xref: /linux/net/ipv4/af_inet.c (revision 3b27d13940c3710a1128527c43719cb0bb05d73b)
1 /*
2  * INET		An implementation of the TCP/IP protocol suite for the LINUX
3  *		operating system.  INET is implemented using the  BSD Socket
4  *		interface as the means of communication with the user level.
5  *
6  *		PF_INET protocol family socket handler.
7  *
8  * Authors:	Ross Biro
9  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10  *		Florian La Roche, <flla@stud.uni-sb.de>
11  *		Alan Cox, <A.Cox@swansea.ac.uk>
12  *
13  * Changes (see also sock.c)
14  *
15  *		piggy,
16  *		Karl Knutson	:	Socket protocol table
17  *		A.N.Kuznetsov	:	Socket death error in accept().
18  *		John Richardson :	Fix non blocking error in connect()
19  *					so sockets that fail to connect
20  *					don't return -EINPROGRESS.
21  *		Alan Cox	:	Asynchronous I/O support
22  *		Alan Cox	:	Keep correct socket pointer on sock
23  *					structures
24  *					when accept() ed
25  *		Alan Cox	:	Semantics of SO_LINGER aren't state
26  *					moved to close when you look carefully.
27  *					With this fixed and the accept bug fixed
28  *					some RPC stuff seems happier.
29  *		Niibe Yutaka	:	4.4BSD style write async I/O
30  *		Alan Cox,
31  *		Tony Gale 	:	Fixed reuse semantics.
32  *		Alan Cox	:	bind() shouldn't abort existing but dead
33  *					sockets. Stops FTP netin:.. I hope.
34  *		Alan Cox	:	bind() works correctly for RAW sockets.
35  *					Note that FreeBSD at least was broken
36  *					in this respect so be careful with
37  *					compatibility tests...
38  *		Alan Cox	:	routing cache support
39  *		Alan Cox	:	memzero the socket structure for
40  *					compactness.
41  *		Matt Day	:	nonblock connect error handler
42  *		Alan Cox	:	Allow large numbers of pending sockets
43  *					(eg for big web sites), but only if
44  *					specifically application requested.
45  *		Alan Cox	:	New buffering throughout IP. Used
46  *					dumbly.
47  *		Alan Cox	:	New buffering now used smartly.
48  *		Alan Cox	:	BSD rather than common sense
49  *					interpretation of listen.
50  *		Germano Caronni	:	Assorted small races.
51  *		Alan Cox	:	sendmsg/recvmsg basic support.
52  *		Alan Cox	:	Only sendmsg/recvmsg now supported.
53  *		Alan Cox	:	Locked down bind (see security list).
54  *		Alan Cox	:	Loosened bind a little.
55  *		Mike McLagan	:	ADD/DEL DLCI Ioctls
56  *	Willy Konynenberg	:	Transparent proxying support.
57  *		David S. Miller	:	New socket lookup architecture.
58  *					Some other random speedups.
59  *		Cyrus Durgin	:	Cleaned up file for kmod hacks.
60  *		Andi Kleen	:	Fix inet_stream_connect TCP race.
61  *
62  *		This program is free software; you can redistribute it and/or
63  *		modify it under the terms of the GNU General Public License
64  *		as published by the Free Software Foundation; either version
65  *		2 of the License, or (at your option) any later version.
66  */
67 
68 #define pr_fmt(fmt) "IPv4: " fmt
69 
70 #include <linux/err.h>
71 #include <linux/errno.h>
72 #include <linux/types.h>
73 #include <linux/socket.h>
74 #include <linux/in.h>
75 #include <linux/kernel.h>
76 #include <linux/module.h>
77 #include <linux/sched.h>
78 #include <linux/timer.h>
79 #include <linux/string.h>
80 #include <linux/sockios.h>
81 #include <linux/net.h>
82 #include <linux/capability.h>
83 #include <linux/fcntl.h>
84 #include <linux/mm.h>
85 #include <linux/interrupt.h>
86 #include <linux/stat.h>
87 #include <linux/init.h>
88 #include <linux/poll.h>
89 #include <linux/netfilter_ipv4.h>
90 #include <linux/random.h>
91 #include <linux/slab.h>
92 
93 #include <asm/uaccess.h>
94 
95 #include <linux/inet.h>
96 #include <linux/igmp.h>
97 #include <linux/inetdevice.h>
98 #include <linux/netdevice.h>
99 #include <net/checksum.h>
100 #include <net/ip.h>
101 #include <net/protocol.h>
102 #include <net/arp.h>
103 #include <net/route.h>
104 #include <net/ip_fib.h>
105 #include <net/inet_connection_sock.h>
106 #include <net/tcp.h>
107 #include <net/udp.h>
108 #include <net/udplite.h>
109 #include <net/ping.h>
110 #include <linux/skbuff.h>
111 #include <net/sock.h>
112 #include <net/raw.h>
113 #include <net/icmp.h>
114 #include <net/inet_common.h>
115 #include <net/xfrm.h>
116 #include <net/net_namespace.h>
117 #include <net/secure_seq.h>
118 #ifdef CONFIG_IP_MROUTE
119 #include <linux/mroute.h>
120 #endif
121 
122 
123 /* The inetsw table contains everything that inet_create needs to
124  * build a new socket.
125  */
126 static struct list_head inetsw[SOCK_MAX];
127 static DEFINE_SPINLOCK(inetsw_lock);
128 
129 /* New destruction routine */
130 
131 void inet_sock_destruct(struct sock *sk)
132 {
133 	struct inet_sock *inet = inet_sk(sk);
134 
135 	__skb_queue_purge(&sk->sk_receive_queue);
136 	__skb_queue_purge(&sk->sk_error_queue);
137 
138 	sk_mem_reclaim(sk);
139 
140 	if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
141 		pr_err("Attempt to release TCP socket in state %d %p\n",
142 		       sk->sk_state, sk);
143 		return;
144 	}
145 	if (!sock_flag(sk, SOCK_DEAD)) {
146 		pr_err("Attempt to release alive inet socket %p\n", sk);
147 		return;
148 	}
149 
150 	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
151 	WARN_ON(atomic_read(&sk->sk_wmem_alloc));
152 	WARN_ON(sk->sk_wmem_queued);
153 	WARN_ON(sk->sk_forward_alloc);
154 
155 	kfree(rcu_dereference_protected(inet->inet_opt, 1));
156 	dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
157 	dst_release(sk->sk_rx_dst);
158 	sk_refcnt_debug_dec(sk);
159 }
160 EXPORT_SYMBOL(inet_sock_destruct);
161 
162 /*
163  *	The routines beyond this point handle the behaviour of an AF_INET
164  *	socket object. Mostly it punts to the subprotocols of IP to do
165  *	the work.
166  */
167 
168 /*
169  *	Automatically bind an unbound socket.
170  */
171 
172 static int inet_autobind(struct sock *sk)
173 {
174 	struct inet_sock *inet;
175 	/* We may need to bind the socket. */
176 	lock_sock(sk);
177 	inet = inet_sk(sk);
178 	if (!inet->inet_num) {
179 		if (sk->sk_prot->get_port(sk, 0)) {
180 			release_sock(sk);
181 			return -EAGAIN;
182 		}
183 		inet->inet_sport = htons(inet->inet_num);
184 	}
185 	release_sock(sk);
186 	return 0;
187 }
188 
189 /*
190  *	Move a socket into listening state.
191  */
192 int inet_listen(struct socket *sock, int backlog)
193 {
194 	struct sock *sk = sock->sk;
195 	unsigned char old_state;
196 	int err;
197 
198 	lock_sock(sk);
199 
200 	err = -EINVAL;
201 	if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
202 		goto out;
203 
204 	old_state = sk->sk_state;
205 	if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
206 		goto out;
207 
208 	/* Really, if the socket is already in listen state
209 	 * we can only allow the backlog to be adjusted.
210 	 */
211 	if (old_state != TCP_LISTEN) {
212 		/* Check special setups for testing purpose to enable TFO w/o
213 		 * requiring TCP_FASTOPEN sockopt.
214 		 * Note that only TCP sockets (SOCK_STREAM) will reach here.
215 		 * Also fastopenq may already been allocated because this
216 		 * socket was in TCP_LISTEN state previously but was
217 		 * shutdown() (rather than close()).
218 		 */
219 		if ((sysctl_tcp_fastopen & TFO_SERVER_ENABLE) != 0 &&
220 		    !inet_csk(sk)->icsk_accept_queue.fastopenq) {
221 			if ((sysctl_tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) != 0)
222 				err = fastopen_init_queue(sk, backlog);
223 			else if ((sysctl_tcp_fastopen &
224 				  TFO_SERVER_WO_SOCKOPT2) != 0)
225 				err = fastopen_init_queue(sk,
226 				    ((uint)sysctl_tcp_fastopen) >> 16);
227 			else
228 				err = 0;
229 			if (err)
230 				goto out;
231 
232 			tcp_fastopen_init_key_once(true);
233 		}
234 		err = inet_csk_listen_start(sk, backlog);
235 		if (err)
236 			goto out;
237 	}
238 	sk->sk_max_ack_backlog = backlog;
239 	err = 0;
240 
241 out:
242 	release_sock(sk);
243 	return err;
244 }
245 EXPORT_SYMBOL(inet_listen);
246 
247 /*
248  *	Create an inet socket.
249  */
250 
251 static int inet_create(struct net *net, struct socket *sock, int protocol,
252 		       int kern)
253 {
254 	struct sock *sk;
255 	struct inet_protosw *answer;
256 	struct inet_sock *inet;
257 	struct proto *answer_prot;
258 	unsigned char answer_flags;
259 	int try_loading_module = 0;
260 	int err;
261 
262 	sock->state = SS_UNCONNECTED;
263 
264 	/* Look for the requested type/protocol pair. */
265 lookup_protocol:
266 	err = -ESOCKTNOSUPPORT;
267 	rcu_read_lock();
268 	list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
269 
270 		err = 0;
271 		/* Check the non-wild match. */
272 		if (protocol == answer->protocol) {
273 			if (protocol != IPPROTO_IP)
274 				break;
275 		} else {
276 			/* Check for the two wild cases. */
277 			if (IPPROTO_IP == protocol) {
278 				protocol = answer->protocol;
279 				break;
280 			}
281 			if (IPPROTO_IP == answer->protocol)
282 				break;
283 		}
284 		err = -EPROTONOSUPPORT;
285 	}
286 
287 	if (unlikely(err)) {
288 		if (try_loading_module < 2) {
289 			rcu_read_unlock();
290 			/*
291 			 * Be more specific, e.g. net-pf-2-proto-132-type-1
292 			 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
293 			 */
294 			if (++try_loading_module == 1)
295 				request_module("net-pf-%d-proto-%d-type-%d",
296 					       PF_INET, protocol, sock->type);
297 			/*
298 			 * Fall back to generic, e.g. net-pf-2-proto-132
299 			 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
300 			 */
301 			else
302 				request_module("net-pf-%d-proto-%d",
303 					       PF_INET, protocol);
304 			goto lookup_protocol;
305 		} else
306 			goto out_rcu_unlock;
307 	}
308 
309 	err = -EPERM;
310 	if (sock->type == SOCK_RAW && !kern &&
311 	    !ns_capable(net->user_ns, CAP_NET_RAW))
312 		goto out_rcu_unlock;
313 
314 	sock->ops = answer->ops;
315 	answer_prot = answer->prot;
316 	answer_flags = answer->flags;
317 	rcu_read_unlock();
318 
319 	WARN_ON(!answer_prot->slab);
320 
321 	err = -ENOBUFS;
322 	sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
323 	if (!sk)
324 		goto out;
325 
326 	err = 0;
327 	if (INET_PROTOSW_REUSE & answer_flags)
328 		sk->sk_reuse = SK_CAN_REUSE;
329 
330 	inet = inet_sk(sk);
331 	inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
332 
333 	inet->nodefrag = 0;
334 
335 	if (SOCK_RAW == sock->type) {
336 		inet->inet_num = protocol;
337 		if (IPPROTO_RAW == protocol)
338 			inet->hdrincl = 1;
339 	}
340 
341 	if (net->ipv4.sysctl_ip_no_pmtu_disc)
342 		inet->pmtudisc = IP_PMTUDISC_DONT;
343 	else
344 		inet->pmtudisc = IP_PMTUDISC_WANT;
345 
346 	inet->inet_id = 0;
347 
348 	sock_init_data(sock, sk);
349 
350 	sk->sk_destruct	   = inet_sock_destruct;
351 	sk->sk_protocol	   = protocol;
352 	sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
353 
354 	inet->uc_ttl	= -1;
355 	inet->mc_loop	= 1;
356 	inet->mc_ttl	= 1;
357 	inet->mc_all	= 1;
358 	inet->mc_index	= 0;
359 	inet->mc_list	= NULL;
360 	inet->rcv_tos	= 0;
361 
362 	sk_refcnt_debug_inc(sk);
363 
364 	if (inet->inet_num) {
365 		/* It assumes that any protocol which allows
366 		 * the user to assign a number at socket
367 		 * creation time automatically
368 		 * shares.
369 		 */
370 		inet->inet_sport = htons(inet->inet_num);
371 		/* Add to protocol hash chains. */
372 		sk->sk_prot->hash(sk);
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 	int err;
430 
431 	/* If the socket has its own bind function then use it. (RAW) */
432 	if (sk->sk_prot->bind) {
433 		err = sk->sk_prot->bind(sk, uaddr, addr_len);
434 		goto out;
435 	}
436 	err = -EINVAL;
437 	if (addr_len < sizeof(struct sockaddr_in))
438 		goto out;
439 
440 	if (addr->sin_family != AF_INET) {
441 		/* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
442 		 * only if s_addr is INADDR_ANY.
443 		 */
444 		err = -EAFNOSUPPORT;
445 		if (addr->sin_family != AF_UNSPEC ||
446 		    addr->sin_addr.s_addr != htonl(INADDR_ANY))
447 			goto out;
448 	}
449 
450 	chk_addr_ret = inet_addr_type(net, addr->sin_addr.s_addr);
451 
452 	/* Not specified by any standard per-se, however it breaks too
453 	 * many applications when removed.  It is unfortunate since
454 	 * allowing applications to make a non-local bind solves
455 	 * several problems with systems using dynamic addressing.
456 	 * (ie. your servers still start up even if your ISDN link
457 	 *  is temporarily down)
458 	 */
459 	err = -EADDRNOTAVAIL;
460 	if (!net->ipv4.sysctl_ip_nonlocal_bind &&
461 	    !(inet->freebind || inet->transparent) &&
462 	    addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
463 	    chk_addr_ret != RTN_LOCAL &&
464 	    chk_addr_ret != RTN_MULTICAST &&
465 	    chk_addr_ret != RTN_BROADCAST)
466 		goto out;
467 
468 	snum = ntohs(addr->sin_port);
469 	err = -EACCES;
470 	if (snum && snum < PROT_SOCK &&
471 	    !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
472 		goto out;
473 
474 	/*      We keep a pair of addresses. rcv_saddr is the one
475 	 *      used by hash lookups, and saddr is used for transmit.
476 	 *
477 	 *      In the BSD API these are the same except where it
478 	 *      would be illegal to use them (multicast/broadcast) in
479 	 *      which case the sending device address is used.
480 	 */
481 	lock_sock(sk);
482 
483 	/* Check these errors (active socket, double bind). */
484 	err = -EINVAL;
485 	if (sk->sk_state != TCP_CLOSE || inet->inet_num)
486 		goto out_release_sock;
487 
488 	inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
489 	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
490 		inet->inet_saddr = 0;  /* Use device */
491 
492 	/* Make sure we are allowed to bind here. */
493 	if ((snum || !inet->bind_address_no_port) &&
494 	    sk->sk_prot->get_port(sk, snum)) {
495 		inet->inet_saddr = inet->inet_rcv_saddr = 0;
496 		err = -EADDRINUSE;
497 		goto out_release_sock;
498 	}
499 
500 	if (inet->inet_rcv_saddr)
501 		sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
502 	if (snum)
503 		sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
504 	inet->inet_sport = htons(inet->inet_num);
505 	inet->inet_daddr = 0;
506 	inet->inet_dport = 0;
507 	sk_dst_reset(sk);
508 	err = 0;
509 out_release_sock:
510 	release_sock(sk);
511 out:
512 	return err;
513 }
514 EXPORT_SYMBOL(inet_bind);
515 
516 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
517 		       int addr_len, int flags)
518 {
519 	struct sock *sk = sock->sk;
520 
521 	if (addr_len < sizeof(uaddr->sa_family))
522 		return -EINVAL;
523 	if (uaddr->sa_family == AF_UNSPEC)
524 		return sk->sk_prot->disconnect(sk, flags);
525 
526 	if (!inet_sk(sk)->inet_num && inet_autobind(sk))
527 		return -EAGAIN;
528 	return sk->sk_prot->connect(sk, uaddr, addr_len);
529 }
530 EXPORT_SYMBOL(inet_dgram_connect);
531 
532 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
533 {
534 	DEFINE_WAIT(wait);
535 
536 	prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
537 	sk->sk_write_pending += writebias;
538 
539 	/* Basic assumption: if someone sets sk->sk_err, he _must_
540 	 * change state of the socket from TCP_SYN_*.
541 	 * Connect() does not allow to get error notifications
542 	 * without closing the socket.
543 	 */
544 	while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
545 		release_sock(sk);
546 		timeo = schedule_timeout(timeo);
547 		lock_sock(sk);
548 		if (signal_pending(current) || !timeo)
549 			break;
550 		prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
551 	}
552 	finish_wait(sk_sleep(sk), &wait);
553 	sk->sk_write_pending -= writebias;
554 	return timeo;
555 }
556 
557 /*
558  *	Connect to a remote host. There is regrettably still a little
559  *	TCP 'magic' in here.
560  */
561 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
562 			  int addr_len, int flags)
563 {
564 	struct sock *sk = sock->sk;
565 	int err;
566 	long timeo;
567 
568 	if (addr_len < sizeof(uaddr->sa_family))
569 		return -EINVAL;
570 
571 	if (uaddr->sa_family == AF_UNSPEC) {
572 		err = sk->sk_prot->disconnect(sk, flags);
573 		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
574 		goto out;
575 	}
576 
577 	switch (sock->state) {
578 	default:
579 		err = -EINVAL;
580 		goto out;
581 	case SS_CONNECTED:
582 		err = -EISCONN;
583 		goto out;
584 	case SS_CONNECTING:
585 		err = -EALREADY;
586 		/* Fall out of switch with err, set for this state */
587 		break;
588 	case SS_UNCONNECTED:
589 		err = -EISCONN;
590 		if (sk->sk_state != TCP_CLOSE)
591 			goto out;
592 
593 		err = sk->sk_prot->connect(sk, uaddr, addr_len);
594 		if (err < 0)
595 			goto out;
596 
597 		sock->state = SS_CONNECTING;
598 
599 		/* Just entered SS_CONNECTING state; the only
600 		 * difference is that return value in non-blocking
601 		 * case is EINPROGRESS, rather than EALREADY.
602 		 */
603 		err = -EINPROGRESS;
604 		break;
605 	}
606 
607 	timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
608 
609 	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
610 		int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
611 				tcp_sk(sk)->fastopen_req &&
612 				tcp_sk(sk)->fastopen_req->data ? 1 : 0;
613 
614 		/* Error code is set above */
615 		if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
616 			goto out;
617 
618 		err = sock_intr_errno(timeo);
619 		if (signal_pending(current))
620 			goto out;
621 	}
622 
623 	/* Connection was closed by RST, timeout, ICMP error
624 	 * or another process disconnected us.
625 	 */
626 	if (sk->sk_state == TCP_CLOSE)
627 		goto sock_error;
628 
629 	/* sk->sk_err may be not zero now, if RECVERR was ordered by user
630 	 * and error was received after socket entered established state.
631 	 * Hence, it is handled normally after connect() return successfully.
632 	 */
633 
634 	sock->state = SS_CONNECTED;
635 	err = 0;
636 out:
637 	return err;
638 
639 sock_error:
640 	err = sock_error(sk) ? : -ECONNABORTED;
641 	sock->state = SS_UNCONNECTED;
642 	if (sk->sk_prot->disconnect(sk, flags))
643 		sock->state = SS_DISCONNECTING;
644 	goto out;
645 }
646 EXPORT_SYMBOL(__inet_stream_connect);
647 
648 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
649 			int addr_len, int flags)
650 {
651 	int err;
652 
653 	lock_sock(sock->sk);
654 	err = __inet_stream_connect(sock, uaddr, addr_len, flags);
655 	release_sock(sock->sk);
656 	return err;
657 }
658 EXPORT_SYMBOL(inet_stream_connect);
659 
660 /*
661  *	Accept a pending connection. The TCP layer now gives BSD semantics.
662  */
663 
664 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
665 {
666 	struct sock *sk1 = sock->sk;
667 	int err = -EINVAL;
668 	struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
669 
670 	if (!sk2)
671 		goto do_err;
672 
673 	lock_sock(sk2);
674 
675 	sock_rps_record_flow(sk2);
676 	WARN_ON(!((1 << sk2->sk_state) &
677 		  (TCPF_ESTABLISHED | TCPF_SYN_RECV |
678 		  TCPF_CLOSE_WAIT | TCPF_CLOSE)));
679 
680 	sock_graft(sk2, newsock);
681 
682 	newsock->state = SS_CONNECTED;
683 	err = 0;
684 	release_sock(sk2);
685 do_err:
686 	return err;
687 }
688 EXPORT_SYMBOL(inet_accept);
689 
690 
691 /*
692  *	This does both peername and sockname.
693  */
694 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
695 			int *uaddr_len, int peer)
696 {
697 	struct sock *sk		= sock->sk;
698 	struct inet_sock *inet	= inet_sk(sk);
699 	DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
700 
701 	sin->sin_family = AF_INET;
702 	if (peer) {
703 		if (!inet->inet_dport ||
704 		    (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
705 		     peer == 1))
706 			return -ENOTCONN;
707 		sin->sin_port = inet->inet_dport;
708 		sin->sin_addr.s_addr = inet->inet_daddr;
709 	} else {
710 		__be32 addr = inet->inet_rcv_saddr;
711 		if (!addr)
712 			addr = inet->inet_saddr;
713 		sin->sin_port = inet->inet_sport;
714 		sin->sin_addr.s_addr = addr;
715 	}
716 	memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
717 	*uaddr_len = sizeof(*sin);
718 	return 0;
719 }
720 EXPORT_SYMBOL(inet_getname);
721 
722 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
723 {
724 	struct sock *sk = sock->sk;
725 
726 	sock_rps_record_flow(sk);
727 
728 	/* We may need to bind the socket. */
729 	if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
730 	    inet_autobind(sk))
731 		return -EAGAIN;
732 
733 	return sk->sk_prot->sendmsg(sk, msg, size);
734 }
735 EXPORT_SYMBOL(inet_sendmsg);
736 
737 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
738 		      size_t size, int flags)
739 {
740 	struct sock *sk = sock->sk;
741 
742 	sock_rps_record_flow(sk);
743 
744 	/* We may need to bind the socket. */
745 	if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
746 	    inet_autobind(sk))
747 		return -EAGAIN;
748 
749 	if (sk->sk_prot->sendpage)
750 		return sk->sk_prot->sendpage(sk, page, offset, size, flags);
751 	return sock_no_sendpage(sock, page, offset, size, flags);
752 }
753 EXPORT_SYMBOL(inet_sendpage);
754 
755 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
756 		 int flags)
757 {
758 	struct sock *sk = sock->sk;
759 	int addr_len = 0;
760 	int err;
761 
762 	sock_rps_record_flow(sk);
763 
764 	err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
765 				   flags & ~MSG_DONTWAIT, &addr_len);
766 	if (err >= 0)
767 		msg->msg_namelen = addr_len;
768 	return err;
769 }
770 EXPORT_SYMBOL(inet_recvmsg);
771 
772 int inet_shutdown(struct socket *sock, int how)
773 {
774 	struct sock *sk = sock->sk;
775 	int err = 0;
776 
777 	/* This should really check to make sure
778 	 * the socket is a TCP socket. (WHY AC...)
779 	 */
780 	how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
781 		       1->2 bit 2 snds.
782 		       2->3 */
783 	if ((how & ~SHUTDOWN_MASK) || !how)	/* MAXINT->0 */
784 		return -EINVAL;
785 
786 	lock_sock(sk);
787 	if (sock->state == SS_CONNECTING) {
788 		if ((1 << sk->sk_state) &
789 		    (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
790 			sock->state = SS_DISCONNECTING;
791 		else
792 			sock->state = SS_CONNECTED;
793 	}
794 
795 	switch (sk->sk_state) {
796 	case TCP_CLOSE:
797 		err = -ENOTCONN;
798 		/* Hack to wake up other listeners, who can poll for
799 		   POLLHUP, even on eg. unconnected UDP sockets -- RR */
800 	default:
801 		sk->sk_shutdown |= how;
802 		if (sk->sk_prot->shutdown)
803 			sk->sk_prot->shutdown(sk, how);
804 		break;
805 
806 	/* Remaining two branches are temporary solution for missing
807 	 * close() in multithreaded environment. It is _not_ a good idea,
808 	 * but we have no choice until close() is repaired at VFS level.
809 	 */
810 	case TCP_LISTEN:
811 		if (!(how & RCV_SHUTDOWN))
812 			break;
813 		/* Fall through */
814 	case TCP_SYN_SENT:
815 		err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
816 		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
817 		break;
818 	}
819 
820 	/* Wake up anyone sleeping in poll. */
821 	sk->sk_state_change(sk);
822 	release_sock(sk);
823 	return err;
824 }
825 EXPORT_SYMBOL(inet_shutdown);
826 
827 /*
828  *	ioctl() calls you can issue on an INET socket. Most of these are
829  *	device configuration and stuff and very rarely used. Some ioctls
830  *	pass on to the socket itself.
831  *
832  *	NOTE: I like the idea of a module for the config stuff. ie ifconfig
833  *	loads the devconfigure module does its configuring and unloads it.
834  *	There's a good 20K of config code hanging around the kernel.
835  */
836 
837 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
838 {
839 	struct sock *sk = sock->sk;
840 	int err = 0;
841 	struct net *net = sock_net(sk);
842 
843 	switch (cmd) {
844 	case SIOCGSTAMP:
845 		err = sock_get_timestamp(sk, (struct timeval __user *)arg);
846 		break;
847 	case SIOCGSTAMPNS:
848 		err = sock_get_timestampns(sk, (struct timespec __user *)arg);
849 		break;
850 	case SIOCADDRT:
851 	case SIOCDELRT:
852 	case SIOCRTMSG:
853 		err = ip_rt_ioctl(net, cmd, (void __user *)arg);
854 		break;
855 	case SIOCDARP:
856 	case SIOCGARP:
857 	case SIOCSARP:
858 		err = arp_ioctl(net, cmd, (void __user *)arg);
859 		break;
860 	case SIOCGIFADDR:
861 	case SIOCSIFADDR:
862 	case SIOCGIFBRDADDR:
863 	case SIOCSIFBRDADDR:
864 	case SIOCGIFNETMASK:
865 	case SIOCSIFNETMASK:
866 	case SIOCGIFDSTADDR:
867 	case SIOCSIFDSTADDR:
868 	case SIOCSIFPFLAGS:
869 	case SIOCGIFPFLAGS:
870 	case SIOCSIFFLAGS:
871 		err = devinet_ioctl(net, cmd, (void __user *)arg);
872 		break;
873 	default:
874 		if (sk->sk_prot->ioctl)
875 			err = sk->sk_prot->ioctl(sk, cmd, arg);
876 		else
877 			err = -ENOIOCTLCMD;
878 		break;
879 	}
880 	return err;
881 }
882 EXPORT_SYMBOL(inet_ioctl);
883 
884 #ifdef CONFIG_COMPAT
885 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
886 {
887 	struct sock *sk = sock->sk;
888 	int err = -ENOIOCTLCMD;
889 
890 	if (sk->sk_prot->compat_ioctl)
891 		err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
892 
893 	return err;
894 }
895 #endif
896 
897 const struct proto_ops inet_stream_ops = {
898 	.family		   = PF_INET,
899 	.owner		   = THIS_MODULE,
900 	.release	   = inet_release,
901 	.bind		   = inet_bind,
902 	.connect	   = inet_stream_connect,
903 	.socketpair	   = sock_no_socketpair,
904 	.accept		   = inet_accept,
905 	.getname	   = inet_getname,
906 	.poll		   = tcp_poll,
907 	.ioctl		   = inet_ioctl,
908 	.listen		   = inet_listen,
909 	.shutdown	   = inet_shutdown,
910 	.setsockopt	   = sock_common_setsockopt,
911 	.getsockopt	   = sock_common_getsockopt,
912 	.sendmsg	   = inet_sendmsg,
913 	.recvmsg	   = inet_recvmsg,
914 	.mmap		   = sock_no_mmap,
915 	.sendpage	   = inet_sendpage,
916 	.splice_read	   = tcp_splice_read,
917 #ifdef CONFIG_COMPAT
918 	.compat_setsockopt = compat_sock_common_setsockopt,
919 	.compat_getsockopt = compat_sock_common_getsockopt,
920 	.compat_ioctl	   = inet_compat_ioctl,
921 #endif
922 };
923 EXPORT_SYMBOL(inet_stream_ops);
924 
925 const struct proto_ops inet_dgram_ops = {
926 	.family		   = PF_INET,
927 	.owner		   = THIS_MODULE,
928 	.release	   = inet_release,
929 	.bind		   = inet_bind,
930 	.connect	   = inet_dgram_connect,
931 	.socketpair	   = sock_no_socketpair,
932 	.accept		   = sock_no_accept,
933 	.getname	   = inet_getname,
934 	.poll		   = udp_poll,
935 	.ioctl		   = inet_ioctl,
936 	.listen		   = sock_no_listen,
937 	.shutdown	   = inet_shutdown,
938 	.setsockopt	   = sock_common_setsockopt,
939 	.getsockopt	   = sock_common_getsockopt,
940 	.sendmsg	   = inet_sendmsg,
941 	.recvmsg	   = inet_recvmsg,
942 	.mmap		   = sock_no_mmap,
943 	.sendpage	   = inet_sendpage,
944 #ifdef CONFIG_COMPAT
945 	.compat_setsockopt = compat_sock_common_setsockopt,
946 	.compat_getsockopt = compat_sock_common_getsockopt,
947 	.compat_ioctl	   = inet_compat_ioctl,
948 #endif
949 };
950 EXPORT_SYMBOL(inet_dgram_ops);
951 
952 /*
953  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
954  * udp_poll
955  */
956 static const struct proto_ops inet_sockraw_ops = {
957 	.family		   = PF_INET,
958 	.owner		   = THIS_MODULE,
959 	.release	   = inet_release,
960 	.bind		   = inet_bind,
961 	.connect	   = inet_dgram_connect,
962 	.socketpair	   = sock_no_socketpair,
963 	.accept		   = sock_no_accept,
964 	.getname	   = inet_getname,
965 	.poll		   = datagram_poll,
966 	.ioctl		   = inet_ioctl,
967 	.listen		   = sock_no_listen,
968 	.shutdown	   = inet_shutdown,
969 	.setsockopt	   = sock_common_setsockopt,
970 	.getsockopt	   = sock_common_getsockopt,
971 	.sendmsg	   = inet_sendmsg,
972 	.recvmsg	   = inet_recvmsg,
973 	.mmap		   = sock_no_mmap,
974 	.sendpage	   = inet_sendpage,
975 #ifdef CONFIG_COMPAT
976 	.compat_setsockopt = compat_sock_common_setsockopt,
977 	.compat_getsockopt = compat_sock_common_getsockopt,
978 	.compat_ioctl	   = inet_compat_ioctl,
979 #endif
980 };
981 
982 static const struct net_proto_family inet_family_ops = {
983 	.family = PF_INET,
984 	.create = inet_create,
985 	.owner	= THIS_MODULE,
986 };
987 
988 /* Upon startup we insert all the elements in inetsw_array[] into
989  * the linked list inetsw.
990  */
991 static struct inet_protosw inetsw_array[] =
992 {
993 	{
994 		.type =       SOCK_STREAM,
995 		.protocol =   IPPROTO_TCP,
996 		.prot =       &tcp_prot,
997 		.ops =        &inet_stream_ops,
998 		.flags =      INET_PROTOSW_PERMANENT |
999 			      INET_PROTOSW_ICSK,
1000 	},
1001 
1002 	{
1003 		.type =       SOCK_DGRAM,
1004 		.protocol =   IPPROTO_UDP,
1005 		.prot =       &udp_prot,
1006 		.ops =        &inet_dgram_ops,
1007 		.flags =      INET_PROTOSW_PERMANENT,
1008        },
1009 
1010        {
1011 		.type =       SOCK_DGRAM,
1012 		.protocol =   IPPROTO_ICMP,
1013 		.prot =       &ping_prot,
1014 		.ops =        &inet_dgram_ops,
1015 		.flags =      INET_PROTOSW_REUSE,
1016        },
1017 
1018        {
1019 	       .type =       SOCK_RAW,
1020 	       .protocol =   IPPROTO_IP,	/* wild card */
1021 	       .prot =       &raw_prot,
1022 	       .ops =        &inet_sockraw_ops,
1023 	       .flags =      INET_PROTOSW_REUSE,
1024        }
1025 };
1026 
1027 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1028 
1029 void inet_register_protosw(struct inet_protosw *p)
1030 {
1031 	struct list_head *lh;
1032 	struct inet_protosw *answer;
1033 	int protocol = p->protocol;
1034 	struct list_head *last_perm;
1035 
1036 	spin_lock_bh(&inetsw_lock);
1037 
1038 	if (p->type >= SOCK_MAX)
1039 		goto out_illegal;
1040 
1041 	/* If we are trying to override a permanent protocol, bail. */
1042 	answer = NULL;
1043 	last_perm = &inetsw[p->type];
1044 	list_for_each(lh, &inetsw[p->type]) {
1045 		answer = list_entry(lh, struct inet_protosw, list);
1046 
1047 		/* Check only the non-wild match. */
1048 		if (INET_PROTOSW_PERMANENT & answer->flags) {
1049 			if (protocol == answer->protocol)
1050 				break;
1051 			last_perm = lh;
1052 		}
1053 
1054 		answer = NULL;
1055 	}
1056 	if (answer)
1057 		goto out_permanent;
1058 
1059 	/* Add the new entry after the last permanent entry if any, so that
1060 	 * the new entry does not override a permanent entry when matched with
1061 	 * a wild-card protocol. But it is allowed to override any existing
1062 	 * non-permanent entry.  This means that when we remove this entry, the
1063 	 * system automatically returns to the old behavior.
1064 	 */
1065 	list_add_rcu(&p->list, last_perm);
1066 out:
1067 	spin_unlock_bh(&inetsw_lock);
1068 
1069 	return;
1070 
1071 out_permanent:
1072 	pr_err("Attempt to override permanent protocol %d\n", protocol);
1073 	goto out;
1074 
1075 out_illegal:
1076 	pr_err("Ignoring attempt to register invalid socket type %d\n",
1077 	       p->type);
1078 	goto out;
1079 }
1080 EXPORT_SYMBOL(inet_register_protosw);
1081 
1082 void inet_unregister_protosw(struct inet_protosw *p)
1083 {
1084 	if (INET_PROTOSW_PERMANENT & p->flags) {
1085 		pr_err("Attempt to unregister permanent protocol %d\n",
1086 		       p->protocol);
1087 	} else {
1088 		spin_lock_bh(&inetsw_lock);
1089 		list_del_rcu(&p->list);
1090 		spin_unlock_bh(&inetsw_lock);
1091 
1092 		synchronize_net();
1093 	}
1094 }
1095 EXPORT_SYMBOL(inet_unregister_protosw);
1096 
1097 /*
1098  *      Shall we try to damage output packets if routing dev changes?
1099  */
1100 
1101 int sysctl_ip_dynaddr __read_mostly;
1102 
1103 static int inet_sk_reselect_saddr(struct sock *sk)
1104 {
1105 	struct inet_sock *inet = inet_sk(sk);
1106 	__be32 old_saddr = inet->inet_saddr;
1107 	__be32 daddr = inet->inet_daddr;
1108 	struct flowi4 *fl4;
1109 	struct rtable *rt;
1110 	__be32 new_saddr;
1111 	struct ip_options_rcu *inet_opt;
1112 
1113 	inet_opt = rcu_dereference_protected(inet->inet_opt,
1114 					     sock_owned_by_user(sk));
1115 	if (inet_opt && inet_opt->opt.srr)
1116 		daddr = inet_opt->opt.faddr;
1117 
1118 	/* Query new route. */
1119 	fl4 = &inet->cork.fl.u.ip4;
1120 	rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1121 			      sk->sk_bound_dev_if, sk->sk_protocol,
1122 			      inet->inet_sport, inet->inet_dport, sk);
1123 	if (IS_ERR(rt))
1124 		return PTR_ERR(rt);
1125 
1126 	sk_setup_caps(sk, &rt->dst);
1127 
1128 	new_saddr = fl4->saddr;
1129 
1130 	if (new_saddr == old_saddr)
1131 		return 0;
1132 
1133 	if (sysctl_ip_dynaddr > 1) {
1134 		pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1135 			__func__, &old_saddr, &new_saddr);
1136 	}
1137 
1138 	inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1139 
1140 	/*
1141 	 * XXX The only one ugly spot where we need to
1142 	 * XXX really change the sockets identity after
1143 	 * XXX it has entered the hashes. -DaveM
1144 	 *
1145 	 * Besides that, it does not check for connection
1146 	 * uniqueness. Wait for troubles.
1147 	 */
1148 	__sk_prot_rehash(sk);
1149 	return 0;
1150 }
1151 
1152 int inet_sk_rebuild_header(struct sock *sk)
1153 {
1154 	struct inet_sock *inet = inet_sk(sk);
1155 	struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1156 	__be32 daddr;
1157 	struct ip_options_rcu *inet_opt;
1158 	struct flowi4 *fl4;
1159 	int err;
1160 
1161 	/* Route is OK, nothing to do. */
1162 	if (rt)
1163 		return 0;
1164 
1165 	/* Reroute. */
1166 	rcu_read_lock();
1167 	inet_opt = rcu_dereference(inet->inet_opt);
1168 	daddr = inet->inet_daddr;
1169 	if (inet_opt && inet_opt->opt.srr)
1170 		daddr = inet_opt->opt.faddr;
1171 	rcu_read_unlock();
1172 	fl4 = &inet->cork.fl.u.ip4;
1173 	rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1174 				   inet->inet_dport, inet->inet_sport,
1175 				   sk->sk_protocol, RT_CONN_FLAGS(sk),
1176 				   sk->sk_bound_dev_if);
1177 	if (!IS_ERR(rt)) {
1178 		err = 0;
1179 		sk_setup_caps(sk, &rt->dst);
1180 	} else {
1181 		err = PTR_ERR(rt);
1182 
1183 		/* Routing failed... */
1184 		sk->sk_route_caps = 0;
1185 		/*
1186 		 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1187 		 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1188 		 */
1189 		if (!sysctl_ip_dynaddr ||
1190 		    sk->sk_state != TCP_SYN_SENT ||
1191 		    (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1192 		    (err = inet_sk_reselect_saddr(sk)) != 0)
1193 			sk->sk_err_soft = -err;
1194 	}
1195 
1196 	return err;
1197 }
1198 EXPORT_SYMBOL(inet_sk_rebuild_header);
1199 
1200 static struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1201 					netdev_features_t features)
1202 {
1203 	struct sk_buff *segs = ERR_PTR(-EINVAL);
1204 	const struct net_offload *ops;
1205 	unsigned int offset = 0;
1206 	bool udpfrag, encap;
1207 	struct iphdr *iph;
1208 	int proto;
1209 	int nhoff;
1210 	int ihl;
1211 	int id;
1212 
1213 	if (unlikely(skb_shinfo(skb)->gso_type &
1214 		     ~(SKB_GSO_TCPV4 |
1215 		       SKB_GSO_UDP |
1216 		       SKB_GSO_DODGY |
1217 		       SKB_GSO_TCP_ECN |
1218 		       SKB_GSO_GRE |
1219 		       SKB_GSO_GRE_CSUM |
1220 		       SKB_GSO_IPIP |
1221 		       SKB_GSO_SIT |
1222 		       SKB_GSO_TCPV6 |
1223 		       SKB_GSO_UDP_TUNNEL |
1224 		       SKB_GSO_UDP_TUNNEL_CSUM |
1225 		       SKB_GSO_TUNNEL_REMCSUM |
1226 		       0)))
1227 		goto out;
1228 
1229 	skb_reset_network_header(skb);
1230 	nhoff = skb_network_header(skb) - skb_mac_header(skb);
1231 	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1232 		goto out;
1233 
1234 	iph = ip_hdr(skb);
1235 	ihl = iph->ihl * 4;
1236 	if (ihl < sizeof(*iph))
1237 		goto out;
1238 
1239 	id = ntohs(iph->id);
1240 	proto = iph->protocol;
1241 
1242 	/* Warning: after this point, iph might be no longer valid */
1243 	if (unlikely(!pskb_may_pull(skb, ihl)))
1244 		goto out;
1245 	__skb_pull(skb, ihl);
1246 
1247 	encap = SKB_GSO_CB(skb)->encap_level > 0;
1248 	if (encap)
1249 		features &= skb->dev->hw_enc_features;
1250 	SKB_GSO_CB(skb)->encap_level += ihl;
1251 
1252 	skb_reset_transport_header(skb);
1253 
1254 	segs = ERR_PTR(-EPROTONOSUPPORT);
1255 
1256 	if (skb->encapsulation &&
1257 	    skb_shinfo(skb)->gso_type & (SKB_GSO_SIT|SKB_GSO_IPIP))
1258 		udpfrag = proto == IPPROTO_UDP && encap;
1259 	else
1260 		udpfrag = proto == IPPROTO_UDP && !skb->encapsulation;
1261 
1262 	ops = rcu_dereference(inet_offloads[proto]);
1263 	if (likely(ops && ops->callbacks.gso_segment))
1264 		segs = ops->callbacks.gso_segment(skb, features);
1265 
1266 	if (IS_ERR_OR_NULL(segs))
1267 		goto out;
1268 
1269 	skb = segs;
1270 	do {
1271 		iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1272 		if (udpfrag) {
1273 			iph->id = htons(id);
1274 			iph->frag_off = htons(offset >> 3);
1275 			if (skb->next)
1276 				iph->frag_off |= htons(IP_MF);
1277 			offset += skb->len - nhoff - ihl;
1278 		} else {
1279 			iph->id = htons(id++);
1280 		}
1281 		iph->tot_len = htons(skb->len - nhoff);
1282 		ip_send_check(iph);
1283 		if (encap)
1284 			skb_reset_inner_headers(skb);
1285 		skb->network_header = (u8 *)iph - skb->head;
1286 	} while ((skb = skb->next));
1287 
1288 out:
1289 	return segs;
1290 }
1291 
1292 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1293 					 struct sk_buff *skb)
1294 {
1295 	const struct net_offload *ops;
1296 	struct sk_buff **pp = NULL;
1297 	struct sk_buff *p;
1298 	const struct iphdr *iph;
1299 	unsigned int hlen;
1300 	unsigned int off;
1301 	unsigned int id;
1302 	int flush = 1;
1303 	int proto;
1304 
1305 	off = skb_gro_offset(skb);
1306 	hlen = off + sizeof(*iph);
1307 	iph = skb_gro_header_fast(skb, off);
1308 	if (skb_gro_header_hard(skb, hlen)) {
1309 		iph = skb_gro_header_slow(skb, hlen, off);
1310 		if (unlikely(!iph))
1311 			goto out;
1312 	}
1313 
1314 	proto = iph->protocol;
1315 
1316 	rcu_read_lock();
1317 	ops = rcu_dereference(inet_offloads[proto]);
1318 	if (!ops || !ops->callbacks.gro_receive)
1319 		goto out_unlock;
1320 
1321 	if (*(u8 *)iph != 0x45)
1322 		goto out_unlock;
1323 
1324 	if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1325 		goto out_unlock;
1326 
1327 	id = ntohl(*(__be32 *)&iph->id);
1328 	flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1329 	id >>= 16;
1330 
1331 	for (p = *head; p; p = p->next) {
1332 		struct iphdr *iph2;
1333 
1334 		if (!NAPI_GRO_CB(p)->same_flow)
1335 			continue;
1336 
1337 		iph2 = (struct iphdr *)(p->data + off);
1338 		/* The above works because, with the exception of the top
1339 		 * (inner most) layer, we only aggregate pkts with the same
1340 		 * hdr length so all the hdrs we'll need to verify will start
1341 		 * at the same offset.
1342 		 */
1343 		if ((iph->protocol ^ iph2->protocol) |
1344 		    ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1345 		    ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1346 			NAPI_GRO_CB(p)->same_flow = 0;
1347 			continue;
1348 		}
1349 
1350 		/* All fields must match except length and checksum. */
1351 		NAPI_GRO_CB(p)->flush |=
1352 			(iph->ttl ^ iph2->ttl) |
1353 			(iph->tos ^ iph2->tos) |
1354 			((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1355 
1356 		/* Save the IP ID check to be included later when we get to
1357 		 * the transport layer so only the inner most IP ID is checked.
1358 		 * This is because some GSO/TSO implementations do not
1359 		 * correctly increment the IP ID for the outer hdrs.
1360 		 */
1361 		NAPI_GRO_CB(p)->flush_id =
1362 			    ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1363 		NAPI_GRO_CB(p)->flush |= flush;
1364 	}
1365 
1366 	NAPI_GRO_CB(skb)->flush |= flush;
1367 	skb_set_network_header(skb, off);
1368 	/* The above will be needed by the transport layer if there is one
1369 	 * immediately following this IP hdr.
1370 	 */
1371 
1372 	/* Note : No need to call skb_gro_postpull_rcsum() here,
1373 	 * as we already checked checksum over ipv4 header was 0
1374 	 */
1375 	skb_gro_pull(skb, sizeof(*iph));
1376 	skb_set_transport_header(skb, skb_gro_offset(skb));
1377 
1378 	pp = ops->callbacks.gro_receive(head, skb);
1379 
1380 out_unlock:
1381 	rcu_read_unlock();
1382 
1383 out:
1384 	NAPI_GRO_CB(skb)->flush |= flush;
1385 
1386 	return pp;
1387 }
1388 
1389 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1390 {
1391 	if (sk->sk_family == AF_INET)
1392 		return ip_recv_error(sk, msg, len, addr_len);
1393 #if IS_ENABLED(CONFIG_IPV6)
1394 	if (sk->sk_family == AF_INET6)
1395 		return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1396 #endif
1397 	return -EINVAL;
1398 }
1399 
1400 static int inet_gro_complete(struct sk_buff *skb, int nhoff)
1401 {
1402 	__be16 newlen = htons(skb->len - nhoff);
1403 	struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1404 	const struct net_offload *ops;
1405 	int proto = iph->protocol;
1406 	int err = -ENOSYS;
1407 
1408 	if (skb->encapsulation)
1409 		skb_set_inner_network_header(skb, nhoff);
1410 
1411 	csum_replace2(&iph->check, iph->tot_len, newlen);
1412 	iph->tot_len = newlen;
1413 
1414 	rcu_read_lock();
1415 	ops = rcu_dereference(inet_offloads[proto]);
1416 	if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1417 		goto out_unlock;
1418 
1419 	/* Only need to add sizeof(*iph) to get to the next hdr below
1420 	 * because any hdr with option will have been flushed in
1421 	 * inet_gro_receive().
1422 	 */
1423 	err = ops->callbacks.gro_complete(skb, nhoff + sizeof(*iph));
1424 
1425 out_unlock:
1426 	rcu_read_unlock();
1427 
1428 	return err;
1429 }
1430 
1431 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1432 			 unsigned short type, unsigned char protocol,
1433 			 struct net *net)
1434 {
1435 	struct socket *sock;
1436 	int rc = sock_create_kern(net, family, type, protocol, &sock);
1437 
1438 	if (rc == 0) {
1439 		*sk = sock->sk;
1440 		(*sk)->sk_allocation = GFP_ATOMIC;
1441 		/*
1442 		 * Unhash it so that IP input processing does not even see it,
1443 		 * we do not wish this socket to see incoming packets.
1444 		 */
1445 		(*sk)->sk_prot->unhash(*sk);
1446 	}
1447 	return rc;
1448 }
1449 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1450 
1451 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1452 {
1453 	unsigned long res = 0;
1454 	int i;
1455 
1456 	for_each_possible_cpu(i)
1457 		res += *(((unsigned long *) per_cpu_ptr(mib, i)) + offt);
1458 	return res;
1459 }
1460 EXPORT_SYMBOL_GPL(snmp_fold_field);
1461 
1462 #if BITS_PER_LONG==32
1463 
1464 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1465 {
1466 	u64 res = 0;
1467 	int cpu;
1468 
1469 	for_each_possible_cpu(cpu) {
1470 		void *bhptr;
1471 		struct u64_stats_sync *syncp;
1472 		u64 v;
1473 		unsigned int start;
1474 
1475 		bhptr = per_cpu_ptr(mib, cpu);
1476 		syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1477 		do {
1478 			start = u64_stats_fetch_begin_irq(syncp);
1479 			v = *(((u64 *) bhptr) + offt);
1480 		} while (u64_stats_fetch_retry_irq(syncp, start));
1481 
1482 		res += v;
1483 	}
1484 	return res;
1485 }
1486 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1487 #endif
1488 
1489 #ifdef CONFIG_IP_MULTICAST
1490 static const struct net_protocol igmp_protocol = {
1491 	.handler =	igmp_rcv,
1492 	.netns_ok =	1,
1493 };
1494 #endif
1495 
1496 static const struct net_protocol tcp_protocol = {
1497 	.early_demux	=	tcp_v4_early_demux,
1498 	.handler	=	tcp_v4_rcv,
1499 	.err_handler	=	tcp_v4_err,
1500 	.no_policy	=	1,
1501 	.netns_ok	=	1,
1502 	.icmp_strict_tag_validation = 1,
1503 };
1504 
1505 static const struct net_protocol udp_protocol = {
1506 	.early_demux =	udp_v4_early_demux,
1507 	.handler =	udp_rcv,
1508 	.err_handler =	udp_err,
1509 	.no_policy =	1,
1510 	.netns_ok =	1,
1511 };
1512 
1513 static const struct net_protocol icmp_protocol = {
1514 	.handler =	icmp_rcv,
1515 	.err_handler =	icmp_err,
1516 	.no_policy =	1,
1517 	.netns_ok =	1,
1518 };
1519 
1520 static __net_init int ipv4_mib_init_net(struct net *net)
1521 {
1522 	int i;
1523 
1524 	net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1525 	if (!net->mib.tcp_statistics)
1526 		goto err_tcp_mib;
1527 	net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1528 	if (!net->mib.ip_statistics)
1529 		goto err_ip_mib;
1530 
1531 	for_each_possible_cpu(i) {
1532 		struct ipstats_mib *af_inet_stats;
1533 		af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1534 		u64_stats_init(&af_inet_stats->syncp);
1535 	}
1536 
1537 	net->mib.net_statistics = alloc_percpu(struct linux_mib);
1538 	if (!net->mib.net_statistics)
1539 		goto err_net_mib;
1540 	net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1541 	if (!net->mib.udp_statistics)
1542 		goto err_udp_mib;
1543 	net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1544 	if (!net->mib.udplite_statistics)
1545 		goto err_udplite_mib;
1546 	net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1547 	if (!net->mib.icmp_statistics)
1548 		goto err_icmp_mib;
1549 	net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1550 					      GFP_KERNEL);
1551 	if (!net->mib.icmpmsg_statistics)
1552 		goto err_icmpmsg_mib;
1553 
1554 	tcp_mib_init(net);
1555 	return 0;
1556 
1557 err_icmpmsg_mib:
1558 	free_percpu(net->mib.icmp_statistics);
1559 err_icmp_mib:
1560 	free_percpu(net->mib.udplite_statistics);
1561 err_udplite_mib:
1562 	free_percpu(net->mib.udp_statistics);
1563 err_udp_mib:
1564 	free_percpu(net->mib.net_statistics);
1565 err_net_mib:
1566 	free_percpu(net->mib.ip_statistics);
1567 err_ip_mib:
1568 	free_percpu(net->mib.tcp_statistics);
1569 err_tcp_mib:
1570 	return -ENOMEM;
1571 }
1572 
1573 static __net_exit void ipv4_mib_exit_net(struct net *net)
1574 {
1575 	kfree(net->mib.icmpmsg_statistics);
1576 	free_percpu(net->mib.icmp_statistics);
1577 	free_percpu(net->mib.udplite_statistics);
1578 	free_percpu(net->mib.udp_statistics);
1579 	free_percpu(net->mib.net_statistics);
1580 	free_percpu(net->mib.ip_statistics);
1581 	free_percpu(net->mib.tcp_statistics);
1582 }
1583 
1584 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1585 	.init = ipv4_mib_init_net,
1586 	.exit = ipv4_mib_exit_net,
1587 };
1588 
1589 static int __init init_ipv4_mibs(void)
1590 {
1591 	return register_pernet_subsys(&ipv4_mib_ops);
1592 }
1593 
1594 static __net_init int inet_init_net(struct net *net)
1595 {
1596 	/*
1597 	 * Set defaults for local port range
1598 	 */
1599 	seqlock_init(&net->ipv4.ip_local_ports.lock);
1600 	net->ipv4.ip_local_ports.range[0] =  32768;
1601 	net->ipv4.ip_local_ports.range[1] =  60999;
1602 
1603 	seqlock_init(&net->ipv4.ping_group_range.lock);
1604 	/*
1605 	 * Sane defaults - nobody may create ping sockets.
1606 	 * Boot scripts should set this to distro-specific group.
1607 	 */
1608 	net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1609 	net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1610 	return 0;
1611 }
1612 
1613 static __net_exit void inet_exit_net(struct net *net)
1614 {
1615 }
1616 
1617 static __net_initdata struct pernet_operations af_inet_ops = {
1618 	.init = inet_init_net,
1619 	.exit = inet_exit_net,
1620 };
1621 
1622 static int __init init_inet_pernet_ops(void)
1623 {
1624 	return register_pernet_subsys(&af_inet_ops);
1625 }
1626 
1627 static int ipv4_proc_init(void);
1628 
1629 /*
1630  *	IP protocol layer initialiser
1631  */
1632 
1633 static struct packet_offload ip_packet_offload __read_mostly = {
1634 	.type = cpu_to_be16(ETH_P_IP),
1635 	.callbacks = {
1636 		.gso_segment = inet_gso_segment,
1637 		.gro_receive = inet_gro_receive,
1638 		.gro_complete = inet_gro_complete,
1639 	},
1640 };
1641 
1642 static const struct net_offload ipip_offload = {
1643 	.callbacks = {
1644 		.gso_segment	= inet_gso_segment,
1645 		.gro_receive	= inet_gro_receive,
1646 		.gro_complete	= inet_gro_complete,
1647 	},
1648 };
1649 
1650 static int __init ipv4_offload_init(void)
1651 {
1652 	/*
1653 	 * Add offloads
1654 	 */
1655 	if (udpv4_offload_init() < 0)
1656 		pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1657 	if (tcpv4_offload_init() < 0)
1658 		pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1659 
1660 	dev_add_offload(&ip_packet_offload);
1661 	inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1662 	return 0;
1663 }
1664 
1665 fs_initcall(ipv4_offload_init);
1666 
1667 static struct packet_type ip_packet_type __read_mostly = {
1668 	.type = cpu_to_be16(ETH_P_IP),
1669 	.func = ip_rcv,
1670 };
1671 
1672 static int __init inet_init(void)
1673 {
1674 	struct inet_protosw *q;
1675 	struct list_head *r;
1676 	int rc = -EINVAL;
1677 
1678 	sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1679 
1680 	rc = proto_register(&tcp_prot, 1);
1681 	if (rc)
1682 		goto out;
1683 
1684 	rc = proto_register(&udp_prot, 1);
1685 	if (rc)
1686 		goto out_unregister_tcp_proto;
1687 
1688 	rc = proto_register(&raw_prot, 1);
1689 	if (rc)
1690 		goto out_unregister_udp_proto;
1691 
1692 	rc = proto_register(&ping_prot, 1);
1693 	if (rc)
1694 		goto out_unregister_raw_proto;
1695 
1696 	/*
1697 	 *	Tell SOCKET that we are alive...
1698 	 */
1699 
1700 	(void)sock_register(&inet_family_ops);
1701 
1702 #ifdef CONFIG_SYSCTL
1703 	ip_static_sysctl_init();
1704 #endif
1705 
1706 	/*
1707 	 *	Add all the base protocols.
1708 	 */
1709 
1710 	if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1711 		pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1712 	if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1713 		pr_crit("%s: Cannot add UDP protocol\n", __func__);
1714 	if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1715 		pr_crit("%s: Cannot add TCP protocol\n", __func__);
1716 #ifdef CONFIG_IP_MULTICAST
1717 	if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1718 		pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1719 #endif
1720 
1721 	/* Register the socket-side information for inet_create. */
1722 	for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1723 		INIT_LIST_HEAD(r);
1724 
1725 	for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1726 		inet_register_protosw(q);
1727 
1728 	/*
1729 	 *	Set the ARP module up
1730 	 */
1731 
1732 	arp_init();
1733 
1734 	/*
1735 	 *	Set the IP module up
1736 	 */
1737 
1738 	ip_init();
1739 
1740 	tcp_v4_init();
1741 
1742 	/* Setup TCP slab cache for open requests. */
1743 	tcp_init();
1744 
1745 	/* Setup UDP memory threshold */
1746 	udp_init();
1747 
1748 	/* Add UDP-Lite (RFC 3828) */
1749 	udplite4_register();
1750 
1751 	ping_init();
1752 
1753 	/*
1754 	 *	Set the ICMP layer up
1755 	 */
1756 
1757 	if (icmp_init() < 0)
1758 		panic("Failed to create the ICMP control socket.\n");
1759 
1760 	/*
1761 	 *	Initialise the multicast router
1762 	 */
1763 #if defined(CONFIG_IP_MROUTE)
1764 	if (ip_mr_init())
1765 		pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1766 #endif
1767 
1768 	if (init_inet_pernet_ops())
1769 		pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
1770 	/*
1771 	 *	Initialise per-cpu ipv4 mibs
1772 	 */
1773 
1774 	if (init_ipv4_mibs())
1775 		pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1776 
1777 	ipv4_proc_init();
1778 
1779 	ipfrag_init();
1780 
1781 	dev_add_pack(&ip_packet_type);
1782 
1783 	rc = 0;
1784 out:
1785 	return rc;
1786 out_unregister_raw_proto:
1787 	proto_unregister(&raw_prot);
1788 out_unregister_udp_proto:
1789 	proto_unregister(&udp_prot);
1790 out_unregister_tcp_proto:
1791 	proto_unregister(&tcp_prot);
1792 	goto out;
1793 }
1794 
1795 fs_initcall(inet_init);
1796 
1797 /* ------------------------------------------------------------------------ */
1798 
1799 #ifdef CONFIG_PROC_FS
1800 static int __init ipv4_proc_init(void)
1801 {
1802 	int rc = 0;
1803 
1804 	if (raw_proc_init())
1805 		goto out_raw;
1806 	if (tcp4_proc_init())
1807 		goto out_tcp;
1808 	if (udp4_proc_init())
1809 		goto out_udp;
1810 	if (ping_proc_init())
1811 		goto out_ping;
1812 	if (ip_misc_proc_init())
1813 		goto out_misc;
1814 out:
1815 	return rc;
1816 out_misc:
1817 	ping_proc_exit();
1818 out_ping:
1819 	udp4_proc_exit();
1820 out_udp:
1821 	tcp4_proc_exit();
1822 out_tcp:
1823 	raw_proc_exit();
1824 out_raw:
1825 	rc = -ENOMEM;
1826 	goto out;
1827 }
1828 
1829 #else /* CONFIG_PROC_FS */
1830 static int __init ipv4_proc_init(void)
1831 {
1832 	return 0;
1833 }
1834 #endif /* CONFIG_PROC_FS */
1835 
1836 MODULE_ALIAS_NETPROTO(PF_INET);
1837 
1838