• 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/tcp.h>
103 #include <net/udp.h>
104 #include <net/udplite.h>
105 #include <net/ping.h>
106 #include <linux/skbuff.h>
107 #include <net/sock.h>
108 #include <net/raw.h>
109 #include <net/icmp.h>
110 #include <net/inet_common.h>
111 #include <net/ip_tunnels.h>
112 #include <net/xfrm.h>
113 #include <net/net_namespace.h>
114 #include <net/secure_seq.h>
115 #ifdef CONFIG_IP_MROUTE
116 #include <linux/mroute.h>
117 #endif
118 #include <net/l3mdev.h>
119 #include <net/compat.h>
120 
121 #include <trace/events/sock.h>
122 
123 /* The inetsw table contains everything that inet_create needs to
124  * build a new socket.
125  */
126 static struct list_head inetsw[SOCK_MAX];
127 static DEFINE_SPINLOCK(inetsw_lock);
128 
129 /* New destruction routine */
130 
inet_sock_destruct(struct sock * sk)131 void inet_sock_destruct(struct sock *sk)
132 {
133 	struct inet_sock *inet = inet_sk(sk);
134 
135 	__skb_queue_purge(&sk->sk_receive_queue);
136 	if (sk->sk_rx_skb_cache) {
137 		__kfree_skb(sk->sk_rx_skb_cache);
138 		sk->sk_rx_skb_cache = NULL;
139 	}
140 	__skb_queue_purge(&sk->sk_error_queue);
141 
142 	sk_mem_reclaim(sk);
143 
144 	if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
145 		pr_err("Attempt to release TCP socket in state %d %p\n",
146 		       sk->sk_state, sk);
147 		return;
148 	}
149 	if (!sock_flag(sk, SOCK_DEAD)) {
150 		pr_err("Attempt to release alive inet socket %p\n", sk);
151 		return;
152 	}
153 
154 	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
155 	WARN_ON(refcount_read(&sk->sk_wmem_alloc));
156 	WARN_ON(sk->sk_wmem_queued);
157 	WARN_ON(sk->sk_forward_alloc);
158 
159 	kfree(rcu_dereference_protected(inet->inet_opt, 1));
160 	dst_release(rcu_dereference_protected(sk->sk_dst_cache, 1));
161 	dst_release(rcu_dereference_protected(sk->sk_rx_dst, 1));
162 	sk_refcnt_debug_dec(sk);
163 }
164 EXPORT_SYMBOL(inet_sock_destruct);
165 
166 /*
167  *	The routines beyond this point handle the behaviour of an AF_INET
168  *	socket object. Mostly it punts to the subprotocols of IP to do
169  *	the work.
170  */
171 
172 /*
173  *	Automatically bind an unbound socket.
174  */
175 
inet_autobind(struct sock * sk)176 static int inet_autobind(struct sock *sk)
177 {
178 	struct inet_sock *inet;
179 	/* We may need to bind the socket. */
180 	lock_sock(sk);
181 	inet = inet_sk(sk);
182 	if (!inet->inet_num) {
183 		if (sk->sk_prot->get_port(sk, 0)) {
184 			release_sock(sk);
185 			return -EAGAIN;
186 		}
187 		inet->inet_sport = htons(inet->inet_num);
188 	}
189 	release_sock(sk);
190 	return 0;
191 }
192 
193 /*
194  *	Move a socket into listening state.
195  */
inet_listen(struct socket * sock,int backlog)196 int inet_listen(struct socket *sock, int backlog)
197 {
198 	struct sock *sk = sock->sk;
199 	unsigned char old_state;
200 	int err, tcp_fastopen;
201 
202 	lock_sock(sk);
203 
204 	err = -EINVAL;
205 	if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
206 		goto out;
207 
208 	old_state = sk->sk_state;
209 	if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
210 		goto out;
211 
212 	WRITE_ONCE(sk->sk_max_ack_backlog, backlog);
213 	/* Really, if the socket is already in listen state
214 	 * we can only allow the backlog to be adjusted.
215 	 */
216 	if (old_state != TCP_LISTEN) {
217 		/* Enable TFO w/o requiring TCP_FASTOPEN socket option.
218 		 * Note that only TCP sockets (SOCK_STREAM) will reach here.
219 		 * Also fastopen backlog may already been set via the option
220 		 * because the socket was in TCP_LISTEN state previously but
221 		 * was shutdown() rather than close().
222 		 */
223 		tcp_fastopen = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fastopen);
224 		if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
225 		    (tcp_fastopen & TFO_SERVER_ENABLE) &&
226 		    !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
227 			fastopen_queue_tune(sk, backlog);
228 			tcp_fastopen_init_key_once(sock_net(sk));
229 		}
230 
231 		err = inet_csk_listen_start(sk, backlog);
232 		if (err)
233 			goto out;
234 		tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL);
235 	}
236 	err = 0;
237 
238 out:
239 	release_sock(sk);
240 	return err;
241 }
242 EXPORT_SYMBOL(inet_listen);
243 
244 /*
245  *	Create an inet socket.
246  */
247 
inet_create(struct net * net,struct socket * sock,int protocol,int kern)248 static int inet_create(struct net *net, struct socket *sock, int protocol,
249 		       int kern)
250 {
251 	struct sock *sk;
252 	struct inet_protosw *answer;
253 	struct inet_sock *inet;
254 	struct proto *answer_prot;
255 	unsigned char answer_flags;
256 	int try_loading_module = 0;
257 	int err;
258 
259 	if (protocol < 0 || protocol >= IPPROTO_MAX)
260 		return -EINVAL;
261 
262 	sock->state = SS_UNCONNECTED;
263 
264 	/* Look for the requested type/protocol pair. */
265 lookup_protocol:
266 	err = -ESOCKTNOSUPPORT;
267 	rcu_read_lock();
268 	list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
269 
270 		err = 0;
271 		/* Check the non-wild match. */
272 		if (protocol == answer->protocol) {
273 			if (protocol != IPPROTO_IP)
274 				break;
275 		} else {
276 			/* Check for the two wild cases. */
277 			if (IPPROTO_IP == protocol) {
278 				protocol = answer->protocol;
279 				break;
280 			}
281 			if (IPPROTO_IP == answer->protocol)
282 				break;
283 		}
284 		err = -EPROTONOSUPPORT;
285 	}
286 
287 	if (unlikely(err)) {
288 		if (try_loading_module < 2) {
289 			rcu_read_unlock();
290 			/*
291 			 * Be more specific, e.g. net-pf-2-proto-132-type-1
292 			 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
293 			 */
294 			if (++try_loading_module == 1)
295 				request_module("net-pf-%d-proto-%d-type-%d",
296 					       PF_INET, protocol, sock->type);
297 			/*
298 			 * Fall back to generic, e.g. net-pf-2-proto-132
299 			 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
300 			 */
301 			else
302 				request_module("net-pf-%d-proto-%d",
303 					       PF_INET, protocol);
304 			goto lookup_protocol;
305 		} else
306 			goto out_rcu_unlock;
307 	}
308 
309 	err = -EPERM;
310 	if (sock->type == SOCK_RAW && !kern &&
311 	    !ns_capable(net->user_ns, CAP_NET_RAW))
312 		goto out_rcu_unlock;
313 
314 	sock->ops = answer->ops;
315 	answer_prot = answer->prot;
316 	answer_flags = answer->flags;
317 	rcu_read_unlock();
318 
319 	WARN_ON(!answer_prot->slab);
320 
321 	err = -ENOBUFS;
322 	sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
323 	if (!sk)
324 		goto out;
325 
326 	err = 0;
327 	if (INET_PROTOSW_REUSE & answer_flags)
328 		sk->sk_reuse = SK_CAN_REUSE;
329 
330 	if (INET_PROTOSW_ICSK & answer_flags)
331 		inet_init_csk_locks(sk);
332 
333 	inet = inet_sk(sk);
334 	inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
335 
336 	inet->nodefrag = 0;
337 
338 	if (SOCK_RAW == sock->type) {
339 		inet->inet_num = protocol;
340 		if (IPPROTO_RAW == protocol)
341 			inet->hdrincl = 1;
342 	}
343 
344 	if (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc))
345 		inet->pmtudisc = IP_PMTUDISC_DONT;
346 	else
347 		inet->pmtudisc = IP_PMTUDISC_WANT;
348 
349 	inet->inet_id = 0;
350 
351 	sock_init_data(sock, sk);
352 
353 	sk->sk_destruct	   = inet_sock_destruct;
354 	sk->sk_protocol	   = protocol;
355 	sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
356 
357 	inet->uc_ttl	= -1;
358 	inet->mc_loop	= 1;
359 	inet->mc_ttl	= 1;
360 	inet->mc_all	= 1;
361 	inet->mc_index	= 0;
362 	inet->mc_list	= NULL;
363 	inet->rcv_tos	= 0;
364 
365 	sk_refcnt_debug_inc(sk);
366 
367 	if (inet->inet_num) {
368 		/* It assumes that any protocol which allows
369 		 * the user to assign a number at socket
370 		 * creation time automatically
371 		 * shares.
372 		 */
373 		inet->inet_sport = htons(inet->inet_num);
374 		/* Add to protocol hash chains. */
375 		err = sk->sk_prot->hash(sk);
376 		if (err) {
377 			sk_common_release(sk);
378 			goto out;
379 		}
380 	}
381 
382 	if (sk->sk_prot->init) {
383 		err = sk->sk_prot->init(sk);
384 		if (err) {
385 			sk_common_release(sk);
386 			goto out;
387 		}
388 	}
389 
390 	if (!kern) {
391 		err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
392 		if (err) {
393 			sk_common_release(sk);
394 			goto out;
395 		}
396 	}
397 out:
398 	return err;
399 out_rcu_unlock:
400 	rcu_read_unlock();
401 	goto out;
402 }
403 
404 
405 /*
406  *	The peer socket should always be NULL (or else). When we call this
407  *	function we are destroying the object and from then on nobody
408  *	should refer to it.
409  */
inet_release(struct socket * sock)410 int inet_release(struct socket *sock)
411 {
412 	struct sock *sk = sock->sk;
413 
414 	if (sk) {
415 		long timeout;
416 
417 		if (!sk->sk_kern_sock)
418 			BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk);
419 
420 		/* Applications forget to leave groups before exiting */
421 		ip_mc_drop_socket(sk);
422 
423 		/* If linger is set, we don't return until the close
424 		 * is complete.  Otherwise we return immediately. The
425 		 * actually closing is done the same either way.
426 		 *
427 		 * If the close is due to the process exiting, we never
428 		 * linger..
429 		 */
430 		timeout = 0;
431 		if (sock_flag(sk, SOCK_LINGER) &&
432 		    !(current->flags & PF_EXITING))
433 			timeout = sk->sk_lingertime;
434 		sk->sk_prot->close(sk, timeout);
435 		sock->sk = NULL;
436 	}
437 	return 0;
438 }
439 EXPORT_SYMBOL(inet_release);
440 
inet_bind(struct socket * sock,struct sockaddr * uaddr,int addr_len)441 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
442 {
443 	struct sock *sk = sock->sk;
444 	int err;
445 
446 	/* If the socket has its own bind function then use it. (RAW) */
447 	if (sk->sk_prot->bind) {
448 		return sk->sk_prot->bind(sk, uaddr, addr_len);
449 	}
450 	if (addr_len < sizeof(struct sockaddr_in))
451 		return -EINVAL;
452 
453 	/* BPF prog is run before any checks are done so that if the prog
454 	 * changes context in a wrong way it will be caught.
455 	 */
456 	err = BPF_CGROUP_RUN_PROG_INET4_BIND(sk, uaddr);
457 	if (err)
458 		return err;
459 
460 	return __inet_bind(sk, uaddr, addr_len, BIND_WITH_LOCK);
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_can_nonlocal_bind(net, inet) &&
497 	    addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
498 	    chk_addr_ret != RTN_LOCAL &&
499 	    chk_addr_ret != RTN_MULTICAST &&
500 	    chk_addr_ret != RTN_BROADCAST)
501 		goto out;
502 
503 	snum = ntohs(addr->sin_port);
504 	err = -EACCES;
505 	if (snum && inet_port_requires_bind_service(net, snum) &&
506 	    !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
507 		goto out;
508 
509 	/*      We keep a pair of addresses. rcv_saddr is the one
510 	 *      used by hash lookups, and saddr is used for transmit.
511 	 *
512 	 *      In the BSD API these are the same except where it
513 	 *      would be illegal to use them (multicast/broadcast) in
514 	 *      which case the sending device address is used.
515 	 */
516 	if (flags & BIND_WITH_LOCK)
517 		lock_sock(sk);
518 
519 	/* Check these errors (active socket, double bind). */
520 	err = -EINVAL;
521 	if (sk->sk_state != TCP_CLOSE || inet->inet_num)
522 		goto out_release_sock;
523 
524 	inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
525 	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
526 		inet->inet_saddr = 0;  /* Use device */
527 
528 	/* Make sure we are allowed to bind here. */
529 	if (snum || !(inet->bind_address_no_port ||
530 		      (flags & BIND_FORCE_ADDRESS_NO_PORT))) {
531 		if (sk->sk_prot->get_port(sk, snum)) {
532 			inet->inet_saddr = inet->inet_rcv_saddr = 0;
533 			err = -EADDRINUSE;
534 			goto out_release_sock;
535 		}
536 		if (!(flags & BIND_FROM_BPF)) {
537 			err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
538 			if (err) {
539 				inet->inet_saddr = inet->inet_rcv_saddr = 0;
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 
596 	/* Basic assumption: if someone sets sk->sk_err, he _must_
597 	 * change state of the socket from TCP_SYN_*.
598 	 * Connect() does not allow to get error notifications
599 	 * without closing the socket.
600 	 */
601 	while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
602 		release_sock(sk);
603 		timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
604 		lock_sock(sk);
605 		if (signal_pending(current) || !timeo)
606 			break;
607 	}
608 	remove_wait_queue(sk_sleep(sk), &wait);
609 	sk->sk_write_pending -= writebias;
610 	return timeo;
611 }
612 
613 /*
614  *	Connect to a remote host. There is regrettably still a little
615  *	TCP 'magic' in here.
616  */
__inet_stream_connect(struct socket * sock,struct sockaddr * uaddr,int addr_len,int flags,int is_sendmsg)617 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
618 			  int addr_len, int flags, int is_sendmsg)
619 {
620 	struct sock *sk = sock->sk;
621 	int err;
622 	long timeo;
623 
624 	/*
625 	 * uaddr can be NULL and addr_len can be 0 if:
626 	 * sk is a TCP fastopen active socket and
627 	 * TCP_FASTOPEN_CONNECT sockopt is set and
628 	 * we already have a valid cookie for this socket.
629 	 * In this case, user can call write() after connect().
630 	 * write() will invoke tcp_sendmsg_fastopen() which calls
631 	 * __inet_stream_connect().
632 	 */
633 	if (uaddr) {
634 		if (addr_len < sizeof(uaddr->sa_family))
635 			return -EINVAL;
636 
637 		if (uaddr->sa_family == AF_UNSPEC) {
638 			err = sk->sk_prot->disconnect(sk, flags);
639 			sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
640 			goto out;
641 		}
642 	}
643 
644 	switch (sock->state) {
645 	default:
646 		err = -EINVAL;
647 		goto out;
648 	case SS_CONNECTED:
649 		err = -EISCONN;
650 		goto out;
651 	case SS_CONNECTING:
652 		if (inet_sk(sk)->defer_connect)
653 			err = is_sendmsg ? -EINPROGRESS : -EISCONN;
654 		else
655 			err = -EALREADY;
656 		/* Fall out of switch with err, set for this state */
657 		break;
658 	case SS_UNCONNECTED:
659 		err = -EISCONN;
660 		if (sk->sk_state != TCP_CLOSE)
661 			goto out;
662 
663 		if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
664 			err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
665 			if (err)
666 				goto out;
667 		}
668 
669 		err = sk->sk_prot->connect(sk, uaddr, addr_len);
670 		if (err < 0)
671 			goto out;
672 
673 		sock->state = SS_CONNECTING;
674 
675 		if (!err && inet_sk(sk)->defer_connect)
676 			goto out;
677 
678 		/* Just entered SS_CONNECTING state; the only
679 		 * difference is that return value in non-blocking
680 		 * case is EINPROGRESS, rather than EALREADY.
681 		 */
682 		err = -EINPROGRESS;
683 		break;
684 	}
685 
686 	timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
687 
688 	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
689 		int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
690 				tcp_sk(sk)->fastopen_req &&
691 				tcp_sk(sk)->fastopen_req->data ? 1 : 0;
692 
693 		/* Error code is set above */
694 		if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
695 			goto out;
696 
697 		err = sock_intr_errno(timeo);
698 		if (signal_pending(current))
699 			goto out;
700 	}
701 
702 	/* Connection was closed by RST, timeout, ICMP error
703 	 * or another process disconnected us.
704 	 */
705 	if (sk->sk_state == TCP_CLOSE)
706 		goto sock_error;
707 
708 	/* sk->sk_err may be not zero now, if RECVERR was ordered by user
709 	 * and error was received after socket entered established state.
710 	 * Hence, it is handled normally after connect() return successfully.
711 	 */
712 
713 	sock->state = SS_CONNECTED;
714 	err = 0;
715 out:
716 	return err;
717 
718 sock_error:
719 	err = sock_error(sk) ? : -ECONNABORTED;
720 	sock->state = SS_UNCONNECTED;
721 	if (sk->sk_prot->disconnect(sk, flags))
722 		sock->state = SS_DISCONNECTING;
723 	goto out;
724 }
725 EXPORT_SYMBOL(__inet_stream_connect);
726 
inet_stream_connect(struct socket * sock,struct sockaddr * uaddr,int addr_len,int flags)727 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
728 			int addr_len, int flags)
729 {
730 	int err;
731 
732 	lock_sock(sock->sk);
733 	err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
734 	release_sock(sock->sk);
735 	return err;
736 }
737 EXPORT_SYMBOL(inet_stream_connect);
738 
739 /*
740  *	Accept a pending connection. The TCP layer now gives BSD semantics.
741  */
742 
inet_accept(struct socket * sock,struct socket * newsock,int flags,bool kern)743 int inet_accept(struct socket *sock, struct socket *newsock, int flags,
744 		bool kern)
745 {
746 	struct sock *sk1 = sock->sk, *sk2;
747 	int err = -EINVAL;
748 
749 	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
750 	sk2 = READ_ONCE(sk1->sk_prot)->accept(sk1, flags, &err, kern);
751 	if (!sk2)
752 		goto do_err;
753 
754 	lock_sock(sk2);
755 
756 	sock_rps_record_flow(sk2);
757 	WARN_ON(!((1 << sk2->sk_state) &
758 		  (TCPF_ESTABLISHED | TCPF_SYN_RECV |
759 		  TCPF_CLOSE_WAIT | TCPF_CLOSE)));
760 
761 	sock_graft(sk2, newsock);
762 
763 	newsock->state = SS_CONNECTED;
764 	err = 0;
765 	release_sock(sk2);
766 do_err:
767 	return err;
768 }
769 EXPORT_SYMBOL(inet_accept);
770 
771 /*
772  *	This does both peername and sockname.
773  */
inet_getname(struct socket * sock,struct sockaddr * uaddr,int peer)774 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
775 		 int peer)
776 {
777 	struct sock *sk		= sock->sk;
778 	struct inet_sock *inet	= inet_sk(sk);
779 	DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
780 
781 	sin->sin_family = AF_INET;
782 	if (peer) {
783 		if (!inet->inet_dport ||
784 		    (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
785 		     peer == 1))
786 			return -ENOTCONN;
787 		sin->sin_port = inet->inet_dport;
788 		sin->sin_addr.s_addr = inet->inet_daddr;
789 	} else {
790 		__be32 addr = inet->inet_rcv_saddr;
791 		if (!addr)
792 			addr = inet->inet_saddr;
793 		sin->sin_port = inet->inet_sport;
794 		sin->sin_addr.s_addr = addr;
795 	}
796 	if (cgroup_bpf_enabled)
797 		BPF_CGROUP_RUN_SA_PROG_LOCK(sk, (struct sockaddr *)sin,
798 					    peer ? BPF_CGROUP_INET4_GETPEERNAME :
799 						   BPF_CGROUP_INET4_GETSOCKNAME,
800 					    NULL);
801 	memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
802 	return sizeof(*sin);
803 }
804 EXPORT_SYMBOL(inet_getname);
805 
inet_send_prepare(struct sock * sk)806 int inet_send_prepare(struct sock *sk)
807 {
808 	sock_rps_record_flow(sk);
809 
810 	/* We may need to bind the socket. */
811 	if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind &&
812 	    inet_autobind(sk))
813 		return -EAGAIN;
814 
815 	return 0;
816 }
817 EXPORT_SYMBOL_GPL(inet_send_prepare);
818 
inet_sendmsg(struct socket * sock,struct msghdr * msg,size_t size)819 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
820 {
821 	struct sock *sk = sock->sk;
822 
823 	if (unlikely(inet_send_prepare(sk)))
824 		return -EAGAIN;
825 
826 	return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
827 			       sk, msg, size);
828 }
829 EXPORT_SYMBOL(inet_sendmsg);
830 
inet_sendpage(struct socket * sock,struct page * page,int offset,size_t size,int flags)831 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
832 		      size_t size, int flags)
833 {
834 	struct sock *sk = sock->sk;
835 	const struct proto *prot;
836 
837 	if (unlikely(inet_send_prepare(sk)))
838 		return -EAGAIN;
839 
840 	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
841 	prot = READ_ONCE(sk->sk_prot);
842 	if (prot->sendpage)
843 		return prot->sendpage(sk, page, offset, size, flags);
844 	return sock_no_sendpage(sock, page, offset, size, flags);
845 }
846 EXPORT_SYMBOL(inet_sendpage);
847 
848 INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
849 					  size_t, int, int, int *));
inet_recvmsg(struct socket * sock,struct msghdr * msg,size_t size,int flags)850 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
851 		 int flags)
852 {
853 	struct sock *sk = sock->sk;
854 	int addr_len = 0;
855 	int err;
856 
857 	if (likely(!(flags & MSG_ERRQUEUE)))
858 		sock_rps_record_flow(sk);
859 
860 	err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
861 			      sk, msg, size, flags & MSG_DONTWAIT,
862 			      flags & ~MSG_DONTWAIT, &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 (copy_from_user(&ifr, p, sizeof(struct ifreq)))
965 			return -EFAULT;
966 		err = devinet_ioctl(net, cmd, &ifr);
967 		if (!err && copy_to_user(p, &ifr, sizeof(struct ifreq)))
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 (copy_from_user(&ifr, p, sizeof(struct ifreq)))
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 	.sendmsg_locked    = tcp_sendmsg_locked,
1055 	.sendpage_locked   = tcp_sendpage_locked,
1056 	.peek_len	   = tcp_peek_len,
1057 #ifdef CONFIG_COMPAT
1058 	.compat_ioctl	   = inet_compat_ioctl,
1059 #endif
1060 	.set_rcvlowat	   = tcp_set_rcvlowat,
1061 };
1062 EXPORT_SYMBOL(inet_stream_ops);
1063 
1064 const struct proto_ops inet_dgram_ops = {
1065 	.family		   = PF_INET,
1066 	.owner		   = THIS_MODULE,
1067 	.release	   = inet_release,
1068 	.bind		   = inet_bind,
1069 	.connect	   = inet_dgram_connect,
1070 	.socketpair	   = sock_no_socketpair,
1071 	.accept		   = sock_no_accept,
1072 	.getname	   = inet_getname,
1073 	.poll		   = udp_poll,
1074 	.ioctl		   = inet_ioctl,
1075 	.gettstamp	   = sock_gettstamp,
1076 	.listen		   = sock_no_listen,
1077 	.shutdown	   = inet_shutdown,
1078 	.setsockopt	   = sock_common_setsockopt,
1079 	.getsockopt	   = sock_common_getsockopt,
1080 	.sendmsg	   = inet_sendmsg,
1081 	.recvmsg	   = inet_recvmsg,
1082 	.mmap		   = sock_no_mmap,
1083 	.sendpage	   = inet_sendpage,
1084 	.set_peek_off	   = sk_set_peek_off,
1085 #ifdef CONFIG_COMPAT
1086 	.compat_ioctl	   = inet_compat_ioctl,
1087 #endif
1088 };
1089 EXPORT_SYMBOL(inet_dgram_ops);
1090 
1091 /*
1092  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1093  * udp_poll
1094  */
1095 static const struct proto_ops inet_sockraw_ops = {
1096 	.family		   = PF_INET,
1097 	.owner		   = THIS_MODULE,
1098 	.release	   = inet_release,
1099 	.bind		   = inet_bind,
1100 	.connect	   = inet_dgram_connect,
1101 	.socketpair	   = sock_no_socketpair,
1102 	.accept		   = sock_no_accept,
1103 	.getname	   = inet_getname,
1104 	.poll		   = datagram_poll,
1105 	.ioctl		   = inet_ioctl,
1106 	.gettstamp	   = sock_gettstamp,
1107 	.listen		   = sock_no_listen,
1108 	.shutdown	   = inet_shutdown,
1109 	.setsockopt	   = sock_common_setsockopt,
1110 	.getsockopt	   = sock_common_getsockopt,
1111 	.sendmsg	   = inet_sendmsg,
1112 	.recvmsg	   = inet_recvmsg,
1113 	.mmap		   = sock_no_mmap,
1114 	.sendpage	   = inet_sendpage,
1115 #ifdef CONFIG_COMPAT
1116 	.compat_ioctl	   = inet_compat_ioctl,
1117 #endif
1118 };
1119 
1120 static const struct net_proto_family inet_family_ops = {
1121 	.family = PF_INET,
1122 	.create = inet_create,
1123 	.owner	= THIS_MODULE,
1124 };
1125 
1126 /* Upon startup we insert all the elements in inetsw_array[] into
1127  * the linked list inetsw.
1128  */
1129 static struct inet_protosw inetsw_array[] =
1130 {
1131 	{
1132 		.type =       SOCK_STREAM,
1133 		.protocol =   IPPROTO_TCP,
1134 		.prot =       &tcp_prot,
1135 		.ops =        &inet_stream_ops,
1136 		.flags =      INET_PROTOSW_PERMANENT |
1137 			      INET_PROTOSW_ICSK,
1138 	},
1139 
1140 	{
1141 		.type =       SOCK_DGRAM,
1142 		.protocol =   IPPROTO_UDP,
1143 		.prot =       &udp_prot,
1144 		.ops =        &inet_dgram_ops,
1145 		.flags =      INET_PROTOSW_PERMANENT,
1146        },
1147 
1148        {
1149 		.type =       SOCK_DGRAM,
1150 		.protocol =   IPPROTO_ICMP,
1151 		.prot =       &ping_prot,
1152 		.ops =        &inet_sockraw_ops,
1153 		.flags =      INET_PROTOSW_REUSE,
1154        },
1155 
1156        {
1157 	       .type =       SOCK_RAW,
1158 	       .protocol =   IPPROTO_IP,	/* wild card */
1159 	       .prot =       &raw_prot,
1160 	       .ops =        &inet_sockraw_ops,
1161 	       .flags =      INET_PROTOSW_REUSE,
1162        }
1163 };
1164 
1165 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1166 
inet_register_protosw(struct inet_protosw * p)1167 void inet_register_protosw(struct inet_protosw *p)
1168 {
1169 	struct list_head *lh;
1170 	struct inet_protosw *answer;
1171 	int protocol = p->protocol;
1172 	struct list_head *last_perm;
1173 
1174 	spin_lock_bh(&inetsw_lock);
1175 
1176 	if (p->type >= SOCK_MAX)
1177 		goto out_illegal;
1178 
1179 	/* If we are trying to override a permanent protocol, bail. */
1180 	last_perm = &inetsw[p->type];
1181 	list_for_each(lh, &inetsw[p->type]) {
1182 		answer = list_entry(lh, struct inet_protosw, list);
1183 		/* Check only the non-wild match. */
1184 		if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1185 			break;
1186 		if (protocol == answer->protocol)
1187 			goto out_permanent;
1188 		last_perm = lh;
1189 	}
1190 
1191 	/* Add the new entry after the last permanent entry if any, so that
1192 	 * the new entry does not override a permanent entry when matched with
1193 	 * a wild-card protocol. But it is allowed to override any existing
1194 	 * non-permanent entry.  This means that when we remove this entry, the
1195 	 * system automatically returns to the old behavior.
1196 	 */
1197 	list_add_rcu(&p->list, last_perm);
1198 out:
1199 	spin_unlock_bh(&inetsw_lock);
1200 
1201 	return;
1202 
1203 out_permanent:
1204 	pr_err("Attempt to override permanent protocol %d\n", protocol);
1205 	goto out;
1206 
1207 out_illegal:
1208 	pr_err("Ignoring attempt to register invalid socket type %d\n",
1209 	       p->type);
1210 	goto out;
1211 }
1212 EXPORT_SYMBOL(inet_register_protosw);
1213 
inet_unregister_protosw(struct inet_protosw * p)1214 void inet_unregister_protosw(struct inet_protosw *p)
1215 {
1216 	if (INET_PROTOSW_PERMANENT & p->flags) {
1217 		pr_err("Attempt to unregister permanent protocol %d\n",
1218 		       p->protocol);
1219 	} else {
1220 		spin_lock_bh(&inetsw_lock);
1221 		list_del_rcu(&p->list);
1222 		spin_unlock_bh(&inetsw_lock);
1223 
1224 		synchronize_net();
1225 	}
1226 }
1227 EXPORT_SYMBOL(inet_unregister_protosw);
1228 
inet_sk_reselect_saddr(struct sock * sk)1229 static int inet_sk_reselect_saddr(struct sock *sk)
1230 {
1231 	struct inet_sock *inet = inet_sk(sk);
1232 	__be32 old_saddr = inet->inet_saddr;
1233 	__be32 daddr = inet->inet_daddr;
1234 	struct flowi4 *fl4;
1235 	struct rtable *rt;
1236 	__be32 new_saddr;
1237 	struct ip_options_rcu *inet_opt;
1238 
1239 	inet_opt = rcu_dereference_protected(inet->inet_opt,
1240 					     lockdep_sock_is_held(sk));
1241 	if (inet_opt && inet_opt->opt.srr)
1242 		daddr = inet_opt->opt.faddr;
1243 
1244 	/* Query new route. */
1245 	fl4 = &inet->cork.fl.u.ip4;
1246 	rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1247 			      sk->sk_bound_dev_if, sk->sk_protocol,
1248 			      inet->inet_sport, inet->inet_dport, sk);
1249 	if (IS_ERR(rt))
1250 		return PTR_ERR(rt);
1251 
1252 	sk_setup_caps(sk, &rt->dst);
1253 
1254 	new_saddr = fl4->saddr;
1255 
1256 	if (new_saddr == old_saddr)
1257 		return 0;
1258 
1259 	if (READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) > 1) {
1260 		pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1261 			__func__, &old_saddr, &new_saddr);
1262 	}
1263 
1264 	inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1265 
1266 	/*
1267 	 * XXX The only one ugly spot where we need to
1268 	 * XXX really change the sockets identity after
1269 	 * XXX it has entered the hashes. -DaveM
1270 	 *
1271 	 * Besides that, it does not check for connection
1272 	 * uniqueness. Wait for troubles.
1273 	 */
1274 	return __sk_prot_rehash(sk);
1275 }
1276 
inet_sk_rebuild_header(struct sock * sk)1277 int inet_sk_rebuild_header(struct sock *sk)
1278 {
1279 	struct inet_sock *inet = inet_sk(sk);
1280 	struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1281 	__be32 daddr;
1282 	struct ip_options_rcu *inet_opt;
1283 	struct flowi4 *fl4;
1284 	int err;
1285 
1286 	/* Route is OK, nothing to do. */
1287 	if (rt)
1288 		return 0;
1289 
1290 	/* Reroute. */
1291 	rcu_read_lock();
1292 	inet_opt = rcu_dereference(inet->inet_opt);
1293 	daddr = inet->inet_daddr;
1294 	if (inet_opt && inet_opt->opt.srr)
1295 		daddr = inet_opt->opt.faddr;
1296 	rcu_read_unlock();
1297 	fl4 = &inet->cork.fl.u.ip4;
1298 	rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1299 				   inet->inet_dport, inet->inet_sport,
1300 				   sk->sk_protocol, RT_CONN_FLAGS(sk),
1301 				   sk->sk_bound_dev_if);
1302 	if (!IS_ERR(rt)) {
1303 		err = 0;
1304 		sk_setup_caps(sk, &rt->dst);
1305 	} else {
1306 		err = PTR_ERR(rt);
1307 
1308 		/* Routing failed... */
1309 		sk->sk_route_caps = 0;
1310 		/*
1311 		 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1312 		 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1313 		 */
1314 		if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) ||
1315 		    sk->sk_state != TCP_SYN_SENT ||
1316 		    (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1317 		    (err = inet_sk_reselect_saddr(sk)) != 0)
1318 			sk->sk_err_soft = -err;
1319 	}
1320 
1321 	return err;
1322 }
1323 EXPORT_SYMBOL(inet_sk_rebuild_header);
1324 
inet_sk_set_state(struct sock * sk,int state)1325 void inet_sk_set_state(struct sock *sk, int state)
1326 {
1327 	trace_inet_sock_set_state(sk, sk->sk_state, state);
1328 	sk->sk_state = state;
1329 }
1330 EXPORT_SYMBOL(inet_sk_set_state);
1331 
inet_sk_state_store(struct sock * sk,int newstate)1332 void inet_sk_state_store(struct sock *sk, int newstate)
1333 {
1334 	trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1335 	smp_store_release(&sk->sk_state, newstate);
1336 }
1337 
inet_gso_segment(struct sk_buff * skb,netdev_features_t features)1338 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1339 				 netdev_features_t features)
1340 {
1341 	bool udpfrag = false, fixedid = false, gso_partial, encap;
1342 	struct sk_buff *segs = ERR_PTR(-EINVAL);
1343 	const struct net_offload *ops;
1344 	unsigned int offset = 0;
1345 	struct iphdr *iph;
1346 	int proto, tot_len;
1347 	int nhoff;
1348 	int ihl;
1349 	int id;
1350 
1351 	skb_reset_network_header(skb);
1352 	nhoff = skb_network_header(skb) - skb_mac_header(skb);
1353 	if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1354 		goto out;
1355 
1356 	iph = ip_hdr(skb);
1357 	ihl = iph->ihl * 4;
1358 	if (ihl < sizeof(*iph))
1359 		goto out;
1360 
1361 	id = ntohs(iph->id);
1362 	proto = iph->protocol;
1363 
1364 	/* Warning: after this point, iph might be no longer valid */
1365 	if (unlikely(!pskb_may_pull(skb, ihl)))
1366 		goto out;
1367 	__skb_pull(skb, ihl);
1368 
1369 	encap = SKB_GSO_CB(skb)->encap_level > 0;
1370 	if (encap)
1371 		features &= skb->dev->hw_enc_features;
1372 	SKB_GSO_CB(skb)->encap_level += ihl;
1373 
1374 	skb_reset_transport_header(skb);
1375 
1376 	segs = ERR_PTR(-EPROTONOSUPPORT);
1377 
1378 	if (!skb->encapsulation || encap) {
1379 		udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1380 		fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1381 
1382 		/* fixed ID is invalid if DF bit is not set */
1383 		if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1384 			goto out;
1385 	}
1386 
1387 	ops = rcu_dereference(inet_offloads[proto]);
1388 	if (likely(ops && ops->callbacks.gso_segment)) {
1389 		segs = ops->callbacks.gso_segment(skb, features);
1390 		if (!segs)
1391 			skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
1392 	}
1393 
1394 	if (IS_ERR_OR_NULL(segs))
1395 		goto out;
1396 
1397 	gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1398 
1399 	skb = segs;
1400 	do {
1401 		iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1402 		if (udpfrag) {
1403 			iph->frag_off = htons(offset >> 3);
1404 			if (skb->next)
1405 				iph->frag_off |= htons(IP_MF);
1406 			offset += skb->len - nhoff - ihl;
1407 			tot_len = skb->len - nhoff;
1408 		} else if (skb_is_gso(skb)) {
1409 			if (!fixedid) {
1410 				iph->id = htons(id);
1411 				id += skb_shinfo(skb)->gso_segs;
1412 			}
1413 
1414 			if (gso_partial)
1415 				tot_len = skb_shinfo(skb)->gso_size +
1416 					  SKB_GSO_CB(skb)->data_offset +
1417 					  skb->head - (unsigned char *)iph;
1418 			else
1419 				tot_len = skb->len - nhoff;
1420 		} else {
1421 			if (!fixedid)
1422 				iph->id = htons(id++);
1423 			tot_len = skb->len - nhoff;
1424 		}
1425 		iph->tot_len = htons(tot_len);
1426 		ip_send_check(iph);
1427 		if (encap)
1428 			skb_reset_inner_headers(skb);
1429 		skb->network_header = (u8 *)iph - skb->head;
1430 		skb_reset_mac_len(skb);
1431 	} while ((skb = skb->next));
1432 
1433 out:
1434 	return segs;
1435 }
1436 EXPORT_SYMBOL(inet_gso_segment);
1437 
ipip_gso_segment(struct sk_buff * skb,netdev_features_t features)1438 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1439 					netdev_features_t features)
1440 {
1441 	if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1442 		return ERR_PTR(-EINVAL);
1443 
1444 	return inet_gso_segment(skb, features);
1445 }
1446 
inet_gro_receive(struct list_head * head,struct sk_buff * skb)1447 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1448 {
1449 	const struct net_offload *ops;
1450 	struct sk_buff *pp = NULL;
1451 	const struct iphdr *iph;
1452 	struct sk_buff *p;
1453 	unsigned int hlen;
1454 	unsigned int off;
1455 	unsigned int id;
1456 	int flush = 1;
1457 	int proto;
1458 
1459 	off = skb_gro_offset(skb);
1460 	hlen = off + sizeof(*iph);
1461 	iph = skb_gro_header_fast(skb, off);
1462 	if (skb_gro_header_hard(skb, hlen)) {
1463 		iph = skb_gro_header_slow(skb, hlen, off);
1464 		if (unlikely(!iph))
1465 			goto out;
1466 	}
1467 
1468 	proto = iph->protocol;
1469 
1470 	rcu_read_lock();
1471 	ops = rcu_dereference(inet_offloads[proto]);
1472 	if (!ops || !ops->callbacks.gro_receive)
1473 		goto out_unlock;
1474 
1475 	if (*(u8 *)iph != 0x45)
1476 		goto out_unlock;
1477 
1478 	if (ip_is_fragment(iph))
1479 		goto out_unlock;
1480 
1481 	if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1482 		goto out_unlock;
1483 
1484 	id = ntohl(*(__be32 *)&iph->id);
1485 	flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1486 	id >>= 16;
1487 
1488 	list_for_each_entry(p, head, list) {
1489 		struct iphdr *iph2;
1490 		u16 flush_id;
1491 
1492 		if (!NAPI_GRO_CB(p)->same_flow)
1493 			continue;
1494 
1495 		iph2 = (struct iphdr *)(p->data + off);
1496 		/* The above works because, with the exception of the top
1497 		 * (inner most) layer, we only aggregate pkts with the same
1498 		 * hdr length so all the hdrs we'll need to verify will start
1499 		 * at the same offset.
1500 		 */
1501 		if ((iph->protocol ^ iph2->protocol) |
1502 		    ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1503 		    ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1504 			NAPI_GRO_CB(p)->same_flow = 0;
1505 			continue;
1506 		}
1507 
1508 		/* All fields must match except length and checksum. */
1509 		NAPI_GRO_CB(p)->flush |=
1510 			(iph->ttl ^ iph2->ttl) |
1511 			(iph->tos ^ iph2->tos) |
1512 			((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1513 
1514 		NAPI_GRO_CB(p)->flush |= flush;
1515 
1516 		/* We need to store of the IP ID check to be included later
1517 		 * when we can verify that this packet does in fact belong
1518 		 * to a given flow.
1519 		 */
1520 		flush_id = (u16)(id - ntohs(iph2->id));
1521 
1522 		/* This bit of code makes it much easier for us to identify
1523 		 * the cases where we are doing atomic vs non-atomic IP ID
1524 		 * checks.  Specifically an atomic check can return IP ID
1525 		 * values 0 - 0xFFFF, while a non-atomic check can only
1526 		 * return 0 or 0xFFFF.
1527 		 */
1528 		if (!NAPI_GRO_CB(p)->is_atomic ||
1529 		    !(iph->frag_off & htons(IP_DF))) {
1530 			flush_id ^= NAPI_GRO_CB(p)->count;
1531 			flush_id = flush_id ? 0xFFFF : 0;
1532 		}
1533 
1534 		/* If the previous IP ID value was based on an atomic
1535 		 * datagram we can overwrite the value and ignore it.
1536 		 */
1537 		if (NAPI_GRO_CB(skb)->is_atomic)
1538 			NAPI_GRO_CB(p)->flush_id = flush_id;
1539 		else
1540 			NAPI_GRO_CB(p)->flush_id |= flush_id;
1541 	}
1542 
1543 	NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1544 	NAPI_GRO_CB(skb)->flush |= flush;
1545 	skb_set_network_header(skb, off);
1546 	/* The above will be needed by the transport layer if there is one
1547 	 * immediately following this IP hdr.
1548 	 */
1549 
1550 	/* Note : No need to call skb_gro_postpull_rcsum() here,
1551 	 * as we already checked checksum over ipv4 header was 0
1552 	 */
1553 	skb_gro_pull(skb, sizeof(*iph));
1554 	skb_set_transport_header(skb, skb_gro_offset(skb));
1555 
1556 	pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1557 				       ops->callbacks.gro_receive, head, skb);
1558 
1559 out_unlock:
1560 	rcu_read_unlock();
1561 
1562 out:
1563 	skb_gro_flush_final(skb, pp, flush);
1564 
1565 	return pp;
1566 }
1567 EXPORT_SYMBOL(inet_gro_receive);
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 	rcu_read_lock();
1636 	ops = rcu_dereference(inet_offloads[proto]);
1637 	if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1638 		goto out_unlock;
1639 
1640 	/* Only need to add sizeof(*iph) to get to the next hdr below
1641 	 * because any hdr with option will have been flushed in
1642 	 * inet_gro_receive().
1643 	 */
1644 	err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1645 			      tcp4_gro_complete, udp4_gro_complete,
1646 			      skb, nhoff + sizeof(*iph));
1647 
1648 out_unlock:
1649 	rcu_read_unlock();
1650 
1651 	return err;
1652 }
1653 EXPORT_SYMBOL(inet_gro_complete);
1654 
ipip_gro_complete(struct sk_buff * skb,int nhoff)1655 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1656 {
1657 	skb->encapsulation = 1;
1658 	skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1659 	return inet_gro_complete(skb, nhoff);
1660 }
1661 
inet_ctl_sock_create(struct sock ** sk,unsigned short family,unsigned short type,unsigned char protocol,struct net * net)1662 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1663 			 unsigned short type, unsigned char protocol,
1664 			 struct net *net)
1665 {
1666 	struct socket *sock;
1667 	int rc = sock_create_kern(net, family, type, protocol, &sock);
1668 
1669 	if (rc == 0) {
1670 		*sk = sock->sk;
1671 		(*sk)->sk_allocation = GFP_ATOMIC;
1672 		/*
1673 		 * Unhash it so that IP input processing does not even see it,
1674 		 * we do not wish this socket to see incoming packets.
1675 		 */
1676 		(*sk)->sk_prot->unhash(*sk);
1677 	}
1678 	return rc;
1679 }
1680 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1681 
snmp_get_cpu_field(void __percpu * mib,int cpu,int offt)1682 u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt)
1683 {
1684 	return  *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt);
1685 }
1686 EXPORT_SYMBOL_GPL(snmp_get_cpu_field);
1687 
snmp_fold_field(void __percpu * mib,int offt)1688 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1689 {
1690 	unsigned long res = 0;
1691 	int i;
1692 
1693 	for_each_possible_cpu(i)
1694 		res += snmp_get_cpu_field(mib, i, offt);
1695 	return res;
1696 }
1697 EXPORT_SYMBOL_GPL(snmp_fold_field);
1698 
1699 #if BITS_PER_LONG==32
1700 
snmp_get_cpu_field64(void __percpu * mib,int cpu,int offt,size_t syncp_offset)1701 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1702 			 size_t syncp_offset)
1703 {
1704 	void *bhptr;
1705 	struct u64_stats_sync *syncp;
1706 	u64 v;
1707 	unsigned int start;
1708 
1709 	bhptr = per_cpu_ptr(mib, cpu);
1710 	syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1711 	do {
1712 		start = u64_stats_fetch_begin_irq(syncp);
1713 		v = *(((u64 *)bhptr) + offt);
1714 	} while (u64_stats_fetch_retry_irq(syncp, start));
1715 
1716 	return v;
1717 }
1718 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1719 
snmp_fold_field64(void __percpu * mib,int offt,size_t syncp_offset)1720 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1721 {
1722 	u64 res = 0;
1723 	int cpu;
1724 
1725 	for_each_possible_cpu(cpu) {
1726 		res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1727 	}
1728 	return res;
1729 }
1730 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1731 #endif
1732 
1733 #ifdef CONFIG_IP_MULTICAST
1734 static const struct net_protocol igmp_protocol = {
1735 	.handler =	igmp_rcv,
1736 	.netns_ok =	1,
1737 };
1738 #endif
1739 
1740 static const struct net_protocol tcp_protocol = {
1741 	.handler	=	tcp_v4_rcv,
1742 	.err_handler	=	tcp_v4_err,
1743 	.no_policy	=	1,
1744 	.netns_ok	=	1,
1745 	.icmp_strict_tag_validation = 1,
1746 };
1747 
1748 static const struct net_protocol udp_protocol = {
1749 	.handler =	udp_rcv,
1750 	.err_handler =	udp_err,
1751 	.no_policy =	1,
1752 	.netns_ok =	1,
1753 };
1754 
1755 static const struct net_protocol icmp_protocol = {
1756 	.handler =	icmp_rcv,
1757 	.err_handler =	icmp_err,
1758 	.no_policy =	1,
1759 	.netns_ok =	1,
1760 };
1761 
ipv4_mib_init_net(struct net * net)1762 static __net_init int ipv4_mib_init_net(struct net *net)
1763 {
1764 	int i;
1765 
1766 	net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1767 	if (!net->mib.tcp_statistics)
1768 		goto err_tcp_mib;
1769 	net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1770 	if (!net->mib.ip_statistics)
1771 		goto err_ip_mib;
1772 
1773 	for_each_possible_cpu(i) {
1774 		struct ipstats_mib *af_inet_stats;
1775 		af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1776 		u64_stats_init(&af_inet_stats->syncp);
1777 	}
1778 
1779 	net->mib.net_statistics = alloc_percpu(struct linux_mib);
1780 	if (!net->mib.net_statistics)
1781 		goto err_net_mib;
1782 	net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1783 	if (!net->mib.udp_statistics)
1784 		goto err_udp_mib;
1785 	net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1786 	if (!net->mib.udplite_statistics)
1787 		goto err_udplite_mib;
1788 	net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1789 	if (!net->mib.icmp_statistics)
1790 		goto err_icmp_mib;
1791 	net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1792 					      GFP_KERNEL);
1793 	if (!net->mib.icmpmsg_statistics)
1794 		goto err_icmpmsg_mib;
1795 
1796 	tcp_mib_init(net);
1797 	return 0;
1798 
1799 err_icmpmsg_mib:
1800 	free_percpu(net->mib.icmp_statistics);
1801 err_icmp_mib:
1802 	free_percpu(net->mib.udplite_statistics);
1803 err_udplite_mib:
1804 	free_percpu(net->mib.udp_statistics);
1805 err_udp_mib:
1806 	free_percpu(net->mib.net_statistics);
1807 err_net_mib:
1808 	free_percpu(net->mib.ip_statistics);
1809 err_ip_mib:
1810 	free_percpu(net->mib.tcp_statistics);
1811 err_tcp_mib:
1812 	return -ENOMEM;
1813 }
1814 
ipv4_mib_exit_net(struct net * net)1815 static __net_exit void ipv4_mib_exit_net(struct net *net)
1816 {
1817 	kfree(net->mib.icmpmsg_statistics);
1818 	free_percpu(net->mib.icmp_statistics);
1819 	free_percpu(net->mib.udplite_statistics);
1820 	free_percpu(net->mib.udp_statistics);
1821 	free_percpu(net->mib.net_statistics);
1822 	free_percpu(net->mib.ip_statistics);
1823 	free_percpu(net->mib.tcp_statistics);
1824 #ifdef CONFIG_MPTCP
1825 	/* allocated on demand, see mptcp_init_sock() */
1826 	free_percpu(net->mib.mptcp_statistics);
1827 #endif
1828 }
1829 
1830 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1831 	.init = ipv4_mib_init_net,
1832 	.exit = ipv4_mib_exit_net,
1833 };
1834 
init_ipv4_mibs(void)1835 static int __init init_ipv4_mibs(void)
1836 {
1837 	return register_pernet_subsys(&ipv4_mib_ops);
1838 }
1839 
inet_init_net(struct net * net)1840 static __net_init int inet_init_net(struct net *net)
1841 {
1842 	/*
1843 	 * Set defaults for local port range
1844 	 */
1845 	seqlock_init(&net->ipv4.ip_local_ports.lock);
1846 	net->ipv4.ip_local_ports.range[0] =  32768;
1847 	net->ipv4.ip_local_ports.range[1] =  60999;
1848 
1849 	seqlock_init(&net->ipv4.ping_group_range.lock);
1850 	/*
1851 	 * Sane defaults - nobody may create ping sockets.
1852 	 * Boot scripts should set this to distro-specific group.
1853 	 */
1854 	net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1855 	net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1856 
1857 	/* Default values for sysctl-controlled parameters.
1858 	 * We set them here, in case sysctl is not compiled.
1859 	 */
1860 	net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1861 	net->ipv4.sysctl_ip_fwd_update_priority = 1;
1862 	net->ipv4.sysctl_ip_dynaddr = 0;
1863 	net->ipv4.sysctl_ip_early_demux = 1;
1864 	net->ipv4.sysctl_udp_early_demux = 1;
1865 	net->ipv4.sysctl_tcp_early_demux = 1;
1866 	net->ipv4.sysctl_nexthop_compat_mode = 1;
1867 #ifdef CONFIG_SYSCTL
1868 	net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1869 #endif
1870 
1871 	/* Some igmp sysctl, whose values are always used */
1872 	net->ipv4.sysctl_igmp_max_memberships = 20;
1873 	net->ipv4.sysctl_igmp_max_msf = 10;
1874 	/* IGMP reports for link-local multicast groups are enabled by default */
1875 	net->ipv4.sysctl_igmp_llm_reports = 1;
1876 	net->ipv4.sysctl_igmp_qrv = 2;
1877 
1878 	return 0;
1879 }
1880 
1881 static __net_initdata struct pernet_operations af_inet_ops = {
1882 	.init = inet_init_net,
1883 };
1884 
init_inet_pernet_ops(void)1885 static int __init init_inet_pernet_ops(void)
1886 {
1887 	return register_pernet_subsys(&af_inet_ops);
1888 }
1889 
1890 static int ipv4_proc_init(void);
1891 
1892 /*
1893  *	IP protocol layer initialiser
1894  */
1895 
1896 static struct packet_offload ip_packet_offload __read_mostly = {
1897 	.type = cpu_to_be16(ETH_P_IP),
1898 	.callbacks = {
1899 		.gso_segment = inet_gso_segment,
1900 		.gro_receive = inet_gro_receive,
1901 		.gro_complete = inet_gro_complete,
1902 	},
1903 };
1904 
1905 static const struct net_offload ipip_offload = {
1906 	.callbacks = {
1907 		.gso_segment	= ipip_gso_segment,
1908 		.gro_receive	= ipip_gro_receive,
1909 		.gro_complete	= ipip_gro_complete,
1910 	},
1911 };
1912 
ipip_offload_init(void)1913 static int __init ipip_offload_init(void)
1914 {
1915 	return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1916 }
1917 
ipv4_offload_init(void)1918 static int __init ipv4_offload_init(void)
1919 {
1920 	/*
1921 	 * Add offloads
1922 	 */
1923 	if (udpv4_offload_init() < 0)
1924 		pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1925 	if (tcpv4_offload_init() < 0)
1926 		pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1927 	if (ipip_offload_init() < 0)
1928 		pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1929 
1930 	dev_add_offload(&ip_packet_offload);
1931 	return 0;
1932 }
1933 
1934 fs_initcall(ipv4_offload_init);
1935 
1936 static struct packet_type ip_packet_type __read_mostly = {
1937 	.type = cpu_to_be16(ETH_P_IP),
1938 	.func = ip_rcv,
1939 	.list_func = ip_list_rcv,
1940 };
1941 
inet_init(void)1942 static int __init inet_init(void)
1943 {
1944 	struct inet_protosw *q;
1945 	struct list_head *r;
1946 	int rc;
1947 
1948 	sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1949 
1950 	rc = proto_register(&tcp_prot, 1);
1951 	if (rc)
1952 		goto out;
1953 
1954 	rc = proto_register(&udp_prot, 1);
1955 	if (rc)
1956 		goto out_unregister_tcp_proto;
1957 
1958 	rc = proto_register(&raw_prot, 1);
1959 	if (rc)
1960 		goto out_unregister_udp_proto;
1961 
1962 	rc = proto_register(&ping_prot, 1);
1963 	if (rc)
1964 		goto out_unregister_raw_proto;
1965 
1966 	/*
1967 	 *	Tell SOCKET that we are alive...
1968 	 */
1969 
1970 	(void)sock_register(&inet_family_ops);
1971 
1972 #ifdef CONFIG_SYSCTL
1973 	ip_static_sysctl_init();
1974 #endif
1975 
1976 	/*
1977 	 *	Add all the base protocols.
1978 	 */
1979 
1980 	if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1981 		pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1982 	if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1983 		pr_crit("%s: Cannot add UDP protocol\n", __func__);
1984 	if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1985 		pr_crit("%s: Cannot add TCP protocol\n", __func__);
1986 #ifdef CONFIG_IP_MULTICAST
1987 	if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1988 		pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1989 #endif
1990 
1991 	/* Register the socket-side information for inet_create. */
1992 	for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1993 		INIT_LIST_HEAD(r);
1994 
1995 	for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1996 		inet_register_protosw(q);
1997 
1998 	/*
1999 	 *	Set the ARP module up
2000 	 */
2001 
2002 	arp_init();
2003 
2004 	/*
2005 	 *	Set the IP module up
2006 	 */
2007 
2008 	ip_init();
2009 
2010 	/* Initialise per-cpu ipv4 mibs */
2011 	if (init_ipv4_mibs())
2012 		panic("%s: Cannot init ipv4 mibs\n", __func__);
2013 
2014 	/* Setup TCP slab cache for open requests. */
2015 	tcp_init();
2016 
2017 	/* Setup UDP memory threshold */
2018 	udp_init();
2019 
2020 	/* Add UDP-Lite (RFC 3828) */
2021 	udplite4_register();
2022 
2023 	raw_init();
2024 
2025 	ping_init();
2026 
2027 	/*
2028 	 *	Set the ICMP layer up
2029 	 */
2030 
2031 	if (icmp_init() < 0)
2032 		panic("Failed to create the ICMP control socket.\n");
2033 
2034 	/*
2035 	 *	Initialise the multicast router
2036 	 */
2037 #if defined(CONFIG_IP_MROUTE)
2038 	if (ip_mr_init())
2039 		pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2040 #endif
2041 
2042 	if (init_inet_pernet_ops())
2043 		pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2044 
2045 	ipv4_proc_init();
2046 
2047 	ipfrag_init();
2048 
2049 	dev_add_pack(&ip_packet_type);
2050 
2051 	ip_tunnel_core_init();
2052 
2053 	rc = 0;
2054 out:
2055 	return rc;
2056 out_unregister_raw_proto:
2057 	proto_unregister(&raw_prot);
2058 out_unregister_udp_proto:
2059 	proto_unregister(&udp_prot);
2060 out_unregister_tcp_proto:
2061 	proto_unregister(&tcp_prot);
2062 	goto out;
2063 }
2064 
2065 fs_initcall(inet_init);
2066 
2067 /* ------------------------------------------------------------------------ */
2068 
2069 #ifdef CONFIG_PROC_FS
ipv4_proc_init(void)2070 static int __init ipv4_proc_init(void)
2071 {
2072 	int rc = 0;
2073 
2074 	if (raw_proc_init())
2075 		goto out_raw;
2076 	if (tcp4_proc_init())
2077 		goto out_tcp;
2078 	if (udp4_proc_init())
2079 		goto out_udp;
2080 	if (ping_proc_init())
2081 		goto out_ping;
2082 	if (ip_misc_proc_init())
2083 		goto out_misc;
2084 out:
2085 	return rc;
2086 out_misc:
2087 	ping_proc_exit();
2088 out_ping:
2089 	udp4_proc_exit();
2090 out_udp:
2091 	tcp4_proc_exit();
2092 out_tcp:
2093 	raw_proc_exit();
2094 out_raw:
2095 	rc = -ENOMEM;
2096 	goto out;
2097 }
2098 
2099 #else /* CONFIG_PROC_FS */
ipv4_proc_init(void)2100 static int __init ipv4_proc_init(void)
2101 {
2102 	return 0;
2103 }
2104 #endif /* CONFIG_PROC_FS */
2105