xref: /openbmc/linux/net/ipv4/af_inet.c (revision b96fc2f3)
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/vrf.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) {
223 			if ((sysctl_tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) != 0)
224 				err = fastopen_init_queue(sk, backlog);
225 			else if ((sysctl_tcp_fastopen &
226 				  TFO_SERVER_WO_SOCKOPT2) != 0)
227 				err = fastopen_init_queue(sk,
228 				    ((uint)sysctl_tcp_fastopen) >> 16);
229 			else
230 				err = 0;
231 			if (err)
232 				goto out;
233 
234 			tcp_fastopen_init_key_once(true);
235 		}
236 		err = inet_csk_listen_start(sk, backlog);
237 		if (err)
238 			goto out;
239 	}
240 	sk->sk_max_ack_backlog = backlog;
241 	err = 0;
242 
243 out:
244 	release_sock(sk);
245 	return err;
246 }
247 EXPORT_SYMBOL(inet_listen);
248 
249 /*
250  *	Create an inet socket.
251  */
252 
253 static int inet_create(struct net *net, struct socket *sock, int protocol,
254 		       int kern)
255 {
256 	struct sock *sk;
257 	struct inet_protosw *answer;
258 	struct inet_sock *inet;
259 	struct proto *answer_prot;
260 	unsigned char answer_flags;
261 	int try_loading_module = 0;
262 	int err;
263 
264 	sock->state = SS_UNCONNECTED;
265 
266 	/* Look for the requested type/protocol pair. */
267 lookup_protocol:
268 	err = -ESOCKTNOSUPPORT;
269 	rcu_read_lock();
270 	list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
271 
272 		err = 0;
273 		/* Check the non-wild match. */
274 		if (protocol == answer->protocol) {
275 			if (protocol != IPPROTO_IP)
276 				break;
277 		} else {
278 			/* Check for the two wild cases. */
279 			if (IPPROTO_IP == protocol) {
280 				protocol = answer->protocol;
281 				break;
282 			}
283 			if (IPPROTO_IP == answer->protocol)
284 				break;
285 		}
286 		err = -EPROTONOSUPPORT;
287 	}
288 
289 	if (unlikely(err)) {
290 		if (try_loading_module < 2) {
291 			rcu_read_unlock();
292 			/*
293 			 * Be more specific, e.g. net-pf-2-proto-132-type-1
294 			 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
295 			 */
296 			if (++try_loading_module == 1)
297 				request_module("net-pf-%d-proto-%d-type-%d",
298 					       PF_INET, protocol, sock->type);
299 			/*
300 			 * Fall back to generic, e.g. net-pf-2-proto-132
301 			 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
302 			 */
303 			else
304 				request_module("net-pf-%d-proto-%d",
305 					       PF_INET, protocol);
306 			goto lookup_protocol;
307 		} else
308 			goto out_rcu_unlock;
309 	}
310 
311 	err = -EPERM;
312 	if (sock->type == SOCK_RAW && !kern &&
313 	    !ns_capable(net->user_ns, CAP_NET_RAW))
314 		goto out_rcu_unlock;
315 
316 	sock->ops = answer->ops;
317 	answer_prot = answer->prot;
318 	answer_flags = answer->flags;
319 	rcu_read_unlock();
320 
321 	WARN_ON(!answer_prot->slab);
322 
323 	err = -ENOBUFS;
324 	sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
325 	if (!sk)
326 		goto out;
327 
328 	err = 0;
329 	if (INET_PROTOSW_REUSE & answer_flags)
330 		sk->sk_reuse = SK_CAN_REUSE;
331 
332 	inet = inet_sk(sk);
333 	inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
334 
335 	inet->nodefrag = 0;
336 
337 	if (SOCK_RAW == sock->type) {
338 		inet->inet_num = protocol;
339 		if (IPPROTO_RAW == protocol)
340 			inet->hdrincl = 1;
341 	}
342 
343 	if (net->ipv4.sysctl_ip_no_pmtu_disc)
344 		inet->pmtudisc = IP_PMTUDISC_DONT;
345 	else
346 		inet->pmtudisc = IP_PMTUDISC_WANT;
347 
348 	inet->inet_id = 0;
349 
350 	sock_init_data(sock, sk);
351 
352 	sk->sk_destruct	   = inet_sock_destruct;
353 	sk->sk_protocol	   = protocol;
354 	sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
355 
356 	inet->uc_ttl	= -1;
357 	inet->mc_loop	= 1;
358 	inet->mc_ttl	= 1;
359 	inet->mc_all	= 1;
360 	inet->mc_index	= 0;
361 	inet->mc_list	= NULL;
362 	inet->rcv_tos	= 0;
363 
364 	sk_refcnt_debug_inc(sk);
365 
366 	if (inet->inet_num) {
367 		/* It assumes that any protocol which allows
368 		 * the user to assign a number at socket
369 		 * creation time automatically
370 		 * shares.
371 		 */
372 		inet->inet_sport = htons(inet->inet_num);
373 		/* Add to protocol hash chains. */
374 		sk->sk_prot->hash(sk);
375 	}
376 
377 	if (sk->sk_prot->init) {
378 		err = sk->sk_prot->init(sk);
379 		if (err)
380 			sk_common_release(sk);
381 	}
382 out:
383 	return err;
384 out_rcu_unlock:
385 	rcu_read_unlock();
386 	goto out;
387 }
388 
389 
390 /*
391  *	The peer socket should always be NULL (or else). When we call this
392  *	function we are destroying the object and from then on nobody
393  *	should refer to it.
394  */
395 int inet_release(struct socket *sock)
396 {
397 	struct sock *sk = sock->sk;
398 
399 	if (sk) {
400 		long timeout;
401 
402 		/* Applications forget to leave groups before exiting */
403 		ip_mc_drop_socket(sk);
404 
405 		/* If linger is set, we don't return until the close
406 		 * is complete.  Otherwise we return immediately. The
407 		 * actually closing is done the same either way.
408 		 *
409 		 * If the close is due to the process exiting, we never
410 		 * linger..
411 		 */
412 		timeout = 0;
413 		if (sock_flag(sk, SOCK_LINGER) &&
414 		    !(current->flags & PF_EXITING))
415 			timeout = sk->sk_lingertime;
416 		sock->sk = NULL;
417 		sk->sk_prot->close(sk, timeout);
418 	}
419 	return 0;
420 }
421 EXPORT_SYMBOL(inet_release);
422 
423 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
424 {
425 	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
426 	struct sock *sk = sock->sk;
427 	struct inet_sock *inet = inet_sk(sk);
428 	struct net *net = sock_net(sk);
429 	unsigned short snum;
430 	int chk_addr_ret;
431 	u32 tb_id = RT_TABLE_LOCAL;
432 	int err;
433 
434 	/* If the socket has its own bind function then use it. (RAW) */
435 	if (sk->sk_prot->bind) {
436 		err = sk->sk_prot->bind(sk, uaddr, addr_len);
437 		goto out;
438 	}
439 	err = -EINVAL;
440 	if (addr_len < sizeof(struct sockaddr_in))
441 		goto out;
442 
443 	if (addr->sin_family != AF_INET) {
444 		/* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
445 		 * only if s_addr is INADDR_ANY.
446 		 */
447 		err = -EAFNOSUPPORT;
448 		if (addr->sin_family != AF_UNSPEC ||
449 		    addr->sin_addr.s_addr != htonl(INADDR_ANY))
450 			goto out;
451 	}
452 
453 	tb_id = vrf_dev_table_ifindex(net, sk->sk_bound_dev_if) ? : tb_id;
454 	chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
455 
456 	/* Not specified by any standard per-se, however it breaks too
457 	 * many applications when removed.  It is unfortunate since
458 	 * allowing applications to make a non-local bind solves
459 	 * several problems with systems using dynamic addressing.
460 	 * (ie. your servers still start up even if your ISDN link
461 	 *  is temporarily down)
462 	 */
463 	err = -EADDRNOTAVAIL;
464 	if (!net->ipv4.sysctl_ip_nonlocal_bind &&
465 	    !(inet->freebind || inet->transparent) &&
466 	    addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
467 	    chk_addr_ret != RTN_LOCAL &&
468 	    chk_addr_ret != RTN_MULTICAST &&
469 	    chk_addr_ret != RTN_BROADCAST)
470 		goto out;
471 
472 	snum = ntohs(addr->sin_port);
473 	err = -EACCES;
474 	if (snum && snum < PROT_SOCK &&
475 	    !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
476 		goto out;
477 
478 	/*      We keep a pair of addresses. rcv_saddr is the one
479 	 *      used by hash lookups, and saddr is used for transmit.
480 	 *
481 	 *      In the BSD API these are the same except where it
482 	 *      would be illegal to use them (multicast/broadcast) in
483 	 *      which case the sending device address is used.
484 	 */
485 	lock_sock(sk);
486 
487 	/* Check these errors (active socket, double bind). */
488 	err = -EINVAL;
489 	if (sk->sk_state != TCP_CLOSE || inet->inet_num)
490 		goto out_release_sock;
491 
492 	inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
493 	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
494 		inet->inet_saddr = 0;  /* Use device */
495 
496 	/* Make sure we are allowed to bind here. */
497 	if ((snum || !inet->bind_address_no_port) &&
498 	    sk->sk_prot->get_port(sk, snum)) {
499 		inet->inet_saddr = inet->inet_rcv_saddr = 0;
500 		err = -EADDRINUSE;
501 		goto out_release_sock;
502 	}
503 
504 	if (inet->inet_rcv_saddr)
505 		sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
506 	if (snum)
507 		sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
508 	inet->inet_sport = htons(inet->inet_num);
509 	inet->inet_daddr = 0;
510 	inet->inet_dport = 0;
511 	sk_dst_reset(sk);
512 	err = 0;
513 out_release_sock:
514 	release_sock(sk);
515 out:
516 	return err;
517 }
518 EXPORT_SYMBOL(inet_bind);
519 
520 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
521 		       int addr_len, int flags)
522 {
523 	struct sock *sk = sock->sk;
524 
525 	if (addr_len < sizeof(uaddr->sa_family))
526 		return -EINVAL;
527 	if (uaddr->sa_family == AF_UNSPEC)
528 		return sk->sk_prot->disconnect(sk, flags);
529 
530 	if (!inet_sk(sk)->inet_num && inet_autobind(sk))
531 		return -EAGAIN;
532 	return sk->sk_prot->connect(sk, uaddr, addr_len);
533 }
534 EXPORT_SYMBOL(inet_dgram_connect);
535 
536 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
537 {
538 	DEFINE_WAIT(wait);
539 
540 	prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
541 	sk->sk_write_pending += writebias;
542 
543 	/* Basic assumption: if someone sets sk->sk_err, he _must_
544 	 * change state of the socket from TCP_SYN_*.
545 	 * Connect() does not allow to get error notifications
546 	 * without closing the socket.
547 	 */
548 	while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
549 		release_sock(sk);
550 		timeo = schedule_timeout(timeo);
551 		lock_sock(sk);
552 		if (signal_pending(current) || !timeo)
553 			break;
554 		prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
555 	}
556 	finish_wait(sk_sleep(sk), &wait);
557 	sk->sk_write_pending -= writebias;
558 	return timeo;
559 }
560 
561 /*
562  *	Connect to a remote host. There is regrettably still a little
563  *	TCP 'magic' in here.
564  */
565 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
566 			  int addr_len, int flags)
567 {
568 	struct sock *sk = sock->sk;
569 	int err;
570 	long timeo;
571 
572 	if (addr_len < sizeof(uaddr->sa_family))
573 		return -EINVAL;
574 
575 	if (uaddr->sa_family == AF_UNSPEC) {
576 		err = sk->sk_prot->disconnect(sk, flags);
577 		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
578 		goto out;
579 	}
580 
581 	switch (sock->state) {
582 	default:
583 		err = -EINVAL;
584 		goto out;
585 	case SS_CONNECTED:
586 		err = -EISCONN;
587 		goto out;
588 	case SS_CONNECTING:
589 		err = -EALREADY;
590 		/* Fall out of switch with err, set for this state */
591 		break;
592 	case SS_UNCONNECTED:
593 		err = -EISCONN;
594 		if (sk->sk_state != TCP_CLOSE)
595 			goto out;
596 
597 		err = sk->sk_prot->connect(sk, uaddr, addr_len);
598 		if (err < 0)
599 			goto out;
600 
601 		sock->state = SS_CONNECTING;
602 
603 		/* Just entered SS_CONNECTING state; the only
604 		 * difference is that return value in non-blocking
605 		 * case is EINPROGRESS, rather than EALREADY.
606 		 */
607 		err = -EINPROGRESS;
608 		break;
609 	}
610 
611 	timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
612 
613 	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
614 		int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
615 				tcp_sk(sk)->fastopen_req &&
616 				tcp_sk(sk)->fastopen_req->data ? 1 : 0;
617 
618 		/* Error code is set above */
619 		if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
620 			goto out;
621 
622 		err = sock_intr_errno(timeo);
623 		if (signal_pending(current))
624 			goto out;
625 	}
626 
627 	/* Connection was closed by RST, timeout, ICMP error
628 	 * or another process disconnected us.
629 	 */
630 	if (sk->sk_state == TCP_CLOSE)
631 		goto sock_error;
632 
633 	/* sk->sk_err may be not zero now, if RECVERR was ordered by user
634 	 * and error was received after socket entered established state.
635 	 * Hence, it is handled normally after connect() return successfully.
636 	 */
637 
638 	sock->state = SS_CONNECTED;
639 	err = 0;
640 out:
641 	return err;
642 
643 sock_error:
644 	err = sock_error(sk) ? : -ECONNABORTED;
645 	sock->state = SS_UNCONNECTED;
646 	if (sk->sk_prot->disconnect(sk, flags))
647 		sock->state = SS_DISCONNECTING;
648 	goto out;
649 }
650 EXPORT_SYMBOL(__inet_stream_connect);
651 
652 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
653 			int addr_len, int flags)
654 {
655 	int err;
656 
657 	lock_sock(sock->sk);
658 	err = __inet_stream_connect(sock, uaddr, addr_len, flags);
659 	release_sock(sock->sk);
660 	return err;
661 }
662 EXPORT_SYMBOL(inet_stream_connect);
663 
664 /*
665  *	Accept a pending connection. The TCP layer now gives BSD semantics.
666  */
667 
668 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
669 {
670 	struct sock *sk1 = sock->sk;
671 	int err = -EINVAL;
672 	struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
673 
674 	if (!sk2)
675 		goto do_err;
676 
677 	lock_sock(sk2);
678 
679 	sock_rps_record_flow(sk2);
680 	WARN_ON(!((1 << sk2->sk_state) &
681 		  (TCPF_ESTABLISHED | TCPF_SYN_RECV |
682 		  TCPF_CLOSE_WAIT | TCPF_CLOSE)));
683 
684 	sock_graft(sk2, newsock);
685 
686 	newsock->state = SS_CONNECTED;
687 	err = 0;
688 	release_sock(sk2);
689 do_err:
690 	return err;
691 }
692 EXPORT_SYMBOL(inet_accept);
693 
694 
695 /*
696  *	This does both peername and sockname.
697  */
698 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
699 			int *uaddr_len, int peer)
700 {
701 	struct sock *sk		= sock->sk;
702 	struct inet_sock *inet	= inet_sk(sk);
703 	DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
704 
705 	sin->sin_family = AF_INET;
706 	if (peer) {
707 		if (!inet->inet_dport ||
708 		    (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
709 		     peer == 1))
710 			return -ENOTCONN;
711 		sin->sin_port = inet->inet_dport;
712 		sin->sin_addr.s_addr = inet->inet_daddr;
713 	} else {
714 		__be32 addr = inet->inet_rcv_saddr;
715 		if (!addr)
716 			addr = inet->inet_saddr;
717 		sin->sin_port = inet->inet_sport;
718 		sin->sin_addr.s_addr = addr;
719 	}
720 	memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
721 	*uaddr_len = sizeof(*sin);
722 	return 0;
723 }
724 EXPORT_SYMBOL(inet_getname);
725 
726 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
727 {
728 	struct sock *sk = sock->sk;
729 
730 	sock_rps_record_flow(sk);
731 
732 	/* We may need to bind the socket. */
733 	if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
734 	    inet_autobind(sk))
735 		return -EAGAIN;
736 
737 	return sk->sk_prot->sendmsg(sk, msg, size);
738 }
739 EXPORT_SYMBOL(inet_sendmsg);
740 
741 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
742 		      size_t size, int flags)
743 {
744 	struct sock *sk = sock->sk;
745 
746 	sock_rps_record_flow(sk);
747 
748 	/* We may need to bind the socket. */
749 	if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
750 	    inet_autobind(sk))
751 		return -EAGAIN;
752 
753 	if (sk->sk_prot->sendpage)
754 		return sk->sk_prot->sendpage(sk, page, offset, size, flags);
755 	return sock_no_sendpage(sock, page, offset, size, flags);
756 }
757 EXPORT_SYMBOL(inet_sendpage);
758 
759 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
760 		 int flags)
761 {
762 	struct sock *sk = sock->sk;
763 	int addr_len = 0;
764 	int err;
765 
766 	sock_rps_record_flow(sk);
767 
768 	err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
769 				   flags & ~MSG_DONTWAIT, &addr_len);
770 	if (err >= 0)
771 		msg->msg_namelen = addr_len;
772 	return err;
773 }
774 EXPORT_SYMBOL(inet_recvmsg);
775 
776 int inet_shutdown(struct socket *sock, int how)
777 {
778 	struct sock *sk = sock->sk;
779 	int err = 0;
780 
781 	/* This should really check to make sure
782 	 * the socket is a TCP socket. (WHY AC...)
783 	 */
784 	how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
785 		       1->2 bit 2 snds.
786 		       2->3 */
787 	if ((how & ~SHUTDOWN_MASK) || !how)	/* MAXINT->0 */
788 		return -EINVAL;
789 
790 	lock_sock(sk);
791 	if (sock->state == SS_CONNECTING) {
792 		if ((1 << sk->sk_state) &
793 		    (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
794 			sock->state = SS_DISCONNECTING;
795 		else
796 			sock->state = SS_CONNECTED;
797 	}
798 
799 	switch (sk->sk_state) {
800 	case TCP_CLOSE:
801 		err = -ENOTCONN;
802 		/* Hack to wake up other listeners, who can poll for
803 		   POLLHUP, even on eg. unconnected UDP sockets -- RR */
804 	default:
805 		sk->sk_shutdown |= how;
806 		if (sk->sk_prot->shutdown)
807 			sk->sk_prot->shutdown(sk, how);
808 		break;
809 
810 	/* Remaining two branches are temporary solution for missing
811 	 * close() in multithreaded environment. It is _not_ a good idea,
812 	 * but we have no choice until close() is repaired at VFS level.
813 	 */
814 	case TCP_LISTEN:
815 		if (!(how & RCV_SHUTDOWN))
816 			break;
817 		/* Fall through */
818 	case TCP_SYN_SENT:
819 		err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
820 		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
821 		break;
822 	}
823 
824 	/* Wake up anyone sleeping in poll. */
825 	sk->sk_state_change(sk);
826 	release_sock(sk);
827 	return err;
828 }
829 EXPORT_SYMBOL(inet_shutdown);
830 
831 /*
832  *	ioctl() calls you can issue on an INET socket. Most of these are
833  *	device configuration and stuff and very rarely used. Some ioctls
834  *	pass on to the socket itself.
835  *
836  *	NOTE: I like the idea of a module for the config stuff. ie ifconfig
837  *	loads the devconfigure module does its configuring and unloads it.
838  *	There's a good 20K of config code hanging around the kernel.
839  */
840 
841 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
842 {
843 	struct sock *sk = sock->sk;
844 	int err = 0;
845 	struct net *net = sock_net(sk);
846 
847 	switch (cmd) {
848 	case SIOCGSTAMP:
849 		err = sock_get_timestamp(sk, (struct timeval __user *)arg);
850 		break;
851 	case SIOCGSTAMPNS:
852 		err = sock_get_timestampns(sk, (struct timespec __user *)arg);
853 		break;
854 	case SIOCADDRT:
855 	case SIOCDELRT:
856 	case SIOCRTMSG:
857 		err = ip_rt_ioctl(net, cmd, (void __user *)arg);
858 		break;
859 	case SIOCDARP:
860 	case SIOCGARP:
861 	case SIOCSARP:
862 		err = arp_ioctl(net, cmd, (void __user *)arg);
863 		break;
864 	case SIOCGIFADDR:
865 	case SIOCSIFADDR:
866 	case SIOCGIFBRDADDR:
867 	case SIOCSIFBRDADDR:
868 	case SIOCGIFNETMASK:
869 	case SIOCSIFNETMASK:
870 	case SIOCGIFDSTADDR:
871 	case SIOCSIFDSTADDR:
872 	case SIOCSIFPFLAGS:
873 	case SIOCGIFPFLAGS:
874 	case SIOCSIFFLAGS:
875 		err = devinet_ioctl(net, cmd, (void __user *)arg);
876 		break;
877 	default:
878 		if (sk->sk_prot->ioctl)
879 			err = sk->sk_prot->ioctl(sk, cmd, arg);
880 		else
881 			err = -ENOIOCTLCMD;
882 		break;
883 	}
884 	return err;
885 }
886 EXPORT_SYMBOL(inet_ioctl);
887 
888 #ifdef CONFIG_COMPAT
889 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
890 {
891 	struct sock *sk = sock->sk;
892 	int err = -ENOIOCTLCMD;
893 
894 	if (sk->sk_prot->compat_ioctl)
895 		err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
896 
897 	return err;
898 }
899 #endif
900 
901 const struct proto_ops inet_stream_ops = {
902 	.family		   = PF_INET,
903 	.owner		   = THIS_MODULE,
904 	.release	   = inet_release,
905 	.bind		   = inet_bind,
906 	.connect	   = inet_stream_connect,
907 	.socketpair	   = sock_no_socketpair,
908 	.accept		   = inet_accept,
909 	.getname	   = inet_getname,
910 	.poll		   = tcp_poll,
911 	.ioctl		   = inet_ioctl,
912 	.listen		   = inet_listen,
913 	.shutdown	   = inet_shutdown,
914 	.setsockopt	   = sock_common_setsockopt,
915 	.getsockopt	   = sock_common_getsockopt,
916 	.sendmsg	   = inet_sendmsg,
917 	.recvmsg	   = inet_recvmsg,
918 	.mmap		   = sock_no_mmap,
919 	.sendpage	   = inet_sendpage,
920 	.splice_read	   = tcp_splice_read,
921 #ifdef CONFIG_COMPAT
922 	.compat_setsockopt = compat_sock_common_setsockopt,
923 	.compat_getsockopt = compat_sock_common_getsockopt,
924 	.compat_ioctl	   = inet_compat_ioctl,
925 #endif
926 };
927 EXPORT_SYMBOL(inet_stream_ops);
928 
929 const struct proto_ops inet_dgram_ops = {
930 	.family		   = PF_INET,
931 	.owner		   = THIS_MODULE,
932 	.release	   = inet_release,
933 	.bind		   = inet_bind,
934 	.connect	   = inet_dgram_connect,
935 	.socketpair	   = sock_no_socketpair,
936 	.accept		   = sock_no_accept,
937 	.getname	   = inet_getname,
938 	.poll		   = udp_poll,
939 	.ioctl		   = inet_ioctl,
940 	.listen		   = sock_no_listen,
941 	.shutdown	   = inet_shutdown,
942 	.setsockopt	   = sock_common_setsockopt,
943 	.getsockopt	   = sock_common_getsockopt,
944 	.sendmsg	   = inet_sendmsg,
945 	.recvmsg	   = inet_recvmsg,
946 	.mmap		   = sock_no_mmap,
947 	.sendpage	   = inet_sendpage,
948 #ifdef CONFIG_COMPAT
949 	.compat_setsockopt = compat_sock_common_setsockopt,
950 	.compat_getsockopt = compat_sock_common_getsockopt,
951 	.compat_ioctl	   = inet_compat_ioctl,
952 #endif
953 };
954 EXPORT_SYMBOL(inet_dgram_ops);
955 
956 /*
957  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
958  * udp_poll
959  */
960 static const struct proto_ops inet_sockraw_ops = {
961 	.family		   = PF_INET,
962 	.owner		   = THIS_MODULE,
963 	.release	   = inet_release,
964 	.bind		   = inet_bind,
965 	.connect	   = inet_dgram_connect,
966 	.socketpair	   = sock_no_socketpair,
967 	.accept		   = sock_no_accept,
968 	.getname	   = inet_getname,
969 	.poll		   = datagram_poll,
970 	.ioctl		   = inet_ioctl,
971 	.listen		   = sock_no_listen,
972 	.shutdown	   = inet_shutdown,
973 	.setsockopt	   = sock_common_setsockopt,
974 	.getsockopt	   = sock_common_getsockopt,
975 	.sendmsg	   = inet_sendmsg,
976 	.recvmsg	   = inet_recvmsg,
977 	.mmap		   = sock_no_mmap,
978 	.sendpage	   = inet_sendpage,
979 #ifdef CONFIG_COMPAT
980 	.compat_setsockopt = compat_sock_common_setsockopt,
981 	.compat_getsockopt = compat_sock_common_getsockopt,
982 	.compat_ioctl	   = inet_compat_ioctl,
983 #endif
984 };
985 
986 static const struct net_proto_family inet_family_ops = {
987 	.family = PF_INET,
988 	.create = inet_create,
989 	.owner	= THIS_MODULE,
990 };
991 
992 /* Upon startup we insert all the elements in inetsw_array[] into
993  * the linked list inetsw.
994  */
995 static struct inet_protosw inetsw_array[] =
996 {
997 	{
998 		.type =       SOCK_STREAM,
999 		.protocol =   IPPROTO_TCP,
1000 		.prot =       &tcp_prot,
1001 		.ops =        &inet_stream_ops,
1002 		.flags =      INET_PROTOSW_PERMANENT |
1003 			      INET_PROTOSW_ICSK,
1004 	},
1005 
1006 	{
1007 		.type =       SOCK_DGRAM,
1008 		.protocol =   IPPROTO_UDP,
1009 		.prot =       &udp_prot,
1010 		.ops =        &inet_dgram_ops,
1011 		.flags =      INET_PROTOSW_PERMANENT,
1012        },
1013 
1014        {
1015 		.type =       SOCK_DGRAM,
1016 		.protocol =   IPPROTO_ICMP,
1017 		.prot =       &ping_prot,
1018 		.ops =        &inet_dgram_ops,
1019 		.flags =      INET_PROTOSW_REUSE,
1020        },
1021 
1022        {
1023 	       .type =       SOCK_RAW,
1024 	       .protocol =   IPPROTO_IP,	/* wild card */
1025 	       .prot =       &raw_prot,
1026 	       .ops =        &inet_sockraw_ops,
1027 	       .flags =      INET_PROTOSW_REUSE,
1028        }
1029 };
1030 
1031 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1032 
1033 void inet_register_protosw(struct inet_protosw *p)
1034 {
1035 	struct list_head *lh;
1036 	struct inet_protosw *answer;
1037 	int protocol = p->protocol;
1038 	struct list_head *last_perm;
1039 
1040 	spin_lock_bh(&inetsw_lock);
1041 
1042 	if (p->type >= SOCK_MAX)
1043 		goto out_illegal;
1044 
1045 	/* If we are trying to override a permanent protocol, bail. */
1046 	answer = NULL;
1047 	last_perm = &inetsw[p->type];
1048 	list_for_each(lh, &inetsw[p->type]) {
1049 		answer = list_entry(lh, struct inet_protosw, list);
1050 
1051 		/* Check only the non-wild match. */
1052 		if (INET_PROTOSW_PERMANENT & answer->flags) {
1053 			if (protocol == answer->protocol)
1054 				break;
1055 			last_perm = lh;
1056 		}
1057 
1058 		answer = NULL;
1059 	}
1060 	if (answer)
1061 		goto out_permanent;
1062 
1063 	/* Add the new entry after the last permanent entry if any, so that
1064 	 * the new entry does not override a permanent entry when matched with
1065 	 * a wild-card protocol. But it is allowed to override any existing
1066 	 * non-permanent entry.  This means that when we remove this entry, the
1067 	 * system automatically returns to the old behavior.
1068 	 */
1069 	list_add_rcu(&p->list, last_perm);
1070 out:
1071 	spin_unlock_bh(&inetsw_lock);
1072 
1073 	return;
1074 
1075 out_permanent:
1076 	pr_err("Attempt to override permanent protocol %d\n", protocol);
1077 	goto out;
1078 
1079 out_illegal:
1080 	pr_err("Ignoring attempt to register invalid socket type %d\n",
1081 	       p->type);
1082 	goto out;
1083 }
1084 EXPORT_SYMBOL(inet_register_protosw);
1085 
1086 void inet_unregister_protosw(struct inet_protosw *p)
1087 {
1088 	if (INET_PROTOSW_PERMANENT & p->flags) {
1089 		pr_err("Attempt to unregister permanent protocol %d\n",
1090 		       p->protocol);
1091 	} else {
1092 		spin_lock_bh(&inetsw_lock);
1093 		list_del_rcu(&p->list);
1094 		spin_unlock_bh(&inetsw_lock);
1095 
1096 		synchronize_net();
1097 	}
1098 }
1099 EXPORT_SYMBOL(inet_unregister_protosw);
1100 
1101 /*
1102  *      Shall we try to damage output packets if routing dev changes?
1103  */
1104 
1105 int sysctl_ip_dynaddr __read_mostly;
1106 
1107 static int inet_sk_reselect_saddr(struct sock *sk)
1108 {
1109 	struct inet_sock *inet = inet_sk(sk);
1110 	__be32 old_saddr = inet->inet_saddr;
1111 	__be32 daddr = inet->inet_daddr;
1112 	struct flowi4 *fl4;
1113 	struct rtable *rt;
1114 	__be32 new_saddr;
1115 	struct ip_options_rcu *inet_opt;
1116 
1117 	inet_opt = rcu_dereference_protected(inet->inet_opt,
1118 					     sock_owned_by_user(sk));
1119 	if (inet_opt && inet_opt->opt.srr)
1120 		daddr = inet_opt->opt.faddr;
1121 
1122 	/* Query new route. */
1123 	fl4 = &inet->cork.fl.u.ip4;
1124 	rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1125 			      sk->sk_bound_dev_if, sk->sk_protocol,
1126 			      inet->inet_sport, inet->inet_dport, sk);
1127 	if (IS_ERR(rt))
1128 		return PTR_ERR(rt);
1129 
1130 	sk_setup_caps(sk, &rt->dst);
1131 
1132 	new_saddr = fl4->saddr;
1133 
1134 	if (new_saddr == old_saddr)
1135 		return 0;
1136 
1137 	if (sysctl_ip_dynaddr > 1) {
1138 		pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1139 			__func__, &old_saddr, &new_saddr);
1140 	}
1141 
1142 	inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1143 
1144 	/*
1145 	 * XXX The only one ugly spot where we need to
1146 	 * XXX really change the sockets identity after
1147 	 * XXX it has entered the hashes. -DaveM
1148 	 *
1149 	 * Besides that, it does not check for connection
1150 	 * uniqueness. Wait for troubles.
1151 	 */
1152 	__sk_prot_rehash(sk);
1153 	return 0;
1154 }
1155 
1156 int inet_sk_rebuild_header(struct sock *sk)
1157 {
1158 	struct inet_sock *inet = inet_sk(sk);
1159 	struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1160 	__be32 daddr;
1161 	struct ip_options_rcu *inet_opt;
1162 	struct flowi4 *fl4;
1163 	int err;
1164 
1165 	/* Route is OK, nothing to do. */
1166 	if (rt)
1167 		return 0;
1168 
1169 	/* Reroute. */
1170 	rcu_read_lock();
1171 	inet_opt = rcu_dereference(inet->inet_opt);
1172 	daddr = inet->inet_daddr;
1173 	if (inet_opt && inet_opt->opt.srr)
1174 		daddr = inet_opt->opt.faddr;
1175 	rcu_read_unlock();
1176 	fl4 = &inet->cork.fl.u.ip4;
1177 	rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1178 				   inet->inet_dport, inet->inet_sport,
1179 				   sk->sk_protocol, RT_CONN_FLAGS(sk),
1180 				   sk->sk_bound_dev_if);
1181 	if (!IS_ERR(rt)) {
1182 		err = 0;
1183 		sk_setup_caps(sk, &rt->dst);
1184 	} else {
1185 		err = PTR_ERR(rt);
1186 
1187 		/* Routing failed... */
1188 		sk->sk_route_caps = 0;
1189 		/*
1190 		 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1191 		 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1192 		 */
1193 		if (!sysctl_ip_dynaddr ||
1194 		    sk->sk_state != TCP_SYN_SENT ||
1195 		    (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1196 		    (err = inet_sk_reselect_saddr(sk)) != 0)
1197 			sk->sk_err_soft = -err;
1198 	}
1199 
1200 	return err;
1201 }
1202 EXPORT_SYMBOL(inet_sk_rebuild_header);
1203 
1204 static struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1205 					netdev_features_t features)
1206 {
1207 	struct sk_buff *segs = ERR_PTR(-EINVAL);
1208 	const struct net_offload *ops;
1209 	unsigned int offset = 0;
1210 	bool udpfrag, encap;
1211 	struct iphdr *iph;
1212 	int proto;
1213 	int nhoff;
1214 	int ihl;
1215 	int id;
1216 
1217 	if (unlikely(skb_shinfo(skb)->gso_type &
1218 		     ~(SKB_GSO_TCPV4 |
1219 		       SKB_GSO_UDP |
1220 		       SKB_GSO_DODGY |
1221 		       SKB_GSO_TCP_ECN |
1222 		       SKB_GSO_GRE |
1223 		       SKB_GSO_GRE_CSUM |
1224 		       SKB_GSO_IPIP |
1225 		       SKB_GSO_SIT |
1226 		       SKB_GSO_TCPV6 |
1227 		       SKB_GSO_UDP_TUNNEL |
1228 		       SKB_GSO_UDP_TUNNEL_CSUM |
1229 		       SKB_GSO_TUNNEL_REMCSUM |
1230 		       0)))
1231 		goto out;
1232 
1233 	skb_reset_network_header(skb);
1234 	nhoff = skb_network_header(skb) - skb_mac_header(skb);
1235 	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1236 		goto out;
1237 
1238 	iph = ip_hdr(skb);
1239 	ihl = iph->ihl * 4;
1240 	if (ihl < sizeof(*iph))
1241 		goto out;
1242 
1243 	id = ntohs(iph->id);
1244 	proto = iph->protocol;
1245 
1246 	/* Warning: after this point, iph might be no longer valid */
1247 	if (unlikely(!pskb_may_pull(skb, ihl)))
1248 		goto out;
1249 	__skb_pull(skb, ihl);
1250 
1251 	encap = SKB_GSO_CB(skb)->encap_level > 0;
1252 	if (encap)
1253 		features &= skb->dev->hw_enc_features;
1254 	SKB_GSO_CB(skb)->encap_level += ihl;
1255 
1256 	skb_reset_transport_header(skb);
1257 
1258 	segs = ERR_PTR(-EPROTONOSUPPORT);
1259 
1260 	if (skb->encapsulation &&
1261 	    skb_shinfo(skb)->gso_type & (SKB_GSO_SIT|SKB_GSO_IPIP))
1262 		udpfrag = proto == IPPROTO_UDP && encap;
1263 	else
1264 		udpfrag = proto == IPPROTO_UDP && !skb->encapsulation;
1265 
1266 	ops = rcu_dereference(inet_offloads[proto]);
1267 	if (likely(ops && ops->callbacks.gso_segment))
1268 		segs = ops->callbacks.gso_segment(skb, features);
1269 
1270 	if (IS_ERR_OR_NULL(segs))
1271 		goto out;
1272 
1273 	skb = segs;
1274 	do {
1275 		iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1276 		if (udpfrag) {
1277 			iph->id = htons(id);
1278 			iph->frag_off = htons(offset >> 3);
1279 			if (skb->next)
1280 				iph->frag_off |= htons(IP_MF);
1281 			offset += skb->len - nhoff - ihl;
1282 		} else {
1283 			iph->id = htons(id++);
1284 		}
1285 		iph->tot_len = htons(skb->len - nhoff);
1286 		ip_send_check(iph);
1287 		if (encap)
1288 			skb_reset_inner_headers(skb);
1289 		skb->network_header = (u8 *)iph - skb->head;
1290 	} while ((skb = skb->next));
1291 
1292 out:
1293 	return segs;
1294 }
1295 
1296 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1297 					 struct sk_buff *skb)
1298 {
1299 	const struct net_offload *ops;
1300 	struct sk_buff **pp = NULL;
1301 	struct sk_buff *p;
1302 	const struct iphdr *iph;
1303 	unsigned int hlen;
1304 	unsigned int off;
1305 	unsigned int id;
1306 	int flush = 1;
1307 	int proto;
1308 
1309 	off = skb_gro_offset(skb);
1310 	hlen = off + sizeof(*iph);
1311 	iph = skb_gro_header_fast(skb, off);
1312 	if (skb_gro_header_hard(skb, hlen)) {
1313 		iph = skb_gro_header_slow(skb, hlen, off);
1314 		if (unlikely(!iph))
1315 			goto out;
1316 	}
1317 
1318 	proto = iph->protocol;
1319 
1320 	rcu_read_lock();
1321 	ops = rcu_dereference(inet_offloads[proto]);
1322 	if (!ops || !ops->callbacks.gro_receive)
1323 		goto out_unlock;
1324 
1325 	if (*(u8 *)iph != 0x45)
1326 		goto out_unlock;
1327 
1328 	if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1329 		goto out_unlock;
1330 
1331 	id = ntohl(*(__be32 *)&iph->id);
1332 	flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1333 	id >>= 16;
1334 
1335 	for (p = *head; p; p = p->next) {
1336 		struct iphdr *iph2;
1337 
1338 		if (!NAPI_GRO_CB(p)->same_flow)
1339 			continue;
1340 
1341 		iph2 = (struct iphdr *)(p->data + off);
1342 		/* The above works because, with the exception of the top
1343 		 * (inner most) layer, we only aggregate pkts with the same
1344 		 * hdr length so all the hdrs we'll need to verify will start
1345 		 * at the same offset.
1346 		 */
1347 		if ((iph->protocol ^ iph2->protocol) |
1348 		    ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1349 		    ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1350 			NAPI_GRO_CB(p)->same_flow = 0;
1351 			continue;
1352 		}
1353 
1354 		/* All fields must match except length and checksum. */
1355 		NAPI_GRO_CB(p)->flush |=
1356 			(iph->ttl ^ iph2->ttl) |
1357 			(iph->tos ^ iph2->tos) |
1358 			((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1359 
1360 		/* Save the IP ID check to be included later when we get to
1361 		 * the transport layer so only the inner most IP ID is checked.
1362 		 * This is because some GSO/TSO implementations do not
1363 		 * correctly increment the IP ID for the outer hdrs.
1364 		 */
1365 		NAPI_GRO_CB(p)->flush_id =
1366 			    ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1367 		NAPI_GRO_CB(p)->flush |= flush;
1368 	}
1369 
1370 	NAPI_GRO_CB(skb)->flush |= flush;
1371 	skb_set_network_header(skb, off);
1372 	/* The above will be needed by the transport layer if there is one
1373 	 * immediately following this IP hdr.
1374 	 */
1375 
1376 	/* Note : No need to call skb_gro_postpull_rcsum() here,
1377 	 * as we already checked checksum over ipv4 header was 0
1378 	 */
1379 	skb_gro_pull(skb, sizeof(*iph));
1380 	skb_set_transport_header(skb, skb_gro_offset(skb));
1381 
1382 	pp = ops->callbacks.gro_receive(head, skb);
1383 
1384 out_unlock:
1385 	rcu_read_unlock();
1386 
1387 out:
1388 	NAPI_GRO_CB(skb)->flush |= flush;
1389 
1390 	return pp;
1391 }
1392 
1393 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1394 {
1395 	if (sk->sk_family == AF_INET)
1396 		return ip_recv_error(sk, msg, len, addr_len);
1397 #if IS_ENABLED(CONFIG_IPV6)
1398 	if (sk->sk_family == AF_INET6)
1399 		return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1400 #endif
1401 	return -EINVAL;
1402 }
1403 
1404 static int inet_gro_complete(struct sk_buff *skb, int nhoff)
1405 {
1406 	__be16 newlen = htons(skb->len - nhoff);
1407 	struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1408 	const struct net_offload *ops;
1409 	int proto = iph->protocol;
1410 	int err = -ENOSYS;
1411 
1412 	if (skb->encapsulation)
1413 		skb_set_inner_network_header(skb, nhoff);
1414 
1415 	csum_replace2(&iph->check, iph->tot_len, newlen);
1416 	iph->tot_len = newlen;
1417 
1418 	rcu_read_lock();
1419 	ops = rcu_dereference(inet_offloads[proto]);
1420 	if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1421 		goto out_unlock;
1422 
1423 	/* Only need to add sizeof(*iph) to get to the next hdr below
1424 	 * because any hdr with option will have been flushed in
1425 	 * inet_gro_receive().
1426 	 */
1427 	err = ops->callbacks.gro_complete(skb, nhoff + sizeof(*iph));
1428 
1429 out_unlock:
1430 	rcu_read_unlock();
1431 
1432 	return err;
1433 }
1434 
1435 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1436 			 unsigned short type, unsigned char protocol,
1437 			 struct net *net)
1438 {
1439 	struct socket *sock;
1440 	int rc = sock_create_kern(net, family, type, protocol, &sock);
1441 
1442 	if (rc == 0) {
1443 		*sk = sock->sk;
1444 		(*sk)->sk_allocation = GFP_ATOMIC;
1445 		/*
1446 		 * Unhash it so that IP input processing does not even see it,
1447 		 * we do not wish this socket to see incoming packets.
1448 		 */
1449 		(*sk)->sk_prot->unhash(*sk);
1450 	}
1451 	return rc;
1452 }
1453 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1454 
1455 u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt)
1456 {
1457 	return  *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt);
1458 }
1459 EXPORT_SYMBOL_GPL(snmp_get_cpu_field);
1460 
1461 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1462 {
1463 	unsigned long res = 0;
1464 	int i;
1465 
1466 	for_each_possible_cpu(i)
1467 		res += snmp_get_cpu_field(mib, i, offt);
1468 	return res;
1469 }
1470 EXPORT_SYMBOL_GPL(snmp_fold_field);
1471 
1472 #if BITS_PER_LONG==32
1473 
1474 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1475 			 size_t syncp_offset)
1476 {
1477 	void *bhptr;
1478 	struct u64_stats_sync *syncp;
1479 	u64 v;
1480 	unsigned int start;
1481 
1482 	bhptr = per_cpu_ptr(mib, cpu);
1483 	syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1484 	do {
1485 		start = u64_stats_fetch_begin_irq(syncp);
1486 		v = *(((u64 *)bhptr) + offt);
1487 	} while (u64_stats_fetch_retry_irq(syncp, start));
1488 
1489 	return v;
1490 }
1491 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1492 
1493 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1494 {
1495 	u64 res = 0;
1496 	int cpu;
1497 
1498 	for_each_possible_cpu(cpu) {
1499 		res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1500 	}
1501 	return res;
1502 }
1503 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1504 #endif
1505 
1506 #ifdef CONFIG_IP_MULTICAST
1507 static const struct net_protocol igmp_protocol = {
1508 	.handler =	igmp_rcv,
1509 	.netns_ok =	1,
1510 };
1511 #endif
1512 
1513 static const struct net_protocol tcp_protocol = {
1514 	.early_demux	=	tcp_v4_early_demux,
1515 	.handler	=	tcp_v4_rcv,
1516 	.err_handler	=	tcp_v4_err,
1517 	.no_policy	=	1,
1518 	.netns_ok	=	1,
1519 	.icmp_strict_tag_validation = 1,
1520 };
1521 
1522 static const struct net_protocol udp_protocol = {
1523 	.early_demux =	udp_v4_early_demux,
1524 	.handler =	udp_rcv,
1525 	.err_handler =	udp_err,
1526 	.no_policy =	1,
1527 	.netns_ok =	1,
1528 };
1529 
1530 static const struct net_protocol icmp_protocol = {
1531 	.handler =	icmp_rcv,
1532 	.err_handler =	icmp_err,
1533 	.no_policy =	1,
1534 	.netns_ok =	1,
1535 };
1536 
1537 static __net_init int ipv4_mib_init_net(struct net *net)
1538 {
1539 	int i;
1540 
1541 	net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1542 	if (!net->mib.tcp_statistics)
1543 		goto err_tcp_mib;
1544 	net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1545 	if (!net->mib.ip_statistics)
1546 		goto err_ip_mib;
1547 
1548 	for_each_possible_cpu(i) {
1549 		struct ipstats_mib *af_inet_stats;
1550 		af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1551 		u64_stats_init(&af_inet_stats->syncp);
1552 	}
1553 
1554 	net->mib.net_statistics = alloc_percpu(struct linux_mib);
1555 	if (!net->mib.net_statistics)
1556 		goto err_net_mib;
1557 	net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1558 	if (!net->mib.udp_statistics)
1559 		goto err_udp_mib;
1560 	net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1561 	if (!net->mib.udplite_statistics)
1562 		goto err_udplite_mib;
1563 	net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1564 	if (!net->mib.icmp_statistics)
1565 		goto err_icmp_mib;
1566 	net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1567 					      GFP_KERNEL);
1568 	if (!net->mib.icmpmsg_statistics)
1569 		goto err_icmpmsg_mib;
1570 
1571 	tcp_mib_init(net);
1572 	return 0;
1573 
1574 err_icmpmsg_mib:
1575 	free_percpu(net->mib.icmp_statistics);
1576 err_icmp_mib:
1577 	free_percpu(net->mib.udplite_statistics);
1578 err_udplite_mib:
1579 	free_percpu(net->mib.udp_statistics);
1580 err_udp_mib:
1581 	free_percpu(net->mib.net_statistics);
1582 err_net_mib:
1583 	free_percpu(net->mib.ip_statistics);
1584 err_ip_mib:
1585 	free_percpu(net->mib.tcp_statistics);
1586 err_tcp_mib:
1587 	return -ENOMEM;
1588 }
1589 
1590 static __net_exit void ipv4_mib_exit_net(struct net *net)
1591 {
1592 	kfree(net->mib.icmpmsg_statistics);
1593 	free_percpu(net->mib.icmp_statistics);
1594 	free_percpu(net->mib.udplite_statistics);
1595 	free_percpu(net->mib.udp_statistics);
1596 	free_percpu(net->mib.net_statistics);
1597 	free_percpu(net->mib.ip_statistics);
1598 	free_percpu(net->mib.tcp_statistics);
1599 }
1600 
1601 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1602 	.init = ipv4_mib_init_net,
1603 	.exit = ipv4_mib_exit_net,
1604 };
1605 
1606 static int __init init_ipv4_mibs(void)
1607 {
1608 	return register_pernet_subsys(&ipv4_mib_ops);
1609 }
1610 
1611 static __net_init int inet_init_net(struct net *net)
1612 {
1613 	/*
1614 	 * Set defaults for local port range
1615 	 */
1616 	seqlock_init(&net->ipv4.ip_local_ports.lock);
1617 	net->ipv4.ip_local_ports.range[0] =  32768;
1618 	net->ipv4.ip_local_ports.range[1] =  60999;
1619 
1620 	seqlock_init(&net->ipv4.ping_group_range.lock);
1621 	/*
1622 	 * Sane defaults - nobody may create ping sockets.
1623 	 * Boot scripts should set this to distro-specific group.
1624 	 */
1625 	net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1626 	net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1627 	return 0;
1628 }
1629 
1630 static __net_exit void inet_exit_net(struct net *net)
1631 {
1632 }
1633 
1634 static __net_initdata struct pernet_operations af_inet_ops = {
1635 	.init = inet_init_net,
1636 	.exit = inet_exit_net,
1637 };
1638 
1639 static int __init init_inet_pernet_ops(void)
1640 {
1641 	return register_pernet_subsys(&af_inet_ops);
1642 }
1643 
1644 static int ipv4_proc_init(void);
1645 
1646 /*
1647  *	IP protocol layer initialiser
1648  */
1649 
1650 static struct packet_offload ip_packet_offload __read_mostly = {
1651 	.type = cpu_to_be16(ETH_P_IP),
1652 	.callbacks = {
1653 		.gso_segment = inet_gso_segment,
1654 		.gro_receive = inet_gro_receive,
1655 		.gro_complete = inet_gro_complete,
1656 	},
1657 };
1658 
1659 static const struct net_offload ipip_offload = {
1660 	.callbacks = {
1661 		.gso_segment	= inet_gso_segment,
1662 		.gro_receive	= inet_gro_receive,
1663 		.gro_complete	= inet_gro_complete,
1664 	},
1665 };
1666 
1667 static int __init ipv4_offload_init(void)
1668 {
1669 	/*
1670 	 * Add offloads
1671 	 */
1672 	if (udpv4_offload_init() < 0)
1673 		pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1674 	if (tcpv4_offload_init() < 0)
1675 		pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1676 
1677 	dev_add_offload(&ip_packet_offload);
1678 	inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1679 	return 0;
1680 }
1681 
1682 fs_initcall(ipv4_offload_init);
1683 
1684 static struct packet_type ip_packet_type __read_mostly = {
1685 	.type = cpu_to_be16(ETH_P_IP),
1686 	.func = ip_rcv,
1687 };
1688 
1689 static int __init inet_init(void)
1690 {
1691 	struct inet_protosw *q;
1692 	struct list_head *r;
1693 	int rc = -EINVAL;
1694 
1695 	sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1696 
1697 	rc = proto_register(&tcp_prot, 1);
1698 	if (rc)
1699 		goto out;
1700 
1701 	rc = proto_register(&udp_prot, 1);
1702 	if (rc)
1703 		goto out_unregister_tcp_proto;
1704 
1705 	rc = proto_register(&raw_prot, 1);
1706 	if (rc)
1707 		goto out_unregister_udp_proto;
1708 
1709 	rc = proto_register(&ping_prot, 1);
1710 	if (rc)
1711 		goto out_unregister_raw_proto;
1712 
1713 	/*
1714 	 *	Tell SOCKET that we are alive...
1715 	 */
1716 
1717 	(void)sock_register(&inet_family_ops);
1718 
1719 #ifdef CONFIG_SYSCTL
1720 	ip_static_sysctl_init();
1721 #endif
1722 
1723 	/*
1724 	 *	Add all the base protocols.
1725 	 */
1726 
1727 	if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1728 		pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1729 	if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1730 		pr_crit("%s: Cannot add UDP protocol\n", __func__);
1731 	if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1732 		pr_crit("%s: Cannot add TCP protocol\n", __func__);
1733 #ifdef CONFIG_IP_MULTICAST
1734 	if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1735 		pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1736 #endif
1737 
1738 	/* Register the socket-side information for inet_create. */
1739 	for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1740 		INIT_LIST_HEAD(r);
1741 
1742 	for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1743 		inet_register_protosw(q);
1744 
1745 	/*
1746 	 *	Set the ARP module up
1747 	 */
1748 
1749 	arp_init();
1750 
1751 	/*
1752 	 *	Set the IP module up
1753 	 */
1754 
1755 	ip_init();
1756 
1757 	tcp_v4_init();
1758 
1759 	/* Setup TCP slab cache for open requests. */
1760 	tcp_init();
1761 
1762 	/* Setup UDP memory threshold */
1763 	udp_init();
1764 
1765 	/* Add UDP-Lite (RFC 3828) */
1766 	udplite4_register();
1767 
1768 	ping_init();
1769 
1770 	/*
1771 	 *	Set the ICMP layer up
1772 	 */
1773 
1774 	if (icmp_init() < 0)
1775 		panic("Failed to create the ICMP control socket.\n");
1776 
1777 	/*
1778 	 *	Initialise the multicast router
1779 	 */
1780 #if defined(CONFIG_IP_MROUTE)
1781 	if (ip_mr_init())
1782 		pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1783 #endif
1784 
1785 	if (init_inet_pernet_ops())
1786 		pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
1787 	/*
1788 	 *	Initialise per-cpu ipv4 mibs
1789 	 */
1790 
1791 	if (init_ipv4_mibs())
1792 		pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1793 
1794 	ipv4_proc_init();
1795 
1796 	ipfrag_init();
1797 
1798 	dev_add_pack(&ip_packet_type);
1799 
1800 	ip_tunnel_core_init();
1801 
1802 	rc = 0;
1803 out:
1804 	return rc;
1805 out_unregister_raw_proto:
1806 	proto_unregister(&raw_prot);
1807 out_unregister_udp_proto:
1808 	proto_unregister(&udp_prot);
1809 out_unregister_tcp_proto:
1810 	proto_unregister(&tcp_prot);
1811 	goto out;
1812 }
1813 
1814 fs_initcall(inet_init);
1815 
1816 /* ------------------------------------------------------------------------ */
1817 
1818 #ifdef CONFIG_PROC_FS
1819 static int __init ipv4_proc_init(void)
1820 {
1821 	int rc = 0;
1822 
1823 	if (raw_proc_init())
1824 		goto out_raw;
1825 	if (tcp4_proc_init())
1826 		goto out_tcp;
1827 	if (udp4_proc_init())
1828 		goto out_udp;
1829 	if (ping_proc_init())
1830 		goto out_ping;
1831 	if (ip_misc_proc_init())
1832 		goto out_misc;
1833 out:
1834 	return rc;
1835 out_misc:
1836 	ping_proc_exit();
1837 out_ping:
1838 	udp4_proc_exit();
1839 out_udp:
1840 	tcp4_proc_exit();
1841 out_tcp:
1842 	raw_proc_exit();
1843 out_raw:
1844 	rc = -ENOMEM;
1845 	goto out;
1846 }
1847 
1848 #else /* CONFIG_PROC_FS */
1849 static int __init ipv4_proc_init(void)
1850 {
1851 	return 0;
1852 }
1853 #endif /* CONFIG_PROC_FS */
1854 
1855 MODULE_ALIAS_NETPROTO(PF_INET);
1856 
1857