• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 	if (INET_PROTOSW_ICSK & answer_flags)
328 		inet_init_csk_locks(sk);
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 (READ_ONCE(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 	sk->sk_txrehash = READ_ONCE(net->core.sysctl_txrehash);
354 
355 	inet->uc_ttl	= -1;
356 	inet->mc_loop	= 1;
357 	inet->mc_ttl	= 1;
358 	inet->mc_all	= 1;
359 	inet->mc_index	= 0;
360 	inet->mc_list	= NULL;
361 	inet->rcv_tos	= 0;
362 
363 	sk_refcnt_debug_inc(sk);
364 
365 	if (inet->inet_num) {
366 		/* It assumes that any protocol which allows
367 		 * the user to assign a number at socket
368 		 * creation time automatically
369 		 * shares.
370 		 */
371 		inet->inet_sport = htons(inet->inet_num);
372 		/* Add to protocol hash chains. */
373 		err = sk->sk_prot->hash(sk);
374 		if (err) {
375 			sk_common_release(sk);
376 			goto out;
377 		}
378 	}
379 
380 	if (sk->sk_prot->init) {
381 		err = sk->sk_prot->init(sk);
382 		if (err) {
383 			sk_common_release(sk);
384 			goto out;
385 		}
386 	}
387 
388 	if (!kern) {
389 		err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
390 		if (err) {
391 			sk_common_release(sk);
392 			goto out;
393 		}
394 	}
395 out:
396 	return err;
397 out_rcu_unlock:
398 	rcu_read_unlock();
399 	goto out;
400 }
401 
402 
403 /*
404  *	The peer socket should always be NULL (or else). When we call this
405  *	function we are destroying the object and from then on nobody
406  *	should refer to it.
407  */
inet_release(struct socket * sock)408 int inet_release(struct socket *sock)
409 {
410 	struct sock *sk = sock->sk;
411 
412 	if (sk) {
413 		long timeout;
414 
415 		if (!sk->sk_kern_sock)
416 			BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk);
417 
418 		/* Applications forget to leave groups before exiting */
419 		ip_mc_drop_socket(sk);
420 
421 		/* If linger is set, we don't return until the close
422 		 * is complete.  Otherwise we return immediately. The
423 		 * actually closing is done the same either way.
424 		 *
425 		 * If the close is due to the process exiting, we never
426 		 * linger..
427 		 */
428 		timeout = 0;
429 		if (sock_flag(sk, SOCK_LINGER) &&
430 		    !(current->flags & PF_EXITING))
431 			timeout = sk->sk_lingertime;
432 		sk->sk_prot->close(sk, timeout);
433 		sock->sk = NULL;
434 	}
435 	return 0;
436 }
437 EXPORT_SYMBOL(inet_release);
438 
inet_bind(struct socket * sock,struct sockaddr * uaddr,int addr_len)439 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
440 {
441 	struct sock *sk = sock->sk;
442 	u32 flags = BIND_WITH_LOCK;
443 	int err;
444 
445 	/* If the socket has its own bind function then use it. (RAW) */
446 	if (sk->sk_prot->bind) {
447 		return sk->sk_prot->bind(sk, uaddr, addr_len);
448 	}
449 	if (addr_len < sizeof(struct sockaddr_in))
450 		return -EINVAL;
451 
452 	/* BPF prog is run before any checks are done so that if the prog
453 	 * changes context in a wrong way it will be caught.
454 	 */
455 	err = BPF_CGROUP_RUN_PROG_INET_BIND_LOCK(sk, uaddr,
456 						 CGROUP_INET4_BIND, &flags);
457 	if (err)
458 		return err;
459 
460 	return __inet_bind(sk, uaddr, addr_len, flags);
461 }
462 EXPORT_SYMBOL(inet_bind);
463 
__inet_bind(struct sock * sk,struct sockaddr * uaddr,int addr_len,u32 flags)464 int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
465 		u32 flags)
466 {
467 	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
468 	struct inet_sock *inet = inet_sk(sk);
469 	struct net *net = sock_net(sk);
470 	unsigned short snum;
471 	int chk_addr_ret;
472 	u32 tb_id = RT_TABLE_LOCAL;
473 	int err;
474 
475 	if (addr->sin_family != AF_INET) {
476 		/* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
477 		 * only if s_addr is INADDR_ANY.
478 		 */
479 		err = -EAFNOSUPPORT;
480 		if (addr->sin_family != AF_UNSPEC ||
481 		    addr->sin_addr.s_addr != htonl(INADDR_ANY))
482 			goto out;
483 	}
484 
485 	tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
486 	chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
487 
488 	/* Not specified by any standard per-se, however it breaks too
489 	 * many applications when removed.  It is unfortunate since
490 	 * allowing applications to make a non-local bind solves
491 	 * several problems with systems using dynamic addressing.
492 	 * (ie. your servers still start up even if your ISDN link
493 	 *  is temporarily down)
494 	 */
495 	err = -EADDRNOTAVAIL;
496 	if (!inet_addr_valid_or_nonlocal(net, inet, addr->sin_addr.s_addr,
497 	                                 chk_addr_ret))
498 		goto out;
499 
500 	snum = ntohs(addr->sin_port);
501 	err = -EACCES;
502 	if (!(flags & BIND_NO_CAP_NET_BIND_SERVICE) &&
503 	    snum && inet_port_requires_bind_service(net, snum) &&
504 	    !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
505 		goto out;
506 
507 	/*      We keep a pair of addresses. rcv_saddr is the one
508 	 *      used by hash lookups, and saddr is used for transmit.
509 	 *
510 	 *      In the BSD API these are the same except where it
511 	 *      would be illegal to use them (multicast/broadcast) in
512 	 *      which case the sending device address is used.
513 	 */
514 	if (flags & BIND_WITH_LOCK)
515 		lock_sock(sk);
516 
517 	/* Check these errors (active socket, double bind). */
518 	err = -EINVAL;
519 	if (sk->sk_state != TCP_CLOSE || inet->inet_num)
520 		goto out_release_sock;
521 
522 	inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
523 	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
524 		inet->inet_saddr = 0;  /* Use device */
525 
526 	/* Make sure we are allowed to bind here. */
527 	if (snum || !(inet->bind_address_no_port ||
528 		      (flags & BIND_FORCE_ADDRESS_NO_PORT))) {
529 		err = sk->sk_prot->get_port(sk, snum);
530 		if (err) {
531 			inet->inet_saddr = inet->inet_rcv_saddr = 0;
532 			goto out_release_sock;
533 		}
534 		if (!(flags & BIND_FROM_BPF)) {
535 			err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
536 			if (err) {
537 				inet->inet_saddr = inet->inet_rcv_saddr = 0;
538 				if (sk->sk_prot->put_port)
539 					sk->sk_prot->put_port(sk);
540 				goto out_release_sock;
541 			}
542 		}
543 	}
544 
545 	if (inet->inet_rcv_saddr)
546 		sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
547 	if (snum)
548 		sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
549 	inet->inet_sport = htons(inet->inet_num);
550 	inet->inet_daddr = 0;
551 	inet->inet_dport = 0;
552 	sk_dst_reset(sk);
553 	err = 0;
554 out_release_sock:
555 	if (flags & BIND_WITH_LOCK)
556 		release_sock(sk);
557 out:
558 	return err;
559 }
560 
inet_dgram_connect(struct socket * sock,struct sockaddr * uaddr,int addr_len,int flags)561 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
562 		       int addr_len, int flags)
563 {
564 	struct sock *sk = sock->sk;
565 	const struct proto *prot;
566 	int err;
567 
568 	if (addr_len < sizeof(uaddr->sa_family))
569 		return -EINVAL;
570 
571 	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
572 	prot = READ_ONCE(sk->sk_prot);
573 
574 	if (uaddr->sa_family == AF_UNSPEC)
575 		return prot->disconnect(sk, flags);
576 
577 	if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
578 		err = prot->pre_connect(sk, uaddr, addr_len);
579 		if (err)
580 			return err;
581 	}
582 
583 	if (data_race(!inet_sk(sk)->inet_num) && inet_autobind(sk))
584 		return -EAGAIN;
585 	return prot->connect(sk, uaddr, addr_len);
586 }
587 EXPORT_SYMBOL(inet_dgram_connect);
588 
inet_wait_for_connect(struct sock * sk,long timeo,int writebias)589 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
590 {
591 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
592 
593 	add_wait_queue(sk_sleep(sk), &wait);
594 	sk->sk_write_pending += writebias;
595 	sk->sk_wait_pending++;
596 
597 	/* Basic assumption: if someone sets sk->sk_err, he _must_
598 	 * change state of the socket from TCP_SYN_*.
599 	 * Connect() does not allow to get error notifications
600 	 * without closing the socket.
601 	 */
602 	while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
603 		release_sock(sk);
604 		timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
605 		lock_sock(sk);
606 		if (signal_pending(current) || !timeo)
607 			break;
608 	}
609 	remove_wait_queue(sk_sleep(sk), &wait);
610 	sk->sk_write_pending -= writebias;
611 	sk->sk_wait_pending--;
612 	return timeo;
613 }
614 
615 /*
616  *	Connect to a remote host. There is regrettably still a little
617  *	TCP 'magic' in here.
618  */
__inet_stream_connect(struct socket * sock,struct sockaddr * uaddr,int addr_len,int flags,int is_sendmsg)619 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
620 			  int addr_len, int flags, int is_sendmsg)
621 {
622 	struct sock *sk = sock->sk;
623 	int err;
624 	long timeo;
625 
626 	/*
627 	 * uaddr can be NULL and addr_len can be 0 if:
628 	 * sk is a TCP fastopen active socket and
629 	 * TCP_FASTOPEN_CONNECT sockopt is set and
630 	 * we already have a valid cookie for this socket.
631 	 * In this case, user can call write() after connect().
632 	 * write() will invoke tcp_sendmsg_fastopen() which calls
633 	 * __inet_stream_connect().
634 	 */
635 	if (uaddr) {
636 		if (addr_len < sizeof(uaddr->sa_family))
637 			return -EINVAL;
638 
639 		if (uaddr->sa_family == AF_UNSPEC) {
640 			err = sk->sk_prot->disconnect(sk, flags);
641 			sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
642 			goto out;
643 		}
644 	}
645 
646 	switch (sock->state) {
647 	default:
648 		err = -EINVAL;
649 		goto out;
650 	case SS_CONNECTED:
651 		err = -EISCONN;
652 		goto out;
653 	case SS_CONNECTING:
654 		if (inet_sk(sk)->defer_connect)
655 			err = is_sendmsg ? -EINPROGRESS : -EISCONN;
656 		else
657 			err = -EALREADY;
658 		/* Fall out of switch with err, set for this state */
659 		break;
660 	case SS_UNCONNECTED:
661 		err = -EISCONN;
662 		if (sk->sk_state != TCP_CLOSE)
663 			goto out;
664 
665 		if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
666 			err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
667 			if (err)
668 				goto out;
669 		}
670 
671 		err = sk->sk_prot->connect(sk, uaddr, addr_len);
672 		if (err < 0)
673 			goto out;
674 
675 		sock->state = SS_CONNECTING;
676 
677 		if (!err && inet_sk(sk)->defer_connect)
678 			goto out;
679 
680 		/* Just entered SS_CONNECTING state; the only
681 		 * difference is that return value in non-blocking
682 		 * case is EINPROGRESS, rather than EALREADY.
683 		 */
684 		err = -EINPROGRESS;
685 		break;
686 	}
687 
688 	timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
689 
690 	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
691 		int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
692 				tcp_sk(sk)->fastopen_req &&
693 				tcp_sk(sk)->fastopen_req->data ? 1 : 0;
694 
695 		/* Error code is set above */
696 		if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
697 			goto out;
698 
699 		err = sock_intr_errno(timeo);
700 		if (signal_pending(current))
701 			goto out;
702 	}
703 
704 	/* Connection was closed by RST, timeout, ICMP error
705 	 * or another process disconnected us.
706 	 */
707 	if (sk->sk_state == TCP_CLOSE)
708 		goto sock_error;
709 
710 	/* sk->sk_err may be not zero now, if RECVERR was ordered by user
711 	 * and error was received after socket entered established state.
712 	 * Hence, it is handled normally after connect() return successfully.
713 	 */
714 
715 	sock->state = SS_CONNECTED;
716 	err = 0;
717 out:
718 	return err;
719 
720 sock_error:
721 	err = sock_error(sk) ? : -ECONNABORTED;
722 	sock->state = SS_UNCONNECTED;
723 	if (sk->sk_prot->disconnect(sk, flags))
724 		sock->state = SS_DISCONNECTING;
725 	goto out;
726 }
727 EXPORT_SYMBOL(__inet_stream_connect);
728 
inet_stream_connect(struct socket * sock,struct sockaddr * uaddr,int addr_len,int flags)729 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
730 			int addr_len, int flags)
731 {
732 	int err;
733 
734 	lock_sock(sock->sk);
735 	err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
736 	release_sock(sock->sk);
737 	return err;
738 }
739 EXPORT_SYMBOL(inet_stream_connect);
740 
741 /*
742  *	Accept a pending connection. The TCP layer now gives BSD semantics.
743  */
744 
inet_accept(struct socket * sock,struct socket * newsock,int flags,bool kern)745 int inet_accept(struct socket *sock, struct socket *newsock, int flags,
746 		bool kern)
747 {
748 	struct sock *sk1 = sock->sk, *sk2;
749 	int err = -EINVAL;
750 
751 	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
752 	sk2 = READ_ONCE(sk1->sk_prot)->accept(sk1, flags, &err, kern);
753 	if (!sk2)
754 		goto do_err;
755 
756 	lock_sock(sk2);
757 
758 	sock_rps_record_flow(sk2);
759 	WARN_ON(!((1 << sk2->sk_state) &
760 		  (TCPF_ESTABLISHED | TCPF_SYN_RECV |
761 		  TCPF_CLOSE_WAIT | TCPF_CLOSE)));
762 
763 	if (test_bit(SOCK_SUPPORT_ZC, &sock->flags))
764 		set_bit(SOCK_SUPPORT_ZC, &newsock->flags);
765 	sock_graft(sk2, newsock);
766 
767 	newsock->state = SS_CONNECTED;
768 	err = 0;
769 	release_sock(sk2);
770 do_err:
771 	return err;
772 }
773 EXPORT_SYMBOL(inet_accept);
774 
775 /*
776  *	This does both peername and sockname.
777  */
inet_getname(struct socket * sock,struct sockaddr * uaddr,int peer)778 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
779 		 int peer)
780 {
781 	struct sock *sk		= sock->sk;
782 	struct inet_sock *inet	= inet_sk(sk);
783 	DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
784 
785 	sin->sin_family = AF_INET;
786 	lock_sock(sk);
787 	if (peer) {
788 		if (!inet->inet_dport ||
789 		    (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
790 		     peer == 1)) {
791 			release_sock(sk);
792 			return -ENOTCONN;
793 		}
794 		sin->sin_port = inet->inet_dport;
795 		sin->sin_addr.s_addr = inet->inet_daddr;
796 		BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin,
797 				       CGROUP_INET4_GETPEERNAME);
798 	} else {
799 		__be32 addr = inet->inet_rcv_saddr;
800 		if (!addr)
801 			addr = inet->inet_saddr;
802 		sin->sin_port = inet->inet_sport;
803 		sin->sin_addr.s_addr = addr;
804 		BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin,
805 				       CGROUP_INET4_GETSOCKNAME);
806 	}
807 	release_sock(sk);
808 	memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
809 	return sizeof(*sin);
810 }
811 EXPORT_SYMBOL(inet_getname);
812 
inet_send_prepare(struct sock * sk)813 int inet_send_prepare(struct sock *sk)
814 {
815 	sock_rps_record_flow(sk);
816 
817 	/* We may need to bind the socket. */
818 	if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind &&
819 	    inet_autobind(sk))
820 		return -EAGAIN;
821 
822 	return 0;
823 }
824 EXPORT_SYMBOL_GPL(inet_send_prepare);
825 
inet_sendmsg(struct socket * sock,struct msghdr * msg,size_t size)826 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
827 {
828 	struct sock *sk = sock->sk;
829 
830 	if (unlikely(inet_send_prepare(sk)))
831 		return -EAGAIN;
832 
833 	return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
834 			       sk, msg, size);
835 }
836 EXPORT_SYMBOL(inet_sendmsg);
837 
inet_sendpage(struct socket * sock,struct page * page,int offset,size_t size,int flags)838 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
839 		      size_t size, int flags)
840 {
841 	struct sock *sk = sock->sk;
842 	const struct proto *prot;
843 
844 	if (unlikely(inet_send_prepare(sk)))
845 		return -EAGAIN;
846 
847 	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
848 	prot = READ_ONCE(sk->sk_prot);
849 	if (prot->sendpage)
850 		return prot->sendpage(sk, page, offset, size, flags);
851 	return sock_no_sendpage(sock, page, offset, size, flags);
852 }
853 EXPORT_SYMBOL(inet_sendpage);
854 
855 INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
856 					  size_t, int, int *));
inet_recvmsg(struct socket * sock,struct msghdr * msg,size_t size,int flags)857 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
858 		 int flags)
859 {
860 	struct sock *sk = sock->sk;
861 	int addr_len = 0;
862 	int err;
863 
864 	if (likely(!(flags & MSG_ERRQUEUE)))
865 		sock_rps_record_flow(sk);
866 
867 	err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
868 			      sk, msg, size, flags, &addr_len);
869 	if (err >= 0)
870 		msg->msg_namelen = addr_len;
871 	return err;
872 }
873 EXPORT_SYMBOL(inet_recvmsg);
874 
inet_shutdown(struct socket * sock,int how)875 int inet_shutdown(struct socket *sock, int how)
876 {
877 	struct sock *sk = sock->sk;
878 	int err = 0;
879 
880 	/* This should really check to make sure
881 	 * the socket is a TCP socket. (WHY AC...)
882 	 */
883 	how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
884 		       1->2 bit 2 snds.
885 		       2->3 */
886 	if ((how & ~SHUTDOWN_MASK) || !how)	/* MAXINT->0 */
887 		return -EINVAL;
888 
889 	lock_sock(sk);
890 	if (sock->state == SS_CONNECTING) {
891 		if ((1 << sk->sk_state) &
892 		    (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
893 			sock->state = SS_DISCONNECTING;
894 		else
895 			sock->state = SS_CONNECTED;
896 	}
897 
898 	switch (sk->sk_state) {
899 	case TCP_CLOSE:
900 		err = -ENOTCONN;
901 		/* Hack to wake up other listeners, who can poll for
902 		   EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
903 		fallthrough;
904 	default:
905 		WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | how);
906 		if (sk->sk_prot->shutdown)
907 			sk->sk_prot->shutdown(sk, how);
908 		break;
909 
910 	/* Remaining two branches are temporary solution for missing
911 	 * close() in multithreaded environment. It is _not_ a good idea,
912 	 * but we have no choice until close() is repaired at VFS level.
913 	 */
914 	case TCP_LISTEN:
915 		if (!(how & RCV_SHUTDOWN))
916 			break;
917 		fallthrough;
918 	case TCP_SYN_SENT:
919 		err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
920 		sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
921 		break;
922 	}
923 
924 	/* Wake up anyone sleeping in poll. */
925 	sk->sk_state_change(sk);
926 	release_sock(sk);
927 	return err;
928 }
929 EXPORT_SYMBOL(inet_shutdown);
930 
931 /*
932  *	ioctl() calls you can issue on an INET socket. Most of these are
933  *	device configuration and stuff and very rarely used. Some ioctls
934  *	pass on to the socket itself.
935  *
936  *	NOTE: I like the idea of a module for the config stuff. ie ifconfig
937  *	loads the devconfigure module does its configuring and unloads it.
938  *	There's a good 20K of config code hanging around the kernel.
939  */
940 
inet_ioctl(struct socket * sock,unsigned int cmd,unsigned long arg)941 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
942 {
943 	struct sock *sk = sock->sk;
944 	int err = 0;
945 	struct net *net = sock_net(sk);
946 	void __user *p = (void __user *)arg;
947 	struct ifreq ifr;
948 	struct rtentry rt;
949 
950 	switch (cmd) {
951 	case SIOCADDRT:
952 	case SIOCDELRT:
953 		if (copy_from_user(&rt, p, sizeof(struct rtentry)))
954 			return -EFAULT;
955 		err = ip_rt_ioctl(net, cmd, &rt);
956 		break;
957 	case SIOCRTMSG:
958 		err = -EINVAL;
959 		break;
960 	case SIOCDARP:
961 	case SIOCGARP:
962 	case SIOCSARP:
963 		err = arp_ioctl(net, cmd, (void __user *)arg);
964 		break;
965 	case SIOCGIFADDR:
966 	case SIOCGIFBRDADDR:
967 	case SIOCGIFNETMASK:
968 	case SIOCGIFDSTADDR:
969 	case SIOCGIFPFLAGS:
970 		if (get_user_ifreq(&ifr, NULL, p))
971 			return -EFAULT;
972 		err = devinet_ioctl(net, cmd, &ifr);
973 		if (!err && put_user_ifreq(&ifr, p))
974 			err = -EFAULT;
975 		break;
976 
977 	case SIOCSIFADDR:
978 	case SIOCSIFBRDADDR:
979 	case SIOCSIFNETMASK:
980 	case SIOCSIFDSTADDR:
981 	case SIOCSIFPFLAGS:
982 	case SIOCSIFFLAGS:
983 		if (get_user_ifreq(&ifr, NULL, p))
984 			return -EFAULT;
985 		err = devinet_ioctl(net, cmd, &ifr);
986 		break;
987 	default:
988 		if (sk->sk_prot->ioctl)
989 			err = sk->sk_prot->ioctl(sk, cmd, arg);
990 		else
991 			err = -ENOIOCTLCMD;
992 		break;
993 	}
994 	return err;
995 }
996 EXPORT_SYMBOL(inet_ioctl);
997 
998 #ifdef CONFIG_COMPAT
inet_compat_routing_ioctl(struct sock * sk,unsigned int cmd,struct compat_rtentry __user * ur)999 static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd,
1000 		struct compat_rtentry __user *ur)
1001 {
1002 	compat_uptr_t rtdev;
1003 	struct rtentry rt;
1004 
1005 	if (copy_from_user(&rt.rt_dst, &ur->rt_dst,
1006 			3 * sizeof(struct sockaddr)) ||
1007 	    get_user(rt.rt_flags, &ur->rt_flags) ||
1008 	    get_user(rt.rt_metric, &ur->rt_metric) ||
1009 	    get_user(rt.rt_mtu, &ur->rt_mtu) ||
1010 	    get_user(rt.rt_window, &ur->rt_window) ||
1011 	    get_user(rt.rt_irtt, &ur->rt_irtt) ||
1012 	    get_user(rtdev, &ur->rt_dev))
1013 		return -EFAULT;
1014 
1015 	rt.rt_dev = compat_ptr(rtdev);
1016 	return ip_rt_ioctl(sock_net(sk), cmd, &rt);
1017 }
1018 
inet_compat_ioctl(struct socket * sock,unsigned int cmd,unsigned long arg)1019 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1020 {
1021 	void __user *argp = compat_ptr(arg);
1022 	struct sock *sk = sock->sk;
1023 
1024 	switch (cmd) {
1025 	case SIOCADDRT:
1026 	case SIOCDELRT:
1027 		return inet_compat_routing_ioctl(sk, cmd, argp);
1028 	default:
1029 		if (!sk->sk_prot->compat_ioctl)
1030 			return -ENOIOCTLCMD;
1031 		return sk->sk_prot->compat_ioctl(sk, cmd, arg);
1032 	}
1033 }
1034 #endif /* CONFIG_COMPAT */
1035 
1036 const struct proto_ops inet_stream_ops = {
1037 	.family		   = PF_INET,
1038 	.owner		   = THIS_MODULE,
1039 	.release	   = inet_release,
1040 	.bind		   = inet_bind,
1041 	.connect	   = inet_stream_connect,
1042 	.socketpair	   = sock_no_socketpair,
1043 	.accept		   = inet_accept,
1044 	.getname	   = inet_getname,
1045 	.poll		   = tcp_poll,
1046 	.ioctl		   = inet_ioctl,
1047 	.gettstamp	   = sock_gettstamp,
1048 	.listen		   = inet_listen,
1049 	.shutdown	   = inet_shutdown,
1050 	.setsockopt	   = sock_common_setsockopt,
1051 	.getsockopt	   = sock_common_getsockopt,
1052 	.sendmsg	   = inet_sendmsg,
1053 	.recvmsg	   = inet_recvmsg,
1054 #ifdef CONFIG_MMU
1055 	.mmap		   = tcp_mmap,
1056 #endif
1057 	.sendpage	   = inet_sendpage,
1058 	.splice_read	   = tcp_splice_read,
1059 	.read_sock	   = tcp_read_sock,
1060 	.read_skb	   = tcp_read_skb,
1061 	.sendmsg_locked    = tcp_sendmsg_locked,
1062 	.sendpage_locked   = tcp_sendpage_locked,
1063 	.peek_len	   = tcp_peek_len,
1064 #ifdef CONFIG_COMPAT
1065 	.compat_ioctl	   = inet_compat_ioctl,
1066 #endif
1067 	.set_rcvlowat	   = tcp_set_rcvlowat,
1068 };
1069 EXPORT_SYMBOL(inet_stream_ops);
1070 
1071 const struct proto_ops inet_dgram_ops = {
1072 	.family		   = PF_INET,
1073 	.owner		   = THIS_MODULE,
1074 	.release	   = inet_release,
1075 	.bind		   = inet_bind,
1076 	.connect	   = inet_dgram_connect,
1077 	.socketpair	   = sock_no_socketpair,
1078 	.accept		   = sock_no_accept,
1079 	.getname	   = inet_getname,
1080 	.poll		   = udp_poll,
1081 	.ioctl		   = inet_ioctl,
1082 	.gettstamp	   = sock_gettstamp,
1083 	.listen		   = sock_no_listen,
1084 	.shutdown	   = inet_shutdown,
1085 	.setsockopt	   = sock_common_setsockopt,
1086 	.getsockopt	   = sock_common_getsockopt,
1087 	.sendmsg	   = inet_sendmsg,
1088 	.read_skb	   = udp_read_skb,
1089 	.recvmsg	   = inet_recvmsg,
1090 	.mmap		   = sock_no_mmap,
1091 	.sendpage	   = inet_sendpage,
1092 	.set_peek_off	   = sk_set_peek_off,
1093 #ifdef CONFIG_COMPAT
1094 	.compat_ioctl	   = inet_compat_ioctl,
1095 #endif
1096 };
1097 EXPORT_SYMBOL(inet_dgram_ops);
1098 
1099 /*
1100  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1101  * udp_poll
1102  */
1103 static const struct proto_ops inet_sockraw_ops = {
1104 	.family		   = PF_INET,
1105 	.owner		   = THIS_MODULE,
1106 	.release	   = inet_release,
1107 	.bind		   = inet_bind,
1108 	.connect	   = inet_dgram_connect,
1109 	.socketpair	   = sock_no_socketpair,
1110 	.accept		   = sock_no_accept,
1111 	.getname	   = inet_getname,
1112 	.poll		   = datagram_poll,
1113 	.ioctl		   = inet_ioctl,
1114 	.gettstamp	   = sock_gettstamp,
1115 	.listen		   = sock_no_listen,
1116 	.shutdown	   = inet_shutdown,
1117 	.setsockopt	   = sock_common_setsockopt,
1118 	.getsockopt	   = sock_common_getsockopt,
1119 	.sendmsg	   = inet_sendmsg,
1120 	.recvmsg	   = inet_recvmsg,
1121 	.mmap		   = sock_no_mmap,
1122 	.sendpage	   = inet_sendpage,
1123 #ifdef CONFIG_COMPAT
1124 	.compat_ioctl	   = inet_compat_ioctl,
1125 #endif
1126 };
1127 
1128 static const struct net_proto_family inet_family_ops = {
1129 	.family = PF_INET,
1130 	.create = inet_create,
1131 	.owner	= THIS_MODULE,
1132 };
1133 
1134 /* Upon startup we insert all the elements in inetsw_array[] into
1135  * the linked list inetsw.
1136  */
1137 static struct inet_protosw inetsw_array[] =
1138 {
1139 	{
1140 		.type =       SOCK_STREAM,
1141 		.protocol =   IPPROTO_TCP,
1142 		.prot =       &tcp_prot,
1143 		.ops =        &inet_stream_ops,
1144 		.flags =      INET_PROTOSW_PERMANENT |
1145 			      INET_PROTOSW_ICSK,
1146 	},
1147 
1148 	{
1149 		.type =       SOCK_DGRAM,
1150 		.protocol =   IPPROTO_UDP,
1151 		.prot =       &udp_prot,
1152 		.ops =        &inet_dgram_ops,
1153 		.flags =      INET_PROTOSW_PERMANENT,
1154        },
1155 
1156        {
1157 		.type =       SOCK_DGRAM,
1158 		.protocol =   IPPROTO_ICMP,
1159 		.prot =       &ping_prot,
1160 		.ops =        &inet_sockraw_ops,
1161 		.flags =      INET_PROTOSW_REUSE,
1162        },
1163 
1164        {
1165 	       .type =       SOCK_RAW,
1166 	       .protocol =   IPPROTO_IP,	/* wild card */
1167 	       .prot =       &raw_prot,
1168 	       .ops =        &inet_sockraw_ops,
1169 	       .flags =      INET_PROTOSW_REUSE,
1170        }
1171 };
1172 
1173 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1174 
inet_register_protosw(struct inet_protosw * p)1175 void inet_register_protosw(struct inet_protosw *p)
1176 {
1177 	struct list_head *lh;
1178 	struct inet_protosw *answer;
1179 	int protocol = p->protocol;
1180 	struct list_head *last_perm;
1181 
1182 	spin_lock_bh(&inetsw_lock);
1183 
1184 	if (p->type >= SOCK_MAX)
1185 		goto out_illegal;
1186 
1187 	/* If we are trying to override a permanent protocol, bail. */
1188 	last_perm = &inetsw[p->type];
1189 	list_for_each(lh, &inetsw[p->type]) {
1190 		answer = list_entry(lh, struct inet_protosw, list);
1191 		/* Check only the non-wild match. */
1192 		if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1193 			break;
1194 		if (protocol == answer->protocol)
1195 			goto out_permanent;
1196 		last_perm = lh;
1197 	}
1198 
1199 	/* Add the new entry after the last permanent entry if any, so that
1200 	 * the new entry does not override a permanent entry when matched with
1201 	 * a wild-card protocol. But it is allowed to override any existing
1202 	 * non-permanent entry.  This means that when we remove this entry, the
1203 	 * system automatically returns to the old behavior.
1204 	 */
1205 	list_add_rcu(&p->list, last_perm);
1206 out:
1207 	spin_unlock_bh(&inetsw_lock);
1208 
1209 	return;
1210 
1211 out_permanent:
1212 	pr_err("Attempt to override permanent protocol %d\n", protocol);
1213 	goto out;
1214 
1215 out_illegal:
1216 	pr_err("Ignoring attempt to register invalid socket type %d\n",
1217 	       p->type);
1218 	goto out;
1219 }
1220 EXPORT_SYMBOL(inet_register_protosw);
1221 
inet_unregister_protosw(struct inet_protosw * p)1222 void inet_unregister_protosw(struct inet_protosw *p)
1223 {
1224 	if (INET_PROTOSW_PERMANENT & p->flags) {
1225 		pr_err("Attempt to unregister permanent protocol %d\n",
1226 		       p->protocol);
1227 	} else {
1228 		spin_lock_bh(&inetsw_lock);
1229 		list_del_rcu(&p->list);
1230 		spin_unlock_bh(&inetsw_lock);
1231 
1232 		synchronize_net();
1233 	}
1234 }
1235 EXPORT_SYMBOL(inet_unregister_protosw);
1236 
inet_sk_reselect_saddr(struct sock * sk)1237 static int inet_sk_reselect_saddr(struct sock *sk)
1238 {
1239 	struct inet_sock *inet = inet_sk(sk);
1240 	__be32 old_saddr = inet->inet_saddr;
1241 	__be32 daddr = inet->inet_daddr;
1242 	struct flowi4 *fl4;
1243 	struct rtable *rt;
1244 	__be32 new_saddr;
1245 	struct ip_options_rcu *inet_opt;
1246 	int err;
1247 
1248 	inet_opt = rcu_dereference_protected(inet->inet_opt,
1249 					     lockdep_sock_is_held(sk));
1250 	if (inet_opt && inet_opt->opt.srr)
1251 		daddr = inet_opt->opt.faddr;
1252 
1253 	/* Query new route. */
1254 	fl4 = &inet->cork.fl.u.ip4;
1255 	rt = ip_route_connect(fl4, daddr, 0, sk->sk_bound_dev_if,
1256 			      sk->sk_protocol, inet->inet_sport,
1257 			      inet->inet_dport, sk);
1258 	if (IS_ERR(rt))
1259 		return PTR_ERR(rt);
1260 
1261 	new_saddr = fl4->saddr;
1262 
1263 	if (new_saddr == old_saddr) {
1264 		sk_setup_caps(sk, &rt->dst);
1265 		return 0;
1266 	}
1267 
1268 	err = inet_bhash2_update_saddr(sk, &new_saddr, AF_INET);
1269 	if (err) {
1270 		ip_rt_put(rt);
1271 		return err;
1272 	}
1273 
1274 	sk_setup_caps(sk, &rt->dst);
1275 
1276 	if (READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) > 1) {
1277 		pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1278 			__func__, &old_saddr, &new_saddr);
1279 	}
1280 
1281 	/*
1282 	 * XXX The only one ugly spot where we need to
1283 	 * XXX really change the sockets identity after
1284 	 * XXX it has entered the hashes. -DaveM
1285 	 *
1286 	 * Besides that, it does not check for connection
1287 	 * uniqueness. Wait for troubles.
1288 	 */
1289 	return __sk_prot_rehash(sk);
1290 }
1291 
inet_sk_rebuild_header(struct sock * sk)1292 int inet_sk_rebuild_header(struct sock *sk)
1293 {
1294 	struct inet_sock *inet = inet_sk(sk);
1295 	struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1296 	__be32 daddr;
1297 	struct ip_options_rcu *inet_opt;
1298 	struct flowi4 *fl4;
1299 	int err;
1300 
1301 	/* Route is OK, nothing to do. */
1302 	if (rt)
1303 		return 0;
1304 
1305 	/* Reroute. */
1306 	rcu_read_lock();
1307 	inet_opt = rcu_dereference(inet->inet_opt);
1308 	daddr = inet->inet_daddr;
1309 	if (inet_opt && inet_opt->opt.srr)
1310 		daddr = inet_opt->opt.faddr;
1311 	rcu_read_unlock();
1312 	fl4 = &inet->cork.fl.u.ip4;
1313 	rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1314 				   inet->inet_dport, inet->inet_sport,
1315 				   sk->sk_protocol, RT_CONN_FLAGS(sk),
1316 				   sk->sk_bound_dev_if);
1317 	if (!IS_ERR(rt)) {
1318 		err = 0;
1319 		sk_setup_caps(sk, &rt->dst);
1320 	} else {
1321 		err = PTR_ERR(rt);
1322 
1323 		/* Routing failed... */
1324 		sk->sk_route_caps = 0;
1325 		/*
1326 		 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1327 		 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1328 		 */
1329 		if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) ||
1330 		    sk->sk_state != TCP_SYN_SENT ||
1331 		    (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1332 		    (err = inet_sk_reselect_saddr(sk)) != 0)
1333 			sk->sk_err_soft = -err;
1334 	}
1335 
1336 	return err;
1337 }
1338 EXPORT_SYMBOL(inet_sk_rebuild_header);
1339 
inet_sk_set_state(struct sock * sk,int state)1340 void inet_sk_set_state(struct sock *sk, int state)
1341 {
1342 	trace_inet_sock_set_state(sk, sk->sk_state, state);
1343 	sk->sk_state = state;
1344 }
1345 EXPORT_SYMBOL(inet_sk_set_state);
1346 
inet_sk_state_store(struct sock * sk,int newstate)1347 void inet_sk_state_store(struct sock *sk, int newstate)
1348 {
1349 	trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1350 	smp_store_release(&sk->sk_state, newstate);
1351 }
1352 
inet_gso_segment(struct sk_buff * skb,netdev_features_t features)1353 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1354 				 netdev_features_t features)
1355 {
1356 	bool udpfrag = false, fixedid = false, gso_partial, encap;
1357 	struct sk_buff *segs = ERR_PTR(-EINVAL);
1358 	const struct net_offload *ops;
1359 	unsigned int offset = 0;
1360 	struct iphdr *iph;
1361 	int proto, tot_len;
1362 	int nhoff;
1363 	int ihl;
1364 	int id;
1365 
1366 	skb_reset_network_header(skb);
1367 	nhoff = skb_network_header(skb) - skb_mac_header(skb);
1368 	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1369 		goto out;
1370 
1371 	iph = ip_hdr(skb);
1372 	ihl = iph->ihl * 4;
1373 	if (ihl < sizeof(*iph))
1374 		goto out;
1375 
1376 	id = ntohs(iph->id);
1377 	proto = iph->protocol;
1378 
1379 	/* Warning: after this point, iph might be no longer valid */
1380 	if (unlikely(!pskb_may_pull(skb, ihl)))
1381 		goto out;
1382 	__skb_pull(skb, ihl);
1383 
1384 	encap = SKB_GSO_CB(skb)->encap_level > 0;
1385 	if (encap)
1386 		features &= skb->dev->hw_enc_features;
1387 	SKB_GSO_CB(skb)->encap_level += ihl;
1388 
1389 	skb_reset_transport_header(skb);
1390 
1391 	segs = ERR_PTR(-EPROTONOSUPPORT);
1392 
1393 	if (!skb->encapsulation || encap) {
1394 		udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1395 		fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1396 
1397 		/* fixed ID is invalid if DF bit is not set */
1398 		if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1399 			goto out;
1400 	}
1401 
1402 	ops = rcu_dereference(inet_offloads[proto]);
1403 	if (likely(ops && ops->callbacks.gso_segment)) {
1404 		segs = ops->callbacks.gso_segment(skb, features);
1405 		if (!segs)
1406 			skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
1407 	}
1408 
1409 	if (IS_ERR_OR_NULL(segs))
1410 		goto out;
1411 
1412 	gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1413 
1414 	skb = segs;
1415 	do {
1416 		iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1417 		if (udpfrag) {
1418 			iph->frag_off = htons(offset >> 3);
1419 			if (skb->next)
1420 				iph->frag_off |= htons(IP_MF);
1421 			offset += skb->len - nhoff - ihl;
1422 			tot_len = skb->len - nhoff;
1423 		} else if (skb_is_gso(skb)) {
1424 			if (!fixedid) {
1425 				iph->id = htons(id);
1426 				id += skb_shinfo(skb)->gso_segs;
1427 			}
1428 
1429 			if (gso_partial)
1430 				tot_len = skb_shinfo(skb)->gso_size +
1431 					  SKB_GSO_CB(skb)->data_offset +
1432 					  skb->head - (unsigned char *)iph;
1433 			else
1434 				tot_len = skb->len - nhoff;
1435 		} else {
1436 			if (!fixedid)
1437 				iph->id = htons(id++);
1438 			tot_len = skb->len - nhoff;
1439 		}
1440 		iph->tot_len = htons(tot_len);
1441 		ip_send_check(iph);
1442 		if (encap)
1443 			skb_reset_inner_headers(skb);
1444 		skb->network_header = (u8 *)iph - skb->head;
1445 		skb_reset_mac_len(skb);
1446 	} while ((skb = skb->next));
1447 
1448 out:
1449 	return segs;
1450 }
1451 
ipip_gso_segment(struct sk_buff * skb,netdev_features_t features)1452 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1453 					netdev_features_t features)
1454 {
1455 	if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1456 		return ERR_PTR(-EINVAL);
1457 
1458 	return inet_gso_segment(skb, features);
1459 }
1460 
inet_gro_receive(struct list_head * head,struct sk_buff * skb)1461 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1462 {
1463 	const struct net_offload *ops;
1464 	struct sk_buff *pp = NULL;
1465 	const struct iphdr *iph;
1466 	struct sk_buff *p;
1467 	unsigned int hlen;
1468 	unsigned int off;
1469 	unsigned int id;
1470 	int flush = 1;
1471 	int proto;
1472 
1473 	off = skb_gro_offset(skb);
1474 	hlen = off + sizeof(*iph);
1475 	iph = skb_gro_header(skb, hlen, off);
1476 	if (unlikely(!iph))
1477 		goto out;
1478 
1479 	proto = iph->protocol;
1480 
1481 	ops = rcu_dereference(inet_offloads[proto]);
1482 	if (!ops || !ops->callbacks.gro_receive)
1483 		goto out;
1484 
1485 	if (*(u8 *)iph != 0x45)
1486 		goto out;
1487 
1488 	if (ip_is_fragment(iph))
1489 		goto out;
1490 
1491 	if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1492 		goto out;
1493 
1494 	id = ntohl(*(__be32 *)&iph->id);
1495 	flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1496 	id >>= 16;
1497 
1498 	list_for_each_entry(p, head, list) {
1499 		struct iphdr *iph2;
1500 		u16 flush_id;
1501 
1502 		if (!NAPI_GRO_CB(p)->same_flow)
1503 			continue;
1504 
1505 		iph2 = (struct iphdr *)(p->data + off);
1506 		/* The above works because, with the exception of the top
1507 		 * (inner most) layer, we only aggregate pkts with the same
1508 		 * hdr length so all the hdrs we'll need to verify will start
1509 		 * at the same offset.
1510 		 */
1511 		if ((iph->protocol ^ iph2->protocol) |
1512 		    ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1513 		    ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1514 			NAPI_GRO_CB(p)->same_flow = 0;
1515 			continue;
1516 		}
1517 
1518 		/* All fields must match except length and checksum. */
1519 		NAPI_GRO_CB(p)->flush |=
1520 			(iph->ttl ^ iph2->ttl) |
1521 			(iph->tos ^ iph2->tos) |
1522 			((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1523 
1524 		NAPI_GRO_CB(p)->flush |= flush;
1525 
1526 		/* We need to store of the IP ID check to be included later
1527 		 * when we can verify that this packet does in fact belong
1528 		 * to a given flow.
1529 		 */
1530 		flush_id = (u16)(id - ntohs(iph2->id));
1531 
1532 		/* This bit of code makes it much easier for us to identify
1533 		 * the cases where we are doing atomic vs non-atomic IP ID
1534 		 * checks.  Specifically an atomic check can return IP ID
1535 		 * values 0 - 0xFFFF, while a non-atomic check can only
1536 		 * return 0 or 0xFFFF.
1537 		 */
1538 		if (!NAPI_GRO_CB(p)->is_atomic ||
1539 		    !(iph->frag_off & htons(IP_DF))) {
1540 			flush_id ^= NAPI_GRO_CB(p)->count;
1541 			flush_id = flush_id ? 0xFFFF : 0;
1542 		}
1543 
1544 		/* If the previous IP ID value was based on an atomic
1545 		 * datagram we can overwrite the value and ignore it.
1546 		 */
1547 		if (NAPI_GRO_CB(skb)->is_atomic)
1548 			NAPI_GRO_CB(p)->flush_id = flush_id;
1549 		else
1550 			NAPI_GRO_CB(p)->flush_id |= flush_id;
1551 	}
1552 
1553 	NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1554 	NAPI_GRO_CB(skb)->flush |= flush;
1555 	skb_set_network_header(skb, off);
1556 	/* The above will be needed by the transport layer if there is one
1557 	 * immediately following this IP hdr.
1558 	 */
1559 
1560 	/* Note : No need to call skb_gro_postpull_rcsum() here,
1561 	 * as we already checked checksum over ipv4 header was 0
1562 	 */
1563 	skb_gro_pull(skb, sizeof(*iph));
1564 	skb_set_transport_header(skb, skb_gro_offset(skb));
1565 
1566 	pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1567 				       ops->callbacks.gro_receive, head, skb);
1568 
1569 out:
1570 	skb_gro_flush_final(skb, pp, flush);
1571 
1572 	return pp;
1573 }
1574 
ipip_gro_receive(struct list_head * head,struct sk_buff * skb)1575 static struct sk_buff *ipip_gro_receive(struct list_head *head,
1576 					struct sk_buff *skb)
1577 {
1578 	if (NAPI_GRO_CB(skb)->encap_mark) {
1579 		NAPI_GRO_CB(skb)->flush = 1;
1580 		return NULL;
1581 	}
1582 
1583 	NAPI_GRO_CB(skb)->encap_mark = 1;
1584 
1585 	return inet_gro_receive(head, skb);
1586 }
1587 
1588 #define SECONDS_PER_DAY	86400
1589 
1590 /* inet_current_timestamp - Return IP network timestamp
1591  *
1592  * Return milliseconds since midnight in network byte order.
1593  */
inet_current_timestamp(void)1594 __be32 inet_current_timestamp(void)
1595 {
1596 	u32 secs;
1597 	u32 msecs;
1598 	struct timespec64 ts;
1599 
1600 	ktime_get_real_ts64(&ts);
1601 
1602 	/* Get secs since midnight. */
1603 	(void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1604 	/* Convert to msecs. */
1605 	msecs = secs * MSEC_PER_SEC;
1606 	/* Convert nsec to msec. */
1607 	msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1608 
1609 	/* Convert to network byte order. */
1610 	return htonl(msecs);
1611 }
1612 EXPORT_SYMBOL(inet_current_timestamp);
1613 
inet_recv_error(struct sock * sk,struct msghdr * msg,int len,int * addr_len)1614 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1615 {
1616 	unsigned int family = READ_ONCE(sk->sk_family);
1617 
1618 	if (family == AF_INET)
1619 		return ip_recv_error(sk, msg, len, addr_len);
1620 #if IS_ENABLED(CONFIG_IPV6)
1621 	if (family == AF_INET6)
1622 		return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1623 #endif
1624 	return -EINVAL;
1625 }
1626 EXPORT_SYMBOL(inet_recv_error);
1627 
inet_gro_complete(struct sk_buff * skb,int nhoff)1628 int inet_gro_complete(struct sk_buff *skb, int nhoff)
1629 {
1630 	__be16 newlen = htons(skb->len - nhoff);
1631 	struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1632 	const struct net_offload *ops;
1633 	int proto = iph->protocol;
1634 	int err = -ENOSYS;
1635 
1636 	if (skb->encapsulation) {
1637 		skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1638 		skb_set_inner_network_header(skb, nhoff);
1639 	}
1640 
1641 	csum_replace2(&iph->check, iph->tot_len, newlen);
1642 	iph->tot_len = newlen;
1643 
1644 	ops = rcu_dereference(inet_offloads[proto]);
1645 	if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1646 		goto out;
1647 
1648 	/* Only need to add sizeof(*iph) to get to the next hdr below
1649 	 * because any hdr with option will have been flushed in
1650 	 * inet_gro_receive().
1651 	 */
1652 	err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1653 			      tcp4_gro_complete, udp4_gro_complete,
1654 			      skb, nhoff + sizeof(*iph));
1655 
1656 out:
1657 	return err;
1658 }
1659 
ipip_gro_complete(struct sk_buff * skb,int nhoff)1660 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1661 {
1662 	skb->encapsulation = 1;
1663 	skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1664 	return inet_gro_complete(skb, nhoff);
1665 }
1666 
inet_ctl_sock_create(struct sock ** sk,unsigned short family,unsigned short type,unsigned char protocol,struct net * net)1667 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1668 			 unsigned short type, unsigned char protocol,
1669 			 struct net *net)
1670 {
1671 	struct socket *sock;
1672 	int rc = sock_create_kern(net, family, type, protocol, &sock);
1673 
1674 	if (rc == 0) {
1675 		*sk = sock->sk;
1676 		(*sk)->sk_allocation = GFP_ATOMIC;
1677 		/*
1678 		 * Unhash it so that IP input processing does not even see it,
1679 		 * we do not wish this socket to see incoming packets.
1680 		 */
1681 		(*sk)->sk_prot->unhash(*sk);
1682 	}
1683 	return rc;
1684 }
1685 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1686 
snmp_fold_field(void __percpu * mib,int offt)1687 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1688 {
1689 	unsigned long res = 0;
1690 	int i;
1691 
1692 	for_each_possible_cpu(i)
1693 		res += snmp_get_cpu_field(mib, i, offt);
1694 	return res;
1695 }
1696 EXPORT_SYMBOL_GPL(snmp_fold_field);
1697 
1698 #if BITS_PER_LONG==32
1699 
snmp_get_cpu_field64(void __percpu * mib,int cpu,int offt,size_t syncp_offset)1700 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1701 			 size_t syncp_offset)
1702 {
1703 	void *bhptr;
1704 	struct u64_stats_sync *syncp;
1705 	u64 v;
1706 	unsigned int start;
1707 
1708 	bhptr = per_cpu_ptr(mib, cpu);
1709 	syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1710 	do {
1711 		start = u64_stats_fetch_begin_irq(syncp);
1712 		v = *(((u64 *)bhptr) + offt);
1713 	} while (u64_stats_fetch_retry_irq(syncp, start));
1714 
1715 	return v;
1716 }
1717 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1718 
snmp_fold_field64(void __percpu * mib,int offt,size_t syncp_offset)1719 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1720 {
1721 	u64 res = 0;
1722 	int cpu;
1723 
1724 	for_each_possible_cpu(cpu) {
1725 		res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1726 	}
1727 	return res;
1728 }
1729 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1730 #endif
1731 
1732 #ifdef CONFIG_IP_MULTICAST
1733 static const struct net_protocol igmp_protocol = {
1734 	.handler =	igmp_rcv,
1735 };
1736 #endif
1737 
1738 static const struct net_protocol tcp_protocol = {
1739 	.handler	=	tcp_v4_rcv,
1740 	.err_handler	=	tcp_v4_err,
1741 	.no_policy	=	1,
1742 	.icmp_strict_tag_validation = 1,
1743 };
1744 
1745 static const struct net_protocol udp_protocol = {
1746 	.handler =	udp_rcv,
1747 	.err_handler =	udp_err,
1748 	.no_policy =	1,
1749 };
1750 
1751 static const struct net_protocol icmp_protocol = {
1752 	.handler =	icmp_rcv,
1753 	.err_handler =	icmp_err,
1754 	.no_policy =	1,
1755 };
1756 
ipv4_mib_init_net(struct net * net)1757 static __net_init int ipv4_mib_init_net(struct net *net)
1758 {
1759 	int i;
1760 
1761 	net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1762 	if (!net->mib.tcp_statistics)
1763 		goto err_tcp_mib;
1764 	net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1765 	if (!net->mib.ip_statistics)
1766 		goto err_ip_mib;
1767 
1768 	for_each_possible_cpu(i) {
1769 		struct ipstats_mib *af_inet_stats;
1770 		af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1771 		u64_stats_init(&af_inet_stats->syncp);
1772 	}
1773 
1774 	net->mib.net_statistics = alloc_percpu(struct linux_mib);
1775 	if (!net->mib.net_statistics)
1776 		goto err_net_mib;
1777 	net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1778 	if (!net->mib.udp_statistics)
1779 		goto err_udp_mib;
1780 	net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1781 	if (!net->mib.udplite_statistics)
1782 		goto err_udplite_mib;
1783 	net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1784 	if (!net->mib.icmp_statistics)
1785 		goto err_icmp_mib;
1786 	net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1787 					      GFP_KERNEL);
1788 	if (!net->mib.icmpmsg_statistics)
1789 		goto err_icmpmsg_mib;
1790 
1791 	tcp_mib_init(net);
1792 	return 0;
1793 
1794 err_icmpmsg_mib:
1795 	free_percpu(net->mib.icmp_statistics);
1796 err_icmp_mib:
1797 	free_percpu(net->mib.udplite_statistics);
1798 err_udplite_mib:
1799 	free_percpu(net->mib.udp_statistics);
1800 err_udp_mib:
1801 	free_percpu(net->mib.net_statistics);
1802 err_net_mib:
1803 	free_percpu(net->mib.ip_statistics);
1804 err_ip_mib:
1805 	free_percpu(net->mib.tcp_statistics);
1806 err_tcp_mib:
1807 	return -ENOMEM;
1808 }
1809 
ipv4_mib_exit_net(struct net * net)1810 static __net_exit void ipv4_mib_exit_net(struct net *net)
1811 {
1812 	kfree(net->mib.icmpmsg_statistics);
1813 	free_percpu(net->mib.icmp_statistics);
1814 	free_percpu(net->mib.udplite_statistics);
1815 	free_percpu(net->mib.udp_statistics);
1816 	free_percpu(net->mib.net_statistics);
1817 	free_percpu(net->mib.ip_statistics);
1818 	free_percpu(net->mib.tcp_statistics);
1819 #ifdef CONFIG_MPTCP
1820 	/* allocated on demand, see mptcp_init_sock() */
1821 	free_percpu(net->mib.mptcp_statistics);
1822 #endif
1823 }
1824 
1825 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1826 	.init = ipv4_mib_init_net,
1827 	.exit = ipv4_mib_exit_net,
1828 };
1829 
init_ipv4_mibs(void)1830 static int __init init_ipv4_mibs(void)
1831 {
1832 	return register_pernet_subsys(&ipv4_mib_ops);
1833 }
1834 
inet_init_net(struct net * net)1835 static __net_init int inet_init_net(struct net *net)
1836 {
1837 	/*
1838 	 * Set defaults for local port range
1839 	 */
1840 	seqlock_init(&net->ipv4.ip_local_ports.lock);
1841 	net->ipv4.ip_local_ports.range[0] =  32768;
1842 	net->ipv4.ip_local_ports.range[1] =  60999;
1843 
1844 	seqlock_init(&net->ipv4.ping_group_range.lock);
1845 	/*
1846 	 * Sane defaults - nobody may create ping sockets.
1847 	 * Boot scripts should set this to distro-specific group.
1848 	 */
1849 	net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1850 	net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1851 
1852 	/* Default values for sysctl-controlled parameters.
1853 	 * We set them here, in case sysctl is not compiled.
1854 	 */
1855 	net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1856 	net->ipv4.sysctl_ip_fwd_update_priority = 1;
1857 	net->ipv4.sysctl_ip_dynaddr = 0;
1858 	net->ipv4.sysctl_ip_early_demux = 1;
1859 	net->ipv4.sysctl_udp_early_demux = 1;
1860 	net->ipv4.sysctl_tcp_early_demux = 1;
1861 	net->ipv4.sysctl_nexthop_compat_mode = 1;
1862 #ifdef CONFIG_SYSCTL
1863 	net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1864 #endif
1865 
1866 	/* Some igmp sysctl, whose values are always used */
1867 	net->ipv4.sysctl_igmp_max_memberships = 20;
1868 	net->ipv4.sysctl_igmp_max_msf = 10;
1869 	/* IGMP reports for link-local multicast groups are enabled by default */
1870 	net->ipv4.sysctl_igmp_llm_reports = 1;
1871 	net->ipv4.sysctl_igmp_qrv = 2;
1872 
1873 	net->ipv4.sysctl_fib_notify_on_flag_change = 0;
1874 
1875 	return 0;
1876 }
1877 
1878 static __net_initdata struct pernet_operations af_inet_ops = {
1879 	.init = inet_init_net,
1880 };
1881 
init_inet_pernet_ops(void)1882 static int __init init_inet_pernet_ops(void)
1883 {
1884 	return register_pernet_subsys(&af_inet_ops);
1885 }
1886 
1887 static int ipv4_proc_init(void);
1888 
1889 /*
1890  *	IP protocol layer initialiser
1891  */
1892 
1893 static struct packet_offload ip_packet_offload __read_mostly = {
1894 	.type = cpu_to_be16(ETH_P_IP),
1895 	.callbacks = {
1896 		.gso_segment = inet_gso_segment,
1897 		.gro_receive = inet_gro_receive,
1898 		.gro_complete = inet_gro_complete,
1899 	},
1900 };
1901 
1902 static const struct net_offload ipip_offload = {
1903 	.callbacks = {
1904 		.gso_segment	= ipip_gso_segment,
1905 		.gro_receive	= ipip_gro_receive,
1906 		.gro_complete	= ipip_gro_complete,
1907 	},
1908 };
1909 
ipip_offload_init(void)1910 static int __init ipip_offload_init(void)
1911 {
1912 	return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1913 }
1914 
ipv4_offload_init(void)1915 static int __init ipv4_offload_init(void)
1916 {
1917 	/*
1918 	 * Add offloads
1919 	 */
1920 	if (udpv4_offload_init() < 0)
1921 		pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1922 	if (tcpv4_offload_init() < 0)
1923 		pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1924 	if (ipip_offload_init() < 0)
1925 		pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1926 
1927 	dev_add_offload(&ip_packet_offload);
1928 	return 0;
1929 }
1930 
1931 fs_initcall(ipv4_offload_init);
1932 
1933 static struct packet_type ip_packet_type __read_mostly = {
1934 	.type = cpu_to_be16(ETH_P_IP),
1935 	.func = ip_rcv,
1936 	.list_func = ip_list_rcv,
1937 };
1938 
inet_init(void)1939 static int __init inet_init(void)
1940 {
1941 	struct inet_protosw *q;
1942 	struct list_head *r;
1943 	int rc;
1944 
1945 	sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1946 
1947 	raw_hashinfo_init(&raw_v4_hashinfo);
1948 
1949 	rc = proto_register(&tcp_prot, 1);
1950 	if (rc)
1951 		goto out;
1952 
1953 	rc = proto_register(&udp_prot, 1);
1954 	if (rc)
1955 		goto out_unregister_tcp_proto;
1956 
1957 	rc = proto_register(&raw_prot, 1);
1958 	if (rc)
1959 		goto out_unregister_udp_proto;
1960 
1961 	rc = proto_register(&ping_prot, 1);
1962 	if (rc)
1963 		goto out_unregister_raw_proto;
1964 
1965 	/*
1966 	 *	Tell SOCKET that we are alive...
1967 	 */
1968 
1969 	(void)sock_register(&inet_family_ops);
1970 
1971 #ifdef CONFIG_SYSCTL
1972 	ip_static_sysctl_init();
1973 #endif
1974 
1975 	/*
1976 	 *	Add all the base protocols.
1977 	 */
1978 
1979 	if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1980 		pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1981 	if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1982 		pr_crit("%s: Cannot add UDP protocol\n", __func__);
1983 	if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1984 		pr_crit("%s: Cannot add TCP protocol\n", __func__);
1985 #ifdef CONFIG_IP_MULTICAST
1986 	if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1987 		pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1988 #endif
1989 
1990 	/* Register the socket-side information for inet_create. */
1991 	for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1992 		INIT_LIST_HEAD(r);
1993 
1994 	for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1995 		inet_register_protosw(q);
1996 
1997 	/*
1998 	 *	Set the ARP module up
1999 	 */
2000 
2001 	arp_init();
2002 
2003 	/*
2004 	 *	Set the IP module up
2005 	 */
2006 
2007 	ip_init();
2008 
2009 	/* Initialise per-cpu ipv4 mibs */
2010 	if (init_ipv4_mibs())
2011 		panic("%s: Cannot init ipv4 mibs\n", __func__);
2012 
2013 	/* Setup TCP slab cache for open requests. */
2014 	tcp_init();
2015 
2016 	/* Setup UDP memory threshold */
2017 	udp_init();
2018 
2019 	/* Add UDP-Lite (RFC 3828) */
2020 	udplite4_register();
2021 
2022 	raw_init();
2023 
2024 	ping_init();
2025 
2026 	/*
2027 	 *	Set the ICMP layer up
2028 	 */
2029 
2030 	if (icmp_init() < 0)
2031 		panic("Failed to create the ICMP control socket.\n");
2032 
2033 	/*
2034 	 *	Initialise the multicast router
2035 	 */
2036 #if defined(CONFIG_IP_MROUTE)
2037 	if (ip_mr_init())
2038 		pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2039 #endif
2040 
2041 	if (init_inet_pernet_ops())
2042 		pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2043 
2044 	ipv4_proc_init();
2045 
2046 	ipfrag_init();
2047 
2048 	dev_add_pack(&ip_packet_type);
2049 
2050 	ip_tunnel_core_init();
2051 
2052 	rc = 0;
2053 out:
2054 	return rc;
2055 out_unregister_raw_proto:
2056 	proto_unregister(&raw_prot);
2057 out_unregister_udp_proto:
2058 	proto_unregister(&udp_prot);
2059 out_unregister_tcp_proto:
2060 	proto_unregister(&tcp_prot);
2061 	goto out;
2062 }
2063 
2064 fs_initcall(inet_init);
2065 
2066 /* ------------------------------------------------------------------------ */
2067 
2068 #ifdef CONFIG_PROC_FS
ipv4_proc_init(void)2069 static int __init ipv4_proc_init(void)
2070 {
2071 	int rc = 0;
2072 
2073 	if (raw_proc_init())
2074 		goto out_raw;
2075 	if (tcp4_proc_init())
2076 		goto out_tcp;
2077 	if (udp4_proc_init())
2078 		goto out_udp;
2079 	if (ping_proc_init())
2080 		goto out_ping;
2081 	if (ip_misc_proc_init())
2082 		goto out_misc;
2083 out:
2084 	return rc;
2085 out_misc:
2086 	ping_proc_exit();
2087 out_ping:
2088 	udp4_proc_exit();
2089 out_udp:
2090 	tcp4_proc_exit();
2091 out_tcp:
2092 	raw_proc_exit();
2093 out_raw:
2094 	rc = -ENOMEM;
2095 	goto out;
2096 }
2097 
2098 #else /* CONFIG_PROC_FS */
ipv4_proc_init(void)2099 static int __init ipv4_proc_init(void)
2100 {
2101 	return 0;
2102 }
2103 #endif /* CONFIG_PROC_FS */
2104