 */
211 	if (old_state != TCP_LISTEN) {
212 		/* Check special setups for testing purpose to enable TFO w/o
213 		 * requiring TCP_FASTOPEN sockopt.
214 		 * Note that only TCP sockets (SOCK_STREAM) will reach here.
215 		 * Also fastopenq may already been allocated because this
216 		 * socket was in TCP_LISTEN state previously but was
217 		 * shutdown() (rather than close()).
218 		 */
219 		if ((sysctl_tcp_fastopen & TFO_SERVER_ENABLE) != 0 &&
220 		    inet_csk(sk)->icsk_accept_queue.fastopenq == NULL) {
221 			if ((sysctl_tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) != 0)
222 				err = fastopen_init_queue(sk, backlog);
223 			else if ((sysctl_tcp_fastopen &
224 				  TFO_SERVER_WO_SOCKOPT2) != 0)
225 				err = fastopen_init_queue(sk,
226 				    ((uint)sysctl_tcp_fastopen) >> 16);
227 			else
228 				err = 0;
229 			if (err)
230 				goto out;
231 		}
232 		err = inet_csk_listen_start(sk, backlog);
233 		if (err)
234 			goto out;
235 	}
236 	sk->sk_max_ack_backlog = backlog;
237 	err = 0;
238 
239 out:
240 	release_sock(sk);
241 	return err;
242 }
243 EXPORT_SYMBOL(inet_listen);
244 
245 /*
246  *	Create an inet socket.
247  */
248 
249 static int inet_create(struct net *net, struct socket *sock, int protocol,
250 		       int kern)
251 {
252 	struct sock *sk;
253 	struct inet_protosw *answer;
254 	struct inet_sock *inet;
255 	struct proto *answer_prot;
256 	unsigned char answer_flags;
257 	int try_loading_module = 0;
258 	int err;
259 
260 	sock->state = SS_UNCONNECTED;
261 
262 	/* Look for the requested type/protocol pair. */
263 lookup_protocol:
264 	err = -ESOCKTNOSUPPORT;
265 	rcu_read_lock();
266 	list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
267 
268 		err = 0;
269 		/* Check the non-wild match. */
270 		if (protocol == answer->protocol) {
271 			if (protocol != IPPROTO_IP)
272 				break;
273 		} else {
274 			/* Check for the two wild cases. */
275 			if (IPPROTO_IP == protocol) {
276 				protocol = answer->protocol;
277 				break;
278 			}
279 			if (IPPROTO_IP == answer->protocol)
280 				break;
281 		}
282 		err = -EPROTONOSUPPORT;
283 	}
284 
285 	if (unlikely(err)) {
286 		if (try_loading_module < 2) {
287 			rcu_read_unlock();
288 			/*
289 			 * Be more specific, e.g. net-pf-2-proto-132-type-1
290 			 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
291 			 */
292 			if (++try_loading_module == 1)
293 				request_module("net-pf-%d-proto-%d-type-%d",
294 					       PF_INET, protocol, sock->type);
295 			/*
296 			 * Fall back to generic, e.g. net-pf-2-proto-132
297 			 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
298 			 */
299 			else
300 				request_module("net-pf-%d-proto-%d",
301 					       PF_INET, protocol);
302 			goto lookup_protocol;
303 		} else
304 			goto out_rcu_unlock;
305 	}
306 
307 	err = -EPERM;
308 	if (sock->type == SOCK_RAW && !kern &&
309 	    !ns_capable(net->user_ns, CAP_NET_RAW))
310 		goto out_rcu_unlock;
311 
312 	sock->ops = answer->ops;
313 	answer_prot = answer->prot;
314 	answer_flags = answer->flags;
315 	rcu_read_unlock();
316 
317 	WARN_ON(answer_prot->slab == NULL);
318 
319 	err = -ENOBUFS;
320 	sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
321 	if (sk == NULL)
322 		goto out;
323 
324 	err = 0;
325 	if (INET_PROTOSW_REUSE & answer_flags)
326 		sk->sk_reuse = SK_CAN_REUSE;
327 
328 	inet = inet_sk(sk);
329 	inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
330 
331 	inet->nodefrag = 0;
332 
333 	if (SOCK_RAW == sock->type) {
334 		inet->inet_num = protocol;
335 		if (IPPROTO_RAW == protocol)
336 			inet->hdrincl = 1;
337 	}
338 
339 	if (net->ipv4.sysctl_ip_no_pmtu_disc)
340 		inet->pmtudisc = IP_PMTUDISC_DONT;
341 	else
342 		inet->pmtudisc = IP_PMTUDISC_WANT;
343 
344 	inet->inet_id = 0;
345 
346 	sock_init_data(sock, sk);
347 
348 	sk->sk_destruct	   = inet_sock_destruct;
349 	sk->sk_protocol	   = protocol;
350 	sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
351 
352 	inet->uc_ttl	= -1;
353 	inet->mc_loop	= 1;
354 	inet->mc_ttl	= 1;
355 	inet->mc_all	= 1;
356 	inet->mc_index	= 0;
357 	inet->mc_list	= NULL;
358 	inet->rcv_tos	= 0;
359 
360 	sk_refcnt_debug_inc(sk);
361 
362 	if (inet->inet_num) {
363 		/* It assumes that any protocol which allows
364 		 * the user to assign a number at socket
365 		 * creation time automatically
366 		 * shares.
367 		 */
368 		inet->inet_sport = htons(inet->inet_num);
369 		/* Add to protocol hash chains. */
370 		sk->sk_prot->hash(sk);
371 	}
372 
373 	if (sk->sk_prot->init) {
374 		err = sk->sk_prot->init(sk);
375 		if (err)
376 			sk_common_release(sk);
377 	}
378 out:
379 	return err;
380 out_rcu_unlock:
381 	rcu_read_unlock();
382 	goto out;
383 }
384 
385 
386 /*
387  *	The peer socket should always be NULL (or else). When we call this
388  *	function we are destroying the object and from then on nobody
389  *	should refer to it.
390  */
391 int inet_release(struct socket *sock)
392 {
393 	struct sock *sk = sock->sk;
394 
395 	if (sk) {
396 		long timeout;
397 
398 		/* Applications forget to leave groups before exiting */
399 		ip_mc_drop_socket(sk);
400 
401 		/* If linger is set, we don't return until the close
402 		 * is complete.  Otherwise we return immediately. The
403 		 * actually closing is done the same either way.
404 		 *
405 		 * If the close is due to the process exiting, we never
406 		 * linger..
407 		 */
408 		timeout = 0;
409 		if (sock_flag(sk, SOCK_LINGER) &&
410 		    !(current->flags & PF_EXITING))
411 			timeout = sk->sk_lingertime;
412 		sock->sk = NULL;
413 		sk->sk_prot->close(sk, timeout);
414 	}
415 	return 0;
416 }
417 EXPORT_SYMBOL(inet_release);
418 
419 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
420 {
421 	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
422 	struct sock *sk = sock->sk;
423 	struct inet_sock *inet = inet_sk(sk);
424 	struct net *net = sock_net(sk);
425 	unsigned short snum;
426 	int chk_addr_ret;
427 	int err;
428 
429 	/* If the socket has its own bind function then use it. (RAW) */
430 	if (sk->sk_prot->bind) {
431 		err = sk->sk_prot->bind(sk, uaddr, addr_len);
432 		goto out;
433 	}
434 	err = -EINVAL;
435 	if (addr_len < sizeof(struct sockaddr_in))
436 		goto out;
437 
438 	if (addr->sin_family != AF_INET) {
439 		/* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
440 		 * only if s_addr is INADDR_ANY.
441 		 */
442 		err = -EAFNOSUPPORT;
443 		if (addr->sin_family != AF_UNSPEC ||
444 		    addr->sin_addr.s_addr != htonl(INADDR_ANY))
445 			goto out;
446 	}
447 
448 	chk_addr_ret = inet_addr_type(net, addr->sin_addr.s_addr);
449 
450 	/* Not specified by any standard per-se, however it breaks too
451 	 * many applications when removed.  It is unfortunate since
452 	 * allowing applications to make a non-local bind solves
453 	 * several problems with systems using dynamic addressing.
454 	 * (ie. your servers still start up even if your ISDN link
455 	 *  is temporarily down)
456 	 */
457 	err = -EADDRNOTAVAIL;
458 	if (!net->ipv4.sysctl_ip_nonlocal_bind &&
459 	    !(inet->freebind || inet->transparent) &&
460 	    addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
461 	    chk_addr_ret != RTN_LOCAL &&
462 	    chk_addr_ret != RTN_MULTICAST &&
463 	    chk_addr_ret != RTN_BROADCAST)
464 		goto out;
465 
466 	snum = ntohs(addr->sin_port);
467 	err = -EACCES;
468 	if (snum && snum < PROT_SOCK &&
469 	    !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
470 		goto out;
471 
472 	/*      We keep a pair of addresses. rcv_saddr is the one
473 	 *      used by hash lookups, and saddr is used for transmit.
474 	 *
475 	 *      In the BSD API these are the same except where it
476 	 *      would be illegal to use them (multicast/broadcast) in
477 	 *      which case the sending device address is used.
478 	 */
479 	lock_sock(sk);
480 
481 	/* Check these errors (active socket, double bind). */
482 	err = -EINVAL;
483 	if (sk->sk_state != TCP_CLOSE || inet->inet_num)
484 		goto out_release_sock;
485 
486 	inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
487 	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
488 		inet->inet_saddr = 0;  /* Use device */
489 
490 	/* Make sure we are allowed to bind here. */
491 	if (sk->sk_prot->get_port(sk, snum)) {
492 		inet->inet_saddr = inet->inet_rcv_saddr = 0;
493 		err = -EADDRINUSE;
494 		goto out_release_sock;
495 	}
496 
497 	if (inet->inet_rcv_saddr)
498 		sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
499 	if (snum)
500 		sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
501 	inet->inet_sport = htons(inet->inet_num);
502 	inet->inet_daddr = 0;
503 	inet->inet_dport = 0;
504 	sk_dst_reset(sk);
505 	err = 0;
506 out_release_sock:
507 	release_sock(sk);
508 out:
509 	return err;
510 }
511 EXPORT_SYMBOL(inet_bind);
512 
513 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
514 		       int addr_len, int flags)
515 {
516 	struct sock *sk = sock->sk;
517 
518 	if (addr_len < sizeof(uaddr->sa_family))
519 		return -EINVAL;
520 	if (uaddr->sa_family == AF_UNSPEC)
521 		return sk->sk_prot->disconnect(sk, flags);
522 
523 	if (!inet_sk(sk)->inet_num && inet_autobind(sk))
524 		return -EAGAIN;
525 	return sk->sk_prot->connect(sk, uaddr, addr_len);
526 }
527 EXPORT_SYMBOL(inet_dgram_connect);
528 
529 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
530 {
531 	DEFINE_WAIT(wait);
532 
533 	prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
534 	sk->sk_write_pending += writebias;
535 
536 	/* Basic assumption: if someone sets sk->sk_err, he _must_
537 	 * change state of the socket from TCP_SYN_*.
538 	 * Connect() does not allow to get error notifications
539 	 * without closing the socket.
540 	 */
541 	while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
542 		release_sock(sk);
543 		timeo = schedule_timeout(timeo);
544 		lock_sock(sk);
545 		if (signal_pending(current) || !timeo)
546 			break;
547 		prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
548 	}
549 	finish_wait(sk_sleep(sk), &wait);
550 	sk->sk_write_pending -= writebias;
551 	return timeo;
552 }
553 
554 /*
555  *	Connect to a remote host. There is regrettably still a little
556  *	TCP 'magic' in here.
557  */
558 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
559 			  int addr_len, int flags)
560 {
561 	struct sock *sk = sock->sk;
562 	int err;
563 	long timeo;
564 
565 	if (addr_len < sizeof(uaddr->sa_family))
566 		return -EINVAL;
567 
568 	if (uaddr->sa_family == AF_UNSPEC) {
569 		err = sk->sk_prot->disconnect(sk, flags);
570 		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
571 		goto out;
572 	}
573 
574 	switch (sock->state) {
575 	default:
576 		err = -EINVAL;
577 		goto out;
578 	case SS_CONNECTED:
579 		err = -EISCONN;
580 		goto out;
581 	case SS_CONNECTING:
582 		err = -EALREADY;
583 		/* Fall out of switch with err, set for this state */
584 		break;
585 	case SS_UNCONNECTED:
586 		err = -EISCONN;
587 		if (sk->sk_state != TCP_CLOSE)
588 			goto out;
589 
590 		err = sk->sk_prot->connect(sk, uaddr, addr_len);
591 		if (err < 0)
592 			goto out;
593 
594 		sock->state = SS_CONNECTING;
595 
596 		/* Just entered SS_CONNECTING state; the only
597 		 * difference is that return value in non-blocking
598 		 * case is EINPROGRESS, rather than EALREADY.
599 		 */
600 		err = -EINPROGRESS;
601 		break;
602 	}
603 
604 	timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
605 
606 	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
607 		int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
608 				tcp_sk(sk)->fastopen_req &&
609 				tcp_sk(sk)->fastopen_req->data ? 1 : 0;
610 
611 		/* Error code is set above */
612 		if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
613 			goto out;
614 
615 		err = sock_intr_errno(timeo);
616 		if (signal_pending(current))
617 			goto out;
618 	}
619 
620 	/* Connection was closed by RST, timeout, ICMP error
621 	 * or another process disconnected us.
622 	 */
623 	if (sk->sk_state == TCP_CLOSE)
624 		goto sock_error;
625 
626 	/* sk->sk_err may be not zero now, if RECVERR was ordered by user
627 	 * and error was received after socket entered established state.
628 	 * Hence, it is handled normally after connect() return successfully.
629 	 */
630 
631 	sock->state = SS_CONNECTED;
632 	err = 0;
633 out:
634 	return err;
635 
636 sock_error:
637 	err = sock_error(sk) ? : -ECONNABORTED;
638 	sock->state = SS_UNCONNECTED;
639 	if (sk->sk_prot->disconnect(sk, flags))
640 		sock->state = SS_DISCONNECTING;
641 	goto out;
642 }
643 EXPORT_SYMBOL(__inet_stream_connect);
644 
645 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
646 			int addr_len, int flags)
647 {
648 	int err;
649 
650 	lock_sock(sock->sk);
651 	err = __inet_stream_connect(sock, uaddr, addr_len, flags);
652 	release_sock(sock->sk);
653 	return err;
654 }
655 EXPORT_SYMBOL(inet_stream_connect);
656 
657 /*
658  *	Accept a pending connection. The TCP layer now gives BSD semantics.
659  */
660 
661 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
662 {
663 	struct sock *sk1 = sock->sk;
664 	int err = -EINVAL;
665 	struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
666 
667 	if (!sk2)
668 		goto do_err;
669 
670 	lock_sock(sk2);
671 
672 	sock_rps_record_flow(sk2);
673 	WARN_ON(!((1 << sk2->sk_state) &
674 		  (TCPF_ESTABLISHED | TCPF_SYN_RECV |
675 		  TCPF_CLOSE_WAIT | TCPF_CLOSE)));
676 
677 	sock_graft(sk2, newsock);
678 
679 	newsock->state = SS_CONNECTED;
680 	err = 0;
681 	release_sock(sk2);
682 do_err:
683 	return err;
684 }
685 EXPORT_SYMBOL(inet_accept);
686 
687 
688 /*
689  *	This does both peername and sockname.
690  */
691 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
692 			int *uaddr_len, int peer)
693 {
694 	struct sock *sk		= sock->sk;
695 	struct inet_sock *inet	= inet_sk(sk);
696 	DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
697 
698 	sin->sin_family = AF_INET;
699 	if (peer) {
700 		if (!inet->inet_dport ||
701 		    (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
702 		     peer == 1))
703 			return -ENOTCONN;
704 		sin->sin_port = inet->inet_dport;
705 		sin->sin_addr.s_addr = inet->inet_daddr;
706 	} else {
707 		__be32 addr = inet->inet_rcv_saddr;
708 		if (!addr)
709 			addr = inet->inet_saddr;
710 		sin->sin_port = inet->inet_sport;
711 		sin->sin_addr.s_addr = addr;
712 	}
713 	memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
714 	*uaddr_len = sizeof(*sin);
715 	return 0;
716 }
717 EXPORT_SYMBOL(inet_getname);
718 
719 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
720 		 size_t size)
721 {
722 	struct sock *sk = sock->sk;
723 
724 	sock_rps_record_flow(sk);
725 
726 	/* We may need to bind the socket. */
727 	if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
728 	    inet_autobind(sk))
729 		return -EAGAIN;
730 
731 	return sk->sk_prot->sendmsg(iocb, sk, msg, size);
732 }
733 EXPORT_SYMBOL(inet_sendmsg);
734 
735 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
736 		      size_t size, int flags)
737 {
738 	struct sock *sk = sock->sk;
739 
740 	sock_rps_record_flow(sk);
741 
742 	/* We may need to bind the socket. */
743 	if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
744 	    inet_autobind(sk))
745 		return -EAGAIN;
746 
747 	if (sk->sk_prot->sendpage)
748 		return sk->sk_prot->sendpage(sk, page, offset, size, flags);
749 	return sock_no_sendpage(sock, page, offset, size, flags);
750 }
751 EXPORT_SYMBOL(inet_sendpage);
752 
753 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
754 		 size_t size, int flags)
755 {
756 	struct sock *sk = sock->sk;
757 	int addr_len = 0;
758 	int err;
759 
760 	sock_rps_record_flow(sk);
761 
762 	err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
763 				   flags & ~MSG_DONTWAIT, &addr_len);
764 	if (err >= 0)
765 		msg->msg_namelen = addr_len;
766 	return err;
767 }
768 EXPORT_SYMBOL(inet_recvmsg);
769 
770 int inet_shutdown(struct socket *sock, int how)
771 {
772 	struct sock *sk = sock->sk;
773 	int err = 0;
774 
775 	/* This should really check to make sure
776 	 * the socket is a TCP socket. (WHY AC...)
777 	 */
778 	how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
779 		       1->2 bit 2 snds.
780 		       2->3 */
781 	if ((how & ~SHUTDOWN_MASK) || !how)	/* MAXINT->0 */
782 		return -EINVAL;
783 
784 	lock_sock(sk);
785 	if (sock->state == SS_CONNECTING) {
786 		if ((1 << sk->sk_state) &
787 		    (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
788 			sock->state = SS_DISCONNECTING;
789 		else
790 			sock->state = SS_CONNECTED;
791 	}
792 
793 	switch (sk->sk_state) {
794 	case TCP_CLOSE:
795 		err = -ENOTCONN;
796 		/* Hack to wake up other listeners, who can poll for
797 		   POLLHUP, even on eg. unconnected UDP sockets -- RR */
798 	default:
799 		sk->sk_shutdown |= how;
800 		if (sk->sk_prot->shutdown)
801 			sk->sk_prot->shutdown(sk, how);
802 		break;
803 
804 	/* Remaining two branches are temporary solution for missing
805 	 * close() in multithreaded environment. It is _not_ a good idea,
806 	 * but we have no choice until close() is repaired at VFS level.
807 	 */
808 	case TCP_LISTEN:
809 		if (!(how & RCV_SHUTDOWN))
810 			break;
811 		/* Fall through */
812 	case TCP_SYN_SENT:
813 		err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
814 		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
815 		break;
816 	}
817 
818 	/* Wake up anyone sleeping in poll. */
819 	sk->sk_state_change(sk);
820 	release_sock(sk);
821 	return err;
822 }
823 EXPORT_SYMBOL(inet_shutdown);
824 
825 /*
826  *	ioctl() calls you can issue on an INET socket. Most of these are
827  *	device configuration and stuff and very rarely used. Some ioctls
828  *	pass on to the socket itself.
829  *
830  *	NOTE: I like the idea of a module for the config stuff. ie ifconfig
831  *	loads the devconfigure module does its configuring and unloads it.
832  *	There's a good 20K of config code hanging around the kernel.
833  */
834 
835 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
836 {
837 	struct sock *sk = sock->sk;
838 	int err = 0;
839 	struct net *net = sock_net(sk);
840 
841 	switch (cmd) {
842 	case SIOCGSTAMP:
843 		err = sock_get_timestamp(sk, (struct timeval __user *)arg);
844 		break;
845 	case SIOCGSTAMPNS:
846 		err = sock_get_timestampns(sk, (struct timespec __user *)arg);
847 		break;
848 	case SIOCADDRT:
849 	case SIOCDELRT:
850 	case SIOCRTMSG:
851 		err = ip_rt_ioctl(net, cmd, (void __user *)arg);
852 		break;
853 	case SIOCDARP:
854 	case SIOCGARP:
855 	case SIOCSARP:
856 		err = arp_ioctl(net, cmd, (void __user *)arg);
857 		break;
858 	case SIOCGIFADDR:
859 	case SIOCSIFADDR:
860 	case SIOCGIFBRDADDR:
861 	case SIOCSIFBRDADDR:
862 	case SIOCGIFNETMASK:
863 	case SIOCSIFNETMASK:
864 	case SIOCGIFDSTADDR:
865 	case SIOCSIFDSTADDR:
866 	case SIOCSIFPFLAGS:
867 	case SIOCGIFPFLAGS:
868 	case SIOCSIFFLAGS:
869 		err = devinet_ioctl(net, cmd, (void __user *)arg);
870 		break;
871 	default:
872 		if (sk->sk_prot->ioctl)
873 			err = sk->sk_prot->ioctl(sk, cmd, arg);
874 		else
875 			err = -ENOIOCTLCMD;
876 		break;
877 	}
878 	return err;
879 }
880 EXPORT_SYMBOL(inet_ioctl);
881 
882 #ifdef CONFIG_COMPAT
883 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
884 {
885 	struct sock *sk = sock->sk;
886 	int err = -ENOIOCTLCMD;
887 
888 	if (sk->sk_prot->compat_ioctl)
889 		err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
890 
891 	return err;
892 }
893 #endif
894 
895 const struct proto_ops inet_stream_ops = {
896 	.family		   = PF_INET,
897 	.owner		   = THIS_MODULE,
898 	.release	   = inet_release,
899 	.bind		   = inet_bind,
900 	.connect	   = inet_stream_connect,
901 	.socketpair	   = sock_no_socketpair,
902 	.accept		   = inet_accept,
903 	.getname	   = inet_getname,
904 	.poll		   = tcp_poll,
905 	.ioctl		   = inet_ioctl,
906 	.listen		   = inet_listen,
907 	.shutdown	   = inet_shutdown,
908 	.setsockopt	   = sock_common_setsockopt,
909 	.getsockopt	   = sock_common_getsockopt,
910 	.sendmsg	   = inet_sendmsg,
911 	.recvmsg	   = inet_recvmsg,
912 	.mmap		   = sock_no_mmap,
913 	.sendpage	   = inet_sendpage,
914 	.splice_read	   = tcp_splice_read,
915 #ifdef CONFIG_COMPAT
916 	.compat_setsockopt = compat_sock_common_setsockopt,
917 	.compat_getsockopt = compat_sock_common_getsockopt,
918 	.compat_ioctl	   = inet_compat_ioctl,
919 #endif
920 };
921 EXPORT_SYMBOL(inet_stream_ops);
922 
923 const struct proto_ops inet_dgram_ops = {
924 	.family		   = PF_INET,
925 	.owner		   = THIS_MODULE,
926 	.release	   = inet_release,
927 	.bind		   = inet_bind,
928 	.connect	   = inet_dgram_connect,
929 	.socketpair	   = sock_no_socketpair,
930 	.accept		   = sock_no_accept,
931 	.getname	   = inet_getname,
932 	.poll		   = udp_poll,
933 	.ioctl		   = inet_ioctl,
934 	.listen		   = sock_no_listen,
935 	.shutdown	   = inet_shutdown,
936 	.setsockopt	   = sock_common_setsockopt,
937 	.getsockopt	   = sock_common_getsockopt,
938 	.sendmsg	   = inet_sendmsg,
939 	.recvmsg	   = inet_recvmsg,
940 	.mmap		   = sock_no_mmap,
941 	.sendpage	   = inet_sendpage,
942 #ifdef CONFIG_COMPAT
943 	.compat_setsockopt = compat_sock_common_setsockopt,
944 	.compat_getsockopt = compat_sock_common_getsockopt,
945 	.compat_ioctl	   = inet_compat_ioctl,
946 #endif
947 };
948 EXPORT_SYMBOL(inet_dgram_ops);
949 
950 /*
951  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
952  * udp_poll
953  */
954 static const struct proto_ops inet_sockraw_ops = {
955 	.family		   = PF_INET,
956 	.owner		   = THIS_MODULE,
957 	.release	   = inet_release,
958 	.bind		   = inet_bind,
959 	.connect	   = inet_dgram_connect,
960 	.socketpair	   = sock_no_socketpair,
961 	.accept		   = sock_no_accept,
962 	.getname	   = inet_getname,
963 	.poll		   = datagram_poll,
964 	.ioctl		   = inet_ioctl,
965 	.listen		   = sock_no_listen,
966 	.shutdown	   = inet_shutdown,
967 	.setsockopt	   = sock_common_setsockopt,
968 	.getsockopt	   = sock_common_getsockopt,
969 	.sendmsg	   = inet_sendmsg,
970 	.recvmsg	   = inet_recvmsg,
971 	.mmap		   = sock_no_mmap,
972 	.sendpage	   = inet_sendpage,
973 #ifdef CONFIG_COMPAT
974 	.compat_setsockopt = compat_sock_common_setsockopt,
975 	.compat_getsockopt = compat_sock_common_getsockopt,
976 	.compat_ioctl	   = inet_compat_ioctl,
977 #endif
978 };
979 
980 static const struct net_proto_family inet_family_ops = {
981 	.family = PF_INET,
982 	.create = inet_create,
983 	.owner	= THIS_MODULE,
984 };
985 
986 /* Upon startup we insert all the elements in inetsw_array[] into
987  * the linked list inetsw.
988  */
989 static struct inet_protosw inetsw_array[] =
990 {
991 	{
992 		.type =       SOCK_STREAM,
993 		.protocol =   IPPROTO_TCP,
994 		.prot =       &tcp_prot,
995 		.ops =        &inet_stream_ops,
996 		.flags =      INET_PROTOSW_PERMANENT |
997 			      INET_PROTOSW_ICSK,
998 	},
999 
1000 	{
1001 		.type =       SOCK_DGRAM,
1002 		.protocol =   IPPROTO_UDP,
1003 		.prot =       &udp_prot,
1004 		.ops =        &inet_dgram_ops,
1005 		.flags =      INET_PROTOSW_PERMANENT,
1006        },
1007 
1008        {
1009 		.type =       SOCK_DGRAM,
1010 		.protocol =   IPPROTO_ICMP,
1011 		.prot =       &ping_prot,
1012 		.ops =        &inet_dgram_ops,
1013 		.flags =      INET_PROTOSW_REUSE,
1014        },
1015 
1016        {
1017 	       .type =       SOCK_RAW,
1018 	       .protocol =   IPPROTO_IP,	/* wild card */
1019 	       .prot =       &raw_prot,
1020 	       .ops =        &inet_sockraw_ops,
1021 	       .flags =      INET_PROTOSW_REUSE,
1022        }
1023 };
1024 
1025 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1026 
1027 void inet_register_protosw(struct inet_protosw *p)
1028 {
1029 	struct list_head *lh;
1030 	struct inet_protosw *answer;
1031 	int protocol = p->protocol;
1032 	struct list_head *last_perm;
1033 
1034 	spin_lock_bh(&inetsw_lock);
1035 
1036 	if (p->type >= SOCK_MAX)
1037 		goto out_illegal;
1038 
1039 	/* If we are trying to override a permanent protocol, bail. */
1040 	answer = NULL;
1041 	last_perm = &inetsw[p->type];
1042 	list_for_each(lh, &inetsw[p->type]) {
1043 		answer = list_entry(lh, struct inet_protosw, list);
1044 
1045 		/* Check only the non-wild match. */
1046 		if (INET_PROTOSW_PERMANENT & answer->flags) {
1047 			if (protocol == answer->protocol)
1048 				break;
1049 			last_perm = lh;
1050 		}
1051 
1052 		answer = NULL;
1053 	}
1054 	if (answer)
1055 		goto out_permanent;
1056 
1057 	/* Add the new entry after the last permanent entry if any, so that
1058 	 * the new entry does not override a permanent entry when matched with
1059 	 * a wild-card protocol. But it is allowed to override any existing
1060 	 * non-permanent entry.  This means that when we remove this entry, the
1061 	 * system automatically returns to the old behavior.
1062 	 */
1063 	list_add_rcu(&p->list, last_perm);
1064 out:
1065 	spin_unlock_bh(&inetsw_lock);
1066 
1067 	return;
1068 
1069 out_permanent:
1070 	pr_err("Attempt to override permanent protocol %d\n", protocol);
1071 	goto out;
1072 
1073 out_illegal:
1074 	pr_err("Ignoring attempt to register invalid socket type %d\n",
1075 	       p->type);
1076 	goto out;
1077 }
1078 EXPORT_SYMBOL(inet_register_protosw);
1079 
1080 void inet_unregister_protosw(struct inet_protosw *p)
1081 {
1082 	if (INET_PROTOSW_PERMANENT & p->flags) {
1083 		pr_err("Attempt to unregister permanent protocol %d\n",
1084 		       p->protocol);
1085 	} else {
1086 		spin_lock_bh(&inetsw_lock);
1087 		list_del_rcu(&p->list);
1088 		spin_unlock_bh(&inetsw_lock);
1089 
1090 		synchronize_net();
1091 	}
1092 }
1093 EXPORT_SYMBOL(inet_unregister_protosw);
1094 
1095 /*
1096  *      Shall we try to damage output packets if routing dev changes?
1097  */
1098 
1099 int sysctl_ip_dynaddr __read_mostly;
1100 
1101 static int inet_sk_reselect_saddr(struct sock *sk)
1102 {
1103 	struct inet_sock *inet = inet_sk(sk);
1104 	__be32 old_saddr = inet->inet_saddr;
1105 	__be32 daddr = inet->inet_daddr;
1106 	struct flowi4 *fl4;
1107 	struct rtable *rt;
1108 	__be32 new_saddr;
1109 	struct ip_options_rcu *inet_opt;
1110 
1111 	inet_opt = rcu_dereference_protected(inet->inet_opt,
1112 					     sock_owned_by_user(sk));
1113 	if (inet_opt && inet_opt->opt.srr)
1114 		daddr = inet_opt->opt.faddr;
1115 
1116 	/* Query new route. */
1117 	fl4 = &inet->cork.fl.u.ip4;
1118 	rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1119 			      sk->sk_bound_dev_if, sk->sk_protocol,
1120 			      inet->inet_sport, inet->inet_dport, sk);
1121 	if (IS_ERR(rt))
1122 		return PTR_ERR(rt);
1123 
1124 	sk_setup_caps(sk, &rt->dst);
1125 
1126 	new_saddr = fl4->saddr;
1127 
1128 	if (new_saddr == old_saddr)
1129 		return 0;
1130 
1131 	if (sysctl_ip_dynaddr > 1) {
1132 		pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1133 			__func__, &old_saddr, &new_saddr);
1134 	}
1135 
1136 	inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1137 
1138 	/*
1139 	 * XXX The only one ugly spot where we need to
1140 	 * XXX really change the sockets identity after
1141 	 * XXX it has entered the hashes. -DaveM
1142 	 *
1143 	 * Besides that, it does not check for connection
1144 	 * uniqueness. Wait for troubles.
1145 	 */
1146 	__sk_prot_rehash(sk);
1147 	return 0;
1148 }
1149 
1150 int inet_sk_rebuild_header(struct sock *sk)
1151 {
1152 	struct inet_sock *inet = inet_sk(sk);
1153 	struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1154 	__be32 daddr;
1155 	struct ip_options_rcu *inet_opt;
1156 	struct flowi4 *fl4;
1157 	int err;
1158 
1159 	/* Route is OK, nothing to do. */
1160 	if (rt)
1161 		return 0;
1162 
1163 	/* Reroute. */
1164 	rcu_read_lock();
1165 	inet_opt = rcu_dereference(inet->inet_opt);
1166 	daddr = inet->inet_daddr;
1167 	if (inet_opt && inet_opt->opt.srr)
1168 		daddr = inet_opt->opt.faddr;
1169 	rcu_read_unlock();
1170 	fl4 = &inet->cork.fl.u.ip4;
1171 	rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1172 				   inet->inet_dport, inet->inet_sport,
1173 				   sk->sk_protocol, RT_CONN_FLAGS(sk),
1174 				   sk->sk_bound_dev_if);
1175 	if (!IS_ERR(rt)) {
1176 		err = 0;
1177 		sk_setup_caps(sk, &rt->dst);
1178 	} else {
1179 		err = PTR_ERR(rt);
1180 
1181 		/* Routing failed... */
1182 		sk->sk_route_caps = 0;
1183 		/*
1184 		 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1185 		 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1186 		 */
1187 		if (!sysctl_ip_dynaddr ||
1188 		    sk->sk_state != TCP_SYN_SENT ||
1189 		    (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1190 		    (err = inet_sk_reselect_saddr(sk)) != 0)
1191 			sk->sk_err_soft = -err;
1192 	}
1193 
1194 	return err;
1195 }
1196 EXPORT_SYMBOL(inet_sk_rebuild_header);
1197 
1198 static struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1199 					netdev_features_t features)
1200 {
1201 	struct sk_buff *segs = ERR_PTR(-EINVAL);
1202 	const struct net_offload *ops;
1203 	unsigned int offset = 0;
1204 	bool udpfrag, encap;
1205 	struct iphdr *iph;
1206 	int proto;
1207 	int nhoff;
1208 	int ihl;
1209 	int id;
1210 
1211 	if (unlikely(skb_shinfo(skb)->gso_type &
1212 		     ~(SKB_GSO_TCPV4 |
1213 		       SKB_GSO_UDP |
1214 		       SKB_GSO_DODGY |
1215 		       SKB_GSO_TCP_ECN |
1216 		       SKB_GSO_GRE |
1217 		       SKB_GSO_GRE_CSUM |
1218 		       SKB_GSO_IPIP |
1219 		       SKB_GSO_SIT |
1220 		       SKB_GSO_TCPV6 |
1221 		       SKB_GSO_UDP_TUNNEL |
1222 		       SKB_GSO_UDP_TUNNEL_CSUM |
1223 		       SKB_GSO_TUNNEL_REMCSUM |
1224 		       0)))
1225 		goto out;
1226 
1227 	skb_reset_network_header(skb);
1228 	nhoff = skb_network_header(skb) - skb_mac_header(skb);
1229 	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1230 		goto out;
1231 
1232 	iph = ip_hdr(skb);
1233 	ihl = iph->ihl * 4;
1234 	if (ihl < sizeof(*iph))
1235 		goto out;
1236 
1237 	id = ntohs(iph->id);
1238 	proto = iph->protocol;
1239 
1240 	/* Warning: after this point, iph might be no longer valid */
1241 	if (unlikely(!pskb_may_pull(skb, ihl)))
1242 		goto out;
1243 	__skb_pull(skb, ihl);
1244 
1245 	encap = SKB_GSO_CB(skb)->encap_level > 0;
1246 	if (encap)
1247 		features &= skb->dev->hw_enc_features;
1248 	SKB_GSO_CB(skb)->encap_level += ihl;
1249 
1250 	skb_reset_transport_header(skb);
1251 
1252 	segs = ERR_PTR(-EPROTONOSUPPORT);
1253 
1254 	if (skb->encapsulation &&
1255 	    skb_shinfo(skb)->gso_type & (SKB_GSO_SIT|SKB_GSO_IPIP))
1256 		udpfrag = proto == IPPROTO_UDP && encap;
1257 	else
1258 		udpfrag = proto == IPPROTO_UDP && !skb->encapsulation;
1259 
1260 	ops = rcu_dereference(inet_offloads[proto]);
1261 	if (likely(ops && ops->callbacks.gso_segment))
1262 		segs = ops->callbacks.gso_segment(skb, features);
1263 
1264 	if (IS_ERR_OR_NULL(segs))
1265 		goto out;
1266 
1267 	skb = segs;
1268 	do {
1269 		iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1270 		if (udpfrag) {
1271 			iph->id = htons(id);
1272 			iph->frag_off = htons(offset >> 3);
1273 			if (skb->next != NULL)
1274 				iph->frag_off |= htons(IP_MF);
1275 			offset += skb->len - nhoff - ihl;
1276 		} else {
1277 			iph->id = htons(id++);
1278 		}
1279 		iph->tot_len = htons(skb->len - nhoff);
1280 		ip_send_check(iph);
1281 		if (encap)
1282 			skb_reset_inner_headers(skb);
1283 		skb->network_header = (u8 *)iph - skb->head;
1284 	} while ((skb = skb->next));
1285 
1286 out:
1287 	return segs;
1288 }
1289 
1290 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1291 					 struct sk_buff *skb)
1292 {
1293 	const struct net_offload *ops;
1294 	struct sk_buff **pp = NULL;
1295 	struct sk_buff *p;
1296 	const struct iphdr *iph;
1297 	unsigned int hlen;
1298 	unsigned int off;
1299 	unsigned int id;
1300 	int flush = 1;
1301 	int proto;
1302 
1303 	off = skb_gro_offset(skb);
1304 	hlen = off + sizeof(*iph);
1305 	iph = skb_gro_header_fast(skb, off);
1306 	if (skb_gro_header_hard(skb, hlen)) {
1307 		iph = skb_gro_header_slow(skb, hlen, off);
1308 		if (unlikely(!iph))
1309 			goto out;
1310 	}
1311 
1312 	proto = iph->protocol;
1313 
1314 	rcu_read_lock();
1315 	ops = rcu_dereference(inet_offloads[proto]);
1316 	if (!ops || !ops->callbacks.gro_receive)
1317 		goto out_unlock;
1318 
1319 	if (*(u8 *)iph != 0x45)
1320 		goto out_unlock;
1321 
1322 	if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1323 		goto out_unlock;
1324 
1325 	id = ntohl(*(__be32 *)&iph->id);
1326 	flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1327 	id >>= 16;
1328 
1329 	for (p = *head; p; p = p->next) {
1330 		struct iphdr *iph2;
1331 
1332 		if (!NAPI_GRO_CB(p)->same_flow)
1333 			continue;
1334 
1335 		iph2 = (struct iphdr *)(p->data + off);
1336 		/* The above works because, with the exception of the top
1337 		 * (inner most) layer, we only aggregate pkts with the same
1338 		 * hdr length so all the hdrs we'll need to verify will start
1339 		 * at the same offset.
1340 		 */
1341 		if ((iph->protocol ^ iph2->protocol) |
1342 		    ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1343 		    ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1344 			NAPI_GRO_CB(p)->same_flow = 0;
1345 			continue;
1346 		}
1347 
1348 		/* All fields must match except length and checksum. */
1349 		NAPI_GRO_CB(p)->flush |=
1350 			(iph->ttl ^ iph2->ttl) |
1351 			(iph->tos ^ iph2->tos) |
1352 			((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1353 
1354 		/* Save the IP ID check to be included later when we get to
1355 		 * the transport layer so only the inner most IP ID is checked.
1356 		 * This is because some GSO/TSO implementations do not
1357 		 * correctly increment the IP ID for the outer hdrs.
1358 		 */
1359 		NAPI_GRO_CB(p)->flush_id =
1360 			    ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1361 		NAPI_GRO_CB(p)->flush |= flush;
1362 	}
1363 
1364 	NAPI_GRO_CB(skb)->flush |= flush;
1365 	skb_set_network_header(skb, off);
1366 	/* The above will be needed by the transport layer if there is one
1367 	 * immediately following this IP hdr.
1368 	 */
1369 
1370 	/* Note : No need to call skb_gro_postpull_rcsum() here,
1371 	 * as we already checked checksum over ipv4 header was 0
1372 	 */
1373 	skb_gro_pull(skb, sizeof(*iph));
1374 	skb_set_transport_header(skb, skb_gro_offset(skb));
1375 
1376 	pp = ops->callbacks.gro_receive(head, skb);
1377 
1378 out_unlock:
1379 	rcu_read_unlock();
1380 
1381 out:
1382 	NAPI_GRO_CB(skb)->flush |= flush;
1383 
1384 	return pp;
1385 }
1386 
1387 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1388 {
1389 	if (sk->sk_family == AF_INET)
1390 		return ip_recv_error(sk, msg, len, addr_len);
1391 #if IS_ENABLED(CONFIG_IPV6)
1392 	if (sk->sk_family == AF_INET6)
1393 		return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1394 #endif
1395 	return -EINVAL;
1396 }
1397 
1398 static int inet_gro_complete(struct sk_buff *skb, int nhoff)
1399 {
1400 	__be16 newlen = htons(skb->len - nhoff);
1401 	struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1402 	const struct net_offload *ops;
1403 	int proto = iph->protocol;
1404 	int err = -ENOSYS;
1405 
1406 	if (skb->encapsulation)
1407 		skb_set_inner_network_header(skb, nhoff);
1408 
1409 	csum_replace2(&iph->check, iph->tot_len, newlen);
1410 	iph->tot_len = newlen;
1411 
1412 	rcu_read_lock();
1413 	ops = rcu_dereference(inet_offloads[proto]);
1414 	if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1415 		goto out_unlock;
1416 
1417 	/* Only need to add sizeof(*iph) to get to the next hdr below
1418 	 * because any hdr with option will have been flushed in
1419 	 * inet_gro_receive().
1420 	 */
1421 	err = ops->callbacks.gro_complete(skb, nhoff + sizeof(*iph));
1422 
1423 out_unlock:
1424 	rcu_read_unlock();
1425 
1426 	return err;
1427 }
1428 
1429 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1430 			 unsigned short type, unsigned char protocol,
1431 			 struct net *net)
1432 {
1433 	struct socket *sock;
1434 	int rc = sock_create_kern(family, type, protocol, &sock);
1435 
1436 	if (rc == 0) {
1437 		*sk = sock->sk;
1438 		(*sk)->sk_allocation = GFP_ATOMIC;
1439 		/*
1440 		 * Unhash it so that IP input processing does not even see it,
1441 		 * we do not wish this socket to see incoming packets.
1442 		 */
1443 		(*sk)->sk_prot->unhash(*sk);
1444 
1445 		sk_change_net(*sk, net);
1446 	}
1447 	return rc;
1448 }
1449 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1450 
1451 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1452 {
1453 	unsigned long res = 0;
1454 	int i;
1455 
1456 	for_each_possible_cpu(i)
1457 		res += *(((unsigned long *) per_cpu_ptr(mib, i)) + offt);
1458 	return res;
1459 }
1460 EXPORT_SYMBOL_GPL(snmp_fold_field);
1461 
1462 #if BITS_PER_LONG==32
1463 
1464 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1465 {
1466 	u64 res = 0;
1467 	int cpu;
1468 
1469 	for_each_possible_cpu(cpu) {
1470 		void *bhptr;
1471 		struct u64_stats_sync *syncp;
1472 		u64 v;
1473 		unsigned int start;
1474 
1475 		bhptr = per_cpu_ptr(mib, cpu);
1476 		syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1477 		do {
1478 			start = u64_stats_fetch_begin_irq(syncp);
1479 			v = *(((u64 *) bhptr) + offt);
1480 		} while (u64_stats_fetch_retry_irq(syncp, start));
1481 
1482 		res += v;
1483 	}
1484 	return res;
1485 }
1486 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1487 #endif
1488 
1489 #ifdef CONFIG_IP_MULTICAST
1490 static const struct net_protocol igmp_protocol = {
1491 	.handler =	igmp_rcv,
1492 	.netns_ok =	1,
1493 };
1494 #endif
1495 
1496 static const struct net_protocol tcp_protocol = {
1497 	.early_demux	=	tcp_v4_early_demux,
1498 	.handler	=	tcp_v4_rcv,
1499 	.err_handler	=	tcp_v4_err,
1500 	.no_policy	=	1,
1501 	.netns_ok	=	1,
1502 	.icmp_strict_tag_validation = 1,
1503 };
1504 
1505 static const struct net_protocol udp_protocol = {
1506 	.early_demux =	udp_v4_early_demux,
1507 	.handler =	udp_rcv,
1508 	.err_handler =	udp_err,
1509 	.no_policy =	1,
1510 	.netns_ok =	1,
1511 };
1512 
1513 static const struct net_protocol icmp_protocol = {
1514 	.handler =	icmp_rcv,
1515 	.err_handler =	icmp_err,
1516 	.no_policy =	1,
1517 	.netns_ok =	1,
1518 };
1519 
1520 static __net_init int ipv4_mib_init_net(struct net *net)
1521 {
1522 	int i;
1523 
1524 	net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1525 	if (!net->mib.tcp_statistics)
1526 		goto err_tcp_mib;
1527 	net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1528 	if (!net->mib.ip_statistics)
1529 		goto err_ip_mib;
1530 
1531 	for_each_possible_cpu(i) {
1532 		struct ipstats_mib *af_inet_stats;
1533 		af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1534 		u64_stats_init(&af_inet_stats->syncp);
1535 	}
1536 
1537 	net->mib.net_statistics = alloc_percpu(struct linux_mib);
1538 	if (!net->mib.net_statistics)
1539 		goto err_net_mib;
1540 	net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1541 	if (!net->mib.udp_statistics)
1542 		goto err_udp_mib;
1543 	net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1544 	if (!net->mib.udplite_statistics)
1545 		goto err_udplite_mib;
1546 	net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1547 	if (!net->mib.icmp_statistics)
1548 		goto err_icmp_mib;
1549 	net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1550 					      GFP_KERNEL);
1551 	if (!net->mib.icmpmsg_statistics)
1552 		goto err_icmpmsg_mib;
1553 
1554 	tcp_mib_init(net);
1555 	return 0;
1556 
1557 err_icmpmsg_mib:
1558 	free_percpu(net->mib.icmp_statistics);
1559 err_icmp_mib:
1560 	free_percpu(net->mib.udplite_statistics);
1561 err_udplite_mib:
1562 	free_percpu(net->mib.udp_statistics);
1563 err_udp_mib:
1564 	free_percpu(net->mib.net_statistics);
1565 err_net_mib:
1566 	free_percpu(net->mib.ip_statistics);
1567 err_ip_mib:
1568 	free_percpu(net->mib.tcp_statistics);
1569 err_tcp_mib:
1570 	return -ENOMEM;
1571 }
1572 
1573 static __net_exit void ipv4_mib_exit_net(struct net *net)
1574 {
1575 	kfree(net->mib.icmpmsg_statistics);
1576 	free_percpu(net->mib.icmp_statistics);
1577 	free_percpu(net->mib.udplite_statistics);
1578 	free_percpu(net->mib.udp_statistics);
1579 	free_percpu(net->mib.net_statistics);
1580 	free_percpu(net->mib.ip_statistics);
1581 	free_percpu(net->mib.tcp_statistics);
1582 }
1583 
1584 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1585 	.init = ipv4_mib_init_net,
1586 	.exit = ipv4_mib_exit_net,
1587 };
1588 
1589 static int __init init_ipv4_mibs(void)
1590 {
1591 	return register_pernet_subsys(&ipv4_mib_ops);
1592 }
1593 
1594 static __net_init int inet_init_net(struct net *net)
1595 {
1596 	/*
1597 	 * Set defaults for local port range
1598 	 */
1599 	seqlock_init(&net->ipv4.ip_local_ports.lock);
1600 	net->ipv4.ip_local_ports.range[0] =  32768;
1601 	net->ipv4.ip_local_ports.range[1] =  61000;
1602 
1603 	seqlock_init(&net->ipv4.ping_group_range.lock);
1604 	/*
1605 	 * Sane defaults - nobody may create ping sockets.
1606 	 * Boot scripts should set this to distro-specific group.
1607 	 */
1608 	net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1609 	net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1610 	return 0;
1611 }
1612 
1613 static __net_exit void inet_exit_net(struct net *net)
1614 {
1615 }
1616 
1617 static __net_initdata struct pernet_operations af_inet_ops = {
1618 	.init = inet_init_net,
1619 	.exit = inet_exit_net,
1620 };
1621 
1622 static int __init init_inet_pernet_ops(void)
1623 {
1624 	return register_pernet_subsys(&af_inet_ops);
1625 }
1626 
1627 static int ipv4_proc_init(void);
1628 
1629 /*
1630  *	IP protocol layer initialiser
1631  */
1632 
1633 static struct packet_offload ip_packet_offload __read_mostly = {
1634 	.type = cpu_to_be16(ETH_P_IP),
1635 	.callbacks = {
1636 		.gso_segment = inet_gso_segment,
1637 		.gro_receive = inet_gro_receive,
1638 		.gro_complete = inet_gro_complete,
1639 	},
1640 };
1641 
1642 static const struct net_offload ipip_offload = {
1643 	.callbacks = {
1644 		.gso_segment	= inet_gso_segment,
1645 		.gro_receive	= inet_gro_receive,
1646 		.gro_complete	= inet_gro_complete,
1647 	},
1648 };
1649 
1650 static int __init ipv4_offload_init(void)
1651 {
1652 	/*
1653 	 * Add offloads
1654 	 */
1655 	if (udpv4_offload_init() < 0)
1656 		pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1657 	if (tcpv4_offload_init() < 0)
1658 		pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1659 
1660 	dev_add_offload(&ip_packet_offload);
1661 	inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1662 	return 0;
1663 }
1664 
1665 fs_initcall(ipv4_offload_init);
1666 
1667 static struct packet_type ip_packet_type __read_mostly = {
1668 	.type = cpu_to_be16(ETH_P_IP),
1669 	.func = ip_rcv,
1670 };
1671 
1672 static int __init inet_init(void)
1673 {
1674 	struct inet_protosw *q;
1675 	struct list_head *r;
1676 	int rc = -EINVAL;
1677 
1678 	BUILD_BUG_ON(sizeof(struct inet_skb_parm) > FIELD_SIZEOF(struct sk_buff, cb));
1679 
1680 	rc = proto_register(&tcp_prot, 1);
1681 	if (rc)
1682 		goto out;
1683 
1684 	rc = proto_register(&udp_prot, 1);
1685 	if (rc)
1686 		goto out_unregister_tcp_proto;
1687 
1688 	rc = proto_register(&raw_prot, 1);
1689 	if (rc)
1690 		goto out_unregister_udp_proto;
1691 
1692 	rc = proto_register(&ping_prot, 1);
1693 	if (rc)
1694 		goto out_unregister_raw_proto;
1695 
1696 	/*
1697 	 *	Tell SOCKET that we are alive...
1698 	 */
1699 
1700 	(void)sock_register(&inet_family_ops);
1701 
1702 #ifdef CONFIG_SYSCTL
1703 	ip_static_sysctl_init();
1704 #endif
1705 
1706 	/*
1707 	 *	Add all the base protocols.
1708 	 */
1709 
1710 	if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1711 		pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1712 	if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1713 		pr_crit("%s: Cannot add UDP protocol\n", __func__);
1714 	if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1715 		pr_crit("%s: Cannot add TCP protocol\n", __func__);
1716 #ifdef CONFIG_IP_MULTICAST
1717 	if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1718 		pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1719 #endif
1720 
1721 	/* Register the socket-side information for inet_create. */
1722 	for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1723 		INIT_LIST_HEAD(r);
1724 
1725 	for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1726 		inet_register_protosw(q);
1727 
1728 	/*
1729 	 *	Set the ARP module up
1730 	 */
1731 
1732 	arp_init();
1733 
1734 	/*
1735 	 *	Set the IP module up
1736 	 */
1737 
1738 	ip_init();
1739 
1740 	tcp_v4_init();
1741 
1742 	/* Setup TCP slab cache for open requests. */
1743 	tcp_init();
1744 
1745 	/* Setup UDP memory threshold */
1746 	udp_init();
1747 
1748 	/* Add UDP-Lite (RFC 3828) */
1749 	udplite4_register();
1750 
1751 	ping_init();
1752 
1753 	/*
1754 	 *	Set the ICMP layer up
1755 	 */
1756 
1757 	if (icmp_init() < 0)
1758 		panic("Failed to create the ICMP control socket.\n");
1759 
1760 	/*
1761 	 *	Initialise the multicast router
1762 	 */
1763 #if defined(CONFIG_IP_MROUTE)
1764 	if (ip_mr_init())
1765 		pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1766 #endif
1767 
1768 	if (init_inet_pernet_ops())
1769 		pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
1770 	/*
1771 	 *	Initialise per-cpu ipv4 mibs
1772 	 */
1773 
1774 	if (init_ipv4_mibs())
1775 		pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1776 
1777 	ipv4_proc_init();
1778 
1779 	ipfrag_init();
1780 
1781 	dev_add_pack(&ip_packet_type);
1782 
1783 	rc = 0;
1784 out:
1785 	return rc;
1786 out_unregister_raw_proto:
1787 	proto_unregister(&raw_prot);
1788 out_unregister_udp_proto:
1789 	proto_unregister(&udp_prot);
1790 out_unregister_tcp_proto:
1791 	proto_unregister(&tcp_prot);
1792 	goto out;
1793 }
1794 
1795 fs_initcall(inet_init);
1796 
1797 /* ------------------------------------------------------------------------ */
1798 
1799 #ifdef CONFIG_PROC_FS
1800 static int __init ipv4_proc_init(void)
1801 {
1802 	int rc = 0;
1803 
1804 	if (raw_proc_init())
1805 		goto out_raw;
1806 	if (tcp4_proc_init())
1807 		goto out_tcp;
1808 	if (udp4_proc_init())
1809 		goto out_udp;
1810 	if (ping_proc_init())
1811 		goto out_ping;
1812 	if (ip_misc_proc_init())
1813 		goto out_misc;
1814 out:
1815 	return rc;
1816 out_misc:
1817 	ping_proc_exit();
1818 out_ping:
1819 	udp4_proc_exit();
1820 out_udp:
1821 	tcp4_proc_exit();
1822 out_tcp:
1823 	raw_proc_exit();
1824 out_raw:
1825 	rc = -ENOMEM;
1826 	goto out;
1827 }
1828 
1829 #else /* CONFIG_PROC_FS */
1830 static int __init ipv4_proc_init(void)
1831 {
1832 	return 0;
1833 }
1834 #endif /* CONFIG_PROC_FS */
1835 
1836 MODULE_ALIAS_NETPROTO(PF_INET);
1837 
1838