Loading...
1/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * PF_INET protocol family socket handler.
7 *
8 * Authors: Ross Biro
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Florian La Roche, <flla@stud.uni-sb.de>
11 * Alan Cox, <A.Cox@swansea.ac.uk>
12 *
13 * Changes (see also sock.c)
14 *
15 * piggy,
16 * Karl Knutson : Socket protocol table
17 * A.N.Kuznetsov : Socket death error in accept().
18 * John Richardson : Fix non blocking error in connect()
19 * so sockets that fail to connect
20 * don't return -EINPROGRESS.
21 * Alan Cox : Asynchronous I/O support
22 * Alan Cox : Keep correct socket pointer on sock
23 * structures
24 * when accept() ed
25 * Alan Cox : Semantics of SO_LINGER aren't state
26 * moved to close when you look carefully.
27 * With this fixed and the accept bug fixed
28 * some RPC stuff seems happier.
29 * Niibe Yutaka : 4.4BSD style write async I/O
30 * Alan Cox,
31 * Tony Gale : Fixed reuse semantics.
32 * Alan Cox : bind() shouldn't abort existing but dead
33 * sockets. Stops FTP netin:.. I hope.
34 * Alan Cox : bind() works correctly for RAW sockets.
35 * Note that FreeBSD at least was broken
36 * in this respect so be careful with
37 * compatibility tests...
38 * Alan Cox : routing cache support
39 * Alan Cox : memzero the socket structure for
40 * compactness.
41 * Matt Day : nonblock connect error handler
42 * Alan Cox : Allow large numbers of pending sockets
43 * (eg for big web sites), but only if
44 * specifically application requested.
45 * Alan Cox : New buffering throughout IP. Used
46 * dumbly.
47 * Alan Cox : New buffering now used smartly.
48 * Alan Cox : BSD rather than common sense
49 * interpretation of listen.
50 * Germano Caronni : Assorted small races.
51 * Alan Cox : sendmsg/recvmsg basic support.
52 * Alan Cox : Only sendmsg/recvmsg now supported.
53 * Alan Cox : Locked down bind (see security list).
54 * Alan Cox : Loosened bind a little.
55 * Mike McLagan : ADD/DEL DLCI Ioctls
56 * Willy Konynenberg : Transparent proxying support.
57 * David S. Miller : New socket lookup architecture.
58 * Some other random speedups.
59 * Cyrus Durgin : Cleaned up file for kmod hacks.
60 * Andi Kleen : Fix inet_stream_connect TCP race.
61 *
62 * This program is free software; you can redistribute it and/or
63 * modify it under the terms of the GNU General Public License
64 * as published by the Free Software Foundation; either version
65 * 2 of the License, or (at your option) any later version.
66 */
67
68#define pr_fmt(fmt) "IPv4: " fmt
69
70#include <linux/err.h>
71#include <linux/errno.h>
72#include <linux/types.h>
73#include <linux/socket.h>
74#include <linux/in.h>
75#include <linux/kernel.h>
76#include <linux/module.h>
77#include <linux/sched.h>
78#include <linux/timer.h>
79#include <linux/string.h>
80#include <linux/sockios.h>
81#include <linux/net.h>
82#include <linux/capability.h>
83#include <linux/fcntl.h>
84#include <linux/mm.h>
85#include <linux/interrupt.h>
86#include <linux/stat.h>
87#include <linux/init.h>
88#include <linux/poll.h>
89#include <linux/netfilter_ipv4.h>
90#include <linux/random.h>
91#include <linux/slab.h>
92
93#include <asm/uaccess.h>
94
95#include <linux/inet.h>
96#include <linux/igmp.h>
97#include <linux/inetdevice.h>
98#include <linux/netdevice.h>
99#include <net/checksum.h>
100#include <net/ip.h>
101#include <net/protocol.h>
102#include <net/arp.h>
103#include <net/route.h>
104#include <net/ip_fib.h>
105#include <net/inet_connection_sock.h>
106#include <net/tcp.h>
107#include <net/udp.h>
108#include <net/udplite.h>
109#include <net/ping.h>
110#include <linux/skbuff.h>
111#include <net/sock.h>
112#include <net/raw.h>
113#include <net/icmp.h>
114#include <net/inet_common.h>
115#include <net/xfrm.h>
116#include <net/net_namespace.h>
117#include <net/secure_seq.h>
118#ifdef CONFIG_IP_MROUTE
119#include <linux/mroute.h>
120#endif
121
122
123/* The inetsw table contains everything that inet_create needs to
124 * build a new socket.
125 */
126static struct list_head inetsw[SOCK_MAX];
127static DEFINE_SPINLOCK(inetsw_lock);
128
129/* New destruction routine */
130
131void inet_sock_destruct(struct sock *sk)
132{
133 struct inet_sock *inet = inet_sk(sk);
134
135 __skb_queue_purge(&sk->sk_receive_queue);
136 __skb_queue_purge(&sk->sk_error_queue);
137
138 sk_mem_reclaim(sk);
139
140 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
141 pr_err("Attempt to release TCP socket in state %d %p\n",
142 sk->sk_state, sk);
143 return;
144 }
145 if (!sock_flag(sk, SOCK_DEAD)) {
146 pr_err("Attempt to release alive inet socket %p\n", sk);
147 return;
148 }
149
150 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
151 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
152 WARN_ON(sk->sk_wmem_queued);
153 WARN_ON(sk->sk_forward_alloc);
154
155 kfree(rcu_dereference_protected(inet->inet_opt, 1));
156 dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
157 dst_release(sk->sk_rx_dst);
158 sk_refcnt_debug_dec(sk);
159}
160EXPORT_SYMBOL(inet_sock_destruct);
161
162/*
163 * The routines beyond this point handle the behaviour of an AF_INET
164 * socket object. Mostly it punts to the subprotocols of IP to do
165 * the work.
166 */
167
168/*
169 * Automatically bind an unbound socket.
170 */
171
172static int inet_autobind(struct sock *sk)
173{
174 struct inet_sock *inet;
175 /* We may need to bind the socket. */
176 lock_sock(sk);
177 inet = inet_sk(sk);
178 if (!inet->inet_num) {
179 if (sk->sk_prot->get_port(sk, 0)) {
180 release_sock(sk);
181 return -EAGAIN;
182 }
183 inet->inet_sport = htons(inet->inet_num);
184 }
185 release_sock(sk);
186 return 0;
187}
188
189/*
190 * Move a socket into listening state.
191 */
192int inet_listen(struct socket *sock, int backlog)
193{
194 struct sock *sk = sock->sk;
195 unsigned char old_state;
196 int err;
197
198 lock_sock(sk);
199
200 err = -EINVAL;
201 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
202 goto out;
203
204 old_state = sk->sk_state;
205 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
206 goto out;
207
208 /* Really, if the socket is already in listen state
209 * we can only allow the backlog to be adjusted.
210 */
211 if (old_state != TCP_LISTEN) {
212 /* Check special setups for testing purpose to enable TFO w/o
213 * requiring TCP_FASTOPEN sockopt.
214 * Note that only TCP sockets (SOCK_STREAM) will reach here.
215 * Also fastopenq may already been allocated because this
216 * socket was in TCP_LISTEN state previously but was
217 * shutdown() (rather than close()).
218 */
219 if ((sysctl_tcp_fastopen & TFO_SERVER_ENABLE) != 0 &&
220 inet_csk(sk)->icsk_accept_queue.fastopenq == NULL) {
221 if ((sysctl_tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) != 0)
222 err = fastopen_init_queue(sk, backlog);
223 else if ((sysctl_tcp_fastopen &
224 TFO_SERVER_WO_SOCKOPT2) != 0)
225 err = fastopen_init_queue(sk,
226 ((uint)sysctl_tcp_fastopen) >> 16);
227 else
228 err = 0;
229 if (err)
230 goto out;
231 }
232 err = inet_csk_listen_start(sk, backlog);
233 if (err)
234 goto out;
235 }
236 sk->sk_max_ack_backlog = backlog;
237 err = 0;
238
239out:
240 release_sock(sk);
241 return err;
242}
243EXPORT_SYMBOL(inet_listen);
244
245/*
246 * Create an inet socket.
247 */
248
249static int inet_create(struct net *net, struct socket *sock, int protocol,
250 int kern)
251{
252 struct sock *sk;
253 struct inet_protosw *answer;
254 struct inet_sock *inet;
255 struct proto *answer_prot;
256 unsigned char answer_flags;
257 char answer_no_check;
258 int try_loading_module = 0;
259 int err;
260
261 sock->state = SS_UNCONNECTED;
262
263 /* Look for the requested type/protocol pair. */
264lookup_protocol:
265 err = -ESOCKTNOSUPPORT;
266 rcu_read_lock();
267 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
268
269 err = 0;
270 /* Check the non-wild match. */
271 if (protocol == answer->protocol) {
272 if (protocol != IPPROTO_IP)
273 break;
274 } else {
275 /* Check for the two wild cases. */
276 if (IPPROTO_IP == protocol) {
277 protocol = answer->protocol;
278 break;
279 }
280 if (IPPROTO_IP == answer->protocol)
281 break;
282 }
283 err = -EPROTONOSUPPORT;
284 }
285
286 if (unlikely(err)) {
287 if (try_loading_module < 2) {
288 rcu_read_unlock();
289 /*
290 * Be more specific, e.g. net-pf-2-proto-132-type-1
291 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
292 */
293 if (++try_loading_module == 1)
294 request_module("net-pf-%d-proto-%d-type-%d",
295 PF_INET, protocol, sock->type);
296 /*
297 * Fall back to generic, e.g. net-pf-2-proto-132
298 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
299 */
300 else
301 request_module("net-pf-%d-proto-%d",
302 PF_INET, protocol);
303 goto lookup_protocol;
304 } else
305 goto out_rcu_unlock;
306 }
307
308 err = -EPERM;
309 if (sock->type == SOCK_RAW && !kern &&
310 !ns_capable(net->user_ns, CAP_NET_RAW))
311 goto out_rcu_unlock;
312
313 sock->ops = answer->ops;
314 answer_prot = answer->prot;
315 answer_no_check = answer->no_check;
316 answer_flags = answer->flags;
317 rcu_read_unlock();
318
319 WARN_ON(answer_prot->slab == NULL);
320
321 err = -ENOBUFS;
322 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
323 if (sk == NULL)
324 goto out;
325
326 err = 0;
327 sk->sk_no_check = answer_no_check;
328 if (INET_PROTOSW_REUSE & answer_flags)
329 sk->sk_reuse = SK_CAN_REUSE;
330
331 inet = inet_sk(sk);
332 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
333
334 inet->nodefrag = 0;
335
336 if (SOCK_RAW == sock->type) {
337 inet->inet_num = protocol;
338 if (IPPROTO_RAW == protocol)
339 inet->hdrincl = 1;
340 }
341
342 if (net->ipv4.sysctl_ip_no_pmtu_disc)
343 inet->pmtudisc = IP_PMTUDISC_DONT;
344 else
345 inet->pmtudisc = IP_PMTUDISC_WANT;
346
347 inet->inet_id = 0;
348
349 sock_init_data(sock, sk);
350
351 sk->sk_destruct = inet_sock_destruct;
352 sk->sk_protocol = protocol;
353 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
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 sk->sk_prot->hash(sk);
374 }
375
376 if (sk->sk_prot->init) {
377 err = sk->sk_prot->init(sk);
378 if (err)
379 sk_common_release(sk);
380 }
381out:
382 return err;
383out_rcu_unlock:
384 rcu_read_unlock();
385 goto out;
386}
387
388
389/*
390 * The peer socket should always be NULL (or else). When we call this
391 * function we are destroying the object and from then on nobody
392 * should refer to it.
393 */
394int inet_release(struct socket *sock)
395{
396 struct sock *sk = sock->sk;
397
398 if (sk) {
399 long timeout;
400
401 sock_rps_reset_flow(sk);
402
403 /* Applications forget to leave groups before exiting */
404 ip_mc_drop_socket(sk);
405
406 /* If linger is set, we don't return until the close
407 * is complete. Otherwise we return immediately. The
408 * actually closing is done the same either way.
409 *
410 * If the close is due to the process exiting, we never
411 * linger..
412 */
413 timeout = 0;
414 if (sock_flag(sk, SOCK_LINGER) &&
415 !(current->flags & PF_EXITING))
416 timeout = sk->sk_lingertime;
417 sock->sk = NULL;
418 sk->sk_prot->close(sk, timeout);
419 }
420 return 0;
421}
422EXPORT_SYMBOL(inet_release);
423
424/* It is off by default, see below. */
425int sysctl_ip_nonlocal_bind __read_mostly;
426EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
427
428int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
429{
430 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
431 struct sock *sk = sock->sk;
432 struct inet_sock *inet = inet_sk(sk);
433 struct net *net = sock_net(sk);
434 unsigned short snum;
435 int chk_addr_ret;
436 int err;
437
438 /* If the socket has its own bind function then use it. (RAW) */
439 if (sk->sk_prot->bind) {
440 err = sk->sk_prot->bind(sk, uaddr, addr_len);
441 goto out;
442 }
443 err = -EINVAL;
444 if (addr_len < sizeof(struct sockaddr_in))
445 goto out;
446
447 if (addr->sin_family != AF_INET) {
448 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
449 * only if s_addr is INADDR_ANY.
450 */
451 err = -EAFNOSUPPORT;
452 if (addr->sin_family != AF_UNSPEC ||
453 addr->sin_addr.s_addr != htonl(INADDR_ANY))
454 goto out;
455 }
456
457 chk_addr_ret = inet_addr_type(net, addr->sin_addr.s_addr);
458
459 /* Not specified by any standard per-se, however it breaks too
460 * many applications when removed. It is unfortunate since
461 * allowing applications to make a non-local bind solves
462 * several problems with systems using dynamic addressing.
463 * (ie. your servers still start up even if your ISDN link
464 * is temporarily down)
465 */
466 err = -EADDRNOTAVAIL;
467 if (!sysctl_ip_nonlocal_bind &&
468 !(inet->freebind || inet->transparent) &&
469 addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
470 chk_addr_ret != RTN_LOCAL &&
471 chk_addr_ret != RTN_MULTICAST &&
472 chk_addr_ret != RTN_BROADCAST)
473 goto out;
474
475 snum = ntohs(addr->sin_port);
476 err = -EACCES;
477 if (snum && snum < PROT_SOCK &&
478 !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
479 goto out;
480
481 /* We keep a pair of addresses. rcv_saddr is the one
482 * used by hash lookups, and saddr is used for transmit.
483 *
484 * In the BSD API these are the same except where it
485 * would be illegal to use them (multicast/broadcast) in
486 * which case the sending device address is used.
487 */
488 lock_sock(sk);
489
490 /* Check these errors (active socket, double bind). */
491 err = -EINVAL;
492 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
493 goto out_release_sock;
494
495 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
496 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
497 inet->inet_saddr = 0; /* Use device */
498
499 /* Make sure we are allowed to bind here. */
500 if (sk->sk_prot->get_port(sk, snum)) {
501 inet->inet_saddr = inet->inet_rcv_saddr = 0;
502 err = -EADDRINUSE;
503 goto out_release_sock;
504 }
505
506 if (inet->inet_rcv_saddr)
507 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
508 if (snum)
509 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
510 inet->inet_sport = htons(inet->inet_num);
511 inet->inet_daddr = 0;
512 inet->inet_dport = 0;
513 sk_dst_reset(sk);
514 err = 0;
515out_release_sock:
516 release_sock(sk);
517out:
518 return err;
519}
520EXPORT_SYMBOL(inet_bind);
521
522int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
523 int addr_len, int flags)
524{
525 struct sock *sk = sock->sk;
526
527 if (addr_len < sizeof(uaddr->sa_family))
528 return -EINVAL;
529 if (uaddr->sa_family == AF_UNSPEC)
530 return sk->sk_prot->disconnect(sk, flags);
531
532 if (!inet_sk(sk)->inet_num && inet_autobind(sk))
533 return -EAGAIN;
534 return sk->sk_prot->connect(sk, uaddr, addr_len);
535}
536EXPORT_SYMBOL(inet_dgram_connect);
537
538static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
539{
540 DEFINE_WAIT(wait);
541
542 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
543 sk->sk_write_pending += writebias;
544
545 /* Basic assumption: if someone sets sk->sk_err, he _must_
546 * change state of the socket from TCP_SYN_*.
547 * Connect() does not allow to get error notifications
548 * without closing the socket.
549 */
550 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
551 release_sock(sk);
552 timeo = schedule_timeout(timeo);
553 lock_sock(sk);
554 if (signal_pending(current) || !timeo)
555 break;
556 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
557 }
558 finish_wait(sk_sleep(sk), &wait);
559 sk->sk_write_pending -= writebias;
560 return timeo;
561}
562
563/*
564 * Connect to a remote host. There is regrettably still a little
565 * TCP 'magic' in here.
566 */
567int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
568 int addr_len, int flags)
569{
570 struct sock *sk = sock->sk;
571 int err;
572 long timeo;
573
574 if (addr_len < sizeof(uaddr->sa_family))
575 return -EINVAL;
576
577 if (uaddr->sa_family == AF_UNSPEC) {
578 err = sk->sk_prot->disconnect(sk, flags);
579 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
580 goto out;
581 }
582
583 switch (sock->state) {
584 default:
585 err = -EINVAL;
586 goto out;
587 case SS_CONNECTED:
588 err = -EISCONN;
589 goto out;
590 case SS_CONNECTING:
591 err = -EALREADY;
592 /* Fall out of switch with err, set for this state */
593 break;
594 case SS_UNCONNECTED:
595 err = -EISCONN;
596 if (sk->sk_state != TCP_CLOSE)
597 goto out;
598
599 err = sk->sk_prot->connect(sk, uaddr, addr_len);
600 if (err < 0)
601 goto out;
602
603 sock->state = SS_CONNECTING;
604
605 /* Just entered SS_CONNECTING state; the only
606 * difference is that return value in non-blocking
607 * case is EINPROGRESS, rather than EALREADY.
608 */
609 err = -EINPROGRESS;
610 break;
611 }
612
613 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
614
615 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
616 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
617 tcp_sk(sk)->fastopen_req &&
618 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
619
620 /* Error code is set above */
621 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
622 goto out;
623
624 err = sock_intr_errno(timeo);
625 if (signal_pending(current))
626 goto out;
627 }
628
629 /* Connection was closed by RST, timeout, ICMP error
630 * or another process disconnected us.
631 */
632 if (sk->sk_state == TCP_CLOSE)
633 goto sock_error;
634
635 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
636 * and error was received after socket entered established state.
637 * Hence, it is handled normally after connect() return successfully.
638 */
639
640 sock->state = SS_CONNECTED;
641 err = 0;
642out:
643 return err;
644
645sock_error:
646 err = sock_error(sk) ? : -ECONNABORTED;
647 sock->state = SS_UNCONNECTED;
648 if (sk->sk_prot->disconnect(sk, flags))
649 sock->state = SS_DISCONNECTING;
650 goto out;
651}
652EXPORT_SYMBOL(__inet_stream_connect);
653
654int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
655 int addr_len, int flags)
656{
657 int err;
658
659 lock_sock(sock->sk);
660 err = __inet_stream_connect(sock, uaddr, addr_len, flags);
661 release_sock(sock->sk);
662 return err;
663}
664EXPORT_SYMBOL(inet_stream_connect);
665
666/*
667 * Accept a pending connection. The TCP layer now gives BSD semantics.
668 */
669
670int inet_accept(struct socket *sock, struct socket *newsock, int flags)
671{
672 struct sock *sk1 = sock->sk;
673 int err = -EINVAL;
674 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
675
676 if (!sk2)
677 goto do_err;
678
679 lock_sock(sk2);
680
681 sock_rps_record_flow(sk2);
682 WARN_ON(!((1 << sk2->sk_state) &
683 (TCPF_ESTABLISHED | TCPF_SYN_RECV |
684 TCPF_CLOSE_WAIT | TCPF_CLOSE)));
685
686 sock_graft(sk2, newsock);
687
688 newsock->state = SS_CONNECTED;
689 err = 0;
690 release_sock(sk2);
691do_err:
692 return err;
693}
694EXPORT_SYMBOL(inet_accept);
695
696
697/*
698 * This does both peername and sockname.
699 */
700int inet_getname(struct socket *sock, struct sockaddr *uaddr,
701 int *uaddr_len, int peer)
702{
703 struct sock *sk = sock->sk;
704 struct inet_sock *inet = inet_sk(sk);
705 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
706
707 sin->sin_family = AF_INET;
708 if (peer) {
709 if (!inet->inet_dport ||
710 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
711 peer == 1))
712 return -ENOTCONN;
713 sin->sin_port = inet->inet_dport;
714 sin->sin_addr.s_addr = inet->inet_daddr;
715 } else {
716 __be32 addr = inet->inet_rcv_saddr;
717 if (!addr)
718 addr = inet->inet_saddr;
719 sin->sin_port = inet->inet_sport;
720 sin->sin_addr.s_addr = addr;
721 }
722 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
723 *uaddr_len = sizeof(*sin);
724 return 0;
725}
726EXPORT_SYMBOL(inet_getname);
727
728int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
729 size_t size)
730{
731 struct sock *sk = sock->sk;
732
733 sock_rps_record_flow(sk);
734
735 /* We may need to bind the socket. */
736 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
737 inet_autobind(sk))
738 return -EAGAIN;
739
740 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
741}
742EXPORT_SYMBOL(inet_sendmsg);
743
744ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
745 size_t size, int flags)
746{
747 struct sock *sk = sock->sk;
748
749 sock_rps_record_flow(sk);
750
751 /* We may need to bind the socket. */
752 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
753 inet_autobind(sk))
754 return -EAGAIN;
755
756 if (sk->sk_prot->sendpage)
757 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
758 return sock_no_sendpage(sock, page, offset, size, flags);
759}
760EXPORT_SYMBOL(inet_sendpage);
761
762int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
763 size_t size, int flags)
764{
765 struct sock *sk = sock->sk;
766 int addr_len = 0;
767 int err;
768
769 sock_rps_record_flow(sk);
770
771 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
772 flags & ~MSG_DONTWAIT, &addr_len);
773 if (err >= 0)
774 msg->msg_namelen = addr_len;
775 return err;
776}
777EXPORT_SYMBOL(inet_recvmsg);
778
779int inet_shutdown(struct socket *sock, int how)
780{
781 struct sock *sk = sock->sk;
782 int err = 0;
783
784 /* This should really check to make sure
785 * the socket is a TCP socket. (WHY AC...)
786 */
787 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
788 1->2 bit 2 snds.
789 2->3 */
790 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
791 return -EINVAL;
792
793 lock_sock(sk);
794 if (sock->state == SS_CONNECTING) {
795 if ((1 << sk->sk_state) &
796 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
797 sock->state = SS_DISCONNECTING;
798 else
799 sock->state = SS_CONNECTED;
800 }
801
802 switch (sk->sk_state) {
803 case TCP_CLOSE:
804 err = -ENOTCONN;
805 /* Hack to wake up other listeners, who can poll for
806 POLLHUP, even on eg. unconnected UDP sockets -- RR */
807 default:
808 sk->sk_shutdown |= how;
809 if (sk->sk_prot->shutdown)
810 sk->sk_prot->shutdown(sk, how);
811 break;
812
813 /* Remaining two branches are temporary solution for missing
814 * close() in multithreaded environment. It is _not_ a good idea,
815 * but we have no choice until close() is repaired at VFS level.
816 */
817 case TCP_LISTEN:
818 if (!(how & RCV_SHUTDOWN))
819 break;
820 /* Fall through */
821 case TCP_SYN_SENT:
822 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
823 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
824 break;
825 }
826
827 /* Wake up anyone sleeping in poll. */
828 sk->sk_state_change(sk);
829 release_sock(sk);
830 return err;
831}
832EXPORT_SYMBOL(inet_shutdown);
833
834/*
835 * ioctl() calls you can issue on an INET socket. Most of these are
836 * device configuration and stuff and very rarely used. Some ioctls
837 * pass on to the socket itself.
838 *
839 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
840 * loads the devconfigure module does its configuring and unloads it.
841 * There's a good 20K of config code hanging around the kernel.
842 */
843
844int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
845{
846 struct sock *sk = sock->sk;
847 int err = 0;
848 struct net *net = sock_net(sk);
849
850 switch (cmd) {
851 case SIOCGSTAMP:
852 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
853 break;
854 case SIOCGSTAMPNS:
855 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
856 break;
857 case SIOCADDRT:
858 case SIOCDELRT:
859 case SIOCRTMSG:
860 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
861 break;
862 case SIOCDARP:
863 case SIOCGARP:
864 case SIOCSARP:
865 err = arp_ioctl(net, cmd, (void __user *)arg);
866 break;
867 case SIOCGIFADDR:
868 case SIOCSIFADDR:
869 case SIOCGIFBRDADDR:
870 case SIOCSIFBRDADDR:
871 case SIOCGIFNETMASK:
872 case SIOCSIFNETMASK:
873 case SIOCGIFDSTADDR:
874 case SIOCSIFDSTADDR:
875 case SIOCSIFPFLAGS:
876 case SIOCGIFPFLAGS:
877 case SIOCSIFFLAGS:
878 err = devinet_ioctl(net, cmd, (void __user *)arg);
879 break;
880 default:
881 if (sk->sk_prot->ioctl)
882 err = sk->sk_prot->ioctl(sk, cmd, arg);
883 else
884 err = -ENOIOCTLCMD;
885 break;
886 }
887 return err;
888}
889EXPORT_SYMBOL(inet_ioctl);
890
891#ifdef CONFIG_COMPAT
892static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
893{
894 struct sock *sk = sock->sk;
895 int err = -ENOIOCTLCMD;
896
897 if (sk->sk_prot->compat_ioctl)
898 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
899
900 return err;
901}
902#endif
903
904const struct proto_ops inet_stream_ops = {
905 .family = PF_INET,
906 .owner = THIS_MODULE,
907 .release = inet_release,
908 .bind = inet_bind,
909 .connect = inet_stream_connect,
910 .socketpair = sock_no_socketpair,
911 .accept = inet_accept,
912 .getname = inet_getname,
913 .poll = tcp_poll,
914 .ioctl = inet_ioctl,
915 .listen = inet_listen,
916 .shutdown = inet_shutdown,
917 .setsockopt = sock_common_setsockopt,
918 .getsockopt = sock_common_getsockopt,
919 .sendmsg = inet_sendmsg,
920 .recvmsg = inet_recvmsg,
921 .mmap = sock_no_mmap,
922 .sendpage = inet_sendpage,
923 .splice_read = tcp_splice_read,
924#ifdef CONFIG_COMPAT
925 .compat_setsockopt = compat_sock_common_setsockopt,
926 .compat_getsockopt = compat_sock_common_getsockopt,
927 .compat_ioctl = inet_compat_ioctl,
928#endif
929};
930EXPORT_SYMBOL(inet_stream_ops);
931
932const struct proto_ops inet_dgram_ops = {
933 .family = PF_INET,
934 .owner = THIS_MODULE,
935 .release = inet_release,
936 .bind = inet_bind,
937 .connect = inet_dgram_connect,
938 .socketpair = sock_no_socketpair,
939 .accept = sock_no_accept,
940 .getname = inet_getname,
941 .poll = udp_poll,
942 .ioctl = inet_ioctl,
943 .listen = sock_no_listen,
944 .shutdown = inet_shutdown,
945 .setsockopt = sock_common_setsockopt,
946 .getsockopt = sock_common_getsockopt,
947 .sendmsg = inet_sendmsg,
948 .recvmsg = inet_recvmsg,
949 .mmap = sock_no_mmap,
950 .sendpage = inet_sendpage,
951#ifdef CONFIG_COMPAT
952 .compat_setsockopt = compat_sock_common_setsockopt,
953 .compat_getsockopt = compat_sock_common_getsockopt,
954 .compat_ioctl = inet_compat_ioctl,
955#endif
956};
957EXPORT_SYMBOL(inet_dgram_ops);
958
959/*
960 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
961 * udp_poll
962 */
963static const struct proto_ops inet_sockraw_ops = {
964 .family = PF_INET,
965 .owner = THIS_MODULE,
966 .release = inet_release,
967 .bind = inet_bind,
968 .connect = inet_dgram_connect,
969 .socketpair = sock_no_socketpair,
970 .accept = sock_no_accept,
971 .getname = inet_getname,
972 .poll = datagram_poll,
973 .ioctl = inet_ioctl,
974 .listen = sock_no_listen,
975 .shutdown = inet_shutdown,
976 .setsockopt = sock_common_setsockopt,
977 .getsockopt = sock_common_getsockopt,
978 .sendmsg = inet_sendmsg,
979 .recvmsg = inet_recvmsg,
980 .mmap = sock_no_mmap,
981 .sendpage = inet_sendpage,
982#ifdef CONFIG_COMPAT
983 .compat_setsockopt = compat_sock_common_setsockopt,
984 .compat_getsockopt = compat_sock_common_getsockopt,
985 .compat_ioctl = inet_compat_ioctl,
986#endif
987};
988
989static const struct net_proto_family inet_family_ops = {
990 .family = PF_INET,
991 .create = inet_create,
992 .owner = THIS_MODULE,
993};
994
995/* Upon startup we insert all the elements in inetsw_array[] into
996 * the linked list inetsw.
997 */
998static struct inet_protosw inetsw_array[] =
999{
1000 {
1001 .type = SOCK_STREAM,
1002 .protocol = IPPROTO_TCP,
1003 .prot = &tcp_prot,
1004 .ops = &inet_stream_ops,
1005 .no_check = 0,
1006 .flags = INET_PROTOSW_PERMANENT |
1007 INET_PROTOSW_ICSK,
1008 },
1009
1010 {
1011 .type = SOCK_DGRAM,
1012 .protocol = IPPROTO_UDP,
1013 .prot = &udp_prot,
1014 .ops = &inet_dgram_ops,
1015 .no_check = UDP_CSUM_DEFAULT,
1016 .flags = INET_PROTOSW_PERMANENT,
1017 },
1018
1019 {
1020 .type = SOCK_DGRAM,
1021 .protocol = IPPROTO_ICMP,
1022 .prot = &ping_prot,
1023 .ops = &inet_dgram_ops,
1024 .no_check = UDP_CSUM_DEFAULT,
1025 .flags = INET_PROTOSW_REUSE,
1026 },
1027
1028 {
1029 .type = SOCK_RAW,
1030 .protocol = IPPROTO_IP, /* wild card */
1031 .prot = &raw_prot,
1032 .ops = &inet_sockraw_ops,
1033 .no_check = UDP_CSUM_DEFAULT,
1034 .flags = INET_PROTOSW_REUSE,
1035 }
1036};
1037
1038#define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1039
1040void inet_register_protosw(struct inet_protosw *p)
1041{
1042 struct list_head *lh;
1043 struct inet_protosw *answer;
1044 int protocol = p->protocol;
1045 struct list_head *last_perm;
1046
1047 spin_lock_bh(&inetsw_lock);
1048
1049 if (p->type >= SOCK_MAX)
1050 goto out_illegal;
1051
1052 /* If we are trying to override a permanent protocol, bail. */
1053 answer = NULL;
1054 last_perm = &inetsw[p->type];
1055 list_for_each(lh, &inetsw[p->type]) {
1056 answer = list_entry(lh, struct inet_protosw, list);
1057
1058 /* Check only the non-wild match. */
1059 if (INET_PROTOSW_PERMANENT & answer->flags) {
1060 if (protocol == answer->protocol)
1061 break;
1062 last_perm = lh;
1063 }
1064
1065 answer = NULL;
1066 }
1067 if (answer)
1068 goto out_permanent;
1069
1070 /* Add the new entry after the last permanent entry if any, so that
1071 * the new entry does not override a permanent entry when matched with
1072 * a wild-card protocol. But it is allowed to override any existing
1073 * non-permanent entry. This means that when we remove this entry, the
1074 * system automatically returns to the old behavior.
1075 */
1076 list_add_rcu(&p->list, last_perm);
1077out:
1078 spin_unlock_bh(&inetsw_lock);
1079
1080 return;
1081
1082out_permanent:
1083 pr_err("Attempt to override permanent protocol %d\n", protocol);
1084 goto out;
1085
1086out_illegal:
1087 pr_err("Ignoring attempt to register invalid socket type %d\n",
1088 p->type);
1089 goto out;
1090}
1091EXPORT_SYMBOL(inet_register_protosw);
1092
1093void inet_unregister_protosw(struct inet_protosw *p)
1094{
1095 if (INET_PROTOSW_PERMANENT & p->flags) {
1096 pr_err("Attempt to unregister permanent protocol %d\n",
1097 p->protocol);
1098 } else {
1099 spin_lock_bh(&inetsw_lock);
1100 list_del_rcu(&p->list);
1101 spin_unlock_bh(&inetsw_lock);
1102
1103 synchronize_net();
1104 }
1105}
1106EXPORT_SYMBOL(inet_unregister_protosw);
1107
1108/*
1109 * Shall we try to damage output packets if routing dev changes?
1110 */
1111
1112int sysctl_ip_dynaddr __read_mostly;
1113
1114static int inet_sk_reselect_saddr(struct sock *sk)
1115{
1116 struct inet_sock *inet = inet_sk(sk);
1117 __be32 old_saddr = inet->inet_saddr;
1118 __be32 daddr = inet->inet_daddr;
1119 struct flowi4 *fl4;
1120 struct rtable *rt;
1121 __be32 new_saddr;
1122 struct ip_options_rcu *inet_opt;
1123
1124 inet_opt = rcu_dereference_protected(inet->inet_opt,
1125 sock_owned_by_user(sk));
1126 if (inet_opt && inet_opt->opt.srr)
1127 daddr = inet_opt->opt.faddr;
1128
1129 /* Query new route. */
1130 fl4 = &inet->cork.fl.u.ip4;
1131 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1132 sk->sk_bound_dev_if, sk->sk_protocol,
1133 inet->inet_sport, inet->inet_dport, sk);
1134 if (IS_ERR(rt))
1135 return PTR_ERR(rt);
1136
1137 sk_setup_caps(sk, &rt->dst);
1138
1139 new_saddr = fl4->saddr;
1140
1141 if (new_saddr == old_saddr)
1142 return 0;
1143
1144 if (sysctl_ip_dynaddr > 1) {
1145 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1146 __func__, &old_saddr, &new_saddr);
1147 }
1148
1149 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1150
1151 /*
1152 * XXX The only one ugly spot where we need to
1153 * XXX really change the sockets identity after
1154 * XXX it has entered the hashes. -DaveM
1155 *
1156 * Besides that, it does not check for connection
1157 * uniqueness. Wait for troubles.
1158 */
1159 __sk_prot_rehash(sk);
1160 return 0;
1161}
1162
1163int inet_sk_rebuild_header(struct sock *sk)
1164{
1165 struct inet_sock *inet = inet_sk(sk);
1166 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1167 __be32 daddr;
1168 struct ip_options_rcu *inet_opt;
1169 struct flowi4 *fl4;
1170 int err;
1171
1172 /* Route is OK, nothing to do. */
1173 if (rt)
1174 return 0;
1175
1176 /* Reroute. */
1177 rcu_read_lock();
1178 inet_opt = rcu_dereference(inet->inet_opt);
1179 daddr = inet->inet_daddr;
1180 if (inet_opt && inet_opt->opt.srr)
1181 daddr = inet_opt->opt.faddr;
1182 rcu_read_unlock();
1183 fl4 = &inet->cork.fl.u.ip4;
1184 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1185 inet->inet_dport, inet->inet_sport,
1186 sk->sk_protocol, RT_CONN_FLAGS(sk),
1187 sk->sk_bound_dev_if);
1188 if (!IS_ERR(rt)) {
1189 err = 0;
1190 sk_setup_caps(sk, &rt->dst);
1191 } else {
1192 err = PTR_ERR(rt);
1193
1194 /* Routing failed... */
1195 sk->sk_route_caps = 0;
1196 /*
1197 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1198 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1199 */
1200 if (!sysctl_ip_dynaddr ||
1201 sk->sk_state != TCP_SYN_SENT ||
1202 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1203 (err = inet_sk_reselect_saddr(sk)) != 0)
1204 sk->sk_err_soft = -err;
1205 }
1206
1207 return err;
1208}
1209EXPORT_SYMBOL(inet_sk_rebuild_header);
1210
1211static int inet_gso_send_check(struct sk_buff *skb)
1212{
1213 const struct net_offload *ops;
1214 const struct iphdr *iph;
1215 int proto;
1216 int ihl;
1217 int err = -EINVAL;
1218
1219 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1220 goto out;
1221
1222 iph = ip_hdr(skb);
1223 ihl = iph->ihl * 4;
1224 if (ihl < sizeof(*iph))
1225 goto out;
1226
1227 proto = iph->protocol;
1228
1229 /* Warning: after this point, iph might be no longer valid */
1230 if (unlikely(!pskb_may_pull(skb, ihl)))
1231 goto out;
1232 __skb_pull(skb, ihl);
1233
1234 skb_reset_transport_header(skb);
1235 err = -EPROTONOSUPPORT;
1236
1237 ops = rcu_dereference(inet_offloads[proto]);
1238 if (likely(ops && ops->callbacks.gso_send_check))
1239 err = ops->callbacks.gso_send_check(skb);
1240
1241out:
1242 return err;
1243}
1244
1245static struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1246 netdev_features_t features)
1247{
1248 struct sk_buff *segs = ERR_PTR(-EINVAL);
1249 const struct net_offload *ops;
1250 unsigned int offset = 0;
1251 bool udpfrag, encap;
1252 struct iphdr *iph;
1253 int proto;
1254 int nhoff;
1255 int ihl;
1256 int id;
1257
1258 if (unlikely(skb_shinfo(skb)->gso_type &
1259 ~(SKB_GSO_TCPV4 |
1260 SKB_GSO_UDP |
1261 SKB_GSO_DODGY |
1262 SKB_GSO_TCP_ECN |
1263 SKB_GSO_GRE |
1264 SKB_GSO_IPIP |
1265 SKB_GSO_SIT |
1266 SKB_GSO_TCPV6 |
1267 SKB_GSO_UDP_TUNNEL |
1268 SKB_GSO_MPLS |
1269 0)))
1270 goto out;
1271
1272 skb_reset_network_header(skb);
1273 nhoff = skb_network_header(skb) - skb_mac_header(skb);
1274 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1275 goto out;
1276
1277 iph = ip_hdr(skb);
1278 ihl = iph->ihl * 4;
1279 if (ihl < sizeof(*iph))
1280 goto out;
1281
1282 id = ntohs(iph->id);
1283 proto = iph->protocol;
1284
1285 /* Warning: after this point, iph might be no longer valid */
1286 if (unlikely(!pskb_may_pull(skb, ihl)))
1287 goto out;
1288 __skb_pull(skb, ihl);
1289
1290 encap = SKB_GSO_CB(skb)->encap_level > 0;
1291 if (encap)
1292 features = skb->dev->hw_enc_features & netif_skb_features(skb);
1293 SKB_GSO_CB(skb)->encap_level += ihl;
1294
1295 skb_reset_transport_header(skb);
1296
1297 segs = ERR_PTR(-EPROTONOSUPPORT);
1298
1299 if (skb->encapsulation &&
1300 skb_shinfo(skb)->gso_type & (SKB_GSO_SIT|SKB_GSO_IPIP))
1301 udpfrag = proto == IPPROTO_UDP && encap;
1302 else
1303 udpfrag = proto == IPPROTO_UDP && !skb->encapsulation;
1304
1305 ops = rcu_dereference(inet_offloads[proto]);
1306 if (likely(ops && ops->callbacks.gso_segment))
1307 segs = ops->callbacks.gso_segment(skb, features);
1308
1309 if (IS_ERR_OR_NULL(segs))
1310 goto out;
1311
1312 skb = segs;
1313 do {
1314 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1315 if (udpfrag) {
1316 iph->id = htons(id);
1317 iph->frag_off = htons(offset >> 3);
1318 if (skb->next != NULL)
1319 iph->frag_off |= htons(IP_MF);
1320 offset += skb->len - nhoff - ihl;
1321 } else {
1322 iph->id = htons(id++);
1323 }
1324 iph->tot_len = htons(skb->len - nhoff);
1325 ip_send_check(iph);
1326 if (encap)
1327 skb_reset_inner_headers(skb);
1328 skb->network_header = (u8 *)iph - skb->head;
1329 } while ((skb = skb->next));
1330
1331out:
1332 return segs;
1333}
1334
1335static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1336 struct sk_buff *skb)
1337{
1338 const struct net_offload *ops;
1339 struct sk_buff **pp = NULL;
1340 struct sk_buff *p;
1341 const struct iphdr *iph;
1342 unsigned int hlen;
1343 unsigned int off;
1344 unsigned int id;
1345 int flush = 1;
1346 int proto;
1347
1348 off = skb_gro_offset(skb);
1349 hlen = off + sizeof(*iph);
1350 iph = skb_gro_header_fast(skb, off);
1351 if (skb_gro_header_hard(skb, hlen)) {
1352 iph = skb_gro_header_slow(skb, hlen, off);
1353 if (unlikely(!iph))
1354 goto out;
1355 }
1356
1357 proto = iph->protocol;
1358
1359 rcu_read_lock();
1360 ops = rcu_dereference(inet_offloads[proto]);
1361 if (!ops || !ops->callbacks.gro_receive)
1362 goto out_unlock;
1363
1364 if (*(u8 *)iph != 0x45)
1365 goto out_unlock;
1366
1367 if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1368 goto out_unlock;
1369
1370 id = ntohl(*(__be32 *)&iph->id);
1371 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1372 id >>= 16;
1373
1374 for (p = *head; p; p = p->next) {
1375 struct iphdr *iph2;
1376
1377 if (!NAPI_GRO_CB(p)->same_flow)
1378 continue;
1379
1380 iph2 = (struct iphdr *)(p->data + off);
1381 /* The above works because, with the exception of the top
1382 * (inner most) layer, we only aggregate pkts with the same
1383 * hdr length so all the hdrs we'll need to verify will start
1384 * at the same offset.
1385 */
1386 if ((iph->protocol ^ iph2->protocol) |
1387 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1388 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1389 NAPI_GRO_CB(p)->same_flow = 0;
1390 continue;
1391 }
1392
1393 /* All fields must match except length and checksum. */
1394 NAPI_GRO_CB(p)->flush |=
1395 (iph->ttl ^ iph2->ttl) |
1396 (iph->tos ^ iph2->tos) |
1397 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1398
1399 /* Save the IP ID check to be included later when we get to
1400 * the transport layer so only the inner most IP ID is checked.
1401 * This is because some GSO/TSO implementations do not
1402 * correctly increment the IP ID for the outer hdrs.
1403 */
1404 NAPI_GRO_CB(p)->flush_id =
1405 ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1406 NAPI_GRO_CB(p)->flush |= flush;
1407 }
1408
1409 NAPI_GRO_CB(skb)->flush |= flush;
1410 skb_set_network_header(skb, off);
1411 /* The above will be needed by the transport layer if there is one
1412 * immediately following this IP hdr.
1413 */
1414
1415 skb_gro_pull(skb, sizeof(*iph));
1416 skb_set_transport_header(skb, skb_gro_offset(skb));
1417
1418 pp = ops->callbacks.gro_receive(head, skb);
1419
1420out_unlock:
1421 rcu_read_unlock();
1422
1423out:
1424 NAPI_GRO_CB(skb)->flush |= flush;
1425
1426 return pp;
1427}
1428
1429static int inet_gro_complete(struct sk_buff *skb, int nhoff)
1430{
1431 __be16 newlen = htons(skb->len - nhoff);
1432 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1433 const struct net_offload *ops;
1434 int proto = iph->protocol;
1435 int err = -ENOSYS;
1436
1437 csum_replace2(&iph->check, iph->tot_len, newlen);
1438 iph->tot_len = newlen;
1439
1440 rcu_read_lock();
1441 ops = rcu_dereference(inet_offloads[proto]);
1442 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1443 goto out_unlock;
1444
1445 /* Only need to add sizeof(*iph) to get to the next hdr below
1446 * because any hdr with option will have been flushed in
1447 * inet_gro_receive().
1448 */
1449 err = ops->callbacks.gro_complete(skb, nhoff + sizeof(*iph));
1450
1451out_unlock:
1452 rcu_read_unlock();
1453
1454 return err;
1455}
1456
1457int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1458 unsigned short type, unsigned char protocol,
1459 struct net *net)
1460{
1461 struct socket *sock;
1462 int rc = sock_create_kern(family, type, protocol, &sock);
1463
1464 if (rc == 0) {
1465 *sk = sock->sk;
1466 (*sk)->sk_allocation = GFP_ATOMIC;
1467 /*
1468 * Unhash it so that IP input processing does not even see it,
1469 * we do not wish this socket to see incoming packets.
1470 */
1471 (*sk)->sk_prot->unhash(*sk);
1472
1473 sk_change_net(*sk, net);
1474 }
1475 return rc;
1476}
1477EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1478
1479unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1480{
1481 unsigned long res = 0;
1482 int i, j;
1483
1484 for_each_possible_cpu(i) {
1485 for (j = 0; j < SNMP_ARRAY_SZ; j++)
1486 res += *(((unsigned long *) per_cpu_ptr(mib[j], i)) + offt);
1487 }
1488 return res;
1489}
1490EXPORT_SYMBOL_GPL(snmp_fold_field);
1491
1492#if BITS_PER_LONG==32
1493
1494u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset)
1495{
1496 u64 res = 0;
1497 int cpu;
1498
1499 for_each_possible_cpu(cpu) {
1500 void *bhptr;
1501 struct u64_stats_sync *syncp;
1502 u64 v;
1503 unsigned int start;
1504
1505 bhptr = per_cpu_ptr(mib[0], cpu);
1506 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1507 do {
1508 start = u64_stats_fetch_begin_irq(syncp);
1509 v = *(((u64 *) bhptr) + offt);
1510 } while (u64_stats_fetch_retry_irq(syncp, start));
1511
1512 res += v;
1513 }
1514 return res;
1515}
1516EXPORT_SYMBOL_GPL(snmp_fold_field64);
1517#endif
1518
1519int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align)
1520{
1521 BUG_ON(ptr == NULL);
1522 ptr[0] = __alloc_percpu(mibsize, align);
1523 if (!ptr[0])
1524 return -ENOMEM;
1525
1526#if SNMP_ARRAY_SZ == 2
1527 ptr[1] = __alloc_percpu(mibsize, align);
1528 if (!ptr[1]) {
1529 free_percpu(ptr[0]);
1530 ptr[0] = NULL;
1531 return -ENOMEM;
1532 }
1533#endif
1534 return 0;
1535}
1536EXPORT_SYMBOL_GPL(snmp_mib_init);
1537
1538#ifdef CONFIG_IP_MULTICAST
1539static const struct net_protocol igmp_protocol = {
1540 .handler = igmp_rcv,
1541 .netns_ok = 1,
1542};
1543#endif
1544
1545static const struct net_protocol tcp_protocol = {
1546 .early_demux = tcp_v4_early_demux,
1547 .handler = tcp_v4_rcv,
1548 .err_handler = tcp_v4_err,
1549 .no_policy = 1,
1550 .netns_ok = 1,
1551 .icmp_strict_tag_validation = 1,
1552};
1553
1554static const struct net_protocol udp_protocol = {
1555 .early_demux = udp_v4_early_demux,
1556 .handler = udp_rcv,
1557 .err_handler = udp_err,
1558 .no_policy = 1,
1559 .netns_ok = 1,
1560};
1561
1562static const struct net_protocol icmp_protocol = {
1563 .handler = icmp_rcv,
1564 .err_handler = icmp_err,
1565 .no_policy = 1,
1566 .netns_ok = 1,
1567};
1568
1569static __net_init int ipv4_mib_init_net(struct net *net)
1570{
1571 int i;
1572
1573 if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1574 sizeof(struct tcp_mib),
1575 __alignof__(struct tcp_mib)) < 0)
1576 goto err_tcp_mib;
1577 if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1578 sizeof(struct ipstats_mib),
1579 __alignof__(struct ipstats_mib)) < 0)
1580 goto err_ip_mib;
1581
1582 for_each_possible_cpu(i) {
1583 struct ipstats_mib *af_inet_stats;
1584 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics[0], i);
1585 u64_stats_init(&af_inet_stats->syncp);
1586#if SNMP_ARRAY_SZ == 2
1587 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics[1], i);
1588 u64_stats_init(&af_inet_stats->syncp);
1589#endif
1590 }
1591
1592 if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1593 sizeof(struct linux_mib),
1594 __alignof__(struct linux_mib)) < 0)
1595 goto err_net_mib;
1596 if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1597 sizeof(struct udp_mib),
1598 __alignof__(struct udp_mib)) < 0)
1599 goto err_udp_mib;
1600 if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1601 sizeof(struct udp_mib),
1602 __alignof__(struct udp_mib)) < 0)
1603 goto err_udplite_mib;
1604 if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1605 sizeof(struct icmp_mib),
1606 __alignof__(struct icmp_mib)) < 0)
1607 goto err_icmp_mib;
1608 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1609 GFP_KERNEL);
1610 if (!net->mib.icmpmsg_statistics)
1611 goto err_icmpmsg_mib;
1612
1613 tcp_mib_init(net);
1614 return 0;
1615
1616err_icmpmsg_mib:
1617 snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1618err_icmp_mib:
1619 snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1620err_udplite_mib:
1621 snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1622err_udp_mib:
1623 snmp_mib_free((void __percpu **)net->mib.net_statistics);
1624err_net_mib:
1625 snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1626err_ip_mib:
1627 snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1628err_tcp_mib:
1629 return -ENOMEM;
1630}
1631
1632static __net_exit void ipv4_mib_exit_net(struct net *net)
1633{
1634 kfree(net->mib.icmpmsg_statistics);
1635 snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1636 snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1637 snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1638 snmp_mib_free((void __percpu **)net->mib.net_statistics);
1639 snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1640 snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1641}
1642
1643static __net_initdata struct pernet_operations ipv4_mib_ops = {
1644 .init = ipv4_mib_init_net,
1645 .exit = ipv4_mib_exit_net,
1646};
1647
1648static int __init init_ipv4_mibs(void)
1649{
1650 return register_pernet_subsys(&ipv4_mib_ops);
1651}
1652
1653static __net_init int inet_init_net(struct net *net)
1654{
1655 /*
1656 * Set defaults for local port range
1657 */
1658 seqlock_init(&net->ipv4.ip_local_ports.lock);
1659 net->ipv4.ip_local_ports.range[0] = 32768;
1660 net->ipv4.ip_local_ports.range[1] = 61000;
1661
1662 seqlock_init(&net->ipv4.ping_group_range.lock);
1663 /*
1664 * Sane defaults - nobody may create ping sockets.
1665 * Boot scripts should set this to distro-specific group.
1666 */
1667 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1668 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1669 return 0;
1670}
1671
1672static __net_exit void inet_exit_net(struct net *net)
1673{
1674}
1675
1676static __net_initdata struct pernet_operations af_inet_ops = {
1677 .init = inet_init_net,
1678 .exit = inet_exit_net,
1679};
1680
1681static int __init init_inet_pernet_ops(void)
1682{
1683 return register_pernet_subsys(&af_inet_ops);
1684}
1685
1686static int ipv4_proc_init(void);
1687
1688/*
1689 * IP protocol layer initialiser
1690 */
1691
1692static struct packet_offload ip_packet_offload __read_mostly = {
1693 .type = cpu_to_be16(ETH_P_IP),
1694 .callbacks = {
1695 .gso_send_check = inet_gso_send_check,
1696 .gso_segment = inet_gso_segment,
1697 .gro_receive = inet_gro_receive,
1698 .gro_complete = inet_gro_complete,
1699 },
1700};
1701
1702static const struct net_offload ipip_offload = {
1703 .callbacks = {
1704 .gso_send_check = inet_gso_send_check,
1705 .gso_segment = inet_gso_segment,
1706 },
1707};
1708
1709static int __init ipv4_offload_init(void)
1710{
1711 /*
1712 * Add offloads
1713 */
1714 if (udpv4_offload_init() < 0)
1715 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1716 if (tcpv4_offload_init() < 0)
1717 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1718
1719 dev_add_offload(&ip_packet_offload);
1720 inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1721 return 0;
1722}
1723
1724fs_initcall(ipv4_offload_init);
1725
1726static struct packet_type ip_packet_type __read_mostly = {
1727 .type = cpu_to_be16(ETH_P_IP),
1728 .func = ip_rcv,
1729};
1730
1731static int __init inet_init(void)
1732{
1733 struct inet_protosw *q;
1734 struct list_head *r;
1735 int rc = -EINVAL;
1736
1737 BUILD_BUG_ON(sizeof(struct inet_skb_parm) > FIELD_SIZEOF(struct sk_buff, cb));
1738
1739 sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL);
1740 if (!sysctl_local_reserved_ports)
1741 goto out;
1742
1743 rc = proto_register(&tcp_prot, 1);
1744 if (rc)
1745 goto out_free_reserved_ports;
1746
1747 rc = proto_register(&udp_prot, 1);
1748 if (rc)
1749 goto out_unregister_tcp_proto;
1750
1751 rc = proto_register(&raw_prot, 1);
1752 if (rc)
1753 goto out_unregister_udp_proto;
1754
1755 rc = proto_register(&ping_prot, 1);
1756 if (rc)
1757 goto out_unregister_raw_proto;
1758
1759 /*
1760 * Tell SOCKET that we are alive...
1761 */
1762
1763 (void)sock_register(&inet_family_ops);
1764
1765#ifdef CONFIG_SYSCTL
1766 ip_static_sysctl_init();
1767#endif
1768
1769 /*
1770 * Add all the base protocols.
1771 */
1772
1773 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1774 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1775 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1776 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1777 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1778 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1779#ifdef CONFIG_IP_MULTICAST
1780 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1781 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1782#endif
1783
1784 /* Register the socket-side information for inet_create. */
1785 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1786 INIT_LIST_HEAD(r);
1787
1788 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1789 inet_register_protosw(q);
1790
1791 /*
1792 * Set the ARP module up
1793 */
1794
1795 arp_init();
1796
1797 /*
1798 * Set the IP module up
1799 */
1800
1801 ip_init();
1802
1803 tcp_v4_init();
1804
1805 /* Setup TCP slab cache for open requests. */
1806 tcp_init();
1807
1808 /* Setup UDP memory threshold */
1809 udp_init();
1810
1811 /* Add UDP-Lite (RFC 3828) */
1812 udplite4_register();
1813
1814 ping_init();
1815
1816 /*
1817 * Set the ICMP layer up
1818 */
1819
1820 if (icmp_init() < 0)
1821 panic("Failed to create the ICMP control socket.\n");
1822
1823 /*
1824 * Initialise the multicast router
1825 */
1826#if defined(CONFIG_IP_MROUTE)
1827 if (ip_mr_init())
1828 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1829#endif
1830
1831 if (init_inet_pernet_ops())
1832 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
1833 /*
1834 * Initialise per-cpu ipv4 mibs
1835 */
1836
1837 if (init_ipv4_mibs())
1838 pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1839
1840 ipv4_proc_init();
1841
1842 ipfrag_init();
1843
1844 dev_add_pack(&ip_packet_type);
1845
1846 rc = 0;
1847out:
1848 return rc;
1849out_unregister_raw_proto:
1850 proto_unregister(&raw_prot);
1851out_unregister_udp_proto:
1852 proto_unregister(&udp_prot);
1853out_unregister_tcp_proto:
1854 proto_unregister(&tcp_prot);
1855out_free_reserved_ports:
1856 kfree(sysctl_local_reserved_ports);
1857 goto out;
1858}
1859
1860fs_initcall(inet_init);
1861
1862/* ------------------------------------------------------------------------ */
1863
1864#ifdef CONFIG_PROC_FS
1865static int __init ipv4_proc_init(void)
1866{
1867 int rc = 0;
1868
1869 if (raw_proc_init())
1870 goto out_raw;
1871 if (tcp4_proc_init())
1872 goto out_tcp;
1873 if (udp4_proc_init())
1874 goto out_udp;
1875 if (ping_proc_init())
1876 goto out_ping;
1877 if (ip_misc_proc_init())
1878 goto out_misc;
1879out:
1880 return rc;
1881out_misc:
1882 ping_proc_exit();
1883out_ping:
1884 udp4_proc_exit();
1885out_udp:
1886 tcp4_proc_exit();
1887out_tcp:
1888 raw_proc_exit();
1889out_raw:
1890 rc = -ENOMEM;
1891 goto out;
1892}
1893
1894#else /* CONFIG_PROC_FS */
1895static int __init ipv4_proc_init(void)
1896{
1897 return 0;
1898}
1899#endif /* CONFIG_PROC_FS */
1900
1901MODULE_ALIAS_NETPROTO(PF_INET);
1902
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 */
127static struct list_head inetsw[SOCK_MAX];
128static DEFINE_SPINLOCK(inetsw_lock);
129
130/* New destruction routine */
131
132void 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}
161EXPORT_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
173static 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 */
193int 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
235out:
236 release_sock(sk);
237 return err;
238}
239EXPORT_SYMBOL(inet_listen);
240
241/*
242 * Create an inet socket.
243 */
244
245static 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. */
262lookup_protocol:
263 err = -ESOCKTNOSUPPORT;
264 rcu_read_lock();
265 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
266
267 err = 0;
268 /* Check the non-wild match. */
269 if (protocol == answer->protocol) {
270 if (protocol != IPPROTO_IP)
271 break;
272 } else {
273 /* Check for the two wild cases. */
274 if (IPPROTO_IP == protocol) {
275 protocol = answer->protocol;
276 break;
277 }
278 if (IPPROTO_IP == answer->protocol)
279 break;
280 }
281 err = -EPROTONOSUPPORT;
282 }
283
284 if (unlikely(err)) {
285 if (try_loading_module < 2) {
286 rcu_read_unlock();
287 /*
288 * Be more specific, e.g. net-pf-2-proto-132-type-1
289 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
290 */
291 if (++try_loading_module == 1)
292 request_module("net-pf-%d-proto-%d-type-%d",
293 PF_INET, protocol, sock->type);
294 /*
295 * Fall back to generic, e.g. net-pf-2-proto-132
296 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
297 */
298 else
299 request_module("net-pf-%d-proto-%d",
300 PF_INET, protocol);
301 goto lookup_protocol;
302 } else
303 goto out_rcu_unlock;
304 }
305
306 err = -EPERM;
307 if (sock->type == SOCK_RAW && !kern &&
308 !ns_capable(net->user_ns, CAP_NET_RAW))
309 goto out_rcu_unlock;
310
311 sock->ops = answer->ops;
312 answer_prot = answer->prot;
313 answer_flags = answer->flags;
314 rcu_read_unlock();
315
316 WARN_ON(!answer_prot->slab);
317
318 err = -ENOMEM;
319 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
320 if (!sk)
321 goto out;
322
323 err = 0;
324 if (INET_PROTOSW_REUSE & answer_flags)
325 sk->sk_reuse = SK_CAN_REUSE;
326
327 inet = inet_sk(sk);
328 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
329
330 inet->nodefrag = 0;
331
332 if (SOCK_RAW == sock->type) {
333 inet->inet_num = protocol;
334 if (IPPROTO_RAW == protocol)
335 inet->hdrincl = 1;
336 }
337
338 if (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc))
339 inet->pmtudisc = IP_PMTUDISC_DONT;
340 else
341 inet->pmtudisc = IP_PMTUDISC_WANT;
342
343 inet->inet_id = 0;
344
345 sock_init_data(sock, sk);
346
347 sk->sk_destruct = inet_sock_destruct;
348 sk->sk_protocol = protocol;
349 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
350 sk->sk_txrehash = READ_ONCE(net->core.sysctl_txrehash);
351
352 inet->uc_ttl = -1;
353 inet->mc_loop = 1;
354 inet->mc_ttl = 1;
355 inet->mc_all = 1;
356 inet->mc_index = 0;
357 inet->mc_list = NULL;
358 inet->rcv_tos = 0;
359
360 sk_refcnt_debug_inc(sk);
361
362 if (inet->inet_num) {
363 /* It assumes that any protocol which allows
364 * the user to assign a number at socket
365 * creation time automatically
366 * shares.
367 */
368 inet->inet_sport = htons(inet->inet_num);
369 /* Add to protocol hash chains. */
370 err = sk->sk_prot->hash(sk);
371 if (err) {
372 sk_common_release(sk);
373 goto out;
374 }
375 }
376
377 if (sk->sk_prot->init) {
378 err = sk->sk_prot->init(sk);
379 if (err) {
380 sk_common_release(sk);
381 goto out;
382 }
383 }
384
385 if (!kern) {
386 err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
387 if (err) {
388 sk_common_release(sk);
389 goto out;
390 }
391 }
392out:
393 return err;
394out_rcu_unlock:
395 rcu_read_unlock();
396 goto out;
397}
398
399
400/*
401 * The peer socket should always be NULL (or else). When we call this
402 * function we are destroying the object and from then on nobody
403 * should refer to it.
404 */
405int inet_release(struct socket *sock)
406{
407 struct sock *sk = sock->sk;
408
409 if (sk) {
410 long timeout;
411
412 if (!sk->sk_kern_sock)
413 BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk);
414
415 /* Applications forget to leave groups before exiting */
416 ip_mc_drop_socket(sk);
417
418 /* If linger is set, we don't return until the close
419 * is complete. Otherwise we return immediately. The
420 * actually closing is done the same either way.
421 *
422 * If the close is due to the process exiting, we never
423 * linger..
424 */
425 timeout = 0;
426 if (sock_flag(sk, SOCK_LINGER) &&
427 !(current->flags & PF_EXITING))
428 timeout = sk->sk_lingertime;
429 sk->sk_prot->close(sk, timeout);
430 sock->sk = NULL;
431 }
432 return 0;
433}
434EXPORT_SYMBOL(inet_release);
435
436int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
437{
438 struct sock *sk = sock->sk;
439 u32 flags = BIND_WITH_LOCK;
440 int err;
441
442 /* If the socket has its own bind function then use it. (RAW) */
443 if (sk->sk_prot->bind) {
444 return sk->sk_prot->bind(sk, uaddr, addr_len);
445 }
446 if (addr_len < sizeof(struct sockaddr_in))
447 return -EINVAL;
448
449 /* BPF prog is run before any checks are done so that if the prog
450 * changes context in a wrong way it will be caught.
451 */
452 err = BPF_CGROUP_RUN_PROG_INET_BIND_LOCK(sk, uaddr,
453 CGROUP_INET4_BIND, &flags);
454 if (err)
455 return err;
456
457 return __inet_bind(sk, uaddr, addr_len, flags);
458}
459EXPORT_SYMBOL(inet_bind);
460
461int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
462 u32 flags)
463{
464 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
465 struct inet_sock *inet = inet_sk(sk);
466 struct net *net = sock_net(sk);
467 unsigned short snum;
468 int chk_addr_ret;
469 u32 tb_id = RT_TABLE_LOCAL;
470 int err;
471
472 if (addr->sin_family != AF_INET) {
473 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
474 * only if s_addr is INADDR_ANY.
475 */
476 err = -EAFNOSUPPORT;
477 if (addr->sin_family != AF_UNSPEC ||
478 addr->sin_addr.s_addr != htonl(INADDR_ANY))
479 goto out;
480 }
481
482 tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
483 chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
484
485 /* Not specified by any standard per-se, however it breaks too
486 * many applications when removed. It is unfortunate since
487 * allowing applications to make a non-local bind solves
488 * several problems with systems using dynamic addressing.
489 * (ie. your servers still start up even if your ISDN link
490 * is temporarily down)
491 */
492 err = -EADDRNOTAVAIL;
493 if (!inet_addr_valid_or_nonlocal(net, inet, addr->sin_addr.s_addr,
494 chk_addr_ret))
495 goto out;
496
497 snum = ntohs(addr->sin_port);
498 err = -EACCES;
499 if (!(flags & BIND_NO_CAP_NET_BIND_SERVICE) &&
500 snum && inet_port_requires_bind_service(net, snum) &&
501 !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
502 goto out;
503
504 /* We keep a pair of addresses. rcv_saddr is the one
505 * used by hash lookups, and saddr is used for transmit.
506 *
507 * In the BSD API these are the same except where it
508 * would be illegal to use them (multicast/broadcast) in
509 * which case the sending device address is used.
510 */
511 if (flags & BIND_WITH_LOCK)
512 lock_sock(sk);
513
514 /* Check these errors (active socket, double bind). */
515 err = -EINVAL;
516 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
517 goto out_release_sock;
518
519 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
520 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
521 inet->inet_saddr = 0; /* Use device */
522
523 /* Make sure we are allowed to bind here. */
524 if (snum || !(inet->bind_address_no_port ||
525 (flags & BIND_FORCE_ADDRESS_NO_PORT))) {
526 err = sk->sk_prot->get_port(sk, snum);
527 if (err) {
528 inet->inet_saddr = inet->inet_rcv_saddr = 0;
529 goto out_release_sock;
530 }
531 if (!(flags & BIND_FROM_BPF)) {
532 err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
533 if (err) {
534 inet->inet_saddr = inet->inet_rcv_saddr = 0;
535 if (sk->sk_prot->put_port)
536 sk->sk_prot->put_port(sk);
537 goto out_release_sock;
538 }
539 }
540 }
541
542 if (inet->inet_rcv_saddr)
543 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
544 if (snum)
545 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
546 inet->inet_sport = htons(inet->inet_num);
547 inet->inet_daddr = 0;
548 inet->inet_dport = 0;
549 sk_dst_reset(sk);
550 err = 0;
551out_release_sock:
552 if (flags & BIND_WITH_LOCK)
553 release_sock(sk);
554out:
555 return err;
556}
557
558int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
559 int addr_len, int flags)
560{
561 struct sock *sk = sock->sk;
562 const struct proto *prot;
563 int err;
564
565 if (addr_len < sizeof(uaddr->sa_family))
566 return -EINVAL;
567
568 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
569 prot = READ_ONCE(sk->sk_prot);
570
571 if (uaddr->sa_family == AF_UNSPEC)
572 return prot->disconnect(sk, flags);
573
574 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
575 err = prot->pre_connect(sk, uaddr, addr_len);
576 if (err)
577 return err;
578 }
579
580 if (data_race(!inet_sk(sk)->inet_num) && inet_autobind(sk))
581 return -EAGAIN;
582 return prot->connect(sk, uaddr, addr_len);
583}
584EXPORT_SYMBOL(inet_dgram_connect);
585
586static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
587{
588 DEFINE_WAIT_FUNC(wait, woken_wake_function);
589
590 add_wait_queue(sk_sleep(sk), &wait);
591 sk->sk_write_pending += writebias;
592
593 /* Basic assumption: if someone sets sk->sk_err, he _must_
594 * change state of the socket from TCP_SYN_*.
595 * Connect() does not allow to get error notifications
596 * without closing the socket.
597 */
598 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
599 release_sock(sk);
600 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
601 lock_sock(sk);
602 if (signal_pending(current) || !timeo)
603 break;
604 }
605 remove_wait_queue(sk_sleep(sk), &wait);
606 sk->sk_write_pending -= writebias;
607 return timeo;
608}
609
610/*
611 * Connect to a remote host. There is regrettably still a little
612 * TCP 'magic' in here.
613 */
614int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
615 int addr_len, int flags, int is_sendmsg)
616{
617 struct sock *sk = sock->sk;
618 int err;
619 long timeo;
620
621 /*
622 * uaddr can be NULL and addr_len can be 0 if:
623 * sk is a TCP fastopen active socket and
624 * TCP_FASTOPEN_CONNECT sockopt is set and
625 * we already have a valid cookie for this socket.
626 * In this case, user can call write() after connect().
627 * write() will invoke tcp_sendmsg_fastopen() which calls
628 * __inet_stream_connect().
629 */
630 if (uaddr) {
631 if (addr_len < sizeof(uaddr->sa_family))
632 return -EINVAL;
633
634 if (uaddr->sa_family == AF_UNSPEC) {
635 err = sk->sk_prot->disconnect(sk, flags);
636 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
637 goto out;
638 }
639 }
640
641 switch (sock->state) {
642 default:
643 err = -EINVAL;
644 goto out;
645 case SS_CONNECTED:
646 err = -EISCONN;
647 goto out;
648 case SS_CONNECTING:
649 if (inet_sk(sk)->defer_connect)
650 err = is_sendmsg ? -EINPROGRESS : -EISCONN;
651 else
652 err = -EALREADY;
653 /* Fall out of switch with err, set for this state */
654 break;
655 case SS_UNCONNECTED:
656 err = -EISCONN;
657 if (sk->sk_state != TCP_CLOSE)
658 goto out;
659
660 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
661 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
662 if (err)
663 goto out;
664 }
665
666 err = sk->sk_prot->connect(sk, uaddr, addr_len);
667 if (err < 0)
668 goto out;
669
670 sock->state = SS_CONNECTING;
671
672 if (!err && inet_sk(sk)->defer_connect)
673 goto out;
674
675 /* Just entered SS_CONNECTING state; the only
676 * difference is that return value in non-blocking
677 * case is EINPROGRESS, rather than EALREADY.
678 */
679 err = -EINPROGRESS;
680 break;
681 }
682
683 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
684
685 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
686 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
687 tcp_sk(sk)->fastopen_req &&
688 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
689
690 /* Error code is set above */
691 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
692 goto out;
693
694 err = sock_intr_errno(timeo);
695 if (signal_pending(current))
696 goto out;
697 }
698
699 /* Connection was closed by RST, timeout, ICMP error
700 * or another process disconnected us.
701 */
702 if (sk->sk_state == TCP_CLOSE)
703 goto sock_error;
704
705 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
706 * and error was received after socket entered established state.
707 * Hence, it is handled normally after connect() return successfully.
708 */
709
710 sock->state = SS_CONNECTED;
711 err = 0;
712out:
713 return err;
714
715sock_error:
716 err = sock_error(sk) ? : -ECONNABORTED;
717 sock->state = SS_UNCONNECTED;
718 if (sk->sk_prot->disconnect(sk, flags))
719 sock->state = SS_DISCONNECTING;
720 goto out;
721}
722EXPORT_SYMBOL(__inet_stream_connect);
723
724int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
725 int addr_len, int flags)
726{
727 int err;
728
729 lock_sock(sock->sk);
730 err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
731 release_sock(sock->sk);
732 return err;
733}
734EXPORT_SYMBOL(inet_stream_connect);
735
736/*
737 * Accept a pending connection. The TCP layer now gives BSD semantics.
738 */
739
740int inet_accept(struct socket *sock, struct socket *newsock, int flags,
741 bool kern)
742{
743 struct sock *sk1 = sock->sk, *sk2;
744 int err = -EINVAL;
745
746 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
747 sk2 = READ_ONCE(sk1->sk_prot)->accept(sk1, flags, &err, kern);
748 if (!sk2)
749 goto do_err;
750
751 lock_sock(sk2);
752
753 sock_rps_record_flow(sk2);
754 WARN_ON(!((1 << sk2->sk_state) &
755 (TCPF_ESTABLISHED | TCPF_SYN_RECV |
756 TCPF_CLOSE_WAIT | TCPF_CLOSE)));
757
758 if (test_bit(SOCK_SUPPORT_ZC, &sock->flags))
759 set_bit(SOCK_SUPPORT_ZC, &newsock->flags);
760 sock_graft(sk2, newsock);
761
762 newsock->state = SS_CONNECTED;
763 err = 0;
764 release_sock(sk2);
765do_err:
766 return err;
767}
768EXPORT_SYMBOL(inet_accept);
769
770/*
771 * This does both peername and sockname.
772 */
773int inet_getname(struct socket *sock, struct sockaddr *uaddr,
774 int peer)
775{
776 struct sock *sk = sock->sk;
777 struct inet_sock *inet = inet_sk(sk);
778 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
779
780 sin->sin_family = AF_INET;
781 lock_sock(sk);
782 if (peer) {
783 if (!inet->inet_dport ||
784 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
785 peer == 1)) {
786 release_sock(sk);
787 return -ENOTCONN;
788 }
789 sin->sin_port = inet->inet_dport;
790 sin->sin_addr.s_addr = inet->inet_daddr;
791 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin,
792 CGROUP_INET4_GETPEERNAME);
793 } else {
794 __be32 addr = inet->inet_rcv_saddr;
795 if (!addr)
796 addr = inet->inet_saddr;
797 sin->sin_port = inet->inet_sport;
798 sin->sin_addr.s_addr = addr;
799 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin,
800 CGROUP_INET4_GETSOCKNAME);
801 }
802 release_sock(sk);
803 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
804 return sizeof(*sin);
805}
806EXPORT_SYMBOL(inet_getname);
807
808int inet_send_prepare(struct sock *sk)
809{
810 sock_rps_record_flow(sk);
811
812 /* We may need to bind the socket. */
813 if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind &&
814 inet_autobind(sk))
815 return -EAGAIN;
816
817 return 0;
818}
819EXPORT_SYMBOL_GPL(inet_send_prepare);
820
821int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
822{
823 struct sock *sk = sock->sk;
824
825 if (unlikely(inet_send_prepare(sk)))
826 return -EAGAIN;
827
828 return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
829 sk, msg, size);
830}
831EXPORT_SYMBOL(inet_sendmsg);
832
833ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
834 size_t size, int flags)
835{
836 struct sock *sk = sock->sk;
837 const struct proto *prot;
838
839 if (unlikely(inet_send_prepare(sk)))
840 return -EAGAIN;
841
842 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
843 prot = READ_ONCE(sk->sk_prot);
844 if (prot->sendpage)
845 return prot->sendpage(sk, page, offset, size, flags);
846 return sock_no_sendpage(sock, page, offset, size, flags);
847}
848EXPORT_SYMBOL(inet_sendpage);
849
850INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
851 size_t, int, int *));
852int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
853 int flags)
854{
855 struct sock *sk = sock->sk;
856 int addr_len = 0;
857 int err;
858
859 if (likely(!(flags & MSG_ERRQUEUE)))
860 sock_rps_record_flow(sk);
861
862 err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
863 sk, msg, size, flags, &addr_len);
864 if (err >= 0)
865 msg->msg_namelen = addr_len;
866 return err;
867}
868EXPORT_SYMBOL(inet_recvmsg);
869
870int inet_shutdown(struct socket *sock, int how)
871{
872 struct sock *sk = sock->sk;
873 int err = 0;
874
875 /* This should really check to make sure
876 * the socket is a TCP socket. (WHY AC...)
877 */
878 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
879 1->2 bit 2 snds.
880 2->3 */
881 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
882 return -EINVAL;
883
884 lock_sock(sk);
885 if (sock->state == SS_CONNECTING) {
886 if ((1 << sk->sk_state) &
887 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
888 sock->state = SS_DISCONNECTING;
889 else
890 sock->state = SS_CONNECTED;
891 }
892
893 switch (sk->sk_state) {
894 case TCP_CLOSE:
895 err = -ENOTCONN;
896 /* Hack to wake up other listeners, who can poll for
897 EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
898 fallthrough;
899 default:
900 sk->sk_shutdown |= how;
901 if (sk->sk_prot->shutdown)
902 sk->sk_prot->shutdown(sk, how);
903 break;
904
905 /* Remaining two branches are temporary solution for missing
906 * close() in multithreaded environment. It is _not_ a good idea,
907 * but we have no choice until close() is repaired at VFS level.
908 */
909 case TCP_LISTEN:
910 if (!(how & RCV_SHUTDOWN))
911 break;
912 fallthrough;
913 case TCP_SYN_SENT:
914 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
915 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
916 break;
917 }
918
919 /* Wake up anyone sleeping in poll. */
920 sk->sk_state_change(sk);
921 release_sock(sk);
922 return err;
923}
924EXPORT_SYMBOL(inet_shutdown);
925
926/*
927 * ioctl() calls you can issue on an INET socket. Most of these are
928 * device configuration and stuff and very rarely used. Some ioctls
929 * pass on to the socket itself.
930 *
931 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
932 * loads the devconfigure module does its configuring and unloads it.
933 * There's a good 20K of config code hanging around the kernel.
934 */
935
936int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
937{
938 struct sock *sk = sock->sk;
939 int err = 0;
940 struct net *net = sock_net(sk);
941 void __user *p = (void __user *)arg;
942 struct ifreq ifr;
943 struct rtentry rt;
944
945 switch (cmd) {
946 case SIOCADDRT:
947 case SIOCDELRT:
948 if (copy_from_user(&rt, p, sizeof(struct rtentry)))
949 return -EFAULT;
950 err = ip_rt_ioctl(net, cmd, &rt);
951 break;
952 case SIOCRTMSG:
953 err = -EINVAL;
954 break;
955 case SIOCDARP:
956 case SIOCGARP:
957 case SIOCSARP:
958 err = arp_ioctl(net, cmd, (void __user *)arg);
959 break;
960 case SIOCGIFADDR:
961 case SIOCGIFBRDADDR:
962 case SIOCGIFNETMASK:
963 case SIOCGIFDSTADDR:
964 case SIOCGIFPFLAGS:
965 if (get_user_ifreq(&ifr, NULL, p))
966 return -EFAULT;
967 err = devinet_ioctl(net, cmd, &ifr);
968 if (!err && put_user_ifreq(&ifr, p))
969 err = -EFAULT;
970 break;
971
972 case SIOCSIFADDR:
973 case SIOCSIFBRDADDR:
974 case SIOCSIFNETMASK:
975 case SIOCSIFDSTADDR:
976 case SIOCSIFPFLAGS:
977 case SIOCSIFFLAGS:
978 if (get_user_ifreq(&ifr, NULL, p))
979 return -EFAULT;
980 err = devinet_ioctl(net, cmd, &ifr);
981 break;
982 default:
983 if (sk->sk_prot->ioctl)
984 err = sk->sk_prot->ioctl(sk, cmd, arg);
985 else
986 err = -ENOIOCTLCMD;
987 break;
988 }
989 return err;
990}
991EXPORT_SYMBOL(inet_ioctl);
992
993#ifdef CONFIG_COMPAT
994static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd,
995 struct compat_rtentry __user *ur)
996{
997 compat_uptr_t rtdev;
998 struct rtentry rt;
999
1000 if (copy_from_user(&rt.rt_dst, &ur->rt_dst,
1001 3 * sizeof(struct sockaddr)) ||
1002 get_user(rt.rt_flags, &ur->rt_flags) ||
1003 get_user(rt.rt_metric, &ur->rt_metric) ||
1004 get_user(rt.rt_mtu, &ur->rt_mtu) ||
1005 get_user(rt.rt_window, &ur->rt_window) ||
1006 get_user(rt.rt_irtt, &ur->rt_irtt) ||
1007 get_user(rtdev, &ur->rt_dev))
1008 return -EFAULT;
1009
1010 rt.rt_dev = compat_ptr(rtdev);
1011 return ip_rt_ioctl(sock_net(sk), cmd, &rt);
1012}
1013
1014static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1015{
1016 void __user *argp = compat_ptr(arg);
1017 struct sock *sk = sock->sk;
1018
1019 switch (cmd) {
1020 case SIOCADDRT:
1021 case SIOCDELRT:
1022 return inet_compat_routing_ioctl(sk, cmd, argp);
1023 default:
1024 if (!sk->sk_prot->compat_ioctl)
1025 return -ENOIOCTLCMD;
1026 return sk->sk_prot->compat_ioctl(sk, cmd, arg);
1027 }
1028}
1029#endif /* CONFIG_COMPAT */
1030
1031const struct proto_ops inet_stream_ops = {
1032 .family = PF_INET,
1033 .owner = THIS_MODULE,
1034 .release = inet_release,
1035 .bind = inet_bind,
1036 .connect = inet_stream_connect,
1037 .socketpair = sock_no_socketpair,
1038 .accept = inet_accept,
1039 .getname = inet_getname,
1040 .poll = tcp_poll,
1041 .ioctl = inet_ioctl,
1042 .gettstamp = sock_gettstamp,
1043 .listen = inet_listen,
1044 .shutdown = inet_shutdown,
1045 .setsockopt = sock_common_setsockopt,
1046 .getsockopt = sock_common_getsockopt,
1047 .sendmsg = inet_sendmsg,
1048 .recvmsg = inet_recvmsg,
1049#ifdef CONFIG_MMU
1050 .mmap = tcp_mmap,
1051#endif
1052 .sendpage = inet_sendpage,
1053 .splice_read = tcp_splice_read,
1054 .read_sock = tcp_read_sock,
1055 .read_skb = tcp_read_skb,
1056 .sendmsg_locked = tcp_sendmsg_locked,
1057 .sendpage_locked = tcp_sendpage_locked,
1058 .peek_len = tcp_peek_len,
1059#ifdef CONFIG_COMPAT
1060 .compat_ioctl = inet_compat_ioctl,
1061#endif
1062 .set_rcvlowat = tcp_set_rcvlowat,
1063};
1064EXPORT_SYMBOL(inet_stream_ops);
1065
1066const struct proto_ops inet_dgram_ops = {
1067 .family = PF_INET,
1068 .owner = THIS_MODULE,
1069 .release = inet_release,
1070 .bind = inet_bind,
1071 .connect = inet_dgram_connect,
1072 .socketpair = sock_no_socketpair,
1073 .accept = sock_no_accept,
1074 .getname = inet_getname,
1075 .poll = udp_poll,
1076 .ioctl = inet_ioctl,
1077 .gettstamp = sock_gettstamp,
1078 .listen = sock_no_listen,
1079 .shutdown = inet_shutdown,
1080 .setsockopt = sock_common_setsockopt,
1081 .getsockopt = sock_common_getsockopt,
1082 .sendmsg = inet_sendmsg,
1083 .read_skb = udp_read_skb,
1084 .recvmsg = inet_recvmsg,
1085 .mmap = sock_no_mmap,
1086 .sendpage = inet_sendpage,
1087 .set_peek_off = sk_set_peek_off,
1088#ifdef CONFIG_COMPAT
1089 .compat_ioctl = inet_compat_ioctl,
1090#endif
1091};
1092EXPORT_SYMBOL(inet_dgram_ops);
1093
1094/*
1095 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1096 * udp_poll
1097 */
1098static const struct proto_ops inet_sockraw_ops = {
1099 .family = PF_INET,
1100 .owner = THIS_MODULE,
1101 .release = inet_release,
1102 .bind = inet_bind,
1103 .connect = inet_dgram_connect,
1104 .socketpair = sock_no_socketpair,
1105 .accept = sock_no_accept,
1106 .getname = inet_getname,
1107 .poll = datagram_poll,
1108 .ioctl = inet_ioctl,
1109 .gettstamp = sock_gettstamp,
1110 .listen = sock_no_listen,
1111 .shutdown = inet_shutdown,
1112 .setsockopt = sock_common_setsockopt,
1113 .getsockopt = sock_common_getsockopt,
1114 .sendmsg = inet_sendmsg,
1115 .recvmsg = inet_recvmsg,
1116 .mmap = sock_no_mmap,
1117 .sendpage = inet_sendpage,
1118#ifdef CONFIG_COMPAT
1119 .compat_ioctl = inet_compat_ioctl,
1120#endif
1121};
1122
1123static const struct net_proto_family inet_family_ops = {
1124 .family = PF_INET,
1125 .create = inet_create,
1126 .owner = THIS_MODULE,
1127};
1128
1129/* Upon startup we insert all the elements in inetsw_array[] into
1130 * the linked list inetsw.
1131 */
1132static struct inet_protosw inetsw_array[] =
1133{
1134 {
1135 .type = SOCK_STREAM,
1136 .protocol = IPPROTO_TCP,
1137 .prot = &tcp_prot,
1138 .ops = &inet_stream_ops,
1139 .flags = INET_PROTOSW_PERMANENT |
1140 INET_PROTOSW_ICSK,
1141 },
1142
1143 {
1144 .type = SOCK_DGRAM,
1145 .protocol = IPPROTO_UDP,
1146 .prot = &udp_prot,
1147 .ops = &inet_dgram_ops,
1148 .flags = INET_PROTOSW_PERMANENT,
1149 },
1150
1151 {
1152 .type = SOCK_DGRAM,
1153 .protocol = IPPROTO_ICMP,
1154 .prot = &ping_prot,
1155 .ops = &inet_sockraw_ops,
1156 .flags = INET_PROTOSW_REUSE,
1157 },
1158
1159 {
1160 .type = SOCK_RAW,
1161 .protocol = IPPROTO_IP, /* wild card */
1162 .prot = &raw_prot,
1163 .ops = &inet_sockraw_ops,
1164 .flags = INET_PROTOSW_REUSE,
1165 }
1166};
1167
1168#define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1169
1170void inet_register_protosw(struct inet_protosw *p)
1171{
1172 struct list_head *lh;
1173 struct inet_protosw *answer;
1174 int protocol = p->protocol;
1175 struct list_head *last_perm;
1176
1177 spin_lock_bh(&inetsw_lock);
1178
1179 if (p->type >= SOCK_MAX)
1180 goto out_illegal;
1181
1182 /* If we are trying to override a permanent protocol, bail. */
1183 last_perm = &inetsw[p->type];
1184 list_for_each(lh, &inetsw[p->type]) {
1185 answer = list_entry(lh, struct inet_protosw, list);
1186 /* Check only the non-wild match. */
1187 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1188 break;
1189 if (protocol == answer->protocol)
1190 goto out_permanent;
1191 last_perm = lh;
1192 }
1193
1194 /* Add the new entry after the last permanent entry if any, so that
1195 * the new entry does not override a permanent entry when matched with
1196 * a wild-card protocol. But it is allowed to override any existing
1197 * non-permanent entry. This means that when we remove this entry, the
1198 * system automatically returns to the old behavior.
1199 */
1200 list_add_rcu(&p->list, last_perm);
1201out:
1202 spin_unlock_bh(&inetsw_lock);
1203
1204 return;
1205
1206out_permanent:
1207 pr_err("Attempt to override permanent protocol %d\n", protocol);
1208 goto out;
1209
1210out_illegal:
1211 pr_err("Ignoring attempt to register invalid socket type %d\n",
1212 p->type);
1213 goto out;
1214}
1215EXPORT_SYMBOL(inet_register_protosw);
1216
1217void inet_unregister_protosw(struct inet_protosw *p)
1218{
1219 if (INET_PROTOSW_PERMANENT & p->flags) {
1220 pr_err("Attempt to unregister permanent protocol %d\n",
1221 p->protocol);
1222 } else {
1223 spin_lock_bh(&inetsw_lock);
1224 list_del_rcu(&p->list);
1225 spin_unlock_bh(&inetsw_lock);
1226
1227 synchronize_net();
1228 }
1229}
1230EXPORT_SYMBOL(inet_unregister_protosw);
1231
1232static int inet_sk_reselect_saddr(struct sock *sk)
1233{
1234 struct inet_sock *inet = inet_sk(sk);
1235 __be32 old_saddr = inet->inet_saddr;
1236 __be32 daddr = inet->inet_daddr;
1237 struct flowi4 *fl4;
1238 struct rtable *rt;
1239 __be32 new_saddr;
1240 struct ip_options_rcu *inet_opt;
1241 int err;
1242
1243 inet_opt = rcu_dereference_protected(inet->inet_opt,
1244 lockdep_sock_is_held(sk));
1245 if (inet_opt && inet_opt->opt.srr)
1246 daddr = inet_opt->opt.faddr;
1247
1248 /* Query new route. */
1249 fl4 = &inet->cork.fl.u.ip4;
1250 rt = ip_route_connect(fl4, daddr, 0, sk->sk_bound_dev_if,
1251 sk->sk_protocol, inet->inet_sport,
1252 inet->inet_dport, sk);
1253 if (IS_ERR(rt))
1254 return PTR_ERR(rt);
1255
1256 new_saddr = fl4->saddr;
1257
1258 if (new_saddr == old_saddr) {
1259 sk_setup_caps(sk, &rt->dst);
1260 return 0;
1261 }
1262
1263 err = inet_bhash2_update_saddr(sk, &new_saddr, AF_INET);
1264 if (err) {
1265 ip_rt_put(rt);
1266 return err;
1267 }
1268
1269 sk_setup_caps(sk, &rt->dst);
1270
1271 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) > 1) {
1272 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1273 __func__, &old_saddr, &new_saddr);
1274 }
1275
1276 /*
1277 * XXX The only one ugly spot where we need to
1278 * XXX really change the sockets identity after
1279 * XXX it has entered the hashes. -DaveM
1280 *
1281 * Besides that, it does not check for connection
1282 * uniqueness. Wait for troubles.
1283 */
1284 return __sk_prot_rehash(sk);
1285}
1286
1287int inet_sk_rebuild_header(struct sock *sk)
1288{
1289 struct inet_sock *inet = inet_sk(sk);
1290 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1291 __be32 daddr;
1292 struct ip_options_rcu *inet_opt;
1293 struct flowi4 *fl4;
1294 int err;
1295
1296 /* Route is OK, nothing to do. */
1297 if (rt)
1298 return 0;
1299
1300 /* Reroute. */
1301 rcu_read_lock();
1302 inet_opt = rcu_dereference(inet->inet_opt);
1303 daddr = inet->inet_daddr;
1304 if (inet_opt && inet_opt->opt.srr)
1305 daddr = inet_opt->opt.faddr;
1306 rcu_read_unlock();
1307 fl4 = &inet->cork.fl.u.ip4;
1308 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1309 inet->inet_dport, inet->inet_sport,
1310 sk->sk_protocol, RT_CONN_FLAGS(sk),
1311 sk->sk_bound_dev_if);
1312 if (!IS_ERR(rt)) {
1313 err = 0;
1314 sk_setup_caps(sk, &rt->dst);
1315 } else {
1316 err = PTR_ERR(rt);
1317
1318 /* Routing failed... */
1319 sk->sk_route_caps = 0;
1320 /*
1321 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1322 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1323 */
1324 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) ||
1325 sk->sk_state != TCP_SYN_SENT ||
1326 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1327 (err = inet_sk_reselect_saddr(sk)) != 0)
1328 sk->sk_err_soft = -err;
1329 }
1330
1331 return err;
1332}
1333EXPORT_SYMBOL(inet_sk_rebuild_header);
1334
1335void inet_sk_set_state(struct sock *sk, int state)
1336{
1337 trace_inet_sock_set_state(sk, sk->sk_state, state);
1338 sk->sk_state = state;
1339}
1340EXPORT_SYMBOL(inet_sk_set_state);
1341
1342void inet_sk_state_store(struct sock *sk, int newstate)
1343{
1344 trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1345 smp_store_release(&sk->sk_state, newstate);
1346}
1347
1348struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1349 netdev_features_t features)
1350{
1351 bool udpfrag = false, fixedid = false, gso_partial, encap;
1352 struct sk_buff *segs = ERR_PTR(-EINVAL);
1353 const struct net_offload *ops;
1354 unsigned int offset = 0;
1355 struct iphdr *iph;
1356 int proto, tot_len;
1357 int nhoff;
1358 int ihl;
1359 int id;
1360
1361 skb_reset_network_header(skb);
1362 nhoff = skb_network_header(skb) - skb_mac_header(skb);
1363 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1364 goto out;
1365
1366 iph = ip_hdr(skb);
1367 ihl = iph->ihl * 4;
1368 if (ihl < sizeof(*iph))
1369 goto out;
1370
1371 id = ntohs(iph->id);
1372 proto = iph->protocol;
1373
1374 /* Warning: after this point, iph might be no longer valid */
1375 if (unlikely(!pskb_may_pull(skb, ihl)))
1376 goto out;
1377 __skb_pull(skb, ihl);
1378
1379 encap = SKB_GSO_CB(skb)->encap_level > 0;
1380 if (encap)
1381 features &= skb->dev->hw_enc_features;
1382 SKB_GSO_CB(skb)->encap_level += ihl;
1383
1384 skb_reset_transport_header(skb);
1385
1386 segs = ERR_PTR(-EPROTONOSUPPORT);
1387
1388 if (!skb->encapsulation || encap) {
1389 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1390 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1391
1392 /* fixed ID is invalid if DF bit is not set */
1393 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1394 goto out;
1395 }
1396
1397 ops = rcu_dereference(inet_offloads[proto]);
1398 if (likely(ops && ops->callbacks.gso_segment)) {
1399 segs = ops->callbacks.gso_segment(skb, features);
1400 if (!segs)
1401 skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
1402 }
1403
1404 if (IS_ERR_OR_NULL(segs))
1405 goto out;
1406
1407 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1408
1409 skb = segs;
1410 do {
1411 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1412 if (udpfrag) {
1413 iph->frag_off = htons(offset >> 3);
1414 if (skb->next)
1415 iph->frag_off |= htons(IP_MF);
1416 offset += skb->len - nhoff - ihl;
1417 tot_len = skb->len - nhoff;
1418 } else if (skb_is_gso(skb)) {
1419 if (!fixedid) {
1420 iph->id = htons(id);
1421 id += skb_shinfo(skb)->gso_segs;
1422 }
1423
1424 if (gso_partial)
1425 tot_len = skb_shinfo(skb)->gso_size +
1426 SKB_GSO_CB(skb)->data_offset +
1427 skb->head - (unsigned char *)iph;
1428 else
1429 tot_len = skb->len - nhoff;
1430 } else {
1431 if (!fixedid)
1432 iph->id = htons(id++);
1433 tot_len = skb->len - nhoff;
1434 }
1435 iph->tot_len = htons(tot_len);
1436 ip_send_check(iph);
1437 if (encap)
1438 skb_reset_inner_headers(skb);
1439 skb->network_header = (u8 *)iph - skb->head;
1440 skb_reset_mac_len(skb);
1441 } while ((skb = skb->next));
1442
1443out:
1444 return segs;
1445}
1446
1447static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1448 netdev_features_t features)
1449{
1450 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1451 return ERR_PTR(-EINVAL);
1452
1453 return inet_gso_segment(skb, features);
1454}
1455
1456struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1457{
1458 const struct net_offload *ops;
1459 struct sk_buff *pp = NULL;
1460 const struct iphdr *iph;
1461 struct sk_buff *p;
1462 unsigned int hlen;
1463 unsigned int off;
1464 unsigned int id;
1465 int flush = 1;
1466 int proto;
1467
1468 off = skb_gro_offset(skb);
1469 hlen = off + sizeof(*iph);
1470 iph = skb_gro_header(skb, hlen, off);
1471 if (unlikely(!iph))
1472 goto out;
1473
1474 proto = iph->protocol;
1475
1476 ops = rcu_dereference(inet_offloads[proto]);
1477 if (!ops || !ops->callbacks.gro_receive)
1478 goto out;
1479
1480 if (*(u8 *)iph != 0x45)
1481 goto out;
1482
1483 if (ip_is_fragment(iph))
1484 goto out;
1485
1486 if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1487 goto out;
1488
1489 id = ntohl(*(__be32 *)&iph->id);
1490 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1491 id >>= 16;
1492
1493 list_for_each_entry(p, head, list) {
1494 struct iphdr *iph2;
1495 u16 flush_id;
1496
1497 if (!NAPI_GRO_CB(p)->same_flow)
1498 continue;
1499
1500 iph2 = (struct iphdr *)(p->data + off);
1501 /* The above works because, with the exception of the top
1502 * (inner most) layer, we only aggregate pkts with the same
1503 * hdr length so all the hdrs we'll need to verify will start
1504 * at the same offset.
1505 */
1506 if ((iph->protocol ^ iph2->protocol) |
1507 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1508 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1509 NAPI_GRO_CB(p)->same_flow = 0;
1510 continue;
1511 }
1512
1513 /* All fields must match except length and checksum. */
1514 NAPI_GRO_CB(p)->flush |=
1515 (iph->ttl ^ iph2->ttl) |
1516 (iph->tos ^ iph2->tos) |
1517 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1518
1519 NAPI_GRO_CB(p)->flush |= flush;
1520
1521 /* We need to store of the IP ID check to be included later
1522 * when we can verify that this packet does in fact belong
1523 * to a given flow.
1524 */
1525 flush_id = (u16)(id - ntohs(iph2->id));
1526
1527 /* This bit of code makes it much easier for us to identify
1528 * the cases where we are doing atomic vs non-atomic IP ID
1529 * checks. Specifically an atomic check can return IP ID
1530 * values 0 - 0xFFFF, while a non-atomic check can only
1531 * return 0 or 0xFFFF.
1532 */
1533 if (!NAPI_GRO_CB(p)->is_atomic ||
1534 !(iph->frag_off & htons(IP_DF))) {
1535 flush_id ^= NAPI_GRO_CB(p)->count;
1536 flush_id = flush_id ? 0xFFFF : 0;
1537 }
1538
1539 /* If the previous IP ID value was based on an atomic
1540 * datagram we can overwrite the value and ignore it.
1541 */
1542 if (NAPI_GRO_CB(skb)->is_atomic)
1543 NAPI_GRO_CB(p)->flush_id = flush_id;
1544 else
1545 NAPI_GRO_CB(p)->flush_id |= flush_id;
1546 }
1547
1548 NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1549 NAPI_GRO_CB(skb)->flush |= flush;
1550 skb_set_network_header(skb, off);
1551 /* The above will be needed by the transport layer if there is one
1552 * immediately following this IP hdr.
1553 */
1554
1555 /* Note : No need to call skb_gro_postpull_rcsum() here,
1556 * as we already checked checksum over ipv4 header was 0
1557 */
1558 skb_gro_pull(skb, sizeof(*iph));
1559 skb_set_transport_header(skb, skb_gro_offset(skb));
1560
1561 pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1562 ops->callbacks.gro_receive, head, skb);
1563
1564out:
1565 skb_gro_flush_final(skb, pp, flush);
1566
1567 return pp;
1568}
1569
1570static struct sk_buff *ipip_gro_receive(struct list_head *head,
1571 struct sk_buff *skb)
1572{
1573 if (NAPI_GRO_CB(skb)->encap_mark) {
1574 NAPI_GRO_CB(skb)->flush = 1;
1575 return NULL;
1576 }
1577
1578 NAPI_GRO_CB(skb)->encap_mark = 1;
1579
1580 return inet_gro_receive(head, skb);
1581}
1582
1583#define SECONDS_PER_DAY 86400
1584
1585/* inet_current_timestamp - Return IP network timestamp
1586 *
1587 * Return milliseconds since midnight in network byte order.
1588 */
1589__be32 inet_current_timestamp(void)
1590{
1591 u32 secs;
1592 u32 msecs;
1593 struct timespec64 ts;
1594
1595 ktime_get_real_ts64(&ts);
1596
1597 /* Get secs since midnight. */
1598 (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1599 /* Convert to msecs. */
1600 msecs = secs * MSEC_PER_SEC;
1601 /* Convert nsec to msec. */
1602 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1603
1604 /* Convert to network byte order. */
1605 return htonl(msecs);
1606}
1607EXPORT_SYMBOL(inet_current_timestamp);
1608
1609int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1610{
1611 if (sk->sk_family == AF_INET)
1612 return ip_recv_error(sk, msg, len, addr_len);
1613#if IS_ENABLED(CONFIG_IPV6)
1614 if (sk->sk_family == AF_INET6)
1615 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1616#endif
1617 return -EINVAL;
1618}
1619
1620int inet_gro_complete(struct sk_buff *skb, int nhoff)
1621{
1622 __be16 newlen = htons(skb->len - nhoff);
1623 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1624 const struct net_offload *ops;
1625 int proto = iph->protocol;
1626 int err = -ENOSYS;
1627
1628 if (skb->encapsulation) {
1629 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1630 skb_set_inner_network_header(skb, nhoff);
1631 }
1632
1633 csum_replace2(&iph->check, iph->tot_len, newlen);
1634 iph->tot_len = newlen;
1635
1636 ops = rcu_dereference(inet_offloads[proto]);
1637 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1638 goto out;
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
1648out:
1649 return err;
1650}
1651
1652static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1653{
1654 skb->encapsulation = 1;
1655 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1656 return inet_gro_complete(skb, nhoff);
1657}
1658
1659int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1660 unsigned short type, unsigned char protocol,
1661 struct net *net)
1662{
1663 struct socket *sock;
1664 int rc = sock_create_kern(net, family, type, protocol, &sock);
1665
1666 if (rc == 0) {
1667 *sk = sock->sk;
1668 (*sk)->sk_allocation = GFP_ATOMIC;
1669 (*sk)->sk_use_task_frag = false;
1670 /*
1671 * Unhash it so that IP input processing does not even see it,
1672 * we do not wish this socket to see incoming packets.
1673 */
1674 (*sk)->sk_prot->unhash(*sk);
1675 }
1676 return rc;
1677}
1678EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1679
1680unsigned long snmp_fold_field(void __percpu *mib, int offt)
1681{
1682 unsigned long res = 0;
1683 int i;
1684
1685 for_each_possible_cpu(i)
1686 res += snmp_get_cpu_field(mib, i, offt);
1687 return res;
1688}
1689EXPORT_SYMBOL_GPL(snmp_fold_field);
1690
1691#if BITS_PER_LONG==32
1692
1693u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1694 size_t syncp_offset)
1695{
1696 void *bhptr;
1697 struct u64_stats_sync *syncp;
1698 u64 v;
1699 unsigned int start;
1700
1701 bhptr = per_cpu_ptr(mib, cpu);
1702 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1703 do {
1704 start = u64_stats_fetch_begin(syncp);
1705 v = *(((u64 *)bhptr) + offt);
1706 } while (u64_stats_fetch_retry(syncp, start));
1707
1708 return v;
1709}
1710EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1711
1712u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1713{
1714 u64 res = 0;
1715 int cpu;
1716
1717 for_each_possible_cpu(cpu) {
1718 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1719 }
1720 return res;
1721}
1722EXPORT_SYMBOL_GPL(snmp_fold_field64);
1723#endif
1724
1725#ifdef CONFIG_IP_MULTICAST
1726static const struct net_protocol igmp_protocol = {
1727 .handler = igmp_rcv,
1728};
1729#endif
1730
1731static const struct net_protocol tcp_protocol = {
1732 .handler = tcp_v4_rcv,
1733 .err_handler = tcp_v4_err,
1734 .no_policy = 1,
1735 .icmp_strict_tag_validation = 1,
1736};
1737
1738static const struct net_protocol udp_protocol = {
1739 .handler = udp_rcv,
1740 .err_handler = udp_err,
1741 .no_policy = 1,
1742};
1743
1744static const struct net_protocol icmp_protocol = {
1745 .handler = icmp_rcv,
1746 .err_handler = icmp_err,
1747 .no_policy = 1,
1748};
1749
1750static __net_init int ipv4_mib_init_net(struct net *net)
1751{
1752 int i;
1753
1754 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1755 if (!net->mib.tcp_statistics)
1756 goto err_tcp_mib;
1757 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1758 if (!net->mib.ip_statistics)
1759 goto err_ip_mib;
1760
1761 for_each_possible_cpu(i) {
1762 struct ipstats_mib *af_inet_stats;
1763 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1764 u64_stats_init(&af_inet_stats->syncp);
1765 }
1766
1767 net->mib.net_statistics = alloc_percpu(struct linux_mib);
1768 if (!net->mib.net_statistics)
1769 goto err_net_mib;
1770 net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1771 if (!net->mib.udp_statistics)
1772 goto err_udp_mib;
1773 net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1774 if (!net->mib.udplite_statistics)
1775 goto err_udplite_mib;
1776 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1777 if (!net->mib.icmp_statistics)
1778 goto err_icmp_mib;
1779 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1780 GFP_KERNEL);
1781 if (!net->mib.icmpmsg_statistics)
1782 goto err_icmpmsg_mib;
1783
1784 tcp_mib_init(net);
1785 return 0;
1786
1787err_icmpmsg_mib:
1788 free_percpu(net->mib.icmp_statistics);
1789err_icmp_mib:
1790 free_percpu(net->mib.udplite_statistics);
1791err_udplite_mib:
1792 free_percpu(net->mib.udp_statistics);
1793err_udp_mib:
1794 free_percpu(net->mib.net_statistics);
1795err_net_mib:
1796 free_percpu(net->mib.ip_statistics);
1797err_ip_mib:
1798 free_percpu(net->mib.tcp_statistics);
1799err_tcp_mib:
1800 return -ENOMEM;
1801}
1802
1803static __net_exit void ipv4_mib_exit_net(struct net *net)
1804{
1805 kfree(net->mib.icmpmsg_statistics);
1806 free_percpu(net->mib.icmp_statistics);
1807 free_percpu(net->mib.udplite_statistics);
1808 free_percpu(net->mib.udp_statistics);
1809 free_percpu(net->mib.net_statistics);
1810 free_percpu(net->mib.ip_statistics);
1811 free_percpu(net->mib.tcp_statistics);
1812#ifdef CONFIG_MPTCP
1813 /* allocated on demand, see mptcp_init_sock() */
1814 free_percpu(net->mib.mptcp_statistics);
1815#endif
1816}
1817
1818static __net_initdata struct pernet_operations ipv4_mib_ops = {
1819 .init = ipv4_mib_init_net,
1820 .exit = ipv4_mib_exit_net,
1821};
1822
1823static int __init init_ipv4_mibs(void)
1824{
1825 return register_pernet_subsys(&ipv4_mib_ops);
1826}
1827
1828static __net_init int inet_init_net(struct net *net)
1829{
1830 /*
1831 * Set defaults for local port range
1832 */
1833 seqlock_init(&net->ipv4.ip_local_ports.lock);
1834 net->ipv4.ip_local_ports.range[0] = 32768;
1835 net->ipv4.ip_local_ports.range[1] = 60999;
1836
1837 seqlock_init(&net->ipv4.ping_group_range.lock);
1838 /*
1839 * Sane defaults - nobody may create ping sockets.
1840 * Boot scripts should set this to distro-specific group.
1841 */
1842 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1843 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1844
1845 /* Default values for sysctl-controlled parameters.
1846 * We set them here, in case sysctl is not compiled.
1847 */
1848 net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1849 net->ipv4.sysctl_ip_fwd_update_priority = 1;
1850 net->ipv4.sysctl_ip_dynaddr = 0;
1851 net->ipv4.sysctl_ip_early_demux = 1;
1852 net->ipv4.sysctl_udp_early_demux = 1;
1853 net->ipv4.sysctl_tcp_early_demux = 1;
1854 net->ipv4.sysctl_nexthop_compat_mode = 1;
1855#ifdef CONFIG_SYSCTL
1856 net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1857#endif
1858
1859 /* Some igmp sysctl, whose values are always used */
1860 net->ipv4.sysctl_igmp_max_memberships = 20;
1861 net->ipv4.sysctl_igmp_max_msf = 10;
1862 /* IGMP reports for link-local multicast groups are enabled by default */
1863 net->ipv4.sysctl_igmp_llm_reports = 1;
1864 net->ipv4.sysctl_igmp_qrv = 2;
1865
1866 net->ipv4.sysctl_fib_notify_on_flag_change = 0;
1867
1868 return 0;
1869}
1870
1871static __net_initdata struct pernet_operations af_inet_ops = {
1872 .init = inet_init_net,
1873};
1874
1875static int __init init_inet_pernet_ops(void)
1876{
1877 return register_pernet_subsys(&af_inet_ops);
1878}
1879
1880static int ipv4_proc_init(void);
1881
1882/*
1883 * IP protocol layer initialiser
1884 */
1885
1886static struct packet_offload ip_packet_offload __read_mostly = {
1887 .type = cpu_to_be16(ETH_P_IP),
1888 .callbacks = {
1889 .gso_segment = inet_gso_segment,
1890 .gro_receive = inet_gro_receive,
1891 .gro_complete = inet_gro_complete,
1892 },
1893};
1894
1895static const struct net_offload ipip_offload = {
1896 .callbacks = {
1897 .gso_segment = ipip_gso_segment,
1898 .gro_receive = ipip_gro_receive,
1899 .gro_complete = ipip_gro_complete,
1900 },
1901};
1902
1903static int __init ipip_offload_init(void)
1904{
1905 return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1906}
1907
1908static int __init ipv4_offload_init(void)
1909{
1910 /*
1911 * Add offloads
1912 */
1913 if (udpv4_offload_init() < 0)
1914 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1915 if (tcpv4_offload_init() < 0)
1916 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1917 if (ipip_offload_init() < 0)
1918 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1919
1920 dev_add_offload(&ip_packet_offload);
1921 return 0;
1922}
1923
1924fs_initcall(ipv4_offload_init);
1925
1926static struct packet_type ip_packet_type __read_mostly = {
1927 .type = cpu_to_be16(ETH_P_IP),
1928 .func = ip_rcv,
1929 .list_func = ip_list_rcv,
1930};
1931
1932static int __init inet_init(void)
1933{
1934 struct inet_protosw *q;
1935 struct list_head *r;
1936 int rc;
1937
1938 sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1939
1940 raw_hashinfo_init(&raw_v4_hashinfo);
1941
1942 rc = proto_register(&tcp_prot, 1);
1943 if (rc)
1944 goto out;
1945
1946 rc = proto_register(&udp_prot, 1);
1947 if (rc)
1948 goto out_unregister_tcp_proto;
1949
1950 rc = proto_register(&raw_prot, 1);
1951 if (rc)
1952 goto out_unregister_udp_proto;
1953
1954 rc = proto_register(&ping_prot, 1);
1955 if (rc)
1956 goto out_unregister_raw_proto;
1957
1958 /*
1959 * Tell SOCKET that we are alive...
1960 */
1961
1962 (void)sock_register(&inet_family_ops);
1963
1964#ifdef CONFIG_SYSCTL
1965 ip_static_sysctl_init();
1966#endif
1967
1968 /*
1969 * Add all the base protocols.
1970 */
1971
1972 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1973 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1974 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1975 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1976 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1977 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1978#ifdef CONFIG_IP_MULTICAST
1979 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1980 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1981#endif
1982
1983 /* Register the socket-side information for inet_create. */
1984 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1985 INIT_LIST_HEAD(r);
1986
1987 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1988 inet_register_protosw(q);
1989
1990 /*
1991 * Set the ARP module up
1992 */
1993
1994 arp_init();
1995
1996 /*
1997 * Set the IP module up
1998 */
1999
2000 ip_init();
2001
2002 /* Initialise per-cpu ipv4 mibs */
2003 if (init_ipv4_mibs())
2004 panic("%s: Cannot init ipv4 mibs\n", __func__);
2005
2006 /* Setup TCP slab cache for open requests. */
2007 tcp_init();
2008
2009 /* Setup UDP memory threshold */
2010 udp_init();
2011
2012 /* Add UDP-Lite (RFC 3828) */
2013 udplite4_register();
2014
2015 raw_init();
2016
2017 ping_init();
2018
2019 /*
2020 * Set the ICMP layer up
2021 */
2022
2023 if (icmp_init() < 0)
2024 panic("Failed to create the ICMP control socket.\n");
2025
2026 /*
2027 * Initialise the multicast router
2028 */
2029#if defined(CONFIG_IP_MROUTE)
2030 if (ip_mr_init())
2031 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2032#endif
2033
2034 if (init_inet_pernet_ops())
2035 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2036
2037 ipv4_proc_init();
2038
2039 ipfrag_init();
2040
2041 dev_add_pack(&ip_packet_type);
2042
2043 ip_tunnel_core_init();
2044
2045 rc = 0;
2046out:
2047 return rc;
2048out_unregister_raw_proto:
2049 proto_unregister(&raw_prot);
2050out_unregister_udp_proto:
2051 proto_unregister(&udp_prot);
2052out_unregister_tcp_proto:
2053 proto_unregister(&tcp_prot);
2054 goto out;
2055}
2056
2057fs_initcall(inet_init);
2058
2059/* ------------------------------------------------------------------------ */
2060
2061#ifdef CONFIG_PROC_FS
2062static int __init ipv4_proc_init(void)
2063{
2064 int rc = 0;
2065
2066 if (raw_proc_init())
2067 goto out_raw;
2068 if (tcp4_proc_init())
2069 goto out_tcp;
2070 if (udp4_proc_init())
2071 goto out_udp;
2072 if (ping_proc_init())
2073 goto out_ping;
2074 if (ip_misc_proc_init())
2075 goto out_misc;
2076out:
2077 return rc;
2078out_misc:
2079 ping_proc_exit();
2080out_ping:
2081 udp4_proc_exit();
2082out_udp:
2083 tcp4_proc_exit();
2084out_tcp:
2085 raw_proc_exit();
2086out_raw:
2087 rc = -ENOMEM;
2088 goto out;
2089}
2090
2091#else /* CONFIG_PROC_FS */
2092static int __init ipv4_proc_init(void)
2093{
2094 return 0;
2095}
2096#endif /* CONFIG_PROC_FS */