1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		PF_INET protocol family socket handler.
8  *
9  * Authors:	Ross Biro
10  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11  *		Florian La Roche, <flla@stud.uni-sb.de>
12  *		Alan Cox, <A.Cox@swansea.ac.uk>
13  *
14  * Changes (see also sock.c)
15  *
16  *		piggy,
17  *		Karl Knutson	:	Socket protocol table
18  *		A.N.Kuznetsov	:	Socket death error in accept().
19  *		John Richardson :	Fix non blocking error in connect()
20  *					so sockets that fail to connect
21  *					don't return -EINPROGRESS.
22  *		Alan Cox	:	Asynchronous I/O support
23  *		Alan Cox	:	Keep correct socket pointer on sock
24  *					structures
25  *					when accept() ed
26  *		Alan Cox	:	Semantics of SO_LINGER aren't state
27  *					moved to close when you look carefully.
28  *					With this fixed and the accept bug fixed
29  *					some RPC stuff seems happier.
30  *		Niibe Yutaka	:	4.4BSD style write async I/O
31  *		Alan Cox,
32  *		Tony Gale 	:	Fixed reuse semantics.
33  *		Alan Cox	:	bind() shouldn't abort existing but dead
34  *					sockets. Stops FTP netin:.. I hope.
35  *		Alan Cox	:	bind() works correctly for RAW sockets.
36  *					Note that FreeBSD at least was broken
37  *					in this respect so be careful with
38  *					compatibility tests...
39  *		Alan Cox	:	routing cache support
40  *		Alan Cox	:	memzero the socket structure for
41  *					compactness.
42  *		Matt Day	:	nonblock connect error handler
43  *		Alan Cox	:	Allow large numbers of pending sockets
44  *					(eg for big web sites), but only if
45  *					specifically application requested.
46  *		Alan Cox	:	New buffering throughout IP. Used
47  *					dumbly.
48  *		Alan Cox	:	New buffering now used smartly.
49  *		Alan Cox	:	BSD rather than common sense
50  *					interpretation of listen.
51  *		Germano Caronni	:	Assorted small races.
52  *		Alan Cox	:	sendmsg/recvmsg basic support.
53  *		Alan Cox	:	Only sendmsg/recvmsg now supported.
54  *		Alan Cox	:	Locked down bind (see security list).
55  *		Alan Cox	:	Loosened bind a little.
56  *		Mike McLagan	:	ADD/DEL DLCI Ioctls
57  *	Willy Konynenberg	:	Transparent proxying support.
58  *		David S. Miller	:	New socket lookup architecture.
59  *					Some other random speedups.
60  *		Cyrus Durgin	:	Cleaned up file for kmod hacks.
61  *		Andi Kleen	:	Fix inet_stream_connect TCP race.
62  */
63 
64 #define pr_fmt(fmt) "IPv4: " fmt
65 
66 #include <linux/err.h>
67 #include <linux/errno.h>
68 #include <linux/types.h>
69 #include <linux/socket.h>
70 #include <linux/in.h>
71 #include <linux/kernel.h>
72 #include <linux/kmod.h>
73 #include <linux/sched.h>
74 #include <linux/timer.h>
75 #include <linux/string.h>
76 #include <linux/sockios.h>
77 #include <linux/net.h>
78 #include <linux/capability.h>
79 #include <linux/fcntl.h>
80 #include <linux/mm.h>
81 #include <linux/interrupt.h>
82 #include <linux/stat.h>
83 #include <linux/init.h>
84 #include <linux/poll.h>
85 #include <linux/netfilter_ipv4.h>
86 #include <linux/random.h>
87 #include <linux/slab.h>
88 
89 #include <linux/uaccess.h>
90 
91 #include <linux/inet.h>
92 #include <linux/igmp.h>
93 #include <linux/inetdevice.h>
94 #include <linux/netdevice.h>
95 #include <net/checksum.h>
96 #include <net/ip.h>
97 #include <net/protocol.h>
98 #include <net/arp.h>
99 #include <net/route.h>
100 #include <net/ip_fib.h>
101 #include <net/inet_connection_sock.h>
102 #include <net/gro.h>
103 #include <net/tcp.h>
104 #include <net/udp.h>
105 #include <net/udplite.h>
106 #include <net/ping.h>
107 #include <linux/skbuff.h>
108 #include <net/sock.h>
109 #include <net/raw.h>
110 #include <net/icmp.h>
111 #include <net/inet_common.h>
112 #include <net/ip_tunnels.h>
113 #include <net/xfrm.h>
114 #include <net/net_namespace.h>
115 #include <net/secure_seq.h>
116 #ifdef CONFIG_IP_MROUTE
117 #include <linux/mroute.h>
118 #endif
119 #include <net/l3mdev.h>
120 #include <net/compat.h>
121 
122 #include <trace/events/sock.h>
123 
124 /* The inetsw table contains everything that inet_create needs to
125  * build a new socket.
126  */
127 static struct list_head inetsw[SOCK_MAX];
128 static DEFINE_SPINLOCK(inetsw_lock);
129 
130 /* New destruction routine */
131 
inet_sock_destruct(struct sock * sk)132 void inet_sock_destruct(struct sock *sk)
133 {
134 	struct inet_sock *inet = inet_sk(sk);
135 
136 	__skb_queue_purge(&sk->sk_receive_queue);
137 	__skb_queue_purge(&sk->sk_error_queue);
138 
139 	sk_mem_reclaim_final(sk);
140 
141 	if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
142 		pr_err("Attempt to release TCP socket in state %d %p\n",
143 		       sk->sk_state, sk);
144 		return;
145 	}
146 	if (!sock_flag(sk, SOCK_DEAD)) {
147 		pr_err("Attempt to release alive inet socket %p\n", sk);
148 		return;
149 	}
150 
151 	WARN_ON_ONCE(atomic_read(&sk->sk_rmem_alloc));
152 	WARN_ON_ONCE(refcount_read(&sk->sk_wmem_alloc));
153 	WARN_ON_ONCE(sk->sk_wmem_queued);
154 	WARN_ON_ONCE(sk_forward_alloc_get(sk));
155 
156 	kfree(rcu_dereference_protected(inet->inet_opt, 1));
157 	dst_release(rcu_dereference_protected(sk->sk_dst_cache, 1));
158 	dst_release(rcu_dereference_protected(sk->sk_rx_dst, 1));
159 	sk_refcnt_debug_dec(sk);
160 }
161 EXPORT_SYMBOL(inet_sock_destruct);
162 
163 /*
164  *	The routines beyond this point handle the behaviour of an AF_INET
165  *	socket object. Mostly it punts to the subprotocols of IP to do
166  *	the work.
167  */
168 
169 /*
170  *	Automatically bind an unbound socket.
171  */
172 
inet_autobind(struct sock * sk)173 static int inet_autobind(struct sock *sk)
174 {
175 	struct inet_sock *inet;
176 	/* We may need to bind the socket. */
177 	lock_sock(sk);
178 	inet = inet_sk(sk);
179 	if (!inet->inet_num) {
180 		if (sk->sk_prot->get_port(sk, 0)) {
181 			release_sock(sk);
182 			return -EAGAIN;
183 		}
184 		inet->inet_sport = htons(inet->inet_num);
185 	}
186 	release_sock(sk);
187 	return 0;
188 }
189 
190 /*
191  *	Move a socket into listening state.
192  */
inet_listen(struct socket * sock,int backlog)193 int inet_listen(struct socket *sock, int backlog)
194 {
195 	struct sock *sk = sock->sk;
196 	unsigned char old_state;
197 	int err, tcp_fastopen;
198 
199 	lock_sock(sk);
200 
201 	err = -EINVAL;
202 	if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
203 		goto out;
204 
205 	old_state = sk->sk_state;
206 	if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
207 		goto out;
208 
209 	WRITE_ONCE(sk->sk_max_ack_backlog, backlog);
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 		tcp_fastopen = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fastopen);
221 		if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
222 		    (tcp_fastopen & TFO_SERVER_ENABLE) &&
223 		    !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
224 			fastopen_queue_tune(sk, backlog);
225 			tcp_fastopen_init_key_once(sock_net(sk));
226 		}
227 
228 		err = inet_csk_listen_start(sk);
229 		if (err)
230 			goto out;
231 		tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL);
232 	}
233 	err = 0;
234 
235 out:
236 	release_sock(sk);
237 	return err;
238 }
239 EXPORT_SYMBOL(inet_listen);
240 
241 /*
242  *	Create an inet socket.
243  */
244 
inet_create(struct net * net,struct socket * sock,int protocol,int kern)245 static int inet_create(struct net *net, struct socket *sock, int protocol,
246 		       int kern)
247 {
248 	struct sock *sk;
249 	struct inet_protosw *answer;
250 	struct inet_sock *inet;
251 	struct proto *answer_prot;
252 	unsigned char answer_flags;
253 	int try_loading_module = 0;
254 	int err;
255 
256 	if (protocol < 0 || protocol >= IPPROTO_MAX)
257 		return -EINVAL;
258 
259 	sock->state = SS_UNCONNECTED;
260 
261 	/* Look for the requested type/protocol pair. */
262 lookup_protocol:
263 	err = -ESOCKTNOSUPPORT;
264 	rcu_read_lock();
265 	list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
266 
267 		err = 0;
268 		/* Check the non-wild match. */
269 		if (protocol == answer->protocol) {
270 			if (protocol != IPPROTO_IP)
271 				break;
272 		} else {
273 			/* Check for the two wild cases. */
274 			if (IPPROTO_IP == protocol) {
275 				protocol = answer->protocol;
276 				break;
277 			}
278 			if (IPPROTO_IP == answer->protocol)
279 				break;
280 		}
281 		err = -EPROTONOSUPPORT;
282 	}
283 
284 	if (unlikely(err)) {
285 		if (try_loading_module < 2) {
286 			rcu_read_unlock();
287 			/*
288 			 * Be more specific, e.g. net-pf-2-proto-132-type-1
289 			 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
290 			 */
291 			if (++try_loading_module == 1)
292 				request_module("net-pf-%d-proto-%d-type-%d",
293 					       PF_INET, protocol, sock->type);
294 			/*
295 			 * Fall back to generic, e.g. net-pf-2-proto-132
296 			 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
297 			 */
298 			else
299 				request_module("net-pf-%d-proto-%d",
300 					       PF_INET, protocol);
301 			goto lookup_protocol;
302 		} else
303 			goto out_rcu_unlock;
304 	}
305 
306 	err = -EPERM;
307 	if (sock->type == SOCK_RAW && !kern &&
308 	    !ns_capable(net->user_ns, CAP_NET_RAW))
309 		goto out_rcu_unlock;
310 
311 	sock->ops = answer->ops;
312 	answer_prot = answer->prot;
313 	answer_flags = answer->flags;
314 	rcu_read_unlock();
315 
316 	WARN_ON(!answer_prot->slab);
317 
318 	err = -ENOMEM;
319 	sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
320 	if (!sk)
321 		goto out;
322 
323 	err = 0;
324 	if (INET_PROTOSW_REUSE & answer_flags)
325 		sk->sk_reuse = SK_CAN_REUSE;
326 
327 	inet = inet_sk(sk);
328 	inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
329 
330 	inet->nodefrag = 0;
331 
332 	if (SOCK_RAW == sock->type) {
333 		inet->inet_num = protocol;
334 		if (IPPROTO_RAW == protocol)
335 			inet->hdrincl = 1;
336 	}
337 
338 	if (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc))
339 		inet->pmtudisc = IP_PMTUDISC_DONT;
340 	else
341 		inet->pmtudisc = IP_PMTUDISC_WANT;
342 
343 	inet->inet_id = 0;
344 
345 	sock_init_data(sock, sk);
346 
347 	sk->sk_destruct	   = inet_sock_destruct;
348 	sk->sk_protocol	   = protocol;
349 	sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
350 
351 	inet->uc_ttl	= -1;
352 	inet->mc_loop	= 1;
353 	inet->mc_ttl	= 1;
354 	inet->mc_all	= 1;
355 	inet->mc_index	= 0;
356 	inet->mc_list	= NULL;
357 	inet->rcv_tos	= 0;
358 
359 	sk_refcnt_debug_inc(sk);
360 
361 	if (inet->inet_num) {
362 		/* It assumes that any protocol which allows
363 		 * the user to assign a number at socket
364 		 * creation time automatically
365 		 * shares.
366 		 */
367 		inet->inet_sport = htons(inet->inet_num);
368 		/* Add to protocol hash chains. */
369 		err = sk->sk_prot->hash(sk);
370 		if (err) {
371 			sk_common_release(sk);
372 			goto out;
373 		}
374 	}
375 
376 	if (sk->sk_prot->init) {
377 		err = sk->sk_prot->init(sk);
378 		if (err) {
379 			sk_common_release(sk);
380 			goto out;
381 		}
382 	}
383 
384 	if (!kern) {
385 		err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
386 		if (err) {
387 			sk_common_release(sk);
388 			goto out;
389 		}
390 	}
391 out:
392 	return err;
393 out_rcu_unlock:
394 	rcu_read_unlock();
395 	goto out;
396 }
397 
398 
399 /*
400  *	The peer socket should always be NULL (or else). When we call this
401  *	function we are destroying the object and from then on nobody
402  *	should refer to it.
403  */
inet_release(struct socket * sock)404 int inet_release(struct socket *sock)
405 {
406 	struct sock *sk = sock->sk;
407 
408 	if (sk) {
409 		long timeout;
410 
411 		if (!sk->sk_kern_sock)
412 			BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk);
413 
414 		/* Applications forget to leave groups before exiting */
415 		ip_mc_drop_socket(sk);
416 
417 		/* If linger is set, we don't return until the close
418 		 * is complete.  Otherwise we return immediately. The
419 		 * actually closing is done the same either way.
420 		 *
421 		 * If the close is due to the process exiting, we never
422 		 * linger..
423 		 */
424 		timeout = 0;
425 		if (sock_flag(sk, SOCK_LINGER) &&
426 		    !(current->flags & PF_EXITING))
427 			timeout = sk->sk_lingertime;
428 		sk->sk_prot->close(sk, timeout);
429 		sock->sk = NULL;
430 	}
431 	return 0;
432 }
433 EXPORT_SYMBOL(inet_release);
434 
inet_bind(struct socket * sock,struct sockaddr * uaddr,int addr_len)435 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
436 {
437 	struct sock *sk = sock->sk;
438 	u32 flags = BIND_WITH_LOCK;
439 	int err;
440 
441 	/* If the socket has its own bind function then use it. (RAW) */
442 	if (sk->sk_prot->bind) {
443 		return sk->sk_prot->bind(sk, uaddr, addr_len);
444 	}
445 	if (addr_len < sizeof(struct sockaddr_in))
446 		return -EINVAL;
447 
448 	/* BPF prog is run before any checks are done so that if the prog
449 	 * changes context in a wrong way it will be caught.
450 	 */
451 	err = BPF_CGROUP_RUN_PROG_INET_BIND_LOCK(sk, uaddr,
452 						 CGROUP_INET4_BIND, &flags);
453 	if (err)
454 		return err;
455 
456 	return __inet_bind(sk, uaddr, addr_len, flags);
457 }
458 EXPORT_SYMBOL(inet_bind);
459 
__inet_bind(struct sock * sk,struct sockaddr * uaddr,int addr_len,u32 flags)460 int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
461 		u32 flags)
462 {
463 	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
464 	struct inet_sock *inet = inet_sk(sk);
465 	struct net *net = sock_net(sk);
466 	unsigned short snum;
467 	int chk_addr_ret;
468 	u32 tb_id = RT_TABLE_LOCAL;
469 	int err;
470 
471 	if (addr->sin_family != AF_INET) {
472 		/* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
473 		 * only if s_addr is INADDR_ANY.
474 		 */
475 		err = -EAFNOSUPPORT;
476 		if (addr->sin_family != AF_UNSPEC ||
477 		    addr->sin_addr.s_addr != htonl(INADDR_ANY))
478 			goto out;
479 	}
480 
481 	tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
482 	chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
483 
484 	/* Not specified by any standard per-se, however it breaks too
485 	 * many applications when removed.  It is unfortunate since
486 	 * allowing applications to make a non-local bind solves
487 	 * several problems with systems using dynamic addressing.
488 	 * (ie. your servers still start up even if your ISDN link
489 	 *  is temporarily down)
490 	 */
491 	err = -EADDRNOTAVAIL;
492 	if (!inet_addr_valid_or_nonlocal(net, inet, addr->sin_addr.s_addr,
493 	                                 chk_addr_ret))
494 		goto out;
495 
496 	snum = ntohs(addr->sin_port);
497 	err = -EACCES;
498 	if (!(flags & BIND_NO_CAP_NET_BIND_SERVICE) &&
499 	    snum && inet_port_requires_bind_service(net, snum) &&
500 	    !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
501 		goto out;
502 
503 	/*      We keep a pair of addresses. rcv_saddr is the one
504 	 *      used by hash lookups, and saddr is used for transmit.
505 	 *
506 	 *      In the BSD API these are the same except where it
507 	 *      would be illegal to use them (multicast/broadcast) in
508 	 *      which case the sending device address is used.
509 	 */
510 	if (flags & BIND_WITH_LOCK)
511 		lock_sock(sk);
512 
513 	/* Check these errors (active socket, double bind). */
514 	err = -EINVAL;
515 	if (sk->sk_state != TCP_CLOSE || inet->inet_num)
516 		goto out_release_sock;
517 
518 	inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
519 	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
520 		inet->inet_saddr = 0;  /* Use device */
521 
522 	/* Make sure we are allowed to bind here. */
523 	if (snum || !(inet->bind_address_no_port ||
524 		      (flags & BIND_FORCE_ADDRESS_NO_PORT))) {
525 		err = sk->sk_prot->get_port(sk, snum);
526 		if (err) {
527 			inet->inet_saddr = inet->inet_rcv_saddr = 0;
528 			goto out_release_sock;
529 		}
530 		if (!(flags & BIND_FROM_BPF)) {
531 			err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
532 			if (err) {
533 				inet->inet_saddr = inet->inet_rcv_saddr = 0;
534 				if (sk->sk_prot->put_port)
535 					sk->sk_prot->put_port(sk);
536 				goto out_release_sock;
537 			}
538 		}
539 	}
540 
541 	if (inet->inet_rcv_saddr)
542 		sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
543 	if (snum)
544 		sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
545 	inet->inet_sport = htons(inet->inet_num);
546 	inet->inet_daddr = 0;
547 	inet->inet_dport = 0;
548 	sk_dst_reset(sk);
549 	err = 0;
550 out_release_sock:
551 	if (flags & BIND_WITH_LOCK)
552 		release_sock(sk);
553 out:
554 	return err;
555 }
556 
inet_dgram_connect(struct socket * sock,struct sockaddr * uaddr,int addr_len,int flags)557 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
558 		       int addr_len, int flags)
559 {
560 	struct sock *sk = sock->sk;
561 	const struct proto *prot;
562 	int err;
563 
564 	if (addr_len < sizeof(uaddr->sa_family))
565 		return -EINVAL;
566 
567 	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
568 	prot = READ_ONCE(sk->sk_prot);
569 
570 	if (uaddr->sa_family == AF_UNSPEC)
571 		return prot->disconnect(sk, flags);
572 
573 	if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
574 		err = prot->pre_connect(sk, uaddr, addr_len);
575 		if (err)
576 			return err;
577 	}
578 
579 	if (data_race(!inet_sk(sk)->inet_num) && inet_autobind(sk))
580 		return -EAGAIN;
581 	return prot->connect(sk, uaddr, addr_len);
582 }
583 EXPORT_SYMBOL(inet_dgram_connect);
584 
inet_wait_for_connect(struct sock * sk,long timeo,int writebias)585 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
586 {
587 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
588 
589 	add_wait_queue(sk_sleep(sk), &wait);
590 	sk->sk_write_pending += writebias;
591 
592 	/* Basic assumption: if someone sets sk->sk_err, he _must_
593 	 * change state of the socket from TCP_SYN_*.
594 	 * Connect() does not allow to get error notifications
595 	 * without closing the socket.
596 	 */
597 	while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
598 		release_sock(sk);
599 		timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
600 		lock_sock(sk);
601 		if (signal_pending(current) || !timeo)
602 			break;
603 	}
604 	remove_wait_queue(sk_sleep(sk), &wait);
605 	sk->sk_write_pending -= writebias;
606 	return timeo;
607 }
608 
609 /*
610  *	Connect to a remote host. There is regrettably still a little
611  *	TCP 'magic' in here.
612  */
__inet_stream_connect(struct socket * sock,struct sockaddr * uaddr,int addr_len,int flags,int is_sendmsg)613 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
614 			  int addr_len, int flags, int is_sendmsg)
615 {
616 	struct sock *sk = sock->sk;
617 	int err;
618 	long timeo;
619 
620 	/*
621 	 * uaddr can be NULL and addr_len can be 0 if:
622 	 * sk is a TCP fastopen active socket and
623 	 * TCP_FASTOPEN_CONNECT sockopt is set and
624 	 * we already have a valid cookie for this socket.
625 	 * In this case, user can call write() after connect().
626 	 * write() will invoke tcp_sendmsg_fastopen() which calls
627 	 * __inet_stream_connect().
628 	 */
629 	if (uaddr) {
630 		if (addr_len < sizeof(uaddr->sa_family))
631 			return -EINVAL;
632 
633 		if (uaddr->sa_family == AF_UNSPEC) {
634 			err = sk->sk_prot->disconnect(sk, flags);
635 			sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
636 			goto out;
637 		}
638 	}
639 
640 	switch (sock->state) {
641 	default:
642 		err = -EINVAL;
643 		goto out;
644 	case SS_CONNECTED:
645 		err = -EISCONN;
646 		goto out;
647 	case SS_CONNECTING:
648 		if (inet_sk(sk)->defer_connect)
649 			err = is_sendmsg ? -EINPROGRESS : -EISCONN;
650 		else
651 			err = -EALREADY;
652 		/* Fall out of switch with err, set for this state */
653 		break;
654 	case SS_UNCONNECTED:
655 		err = -EISCONN;
656 		if (sk->sk_state != TCP_CLOSE)
657 			goto out;
658 
659 		if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
660 			err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
661 			if (err)
662 				goto out;
663 		}
664 
665 		err = sk->sk_prot->connect(sk, uaddr, addr_len);
666 		if (err < 0)
667 			goto out;
668 
669 		sock->state = SS_CONNECTING;
670 
671 		if (!err && inet_sk(sk)->defer_connect)
672 			goto out;
673 
674 		/* Just entered SS_CONNECTING state; the only
675 		 * difference is that return value in non-blocking
676 		 * case is EINPROGRESS, rather than EALREADY.
677 		 */
678 		err = -EINPROGRESS;
679 		break;
680 	}
681 
682 	timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
683 
684 	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
685 		int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
686 				tcp_sk(sk)->fastopen_req &&
687 				tcp_sk(sk)->fastopen_req->data ? 1 : 0;
688 
689 		/* Error code is set above */
690 		if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
691 			goto out;
692 
693 		err = sock_intr_errno(timeo);
694 		if (signal_pending(current))
695 			goto out;
696 	}
697 
698 	/* Connection was closed by RST, timeout, ICMP error
699 	 * or another process disconnected us.
700 	 */
701 	if (sk->sk_state == TCP_CLOSE)
702 		goto sock_error;
703 
704 	/* sk->sk_err may be not zero now, if RECVERR was ordered by user
705 	 * and error was received after socket entered established state.
706 	 * Hence, it is handled normally after connect() return successfully.
707 	 */
708 
709 	sock->state = SS_CONNECTED;
710 	err = 0;
711 out:
712 	return err;
713 
714 sock_error:
715 	err = sock_error(sk) ? : -ECONNABORTED;
716 	sock->state = SS_UNCONNECTED;
717 	if (sk->sk_prot->disconnect(sk, flags))
718 		sock->state = SS_DISCONNECTING;
719 	goto out;
720 }
721 EXPORT_SYMBOL(__inet_stream_connect);
722 
inet_stream_connect(struct socket * sock,struct sockaddr * uaddr,int addr_len,int flags)723 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
724 			int addr_len, int flags)
725 {
726 	int err;
727 
728 	lock_sock(sock->sk);
729 	err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
730 	release_sock(sock->sk);
731 	return err;
732 }
733 EXPORT_SYMBOL(inet_stream_connect);
734 
735 /*
736  *	Accept a pending connection. The TCP layer now gives BSD semantics.
737  */
738 
inet_accept(struct socket * sock,struct socket * newsock,int flags,bool kern)739 int inet_accept(struct socket *sock, struct socket *newsock, int flags,
740 		bool kern)
741 {
742 	struct sock *sk1 = sock->sk, *sk2;
743 	int err = -EINVAL;
744 
745 	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
746 	sk2 = READ_ONCE(sk1->sk_prot)->accept(sk1, flags, &err, kern);
747 	if (!sk2)
748 		goto do_err;
749 
750 	lock_sock(sk2);
751 
752 	sock_rps_record_flow(sk2);
753 	WARN_ON(!((1 << sk2->sk_state) &
754 		  (TCPF_ESTABLISHED | TCPF_SYN_RECV |
755 		  TCPF_CLOSE_WAIT | TCPF_CLOSE)));
756 
757 	if (test_bit(SOCK_SUPPORT_ZC, &sock->flags))
758 		set_bit(SOCK_SUPPORT_ZC, &newsock->flags);
759 	sock_graft(sk2, newsock);
760 
761 	newsock->state = SS_CONNECTED;
762 	err = 0;
763 	release_sock(sk2);
764 do_err:
765 	return err;
766 }
767 EXPORT_SYMBOL(inet_accept);
768 
769 /*
770  *	This does both peername and sockname.
771  */
inet_getname(struct socket * sock,struct sockaddr * uaddr,int peer)772 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
773 		 int peer)
774 {
775 	struct sock *sk		= sock->sk;
776 	struct inet_sock *inet	= inet_sk(sk);
777 	DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
778 
779 	sin->sin_family = AF_INET;
780 	lock_sock(sk);
781 	if (peer) {
782 		if (!inet->inet_dport ||
783 		    (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
784 		     peer == 1)) {
785 			release_sock(sk);
786 			return -ENOTCONN;
787 		}
788 		sin->sin_port = inet->inet_dport;
789 		sin->sin_addr.s_addr = inet->inet_daddr;
790 		BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin,
791 				       CGROUP_INET4_GETPEERNAME);
792 	} else {
793 		__be32 addr = inet->inet_rcv_saddr;
794 		if (!addr)
795 			addr = inet->inet_saddr;
796 		sin->sin_port = inet->inet_sport;
797 		sin->sin_addr.s_addr = addr;
798 		BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin,
799 				       CGROUP_INET4_GETSOCKNAME);
800 	}
801 	release_sock(sk);
802 	memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
803 	return sizeof(*sin);
804 }
805 EXPORT_SYMBOL(inet_getname);
806 
inet_send_prepare(struct sock * sk)807 int inet_send_prepare(struct sock *sk)
808 {
809 	sock_rps_record_flow(sk);
810 
811 	/* We may need to bind the socket. */
812 	if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind &&
813 	    inet_autobind(sk))
814 		return -EAGAIN;
815 
816 	return 0;
817 }
818 EXPORT_SYMBOL_GPL(inet_send_prepare);
819 
inet_sendmsg(struct socket * sock,struct msghdr * msg,size_t size)820 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
821 {
822 	struct sock *sk = sock->sk;
823 
824 	if (unlikely(inet_send_prepare(sk)))
825 		return -EAGAIN;
826 
827 	return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
828 			       sk, msg, size);
829 }
830 EXPORT_SYMBOL(inet_sendmsg);
831 
inet_sendpage(struct socket * sock,struct page * page,int offset,size_t size,int flags)832 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
833 		      size_t size, int flags)
834 {
835 	struct sock *sk = sock->sk;
836 	const struct proto *prot;
837 
838 	if (unlikely(inet_send_prepare(sk)))
839 		return -EAGAIN;
840 
841 	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
842 	prot = READ_ONCE(sk->sk_prot);
843 	if (prot->sendpage)
844 		return prot->sendpage(sk, page, offset, size, flags);
845 	return sock_no_sendpage(sock, page, offset, size, flags);
846 }
847 EXPORT_SYMBOL(inet_sendpage);
848 
849 INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
850 					  size_t, int, int *));
inet_recvmsg(struct socket * sock,struct msghdr * msg,size_t size,int flags)851 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
852 		 int flags)
853 {
854 	struct sock *sk = sock->sk;
855 	int addr_len = 0;
856 	int err;
857 
858 	if (likely(!(flags & MSG_ERRQUEUE)))
859 		sock_rps_record_flow(sk);
860 
861 	err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
862 			      sk, msg, size, flags, &addr_len);
863 	if (err >= 0)
864 		msg->msg_namelen = addr_len;
865 	return err;
866 }
867 EXPORT_SYMBOL(inet_recvmsg);
868 
inet_shutdown(struct socket * sock,int how)869 int inet_shutdown(struct socket *sock, int how)
870 {
871 	struct sock *sk = sock->sk;
872 	int err = 0;
873 
874 	/* This should really check to make sure
875 	 * the socket is a TCP socket. (WHY AC...)
876 	 */
877 	how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
878 		       1->2 bit 2 snds.
879 		       2->3 */
880 	if ((how & ~SHUTDOWN_MASK) || !how)	/* MAXINT->0 */
881 		return -EINVAL;
882 
883 	lock_sock(sk);
884 	if (sock->state == SS_CONNECTING) {
885 		if ((1 << sk->sk_state) &
886 		    (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
887 			sock->state = SS_DISCONNECTING;
888 		else
889 			sock->state = SS_CONNECTED;
890 	}
891 
892 	switch (sk->sk_state) {
893 	case TCP_CLOSE:
894 		err = -ENOTCONN;
895 		/* Hack to wake up other listeners, who can poll for
896 		   EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
897 		fallthrough;
898 	default:
899 		sk->sk_shutdown |= how;
900 		if (sk->sk_prot->shutdown)
901 			sk->sk_prot->shutdown(sk, how);
902 		break;
903 
904 	/* Remaining two branches are temporary solution for missing
905 	 * close() in multithreaded environment. It is _not_ a good idea,
906 	 * but we have no choice until close() is repaired at VFS level.
907 	 */
908 	case TCP_LISTEN:
909 		if (!(how & RCV_SHUTDOWN))
910 			break;
911 		fallthrough;
912 	case TCP_SYN_SENT:
913 		err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
914 		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
915 		break;
916 	}
917 
918 	/* Wake up anyone sleeping in poll. */
919 	sk->sk_state_change(sk);
920 	release_sock(sk);
921 	return err;
922 }
923 EXPORT_SYMBOL(inet_shutdown);
924 
925 /*
926  *	ioctl() calls you can issue on an INET socket. Most of these are
927  *	device configuration and stuff and very rarely used. Some ioctls
928  *	pass on to the socket itself.
929  *
930  *	NOTE: I like the idea of a module for the config stuff. ie ifconfig
931  *	loads the devconfigure module does its configuring and unloads it.
932  *	There's a good 20K of config code hanging around the kernel.
933  */
934 
inet_ioctl(struct socket * sock,unsigned int cmd,unsigned long arg)935 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
936 {
937 	struct sock *sk = sock->sk;
938 	int err = 0;
939 	struct net *net = sock_net(sk);
940 	void __user *p = (void __user *)arg;
941 	struct ifreq ifr;
942 	struct rtentry rt;
943 
944 	switch (cmd) {
945 	case SIOCADDRT:
946 	case SIOCDELRT:
947 		if (copy_from_user(&rt, p, sizeof(struct rtentry)))
948 			return -EFAULT;
949 		err = ip_rt_ioctl(net, cmd, &rt);
950 		break;
951 	case SIOCRTMSG:
952 		err = -EINVAL;
953 		break;
954 	case SIOCDARP:
955 	case SIOCGARP:
956 	case SIOCSARP:
957 		err = arp_ioctl(net, cmd, (void __user *)arg);
958 		break;
959 	case SIOCGIFADDR:
960 	case SIOCGIFBRDADDR:
961 	case SIOCGIFNETMASK:
962 	case SIOCGIFDSTADDR:
963 	case SIOCGIFPFLAGS:
964 		if (get_user_ifreq(&ifr, NULL, p))
965 			return -EFAULT;
966 		err = devinet_ioctl(net, cmd, &ifr);
967 		if (!err && put_user_ifreq(&ifr, p))
968 			err = -EFAULT;
969 		break;
970 
971 	case SIOCSIFADDR:
972 	case SIOCSIFBRDADDR:
973 	case SIOCSIFNETMASK:
974 	case SIOCSIFDSTADDR:
975 	case SIOCSIFPFLAGS:
976 	case SIOCSIFFLAGS:
977 		if (get_user_ifreq(&ifr, NULL, p))
978 			return -EFAULT;
979 		err = devinet_ioctl(net, cmd, &ifr);
980 		break;
981 	default:
982 		if (sk->sk_prot->ioctl)
983 			err = sk->sk_prot->ioctl(sk, cmd, arg);
984 		else
985 			err = -ENOIOCTLCMD;
986 		break;
987 	}
988 	return err;
989 }
990 EXPORT_SYMBOL(inet_ioctl);
991 
992 #ifdef CONFIG_COMPAT
inet_compat_routing_ioctl(struct sock * sk,unsigned int cmd,struct compat_rtentry __user * ur)993 static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd,
994 		struct compat_rtentry __user *ur)
995 {
996 	compat_uptr_t rtdev;
997 	struct rtentry rt;
998 
999 	if (copy_from_user(&rt.rt_dst, &ur->rt_dst,
1000 			3 * sizeof(struct sockaddr)) ||
1001 	    get_user(rt.rt_flags, &ur->rt_flags) ||
1002 	    get_user(rt.rt_metric, &ur->rt_metric) ||
1003 	    get_user(rt.rt_mtu, &ur->rt_mtu) ||
1004 	    get_user(rt.rt_window, &ur->rt_window) ||
1005 	    get_user(rt.rt_irtt, &ur->rt_irtt) ||
1006 	    get_user(rtdev, &ur->rt_dev))
1007 		return -EFAULT;
1008 
1009 	rt.rt_dev = compat_ptr(rtdev);
1010 	return ip_rt_ioctl(sock_net(sk), cmd, &rt);
1011 }
1012 
inet_compat_ioctl(struct socket * sock,unsigned int cmd,unsigned long arg)1013 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1014 {
1015 	void __user *argp = compat_ptr(arg);
1016 	struct sock *sk = sock->sk;
1017 
1018 	switch (cmd) {
1019 	case SIOCADDRT:
1020 	case SIOCDELRT:
1021 		return inet_compat_routing_ioctl(sk, cmd, argp);
1022 	default:
1023 		if (!sk->sk_prot->compat_ioctl)
1024 			return -ENOIOCTLCMD;
1025 		return sk->sk_prot->compat_ioctl(sk, cmd, arg);
1026 	}
1027 }
1028 #endif /* CONFIG_COMPAT */
1029 
1030 const struct proto_ops inet_stream_ops = {
1031 	.family		   = PF_INET,
1032 	.owner		   = THIS_MODULE,
1033 	.release	   = inet_release,
1034 	.bind		   = inet_bind,
1035 	.connect	   = inet_stream_connect,
1036 	.socketpair	   = sock_no_socketpair,
1037 	.accept		   = inet_accept,
1038 	.getname	   = inet_getname,
1039 	.poll		   = tcp_poll,
1040 	.ioctl		   = inet_ioctl,
1041 	.gettstamp	   = sock_gettstamp,
1042 	.listen		   = inet_listen,
1043 	.shutdown	   = inet_shutdown,
1044 	.setsockopt	   = sock_common_setsockopt,
1045 	.getsockopt	   = sock_common_getsockopt,
1046 	.sendmsg	   = inet_sendmsg,
1047 	.recvmsg	   = inet_recvmsg,
1048 #ifdef CONFIG_MMU
1049 	.mmap		   = tcp_mmap,
1050 #endif
1051 	.sendpage	   = inet_sendpage,
1052 	.splice_read	   = tcp_splice_read,
1053 	.read_sock	   = tcp_read_sock,
1054 	.read_skb	   = tcp_read_skb,
1055 	.sendmsg_locked    = tcp_sendmsg_locked,
1056 	.sendpage_locked   = tcp_sendpage_locked,
1057 	.peek_len	   = tcp_peek_len,
1058 #ifdef CONFIG_COMPAT
1059 	.compat_ioctl	   = inet_compat_ioctl,
1060 #endif
1061 	.set_rcvlowat	   = tcp_set_rcvlowat,
1062 };
1063 EXPORT_SYMBOL(inet_stream_ops);
1064 
1065 const struct proto_ops inet_dgram_ops = {
1066 	.family		   = PF_INET,
1067 	.owner		   = THIS_MODULE,
1068 	.release	   = inet_release,
1069 	.bind		   = inet_bind,
1070 	.connect	   = inet_dgram_connect,
1071 	.socketpair	   = sock_no_socketpair,
1072 	.accept		   = sock_no_accept,
1073 	.getname	   = inet_getname,
1074 	.poll		   = udp_poll,
1075 	.ioctl		   = inet_ioctl,
1076 	.gettstamp	   = sock_gettstamp,
1077 	.listen		   = sock_no_listen,
1078 	.shutdown	   = inet_shutdown,
1079 	.setsockopt	   = sock_common_setsockopt,
1080 	.getsockopt	   = sock_common_getsockopt,
1081 	.sendmsg	   = inet_sendmsg,
1082 	.read_skb	   = udp_read_skb,
1083 	.recvmsg	   = inet_recvmsg,
1084 	.mmap		   = sock_no_mmap,
1085 	.sendpage	   = inet_sendpage,
1086 	.set_peek_off	   = sk_set_peek_off,
1087 #ifdef CONFIG_COMPAT
1088 	.compat_ioctl	   = inet_compat_ioctl,
1089 #endif
1090 };
1091 EXPORT_SYMBOL(inet_dgram_ops);
1092 
1093 /*
1094  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1095  * udp_poll
1096  */
1097 static const struct proto_ops inet_sockraw_ops = {
1098 	.family		   = PF_INET,
1099 	.owner		   = THIS_MODULE,
1100 	.release	   = inet_release,
1101 	.bind		   = inet_bind,
1102 	.connect	   = inet_dgram_connect,
1103 	.socketpair	   = sock_no_socketpair,
1104 	.accept		   = sock_no_accept,
1105 	.getname	   = inet_getname,
1106 	.poll		   = datagram_poll,
1107 	.ioctl		   = inet_ioctl,
1108 	.gettstamp	   = sock_gettstamp,
1109 	.listen		   = sock_no_listen,
1110 	.shutdown	   = inet_shutdown,
1111 	.setsockopt	   = sock_common_setsockopt,
1112 	.getsockopt	   = sock_common_getsockopt,
1113 	.sendmsg	   = inet_sendmsg,
1114 	.recvmsg	   = inet_recvmsg,
1115 	.mmap		   = sock_no_mmap,
1116 	.sendpage	   = inet_sendpage,
1117 #ifdef CONFIG_COMPAT
1118 	.compat_ioctl	   = inet_compat_ioctl,
1119 #endif
1120 };
1121 
1122 static const struct net_proto_family inet_family_ops = {
1123 	.family = PF_INET,
1124 	.create = inet_create,
1125 	.owner	= THIS_MODULE,
1126 };
1127 
1128 /* Upon startup we insert all the elements in inetsw_array[] into
1129  * the linked list inetsw.
1130  */
1131 static struct inet_protosw inetsw_array[] =
1132 {
1133 	{
1134 		.type =       SOCK_STREAM,
1135 		.protocol =   IPPROTO_TCP,
1136 		.prot =       &tcp_prot,
1137 		.ops =        &inet_stream_ops,
1138 		.flags =      INET_PROTOSW_PERMANENT |
1139 			      INET_PROTOSW_ICSK,
1140 	},
1141 
1142 	{
1143 		.type =       SOCK_DGRAM,
1144 		.protocol =   IPPROTO_UDP,
1145 		.prot =       &udp_prot,
1146 		.ops =        &inet_dgram_ops,
1147 		.flags =      INET_PROTOSW_PERMANENT,
1148        },
1149 
1150        {
1151 		.type =       SOCK_DGRAM,
1152 		.protocol =   IPPROTO_ICMP,
1153 		.prot =       &ping_prot,
1154 		.ops =        &inet_sockraw_ops,
1155 		.flags =      INET_PROTOSW_REUSE,
1156        },
1157 
1158        {
1159 	       .type =       SOCK_RAW,
1160 	       .protocol =   IPPROTO_IP,	/* wild card */
1161 	       .prot =       &raw_prot,
1162 	       .ops =        &inet_sockraw_ops,
1163 	       .flags =      INET_PROTOSW_REUSE,
1164        }
1165 };
1166 
1167 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1168 
inet_register_protosw(struct inet_protosw * p)1169 void inet_register_protosw(struct inet_protosw *p)
1170 {
1171 	struct list_head *lh;
1172 	struct inet_protosw *answer;
1173 	int protocol = p->protocol;
1174 	struct list_head *last_perm;
1175 
1176 	spin_lock_bh(&inetsw_lock);
1177 
1178 	if (p->type >= SOCK_MAX)
1179 		goto out_illegal;
1180 
1181 	/* If we are trying to override a permanent protocol, bail. */
1182 	last_perm = &inetsw[p->type];
1183 	list_for_each(lh, &inetsw[p->type]) {
1184 		answer = list_entry(lh, struct inet_protosw, list);
1185 		/* Check only the non-wild match. */
1186 		if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1187 			break;
1188 		if (protocol == answer->protocol)
1189 			goto out_permanent;
1190 		last_perm = lh;
1191 	}
1192 
1193 	/* Add the new entry after the last permanent entry if any, so that
1194 	 * the new entry does not override a permanent entry when matched with
1195 	 * a wild-card protocol. But it is allowed to override any existing
1196 	 * non-permanent entry.  This means that when we remove this entry, the
1197 	 * system automatically returns to the old behavior.
1198 	 */
1199 	list_add_rcu(&p->list, last_perm);
1200 out:
1201 	spin_unlock_bh(&inetsw_lock);
1202 
1203 	return;
1204 
1205 out_permanent:
1206 	pr_err("Attempt to override permanent protocol %d\n", protocol);
1207 	goto out;
1208 
1209 out_illegal:
1210 	pr_err("Ignoring attempt to register invalid socket type %d\n",
1211 	       p->type);
1212 	goto out;
1213 }
1214 EXPORT_SYMBOL(inet_register_protosw);
1215 
inet_unregister_protosw(struct inet_protosw * p)1216 void inet_unregister_protosw(struct inet_protosw *p)
1217 {
1218 	if (INET_PROTOSW_PERMANENT & p->flags) {
1219 		pr_err("Attempt to unregister permanent protocol %d\n",
1220 		       p->protocol);
1221 	} else {
1222 		spin_lock_bh(&inetsw_lock);
1223 		list_del_rcu(&p->list);
1224 		spin_unlock_bh(&inetsw_lock);
1225 
1226 		synchronize_net();
1227 	}
1228 }
1229 EXPORT_SYMBOL(inet_unregister_protosw);
1230 
inet_sk_reselect_saddr(struct sock * sk)1231 static int inet_sk_reselect_saddr(struct sock *sk)
1232 {
1233 	struct inet_sock *inet = inet_sk(sk);
1234 	__be32 old_saddr = inet->inet_saddr;
1235 	__be32 daddr = inet->inet_daddr;
1236 	struct flowi4 *fl4;
1237 	struct rtable *rt;
1238 	__be32 new_saddr;
1239 	struct ip_options_rcu *inet_opt;
1240 	int err;
1241 
1242 	inet_opt = rcu_dereference_protected(inet->inet_opt,
1243 					     lockdep_sock_is_held(sk));
1244 	if (inet_opt && inet_opt->opt.srr)
1245 		daddr = inet_opt->opt.faddr;
1246 
1247 	/* Query new route. */
1248 	fl4 = &inet->cork.fl.u.ip4;
1249 	rt = ip_route_connect(fl4, daddr, 0, sk->sk_bound_dev_if,
1250 			      sk->sk_protocol, inet->inet_sport,
1251 			      inet->inet_dport, sk);
1252 	if (IS_ERR(rt))
1253 		return PTR_ERR(rt);
1254 
1255 	new_saddr = fl4->saddr;
1256 
1257 	if (new_saddr == old_saddr) {
1258 		sk_setup_caps(sk, &rt->dst);
1259 		return 0;
1260 	}
1261 
1262 	err = inet_bhash2_update_saddr(sk, &new_saddr, AF_INET);
1263 	if (err) {
1264 		ip_rt_put(rt);
1265 		return err;
1266 	}
1267 
1268 	sk_setup_caps(sk, &rt->dst);
1269 
1270 	if (READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) > 1) {
1271 		pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1272 			__func__, &old_saddr, &new_saddr);
1273 	}
1274 
1275 	/*
1276 	 * XXX The only one ugly spot where we need to
1277 	 * XXX really change the sockets identity after
1278 	 * XXX it has entered the hashes. -DaveM
1279 	 *
1280 	 * Besides that, it does not check for connection
1281 	 * uniqueness. Wait for troubles.
1282 	 */
1283 	return __sk_prot_rehash(sk);
1284 }
1285 
inet_sk_rebuild_header(struct sock * sk)1286 int inet_sk_rebuild_header(struct sock *sk)
1287 {
1288 	struct inet_sock *inet = inet_sk(sk);
1289 	struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1290 	__be32 daddr;
1291 	struct ip_options_rcu *inet_opt;
1292 	struct flowi4 *fl4;
1293 	int err;
1294 
1295 	/* Route is OK, nothing to do. */
1296 	if (rt)
1297 		return 0;
1298 
1299 	/* Reroute. */
1300 	rcu_read_lock();
1301 	inet_opt = rcu_dereference(inet->inet_opt);
1302 	daddr = inet->inet_daddr;
1303 	if (inet_opt && inet_opt->opt.srr)
1304 		daddr = inet_opt->opt.faddr;
1305 	rcu_read_unlock();
1306 	fl4 = &inet->cork.fl.u.ip4;
1307 	rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1308 				   inet->inet_dport, inet->inet_sport,
1309 				   sk->sk_protocol, RT_CONN_FLAGS(sk),
1310 				   sk->sk_bound_dev_if);
1311 	if (!IS_ERR(rt)) {
1312 		err = 0;
1313 		sk_setup_caps(sk, &rt->dst);
1314 	} else {
1315 		err = PTR_ERR(rt);
1316 
1317 		/* Routing failed... */
1318 		sk->sk_route_caps = 0;
1319 		/*
1320 		 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1321 		 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1322 		 */
1323 		if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) ||
1324 		    sk->sk_state != TCP_SYN_SENT ||
1325 		    (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1326 		    (err = inet_sk_reselect_saddr(sk)) != 0)
1327 			sk->sk_err_soft = -err;
1328 	}
1329 
1330 	return err;
1331 }
1332 EXPORT_SYMBOL(inet_sk_rebuild_header);
1333 
inet_sk_set_state(struct sock * sk,int state)1334 void inet_sk_set_state(struct sock *sk, int state)
1335 {
1336 	trace_inet_sock_set_state(sk, sk->sk_state, state);
1337 	sk->sk_state = state;
1338 }
1339 EXPORT_SYMBOL(inet_sk_set_state);
1340 
inet_sk_state_store(struct sock * sk,int newstate)1341 void inet_sk_state_store(struct sock *sk, int newstate)
1342 {
1343 	trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1344 	smp_store_release(&sk->sk_state, newstate);
1345 }
1346 
inet_gso_segment(struct sk_buff * skb,netdev_features_t features)1347 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1348 				 netdev_features_t features)
1349 {
1350 	bool udpfrag = false, fixedid = false, gso_partial, encap;
1351 	struct sk_buff *segs = ERR_PTR(-EINVAL);
1352 	const struct net_offload *ops;
1353 	unsigned int offset = 0;
1354 	struct iphdr *iph;
1355 	int proto, tot_len;
1356 	int nhoff;
1357 	int ihl;
1358 	int id;
1359 
1360 	skb_reset_network_header(skb);
1361 	nhoff = skb_network_header(skb) - skb_mac_header(skb);
1362 	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1363 		goto out;
1364 
1365 	iph = ip_hdr(skb);
1366 	ihl = iph->ihl * 4;
1367 	if (ihl < sizeof(*iph))
1368 		goto out;
1369 
1370 	id = ntohs(iph->id);
1371 	proto = iph->protocol;
1372 
1373 	/* Warning: after this point, iph might be no longer valid */
1374 	if (unlikely(!pskb_may_pull(skb, ihl)))
1375 		goto out;
1376 	__skb_pull(skb, ihl);
1377 
1378 	encap = SKB_GSO_CB(skb)->encap_level > 0;
1379 	if (encap)
1380 		features &= skb->dev->hw_enc_features;
1381 	SKB_GSO_CB(skb)->encap_level += ihl;
1382 
1383 	skb_reset_transport_header(skb);
1384 
1385 	segs = ERR_PTR(-EPROTONOSUPPORT);
1386 
1387 	if (!skb->encapsulation || encap) {
1388 		udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1389 		fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1390 
1391 		/* fixed ID is invalid if DF bit is not set */
1392 		if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1393 			goto out;
1394 	}
1395 
1396 	ops = rcu_dereference(inet_offloads[proto]);
1397 	if (likely(ops && ops->callbacks.gso_segment)) {
1398 		segs = ops->callbacks.gso_segment(skb, features);
1399 		if (!segs)
1400 			skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
1401 	}
1402 
1403 	if (IS_ERR_OR_NULL(segs))
1404 		goto out;
1405 
1406 	gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1407 
1408 	skb = segs;
1409 	do {
1410 		iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1411 		if (udpfrag) {
1412 			iph->frag_off = htons(offset >> 3);
1413 			if (skb->next)
1414 				iph->frag_off |= htons(IP_MF);
1415 			offset += skb->len - nhoff - ihl;
1416 			tot_len = skb->len - nhoff;
1417 		} else if (skb_is_gso(skb)) {
1418 			if (!fixedid) {
1419 				iph->id = htons(id);
1420 				id += skb_shinfo(skb)->gso_segs;
1421 			}
1422 
1423 			if (gso_partial)
1424 				tot_len = skb_shinfo(skb)->gso_size +
1425 					  SKB_GSO_CB(skb)->data_offset +
1426 					  skb->head - (unsigned char *)iph;
1427 			else
1428 				tot_len = skb->len - nhoff;
1429 		} else {
1430 			if (!fixedid)
1431 				iph->id = htons(id++);
1432 			tot_len = skb->len - nhoff;
1433 		}
1434 		iph->tot_len = htons(tot_len);
1435 		ip_send_check(iph);
1436 		if (encap)
1437 			skb_reset_inner_headers(skb);
1438 		skb->network_header = (u8 *)iph - skb->head;
1439 		skb_reset_mac_len(skb);
1440 	} while ((skb = skb->next));
1441 
1442 out:
1443 	return segs;
1444 }
1445 
ipip_gso_segment(struct sk_buff * skb,netdev_features_t features)1446 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1447 					netdev_features_t features)
1448 {
1449 	if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1450 		return ERR_PTR(-EINVAL);
1451 
1452 	return inet_gso_segment(skb, features);
1453 }
1454 
inet_gro_receive(struct list_head * head,struct sk_buff * skb)1455 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1456 {
1457 	const struct net_offload *ops;
1458 	struct sk_buff *pp = NULL;
1459 	const struct iphdr *iph;
1460 	struct sk_buff *p;
1461 	unsigned int hlen;
1462 	unsigned int off;
1463 	unsigned int id;
1464 	int flush = 1;
1465 	int proto;
1466 
1467 	off = skb_gro_offset(skb);
1468 	hlen = off + sizeof(*iph);
1469 	iph = skb_gro_header(skb, hlen, off);
1470 	if (unlikely(!iph))
1471 		goto out;
1472 
1473 	proto = iph->protocol;
1474 
1475 	ops = rcu_dereference(inet_offloads[proto]);
1476 	if (!ops || !ops->callbacks.gro_receive)
1477 		goto out;
1478 
1479 	if (*(u8 *)iph != 0x45)
1480 		goto out;
1481 
1482 	if (ip_is_fragment(iph))
1483 		goto out;
1484 
1485 	if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1486 		goto out;
1487 
1488 	id = ntohl(*(__be32 *)&iph->id);
1489 	flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1490 	id >>= 16;
1491 
1492 	list_for_each_entry(p, head, list) {
1493 		struct iphdr *iph2;
1494 		u16 flush_id;
1495 
1496 		if (!NAPI_GRO_CB(p)->same_flow)
1497 			continue;
1498 
1499 		iph2 = (struct iphdr *)(p->data + off);
1500 		/* The above works because, with the exception of the top
1501 		 * (inner most) layer, we only aggregate pkts with the same
1502 		 * hdr length so all the hdrs we'll need to verify will start
1503 		 * at the same offset.
1504 		 */
1505 		if ((iph->protocol ^ iph2->protocol) |
1506 		    ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1507 		    ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1508 			NAPI_GRO_CB(p)->same_flow = 0;
1509 			continue;
1510 		}
1511 
1512 		/* All fields must match except length and checksum. */
1513 		NAPI_GRO_CB(p)->flush |=
1514 			(iph->ttl ^ iph2->ttl) |
1515 			(iph->tos ^ iph2->tos) |
1516 			((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1517 
1518 		NAPI_GRO_CB(p)->flush |= flush;
1519 
1520 		/* We need to store of the IP ID check to be included later
1521 		 * when we can verify that this packet does in fact belong
1522 		 * to a given flow.
1523 		 */
1524 		flush_id = (u16)(id - ntohs(iph2->id));
1525 
1526 		/* This bit of code makes it much easier for us to identify
1527 		 * the cases where we are doing atomic vs non-atomic IP ID
1528 		 * checks.  Specifically an atomic check can return IP ID
1529 		 * values 0 - 0xFFFF, while a non-atomic check can only
1530 		 * return 0 or 0xFFFF.
1531 		 */
1532 		if (!NAPI_GRO_CB(p)->is_atomic ||
1533 		    !(iph->frag_off & htons(IP_DF))) {
1534 			flush_id ^= NAPI_GRO_CB(p)->count;
1535 			flush_id = flush_id ? 0xFFFF : 0;
1536 		}
1537 
1538 		/* If the previous IP ID value was based on an atomic
1539 		 * datagram we can overwrite the value and ignore it.
1540 		 */
1541 		if (NAPI_GRO_CB(skb)->is_atomic)
1542 			NAPI_GRO_CB(p)->flush_id = flush_id;
1543 		else
1544 			NAPI_GRO_CB(p)->flush_id |= flush_id;
1545 	}
1546 
1547 	NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1548 	NAPI_GRO_CB(skb)->flush |= flush;
1549 	skb_set_network_header(skb, off);
1550 	/* The above will be needed by the transport layer if there is one
1551 	 * immediately following this IP hdr.
1552 	 */
1553 
1554 	/* Note : No need to call skb_gro_postpull_rcsum() here,
1555 	 * as we already checked checksum over ipv4 header was 0
1556 	 */
1557 	skb_gro_pull(skb, sizeof(*iph));
1558 	skb_set_transport_header(skb, skb_gro_offset(skb));
1559 
1560 	pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1561 				       ops->callbacks.gro_receive, head, skb);
1562 
1563 out:
1564 	skb_gro_flush_final(skb, pp, flush);
1565 
1566 	return pp;
1567 }
1568 
ipip_gro_receive(struct list_head * head,struct sk_buff * skb)1569 static struct sk_buff *ipip_gro_receive(struct list_head *head,
1570 					struct sk_buff *skb)
1571 {
1572 	if (NAPI_GRO_CB(skb)->encap_mark) {
1573 		NAPI_GRO_CB(skb)->flush = 1;
1574 		return NULL;
1575 	}
1576 
1577 	NAPI_GRO_CB(skb)->encap_mark = 1;
1578 
1579 	return inet_gro_receive(head, skb);
1580 }
1581 
1582 #define SECONDS_PER_DAY	86400
1583 
1584 /* inet_current_timestamp - Return IP network timestamp
1585  *
1586  * Return milliseconds since midnight in network byte order.
1587  */
inet_current_timestamp(void)1588 __be32 inet_current_timestamp(void)
1589 {
1590 	u32 secs;
1591 	u32 msecs;
1592 	struct timespec64 ts;
1593 
1594 	ktime_get_real_ts64(&ts);
1595 
1596 	/* Get secs since midnight. */
1597 	(void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1598 	/* Convert to msecs. */
1599 	msecs = secs * MSEC_PER_SEC;
1600 	/* Convert nsec to msec. */
1601 	msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1602 
1603 	/* Convert to network byte order. */
1604 	return htonl(msecs);
1605 }
1606 EXPORT_SYMBOL(inet_current_timestamp);
1607 
inet_recv_error(struct sock * sk,struct msghdr * msg,int len,int * addr_len)1608 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1609 {
1610 	if (sk->sk_family == AF_INET)
1611 		return ip_recv_error(sk, msg, len, addr_len);
1612 #if IS_ENABLED(CONFIG_IPV6)
1613 	if (sk->sk_family == AF_INET6)
1614 		return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1615 #endif
1616 	return -EINVAL;
1617 }
1618 
inet_gro_complete(struct sk_buff * skb,int nhoff)1619 int inet_gro_complete(struct sk_buff *skb, int nhoff)
1620 {
1621 	__be16 newlen = htons(skb->len - nhoff);
1622 	struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1623 	const struct net_offload *ops;
1624 	int proto = iph->protocol;
1625 	int err = -ENOSYS;
1626 
1627 	if (skb->encapsulation) {
1628 		skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1629 		skb_set_inner_network_header(skb, nhoff);
1630 	}
1631 
1632 	csum_replace2(&iph->check, iph->tot_len, newlen);
1633 	iph->tot_len = newlen;
1634 
1635 	ops = rcu_dereference(inet_offloads[proto]);
1636 	if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1637 		goto out;
1638 
1639 	/* Only need to add sizeof(*iph) to get to the next hdr below
1640 	 * because any hdr with option will have been flushed in
1641 	 * inet_gro_receive().
1642 	 */
1643 	err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1644 			      tcp4_gro_complete, udp4_gro_complete,
1645 			      skb, nhoff + sizeof(*iph));
1646 
1647 out:
1648 	return err;
1649 }
1650 
ipip_gro_complete(struct sk_buff * skb,int nhoff)1651 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1652 {
1653 	skb->encapsulation = 1;
1654 	skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1655 	return inet_gro_complete(skb, nhoff);
1656 }
1657 
inet_ctl_sock_create(struct sock ** sk,unsigned short family,unsigned short type,unsigned char protocol,struct net * net)1658 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1659 			 unsigned short type, unsigned char protocol,
1660 			 struct net *net)
1661 {
1662 	struct socket *sock;
1663 	int rc = sock_create_kern(net, family, type, protocol, &sock);
1664 
1665 	if (rc == 0) {
1666 		*sk = sock->sk;
1667 		(*sk)->sk_allocation = GFP_ATOMIC;
1668 		/*
1669 		 * Unhash it so that IP input processing does not even see it,
1670 		 * we do not wish this socket to see incoming packets.
1671 		 */
1672 		(*sk)->sk_prot->unhash(*sk);
1673 	}
1674 	return rc;
1675 }
1676 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1677 
snmp_fold_field(void __percpu * mib,int offt)1678 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1679 {
1680 	unsigned long res = 0;
1681 	int i;
1682 
1683 	for_each_possible_cpu(i)
1684 		res += snmp_get_cpu_field(mib, i, offt);
1685 	return res;
1686 }
1687 EXPORT_SYMBOL_GPL(snmp_fold_field);
1688 
1689 #if BITS_PER_LONG==32
1690 
snmp_get_cpu_field64(void __percpu * mib,int cpu,int offt,size_t syncp_offset)1691 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1692 			 size_t syncp_offset)
1693 {
1694 	void *bhptr;
1695 	struct u64_stats_sync *syncp;
1696 	u64 v;
1697 	unsigned int start;
1698 
1699 	bhptr = per_cpu_ptr(mib, cpu);
1700 	syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1701 	do {
1702 		start = u64_stats_fetch_begin_irq(syncp);
1703 		v = *(((u64 *)bhptr) + offt);
1704 	} while (u64_stats_fetch_retry_irq(syncp, start));
1705 
1706 	return v;
1707 }
1708 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1709 
snmp_fold_field64(void __percpu * mib,int offt,size_t syncp_offset)1710 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1711 {
1712 	u64 res = 0;
1713 	int cpu;
1714 
1715 	for_each_possible_cpu(cpu) {
1716 		res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1717 	}
1718 	return res;
1719 }
1720 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1721 #endif
1722 
1723 #ifdef CONFIG_IP_MULTICAST
1724 static const struct net_protocol igmp_protocol = {
1725 	.handler =	igmp_rcv,
1726 };
1727 #endif
1728 
1729 static const struct net_protocol tcp_protocol = {
1730 	.handler	=	tcp_v4_rcv,
1731 	.err_handler	=	tcp_v4_err,
1732 	.no_policy	=	1,
1733 	.icmp_strict_tag_validation = 1,
1734 };
1735 
1736 static const struct net_protocol udp_protocol = {
1737 	.handler =	udp_rcv,
1738 	.err_handler =	udp_err,
1739 	.no_policy =	1,
1740 };
1741 
1742 static const struct net_protocol icmp_protocol = {
1743 	.handler =	icmp_rcv,
1744 	.err_handler =	icmp_err,
1745 	.no_policy =	1,
1746 };
1747 
ipv4_mib_init_net(struct net * net)1748 static __net_init int ipv4_mib_init_net(struct net *net)
1749 {
1750 	int i;
1751 
1752 	net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1753 	if (!net->mib.tcp_statistics)
1754 		goto err_tcp_mib;
1755 	net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1756 	if (!net->mib.ip_statistics)
1757 		goto err_ip_mib;
1758 
1759 	for_each_possible_cpu(i) {
1760 		struct ipstats_mib *af_inet_stats;
1761 		af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1762 		u64_stats_init(&af_inet_stats->syncp);
1763 	}
1764 
1765 	net->mib.net_statistics = alloc_percpu(struct linux_mib);
1766 	if (!net->mib.net_statistics)
1767 		goto err_net_mib;
1768 	net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1769 	if (!net->mib.udp_statistics)
1770 		goto err_udp_mib;
1771 	net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1772 	if (!net->mib.udplite_statistics)
1773 		goto err_udplite_mib;
1774 	net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1775 	if (!net->mib.icmp_statistics)
1776 		goto err_icmp_mib;
1777 	net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1778 					      GFP_KERNEL);
1779 	if (!net->mib.icmpmsg_statistics)
1780 		goto err_icmpmsg_mib;
1781 
1782 	tcp_mib_init(net);
1783 	return 0;
1784 
1785 err_icmpmsg_mib:
1786 	free_percpu(net->mib.icmp_statistics);
1787 err_icmp_mib:
1788 	free_percpu(net->mib.udplite_statistics);
1789 err_udplite_mib:
1790 	free_percpu(net->mib.udp_statistics);
1791 err_udp_mib:
1792 	free_percpu(net->mib.net_statistics);
1793 err_net_mib:
1794 	free_percpu(net->mib.ip_statistics);
1795 err_ip_mib:
1796 	free_percpu(net->mib.tcp_statistics);
1797 err_tcp_mib:
1798 	return -ENOMEM;
1799 }
1800 
ipv4_mib_exit_net(struct net * net)1801 static __net_exit void ipv4_mib_exit_net(struct net *net)
1802 {
1803 	kfree(net->mib.icmpmsg_statistics);
1804 	free_percpu(net->mib.icmp_statistics);
1805 	free_percpu(net->mib.udplite_statistics);
1806 	free_percpu(net->mib.udp_statistics);
1807 	free_percpu(net->mib.net_statistics);
1808 	free_percpu(net->mib.ip_statistics);
1809 	free_percpu(net->mib.tcp_statistics);
1810 #ifdef CONFIG_MPTCP
1811 	/* allocated on demand, see mptcp_init_sock() */
1812 	free_percpu(net->mib.mptcp_statistics);
1813 #endif
1814 }
1815 
1816 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1817 	.init = ipv4_mib_init_net,
1818 	.exit = ipv4_mib_exit_net,
1819 };
1820 
init_ipv4_mibs(void)1821 static int __init init_ipv4_mibs(void)
1822 {
1823 	return register_pernet_subsys(&ipv4_mib_ops);
1824 }
1825 
inet_init_net(struct net * net)1826 static __net_init int inet_init_net(struct net *net)
1827 {
1828 	/*
1829 	 * Set defaults for local port range
1830 	 */
1831 	seqlock_init(&net->ipv4.ip_local_ports.lock);
1832 	net->ipv4.ip_local_ports.range[0] =  32768;
1833 	net->ipv4.ip_local_ports.range[1] =  60999;
1834 
1835 	seqlock_init(&net->ipv4.ping_group_range.lock);
1836 	/*
1837 	 * Sane defaults - nobody may create ping sockets.
1838 	 * Boot scripts should set this to distro-specific group.
1839 	 */
1840 	net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1841 	net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1842 
1843 	/* Default values for sysctl-controlled parameters.
1844 	 * We set them here, in case sysctl is not compiled.
1845 	 */
1846 	net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1847 	net->ipv4.sysctl_ip_fwd_update_priority = 1;
1848 	net->ipv4.sysctl_ip_dynaddr = 0;
1849 	net->ipv4.sysctl_ip_early_demux = 1;
1850 	net->ipv4.sysctl_udp_early_demux = 1;
1851 	net->ipv4.sysctl_tcp_early_demux = 1;
1852 	net->ipv4.sysctl_nexthop_compat_mode = 1;
1853 #ifdef CONFIG_SYSCTL
1854 	net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1855 #endif
1856 
1857 	/* Some igmp sysctl, whose values are always used */
1858 	net->ipv4.sysctl_igmp_max_memberships = 20;
1859 	net->ipv4.sysctl_igmp_max_msf = 10;
1860 	/* IGMP reports for link-local multicast groups are enabled by default */
1861 	net->ipv4.sysctl_igmp_llm_reports = 1;
1862 	net->ipv4.sysctl_igmp_qrv = 2;
1863 
1864 	net->ipv4.sysctl_fib_notify_on_flag_change = 0;
1865 
1866 	return 0;
1867 }
1868 
1869 static __net_initdata struct pernet_operations af_inet_ops = {
1870 	.init = inet_init_net,
1871 };
1872 
init_inet_pernet_ops(void)1873 static int __init init_inet_pernet_ops(void)
1874 {
1875 	return register_pernet_subsys(&af_inet_ops);
1876 }
1877 
1878 static int ipv4_proc_init(void);
1879 
1880 /*
1881  *	IP protocol layer initialiser
1882  */
1883 
1884 static struct packet_offload ip_packet_offload __read_mostly = {
1885 	.type = cpu_to_be16(ETH_P_IP),
1886 	.callbacks = {
1887 		.gso_segment = inet_gso_segment,
1888 		.gro_receive = inet_gro_receive,
1889 		.gro_complete = inet_gro_complete,
1890 	},
1891 };
1892 
1893 static const struct net_offload ipip_offload = {
1894 	.callbacks = {
1895 		.gso_segment	= ipip_gso_segment,
1896 		.gro_receive	= ipip_gro_receive,
1897 		.gro_complete	= ipip_gro_complete,
1898 	},
1899 };
1900 
ipip_offload_init(void)1901 static int __init ipip_offload_init(void)
1902 {
1903 	return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1904 }
1905 
ipv4_offload_init(void)1906 static int __init ipv4_offload_init(void)
1907 {
1908 	/*
1909 	 * Add offloads
1910 	 */
1911 	if (udpv4_offload_init() < 0)
1912 		pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1913 	if (tcpv4_offload_init() < 0)
1914 		pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1915 	if (ipip_offload_init() < 0)
1916 		pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1917 
1918 	dev_add_offload(&ip_packet_offload);
1919 	return 0;
1920 }
1921 
1922 fs_initcall(ipv4_offload_init);
1923 
1924 static struct packet_type ip_packet_type __read_mostly = {
1925 	.type = cpu_to_be16(ETH_P_IP),
1926 	.func = ip_rcv,
1927 	.list_func = ip_list_rcv,
1928 };
1929 
inet_init(void)1930 static int __init inet_init(void)
1931 {
1932 	struct inet_protosw *q;
1933 	struct list_head *r;
1934 	int rc;
1935 
1936 	sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1937 
1938 	raw_hashinfo_init(&raw_v4_hashinfo);
1939 
1940 	rc = proto_register(&tcp_prot, 1);
1941 	if (rc)
1942 		goto out;
1943 
1944 	rc = proto_register(&udp_prot, 1);
1945 	if (rc)
1946 		goto out_unregister_tcp_proto;
1947 
1948 	rc = proto_register(&raw_prot, 1);
1949 	if (rc)
1950 		goto out_unregister_udp_proto;
1951 
1952 	rc = proto_register(&ping_prot, 1);
1953 	if (rc)
1954 		goto out_unregister_raw_proto;
1955 
1956 	/*
1957 	 *	Tell SOCKET that we are alive...
1958 	 */
1959 
1960 	(void)sock_register(&inet_family_ops);
1961 
1962 #ifdef CONFIG_SYSCTL
1963 	ip_static_sysctl_init();
1964 #endif
1965 
1966 	/*
1967 	 *	Add all the base protocols.
1968 	 */
1969 
1970 	if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1971 		pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1972 	if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1973 		pr_crit("%s: Cannot add UDP protocol\n", __func__);
1974 	if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1975 		pr_crit("%s: Cannot add TCP protocol\n", __func__);
1976 #ifdef CONFIG_IP_MULTICAST
1977 	if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1978 		pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1979 #endif
1980 
1981 	/* Register the socket-side information for inet_create. */
1982 	for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1983 		INIT_LIST_HEAD(r);
1984 
1985 	for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1986 		inet_register_protosw(q);
1987 
1988 	/*
1989 	 *	Set the ARP module up
1990 	 */
1991 
1992 	arp_init();
1993 
1994 	/*
1995 	 *	Set the IP module up
1996 	 */
1997 
1998 	ip_init();
1999 
2000 	/* Initialise per-cpu ipv4 mibs */
2001 	if (init_ipv4_mibs())
2002 		panic("%s: Cannot init ipv4 mibs\n", __func__);
2003 
2004 	/* Setup TCP slab cache for open requests. */
2005 	tcp_init();
2006 
2007 	/* Setup UDP memory threshold */
2008 	udp_init();
2009 
2010 	/* Add UDP-Lite (RFC 3828) */
2011 	udplite4_register();
2012 
2013 	raw_init();
2014 
2015 	ping_init();
2016 
2017 	/*
2018 	 *	Set the ICMP layer up
2019 	 */
2020 
2021 	if (icmp_init() < 0)
2022 		panic("Failed to create the ICMP control socket.\n");
2023 
2024 	/*
2025 	 *	Initialise the multicast router
2026 	 */
2027 #if defined(CONFIG_IP_MROUTE)
2028 	if (ip_mr_init())
2029 		pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2030 #endif
2031 
2032 	if (init_inet_pernet_ops())
2033 		pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2034 
2035 	ipv4_proc_init();
2036 
2037 	ipfrag_init();
2038 
2039 	dev_add_pack(&ip_packet_type);
2040 
2041 	ip_tunnel_core_init();
2042 
2043 	rc = 0;
2044 out:
2045 	return rc;
2046 out_unregister_raw_proto:
2047 	proto_unregister(&raw_prot);
2048 out_unregister_udp_proto:
2049 	proto_unregister(&udp_prot);
2050 out_unregister_tcp_proto:
2051 	proto_unregister(&tcp_prot);
2052 	goto out;
2053 }
2054 
2055 fs_initcall(inet_init);
2056 
2057 /* ------------------------------------------------------------------------ */
2058 
2059 #ifdef CONFIG_PROC_FS
ipv4_proc_init(void)2060 static int __init ipv4_proc_init(void)
2061 {
2062 	int rc = 0;
2063 
2064 	if (raw_proc_init())
2065 		goto out_raw;
2066 	if (tcp4_proc_init())
2067 		goto out_tcp;
2068 	if (udp4_proc_init())
2069 		goto out_udp;
2070 	if (ping_proc_init())
2071 		goto out_ping;
2072 	if (ip_misc_proc_init())
2073 		goto out_misc;
2074 out:
2075 	return rc;
2076 out_misc:
2077 	ping_proc_exit();
2078 out_ping:
2079 	udp4_proc_exit();
2080 out_udp:
2081 	tcp4_proc_exit();
2082 out_tcp:
2083 	raw_proc_exit();
2084 out_raw:
2085 	rc = -ENOMEM;
2086 	goto out;
2087 }
2088 
2089 #else /* CONFIG_PROC_FS */
ipv4_proc_init(void)2090 static int __init ipv4_proc_init(void)
2091 {
2092 	return 0;
2093 }
2094 #endif /* CONFIG_PROC_FS */
2095