xref: /openbmc/linux/net/ipv4/af_inet.c (revision 9cfc5c90)
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/ip_tunnels.h>
116 #include <net/xfrm.h>
117 #include <net/net_namespace.h>
118 #include <net/secure_seq.h>
119 #ifdef CONFIG_IP_MROUTE
120 #include <linux/mroute.h>
121 #endif
122 #include <net/l3mdev.h>
123 
124 
125 /* The inetsw table contains everything that inet_create needs to
126  * build a new socket.
127  */
128 static struct list_head inetsw[SOCK_MAX];
129 static DEFINE_SPINLOCK(inetsw_lock);
130 
131 /* New destruction routine */
132 
133 void inet_sock_destruct(struct sock *sk)
134 {
135 	struct inet_sock *inet = inet_sk(sk);
136 
137 	__skb_queue_purge(&sk->sk_receive_queue);
138 	__skb_queue_purge(&sk->sk_error_queue);
139 
140 	sk_mem_reclaim(sk);
141 
142 	if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
143 		pr_err("Attempt to release TCP socket in state %d %p\n",
144 		       sk->sk_state, sk);
145 		return;
146 	}
147 	if (!sock_flag(sk, SOCK_DEAD)) {
148 		pr_err("Attempt to release alive inet socket %p\n", sk);
149 		return;
150 	}
151 
152 	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
153 	WARN_ON(atomic_read(&sk->sk_wmem_alloc));
154 	WARN_ON(sk->sk_wmem_queued);
155 	WARN_ON(sk->sk_forward_alloc);
156 
157 	kfree(rcu_dereference_protected(inet->inet_opt, 1));
158 	dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
159 	dst_release(sk->sk_rx_dst);
160 	sk_refcnt_debug_dec(sk);
161 }
162 EXPORT_SYMBOL(inet_sock_destruct);
163 
164 /*
165  *	The routines beyond this point handle the behaviour of an AF_INET
166  *	socket object. Mostly it punts to the subprotocols of IP to do
167  *	the work.
168  */
169 
170 /*
171  *	Automatically bind an unbound socket.
172  */
173 
174 static int inet_autobind(struct sock *sk)
175 {
176 	struct inet_sock *inet;
177 	/* We may need to bind the socket. */
178 	lock_sock(sk);
179 	inet = inet_sk(sk);
180 	if (!inet->inet_num) {
181 		if (sk->sk_prot->get_port(sk, 0)) {
182 			release_sock(sk);
183 			return -EAGAIN;
184 		}
185 		inet->inet_sport = htons(inet->inet_num);
186 	}
187 	release_sock(sk);
188 	return 0;
189 }
190 
191 /*
192  *	Move a socket into listening state.
193  */
194 int inet_listen(struct socket *sock, int backlog)
195 {
196 	struct sock *sk = sock->sk;
197 	unsigned char old_state;
198 	int err;
199 
200 	lock_sock(sk);
201 
202 	err = -EINVAL;
203 	if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
204 		goto out;
205 
206 	old_state = sk->sk_state;
207 	if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
208 		goto out;
209 
210 	/* Really, if the socket is already in listen state
211 	 * we can only allow the backlog to be adjusted.
212 	 */
213 	if (old_state != TCP_LISTEN) {
214 		/* Check special setups for testing purpose to enable TFO w/o
215 		 * requiring TCP_FASTOPEN sockopt.
216 		 * Note that only TCP sockets (SOCK_STREAM) will reach here.
217 		 * Also fastopenq may already been allocated because this
218 		 * socket was in TCP_LISTEN state previously but was
219 		 * shutdown() (rather than close()).
220 		 */
221 		if ((sysctl_tcp_fastopen & TFO_SERVER_ENABLE) != 0 &&
222 		    !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
223 			if ((sysctl_tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) != 0)
224 				fastopen_queue_tune(sk, backlog);
225 			else if ((sysctl_tcp_fastopen &
226 				  TFO_SERVER_WO_SOCKOPT2) != 0)
227 				fastopen_queue_tune(sk,
228 				    ((uint)sysctl_tcp_fastopen) >> 16);
229 
230 			tcp_fastopen_init_key_once(true);
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);
318 
319 	err = -ENOBUFS;
320 	sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
321 	if (!sk)
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 	u32 tb_id = RT_TABLE_LOCAL;
428 	int err;
429 
430 	/* If the socket has its own bind function then use it. (RAW) */
431 	if (sk->sk_prot->bind) {
432 		err = sk->sk_prot->bind(sk, uaddr, addr_len);
433 		goto out;
434 	}
435 	err = -EINVAL;
436 	if (addr_len < sizeof(struct sockaddr_in))
437 		goto out;
438 
439 	if (addr->sin_family != AF_INET) {
440 		/* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
441 		 * only if s_addr is INADDR_ANY.
442 		 */
443 		err = -EAFNOSUPPORT;
444 		if (addr->sin_family != AF_UNSPEC ||
445 		    addr->sin_addr.s_addr != htonl(INADDR_ANY))
446 			goto out;
447 	}
448 
449 	tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
450 	chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
451 
452 	/* Not specified by any standard per-se, however it breaks too
453 	 * many applications when removed.  It is unfortunate since
454 	 * allowing applications to make a non-local bind solves
455 	 * several problems with systems using dynamic addressing.
456 	 * (ie. your servers still start up even if your ISDN link
457 	 *  is temporarily down)
458 	 */
459 	err = -EADDRNOTAVAIL;
460 	if (!net->ipv4.sysctl_ip_nonlocal_bind &&
461 	    !(inet->freebind || inet->transparent) &&
462 	    addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
463 	    chk_addr_ret != RTN_LOCAL &&
464 	    chk_addr_ret != RTN_MULTICAST &&
465 	    chk_addr_ret != RTN_BROADCAST)
466 		goto out;
467 
468 	snum = ntohs(addr->sin_port);
469 	err = -EACCES;
470 	if (snum && snum < PROT_SOCK &&
471 	    !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
472 		goto out;
473 
474 	/*      We keep a pair of addresses. rcv_saddr is the one
475 	 *      used by hash lookups, and saddr is used for transmit.
476 	 *
477 	 *      In the BSD API these are the same except where it
478 	 *      would be illegal to use them (multicast/broadcast) in
479 	 *      which case the sending device address is used.
480 	 */
481 	lock_sock(sk);
482 
483 	/* Check these errors (active socket, double bind). */
484 	err = -EINVAL;
485 	if (sk->sk_state != TCP_CLOSE || inet->inet_num)
486 		goto out_release_sock;
487 
488 	inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
489 	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
490 		inet->inet_saddr = 0;  /* Use device */
491 
492 	/* Make sure we are allowed to bind here. */
493 	if ((snum || !inet->bind_address_no_port) &&
494 	    sk->sk_prot->get_port(sk, snum)) {
495 		inet->inet_saddr = inet->inet_rcv_saddr = 0;
496 		err = -EADDRINUSE;
497 		goto out_release_sock;
498 	}
499 
500 	if (inet->inet_rcv_saddr)
501 		sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
502 	if (snum)
503 		sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
504 	inet->inet_sport = htons(inet->inet_num);
505 	inet->inet_daddr = 0;
506 	inet->inet_dport = 0;
507 	sk_dst_reset(sk);
508 	err = 0;
509 out_release_sock:
510 	release_sock(sk);
511 out:
512 	return err;
513 }
514 EXPORT_SYMBOL(inet_bind);
515 
516 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
517 		       int addr_len, int flags)
518 {
519 	struct sock *sk = sock->sk;
520 
521 	if (addr_len < sizeof(uaddr->sa_family))
522 		return -EINVAL;
523 	if (uaddr->sa_family == AF_UNSPEC)
524 		return sk->sk_prot->disconnect(sk, flags);
525 
526 	if (!inet_sk(sk)->inet_num && inet_autobind(sk))
527 		return -EAGAIN;
528 	return sk->sk_prot->connect(sk, uaddr, addr_len);
529 }
530 EXPORT_SYMBOL(inet_dgram_connect);
531 
532 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
533 {
534 	DEFINE_WAIT(wait);
535 
536 	prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
537 	sk->sk_write_pending += writebias;
538 
539 	/* Basic assumption: if someone sets sk->sk_err, he _must_
540 	 * change state of the socket from TCP_SYN_*.
541 	 * Connect() does not allow to get error notifications
542 	 * without closing the socket.
543 	 */
544 	while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
545 		release_sock(sk);
546 		timeo = schedule_timeout(timeo);
547 		lock_sock(sk);
548 		if (signal_pending(current) || !timeo)
549 			break;
550 		prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
551 	}
552 	finish_wait(sk_sleep(sk), &wait);
553 	sk->sk_write_pending -= writebias;
554 	return timeo;
555 }
556 
557 /*
558  *	Connect to a remote host. There is regrettably still a little
559  *	TCP 'magic' in here.
560  */
561 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
562 			  int addr_len, int flags)
563 {
564 	struct sock *sk = sock->sk;
565 	int err;
566 	long timeo;
567 
568 	if (addr_len < sizeof(uaddr->sa_family))
569 		return -EINVAL;
570 
571 	if (uaddr->sa_family == AF_UNSPEC) {
572 		err = sk->sk_prot->disconnect(sk, flags);
573 		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
574 		goto out;
575 	}
576 
577 	switch (sock->state) {
578 	default:
579 		err = -EINVAL;
580 		goto out;
581 	case SS_CONNECTED:
582 		err = -EISCONN;
583 		goto out;
584 	case SS_CONNECTING:
585 		err = -EALREADY;
586 		/* Fall out of switch with err, set for this state */
587 		break;
588 	case SS_UNCONNECTED:
589 		err = -EISCONN;
590 		if (sk->sk_state != TCP_CLOSE)
591 			goto out;
592 
593 		err = sk->sk_prot->connect(sk, uaddr, addr_len);
594 		if (err < 0)
595 			goto out;
596 
597 		sock->state = SS_CONNECTING;
598 
599 		/* Just entered SS_CONNECTING state; the only
600 		 * difference is that return value in non-blocking
601 		 * case is EINPROGRESS, rather than EALREADY.
602 		 */
603 		err = -EINPROGRESS;
604 		break;
605 	}
606 
607 	timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
608 
609 	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
610 		int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
611 				tcp_sk(sk)->fastopen_req &&
612 				tcp_sk(sk)->fastopen_req->data ? 1 : 0;
613 
614 		/* Error code is set above */
615 		if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
616 			goto out;
617 
618 		err = sock_intr_errno(timeo);
619 		if (signal_pending(current))
620 			goto out;
621 	}
622 
623 	/* Connection was closed by RST, timeout, ICMP error
624 	 * or another process disconnected us.
625 	 */
626 	if (sk->sk_state == TCP_CLOSE)
627 		goto sock_error;
628 
629 	/* sk->sk_err may be not zero now, if RECVERR was ordered by user
630 	 * and error was received after socket entered established state.
631 	 * Hence, it is handled normally after connect() return successfully.
632 	 */
633 
634 	sock->state = SS_CONNECTED;
635 	err = 0;
636 out:
637 	return err;
638 
639 sock_error:
640 	err = sock_error(sk) ? : -ECONNABORTED;
641 	sock->state = SS_UNCONNECTED;
642 	if (sk->sk_prot->disconnect(sk, flags))
643 		sock->state = SS_DISCONNECTING;
644 	goto out;
645 }
646 EXPORT_SYMBOL(__inet_stream_connect);
647 
648 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
649 			int addr_len, int flags)
650 {
651 	int err;
652 
653 	lock_sock(sock->sk);
654 	err = __inet_stream_connect(sock, uaddr, addr_len, flags);
655 	release_sock(sock->sk);
656 	return err;
657 }
658 EXPORT_SYMBOL(inet_stream_connect);
659 
660 /*
661  *	Accept a pending connection. The TCP layer now gives BSD semantics.
662  */
663 
664 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
665 {
666 	struct sock *sk1 = sock->sk;
667 	int err = -EINVAL;
668 	struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
669 
670 	if (!sk2)
671 		goto do_err;
672 
673 	lock_sock(sk2);
674 
675 	sock_rps_record_flow(sk2);
676 	WARN_ON(!((1 << sk2->sk_state) &
677 		  (TCPF_ESTABLISHED | TCPF_SYN_RECV |
678 		  TCPF_CLOSE_WAIT | TCPF_CLOSE)));
679 
680 	sock_graft(sk2, newsock);
681 
682 	newsock->state = SS_CONNECTED;
683 	err = 0;
684 	release_sock(sk2);
685 do_err:
686 	return err;
687 }
688 EXPORT_SYMBOL(inet_accept);
689 
690 
691 /*
692  *	This does both peername and sockname.
693  */
694 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
695 			int *uaddr_len, int peer)
696 {
697 	struct sock *sk		= sock->sk;
698 	struct inet_sock *inet	= inet_sk(sk);
699 	DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
700 
701 	sin->sin_family = AF_INET;
702 	if (peer) {
703 		if (!inet->inet_dport ||
704 		    (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
705 		     peer == 1))
706 			return -ENOTCONN;
707 		sin->sin_port = inet->inet_dport;
708 		sin->sin_addr.s_addr = inet->inet_daddr;
709 	} else {
710 		__be32 addr = inet->inet_rcv_saddr;
711 		if (!addr)
712 			addr = inet->inet_saddr;
713 		sin->sin_port = inet->inet_sport;
714 		sin->sin_addr.s_addr = addr;
715 	}
716 	memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
717 	*uaddr_len = sizeof(*sin);
718 	return 0;
719 }
720 EXPORT_SYMBOL(inet_getname);
721 
722 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
723 {
724 	struct sock *sk = sock->sk;
725 
726 	sock_rps_record_flow(sk);
727 
728 	/* We may need to bind the socket. */
729 	if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
730 	    inet_autobind(sk))
731 		return -EAGAIN;
732 
733 	return sk->sk_prot->sendmsg(sk, msg, size);
734 }
735 EXPORT_SYMBOL(inet_sendmsg);
736 
737 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
738 		      size_t size, int flags)
739 {
740 	struct sock *sk = sock->sk;
741 
742 	sock_rps_record_flow(sk);
743 
744 	/* We may need to bind the socket. */
745 	if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
746 	    inet_autobind(sk))
747 		return -EAGAIN;
748 
749 	if (sk->sk_prot->sendpage)
750 		return sk->sk_prot->sendpage(sk, page, offset, size, flags);
751 	return sock_no_sendpage(sock, page, offset, size, flags);
752 }
753 EXPORT_SYMBOL(inet_sendpage);
754 
755 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
756 		 int flags)
757 {
758 	struct sock *sk = sock->sk;
759 	int addr_len = 0;
760 	int err;
761 
762 	sock_rps_record_flow(sk);
763 
764 	err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
765 				   flags & ~MSG_DONTWAIT, &addr_len);
766 	if (err >= 0)
767 		msg->msg_namelen = addr_len;
768 	return err;
769 }
770 EXPORT_SYMBOL(inet_recvmsg);
771 
772 int inet_shutdown(struct socket *sock, int how)
773 {
774 	struct sock *sk = sock->sk;
775 	int err = 0;
776 
777 	/* This should really check to make sure
778 	 * the socket is a TCP socket. (WHY AC...)
779 	 */
780 	how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
781 		       1->2 bit 2 snds.
782 		       2->3 */
783 	if ((how & ~SHUTDOWN_MASK) || !how)	/* MAXINT->0 */
784 		return -EINVAL;
785 
786 	lock_sock(sk);
787 	if (sock->state == SS_CONNECTING) {
788 		if ((1 << sk->sk_state) &
789 		    (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
790 			sock->state = SS_DISCONNECTING;
791 		else
792 			sock->state = SS_CONNECTED;
793 	}
794 
795 	switch (sk->sk_state) {
796 	case TCP_CLOSE:
797 		err = -ENOTCONN;
798 		/* Hack to wake up other listeners, who can poll for
799 		   POLLHUP, even on eg. unconnected UDP sockets -- RR */
800 	default:
801 		sk->sk_shutdown |= how;
802 		if (sk->sk_prot->shutdown)
803 			sk->sk_prot->shutdown(sk, how);
804 		break;
805 
806 	/* Remaining two branches are temporary solution for missing
807 	 * close() in multithreaded environment. It is _not_ a good idea,
808 	 * but we have no choice until close() is repaired at VFS level.
809 	 */
810 	case TCP_LISTEN:
811 		if (!(how & RCV_SHUTDOWN))
812 			break;
813 		/* Fall through */
814 	case TCP_SYN_SENT:
815 		err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
816 		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
817 		break;
818 	}
819 
820 	/* Wake up anyone sleeping in poll. */
821 	sk->sk_state_change(sk);
822 	release_sock(sk);
823 	return err;
824 }
825 EXPORT_SYMBOL(inet_shutdown);
826 
827 /*
828  *	ioctl() calls you can issue on an INET socket. Most of these are
829  *	device configuration and stuff and very rarely used. Some ioctls
830  *	pass on to the socket itself.
831  *
832  *	NOTE: I like the idea of a module for the config stuff. ie ifconfig
833  *	loads the devconfigure module does its configuring and unloads it.
834  *	There's a good 20K of config code hanging around the kernel.
835  */
836 
837 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
838 {
839 	struct sock *sk = sock->sk;
840 	int err = 0;
841 	struct net *net = sock_net(sk);
842 
843 	switch (cmd) {
844 	case SIOCGSTAMP:
845 		err = sock_get_timestamp(sk, (struct timeval __user *)arg);
846 		break;
847 	case SIOCGSTAMPNS:
848 		err = sock_get_timestampns(sk, (struct timespec __user *)arg);
849 		break;
850 	case SIOCADDRT:
851 	case SIOCDELRT:
852 	case SIOCRTMSG:
853 		err = ip_rt_ioctl(net, cmd, (void __user *)arg);
854 		break;
855 	case SIOCDARP:
856 	case SIOCGARP:
857 	case SIOCSARP:
858 		err = arp_ioctl(net, cmd, (void __user *)arg);
859 		break;
860 	case SIOCGIFADDR:
861 	case SIOCSIFADDR:
862 	case SIOCGIFBRDADDR:
863 	case SIOCSIFBRDADDR:
864 	case SIOCGIFNETMASK:
865 	case SIOCSIFNETMASK:
866 	case SIOCGIFDSTADDR:
867 	case SIOCSIFDSTADDR:
868 	case SIOCSIFPFLAGS:
869 	case SIOCGIFPFLAGS:
870 	case SIOCSIFFLAGS:
871 		err = devinet_ioctl(net, cmd, (void __user *)arg);
872 		break;
873 	default:
874 		if (sk->sk_prot->ioctl)
875 			err = sk->sk_prot->ioctl(sk, cmd, arg);
876 		else
877 			err = -ENOIOCTLCMD;
878 		break;
879 	}
880 	return err;
881 }
882 EXPORT_SYMBOL(inet_ioctl);
883 
884 #ifdef CONFIG_COMPAT
885 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
886 {
887 	struct sock *sk = sock->sk;
888 	int err = -ENOIOCTLCMD;
889 
890 	if (sk->sk_prot->compat_ioctl)
891 		err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
892 
893 	return err;
894 }
895 #endif
896 
897 const struct proto_ops inet_stream_ops = {
898 	.family		   = PF_INET,
899 	.owner		   = THIS_MODULE,
900 	.release	   = inet_release,
901 	.bind		   = inet_bind,
902 	.connect	   = inet_stream_connect,
903 	.socketpair	   = sock_no_socketpair,
904 	.accept		   = inet_accept,
905 	.getname	   = inet_getname,
906 	.poll		   = tcp_poll,
907 	.ioctl		   = inet_ioctl,
908 	.listen		   = inet_listen,
909 	.shutdown	   = inet_shutdown,
910 	.setsockopt	   = sock_common_setsockopt,
911 	.getsockopt	   = sock_common_getsockopt,
912 	.sendmsg	   = inet_sendmsg,
913 	.recvmsg	   = inet_recvmsg,
914 	.mmap		   = sock_no_mmap,
915 	.sendpage	   = inet_sendpage,
916 	.splice_read	   = tcp_splice_read,
917 #ifdef CONFIG_COMPAT
918 	.compat_setsockopt = compat_sock_common_setsockopt,
919 	.compat_getsockopt = compat_sock_common_getsockopt,
920 	.compat_ioctl	   = inet_compat_ioctl,
921 #endif
922 };
923 EXPORT_SYMBOL(inet_stream_ops);
924 
925 const struct proto_ops inet_dgram_ops = {
926 	.family		   = PF_INET,
927 	.owner		   = THIS_MODULE,
928 	.release	   = inet_release,
929 	.bind		   = inet_bind,
930 	.connect	   = inet_dgram_connect,
931 	.socketpair	   = sock_no_socketpair,
932 	.accept		   = sock_no_accept,
933 	.getname	   = inet_getname,
934 	.poll		   = udp_poll,
935 	.ioctl		   = inet_ioctl,
936 	.listen		   = sock_no_listen,
937 	.shutdown	   = inet_shutdown,
938 	.setsockopt	   = sock_common_setsockopt,
939 	.getsockopt	   = sock_common_getsockopt,
940 	.sendmsg	   = inet_sendmsg,
941 	.recvmsg	   = inet_recvmsg,
942 	.mmap		   = sock_no_mmap,
943 	.sendpage	   = inet_sendpage,
944 #ifdef CONFIG_COMPAT
945 	.compat_setsockopt = compat_sock_common_setsockopt,
946 	.compat_getsockopt = compat_sock_common_getsockopt,
947 	.compat_ioctl	   = inet_compat_ioctl,
948 #endif
949 };
950 EXPORT_SYMBOL(inet_dgram_ops);
951 
952 /*
953  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
954  * udp_poll
955  */
956 static const struct proto_ops inet_sockraw_ops = {
957 	.family		   = PF_INET,
958 	.owner		   = THIS_MODULE,
959 	.release	   = inet_release,
960 	.bind		   = inet_bind,
961 	.connect	   = inet_dgram_connect,
962 	.socketpair	   = sock_no_socketpair,
963 	.accept		   = sock_no_accept,
964 	.getname	   = inet_getname,
965 	.poll		   = datagram_poll,
966 	.ioctl		   = inet_ioctl,
967 	.listen		   = sock_no_listen,
968 	.shutdown	   = inet_shutdown,
969 	.setsockopt	   = sock_common_setsockopt,
970 	.getsockopt	   = sock_common_getsockopt,
971 	.sendmsg	   = inet_sendmsg,
972 	.recvmsg	   = inet_recvmsg,
973 	.mmap		   = sock_no_mmap,
974 	.sendpage	   = inet_sendpage,
975 #ifdef CONFIG_COMPAT
976 	.compat_setsockopt = compat_sock_common_setsockopt,
977 	.compat_getsockopt = compat_sock_common_getsockopt,
978 	.compat_ioctl	   = inet_compat_ioctl,
979 #endif
980 };
981 
982 static const struct net_proto_family inet_family_ops = {
983 	.family = PF_INET,
984 	.create = inet_create,
985 	.owner	= THIS_MODULE,
986 };
987 
988 /* Upon startup we insert all the elements in inetsw_array[] into
989  * the linked list inetsw.
990  */
991 static struct inet_protosw inetsw_array[] =
992 {
993 	{
994 		.type =       SOCK_STREAM,
995 		.protocol =   IPPROTO_TCP,
996 		.prot =       &tcp_prot,
997 		.ops =        &inet_stream_ops,
998 		.flags =      INET_PROTOSW_PERMANENT |
999 			      INET_PROTOSW_ICSK,
1000 	},
1001 
1002 	{
1003 		.type =       SOCK_DGRAM,
1004 		.protocol =   IPPROTO_UDP,
1005 		.prot =       &udp_prot,
1006 		.ops =        &inet_dgram_ops,
1007 		.flags =      INET_PROTOSW_PERMANENT,
1008        },
1009 
1010        {
1011 		.type =       SOCK_DGRAM,
1012 		.protocol =   IPPROTO_ICMP,
1013 		.prot =       &ping_prot,
1014 		.ops =        &inet_dgram_ops,
1015 		.flags =      INET_PROTOSW_REUSE,
1016        },
1017 
1018        {
1019 	       .type =       SOCK_RAW,
1020 	       .protocol =   IPPROTO_IP,	/* wild card */
1021 	       .prot =       &raw_prot,
1022 	       .ops =        &inet_sockraw_ops,
1023 	       .flags =      INET_PROTOSW_REUSE,
1024        }
1025 };
1026 
1027 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1028 
1029 void inet_register_protosw(struct inet_protosw *p)
1030 {
1031 	struct list_head *lh;
1032 	struct inet_protosw *answer;
1033 	int protocol = p->protocol;
1034 	struct list_head *last_perm;
1035 
1036 	spin_lock_bh(&inetsw_lock);
1037 
1038 	if (p->type >= SOCK_MAX)
1039 		goto out_illegal;
1040 
1041 	/* If we are trying to override a permanent protocol, bail. */
1042 	last_perm = &inetsw[p->type];
1043 	list_for_each(lh, &inetsw[p->type]) {
1044 		answer = list_entry(lh, struct inet_protosw, list);
1045 		/* Check only the non-wild match. */
1046 		if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1047 			break;
1048 		if (protocol == answer->protocol)
1049 			goto out_permanent;
1050 		last_perm = lh;
1051 	}
1052 
1053 	/* Add the new entry after the last permanent entry if any, so that
1054 	 * the new entry does not override a permanent entry when matched with
1055 	 * a wild-card protocol. But it is allowed to override any existing
1056 	 * non-permanent entry.  This means that when we remove this entry, the
1057 	 * system automatically returns to the old behavior.
1058 	 */
1059 	list_add_rcu(&p->list, last_perm);
1060 out:
1061 	spin_unlock_bh(&inetsw_lock);
1062 
1063 	return;
1064 
1065 out_permanent:
1066 	pr_err("Attempt to override permanent protocol %d\n", protocol);
1067 	goto out;
1068 
1069 out_illegal:
1070 	pr_err("Ignoring attempt to register invalid socket type %d\n",
1071 	       p->type);
1072 	goto out;
1073 }
1074 EXPORT_SYMBOL(inet_register_protosw);
1075 
1076 void inet_unregister_protosw(struct inet_protosw *p)
1077 {
1078 	if (INET_PROTOSW_PERMANENT & p->flags) {
1079 		pr_err("Attempt to unregister permanent protocol %d\n",
1080 		       p->protocol);
1081 	} else {
1082 		spin_lock_bh(&inetsw_lock);
1083 		list_del_rcu(&p->list);
1084 		spin_unlock_bh(&inetsw_lock);
1085 
1086 		synchronize_net();
1087 	}
1088 }
1089 EXPORT_SYMBOL(inet_unregister_protosw);
1090 
1091 /*
1092  *      Shall we try to damage output packets if routing dev changes?
1093  */
1094 
1095 int sysctl_ip_dynaddr __read_mostly;
1096 
1097 static int inet_sk_reselect_saddr(struct sock *sk)
1098 {
1099 	struct inet_sock *inet = inet_sk(sk);
1100 	__be32 old_saddr = inet->inet_saddr;
1101 	__be32 daddr = inet->inet_daddr;
1102 	struct flowi4 *fl4;
1103 	struct rtable *rt;
1104 	__be32 new_saddr;
1105 	struct ip_options_rcu *inet_opt;
1106 
1107 	inet_opt = rcu_dereference_protected(inet->inet_opt,
1108 					     sock_owned_by_user(sk));
1109 	if (inet_opt && inet_opt->opt.srr)
1110 		daddr = inet_opt->opt.faddr;
1111 
1112 	/* Query new route. */
1113 	fl4 = &inet->cork.fl.u.ip4;
1114 	rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1115 			      sk->sk_bound_dev_if, sk->sk_protocol,
1116 			      inet->inet_sport, inet->inet_dport, sk);
1117 	if (IS_ERR(rt))
1118 		return PTR_ERR(rt);
1119 
1120 	sk_setup_caps(sk, &rt->dst);
1121 
1122 	new_saddr = fl4->saddr;
1123 
1124 	if (new_saddr == old_saddr)
1125 		return 0;
1126 
1127 	if (sysctl_ip_dynaddr > 1) {
1128 		pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1129 			__func__, &old_saddr, &new_saddr);
1130 	}
1131 
1132 	inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1133 
1134 	/*
1135 	 * XXX The only one ugly spot where we need to
1136 	 * XXX really change the sockets identity after
1137 	 * XXX it has entered the hashes. -DaveM
1138 	 *
1139 	 * Besides that, it does not check for connection
1140 	 * uniqueness. Wait for troubles.
1141 	 */
1142 	__sk_prot_rehash(sk);
1143 	return 0;
1144 }
1145 
1146 int inet_sk_rebuild_header(struct sock *sk)
1147 {
1148 	struct inet_sock *inet = inet_sk(sk);
1149 	struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1150 	__be32 daddr;
1151 	struct ip_options_rcu *inet_opt;
1152 	struct flowi4 *fl4;
1153 	int err;
1154 
1155 	/* Route is OK, nothing to do. */
1156 	if (rt)
1157 		return 0;
1158 
1159 	/* Reroute. */
1160 	rcu_read_lock();
1161 	inet_opt = rcu_dereference(inet->inet_opt);
1162 	daddr = inet->inet_daddr;
1163 	if (inet_opt && inet_opt->opt.srr)
1164 		daddr = inet_opt->opt.faddr;
1165 	rcu_read_unlock();
1166 	fl4 = &inet->cork.fl.u.ip4;
1167 	rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1168 				   inet->inet_dport, inet->inet_sport,
1169 				   sk->sk_protocol, RT_CONN_FLAGS(sk),
1170 				   sk->sk_bound_dev_if);
1171 	if (!IS_ERR(rt)) {
1172 		err = 0;
1173 		sk_setup_caps(sk, &rt->dst);
1174 	} else {
1175 		err = PTR_ERR(rt);
1176 
1177 		/* Routing failed... */
1178 		sk->sk_route_caps = 0;
1179 		/*
1180 		 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1181 		 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1182 		 */
1183 		if (!sysctl_ip_dynaddr ||
1184 		    sk->sk_state != TCP_SYN_SENT ||
1185 		    (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1186 		    (err = inet_sk_reselect_saddr(sk)) != 0)
1187 			sk->sk_err_soft = -err;
1188 	}
1189 
1190 	return err;
1191 }
1192 EXPORT_SYMBOL(inet_sk_rebuild_header);
1193 
1194 static struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1195 					netdev_features_t features)
1196 {
1197 	struct sk_buff *segs = ERR_PTR(-EINVAL);
1198 	const struct net_offload *ops;
1199 	unsigned int offset = 0;
1200 	bool udpfrag, encap;
1201 	struct iphdr *iph;
1202 	int proto;
1203 	int nhoff;
1204 	int ihl;
1205 	int id;
1206 
1207 	if (unlikely(skb_shinfo(skb)->gso_type &
1208 		     ~(SKB_GSO_TCPV4 |
1209 		       SKB_GSO_UDP |
1210 		       SKB_GSO_DODGY |
1211 		       SKB_GSO_TCP_ECN |
1212 		       SKB_GSO_GRE |
1213 		       SKB_GSO_GRE_CSUM |
1214 		       SKB_GSO_IPIP |
1215 		       SKB_GSO_SIT |
1216 		       SKB_GSO_TCPV6 |
1217 		       SKB_GSO_UDP_TUNNEL |
1218 		       SKB_GSO_UDP_TUNNEL_CSUM |
1219 		       SKB_GSO_TUNNEL_REMCSUM |
1220 		       0)))
1221 		goto out;
1222 
1223 	skb_reset_network_header(skb);
1224 	nhoff = skb_network_header(skb) - skb_mac_header(skb);
1225 	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1226 		goto out;
1227 
1228 	iph = ip_hdr(skb);
1229 	ihl = iph->ihl * 4;
1230 	if (ihl < sizeof(*iph))
1231 		goto out;
1232 
1233 	id = ntohs(iph->id);
1234 	proto = iph->protocol;
1235 
1236 	/* Warning: after this point, iph might be no longer valid */
1237 	if (unlikely(!pskb_may_pull(skb, ihl)))
1238 		goto out;
1239 	__skb_pull(skb, ihl);
1240 
1241 	encap = SKB_GSO_CB(skb)->encap_level > 0;
1242 	if (encap)
1243 		features &= skb->dev->hw_enc_features;
1244 	SKB_GSO_CB(skb)->encap_level += ihl;
1245 
1246 	skb_reset_transport_header(skb);
1247 
1248 	segs = ERR_PTR(-EPROTONOSUPPORT);
1249 
1250 	if (skb->encapsulation &&
1251 	    skb_shinfo(skb)->gso_type & (SKB_GSO_SIT|SKB_GSO_IPIP))
1252 		udpfrag = proto == IPPROTO_UDP && encap;
1253 	else
1254 		udpfrag = proto == IPPROTO_UDP && !skb->encapsulation;
1255 
1256 	ops = rcu_dereference(inet_offloads[proto]);
1257 	if (likely(ops && ops->callbacks.gso_segment))
1258 		segs = ops->callbacks.gso_segment(skb, features);
1259 
1260 	if (IS_ERR_OR_NULL(segs))
1261 		goto out;
1262 
1263 	skb = segs;
1264 	do {
1265 		iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1266 		if (udpfrag) {
1267 			iph->id = htons(id);
1268 			iph->frag_off = htons(offset >> 3);
1269 			if (skb->next)
1270 				iph->frag_off |= htons(IP_MF);
1271 			offset += skb->len - nhoff - ihl;
1272 		} else {
1273 			iph->id = htons(id++);
1274 		}
1275 		iph->tot_len = htons(skb->len - nhoff);
1276 		ip_send_check(iph);
1277 		if (encap)
1278 			skb_reset_inner_headers(skb);
1279 		skb->network_header = (u8 *)iph - skb->head;
1280 	} while ((skb = skb->next));
1281 
1282 out:
1283 	return segs;
1284 }
1285 
1286 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1287 					 struct sk_buff *skb)
1288 {
1289 	const struct net_offload *ops;
1290 	struct sk_buff **pp = NULL;
1291 	struct sk_buff *p;
1292 	const struct iphdr *iph;
1293 	unsigned int hlen;
1294 	unsigned int off;
1295 	unsigned int id;
1296 	int flush = 1;
1297 	int proto;
1298 
1299 	off = skb_gro_offset(skb);
1300 	hlen = off + sizeof(*iph);
1301 	iph = skb_gro_header_fast(skb, off);
1302 	if (skb_gro_header_hard(skb, hlen)) {
1303 		iph = skb_gro_header_slow(skb, hlen, off);
1304 		if (unlikely(!iph))
1305 			goto out;
1306 	}
1307 
1308 	proto = iph->protocol;
1309 
1310 	rcu_read_lock();
1311 	ops = rcu_dereference(inet_offloads[proto]);
1312 	if (!ops || !ops->callbacks.gro_receive)
1313 		goto out_unlock;
1314 
1315 	if (*(u8 *)iph != 0x45)
1316 		goto out_unlock;
1317 
1318 	if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1319 		goto out_unlock;
1320 
1321 	id = ntohl(*(__be32 *)&iph->id);
1322 	flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1323 	id >>= 16;
1324 
1325 	for (p = *head; p; p = p->next) {
1326 		struct iphdr *iph2;
1327 
1328 		if (!NAPI_GRO_CB(p)->same_flow)
1329 			continue;
1330 
1331 		iph2 = (struct iphdr *)(p->data + off);
1332 		/* The above works because, with the exception of the top
1333 		 * (inner most) layer, we only aggregate pkts with the same
1334 		 * hdr length so all the hdrs we'll need to verify will start
1335 		 * at the same offset.
1336 		 */
1337 		if ((iph->protocol ^ iph2->protocol) |
1338 		    ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1339 		    ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1340 			NAPI_GRO_CB(p)->same_flow = 0;
1341 			continue;
1342 		}
1343 
1344 		/* All fields must match except length and checksum. */
1345 		NAPI_GRO_CB(p)->flush |=
1346 			(iph->ttl ^ iph2->ttl) |
1347 			(iph->tos ^ iph2->tos) |
1348 			((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1349 
1350 		/* Save the IP ID check to be included later when we get to
1351 		 * the transport layer so only the inner most IP ID is checked.
1352 		 * This is because some GSO/TSO implementations do not
1353 		 * correctly increment the IP ID for the outer hdrs.
1354 		 */
1355 		NAPI_GRO_CB(p)->flush_id =
1356 			    ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1357 		NAPI_GRO_CB(p)->flush |= flush;
1358 	}
1359 
1360 	NAPI_GRO_CB(skb)->flush |= flush;
1361 	skb_set_network_header(skb, off);
1362 	/* The above will be needed by the transport layer if there is one
1363 	 * immediately following this IP hdr.
1364 	 */
1365 
1366 	/* Note : No need to call skb_gro_postpull_rcsum() here,
1367 	 * as we already checked checksum over ipv4 header was 0
1368 	 */
1369 	skb_gro_pull(skb, sizeof(*iph));
1370 	skb_set_transport_header(skb, skb_gro_offset(skb));
1371 
1372 	pp = ops->callbacks.gro_receive(head, skb);
1373 
1374 out_unlock:
1375 	rcu_read_unlock();
1376 
1377 out:
1378 	NAPI_GRO_CB(skb)->flush |= flush;
1379 
1380 	return pp;
1381 }
1382 
1383 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1384 {
1385 	if (sk->sk_family == AF_INET)
1386 		return ip_recv_error(sk, msg, len, addr_len);
1387 #if IS_ENABLED(CONFIG_IPV6)
1388 	if (sk->sk_family == AF_INET6)
1389 		return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1390 #endif
1391 	return -EINVAL;
1392 }
1393 
1394 static int inet_gro_complete(struct sk_buff *skb, int nhoff)
1395 {
1396 	__be16 newlen = htons(skb->len - nhoff);
1397 	struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1398 	const struct net_offload *ops;
1399 	int proto = iph->protocol;
1400 	int err = -ENOSYS;
1401 
1402 	if (skb->encapsulation)
1403 		skb_set_inner_network_header(skb, nhoff);
1404 
1405 	csum_replace2(&iph->check, iph->tot_len, newlen);
1406 	iph->tot_len = newlen;
1407 
1408 	rcu_read_lock();
1409 	ops = rcu_dereference(inet_offloads[proto]);
1410 	if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1411 		goto out_unlock;
1412 
1413 	/* Only need to add sizeof(*iph) to get to the next hdr below
1414 	 * because any hdr with option will have been flushed in
1415 	 * inet_gro_receive().
1416 	 */
1417 	err = ops->callbacks.gro_complete(skb, nhoff + sizeof(*iph));
1418 
1419 out_unlock:
1420 	rcu_read_unlock();
1421 
1422 	return err;
1423 }
1424 
1425 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1426 			 unsigned short type, unsigned char protocol,
1427 			 struct net *net)
1428 {
1429 	struct socket *sock;
1430 	int rc = sock_create_kern(net, family, type, protocol, &sock);
1431 
1432 	if (rc == 0) {
1433 		*sk = sock->sk;
1434 		(*sk)->sk_allocation = GFP_ATOMIC;
1435 		/*
1436 		 * Unhash it so that IP input processing does not even see it,
1437 		 * we do not wish this socket to see incoming packets.
1438 		 */
1439 		(*sk)->sk_prot->unhash(*sk);
1440 	}
1441 	return rc;
1442 }
1443 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1444 
1445 u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt)
1446 {
1447 	return  *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt);
1448 }
1449 EXPORT_SYMBOL_GPL(snmp_get_cpu_field);
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 += snmp_get_cpu_field(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_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1465 			 size_t syncp_offset)
1466 {
1467 	void *bhptr;
1468 	struct u64_stats_sync *syncp;
1469 	u64 v;
1470 	unsigned int start;
1471 
1472 	bhptr = per_cpu_ptr(mib, cpu);
1473 	syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1474 	do {
1475 		start = u64_stats_fetch_begin_irq(syncp);
1476 		v = *(((u64 *)bhptr) + offt);
1477 	} while (u64_stats_fetch_retry_irq(syncp, start));
1478 
1479 	return v;
1480 }
1481 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1482 
1483 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1484 {
1485 	u64 res = 0;
1486 	int cpu;
1487 
1488 	for_each_possible_cpu(cpu) {
1489 		res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1490 	}
1491 	return res;
1492 }
1493 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1494 #endif
1495 
1496 #ifdef CONFIG_IP_MULTICAST
1497 static const struct net_protocol igmp_protocol = {
1498 	.handler =	igmp_rcv,
1499 	.netns_ok =	1,
1500 };
1501 #endif
1502 
1503 static const struct net_protocol tcp_protocol = {
1504 	.early_demux	=	tcp_v4_early_demux,
1505 	.handler	=	tcp_v4_rcv,
1506 	.err_handler	=	tcp_v4_err,
1507 	.no_policy	=	1,
1508 	.netns_ok	=	1,
1509 	.icmp_strict_tag_validation = 1,
1510 };
1511 
1512 static const struct net_protocol udp_protocol = {
1513 	.early_demux =	udp_v4_early_demux,
1514 	.handler =	udp_rcv,
1515 	.err_handler =	udp_err,
1516 	.no_policy =	1,
1517 	.netns_ok =	1,
1518 };
1519 
1520 static const struct net_protocol icmp_protocol = {
1521 	.handler =	icmp_rcv,
1522 	.err_handler =	icmp_err,
1523 	.no_policy =	1,
1524 	.netns_ok =	1,
1525 };
1526 
1527 static __net_init int ipv4_mib_init_net(struct net *net)
1528 {
1529 	int i;
1530 
1531 	net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1532 	if (!net->mib.tcp_statistics)
1533 		goto err_tcp_mib;
1534 	net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1535 	if (!net->mib.ip_statistics)
1536 		goto err_ip_mib;
1537 
1538 	for_each_possible_cpu(i) {
1539 		struct ipstats_mib *af_inet_stats;
1540 		af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1541 		u64_stats_init(&af_inet_stats->syncp);
1542 	}
1543 
1544 	net->mib.net_statistics = alloc_percpu(struct linux_mib);
1545 	if (!net->mib.net_statistics)
1546 		goto err_net_mib;
1547 	net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1548 	if (!net->mib.udp_statistics)
1549 		goto err_udp_mib;
1550 	net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1551 	if (!net->mib.udplite_statistics)
1552 		goto err_udplite_mib;
1553 	net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1554 	if (!net->mib.icmp_statistics)
1555 		goto err_icmp_mib;
1556 	net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1557 					      GFP_KERNEL);
1558 	if (!net->mib.icmpmsg_statistics)
1559 		goto err_icmpmsg_mib;
1560 
1561 	tcp_mib_init(net);
1562 	return 0;
1563 
1564 err_icmpmsg_mib:
1565 	free_percpu(net->mib.icmp_statistics);
1566 err_icmp_mib:
1567 	free_percpu(net->mib.udplite_statistics);
1568 err_udplite_mib:
1569 	free_percpu(net->mib.udp_statistics);
1570 err_udp_mib:
1571 	free_percpu(net->mib.net_statistics);
1572 err_net_mib:
1573 	free_percpu(net->mib.ip_statistics);
1574 err_ip_mib:
1575 	free_percpu(net->mib.tcp_statistics);
1576 err_tcp_mib:
1577 	return -ENOMEM;
1578 }
1579 
1580 static __net_exit void ipv4_mib_exit_net(struct net *net)
1581 {
1582 	kfree(net->mib.icmpmsg_statistics);
1583 	free_percpu(net->mib.icmp_statistics);
1584 	free_percpu(net->mib.udplite_statistics);
1585 	free_percpu(net->mib.udp_statistics);
1586 	free_percpu(net->mib.net_statistics);
1587 	free_percpu(net->mib.ip_statistics);
1588 	free_percpu(net->mib.tcp_statistics);
1589 }
1590 
1591 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1592 	.init = ipv4_mib_init_net,
1593 	.exit = ipv4_mib_exit_net,
1594 };
1595 
1596 static int __init init_ipv4_mibs(void)
1597 {
1598 	return register_pernet_subsys(&ipv4_mib_ops);
1599 }
1600 
1601 static __net_init int inet_init_net(struct net *net)
1602 {
1603 	/*
1604 	 * Set defaults for local port range
1605 	 */
1606 	seqlock_init(&net->ipv4.ip_local_ports.lock);
1607 	net->ipv4.ip_local_ports.range[0] =  32768;
1608 	net->ipv4.ip_local_ports.range[1] =  60999;
1609 
1610 	seqlock_init(&net->ipv4.ping_group_range.lock);
1611 	/*
1612 	 * Sane defaults - nobody may create ping sockets.
1613 	 * Boot scripts should set this to distro-specific group.
1614 	 */
1615 	net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1616 	net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1617 	return 0;
1618 }
1619 
1620 static __net_exit void inet_exit_net(struct net *net)
1621 {
1622 }
1623 
1624 static __net_initdata struct pernet_operations af_inet_ops = {
1625 	.init = inet_init_net,
1626 	.exit = inet_exit_net,
1627 };
1628 
1629 static int __init init_inet_pernet_ops(void)
1630 {
1631 	return register_pernet_subsys(&af_inet_ops);
1632 }
1633 
1634 static int ipv4_proc_init(void);
1635 
1636 /*
1637  *	IP protocol layer initialiser
1638  */
1639 
1640 static struct packet_offload ip_packet_offload __read_mostly = {
1641 	.type = cpu_to_be16(ETH_P_IP),
1642 	.callbacks = {
1643 		.gso_segment = inet_gso_segment,
1644 		.gro_receive = inet_gro_receive,
1645 		.gro_complete = inet_gro_complete,
1646 	},
1647 };
1648 
1649 static const struct net_offload ipip_offload = {
1650 	.callbacks = {
1651 		.gso_segment	= inet_gso_segment,
1652 		.gro_receive	= inet_gro_receive,
1653 		.gro_complete	= inet_gro_complete,
1654 	},
1655 };
1656 
1657 static int __init ipv4_offload_init(void)
1658 {
1659 	/*
1660 	 * Add offloads
1661 	 */
1662 	if (udpv4_offload_init() < 0)
1663 		pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1664 	if (tcpv4_offload_init() < 0)
1665 		pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1666 
1667 	dev_add_offload(&ip_packet_offload);
1668 	inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1669 	return 0;
1670 }
1671 
1672 fs_initcall(ipv4_offload_init);
1673 
1674 static struct packet_type ip_packet_type __read_mostly = {
1675 	.type = cpu_to_be16(ETH_P_IP),
1676 	.func = ip_rcv,
1677 };
1678 
1679 static int __init inet_init(void)
1680 {
1681 	struct inet_protosw *q;
1682 	struct list_head *r;
1683 	int rc = -EINVAL;
1684 
1685 	sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1686 
1687 	rc = proto_register(&tcp_prot, 1);
1688 	if (rc)
1689 		goto out;
1690 
1691 	rc = proto_register(&udp_prot, 1);
1692 	if (rc)
1693 		goto out_unregister_tcp_proto;
1694 
1695 	rc = proto_register(&raw_prot, 1);
1696 	if (rc)
1697 		goto out_unregister_udp_proto;
1698 
1699 	rc = proto_register(&ping_prot, 1);
1700 	if (rc)
1701 		goto out_unregister_raw_proto;
1702 
1703 	/*
1704 	 *	Tell SOCKET that we are alive...
1705 	 */
1706 
1707 	(void)sock_register(&inet_family_ops);
1708 
1709 #ifdef CONFIG_SYSCTL
1710 	ip_static_sysctl_init();
1711 #endif
1712 
1713 	/*
1714 	 *	Add all the base protocols.
1715 	 */
1716 
1717 	if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1718 		pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1719 	if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1720 		pr_crit("%s: Cannot add UDP protocol\n", __func__);
1721 	if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1722 		pr_crit("%s: Cannot add TCP protocol\n", __func__);
1723 #ifdef CONFIG_IP_MULTICAST
1724 	if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1725 		pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1726 #endif
1727 
1728 	/* Register the socket-side information for inet_create. */
1729 	for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1730 		INIT_LIST_HEAD(r);
1731 
1732 	for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1733 		inet_register_protosw(q);
1734 
1735 	/*
1736 	 *	Set the ARP module up
1737 	 */
1738 
1739 	arp_init();
1740 
1741 	/*
1742 	 *	Set the IP module up
1743 	 */
1744 
1745 	ip_init();
1746 
1747 	tcp_v4_init();
1748 
1749 	/* Setup TCP slab cache for open requests. */
1750 	tcp_init();
1751 
1752 	/* Setup UDP memory threshold */
1753 	udp_init();
1754 
1755 	/* Add UDP-Lite (RFC 3828) */
1756 	udplite4_register();
1757 
1758 	ping_init();
1759 
1760 	/*
1761 	 *	Set the ICMP layer up
1762 	 */
1763 
1764 	if (icmp_init() < 0)
1765 		panic("Failed to create the ICMP control socket.\n");
1766 
1767 	/*
1768 	 *	Initialise the multicast router
1769 	 */
1770 #if defined(CONFIG_IP_MROUTE)
1771 	if (ip_mr_init())
1772 		pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1773 #endif
1774 
1775 	if (init_inet_pernet_ops())
1776 		pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
1777 	/*
1778 	 *	Initialise per-cpu ipv4 mibs
1779 	 */
1780 
1781 	if (init_ipv4_mibs())
1782 		pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1783 
1784 	ipv4_proc_init();
1785 
1786 	ipfrag_init();
1787 
1788 	dev_add_pack(&ip_packet_type);
1789 
1790 	ip_tunnel_core_init();
1791 
1792 	rc = 0;
1793 out:
1794 	return rc;
1795 out_unregister_raw_proto:
1796 	proto_unregister(&raw_prot);
1797 out_unregister_udp_proto:
1798 	proto_unregister(&udp_prot);
1799 out_unregister_tcp_proto:
1800 	proto_unregister(&tcp_prot);
1801 	goto out;
1802 }
1803 
1804 fs_initcall(inet_init);
1805 
1806 /* ------------------------------------------------------------------------ */
1807 
1808 #ifdef CONFIG_PROC_FS
1809 static int __init ipv4_proc_init(void)
1810 {
1811 	int rc = 0;
1812 
1813 	if (raw_proc_init())
1814 		goto out_raw;
1815 	if (tcp4_proc_init())
1816 		goto out_tcp;
1817 	if (udp4_proc_init())
1818 		goto out_udp;
1819 	if (ping_proc_init())
1820 		goto out_ping;
1821 	if (ip_misc_proc_init())
1822 		goto out_misc;
1823 out:
1824 	return rc;
1825 out_misc:
1826 	ping_proc_exit();
1827 out_ping:
1828 	udp4_proc_exit();
1829 out_udp:
1830 	tcp4_proc_exit();
1831 out_tcp:
1832 	raw_proc_exit();
1833 out_raw:
1834 	rc = -ENOMEM;
1835 	goto out;
1836 }
1837 
1838 #else /* CONFIG_PROC_FS */
1839 static int __init ipv4_proc_init(void)
1840 {
1841 	return 0;
1842 }
1843 #endif /* CONFIG_PROC_FS */
1844 
1845 MODULE_ALIAS_NETPROTO(PF_INET);
1846 
1847