xref: /openbmc/linux/net/ipv4/af_inet.c (revision 6dfcd296)
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 <asm/uaccess.h>
94 
95 #include <linux/inet.h>
96 #include <linux/igmp.h>
97 #include <linux/inetdevice.h>
98 #include <linux/netdevice.h>
99 #include <net/checksum.h>
100 #include <net/ip.h>
101 #include <net/protocol.h>
102 #include <net/arp.h>
103 #include <net/route.h>
104 #include <net/ip_fib.h>
105 #include <net/inet_connection_sock.h>
106 #include <net/tcp.h>
107 #include <net/udp.h>
108 #include <net/udplite.h>
109 #include <net/ping.h>
110 #include <linux/skbuff.h>
111 #include <net/sock.h>
112 #include <net/raw.h>
113 #include <net/icmp.h>
114 #include <net/inet_common.h>
115 #include <net/ip_tunnels.h>
116 #include <net/xfrm.h>
117 #include <net/net_namespace.h>
118 #include <net/secure_seq.h>
119 #ifdef CONFIG_IP_MROUTE
120 #include <linux/mroute.h>
121 #endif
122 #include <net/l3mdev.h>
123 
124 
125 /* The inetsw table contains everything that inet_create needs to
126  * build a new socket.
127  */
128 static struct list_head inetsw[SOCK_MAX];
129 static DEFINE_SPINLOCK(inetsw_lock);
130 
131 /* New destruction routine */
132 
133 void inet_sock_destruct(struct sock *sk)
134 {
135 	struct inet_sock *inet = inet_sk(sk);
136 
137 	__skb_queue_purge(&sk->sk_receive_queue);
138 	__skb_queue_purge(&sk->sk_error_queue);
139 
140 	sk_mem_reclaim(sk);
141 
142 	if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
143 		pr_err("Attempt to release TCP socket in state %d %p\n",
144 		       sk->sk_state, sk);
145 		return;
146 	}
147 	if (!sock_flag(sk, SOCK_DEAD)) {
148 		pr_err("Attempt to release alive inet socket %p\n", sk);
149 		return;
150 	}
151 
152 	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
153 	WARN_ON(atomic_read(&sk->sk_wmem_alloc));
154 	WARN_ON(sk->sk_wmem_queued);
155 	WARN_ON(sk->sk_forward_alloc);
156 
157 	kfree(rcu_dereference_protected(inet->inet_opt, 1));
158 	dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
159 	dst_release(sk->sk_rx_dst);
160 	sk_refcnt_debug_dec(sk);
161 }
162 EXPORT_SYMBOL(inet_sock_destruct);
163 
164 /*
165  *	The routines beyond this point handle the behaviour of an AF_INET
166  *	socket object. Mostly it punts to the subprotocols of IP to do
167  *	the work.
168  */
169 
170 /*
171  *	Automatically bind an unbound socket.
172  */
173 
174 static int inet_autobind(struct sock *sk)
175 {
176 	struct inet_sock *inet;
177 	/* We may need to bind the socket. */
178 	lock_sock(sk);
179 	inet = inet_sk(sk);
180 	if (!inet->inet_num) {
181 		if (sk->sk_prot->get_port(sk, 0)) {
182 			release_sock(sk);
183 			return -EAGAIN;
184 		}
185 		inet->inet_sport = htons(inet->inet_num);
186 	}
187 	release_sock(sk);
188 	return 0;
189 }
190 
191 /*
192  *	Move a socket into listening state.
193  */
194 int inet_listen(struct socket *sock, int backlog)
195 {
196 	struct sock *sk = sock->sk;
197 	unsigned char old_state;
198 	int err;
199 
200 	lock_sock(sk);
201 
202 	err = -EINVAL;
203 	if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
204 		goto out;
205 
206 	old_state = sk->sk_state;
207 	if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
208 		goto out;
209 
210 	/* Really, if the socket is already in listen state
211 	 * we can only allow the backlog to be adjusted.
212 	 */
213 	if (old_state != TCP_LISTEN) {
214 		/* 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 	}
380 out:
381 	return err;
382 out_rcu_unlock:
383 	rcu_read_unlock();
384 	goto out;
385 }
386 
387 
388 /*
389  *	The peer socket should always be NULL (or else). When we call this
390  *	function we are destroying the object and from then on nobody
391  *	should refer to it.
392  */
393 int inet_release(struct socket *sock)
394 {
395 	struct sock *sk = sock->sk;
396 
397 	if (sk) {
398 		long timeout;
399 
400 		/* Applications forget to leave groups before exiting */
401 		ip_mc_drop_socket(sk);
402 
403 		/* If linger is set, we don't return until the close
404 		 * is complete.  Otherwise we return immediately. The
405 		 * actually closing is done the same either way.
406 		 *
407 		 * If the close is due to the process exiting, we never
408 		 * linger..
409 		 */
410 		timeout = 0;
411 		if (sock_flag(sk, SOCK_LINGER) &&
412 		    !(current->flags & PF_EXITING))
413 			timeout = sk->sk_lingertime;
414 		sock->sk = NULL;
415 		sk->sk_prot->close(sk, timeout);
416 	}
417 	return 0;
418 }
419 EXPORT_SYMBOL(inet_release);
420 
421 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
422 {
423 	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
424 	struct sock *sk = sock->sk;
425 	struct inet_sock *inet = inet_sk(sk);
426 	struct net *net = sock_net(sk);
427 	unsigned short snum;
428 	int chk_addr_ret;
429 	u32 tb_id = RT_TABLE_LOCAL;
430 	int err;
431 
432 	/* If the socket has its own bind function then use it. (RAW) */
433 	if (sk->sk_prot->bind) {
434 		err = sk->sk_prot->bind(sk, uaddr, addr_len);
435 		goto out;
436 	}
437 	err = -EINVAL;
438 	if (addr_len < sizeof(struct sockaddr_in))
439 		goto out;
440 
441 	if (addr->sin_family != AF_INET) {
442 		/* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
443 		 * only if s_addr is INADDR_ANY.
444 		 */
445 		err = -EAFNOSUPPORT;
446 		if (addr->sin_family != AF_UNSPEC ||
447 		    addr->sin_addr.s_addr != htonl(INADDR_ANY))
448 			goto out;
449 	}
450 
451 	tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
452 	chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
453 
454 	/* Not specified by any standard per-se, however it breaks too
455 	 * many applications when removed.  It is unfortunate since
456 	 * allowing applications to make a non-local bind solves
457 	 * several problems with systems using dynamic addressing.
458 	 * (ie. your servers still start up even if your ISDN link
459 	 *  is temporarily down)
460 	 */
461 	err = -EADDRNOTAVAIL;
462 	if (!net->ipv4.sysctl_ip_nonlocal_bind &&
463 	    !(inet->freebind || inet->transparent) &&
464 	    addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
465 	    chk_addr_ret != RTN_LOCAL &&
466 	    chk_addr_ret != RTN_MULTICAST &&
467 	    chk_addr_ret != RTN_BROADCAST)
468 		goto out;
469 
470 	snum = ntohs(addr->sin_port);
471 	err = -EACCES;
472 	if (snum && snum < PROT_SOCK &&
473 	    !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
474 		goto out;
475 
476 	/*      We keep a pair of addresses. rcv_saddr is the one
477 	 *      used by hash lookups, and saddr is used for transmit.
478 	 *
479 	 *      In the BSD API these are the same except where it
480 	 *      would be illegal to use them (multicast/broadcast) in
481 	 *      which case the sending device address is used.
482 	 */
483 	lock_sock(sk);
484 
485 	/* Check these errors (active socket, double bind). */
486 	err = -EINVAL;
487 	if (sk->sk_state != TCP_CLOSE || inet->inet_num)
488 		goto out_release_sock;
489 
490 	inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
491 	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
492 		inet->inet_saddr = 0;  /* Use device */
493 
494 	/* Make sure we are allowed to bind here. */
495 	if ((snum || !inet->bind_address_no_port) &&
496 	    sk->sk_prot->get_port(sk, snum)) {
497 		inet->inet_saddr = inet->inet_rcv_saddr = 0;
498 		err = -EADDRINUSE;
499 		goto out_release_sock;
500 	}
501 
502 	if (inet->inet_rcv_saddr)
503 		sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
504 	if (snum)
505 		sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
506 	inet->inet_sport = htons(inet->inet_num);
507 	inet->inet_daddr = 0;
508 	inet->inet_dport = 0;
509 	sk_dst_reset(sk);
510 	err = 0;
511 out_release_sock:
512 	release_sock(sk);
513 out:
514 	return err;
515 }
516 EXPORT_SYMBOL(inet_bind);
517 
518 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
519 		       int addr_len, int flags)
520 {
521 	struct sock *sk = sock->sk;
522 
523 	if (addr_len < sizeof(uaddr->sa_family))
524 		return -EINVAL;
525 	if (uaddr->sa_family == AF_UNSPEC)
526 		return sk->sk_prot->disconnect(sk, flags);
527 
528 	if (!inet_sk(sk)->inet_num && inet_autobind(sk))
529 		return -EAGAIN;
530 	return sk->sk_prot->connect(sk, uaddr, addr_len);
531 }
532 EXPORT_SYMBOL(inet_dgram_connect);
533 
534 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
535 {
536 	DEFINE_WAIT(wait);
537 
538 	prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
539 	sk->sk_write_pending += writebias;
540 
541 	/* Basic assumption: if someone sets sk->sk_err, he _must_
542 	 * change state of the socket from TCP_SYN_*.
543 	 * Connect() does not allow to get error notifications
544 	 * without closing the socket.
545 	 */
546 	while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
547 		release_sock(sk);
548 		timeo = schedule_timeout(timeo);
549 		lock_sock(sk);
550 		if (signal_pending(current) || !timeo)
551 			break;
552 		prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
553 	}
554 	finish_wait(sk_sleep(sk), &wait);
555 	sk->sk_write_pending -= writebias;
556 	return timeo;
557 }
558 
559 /*
560  *	Connect to a remote host. There is regrettably still a little
561  *	TCP 'magic' in here.
562  */
563 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
564 			  int addr_len, int flags)
565 {
566 	struct sock *sk = sock->sk;
567 	int err;
568 	long timeo;
569 
570 	if (addr_len < sizeof(uaddr->sa_family))
571 		return -EINVAL;
572 
573 	if (uaddr->sa_family == AF_UNSPEC) {
574 		err = sk->sk_prot->disconnect(sk, flags);
575 		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
576 		goto out;
577 	}
578 
579 	switch (sock->state) {
580 	default:
581 		err = -EINVAL;
582 		goto out;
583 	case SS_CONNECTED:
584 		err = -EISCONN;
585 		goto out;
586 	case SS_CONNECTING:
587 		err = -EALREADY;
588 		/* Fall out of switch with err, set for this state */
589 		break;
590 	case SS_UNCONNECTED:
591 		err = -EISCONN;
592 		if (sk->sk_state != TCP_CLOSE)
593 			goto out;
594 
595 		err = sk->sk_prot->connect(sk, uaddr, addr_len);
596 		if (err < 0)
597 			goto out;
598 
599 		sock->state = SS_CONNECTING;
600 
601 		/* Just entered SS_CONNECTING state; the only
602 		 * difference is that return value in non-blocking
603 		 * case is EINPROGRESS, rather than EALREADY.
604 		 */
605 		err = -EINPROGRESS;
606 		break;
607 	}
608 
609 	timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
610 
611 	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
612 		int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
613 				tcp_sk(sk)->fastopen_req &&
614 				tcp_sk(sk)->fastopen_req->data ? 1 : 0;
615 
616 		/* Error code is set above */
617 		if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
618 			goto out;
619 
620 		err = sock_intr_errno(timeo);
621 		if (signal_pending(current))
622 			goto out;
623 	}
624 
625 	/* Connection was closed by RST, timeout, ICMP error
626 	 * or another process disconnected us.
627 	 */
628 	if (sk->sk_state == TCP_CLOSE)
629 		goto sock_error;
630 
631 	/* sk->sk_err may be not zero now, if RECVERR was ordered by user
632 	 * and error was received after socket entered established state.
633 	 * Hence, it is handled normally after connect() return successfully.
634 	 */
635 
636 	sock->state = SS_CONNECTED;
637 	err = 0;
638 out:
639 	return err;
640 
641 sock_error:
642 	err = sock_error(sk) ? : -ECONNABORTED;
643 	sock->state = SS_UNCONNECTED;
644 	if (sk->sk_prot->disconnect(sk, flags))
645 		sock->state = SS_DISCONNECTING;
646 	goto out;
647 }
648 EXPORT_SYMBOL(__inet_stream_connect);
649 
650 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
651 			int addr_len, int flags)
652 {
653 	int err;
654 
655 	lock_sock(sock->sk);
656 	err = __inet_stream_connect(sock, uaddr, addr_len, flags);
657 	release_sock(sock->sk);
658 	return err;
659 }
660 EXPORT_SYMBOL(inet_stream_connect);
661 
662 /*
663  *	Accept a pending connection. The TCP layer now gives BSD semantics.
664  */
665 
666 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
667 {
668 	struct sock *sk1 = sock->sk;
669 	int err = -EINVAL;
670 	struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
671 
672 	if (!sk2)
673 		goto do_err;
674 
675 	lock_sock(sk2);
676 
677 	sock_rps_record_flow(sk2);
678 	WARN_ON(!((1 << sk2->sk_state) &
679 		  (TCPF_ESTABLISHED | TCPF_SYN_RECV |
680 		  TCPF_CLOSE_WAIT | TCPF_CLOSE)));
681 
682 	sock_graft(sk2, newsock);
683 
684 	newsock->state = SS_CONNECTED;
685 	err = 0;
686 	release_sock(sk2);
687 do_err:
688 	return err;
689 }
690 EXPORT_SYMBOL(inet_accept);
691 
692 
693 /*
694  *	This does both peername and sockname.
695  */
696 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
697 			int *uaddr_len, int peer)
698 {
699 	struct sock *sk		= sock->sk;
700 	struct inet_sock *inet	= inet_sk(sk);
701 	DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
702 
703 	sin->sin_family = AF_INET;
704 	if (peer) {
705 		if (!inet->inet_dport ||
706 		    (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
707 		     peer == 1))
708 			return -ENOTCONN;
709 		sin->sin_port = inet->inet_dport;
710 		sin->sin_addr.s_addr = inet->inet_daddr;
711 	} else {
712 		__be32 addr = inet->inet_rcv_saddr;
713 		if (!addr)
714 			addr = inet->inet_saddr;
715 		sin->sin_port = inet->inet_sport;
716 		sin->sin_addr.s_addr = addr;
717 	}
718 	memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
719 	*uaddr_len = sizeof(*sin);
720 	return 0;
721 }
722 EXPORT_SYMBOL(inet_getname);
723 
724 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
725 {
726 	struct sock *sk = sock->sk;
727 
728 	sock_rps_record_flow(sk);
729 
730 	/* We may need to bind the socket. */
731 	if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
732 	    inet_autobind(sk))
733 		return -EAGAIN;
734 
735 	return sk->sk_prot->sendmsg(sk, msg, size);
736 }
737 EXPORT_SYMBOL(inet_sendmsg);
738 
739 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
740 		      size_t size, int flags)
741 {
742 	struct sock *sk = sock->sk;
743 
744 	sock_rps_record_flow(sk);
745 
746 	/* We may need to bind the socket. */
747 	if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
748 	    inet_autobind(sk))
749 		return -EAGAIN;
750 
751 	if (sk->sk_prot->sendpage)
752 		return sk->sk_prot->sendpage(sk, page, offset, size, flags);
753 	return sock_no_sendpage(sock, page, offset, size, flags);
754 }
755 EXPORT_SYMBOL(inet_sendpage);
756 
757 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
758 		 int flags)
759 {
760 	struct sock *sk = sock->sk;
761 	int addr_len = 0;
762 	int err;
763 
764 	sock_rps_record_flow(sk);
765 
766 	err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
767 				   flags & ~MSG_DONTWAIT, &addr_len);
768 	if (err >= 0)
769 		msg->msg_namelen = addr_len;
770 	return err;
771 }
772 EXPORT_SYMBOL(inet_recvmsg);
773 
774 int inet_shutdown(struct socket *sock, int how)
775 {
776 	struct sock *sk = sock->sk;
777 	int err = 0;
778 
779 	/* This should really check to make sure
780 	 * the socket is a TCP socket. (WHY AC...)
781 	 */
782 	how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
783 		       1->2 bit 2 snds.
784 		       2->3 */
785 	if ((how & ~SHUTDOWN_MASK) || !how)	/* MAXINT->0 */
786 		return -EINVAL;
787 
788 	lock_sock(sk);
789 	if (sock->state == SS_CONNECTING) {
790 		if ((1 << sk->sk_state) &
791 		    (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
792 			sock->state = SS_DISCONNECTING;
793 		else
794 			sock->state = SS_CONNECTED;
795 	}
796 
797 	switch (sk->sk_state) {
798 	case TCP_CLOSE:
799 		err = -ENOTCONN;
800 		/* Hack to wake up other listeners, who can poll for
801 		   POLLHUP, even on eg. unconnected UDP sockets -- RR */
802 	default:
803 		sk->sk_shutdown |= how;
804 		if (sk->sk_prot->shutdown)
805 			sk->sk_prot->shutdown(sk, how);
806 		break;
807 
808 	/* Remaining two branches are temporary solution for missing
809 	 * close() in multithreaded environment. It is _not_ a good idea,
810 	 * but we have no choice until close() is repaired at VFS level.
811 	 */
812 	case TCP_LISTEN:
813 		if (!(how & RCV_SHUTDOWN))
814 			break;
815 		/* Fall through */
816 	case TCP_SYN_SENT:
817 		err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
818 		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
819 		break;
820 	}
821 
822 	/* Wake up anyone sleeping in poll. */
823 	sk->sk_state_change(sk);
824 	release_sock(sk);
825 	return err;
826 }
827 EXPORT_SYMBOL(inet_shutdown);
828 
829 /*
830  *	ioctl() calls you can issue on an INET socket. Most of these are
831  *	device configuration and stuff and very rarely used. Some ioctls
832  *	pass on to the socket itself.
833  *
834  *	NOTE: I like the idea of a module for the config stuff. ie ifconfig
835  *	loads the devconfigure module does its configuring and unloads it.
836  *	There's a good 20K of config code hanging around the kernel.
837  */
838 
839 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
840 {
841 	struct sock *sk = sock->sk;
842 	int err = 0;
843 	struct net *net = sock_net(sk);
844 
845 	switch (cmd) {
846 	case SIOCGSTAMP:
847 		err = sock_get_timestamp(sk, (struct timeval __user *)arg);
848 		break;
849 	case SIOCGSTAMPNS:
850 		err = sock_get_timestampns(sk, (struct timespec __user *)arg);
851 		break;
852 	case SIOCADDRT:
853 	case SIOCDELRT:
854 	case SIOCRTMSG:
855 		err = ip_rt_ioctl(net, cmd, (void __user *)arg);
856 		break;
857 	case SIOCDARP:
858 	case SIOCGARP:
859 	case SIOCSARP:
860 		err = arp_ioctl(net, cmd, (void __user *)arg);
861 		break;
862 	case SIOCGIFADDR:
863 	case SIOCSIFADDR:
864 	case SIOCGIFBRDADDR:
865 	case SIOCSIFBRDADDR:
866 	case SIOCGIFNETMASK:
867 	case SIOCSIFNETMASK:
868 	case SIOCGIFDSTADDR:
869 	case SIOCSIFDSTADDR:
870 	case SIOCSIFPFLAGS:
871 	case SIOCGIFPFLAGS:
872 	case SIOCSIFFLAGS:
873 		err = devinet_ioctl(net, cmd, (void __user *)arg);
874 		break;
875 	default:
876 		if (sk->sk_prot->ioctl)
877 			err = sk->sk_prot->ioctl(sk, cmd, arg);
878 		else
879 			err = -ENOIOCTLCMD;
880 		break;
881 	}
882 	return err;
883 }
884 EXPORT_SYMBOL(inet_ioctl);
885 
886 #ifdef CONFIG_COMPAT
887 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
888 {
889 	struct sock *sk = sock->sk;
890 	int err = -ENOIOCTLCMD;
891 
892 	if (sk->sk_prot->compat_ioctl)
893 		err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
894 
895 	return err;
896 }
897 #endif
898 
899 const struct proto_ops inet_stream_ops = {
900 	.family		   = PF_INET,
901 	.owner		   = THIS_MODULE,
902 	.release	   = inet_release,
903 	.bind		   = inet_bind,
904 	.connect	   = inet_stream_connect,
905 	.socketpair	   = sock_no_socketpair,
906 	.accept		   = inet_accept,
907 	.getname	   = inet_getname,
908 	.poll		   = tcp_poll,
909 	.ioctl		   = inet_ioctl,
910 	.listen		   = inet_listen,
911 	.shutdown	   = inet_shutdown,
912 	.setsockopt	   = sock_common_setsockopt,
913 	.getsockopt	   = sock_common_getsockopt,
914 	.sendmsg	   = inet_sendmsg,
915 	.recvmsg	   = inet_recvmsg,
916 	.mmap		   = sock_no_mmap,
917 	.sendpage	   = inet_sendpage,
918 	.splice_read	   = tcp_splice_read,
919 	.read_sock	   = tcp_read_sock,
920 	.peek_len	   = tcp_peek_len,
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 	.set_peek_off	   = sk_set_peek_off,
949 #ifdef CONFIG_COMPAT
950 	.compat_setsockopt = compat_sock_common_setsockopt,
951 	.compat_getsockopt = compat_sock_common_getsockopt,
952 	.compat_ioctl	   = inet_compat_ioctl,
953 #endif
954 };
955 EXPORT_SYMBOL(inet_dgram_ops);
956 
957 /*
958  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
959  * udp_poll
960  */
961 static const struct proto_ops inet_sockraw_ops = {
962 	.family		   = PF_INET,
963 	.owner		   = THIS_MODULE,
964 	.release	   = inet_release,
965 	.bind		   = inet_bind,
966 	.connect	   = inet_dgram_connect,
967 	.socketpair	   = sock_no_socketpair,
968 	.accept		   = sock_no_accept,
969 	.getname	   = inet_getname,
970 	.poll		   = datagram_poll,
971 	.ioctl		   = inet_ioctl,
972 	.listen		   = sock_no_listen,
973 	.shutdown	   = inet_shutdown,
974 	.setsockopt	   = sock_common_setsockopt,
975 	.getsockopt	   = sock_common_getsockopt,
976 	.sendmsg	   = inet_sendmsg,
977 	.recvmsg	   = inet_recvmsg,
978 	.mmap		   = sock_no_mmap,
979 	.sendpage	   = inet_sendpage,
980 #ifdef CONFIG_COMPAT
981 	.compat_setsockopt = compat_sock_common_setsockopt,
982 	.compat_getsockopt = compat_sock_common_getsockopt,
983 	.compat_ioctl	   = inet_compat_ioctl,
984 #endif
985 };
986 
987 static const struct net_proto_family inet_family_ops = {
988 	.family = PF_INET,
989 	.create = inet_create,
990 	.owner	= THIS_MODULE,
991 };
992 
993 /* Upon startup we insert all the elements in inetsw_array[] into
994  * the linked list inetsw.
995  */
996 static struct inet_protosw inetsw_array[] =
997 {
998 	{
999 		.type =       SOCK_STREAM,
1000 		.protocol =   IPPROTO_TCP,
1001 		.prot =       &tcp_prot,
1002 		.ops =        &inet_stream_ops,
1003 		.flags =      INET_PROTOSW_PERMANENT |
1004 			      INET_PROTOSW_ICSK,
1005 	},
1006 
1007 	{
1008 		.type =       SOCK_DGRAM,
1009 		.protocol =   IPPROTO_UDP,
1010 		.prot =       &udp_prot,
1011 		.ops =        &inet_dgram_ops,
1012 		.flags =      INET_PROTOSW_PERMANENT,
1013        },
1014 
1015        {
1016 		.type =       SOCK_DGRAM,
1017 		.protocol =   IPPROTO_ICMP,
1018 		.prot =       &ping_prot,
1019 		.ops =        &inet_dgram_ops,
1020 		.flags =      INET_PROTOSW_REUSE,
1021        },
1022 
1023        {
1024 	       .type =       SOCK_RAW,
1025 	       .protocol =   IPPROTO_IP,	/* wild card */
1026 	       .prot =       &raw_prot,
1027 	       .ops =        &inet_sockraw_ops,
1028 	       .flags =      INET_PROTOSW_REUSE,
1029        }
1030 };
1031 
1032 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1033 
1034 void inet_register_protosw(struct inet_protosw *p)
1035 {
1036 	struct list_head *lh;
1037 	struct inet_protosw *answer;
1038 	int protocol = p->protocol;
1039 	struct list_head *last_perm;
1040 
1041 	spin_lock_bh(&inetsw_lock);
1042 
1043 	if (p->type >= SOCK_MAX)
1044 		goto out_illegal;
1045 
1046 	/* If we are trying to override a permanent protocol, bail. */
1047 	last_perm = &inetsw[p->type];
1048 	list_for_each(lh, &inetsw[p->type]) {
1049 		answer = list_entry(lh, struct inet_protosw, list);
1050 		/* Check only the non-wild match. */
1051 		if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1052 			break;
1053 		if (protocol == answer->protocol)
1054 			goto out_permanent;
1055 		last_perm = lh;
1056 	}
1057 
1058 	/* Add the new entry after the last permanent entry if any, so that
1059 	 * the new entry does not override a permanent entry when matched with
1060 	 * a wild-card protocol. But it is allowed to override any existing
1061 	 * non-permanent entry.  This means that when we remove this entry, the
1062 	 * system automatically returns to the old behavior.
1063 	 */
1064 	list_add_rcu(&p->list, last_perm);
1065 out:
1066 	spin_unlock_bh(&inetsw_lock);
1067 
1068 	return;
1069 
1070 out_permanent:
1071 	pr_err("Attempt to override permanent protocol %d\n", protocol);
1072 	goto out;
1073 
1074 out_illegal:
1075 	pr_err("Ignoring attempt to register invalid socket type %d\n",
1076 	       p->type);
1077 	goto out;
1078 }
1079 EXPORT_SYMBOL(inet_register_protosw);
1080 
1081 void inet_unregister_protosw(struct inet_protosw *p)
1082 {
1083 	if (INET_PROTOSW_PERMANENT & p->flags) {
1084 		pr_err("Attempt to unregister permanent protocol %d\n",
1085 		       p->protocol);
1086 	} else {
1087 		spin_lock_bh(&inetsw_lock);
1088 		list_del_rcu(&p->list);
1089 		spin_unlock_bh(&inetsw_lock);
1090 
1091 		synchronize_net();
1092 	}
1093 }
1094 EXPORT_SYMBOL(inet_unregister_protosw);
1095 
1096 static int inet_sk_reselect_saddr(struct sock *sk)
1097 {
1098 	struct inet_sock *inet = inet_sk(sk);
1099 	__be32 old_saddr = inet->inet_saddr;
1100 	__be32 daddr = inet->inet_daddr;
1101 	struct flowi4 *fl4;
1102 	struct rtable *rt;
1103 	__be32 new_saddr;
1104 	struct ip_options_rcu *inet_opt;
1105 
1106 	inet_opt = rcu_dereference_protected(inet->inet_opt,
1107 					     lockdep_sock_is_held(sk));
1108 	if (inet_opt && inet_opt->opt.srr)
1109 		daddr = inet_opt->opt.faddr;
1110 
1111 	/* Query new route. */
1112 	fl4 = &inet->cork.fl.u.ip4;
1113 	rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1114 			      sk->sk_bound_dev_if, sk->sk_protocol,
1115 			      inet->inet_sport, inet->inet_dport, sk);
1116 	if (IS_ERR(rt))
1117 		return PTR_ERR(rt);
1118 
1119 	sk_setup_caps(sk, &rt->dst);
1120 
1121 	new_saddr = fl4->saddr;
1122 
1123 	if (new_saddr == old_saddr)
1124 		return 0;
1125 
1126 	if (sock_net(sk)->ipv4.sysctl_ip_dynaddr > 1) {
1127 		pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1128 			__func__, &old_saddr, &new_saddr);
1129 	}
1130 
1131 	inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1132 
1133 	/*
1134 	 * XXX The only one ugly spot where we need to
1135 	 * XXX really change the sockets identity after
1136 	 * XXX it has entered the hashes. -DaveM
1137 	 *
1138 	 * Besides that, it does not check for connection
1139 	 * uniqueness. Wait for troubles.
1140 	 */
1141 	return __sk_prot_rehash(sk);
1142 }
1143 
1144 int inet_sk_rebuild_header(struct sock *sk)
1145 {
1146 	struct inet_sock *inet = inet_sk(sk);
1147 	struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1148 	__be32 daddr;
1149 	struct ip_options_rcu *inet_opt;
1150 	struct flowi4 *fl4;
1151 	int err;
1152 
1153 	/* Route is OK, nothing to do. */
1154 	if (rt)
1155 		return 0;
1156 
1157 	/* Reroute. */
1158 	rcu_read_lock();
1159 	inet_opt = rcu_dereference(inet->inet_opt);
1160 	daddr = inet->inet_daddr;
1161 	if (inet_opt && inet_opt->opt.srr)
1162 		daddr = inet_opt->opt.faddr;
1163 	rcu_read_unlock();
1164 	fl4 = &inet->cork.fl.u.ip4;
1165 	rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1166 				   inet->inet_dport, inet->inet_sport,
1167 				   sk->sk_protocol, RT_CONN_FLAGS(sk),
1168 				   sk->sk_bound_dev_if);
1169 	if (!IS_ERR(rt)) {
1170 		err = 0;
1171 		sk_setup_caps(sk, &rt->dst);
1172 	} else {
1173 		err = PTR_ERR(rt);
1174 
1175 		/* Routing failed... */
1176 		sk->sk_route_caps = 0;
1177 		/*
1178 		 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1179 		 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1180 		 */
1181 		if (!sock_net(sk)->ipv4.sysctl_ip_dynaddr ||
1182 		    sk->sk_state != TCP_SYN_SENT ||
1183 		    (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1184 		    (err = inet_sk_reselect_saddr(sk)) != 0)
1185 			sk->sk_err_soft = -err;
1186 	}
1187 
1188 	return err;
1189 }
1190 EXPORT_SYMBOL(inet_sk_rebuild_header);
1191 
1192 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1193 				 netdev_features_t features)
1194 {
1195 	bool udpfrag = false, fixedid = false, gso_partial, encap;
1196 	struct sk_buff *segs = ERR_PTR(-EINVAL);
1197 	const struct net_offload *ops;
1198 	unsigned int offset = 0;
1199 	struct iphdr *iph;
1200 	int proto, tot_len;
1201 	int nhoff;
1202 	int ihl;
1203 	int id;
1204 
1205 	skb_reset_network_header(skb);
1206 	nhoff = skb_network_header(skb) - skb_mac_header(skb);
1207 	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1208 		goto out;
1209 
1210 	iph = ip_hdr(skb);
1211 	ihl = iph->ihl * 4;
1212 	if (ihl < sizeof(*iph))
1213 		goto out;
1214 
1215 	id = ntohs(iph->id);
1216 	proto = iph->protocol;
1217 
1218 	/* Warning: after this point, iph might be no longer valid */
1219 	if (unlikely(!pskb_may_pull(skb, ihl)))
1220 		goto out;
1221 	__skb_pull(skb, ihl);
1222 
1223 	encap = SKB_GSO_CB(skb)->encap_level > 0;
1224 	if (encap)
1225 		features &= skb->dev->hw_enc_features;
1226 	SKB_GSO_CB(skb)->encap_level += ihl;
1227 
1228 	skb_reset_transport_header(skb);
1229 
1230 	segs = ERR_PTR(-EPROTONOSUPPORT);
1231 
1232 	if (!skb->encapsulation || encap) {
1233 		udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1234 		fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1235 
1236 		/* fixed ID is invalid if DF bit is not set */
1237 		if (fixedid && !(iph->frag_off & htons(IP_DF)))
1238 			goto out;
1239 	}
1240 
1241 	ops = rcu_dereference(inet_offloads[proto]);
1242 	if (likely(ops && ops->callbacks.gso_segment))
1243 		segs = ops->callbacks.gso_segment(skb, features);
1244 
1245 	if (IS_ERR_OR_NULL(segs))
1246 		goto out;
1247 
1248 	gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1249 
1250 	skb = segs;
1251 	do {
1252 		iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1253 		if (udpfrag) {
1254 			iph->frag_off = htons(offset >> 3);
1255 			if (skb->next)
1256 				iph->frag_off |= htons(IP_MF);
1257 			offset += skb->len - nhoff - ihl;
1258 			tot_len = skb->len - nhoff;
1259 		} else if (skb_is_gso(skb)) {
1260 			if (!fixedid) {
1261 				iph->id = htons(id);
1262 				id += skb_shinfo(skb)->gso_segs;
1263 			}
1264 
1265 			if (gso_partial)
1266 				tot_len = skb_shinfo(skb)->gso_size +
1267 					  SKB_GSO_CB(skb)->data_offset +
1268 					  skb->head - (unsigned char *)iph;
1269 			else
1270 				tot_len = skb->len - nhoff;
1271 		} else {
1272 			if (!fixedid)
1273 				iph->id = htons(id++);
1274 			tot_len = skb->len - nhoff;
1275 		}
1276 		iph->tot_len = htons(tot_len);
1277 		ip_send_check(iph);
1278 		if (encap)
1279 			skb_reset_inner_headers(skb);
1280 		skb->network_header = (u8 *)iph - skb->head;
1281 	} while ((skb = skb->next));
1282 
1283 out:
1284 	return segs;
1285 }
1286 EXPORT_SYMBOL(inet_gso_segment);
1287 
1288 struct sk_buff **inet_gro_receive(struct sk_buff **head, struct sk_buff *skb)
1289 {
1290 	const struct net_offload *ops;
1291 	struct sk_buff **pp = NULL;
1292 	struct sk_buff *p;
1293 	const struct iphdr *iph;
1294 	unsigned int hlen;
1295 	unsigned int off;
1296 	unsigned int id;
1297 	int flush = 1;
1298 	int proto;
1299 
1300 	off = skb_gro_offset(skb);
1301 	hlen = off + sizeof(*iph);
1302 	iph = skb_gro_header_fast(skb, off);
1303 	if (skb_gro_header_hard(skb, hlen)) {
1304 		iph = skb_gro_header_slow(skb, hlen, off);
1305 		if (unlikely(!iph))
1306 			goto out;
1307 	}
1308 
1309 	proto = iph->protocol;
1310 
1311 	rcu_read_lock();
1312 	ops = rcu_dereference(inet_offloads[proto]);
1313 	if (!ops || !ops->callbacks.gro_receive)
1314 		goto out_unlock;
1315 
1316 	if (*(u8 *)iph != 0x45)
1317 		goto out_unlock;
1318 
1319 	if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1320 		goto out_unlock;
1321 
1322 	id = ntohl(*(__be32 *)&iph->id);
1323 	flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1324 	id >>= 16;
1325 
1326 	for (p = *head; p; p = p->next) {
1327 		struct iphdr *iph2;
1328 		u16 flush_id;
1329 
1330 		if (!NAPI_GRO_CB(p)->same_flow)
1331 			continue;
1332 
1333 		iph2 = (struct iphdr *)(p->data + off);
1334 		/* The above works because, with the exception of the top
1335 		 * (inner most) layer, we only aggregate pkts with the same
1336 		 * hdr length so all the hdrs we'll need to verify will start
1337 		 * at the same offset.
1338 		 */
1339 		if ((iph->protocol ^ iph2->protocol) |
1340 		    ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1341 		    ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1342 			NAPI_GRO_CB(p)->same_flow = 0;
1343 			continue;
1344 		}
1345 
1346 		/* All fields must match except length and checksum. */
1347 		NAPI_GRO_CB(p)->flush |=
1348 			(iph->ttl ^ iph2->ttl) |
1349 			(iph->tos ^ iph2->tos) |
1350 			((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1351 
1352 		NAPI_GRO_CB(p)->flush |= flush;
1353 
1354 		/* We need to store of the IP ID check to be included later
1355 		 * when we can verify that this packet does in fact belong
1356 		 * to a given flow.
1357 		 */
1358 		flush_id = (u16)(id - ntohs(iph2->id));
1359 
1360 		/* This bit of code makes it much easier for us to identify
1361 		 * the cases where we are doing atomic vs non-atomic IP ID
1362 		 * checks.  Specifically an atomic check can return IP ID
1363 		 * values 0 - 0xFFFF, while a non-atomic check can only
1364 		 * return 0 or 0xFFFF.
1365 		 */
1366 		if (!NAPI_GRO_CB(p)->is_atomic ||
1367 		    !(iph->frag_off & htons(IP_DF))) {
1368 			flush_id ^= NAPI_GRO_CB(p)->count;
1369 			flush_id = flush_id ? 0xFFFF : 0;
1370 		}
1371 
1372 		/* If the previous IP ID value was based on an atomic
1373 		 * datagram we can overwrite the value and ignore it.
1374 		 */
1375 		if (NAPI_GRO_CB(skb)->is_atomic)
1376 			NAPI_GRO_CB(p)->flush_id = flush_id;
1377 		else
1378 			NAPI_GRO_CB(p)->flush_id |= flush_id;
1379 	}
1380 
1381 	NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1382 	NAPI_GRO_CB(skb)->flush |= flush;
1383 	skb_set_network_header(skb, off);
1384 	/* The above will be needed by the transport layer if there is one
1385 	 * immediately following this IP hdr.
1386 	 */
1387 
1388 	/* Note : No need to call skb_gro_postpull_rcsum() here,
1389 	 * as we already checked checksum over ipv4 header was 0
1390 	 */
1391 	skb_gro_pull(skb, sizeof(*iph));
1392 	skb_set_transport_header(skb, skb_gro_offset(skb));
1393 
1394 	pp = ops->callbacks.gro_receive(head, skb);
1395 
1396 out_unlock:
1397 	rcu_read_unlock();
1398 
1399 out:
1400 	NAPI_GRO_CB(skb)->flush |= flush;
1401 
1402 	return pp;
1403 }
1404 EXPORT_SYMBOL(inet_gro_receive);
1405 
1406 static struct sk_buff **ipip_gro_receive(struct sk_buff **head,
1407 					 struct sk_buff *skb)
1408 {
1409 	if (NAPI_GRO_CB(skb)->encap_mark) {
1410 		NAPI_GRO_CB(skb)->flush = 1;
1411 		return NULL;
1412 	}
1413 
1414 	NAPI_GRO_CB(skb)->encap_mark = 1;
1415 
1416 	return inet_gro_receive(head, skb);
1417 }
1418 
1419 #define SECONDS_PER_DAY	86400
1420 
1421 /* inet_current_timestamp - Return IP network timestamp
1422  *
1423  * Return milliseconds since midnight in network byte order.
1424  */
1425 __be32 inet_current_timestamp(void)
1426 {
1427 	u32 secs;
1428 	u32 msecs;
1429 	struct timespec64 ts;
1430 
1431 	ktime_get_real_ts64(&ts);
1432 
1433 	/* Get secs since midnight. */
1434 	(void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1435 	/* Convert to msecs. */
1436 	msecs = secs * MSEC_PER_SEC;
1437 	/* Convert nsec to msec. */
1438 	msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1439 
1440 	/* Convert to network byte order. */
1441 	return htonl(msecs);
1442 }
1443 EXPORT_SYMBOL(inet_current_timestamp);
1444 
1445 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1446 {
1447 	if (sk->sk_family == AF_INET)
1448 		return ip_recv_error(sk, msg, len, addr_len);
1449 #if IS_ENABLED(CONFIG_IPV6)
1450 	if (sk->sk_family == AF_INET6)
1451 		return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1452 #endif
1453 	return -EINVAL;
1454 }
1455 
1456 int inet_gro_complete(struct sk_buff *skb, int nhoff)
1457 {
1458 	__be16 newlen = htons(skb->len - nhoff);
1459 	struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1460 	const struct net_offload *ops;
1461 	int proto = iph->protocol;
1462 	int err = -ENOSYS;
1463 
1464 	if (skb->encapsulation)
1465 		skb_set_inner_network_header(skb, nhoff);
1466 
1467 	csum_replace2(&iph->check, iph->tot_len, newlen);
1468 	iph->tot_len = newlen;
1469 
1470 	rcu_read_lock();
1471 	ops = rcu_dereference(inet_offloads[proto]);
1472 	if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1473 		goto out_unlock;
1474 
1475 	/* Only need to add sizeof(*iph) to get to the next hdr below
1476 	 * because any hdr with option will have been flushed in
1477 	 * inet_gro_receive().
1478 	 */
1479 	err = ops->callbacks.gro_complete(skb, nhoff + sizeof(*iph));
1480 
1481 out_unlock:
1482 	rcu_read_unlock();
1483 
1484 	return err;
1485 }
1486 EXPORT_SYMBOL(inet_gro_complete);
1487 
1488 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1489 {
1490 	skb->encapsulation = 1;
1491 	skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1492 	return inet_gro_complete(skb, nhoff);
1493 }
1494 
1495 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1496 			 unsigned short type, unsigned char protocol,
1497 			 struct net *net)
1498 {
1499 	struct socket *sock;
1500 	int rc = sock_create_kern(net, family, type, protocol, &sock);
1501 
1502 	if (rc == 0) {
1503 		*sk = sock->sk;
1504 		(*sk)->sk_allocation = GFP_ATOMIC;
1505 		/*
1506 		 * Unhash it so that IP input processing does not even see it,
1507 		 * we do not wish this socket to see incoming packets.
1508 		 */
1509 		(*sk)->sk_prot->unhash(*sk);
1510 	}
1511 	return rc;
1512 }
1513 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1514 
1515 u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt)
1516 {
1517 	return  *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt);
1518 }
1519 EXPORT_SYMBOL_GPL(snmp_get_cpu_field);
1520 
1521 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1522 {
1523 	unsigned long res = 0;
1524 	int i;
1525 
1526 	for_each_possible_cpu(i)
1527 		res += snmp_get_cpu_field(mib, i, offt);
1528 	return res;
1529 }
1530 EXPORT_SYMBOL_GPL(snmp_fold_field);
1531 
1532 #if BITS_PER_LONG==32
1533 
1534 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1535 			 size_t syncp_offset)
1536 {
1537 	void *bhptr;
1538 	struct u64_stats_sync *syncp;
1539 	u64 v;
1540 	unsigned int start;
1541 
1542 	bhptr = per_cpu_ptr(mib, cpu);
1543 	syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1544 	do {
1545 		start = u64_stats_fetch_begin_irq(syncp);
1546 		v = *(((u64 *)bhptr) + offt);
1547 	} while (u64_stats_fetch_retry_irq(syncp, start));
1548 
1549 	return v;
1550 }
1551 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1552 
1553 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1554 {
1555 	u64 res = 0;
1556 	int cpu;
1557 
1558 	for_each_possible_cpu(cpu) {
1559 		res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1560 	}
1561 	return res;
1562 }
1563 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1564 #endif
1565 
1566 #ifdef CONFIG_IP_MULTICAST
1567 static const struct net_protocol igmp_protocol = {
1568 	.handler =	igmp_rcv,
1569 	.netns_ok =	1,
1570 };
1571 #endif
1572 
1573 static const struct net_protocol tcp_protocol = {
1574 	.early_demux	=	tcp_v4_early_demux,
1575 	.handler	=	tcp_v4_rcv,
1576 	.err_handler	=	tcp_v4_err,
1577 	.no_policy	=	1,
1578 	.netns_ok	=	1,
1579 	.icmp_strict_tag_validation = 1,
1580 };
1581 
1582 static const struct net_protocol udp_protocol = {
1583 	.early_demux =	udp_v4_early_demux,
1584 	.handler =	udp_rcv,
1585 	.err_handler =	udp_err,
1586 	.no_policy =	1,
1587 	.netns_ok =	1,
1588 };
1589 
1590 static const struct net_protocol icmp_protocol = {
1591 	.handler =	icmp_rcv,
1592 	.err_handler =	icmp_err,
1593 	.no_policy =	1,
1594 	.netns_ok =	1,
1595 };
1596 
1597 static __net_init int ipv4_mib_init_net(struct net *net)
1598 {
1599 	int i;
1600 
1601 	net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1602 	if (!net->mib.tcp_statistics)
1603 		goto err_tcp_mib;
1604 	net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1605 	if (!net->mib.ip_statistics)
1606 		goto err_ip_mib;
1607 
1608 	for_each_possible_cpu(i) {
1609 		struct ipstats_mib *af_inet_stats;
1610 		af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1611 		u64_stats_init(&af_inet_stats->syncp);
1612 	}
1613 
1614 	net->mib.net_statistics = alloc_percpu(struct linux_mib);
1615 	if (!net->mib.net_statistics)
1616 		goto err_net_mib;
1617 	net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1618 	if (!net->mib.udp_statistics)
1619 		goto err_udp_mib;
1620 	net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1621 	if (!net->mib.udplite_statistics)
1622 		goto err_udplite_mib;
1623 	net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1624 	if (!net->mib.icmp_statistics)
1625 		goto err_icmp_mib;
1626 	net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1627 					      GFP_KERNEL);
1628 	if (!net->mib.icmpmsg_statistics)
1629 		goto err_icmpmsg_mib;
1630 
1631 	tcp_mib_init(net);
1632 	return 0;
1633 
1634 err_icmpmsg_mib:
1635 	free_percpu(net->mib.icmp_statistics);
1636 err_icmp_mib:
1637 	free_percpu(net->mib.udplite_statistics);
1638 err_udplite_mib:
1639 	free_percpu(net->mib.udp_statistics);
1640 err_udp_mib:
1641 	free_percpu(net->mib.net_statistics);
1642 err_net_mib:
1643 	free_percpu(net->mib.ip_statistics);
1644 err_ip_mib:
1645 	free_percpu(net->mib.tcp_statistics);
1646 err_tcp_mib:
1647 	return -ENOMEM;
1648 }
1649 
1650 static __net_exit void ipv4_mib_exit_net(struct net *net)
1651 {
1652 	kfree(net->mib.icmpmsg_statistics);
1653 	free_percpu(net->mib.icmp_statistics);
1654 	free_percpu(net->mib.udplite_statistics);
1655 	free_percpu(net->mib.udp_statistics);
1656 	free_percpu(net->mib.net_statistics);
1657 	free_percpu(net->mib.ip_statistics);
1658 	free_percpu(net->mib.tcp_statistics);
1659 }
1660 
1661 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1662 	.init = ipv4_mib_init_net,
1663 	.exit = ipv4_mib_exit_net,
1664 };
1665 
1666 static int __init init_ipv4_mibs(void)
1667 {
1668 	return register_pernet_subsys(&ipv4_mib_ops);
1669 }
1670 
1671 static __net_init int inet_init_net(struct net *net)
1672 {
1673 	/*
1674 	 * Set defaults for local port range
1675 	 */
1676 	seqlock_init(&net->ipv4.ip_local_ports.lock);
1677 	net->ipv4.ip_local_ports.range[0] =  32768;
1678 	net->ipv4.ip_local_ports.range[1] =  60999;
1679 
1680 	seqlock_init(&net->ipv4.ping_group_range.lock);
1681 	/*
1682 	 * Sane defaults - nobody may create ping sockets.
1683 	 * Boot scripts should set this to distro-specific group.
1684 	 */
1685 	net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1686 	net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1687 
1688 	/* Default values for sysctl-controlled parameters.
1689 	 * We set them here, in case sysctl is not compiled.
1690 	 */
1691 	net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1692 	net->ipv4.sysctl_ip_dynaddr = 0;
1693 	net->ipv4.sysctl_ip_early_demux = 1;
1694 
1695 	return 0;
1696 }
1697 
1698 static __net_exit void inet_exit_net(struct net *net)
1699 {
1700 }
1701 
1702 static __net_initdata struct pernet_operations af_inet_ops = {
1703 	.init = inet_init_net,
1704 	.exit = inet_exit_net,
1705 };
1706 
1707 static int __init init_inet_pernet_ops(void)
1708 {
1709 	return register_pernet_subsys(&af_inet_ops);
1710 }
1711 
1712 static int ipv4_proc_init(void);
1713 
1714 /*
1715  *	IP protocol layer initialiser
1716  */
1717 
1718 static struct packet_offload ip_packet_offload __read_mostly = {
1719 	.type = cpu_to_be16(ETH_P_IP),
1720 	.callbacks = {
1721 		.gso_segment = inet_gso_segment,
1722 		.gro_receive = inet_gro_receive,
1723 		.gro_complete = inet_gro_complete,
1724 	},
1725 };
1726 
1727 static const struct net_offload ipip_offload = {
1728 	.callbacks = {
1729 		.gso_segment	= inet_gso_segment,
1730 		.gro_receive	= ipip_gro_receive,
1731 		.gro_complete	= ipip_gro_complete,
1732 	},
1733 };
1734 
1735 static int __init ipv4_offload_init(void)
1736 {
1737 	/*
1738 	 * Add offloads
1739 	 */
1740 	if (udpv4_offload_init() < 0)
1741 		pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1742 	if (tcpv4_offload_init() < 0)
1743 		pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1744 
1745 	dev_add_offload(&ip_packet_offload);
1746 	inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1747 	return 0;
1748 }
1749 
1750 fs_initcall(ipv4_offload_init);
1751 
1752 static struct packet_type ip_packet_type __read_mostly = {
1753 	.type = cpu_to_be16(ETH_P_IP),
1754 	.func = ip_rcv,
1755 };
1756 
1757 static int __init inet_init(void)
1758 {
1759 	struct inet_protosw *q;
1760 	struct list_head *r;
1761 	int rc = -EINVAL;
1762 
1763 	sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1764 
1765 	rc = proto_register(&tcp_prot, 1);
1766 	if (rc)
1767 		goto out;
1768 
1769 	rc = proto_register(&udp_prot, 1);
1770 	if (rc)
1771 		goto out_unregister_tcp_proto;
1772 
1773 	rc = proto_register(&raw_prot, 1);
1774 	if (rc)
1775 		goto out_unregister_udp_proto;
1776 
1777 	rc = proto_register(&ping_prot, 1);
1778 	if (rc)
1779 		goto out_unregister_raw_proto;
1780 
1781 	/*
1782 	 *	Tell SOCKET that we are alive...
1783 	 */
1784 
1785 	(void)sock_register(&inet_family_ops);
1786 
1787 #ifdef CONFIG_SYSCTL
1788 	ip_static_sysctl_init();
1789 #endif
1790 
1791 	/*
1792 	 *	Add all the base protocols.
1793 	 */
1794 
1795 	if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1796 		pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1797 	if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1798 		pr_crit("%s: Cannot add UDP protocol\n", __func__);
1799 	if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1800 		pr_crit("%s: Cannot add TCP protocol\n", __func__);
1801 #ifdef CONFIG_IP_MULTICAST
1802 	if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1803 		pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1804 #endif
1805 
1806 	/* Register the socket-side information for inet_create. */
1807 	for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1808 		INIT_LIST_HEAD(r);
1809 
1810 	for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1811 		inet_register_protosw(q);
1812 
1813 	/*
1814 	 *	Set the ARP module up
1815 	 */
1816 
1817 	arp_init();
1818 
1819 	/*
1820 	 *	Set the IP module up
1821 	 */
1822 
1823 	ip_init();
1824 
1825 	tcp_v4_init();
1826 
1827 	/* Setup TCP slab cache for open requests. */
1828 	tcp_init();
1829 
1830 	/* Setup UDP memory threshold */
1831 	udp_init();
1832 
1833 	/* Add UDP-Lite (RFC 3828) */
1834 	udplite4_register();
1835 
1836 	ping_init();
1837 
1838 	/*
1839 	 *	Set the ICMP layer up
1840 	 */
1841 
1842 	if (icmp_init() < 0)
1843 		panic("Failed to create the ICMP control socket.\n");
1844 
1845 	/*
1846 	 *	Initialise the multicast router
1847 	 */
1848 #if defined(CONFIG_IP_MROUTE)
1849 	if (ip_mr_init())
1850 		pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1851 #endif
1852 
1853 	if (init_inet_pernet_ops())
1854 		pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
1855 	/*
1856 	 *	Initialise per-cpu ipv4 mibs
1857 	 */
1858 
1859 	if (init_ipv4_mibs())
1860 		pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1861 
1862 	ipv4_proc_init();
1863 
1864 	ipfrag_init();
1865 
1866 	dev_add_pack(&ip_packet_type);
1867 
1868 	ip_tunnel_core_init();
1869 
1870 	rc = 0;
1871 out:
1872 	return rc;
1873 out_unregister_raw_proto:
1874 	proto_unregister(&raw_prot);
1875 out_unregister_udp_proto:
1876 	proto_unregister(&udp_prot);
1877 out_unregister_tcp_proto:
1878 	proto_unregister(&tcp_prot);
1879 	goto out;
1880 }
1881 
1882 fs_initcall(inet_init);
1883 
1884 /* ------------------------------------------------------------------------ */
1885 
1886 #ifdef CONFIG_PROC_FS
1887 static int __init ipv4_proc_init(void)
1888 {
1889 	int rc = 0;
1890 
1891 	if (raw_proc_init())
1892 		goto out_raw;
1893 	if (tcp4_proc_init())
1894 		goto out_tcp;
1895 	if (udp4_proc_init())
1896 		goto out_udp;
1897 	if (ping_proc_init())
1898 		goto out_ping;
1899 	if (ip_misc_proc_init())
1900 		goto out_misc;
1901 out:
1902 	return rc;
1903 out_misc:
1904 	ping_proc_exit();
1905 out_ping:
1906 	udp4_proc_exit();
1907 out_udp:
1908 	tcp4_proc_exit();
1909 out_tcp:
1910 	raw_proc_exit();
1911 out_raw:
1912 	rc = -ENOMEM;
1913 	goto out;
1914 }
1915 
1916 #else /* CONFIG_PROC_FS */
1917 static int __init ipv4_proc_init(void)
1918 {
1919 	return 0;
1920 }
1921 #endif /* CONFIG_PROC_FS */
1922