xref: /openbmc/linux/net/ipv4/af_inet.c (revision 8730046c)
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/kmod.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 <linux/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 		/* Enable TFO w/o requiring TCP_FASTOPEN socket option.
215 		 * Note that only TCP sockets (SOCK_STREAM) will reach here.
216 		 * Also fastopen backlog may already been set via the option
217 		 * because the socket was in TCP_LISTEN state previously but
218 		 * was shutdown() rather than close().
219 		 */
220 		if ((sysctl_tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
221 		    (sysctl_tcp_fastopen & TFO_SERVER_ENABLE) &&
222 		    !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
223 			fastopen_queue_tune(sk, backlog);
224 			tcp_fastopen_init_key_once(true);
225 		}
226 
227 		err = inet_csk_listen_start(sk, backlog);
228 		if (err)
229 			goto out;
230 	}
231 	sk->sk_max_ack_backlog = backlog;
232 	err = 0;
233 
234 out:
235 	release_sock(sk);
236 	return err;
237 }
238 EXPORT_SYMBOL(inet_listen);
239 
240 /*
241  *	Create an inet socket.
242  */
243 
244 static int inet_create(struct net *net, struct socket *sock, int protocol,
245 		       int kern)
246 {
247 	struct sock *sk;
248 	struct inet_protosw *answer;
249 	struct inet_sock *inet;
250 	struct proto *answer_prot;
251 	unsigned char answer_flags;
252 	int try_loading_module = 0;
253 	int err;
254 
255 	if (protocol < 0 || protocol >= IPPROTO_MAX)
256 		return -EINVAL;
257 
258 	sock->state = SS_UNCONNECTED;
259 
260 	/* Look for the requested type/protocol pair. */
261 lookup_protocol:
262 	err = -ESOCKTNOSUPPORT;
263 	rcu_read_lock();
264 	list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
265 
266 		err = 0;
267 		/* Check the non-wild match. */
268 		if (protocol == answer->protocol) {
269 			if (protocol != IPPROTO_IP)
270 				break;
271 		} else {
272 			/* Check for the two wild cases. */
273 			if (IPPROTO_IP == protocol) {
274 				protocol = answer->protocol;
275 				break;
276 			}
277 			if (IPPROTO_IP == answer->protocol)
278 				break;
279 		}
280 		err = -EPROTONOSUPPORT;
281 	}
282 
283 	if (unlikely(err)) {
284 		if (try_loading_module < 2) {
285 			rcu_read_unlock();
286 			/*
287 			 * Be more specific, e.g. net-pf-2-proto-132-type-1
288 			 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
289 			 */
290 			if (++try_loading_module == 1)
291 				request_module("net-pf-%d-proto-%d-type-%d",
292 					       PF_INET, protocol, sock->type);
293 			/*
294 			 * Fall back to generic, e.g. net-pf-2-proto-132
295 			 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
296 			 */
297 			else
298 				request_module("net-pf-%d-proto-%d",
299 					       PF_INET, protocol);
300 			goto lookup_protocol;
301 		} else
302 			goto out_rcu_unlock;
303 	}
304 
305 	err = -EPERM;
306 	if (sock->type == SOCK_RAW && !kern &&
307 	    !ns_capable(net->user_ns, CAP_NET_RAW))
308 		goto out_rcu_unlock;
309 
310 	sock->ops = answer->ops;
311 	answer_prot = answer->prot;
312 	answer_flags = answer->flags;
313 	rcu_read_unlock();
314 
315 	WARN_ON(!answer_prot->slab);
316 
317 	err = -ENOBUFS;
318 	sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
319 	if (!sk)
320 		goto out;
321 
322 	err = 0;
323 	if (INET_PROTOSW_REUSE & answer_flags)
324 		sk->sk_reuse = SK_CAN_REUSE;
325 
326 	inet = inet_sk(sk);
327 	inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
328 
329 	inet->nodefrag = 0;
330 
331 	if (SOCK_RAW == sock->type) {
332 		inet->inet_num = protocol;
333 		if (IPPROTO_RAW == protocol)
334 			inet->hdrincl = 1;
335 	}
336 
337 	if (net->ipv4.sysctl_ip_no_pmtu_disc)
338 		inet->pmtudisc = IP_PMTUDISC_DONT;
339 	else
340 		inet->pmtudisc = IP_PMTUDISC_WANT;
341 
342 	inet->inet_id = 0;
343 
344 	sock_init_data(sock, sk);
345 
346 	sk->sk_destruct	   = inet_sock_destruct;
347 	sk->sk_protocol	   = protocol;
348 	sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
349 
350 	inet->uc_ttl	= -1;
351 	inet->mc_loop	= 1;
352 	inet->mc_ttl	= 1;
353 	inet->mc_all	= 1;
354 	inet->mc_index	= 0;
355 	inet->mc_list	= NULL;
356 	inet->rcv_tos	= 0;
357 
358 	sk_refcnt_debug_inc(sk);
359 
360 	if (inet->inet_num) {
361 		/* It assumes that any protocol which allows
362 		 * the user to assign a number at socket
363 		 * creation time automatically
364 		 * shares.
365 		 */
366 		inet->inet_sport = htons(inet->inet_num);
367 		/* Add to protocol hash chains. */
368 		err = sk->sk_prot->hash(sk);
369 		if (err) {
370 			sk_common_release(sk);
371 			goto out;
372 		}
373 	}
374 
375 	if (sk->sk_prot->init) {
376 		err = sk->sk_prot->init(sk);
377 		if (err) {
378 			sk_common_release(sk);
379 			goto out;
380 		}
381 	}
382 
383 	if (!kern) {
384 		err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
385 		if (err) {
386 			sk_common_release(sk);
387 			goto out;
388 		}
389 	}
390 out:
391 	return err;
392 out_rcu_unlock:
393 	rcu_read_unlock();
394 	goto out;
395 }
396 
397 
398 /*
399  *	The peer socket should always be NULL (or else). When we call this
400  *	function we are destroying the object and from then on nobody
401  *	should refer to it.
402  */
403 int inet_release(struct socket *sock)
404 {
405 	struct sock *sk = sock->sk;
406 
407 	if (sk) {
408 		long timeout;
409 
410 		/* Applications forget to leave groups before exiting */
411 		ip_mc_drop_socket(sk);
412 
413 		/* If linger is set, we don't return until the close
414 		 * is complete.  Otherwise we return immediately. The
415 		 * actually closing is done the same either way.
416 		 *
417 		 * If the close is due to the process exiting, we never
418 		 * linger..
419 		 */
420 		timeout = 0;
421 		if (sock_flag(sk, SOCK_LINGER) &&
422 		    !(current->flags & PF_EXITING))
423 			timeout = sk->sk_lingertime;
424 		sock->sk = NULL;
425 		sk->sk_prot->close(sk, timeout);
426 	}
427 	return 0;
428 }
429 EXPORT_SYMBOL(inet_release);
430 
431 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
432 {
433 	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
434 	struct sock *sk = sock->sk;
435 	struct inet_sock *inet = inet_sk(sk);
436 	struct net *net = sock_net(sk);
437 	unsigned short snum;
438 	int chk_addr_ret;
439 	u32 tb_id = RT_TABLE_LOCAL;
440 	int err;
441 
442 	/* If the socket has its own bind function then use it. (RAW) */
443 	if (sk->sk_prot->bind) {
444 		err = sk->sk_prot->bind(sk, uaddr, addr_len);
445 		goto out;
446 	}
447 	err = -EINVAL;
448 	if (addr_len < sizeof(struct sockaddr_in))
449 		goto out;
450 
451 	if (addr->sin_family != AF_INET) {
452 		/* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
453 		 * only if s_addr is INADDR_ANY.
454 		 */
455 		err = -EAFNOSUPPORT;
456 		if (addr->sin_family != AF_UNSPEC ||
457 		    addr->sin_addr.s_addr != htonl(INADDR_ANY))
458 			goto out;
459 	}
460 
461 	tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
462 	chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
463 
464 	/* Not specified by any standard per-se, however it breaks too
465 	 * many applications when removed.  It is unfortunate since
466 	 * allowing applications to make a non-local bind solves
467 	 * several problems with systems using dynamic addressing.
468 	 * (ie. your servers still start up even if your ISDN link
469 	 *  is temporarily down)
470 	 */
471 	err = -EADDRNOTAVAIL;
472 	if (!net->ipv4.sysctl_ip_nonlocal_bind &&
473 	    !(inet->freebind || inet->transparent) &&
474 	    addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
475 	    chk_addr_ret != RTN_LOCAL &&
476 	    chk_addr_ret != RTN_MULTICAST &&
477 	    chk_addr_ret != RTN_BROADCAST)
478 		goto out;
479 
480 	snum = ntohs(addr->sin_port);
481 	err = -EACCES;
482 	if (snum && snum < PROT_SOCK &&
483 	    !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
484 		goto out;
485 
486 	/*      We keep a pair of addresses. rcv_saddr is the one
487 	 *      used by hash lookups, and saddr is used for transmit.
488 	 *
489 	 *      In the BSD API these are the same except where it
490 	 *      would be illegal to use them (multicast/broadcast) in
491 	 *      which case the sending device address is used.
492 	 */
493 	lock_sock(sk);
494 
495 	/* Check these errors (active socket, double bind). */
496 	err = -EINVAL;
497 	if (sk->sk_state != TCP_CLOSE || inet->inet_num)
498 		goto out_release_sock;
499 
500 	inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
501 	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
502 		inet->inet_saddr = 0;  /* Use device */
503 
504 	/* Make sure we are allowed to bind here. */
505 	if ((snum || !inet->bind_address_no_port) &&
506 	    sk->sk_prot->get_port(sk, snum)) {
507 		inet->inet_saddr = inet->inet_rcv_saddr = 0;
508 		err = -EADDRINUSE;
509 		goto out_release_sock;
510 	}
511 
512 	if (inet->inet_rcv_saddr)
513 		sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
514 	if (snum)
515 		sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
516 	inet->inet_sport = htons(inet->inet_num);
517 	inet->inet_daddr = 0;
518 	inet->inet_dport = 0;
519 	sk_dst_reset(sk);
520 	err = 0;
521 out_release_sock:
522 	release_sock(sk);
523 out:
524 	return err;
525 }
526 EXPORT_SYMBOL(inet_bind);
527 
528 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
529 		       int addr_len, int flags)
530 {
531 	struct sock *sk = sock->sk;
532 
533 	if (addr_len < sizeof(uaddr->sa_family))
534 		return -EINVAL;
535 	if (uaddr->sa_family == AF_UNSPEC)
536 		return sk->sk_prot->disconnect(sk, flags);
537 
538 	if (!inet_sk(sk)->inet_num && inet_autobind(sk))
539 		return -EAGAIN;
540 	return sk->sk_prot->connect(sk, uaddr, addr_len);
541 }
542 EXPORT_SYMBOL(inet_dgram_connect);
543 
544 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
545 {
546 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
547 
548 	add_wait_queue(sk_sleep(sk), &wait);
549 	sk->sk_write_pending += writebias;
550 
551 	/* Basic assumption: if someone sets sk->sk_err, he _must_
552 	 * change state of the socket from TCP_SYN_*.
553 	 * Connect() does not allow to get error notifications
554 	 * without closing the socket.
555 	 */
556 	while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
557 		release_sock(sk);
558 		timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
559 		lock_sock(sk);
560 		if (signal_pending(current) || !timeo)
561 			break;
562 	}
563 	remove_wait_queue(sk_sleep(sk), &wait);
564 	sk->sk_write_pending -= writebias;
565 	return timeo;
566 }
567 
568 /*
569  *	Connect to a remote host. There is regrettably still a little
570  *	TCP 'magic' in here.
571  */
572 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
573 			  int addr_len, int flags)
574 {
575 	struct sock *sk = sock->sk;
576 	int err;
577 	long timeo;
578 
579 	if (addr_len < sizeof(uaddr->sa_family))
580 		return -EINVAL;
581 
582 	if (uaddr->sa_family == AF_UNSPEC) {
583 		err = sk->sk_prot->disconnect(sk, flags);
584 		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
585 		goto out;
586 	}
587 
588 	switch (sock->state) {
589 	default:
590 		err = -EINVAL;
591 		goto out;
592 	case SS_CONNECTED:
593 		err = -EISCONN;
594 		goto out;
595 	case SS_CONNECTING:
596 		err = -EALREADY;
597 		/* Fall out of switch with err, set for this state */
598 		break;
599 	case SS_UNCONNECTED:
600 		err = -EISCONN;
601 		if (sk->sk_state != TCP_CLOSE)
602 			goto out;
603 
604 		err = sk->sk_prot->connect(sk, uaddr, addr_len);
605 		if (err < 0)
606 			goto out;
607 
608 		sock->state = SS_CONNECTING;
609 
610 		/* Just entered SS_CONNECTING state; the only
611 		 * difference is that return value in non-blocking
612 		 * case is EINPROGRESS, rather than EALREADY.
613 		 */
614 		err = -EINPROGRESS;
615 		break;
616 	}
617 
618 	timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
619 
620 	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
621 		int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
622 				tcp_sk(sk)->fastopen_req &&
623 				tcp_sk(sk)->fastopen_req->data ? 1 : 0;
624 
625 		/* Error code is set above */
626 		if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
627 			goto out;
628 
629 		err = sock_intr_errno(timeo);
630 		if (signal_pending(current))
631 			goto out;
632 	}
633 
634 	/* Connection was closed by RST, timeout, ICMP error
635 	 * or another process disconnected us.
636 	 */
637 	if (sk->sk_state == TCP_CLOSE)
638 		goto sock_error;
639 
640 	/* sk->sk_err may be not zero now, if RECVERR was ordered by user
641 	 * and error was received after socket entered established state.
642 	 * Hence, it is handled normally after connect() return successfully.
643 	 */
644 
645 	sock->state = SS_CONNECTED;
646 	err = 0;
647 out:
648 	return err;
649 
650 sock_error:
651 	err = sock_error(sk) ? : -ECONNABORTED;
652 	sock->state = SS_UNCONNECTED;
653 	if (sk->sk_prot->disconnect(sk, flags))
654 		sock->state = SS_DISCONNECTING;
655 	goto out;
656 }
657 EXPORT_SYMBOL(__inet_stream_connect);
658 
659 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
660 			int addr_len, int flags)
661 {
662 	int err;
663 
664 	lock_sock(sock->sk);
665 	err = __inet_stream_connect(sock, uaddr, addr_len, flags);
666 	release_sock(sock->sk);
667 	return err;
668 }
669 EXPORT_SYMBOL(inet_stream_connect);
670 
671 /*
672  *	Accept a pending connection. The TCP layer now gives BSD semantics.
673  */
674 
675 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
676 {
677 	struct sock *sk1 = sock->sk;
678 	int err = -EINVAL;
679 	struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
680 
681 	if (!sk2)
682 		goto do_err;
683 
684 	lock_sock(sk2);
685 
686 	sock_rps_record_flow(sk2);
687 	WARN_ON(!((1 << sk2->sk_state) &
688 		  (TCPF_ESTABLISHED | TCPF_SYN_RECV |
689 		  TCPF_CLOSE_WAIT | TCPF_CLOSE)));
690 
691 	sock_graft(sk2, newsock);
692 
693 	newsock->state = SS_CONNECTED;
694 	err = 0;
695 	release_sock(sk2);
696 do_err:
697 	return err;
698 }
699 EXPORT_SYMBOL(inet_accept);
700 
701 
702 /*
703  *	This does both peername and sockname.
704  */
705 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
706 			int *uaddr_len, int peer)
707 {
708 	struct sock *sk		= sock->sk;
709 	struct inet_sock *inet	= inet_sk(sk);
710 	DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
711 
712 	sin->sin_family = AF_INET;
713 	if (peer) {
714 		if (!inet->inet_dport ||
715 		    (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
716 		     peer == 1))
717 			return -ENOTCONN;
718 		sin->sin_port = inet->inet_dport;
719 		sin->sin_addr.s_addr = inet->inet_daddr;
720 	} else {
721 		__be32 addr = inet->inet_rcv_saddr;
722 		if (!addr)
723 			addr = inet->inet_saddr;
724 		sin->sin_port = inet->inet_sport;
725 		sin->sin_addr.s_addr = addr;
726 	}
727 	memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
728 	*uaddr_len = sizeof(*sin);
729 	return 0;
730 }
731 EXPORT_SYMBOL(inet_getname);
732 
733 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
734 {
735 	struct sock *sk = sock->sk;
736 
737 	sock_rps_record_flow(sk);
738 
739 	/* We may need to bind the socket. */
740 	if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
741 	    inet_autobind(sk))
742 		return -EAGAIN;
743 
744 	return sk->sk_prot->sendmsg(sk, msg, size);
745 }
746 EXPORT_SYMBOL(inet_sendmsg);
747 
748 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
749 		      size_t size, int flags)
750 {
751 	struct sock *sk = sock->sk;
752 
753 	sock_rps_record_flow(sk);
754 
755 	/* We may need to bind the socket. */
756 	if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
757 	    inet_autobind(sk))
758 		return -EAGAIN;
759 
760 	if (sk->sk_prot->sendpage)
761 		return sk->sk_prot->sendpage(sk, page, offset, size, flags);
762 	return sock_no_sendpage(sock, page, offset, size, flags);
763 }
764 EXPORT_SYMBOL(inet_sendpage);
765 
766 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
767 		 int flags)
768 {
769 	struct sock *sk = sock->sk;
770 	int addr_len = 0;
771 	int err;
772 
773 	sock_rps_record_flow(sk);
774 
775 	err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
776 				   flags & ~MSG_DONTWAIT, &addr_len);
777 	if (err >= 0)
778 		msg->msg_namelen = addr_len;
779 	return err;
780 }
781 EXPORT_SYMBOL(inet_recvmsg);
782 
783 int inet_shutdown(struct socket *sock, int how)
784 {
785 	struct sock *sk = sock->sk;
786 	int err = 0;
787 
788 	/* This should really check to make sure
789 	 * the socket is a TCP socket. (WHY AC...)
790 	 */
791 	how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
792 		       1->2 bit 2 snds.
793 		       2->3 */
794 	if ((how & ~SHUTDOWN_MASK) || !how)	/* MAXINT->0 */
795 		return -EINVAL;
796 
797 	lock_sock(sk);
798 	if (sock->state == SS_CONNECTING) {
799 		if ((1 << sk->sk_state) &
800 		    (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
801 			sock->state = SS_DISCONNECTING;
802 		else
803 			sock->state = SS_CONNECTED;
804 	}
805 
806 	switch (sk->sk_state) {
807 	case TCP_CLOSE:
808 		err = -ENOTCONN;
809 		/* Hack to wake up other listeners, who can poll for
810 		   POLLHUP, even on eg. unconnected UDP sockets -- RR */
811 	default:
812 		sk->sk_shutdown |= how;
813 		if (sk->sk_prot->shutdown)
814 			sk->sk_prot->shutdown(sk, how);
815 		break;
816 
817 	/* Remaining two branches are temporary solution for missing
818 	 * close() in multithreaded environment. It is _not_ a good idea,
819 	 * but we have no choice until close() is repaired at VFS level.
820 	 */
821 	case TCP_LISTEN:
822 		if (!(how & RCV_SHUTDOWN))
823 			break;
824 		/* Fall through */
825 	case TCP_SYN_SENT:
826 		err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
827 		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
828 		break;
829 	}
830 
831 	/* Wake up anyone sleeping in poll. */
832 	sk->sk_state_change(sk);
833 	release_sock(sk);
834 	return err;
835 }
836 EXPORT_SYMBOL(inet_shutdown);
837 
838 /*
839  *	ioctl() calls you can issue on an INET socket. Most of these are
840  *	device configuration and stuff and very rarely used. Some ioctls
841  *	pass on to the socket itself.
842  *
843  *	NOTE: I like the idea of a module for the config stuff. ie ifconfig
844  *	loads the devconfigure module does its configuring and unloads it.
845  *	There's a good 20K of config code hanging around the kernel.
846  */
847 
848 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
849 {
850 	struct sock *sk = sock->sk;
851 	int err = 0;
852 	struct net *net = sock_net(sk);
853 
854 	switch (cmd) {
855 	case SIOCGSTAMP:
856 		err = sock_get_timestamp(sk, (struct timeval __user *)arg);
857 		break;
858 	case SIOCGSTAMPNS:
859 		err = sock_get_timestampns(sk, (struct timespec __user *)arg);
860 		break;
861 	case SIOCADDRT:
862 	case SIOCDELRT:
863 	case SIOCRTMSG:
864 		err = ip_rt_ioctl(net, cmd, (void __user *)arg);
865 		break;
866 	case SIOCDARP:
867 	case SIOCGARP:
868 	case SIOCSARP:
869 		err = arp_ioctl(net, cmd, (void __user *)arg);
870 		break;
871 	case SIOCGIFADDR:
872 	case SIOCSIFADDR:
873 	case SIOCGIFBRDADDR:
874 	case SIOCSIFBRDADDR:
875 	case SIOCGIFNETMASK:
876 	case SIOCSIFNETMASK:
877 	case SIOCGIFDSTADDR:
878 	case SIOCSIFDSTADDR:
879 	case SIOCSIFPFLAGS:
880 	case SIOCGIFPFLAGS:
881 	case SIOCSIFFLAGS:
882 		err = devinet_ioctl(net, cmd, (void __user *)arg);
883 		break;
884 	default:
885 		if (sk->sk_prot->ioctl)
886 			err = sk->sk_prot->ioctl(sk, cmd, arg);
887 		else
888 			err = -ENOIOCTLCMD;
889 		break;
890 	}
891 	return err;
892 }
893 EXPORT_SYMBOL(inet_ioctl);
894 
895 #ifdef CONFIG_COMPAT
896 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
897 {
898 	struct sock *sk = sock->sk;
899 	int err = -ENOIOCTLCMD;
900 
901 	if (sk->sk_prot->compat_ioctl)
902 		err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
903 
904 	return err;
905 }
906 #endif
907 
908 const struct proto_ops inet_stream_ops = {
909 	.family		   = PF_INET,
910 	.owner		   = THIS_MODULE,
911 	.release	   = inet_release,
912 	.bind		   = inet_bind,
913 	.connect	   = inet_stream_connect,
914 	.socketpair	   = sock_no_socketpair,
915 	.accept		   = inet_accept,
916 	.getname	   = inet_getname,
917 	.poll		   = tcp_poll,
918 	.ioctl		   = inet_ioctl,
919 	.listen		   = inet_listen,
920 	.shutdown	   = inet_shutdown,
921 	.setsockopt	   = sock_common_setsockopt,
922 	.getsockopt	   = sock_common_getsockopt,
923 	.sendmsg	   = inet_sendmsg,
924 	.recvmsg	   = inet_recvmsg,
925 	.mmap		   = sock_no_mmap,
926 	.sendpage	   = inet_sendpage,
927 	.splice_read	   = tcp_splice_read,
928 	.read_sock	   = tcp_read_sock,
929 	.peek_len	   = tcp_peek_len,
930 #ifdef CONFIG_COMPAT
931 	.compat_setsockopt = compat_sock_common_setsockopt,
932 	.compat_getsockopt = compat_sock_common_getsockopt,
933 	.compat_ioctl	   = inet_compat_ioctl,
934 #endif
935 };
936 EXPORT_SYMBOL(inet_stream_ops);
937 
938 const struct proto_ops inet_dgram_ops = {
939 	.family		   = PF_INET,
940 	.owner		   = THIS_MODULE,
941 	.release	   = inet_release,
942 	.bind		   = inet_bind,
943 	.connect	   = inet_dgram_connect,
944 	.socketpair	   = sock_no_socketpair,
945 	.accept		   = sock_no_accept,
946 	.getname	   = inet_getname,
947 	.poll		   = udp_poll,
948 	.ioctl		   = inet_ioctl,
949 	.listen		   = sock_no_listen,
950 	.shutdown	   = inet_shutdown,
951 	.setsockopt	   = sock_common_setsockopt,
952 	.getsockopt	   = sock_common_getsockopt,
953 	.sendmsg	   = inet_sendmsg,
954 	.recvmsg	   = inet_recvmsg,
955 	.mmap		   = sock_no_mmap,
956 	.sendpage	   = inet_sendpage,
957 	.set_peek_off	   = sk_set_peek_off,
958 #ifdef CONFIG_COMPAT
959 	.compat_setsockopt = compat_sock_common_setsockopt,
960 	.compat_getsockopt = compat_sock_common_getsockopt,
961 	.compat_ioctl	   = inet_compat_ioctl,
962 #endif
963 };
964 EXPORT_SYMBOL(inet_dgram_ops);
965 
966 /*
967  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
968  * udp_poll
969  */
970 static const struct proto_ops inet_sockraw_ops = {
971 	.family		   = PF_INET,
972 	.owner		   = THIS_MODULE,
973 	.release	   = inet_release,
974 	.bind		   = inet_bind,
975 	.connect	   = inet_dgram_connect,
976 	.socketpair	   = sock_no_socketpair,
977 	.accept		   = sock_no_accept,
978 	.getname	   = inet_getname,
979 	.poll		   = datagram_poll,
980 	.ioctl		   = inet_ioctl,
981 	.listen		   = sock_no_listen,
982 	.shutdown	   = inet_shutdown,
983 	.setsockopt	   = sock_common_setsockopt,
984 	.getsockopt	   = sock_common_getsockopt,
985 	.sendmsg	   = inet_sendmsg,
986 	.recvmsg	   = inet_recvmsg,
987 	.mmap		   = sock_no_mmap,
988 	.sendpage	   = inet_sendpage,
989 #ifdef CONFIG_COMPAT
990 	.compat_setsockopt = compat_sock_common_setsockopt,
991 	.compat_getsockopt = compat_sock_common_getsockopt,
992 	.compat_ioctl	   = inet_compat_ioctl,
993 #endif
994 };
995 
996 static const struct net_proto_family inet_family_ops = {
997 	.family = PF_INET,
998 	.create = inet_create,
999 	.owner	= THIS_MODULE,
1000 };
1001 
1002 /* Upon startup we insert all the elements in inetsw_array[] into
1003  * the linked list inetsw.
1004  */
1005 static struct inet_protosw inetsw_array[] =
1006 {
1007 	{
1008 		.type =       SOCK_STREAM,
1009 		.protocol =   IPPROTO_TCP,
1010 		.prot =       &tcp_prot,
1011 		.ops =        &inet_stream_ops,
1012 		.flags =      INET_PROTOSW_PERMANENT |
1013 			      INET_PROTOSW_ICSK,
1014 	},
1015 
1016 	{
1017 		.type =       SOCK_DGRAM,
1018 		.protocol =   IPPROTO_UDP,
1019 		.prot =       &udp_prot,
1020 		.ops =        &inet_dgram_ops,
1021 		.flags =      INET_PROTOSW_PERMANENT,
1022        },
1023 
1024        {
1025 		.type =       SOCK_DGRAM,
1026 		.protocol =   IPPROTO_ICMP,
1027 		.prot =       &ping_prot,
1028 		.ops =        &inet_dgram_ops,
1029 		.flags =      INET_PROTOSW_REUSE,
1030        },
1031 
1032        {
1033 	       .type =       SOCK_RAW,
1034 	       .protocol =   IPPROTO_IP,	/* wild card */
1035 	       .prot =       &raw_prot,
1036 	       .ops =        &inet_sockraw_ops,
1037 	       .flags =      INET_PROTOSW_REUSE,
1038        }
1039 };
1040 
1041 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1042 
1043 void inet_register_protosw(struct inet_protosw *p)
1044 {
1045 	struct list_head *lh;
1046 	struct inet_protosw *answer;
1047 	int protocol = p->protocol;
1048 	struct list_head *last_perm;
1049 
1050 	spin_lock_bh(&inetsw_lock);
1051 
1052 	if (p->type >= SOCK_MAX)
1053 		goto out_illegal;
1054 
1055 	/* If we are trying to override a permanent protocol, bail. */
1056 	last_perm = &inetsw[p->type];
1057 	list_for_each(lh, &inetsw[p->type]) {
1058 		answer = list_entry(lh, struct inet_protosw, list);
1059 		/* Check only the non-wild match. */
1060 		if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1061 			break;
1062 		if (protocol == answer->protocol)
1063 			goto out_permanent;
1064 		last_perm = lh;
1065 	}
1066 
1067 	/* Add the new entry after the last permanent entry if any, so that
1068 	 * the new entry does not override a permanent entry when matched with
1069 	 * a wild-card protocol. But it is allowed to override any existing
1070 	 * non-permanent entry.  This means that when we remove this entry, the
1071 	 * system automatically returns to the old behavior.
1072 	 */
1073 	list_add_rcu(&p->list, last_perm);
1074 out:
1075 	spin_unlock_bh(&inetsw_lock);
1076 
1077 	return;
1078 
1079 out_permanent:
1080 	pr_err("Attempt to override permanent protocol %d\n", protocol);
1081 	goto out;
1082 
1083 out_illegal:
1084 	pr_err("Ignoring attempt to register invalid socket type %d\n",
1085 	       p->type);
1086 	goto out;
1087 }
1088 EXPORT_SYMBOL(inet_register_protosw);
1089 
1090 void inet_unregister_protosw(struct inet_protosw *p)
1091 {
1092 	if (INET_PROTOSW_PERMANENT & p->flags) {
1093 		pr_err("Attempt to unregister permanent protocol %d\n",
1094 		       p->protocol);
1095 	} else {
1096 		spin_lock_bh(&inetsw_lock);
1097 		list_del_rcu(&p->list);
1098 		spin_unlock_bh(&inetsw_lock);
1099 
1100 		synchronize_net();
1101 	}
1102 }
1103 EXPORT_SYMBOL(inet_unregister_protosw);
1104 
1105 static int inet_sk_reselect_saddr(struct sock *sk)
1106 {
1107 	struct inet_sock *inet = inet_sk(sk);
1108 	__be32 old_saddr = inet->inet_saddr;
1109 	__be32 daddr = inet->inet_daddr;
1110 	struct flowi4 *fl4;
1111 	struct rtable *rt;
1112 	__be32 new_saddr;
1113 	struct ip_options_rcu *inet_opt;
1114 
1115 	inet_opt = rcu_dereference_protected(inet->inet_opt,
1116 					     lockdep_sock_is_held(sk));
1117 	if (inet_opt && inet_opt->opt.srr)
1118 		daddr = inet_opt->opt.faddr;
1119 
1120 	/* Query new route. */
1121 	fl4 = &inet->cork.fl.u.ip4;
1122 	rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1123 			      sk->sk_bound_dev_if, sk->sk_protocol,
1124 			      inet->inet_sport, inet->inet_dport, sk);
1125 	if (IS_ERR(rt))
1126 		return PTR_ERR(rt);
1127 
1128 	sk_setup_caps(sk, &rt->dst);
1129 
1130 	new_saddr = fl4->saddr;
1131 
1132 	if (new_saddr == old_saddr)
1133 		return 0;
1134 
1135 	if (sock_net(sk)->ipv4.sysctl_ip_dynaddr > 1) {
1136 		pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1137 			__func__, &old_saddr, &new_saddr);
1138 	}
1139 
1140 	inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1141 
1142 	/*
1143 	 * XXX The only one ugly spot where we need to
1144 	 * XXX really change the sockets identity after
1145 	 * XXX it has entered the hashes. -DaveM
1146 	 *
1147 	 * Besides that, it does not check for connection
1148 	 * uniqueness. Wait for troubles.
1149 	 */
1150 	return __sk_prot_rehash(sk);
1151 }
1152 
1153 int inet_sk_rebuild_header(struct sock *sk)
1154 {
1155 	struct inet_sock *inet = inet_sk(sk);
1156 	struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1157 	__be32 daddr;
1158 	struct ip_options_rcu *inet_opt;
1159 	struct flowi4 *fl4;
1160 	int err;
1161 
1162 	/* Route is OK, nothing to do. */
1163 	if (rt)
1164 		return 0;
1165 
1166 	/* Reroute. */
1167 	rcu_read_lock();
1168 	inet_opt = rcu_dereference(inet->inet_opt);
1169 	daddr = inet->inet_daddr;
1170 	if (inet_opt && inet_opt->opt.srr)
1171 		daddr = inet_opt->opt.faddr;
1172 	rcu_read_unlock();
1173 	fl4 = &inet->cork.fl.u.ip4;
1174 	rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1175 				   inet->inet_dport, inet->inet_sport,
1176 				   sk->sk_protocol, RT_CONN_FLAGS(sk),
1177 				   sk->sk_bound_dev_if);
1178 	if (!IS_ERR(rt)) {
1179 		err = 0;
1180 		sk_setup_caps(sk, &rt->dst);
1181 	} else {
1182 		err = PTR_ERR(rt);
1183 
1184 		/* Routing failed... */
1185 		sk->sk_route_caps = 0;
1186 		/*
1187 		 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1188 		 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1189 		 */
1190 		if (!sock_net(sk)->ipv4.sysctl_ip_dynaddr ||
1191 		    sk->sk_state != TCP_SYN_SENT ||
1192 		    (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1193 		    (err = inet_sk_reselect_saddr(sk)) != 0)
1194 			sk->sk_err_soft = -err;
1195 	}
1196 
1197 	return err;
1198 }
1199 EXPORT_SYMBOL(inet_sk_rebuild_header);
1200 
1201 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1202 				 netdev_features_t features)
1203 {
1204 	bool udpfrag = false, fixedid = false, gso_partial, encap;
1205 	struct sk_buff *segs = ERR_PTR(-EINVAL);
1206 	const struct net_offload *ops;
1207 	unsigned int offset = 0;
1208 	struct iphdr *iph;
1209 	int proto, tot_len;
1210 	int nhoff;
1211 	int ihl;
1212 	int id;
1213 
1214 	skb_reset_network_header(skb);
1215 	nhoff = skb_network_header(skb) - skb_mac_header(skb);
1216 	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1217 		goto out;
1218 
1219 	iph = ip_hdr(skb);
1220 	ihl = iph->ihl * 4;
1221 	if (ihl < sizeof(*iph))
1222 		goto out;
1223 
1224 	id = ntohs(iph->id);
1225 	proto = iph->protocol;
1226 
1227 	/* Warning: after this point, iph might be no longer valid */
1228 	if (unlikely(!pskb_may_pull(skb, ihl)))
1229 		goto out;
1230 	__skb_pull(skb, ihl);
1231 
1232 	encap = SKB_GSO_CB(skb)->encap_level > 0;
1233 	if (encap)
1234 		features &= skb->dev->hw_enc_features;
1235 	SKB_GSO_CB(skb)->encap_level += ihl;
1236 
1237 	skb_reset_transport_header(skb);
1238 
1239 	segs = ERR_PTR(-EPROTONOSUPPORT);
1240 
1241 	if (!skb->encapsulation || encap) {
1242 		udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1243 		fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1244 
1245 		/* fixed ID is invalid if DF bit is not set */
1246 		if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1247 			goto out;
1248 	}
1249 
1250 	ops = rcu_dereference(inet_offloads[proto]);
1251 	if (likely(ops && ops->callbacks.gso_segment))
1252 		segs = ops->callbacks.gso_segment(skb, features);
1253 
1254 	if (IS_ERR_OR_NULL(segs))
1255 		goto out;
1256 
1257 	gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1258 
1259 	skb = segs;
1260 	do {
1261 		iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1262 		if (udpfrag) {
1263 			iph->frag_off = htons(offset >> 3);
1264 			if (skb->next)
1265 				iph->frag_off |= htons(IP_MF);
1266 			offset += skb->len - nhoff - ihl;
1267 			tot_len = skb->len - nhoff;
1268 		} else if (skb_is_gso(skb)) {
1269 			if (!fixedid) {
1270 				iph->id = htons(id);
1271 				id += skb_shinfo(skb)->gso_segs;
1272 			}
1273 
1274 			if (gso_partial)
1275 				tot_len = skb_shinfo(skb)->gso_size +
1276 					  SKB_GSO_CB(skb)->data_offset +
1277 					  skb->head - (unsigned char *)iph;
1278 			else
1279 				tot_len = skb->len - nhoff;
1280 		} else {
1281 			if (!fixedid)
1282 				iph->id = htons(id++);
1283 			tot_len = skb->len - nhoff;
1284 		}
1285 		iph->tot_len = htons(tot_len);
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 EXPORT_SYMBOL(inet_gso_segment);
1296 
1297 struct sk_buff **inet_gro_receive(struct sk_buff **head, 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 		u16 flush_id;
1338 
1339 		if (!NAPI_GRO_CB(p)->same_flow)
1340 			continue;
1341 
1342 		iph2 = (struct iphdr *)(p->data + off);
1343 		/* The above works because, with the exception of the top
1344 		 * (inner most) layer, we only aggregate pkts with the same
1345 		 * hdr length so all the hdrs we'll need to verify will start
1346 		 * at the same offset.
1347 		 */
1348 		if ((iph->protocol ^ iph2->protocol) |
1349 		    ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1350 		    ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1351 			NAPI_GRO_CB(p)->same_flow = 0;
1352 			continue;
1353 		}
1354 
1355 		/* All fields must match except length and checksum. */
1356 		NAPI_GRO_CB(p)->flush |=
1357 			(iph->ttl ^ iph2->ttl) |
1358 			(iph->tos ^ iph2->tos) |
1359 			((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1360 
1361 		NAPI_GRO_CB(p)->flush |= flush;
1362 
1363 		/* We need to store of the IP ID check to be included later
1364 		 * when we can verify that this packet does in fact belong
1365 		 * to a given flow.
1366 		 */
1367 		flush_id = (u16)(id - ntohs(iph2->id));
1368 
1369 		/* This bit of code makes it much easier for us to identify
1370 		 * the cases where we are doing atomic vs non-atomic IP ID
1371 		 * checks.  Specifically an atomic check can return IP ID
1372 		 * values 0 - 0xFFFF, while a non-atomic check can only
1373 		 * return 0 or 0xFFFF.
1374 		 */
1375 		if (!NAPI_GRO_CB(p)->is_atomic ||
1376 		    !(iph->frag_off & htons(IP_DF))) {
1377 			flush_id ^= NAPI_GRO_CB(p)->count;
1378 			flush_id = flush_id ? 0xFFFF : 0;
1379 		}
1380 
1381 		/* If the previous IP ID value was based on an atomic
1382 		 * datagram we can overwrite the value and ignore it.
1383 		 */
1384 		if (NAPI_GRO_CB(skb)->is_atomic)
1385 			NAPI_GRO_CB(p)->flush_id = flush_id;
1386 		else
1387 			NAPI_GRO_CB(p)->flush_id |= flush_id;
1388 	}
1389 
1390 	NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1391 	NAPI_GRO_CB(skb)->flush |= flush;
1392 	skb_set_network_header(skb, off);
1393 	/* The above will be needed by the transport layer if there is one
1394 	 * immediately following this IP hdr.
1395 	 */
1396 
1397 	/* Note : No need to call skb_gro_postpull_rcsum() here,
1398 	 * as we already checked checksum over ipv4 header was 0
1399 	 */
1400 	skb_gro_pull(skb, sizeof(*iph));
1401 	skb_set_transport_header(skb, skb_gro_offset(skb));
1402 
1403 	pp = call_gro_receive(ops->callbacks.gro_receive, head, skb);
1404 
1405 out_unlock:
1406 	rcu_read_unlock();
1407 
1408 out:
1409 	NAPI_GRO_CB(skb)->flush |= flush;
1410 
1411 	return pp;
1412 }
1413 EXPORT_SYMBOL(inet_gro_receive);
1414 
1415 static struct sk_buff **ipip_gro_receive(struct sk_buff **head,
1416 					 struct sk_buff *skb)
1417 {
1418 	if (NAPI_GRO_CB(skb)->encap_mark) {
1419 		NAPI_GRO_CB(skb)->flush = 1;
1420 		return NULL;
1421 	}
1422 
1423 	NAPI_GRO_CB(skb)->encap_mark = 1;
1424 
1425 	return inet_gro_receive(head, skb);
1426 }
1427 
1428 #define SECONDS_PER_DAY	86400
1429 
1430 /* inet_current_timestamp - Return IP network timestamp
1431  *
1432  * Return milliseconds since midnight in network byte order.
1433  */
1434 __be32 inet_current_timestamp(void)
1435 {
1436 	u32 secs;
1437 	u32 msecs;
1438 	struct timespec64 ts;
1439 
1440 	ktime_get_real_ts64(&ts);
1441 
1442 	/* Get secs since midnight. */
1443 	(void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1444 	/* Convert to msecs. */
1445 	msecs = secs * MSEC_PER_SEC;
1446 	/* Convert nsec to msec. */
1447 	msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1448 
1449 	/* Convert to network byte order. */
1450 	return htonl(msecs);
1451 }
1452 EXPORT_SYMBOL(inet_current_timestamp);
1453 
1454 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1455 {
1456 	if (sk->sk_family == AF_INET)
1457 		return ip_recv_error(sk, msg, len, addr_len);
1458 #if IS_ENABLED(CONFIG_IPV6)
1459 	if (sk->sk_family == AF_INET6)
1460 		return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1461 #endif
1462 	return -EINVAL;
1463 }
1464 
1465 int inet_gro_complete(struct sk_buff *skb, int nhoff)
1466 {
1467 	__be16 newlen = htons(skb->len - nhoff);
1468 	struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1469 	const struct net_offload *ops;
1470 	int proto = iph->protocol;
1471 	int err = -ENOSYS;
1472 
1473 	if (skb->encapsulation)
1474 		skb_set_inner_network_header(skb, nhoff);
1475 
1476 	csum_replace2(&iph->check, iph->tot_len, newlen);
1477 	iph->tot_len = newlen;
1478 
1479 	rcu_read_lock();
1480 	ops = rcu_dereference(inet_offloads[proto]);
1481 	if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1482 		goto out_unlock;
1483 
1484 	/* Only need to add sizeof(*iph) to get to the next hdr below
1485 	 * because any hdr with option will have been flushed in
1486 	 * inet_gro_receive().
1487 	 */
1488 	err = ops->callbacks.gro_complete(skb, nhoff + sizeof(*iph));
1489 
1490 out_unlock:
1491 	rcu_read_unlock();
1492 
1493 	return err;
1494 }
1495 EXPORT_SYMBOL(inet_gro_complete);
1496 
1497 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1498 {
1499 	skb->encapsulation = 1;
1500 	skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1501 	return inet_gro_complete(skb, nhoff);
1502 }
1503 
1504 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1505 			 unsigned short type, unsigned char protocol,
1506 			 struct net *net)
1507 {
1508 	struct socket *sock;
1509 	int rc = sock_create_kern(net, family, type, protocol, &sock);
1510 
1511 	if (rc == 0) {
1512 		*sk = sock->sk;
1513 		(*sk)->sk_allocation = GFP_ATOMIC;
1514 		/*
1515 		 * Unhash it so that IP input processing does not even see it,
1516 		 * we do not wish this socket to see incoming packets.
1517 		 */
1518 		(*sk)->sk_prot->unhash(*sk);
1519 	}
1520 	return rc;
1521 }
1522 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1523 
1524 u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt)
1525 {
1526 	return  *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt);
1527 }
1528 EXPORT_SYMBOL_GPL(snmp_get_cpu_field);
1529 
1530 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1531 {
1532 	unsigned long res = 0;
1533 	int i;
1534 
1535 	for_each_possible_cpu(i)
1536 		res += snmp_get_cpu_field(mib, i, offt);
1537 	return res;
1538 }
1539 EXPORT_SYMBOL_GPL(snmp_fold_field);
1540 
1541 #if BITS_PER_LONG==32
1542 
1543 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1544 			 size_t syncp_offset)
1545 {
1546 	void *bhptr;
1547 	struct u64_stats_sync *syncp;
1548 	u64 v;
1549 	unsigned int start;
1550 
1551 	bhptr = per_cpu_ptr(mib, cpu);
1552 	syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1553 	do {
1554 		start = u64_stats_fetch_begin_irq(syncp);
1555 		v = *(((u64 *)bhptr) + offt);
1556 	} while (u64_stats_fetch_retry_irq(syncp, start));
1557 
1558 	return v;
1559 }
1560 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1561 
1562 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1563 {
1564 	u64 res = 0;
1565 	int cpu;
1566 
1567 	for_each_possible_cpu(cpu) {
1568 		res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1569 	}
1570 	return res;
1571 }
1572 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1573 #endif
1574 
1575 #ifdef CONFIG_IP_MULTICAST
1576 static const struct net_protocol igmp_protocol = {
1577 	.handler =	igmp_rcv,
1578 	.netns_ok =	1,
1579 };
1580 #endif
1581 
1582 static const struct net_protocol tcp_protocol = {
1583 	.early_demux	=	tcp_v4_early_demux,
1584 	.handler	=	tcp_v4_rcv,
1585 	.err_handler	=	tcp_v4_err,
1586 	.no_policy	=	1,
1587 	.netns_ok	=	1,
1588 	.icmp_strict_tag_validation = 1,
1589 };
1590 
1591 static const struct net_protocol udp_protocol = {
1592 	.early_demux =	udp_v4_early_demux,
1593 	.handler =	udp_rcv,
1594 	.err_handler =	udp_err,
1595 	.no_policy =	1,
1596 	.netns_ok =	1,
1597 };
1598 
1599 static const struct net_protocol icmp_protocol = {
1600 	.handler =	icmp_rcv,
1601 	.err_handler =	icmp_err,
1602 	.no_policy =	1,
1603 	.netns_ok =	1,
1604 };
1605 
1606 static __net_init int ipv4_mib_init_net(struct net *net)
1607 {
1608 	int i;
1609 
1610 	net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1611 	if (!net->mib.tcp_statistics)
1612 		goto err_tcp_mib;
1613 	net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1614 	if (!net->mib.ip_statistics)
1615 		goto err_ip_mib;
1616 
1617 	for_each_possible_cpu(i) {
1618 		struct ipstats_mib *af_inet_stats;
1619 		af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1620 		u64_stats_init(&af_inet_stats->syncp);
1621 	}
1622 
1623 	net->mib.net_statistics = alloc_percpu(struct linux_mib);
1624 	if (!net->mib.net_statistics)
1625 		goto err_net_mib;
1626 	net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1627 	if (!net->mib.udp_statistics)
1628 		goto err_udp_mib;
1629 	net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1630 	if (!net->mib.udplite_statistics)
1631 		goto err_udplite_mib;
1632 	net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1633 	if (!net->mib.icmp_statistics)
1634 		goto err_icmp_mib;
1635 	net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1636 					      GFP_KERNEL);
1637 	if (!net->mib.icmpmsg_statistics)
1638 		goto err_icmpmsg_mib;
1639 
1640 	tcp_mib_init(net);
1641 	return 0;
1642 
1643 err_icmpmsg_mib:
1644 	free_percpu(net->mib.icmp_statistics);
1645 err_icmp_mib:
1646 	free_percpu(net->mib.udplite_statistics);
1647 err_udplite_mib:
1648 	free_percpu(net->mib.udp_statistics);
1649 err_udp_mib:
1650 	free_percpu(net->mib.net_statistics);
1651 err_net_mib:
1652 	free_percpu(net->mib.ip_statistics);
1653 err_ip_mib:
1654 	free_percpu(net->mib.tcp_statistics);
1655 err_tcp_mib:
1656 	return -ENOMEM;
1657 }
1658 
1659 static __net_exit void ipv4_mib_exit_net(struct net *net)
1660 {
1661 	kfree(net->mib.icmpmsg_statistics);
1662 	free_percpu(net->mib.icmp_statistics);
1663 	free_percpu(net->mib.udplite_statistics);
1664 	free_percpu(net->mib.udp_statistics);
1665 	free_percpu(net->mib.net_statistics);
1666 	free_percpu(net->mib.ip_statistics);
1667 	free_percpu(net->mib.tcp_statistics);
1668 }
1669 
1670 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1671 	.init = ipv4_mib_init_net,
1672 	.exit = ipv4_mib_exit_net,
1673 };
1674 
1675 static int __init init_ipv4_mibs(void)
1676 {
1677 	return register_pernet_subsys(&ipv4_mib_ops);
1678 }
1679 
1680 static __net_init int inet_init_net(struct net *net)
1681 {
1682 	/*
1683 	 * Set defaults for local port range
1684 	 */
1685 	seqlock_init(&net->ipv4.ip_local_ports.lock);
1686 	net->ipv4.ip_local_ports.range[0] =  32768;
1687 	net->ipv4.ip_local_ports.range[1] =  60999;
1688 
1689 	seqlock_init(&net->ipv4.ping_group_range.lock);
1690 	/*
1691 	 * Sane defaults - nobody may create ping sockets.
1692 	 * Boot scripts should set this to distro-specific group.
1693 	 */
1694 	net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1695 	net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1696 
1697 	/* Default values for sysctl-controlled parameters.
1698 	 * We set them here, in case sysctl is not compiled.
1699 	 */
1700 	net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1701 	net->ipv4.sysctl_ip_dynaddr = 0;
1702 	net->ipv4.sysctl_ip_early_demux = 1;
1703 
1704 	return 0;
1705 }
1706 
1707 static __net_exit void inet_exit_net(struct net *net)
1708 {
1709 }
1710 
1711 static __net_initdata struct pernet_operations af_inet_ops = {
1712 	.init = inet_init_net,
1713 	.exit = inet_exit_net,
1714 };
1715 
1716 static int __init init_inet_pernet_ops(void)
1717 {
1718 	return register_pernet_subsys(&af_inet_ops);
1719 }
1720 
1721 static int ipv4_proc_init(void);
1722 
1723 /*
1724  *	IP protocol layer initialiser
1725  */
1726 
1727 static struct packet_offload ip_packet_offload __read_mostly = {
1728 	.type = cpu_to_be16(ETH_P_IP),
1729 	.callbacks = {
1730 		.gso_segment = inet_gso_segment,
1731 		.gro_receive = inet_gro_receive,
1732 		.gro_complete = inet_gro_complete,
1733 	},
1734 };
1735 
1736 static const struct net_offload ipip_offload = {
1737 	.callbacks = {
1738 		.gso_segment	= inet_gso_segment,
1739 		.gro_receive	= ipip_gro_receive,
1740 		.gro_complete	= ipip_gro_complete,
1741 	},
1742 };
1743 
1744 static int __init ipv4_offload_init(void)
1745 {
1746 	/*
1747 	 * Add offloads
1748 	 */
1749 	if (udpv4_offload_init() < 0)
1750 		pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1751 	if (tcpv4_offload_init() < 0)
1752 		pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1753 
1754 	dev_add_offload(&ip_packet_offload);
1755 	inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1756 	return 0;
1757 }
1758 
1759 fs_initcall(ipv4_offload_init);
1760 
1761 static struct packet_type ip_packet_type __read_mostly = {
1762 	.type = cpu_to_be16(ETH_P_IP),
1763 	.func = ip_rcv,
1764 };
1765 
1766 static int __init inet_init(void)
1767 {
1768 	struct inet_protosw *q;
1769 	struct list_head *r;
1770 	int rc = -EINVAL;
1771 
1772 	sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1773 
1774 	rc = proto_register(&tcp_prot, 1);
1775 	if (rc)
1776 		goto out;
1777 
1778 	rc = proto_register(&udp_prot, 1);
1779 	if (rc)
1780 		goto out_unregister_tcp_proto;
1781 
1782 	rc = proto_register(&raw_prot, 1);
1783 	if (rc)
1784 		goto out_unregister_udp_proto;
1785 
1786 	rc = proto_register(&ping_prot, 1);
1787 	if (rc)
1788 		goto out_unregister_raw_proto;
1789 
1790 	/*
1791 	 *	Tell SOCKET that we are alive...
1792 	 */
1793 
1794 	(void)sock_register(&inet_family_ops);
1795 
1796 #ifdef CONFIG_SYSCTL
1797 	ip_static_sysctl_init();
1798 #endif
1799 
1800 	/*
1801 	 *	Add all the base protocols.
1802 	 */
1803 
1804 	if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1805 		pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1806 	if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1807 		pr_crit("%s: Cannot add UDP protocol\n", __func__);
1808 	if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1809 		pr_crit("%s: Cannot add TCP protocol\n", __func__);
1810 #ifdef CONFIG_IP_MULTICAST
1811 	if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1812 		pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1813 #endif
1814 
1815 	/* Register the socket-side information for inet_create. */
1816 	for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1817 		INIT_LIST_HEAD(r);
1818 
1819 	for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1820 		inet_register_protosw(q);
1821 
1822 	/*
1823 	 *	Set the ARP module up
1824 	 */
1825 
1826 	arp_init();
1827 
1828 	/*
1829 	 *	Set the IP module up
1830 	 */
1831 
1832 	ip_init();
1833 
1834 	tcp_v4_init();
1835 
1836 	/* Setup TCP slab cache for open requests. */
1837 	tcp_init();
1838 
1839 	/* Setup UDP memory threshold */
1840 	udp_init();
1841 
1842 	/* Add UDP-Lite (RFC 3828) */
1843 	udplite4_register();
1844 
1845 	ping_init();
1846 
1847 	/*
1848 	 *	Set the ICMP layer up
1849 	 */
1850 
1851 	if (icmp_init() < 0)
1852 		panic("Failed to create the ICMP control socket.\n");
1853 
1854 	/*
1855 	 *	Initialise the multicast router
1856 	 */
1857 #if defined(CONFIG_IP_MROUTE)
1858 	if (ip_mr_init())
1859 		pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1860 #endif
1861 
1862 	if (init_inet_pernet_ops())
1863 		pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
1864 	/*
1865 	 *	Initialise per-cpu ipv4 mibs
1866 	 */
1867 
1868 	if (init_ipv4_mibs())
1869 		pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1870 
1871 	ipv4_proc_init();
1872 
1873 	ipfrag_init();
1874 
1875 	dev_add_pack(&ip_packet_type);
1876 
1877 	ip_tunnel_core_init();
1878 
1879 	rc = 0;
1880 out:
1881 	return rc;
1882 out_unregister_raw_proto:
1883 	proto_unregister(&raw_prot);
1884 out_unregister_udp_proto:
1885 	proto_unregister(&udp_prot);
1886 out_unregister_tcp_proto:
1887 	proto_unregister(&tcp_prot);
1888 	goto out;
1889 }
1890 
1891 fs_initcall(inet_init);
1892 
1893 /* ------------------------------------------------------------------------ */
1894 
1895 #ifdef CONFIG_PROC_FS
1896 static int __init ipv4_proc_init(void)
1897 {
1898 	int rc = 0;
1899 
1900 	if (raw_proc_init())
1901 		goto out_raw;
1902 	if (tcp4_proc_init())
1903 		goto out_tcp;
1904 	if (udp4_proc_init())
1905 		goto out_udp;
1906 	if (ping_proc_init())
1907 		goto out_ping;
1908 	if (ip_misc_proc_init())
1909 		goto out_misc;
1910 out:
1911 	return rc;
1912 out_misc:
1913 	ping_proc_exit();
1914 out_ping:
1915 	udp4_proc_exit();
1916 out_udp:
1917 	tcp4_proc_exit();
1918 out_tcp:
1919 	raw_proc_exit();
1920 out_raw:
1921 	rc = -ENOMEM;
1922 	goto out;
1923 }
1924 
1925 #else /* CONFIG_PROC_FS */
1926 static int __init ipv4_proc_init(void)
1927 {
1928 	return 0;
1929 }
1930 #endif /* CONFIG_PROC_FS */
1931