blob: 0a277453526b9c9730475b77368bfdb0cae2b894 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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 * Version: $Id: af_inet.c,v 1.137 2002/02/01 22:01:03 davem Exp $
9 *
Jesper Juhl02c30a82005-05-05 16:16:16 -070010 * Authors: Ross Biro
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Alan Cox, <A.Cox@swansea.ac.uk>
14 *
15 * Changes (see also sock.c)
16 *
17 * piggy,
18 * Karl Knutson : Socket protocol table
19 * A.N.Kuznetsov : Socket death error in accept().
20 * John Richardson : Fix non blocking error in connect()
21 * so sockets that fail to connect
22 * don't return -EINPROGRESS.
23 * Alan Cox : Asynchronous I/O support
24 * Alan Cox : Keep correct socket pointer on sock
25 * structures
26 * when accept() ed
27 * Alan Cox : Semantics of SO_LINGER aren't state
28 * moved to close when you look carefully.
29 * With this fixed and the accept bug fixed
30 * some RPC stuff seems happier.
31 * Niibe Yutaka : 4.4BSD style write async I/O
32 * Alan Cox,
33 * Tony Gale : Fixed reuse semantics.
34 * Alan Cox : bind() shouldn't abort existing but dead
35 * sockets. Stops FTP netin:.. I hope.
36 * Alan Cox : bind() works correctly for RAW sockets.
37 * Note that FreeBSD at least was broken
38 * in this respect so be careful with
39 * compatibility tests...
40 * Alan Cox : routing cache support
41 * Alan Cox : memzero the socket structure for
42 * compactness.
43 * Matt Day : nonblock connect error handler
44 * Alan Cox : Allow large numbers of pending sockets
45 * (eg for big web sites), but only if
46 * specifically application requested.
47 * Alan Cox : New buffering throughout IP. Used
48 * dumbly.
49 * Alan Cox : New buffering now used smartly.
50 * Alan Cox : BSD rather than common sense
51 * interpretation of listen.
52 * Germano Caronni : Assorted small races.
53 * Alan Cox : sendmsg/recvmsg basic support.
54 * Alan Cox : Only sendmsg/recvmsg now supported.
55 * Alan Cox : Locked down bind (see security list).
56 * Alan Cox : Loosened bind a little.
57 * Mike McLagan : ADD/DEL DLCI Ioctls
58 * Willy Konynenberg : Transparent proxying support.
59 * David S. Miller : New socket lookup architecture.
60 * Some other random speedups.
61 * Cyrus Durgin : Cleaned up file for kmod hacks.
62 * Andi Kleen : Fix inet_stream_connect TCP race.
63 *
64 * This program is free software; you can redistribute it and/or
65 * modify it under the terms of the GNU General Public License
66 * as published by the Free Software Foundation; either version
67 * 2 of the License, or (at your option) any later version.
68 */
69
70#include <linux/config.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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070076#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>
Randy Dunlap4fc268d2006-01-11 12:17:47 -080082#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070083#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
91#include <asm/uaccess.h>
92#include <asm/system.h>
93
94#include <linux/smp_lock.h>
95#include <linux/inet.h>
96#include <linux/igmp.h>
Arnaldo Carvalho de Melo14c85022005-12-27 02:43:12 -020097#include <linux/inetdevice.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070098#include <linux/netdevice.h>
99#include <net/ip.h>
100#include <net/protocol.h>
101#include <net/arp.h>
102#include <net/route.h>
103#include <net/ip_fib.h>
Arnaldo Carvalho de Melo295f7322005-08-09 20:11:56 -0700104#include <net/inet_connection_sock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105#include <net/tcp.h>
106#include <net/udp.h>
107#include <linux/skbuff.h>
108#include <net/sock.h>
109#include <net/raw.h>
110#include <net/icmp.h>
111#include <net/ipip.h>
112#include <net/inet_common.h>
113#include <net/xfrm.h>
114#ifdef CONFIG_IP_MROUTE
115#include <linux/mroute.h>
116#endif
117
Eric Dumazetba899662005-08-26 12:05:31 -0700118DEFINE_SNMP_STAT(struct linux_mib, net_statistics) __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120extern void ip_mc_drop_socket(struct sock *sk);
121
122/* The inetsw table contains everything that inet_create needs to
123 * build a new socket.
124 */
125static struct list_head inetsw[SOCK_MAX];
126static DEFINE_SPINLOCK(inetsw_lock);
127
128/* New destruction routine */
129
130void inet_sock_destruct(struct sock *sk)
131{
132 struct inet_sock *inet = inet_sk(sk);
133
134 __skb_queue_purge(&sk->sk_receive_queue);
135 __skb_queue_purge(&sk->sk_error_queue);
136
137 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
138 printk("Attempt to release TCP socket in state %d %p\n",
139 sk->sk_state, sk);
140 return;
141 }
142 if (!sock_flag(sk, SOCK_DEAD)) {
143 printk("Attempt to release alive inet socket %p\n", sk);
144 return;
145 }
146
147 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
148 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
149 BUG_TRAP(!sk->sk_wmem_queued);
150 BUG_TRAP(!sk->sk_forward_alloc);
151
Jesper Juhla51482b2005-11-08 09:41:34 -0800152 kfree(inet->opt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 dst_release(sk->sk_dst_cache);
Arnaldo Carvalho de Meloe6848972005-08-09 19:45:38 -0700154 sk_refcnt_debug_dec(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155}
156
157/*
158 * The routines beyond this point handle the behaviour of an AF_INET
159 * socket object. Mostly it punts to the subprotocols of IP to do
160 * the work.
161 */
162
163/*
164 * Automatically bind an unbound socket.
165 */
166
167static int inet_autobind(struct sock *sk)
168{
169 struct inet_sock *inet;
170 /* We may need to bind the socket. */
171 lock_sock(sk);
172 inet = inet_sk(sk);
173 if (!inet->num) {
174 if (sk->sk_prot->get_port(sk, 0)) {
175 release_sock(sk);
176 return -EAGAIN;
177 }
178 inet->sport = htons(inet->num);
179 }
180 release_sock(sk);
181 return 0;
182}
183
184/*
185 * Move a socket into listening state.
186 */
187int inet_listen(struct socket *sock, int backlog)
188{
189 struct sock *sk = sock->sk;
190 unsigned char old_state;
191 int err;
192
193 lock_sock(sk);
194
195 err = -EINVAL;
196 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
197 goto out;
198
199 old_state = sk->sk_state;
200 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
201 goto out;
202
203 /* Really, if the socket is already in listen state
204 * we can only allow the backlog to be adjusted.
205 */
206 if (old_state != TCP_LISTEN) {
Arnaldo Carvalho de Melo0a5578c2005-08-09 20:11:41 -0700207 err = inet_csk_listen_start(sk, TCP_SYNQ_HSIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 if (err)
209 goto out;
210 }
211 sk->sk_max_ack_backlog = backlog;
212 err = 0;
213
214out:
215 release_sock(sk);
216 return err;
217}
218
219/*
220 * Create an inet socket.
221 */
222
223static int inet_create(struct socket *sock, int protocol)
224{
225 struct sock *sk;
226 struct list_head *p;
227 struct inet_protosw *answer;
228 struct inet_sock *inet;
229 struct proto *answer_prot;
230 unsigned char answer_flags;
231 char answer_no_check;
Arnaldo Carvalho de Melobb97d312005-08-09 20:19:14 -0700232 int try_loading_module = 0;
Herbert Xu86c8f9d2005-12-02 20:43:26 -0800233 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234
235 sock->state = SS_UNCONNECTED;
236
237 /* Look for the requested type/protocol pair. */
238 answer = NULL;
Arnaldo Carvalho de Melobb97d312005-08-09 20:19:14 -0700239lookup_protocol:
Herbert Xu86c8f9d2005-12-02 20:43:26 -0800240 err = -ESOCKTNOSUPPORT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 rcu_read_lock();
242 list_for_each_rcu(p, &inetsw[sock->type]) {
243 answer = list_entry(p, struct inet_protosw, list);
244
245 /* Check the non-wild match. */
246 if (protocol == answer->protocol) {
247 if (protocol != IPPROTO_IP)
248 break;
249 } else {
250 /* Check for the two wild cases. */
251 if (IPPROTO_IP == protocol) {
252 protocol = answer->protocol;
253 break;
254 }
255 if (IPPROTO_IP == answer->protocol)
256 break;
257 }
Herbert Xu86c8f9d2005-12-02 20:43:26 -0800258 err = -EPROTONOSUPPORT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 answer = NULL;
260 }
261
Arnaldo Carvalho de Melobb97d312005-08-09 20:19:14 -0700262 if (unlikely(answer == NULL)) {
263 if (try_loading_module < 2) {
264 rcu_read_unlock();
265 /*
266 * Be more specific, e.g. net-pf-2-proto-132-type-1
267 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
268 */
269 if (++try_loading_module == 1)
270 request_module("net-pf-%d-proto-%d-type-%d",
271 PF_INET, protocol, sock->type);
272 /*
273 * Fall back to generic, e.g. net-pf-2-proto-132
274 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
275 */
276 else
277 request_module("net-pf-%d-proto-%d",
278 PF_INET, protocol);
279 goto lookup_protocol;
280 } else
281 goto out_rcu_unlock;
282 }
283
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 err = -EPERM;
285 if (answer->capability > 0 && !capable(answer->capability))
286 goto out_rcu_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287
288 sock->ops = answer->ops;
289 answer_prot = answer->prot;
290 answer_no_check = answer->no_check;
291 answer_flags = answer->flags;
292 rcu_read_unlock();
293
294 BUG_TRAP(answer_prot->slab != NULL);
295
296 err = -ENOBUFS;
297 sk = sk_alloc(PF_INET, GFP_KERNEL, answer_prot, 1);
298 if (sk == NULL)
299 goto out;
300
301 err = 0;
302 sk->sk_no_check = answer_no_check;
303 if (INET_PROTOSW_REUSE & answer_flags)
304 sk->sk_reuse = 1;
305
306 inet = inet_sk(sk);
Arnaldo Carvalho de Melod83d8462005-12-13 23:26:10 -0800307 inet->is_icsk = INET_PROTOSW_ICSK & answer_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308
309 if (SOCK_RAW == sock->type) {
310 inet->num = protocol;
311 if (IPPROTO_RAW == protocol)
312 inet->hdrincl = 1;
313 }
314
315 if (ipv4_config.no_pmtu_disc)
316 inet->pmtudisc = IP_PMTUDISC_DONT;
317 else
318 inet->pmtudisc = IP_PMTUDISC_WANT;
319
320 inet->id = 0;
321
322 sock_init_data(sock, sk);
323
324 sk->sk_destruct = inet_sock_destruct;
325 sk->sk_family = PF_INET;
326 sk->sk_protocol = protocol;
327 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
328
329 inet->uc_ttl = -1;
330 inet->mc_loop = 1;
331 inet->mc_ttl = 1;
332 inet->mc_index = 0;
333 inet->mc_list = NULL;
334
Arnaldo Carvalho de Meloe6848972005-08-09 19:45:38 -0700335 sk_refcnt_debug_inc(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336
337 if (inet->num) {
338 /* It assumes that any protocol which allows
339 * the user to assign a number at socket
340 * creation time automatically
341 * shares.
342 */
343 inet->sport = htons(inet->num);
344 /* Add to protocol hash chains. */
345 sk->sk_prot->hash(sk);
346 }
347
348 if (sk->sk_prot->init) {
349 err = sk->sk_prot->init(sk);
350 if (err)
351 sk_common_release(sk);
352 }
353out:
354 return err;
355out_rcu_unlock:
356 rcu_read_unlock();
357 goto out;
358}
359
360
361/*
362 * The peer socket should always be NULL (or else). When we call this
363 * function we are destroying the object and from then on nobody
364 * should refer to it.
365 */
366int inet_release(struct socket *sock)
367{
368 struct sock *sk = sock->sk;
369
370 if (sk) {
371 long timeout;
372
373 /* Applications forget to leave groups before exiting */
374 ip_mc_drop_socket(sk);
375
376 /* If linger is set, we don't return until the close
377 * is complete. Otherwise we return immediately. The
378 * actually closing is done the same either way.
379 *
380 * If the close is due to the process exiting, we never
381 * linger..
382 */
383 timeout = 0;
384 if (sock_flag(sk, SOCK_LINGER) &&
385 !(current->flags & PF_EXITING))
386 timeout = sk->sk_lingertime;
387 sock->sk = NULL;
388 sk->sk_prot->close(sk, timeout);
389 }
390 return 0;
391}
392
393/* It is off by default, see below. */
394int sysctl_ip_nonlocal_bind;
395
396int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
397{
398 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
399 struct sock *sk = sock->sk;
400 struct inet_sock *inet = inet_sk(sk);
401 unsigned short snum;
402 int chk_addr_ret;
403 int err;
404
405 /* If the socket has its own bind function then use it. (RAW) */
406 if (sk->sk_prot->bind) {
407 err = sk->sk_prot->bind(sk, uaddr, addr_len);
408 goto out;
409 }
410 err = -EINVAL;
411 if (addr_len < sizeof(struct sockaddr_in))
412 goto out;
413
414 chk_addr_ret = inet_addr_type(addr->sin_addr.s_addr);
415
416 /* Not specified by any standard per-se, however it breaks too
417 * many applications when removed. It is unfortunate since
418 * allowing applications to make a non-local bind solves
419 * several problems with systems using dynamic addressing.
420 * (ie. your servers still start up even if your ISDN link
421 * is temporarily down)
422 */
423 err = -EADDRNOTAVAIL;
424 if (!sysctl_ip_nonlocal_bind &&
425 !inet->freebind &&
426 addr->sin_addr.s_addr != INADDR_ANY &&
427 chk_addr_ret != RTN_LOCAL &&
428 chk_addr_ret != RTN_MULTICAST &&
429 chk_addr_ret != RTN_BROADCAST)
430 goto out;
431
432 snum = ntohs(addr->sin_port);
433 err = -EACCES;
434 if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
435 goto out;
436
437 /* We keep a pair of addresses. rcv_saddr is the one
438 * used by hash lookups, and saddr is used for transmit.
439 *
440 * In the BSD API these are the same except where it
441 * would be illegal to use them (multicast/broadcast) in
442 * which case the sending device address is used.
443 */
444 lock_sock(sk);
445
446 /* Check these errors (active socket, double bind). */
447 err = -EINVAL;
448 if (sk->sk_state != TCP_CLOSE || inet->num)
449 goto out_release_sock;
450
451 inet->rcv_saddr = inet->saddr = addr->sin_addr.s_addr;
452 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
453 inet->saddr = 0; /* Use device */
454
455 /* Make sure we are allowed to bind here. */
456 if (sk->sk_prot->get_port(sk, snum)) {
457 inet->saddr = inet->rcv_saddr = 0;
458 err = -EADDRINUSE;
459 goto out_release_sock;
460 }
461
462 if (inet->rcv_saddr)
463 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
464 if (snum)
465 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
466 inet->sport = htons(inet->num);
467 inet->daddr = 0;
468 inet->dport = 0;
469 sk_dst_reset(sk);
470 err = 0;
471out_release_sock:
472 release_sock(sk);
473out:
474 return err;
475}
476
477int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
478 int addr_len, int flags)
479{
480 struct sock *sk = sock->sk;
481
482 if (uaddr->sa_family == AF_UNSPEC)
483 return sk->sk_prot->disconnect(sk, flags);
484
485 if (!inet_sk(sk)->num && inet_autobind(sk))
486 return -EAGAIN;
487 return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
488}
489
490static long inet_wait_for_connect(struct sock *sk, long timeo)
491{
492 DEFINE_WAIT(wait);
493
494 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
495
496 /* Basic assumption: if someone sets sk->sk_err, he _must_
497 * change state of the socket from TCP_SYN_*.
498 * Connect() does not allow to get error notifications
499 * without closing the socket.
500 */
501 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
502 release_sock(sk);
503 timeo = schedule_timeout(timeo);
504 lock_sock(sk);
505 if (signal_pending(current) || !timeo)
506 break;
507 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
508 }
509 finish_wait(sk->sk_sleep, &wait);
510 return timeo;
511}
512
513/*
514 * Connect to a remote host. There is regrettably still a little
515 * TCP 'magic' in here.
516 */
517int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
518 int addr_len, int flags)
519{
520 struct sock *sk = sock->sk;
521 int err;
522 long timeo;
523
524 lock_sock(sk);
525
526 if (uaddr->sa_family == AF_UNSPEC) {
527 err = sk->sk_prot->disconnect(sk, flags);
528 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
529 goto out;
530 }
531
532 switch (sock->state) {
533 default:
534 err = -EINVAL;
535 goto out;
536 case SS_CONNECTED:
537 err = -EISCONN;
538 goto out;
539 case SS_CONNECTING:
540 err = -EALREADY;
541 /* Fall out of switch with err, set for this state */
542 break;
543 case SS_UNCONNECTED:
544 err = -EISCONN;
545 if (sk->sk_state != TCP_CLOSE)
546 goto out;
547
548 err = sk->sk_prot->connect(sk, uaddr, addr_len);
549 if (err < 0)
550 goto out;
551
552 sock->state = SS_CONNECTING;
553
554 /* Just entered SS_CONNECTING state; the only
555 * difference is that return value in non-blocking
556 * case is EINPROGRESS, rather than EALREADY.
557 */
558 err = -EINPROGRESS;
559 break;
560 }
561
562 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
563
564 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
565 /* Error code is set above */
566 if (!timeo || !inet_wait_for_connect(sk, timeo))
567 goto out;
568
569 err = sock_intr_errno(timeo);
570 if (signal_pending(current))
571 goto out;
572 }
573
574 /* Connection was closed by RST, timeout, ICMP error
575 * or another process disconnected us.
576 */
577 if (sk->sk_state == TCP_CLOSE)
578 goto sock_error;
579
580 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
581 * and error was received after socket entered established state.
582 * Hence, it is handled normally after connect() return successfully.
583 */
584
585 sock->state = SS_CONNECTED;
586 err = 0;
587out:
588 release_sock(sk);
589 return err;
590
591sock_error:
592 err = sock_error(sk) ? : -ECONNABORTED;
593 sock->state = SS_UNCONNECTED;
594 if (sk->sk_prot->disconnect(sk, flags))
595 sock->state = SS_DISCONNECTING;
596 goto out;
597}
598
599/*
600 * Accept a pending connection. The TCP layer now gives BSD semantics.
601 */
602
603int inet_accept(struct socket *sock, struct socket *newsock, int flags)
604{
605 struct sock *sk1 = sock->sk;
606 int err = -EINVAL;
607 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
608
609 if (!sk2)
610 goto do_err;
611
612 lock_sock(sk2);
613
614 BUG_TRAP((1 << sk2->sk_state) &
615 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE));
616
617 sock_graft(sk2, newsock);
618
619 newsock->state = SS_CONNECTED;
620 err = 0;
621 release_sock(sk2);
622do_err:
623 return err;
624}
625
626
627/*
628 * This does both peername and sockname.
629 */
630int inet_getname(struct socket *sock, struct sockaddr *uaddr,
631 int *uaddr_len, int peer)
632{
633 struct sock *sk = sock->sk;
634 struct inet_sock *inet = inet_sk(sk);
635 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
636
637 sin->sin_family = AF_INET;
638 if (peer) {
639 if (!inet->dport ||
640 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
641 peer == 1))
642 return -ENOTCONN;
643 sin->sin_port = inet->dport;
644 sin->sin_addr.s_addr = inet->daddr;
645 } else {
646 __u32 addr = inet->rcv_saddr;
647 if (!addr)
648 addr = inet->saddr;
649 sin->sin_port = inet->sport;
650 sin->sin_addr.s_addr = addr;
651 }
652 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
653 *uaddr_len = sizeof(*sin);
654 return 0;
655}
656
657int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
658 size_t size)
659{
660 struct sock *sk = sock->sk;
661
662 /* We may need to bind the socket. */
663 if (!inet_sk(sk)->num && inet_autobind(sk))
664 return -EAGAIN;
665
666 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
667}
668
669
670static ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
671{
672 struct sock *sk = sock->sk;
673
674 /* We may need to bind the socket. */
675 if (!inet_sk(sk)->num && inet_autobind(sk))
676 return -EAGAIN;
677
678 if (sk->sk_prot->sendpage)
679 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
680 return sock_no_sendpage(sock, page, offset, size, flags);
681}
682
683
684int inet_shutdown(struct socket *sock, int how)
685{
686 struct sock *sk = sock->sk;
687 int err = 0;
688
689 /* This should really check to make sure
690 * the socket is a TCP socket. (WHY AC...)
691 */
692 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
693 1->2 bit 2 snds.
694 2->3 */
695 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
696 return -EINVAL;
697
698 lock_sock(sk);
699 if (sock->state == SS_CONNECTING) {
700 if ((1 << sk->sk_state) &
701 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
702 sock->state = SS_DISCONNECTING;
703 else
704 sock->state = SS_CONNECTED;
705 }
706
707 switch (sk->sk_state) {
708 case TCP_CLOSE:
709 err = -ENOTCONN;
710 /* Hack to wake up other listeners, who can poll for
711 POLLHUP, even on eg. unconnected UDP sockets -- RR */
712 default:
713 sk->sk_shutdown |= how;
714 if (sk->sk_prot->shutdown)
715 sk->sk_prot->shutdown(sk, how);
716 break;
717
718 /* Remaining two branches are temporary solution for missing
719 * close() in multithreaded environment. It is _not_ a good idea,
720 * but we have no choice until close() is repaired at VFS level.
721 */
722 case TCP_LISTEN:
723 if (!(how & RCV_SHUTDOWN))
724 break;
725 /* Fall through */
726 case TCP_SYN_SENT:
727 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
728 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
729 break;
730 }
731
732 /* Wake up anyone sleeping in poll. */
733 sk->sk_state_change(sk);
734 release_sock(sk);
735 return err;
736}
737
738/*
739 * ioctl() calls you can issue on an INET socket. Most of these are
740 * device configuration and stuff and very rarely used. Some ioctls
741 * pass on to the socket itself.
742 *
743 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
744 * loads the devconfigure module does its configuring and unloads it.
745 * There's a good 20K of config code hanging around the kernel.
746 */
747
748int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
749{
750 struct sock *sk = sock->sk;
751 int err = 0;
752
753 switch (cmd) {
754 case SIOCGSTAMP:
755 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
756 break;
757 case SIOCADDRT:
758 case SIOCDELRT:
759 case SIOCRTMSG:
760 err = ip_rt_ioctl(cmd, (void __user *)arg);
761 break;
762 case SIOCDARP:
763 case SIOCGARP:
764 case SIOCSARP:
765 err = arp_ioctl(cmd, (void __user *)arg);
766 break;
767 case SIOCGIFADDR:
768 case SIOCSIFADDR:
769 case SIOCGIFBRDADDR:
770 case SIOCSIFBRDADDR:
771 case SIOCGIFNETMASK:
772 case SIOCSIFNETMASK:
773 case SIOCGIFDSTADDR:
774 case SIOCSIFDSTADDR:
775 case SIOCSIFPFLAGS:
776 case SIOCGIFPFLAGS:
777 case SIOCSIFFLAGS:
778 err = devinet_ioctl(cmd, (void __user *)arg);
779 break;
780 default:
Christoph Hellwigb5e5fa52006-01-03 14:18:33 -0800781 if (sk->sk_prot->ioctl)
782 err = sk->sk_prot->ioctl(sk, cmd, arg);
783 else
784 err = -ENOIOCTLCMD;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 break;
786 }
787 return err;
788}
789
Eric Dumazet90ddc4f2005-12-22 12:49:22 -0800790const struct proto_ops inet_stream_ops = {
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -0800791 .family = PF_INET,
792 .owner = THIS_MODULE,
793 .release = inet_release,
794 .bind = inet_bind,
795 .connect = inet_stream_connect,
796 .socketpair = sock_no_socketpair,
797 .accept = inet_accept,
798 .getname = inet_getname,
799 .poll = tcp_poll,
800 .ioctl = inet_ioctl,
801 .listen = inet_listen,
802 .shutdown = inet_shutdown,
803 .setsockopt = sock_common_setsockopt,
804 .getsockopt = sock_common_getsockopt,
805 .sendmsg = inet_sendmsg,
806 .recvmsg = sock_common_recvmsg,
807 .mmap = sock_no_mmap,
808 .sendpage = tcp_sendpage,
Dmitry Mishin3fdadf72006-03-20 22:45:21 -0800809#ifdef CONFIG_COMPAT
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -0800810 .compat_setsockopt = compat_sock_common_setsockopt,
811 .compat_getsockopt = compat_sock_common_getsockopt,
Dmitry Mishin3fdadf72006-03-20 22:45:21 -0800812#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813};
814
Eric Dumazet90ddc4f2005-12-22 12:49:22 -0800815const struct proto_ops inet_dgram_ops = {
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -0800816 .family = PF_INET,
817 .owner = THIS_MODULE,
818 .release = inet_release,
819 .bind = inet_bind,
820 .connect = inet_dgram_connect,
821 .socketpair = sock_no_socketpair,
822 .accept = sock_no_accept,
823 .getname = inet_getname,
824 .poll = udp_poll,
825 .ioctl = inet_ioctl,
826 .listen = sock_no_listen,
827 .shutdown = inet_shutdown,
828 .setsockopt = sock_common_setsockopt,
829 .getsockopt = sock_common_getsockopt,
830 .sendmsg = inet_sendmsg,
831 .recvmsg = sock_common_recvmsg,
832 .mmap = sock_no_mmap,
833 .sendpage = inet_sendpage,
Dmitry Mishin3fdadf72006-03-20 22:45:21 -0800834#ifdef CONFIG_COMPAT
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -0800835 .compat_setsockopt = compat_sock_common_setsockopt,
836 .compat_getsockopt = compat_sock_common_getsockopt,
Dmitry Mishin3fdadf72006-03-20 22:45:21 -0800837#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838};
839
840/*
841 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
842 * udp_poll
843 */
Eric Dumazet90ddc4f2005-12-22 12:49:22 -0800844static const struct proto_ops inet_sockraw_ops = {
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -0800845 .family = PF_INET,
846 .owner = THIS_MODULE,
847 .release = inet_release,
848 .bind = inet_bind,
849 .connect = inet_dgram_connect,
850 .socketpair = sock_no_socketpair,
851 .accept = sock_no_accept,
852 .getname = inet_getname,
853 .poll = datagram_poll,
854 .ioctl = inet_ioctl,
855 .listen = sock_no_listen,
856 .shutdown = inet_shutdown,
857 .setsockopt = sock_common_setsockopt,
858 .getsockopt = sock_common_getsockopt,
859 .sendmsg = inet_sendmsg,
860 .recvmsg = sock_common_recvmsg,
861 .mmap = sock_no_mmap,
862 .sendpage = inet_sendpage,
Dmitry Mishin3fdadf72006-03-20 22:45:21 -0800863#ifdef CONFIG_COMPAT
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -0800864 .compat_setsockopt = compat_sock_common_setsockopt,
865 .compat_getsockopt = compat_sock_common_getsockopt,
Dmitry Mishin3fdadf72006-03-20 22:45:21 -0800866#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867};
868
869static struct net_proto_family inet_family_ops = {
870 .family = PF_INET,
871 .create = inet_create,
872 .owner = THIS_MODULE,
873};
874
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875/* Upon startup we insert all the elements in inetsw_array[] into
876 * the linked list inetsw.
877 */
878static struct inet_protosw inetsw_array[] =
879{
880 {
881 .type = SOCK_STREAM,
882 .protocol = IPPROTO_TCP,
883 .prot = &tcp_prot,
884 .ops = &inet_stream_ops,
885 .capability = -1,
886 .no_check = 0,
Arnaldo Carvalho de Melod83d8462005-12-13 23:26:10 -0800887 .flags = INET_PROTOSW_PERMANENT |
888 INET_PROTOSW_ICSK,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 },
890
891 {
892 .type = SOCK_DGRAM,
893 .protocol = IPPROTO_UDP,
894 .prot = &udp_prot,
895 .ops = &inet_dgram_ops,
896 .capability = -1,
897 .no_check = UDP_CSUM_DEFAULT,
898 .flags = INET_PROTOSW_PERMANENT,
899 },
900
901
902 {
903 .type = SOCK_RAW,
904 .protocol = IPPROTO_IP, /* wild card */
905 .prot = &raw_prot,
906 .ops = &inet_sockraw_ops,
907 .capability = CAP_NET_RAW,
908 .no_check = UDP_CSUM_DEFAULT,
909 .flags = INET_PROTOSW_REUSE,
910 }
911};
912
913#define INETSW_ARRAY_LEN (sizeof(inetsw_array) / sizeof(struct inet_protosw))
914
915void inet_register_protosw(struct inet_protosw *p)
916{
917 struct list_head *lh;
918 struct inet_protosw *answer;
919 int protocol = p->protocol;
920 struct list_head *last_perm;
921
922 spin_lock_bh(&inetsw_lock);
923
924 if (p->type >= SOCK_MAX)
925 goto out_illegal;
926
927 /* If we are trying to override a permanent protocol, bail. */
928 answer = NULL;
929 last_perm = &inetsw[p->type];
930 list_for_each(lh, &inetsw[p->type]) {
931 answer = list_entry(lh, struct inet_protosw, list);
932
933 /* Check only the non-wild match. */
934 if (INET_PROTOSW_PERMANENT & answer->flags) {
935 if (protocol == answer->protocol)
936 break;
937 last_perm = lh;
938 }
939
940 answer = NULL;
941 }
942 if (answer)
943 goto out_permanent;
944
945 /* Add the new entry after the last permanent entry if any, so that
946 * the new entry does not override a permanent entry when matched with
947 * a wild-card protocol. But it is allowed to override any existing
948 * non-permanent entry. This means that when we remove this entry, the
949 * system automatically returns to the old behavior.
950 */
951 list_add_rcu(&p->list, last_perm);
952out:
953 spin_unlock_bh(&inetsw_lock);
954
955 synchronize_net();
956
957 return;
958
959out_permanent:
960 printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
961 protocol);
962 goto out;
963
964out_illegal:
965 printk(KERN_ERR
966 "Ignoring attempt to register invalid socket type %d.\n",
967 p->type);
968 goto out;
969}
970
971void inet_unregister_protosw(struct inet_protosw *p)
972{
973 if (INET_PROTOSW_PERMANENT & p->flags) {
974 printk(KERN_ERR
975 "Attempt to unregister permanent protocol %d.\n",
976 p->protocol);
977 } else {
978 spin_lock_bh(&inetsw_lock);
979 list_del_rcu(&p->list);
980 spin_unlock_bh(&inetsw_lock);
981
982 synchronize_net();
983 }
984}
985
Arnaldo Carvalho de Melo32519f12005-08-09 19:50:02 -0700986/*
987 * Shall we try to damage output packets if routing dev changes?
988 */
989
990int sysctl_ip_dynaddr;
991
992static int inet_sk_reselect_saddr(struct sock *sk)
993{
994 struct inet_sock *inet = inet_sk(sk);
995 int err;
996 struct rtable *rt;
997 __u32 old_saddr = inet->saddr;
998 __u32 new_saddr;
999 __u32 daddr = inet->daddr;
1000
1001 if (inet->opt && inet->opt->srr)
1002 daddr = inet->opt->faddr;
1003
1004 /* Query new route. */
1005 err = ip_route_connect(&rt, daddr, 0,
1006 RT_CONN_FLAGS(sk),
1007 sk->sk_bound_dev_if,
1008 sk->sk_protocol,
1009 inet->sport, inet->dport, sk);
1010 if (err)
1011 return err;
1012
1013 sk_setup_caps(sk, &rt->u.dst);
1014
1015 new_saddr = rt->rt_src;
1016
1017 if (new_saddr == old_saddr)
1018 return 0;
1019
1020 if (sysctl_ip_dynaddr > 1) {
1021 printk(KERN_INFO "%s(): shifting inet->"
1022 "saddr from %d.%d.%d.%d to %d.%d.%d.%d\n",
1023 __FUNCTION__,
1024 NIPQUAD(old_saddr),
1025 NIPQUAD(new_saddr));
1026 }
1027
1028 inet->saddr = inet->rcv_saddr = new_saddr;
1029
1030 /*
1031 * XXX The only one ugly spot where we need to
1032 * XXX really change the sockets identity after
1033 * XXX it has entered the hashes. -DaveM
1034 *
1035 * Besides that, it does not check for connection
1036 * uniqueness. Wait for troubles.
1037 */
1038 __sk_prot_rehash(sk);
1039 return 0;
1040}
1041
1042int inet_sk_rebuild_header(struct sock *sk)
1043{
1044 struct inet_sock *inet = inet_sk(sk);
1045 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1046 u32 daddr;
1047 int err;
1048
1049 /* Route is OK, nothing to do. */
1050 if (rt)
1051 return 0;
1052
1053 /* Reroute. */
1054 daddr = inet->daddr;
1055 if (inet->opt && inet->opt->srr)
1056 daddr = inet->opt->faddr;
1057{
1058 struct flowi fl = {
1059 .oif = sk->sk_bound_dev_if,
1060 .nl_u = {
1061 .ip4_u = {
1062 .daddr = daddr,
1063 .saddr = inet->saddr,
1064 .tos = RT_CONN_FLAGS(sk),
1065 },
1066 },
1067 .proto = sk->sk_protocol,
1068 .uli_u = {
1069 .ports = {
1070 .sport = inet->sport,
1071 .dport = inet->dport,
1072 },
1073 },
1074 };
1075
1076 err = ip_route_output_flow(&rt, &fl, sk, 0);
1077}
1078 if (!err)
1079 sk_setup_caps(sk, &rt->u.dst);
1080 else {
1081 /* Routing failed... */
1082 sk->sk_route_caps = 0;
1083 /*
1084 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1085 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1086 */
1087 if (!sysctl_ip_dynaddr ||
1088 sk->sk_state != TCP_SYN_SENT ||
1089 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1090 (err = inet_sk_reselect_saddr(sk)) != 0)
1091 sk->sk_err_soft = -err;
1092 }
1093
1094 return err;
1095}
1096
1097EXPORT_SYMBOL(inet_sk_rebuild_header);
1098
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099#ifdef CONFIG_IP_MULTICAST
1100static struct net_protocol igmp_protocol = {
1101 .handler = igmp_rcv,
1102};
1103#endif
1104
1105static struct net_protocol tcp_protocol = {
1106 .handler = tcp_v4_rcv,
1107 .err_handler = tcp_v4_err,
1108 .no_policy = 1,
1109};
1110
1111static struct net_protocol udp_protocol = {
1112 .handler = udp_rcv,
1113 .err_handler = udp_err,
1114 .no_policy = 1,
1115};
1116
1117static struct net_protocol icmp_protocol = {
1118 .handler = icmp_rcv,
1119};
1120
1121static int __init init_ipv4_mibs(void)
1122{
1123 net_statistics[0] = alloc_percpu(struct linux_mib);
1124 net_statistics[1] = alloc_percpu(struct linux_mib);
1125 ip_statistics[0] = alloc_percpu(struct ipstats_mib);
1126 ip_statistics[1] = alloc_percpu(struct ipstats_mib);
1127 icmp_statistics[0] = alloc_percpu(struct icmp_mib);
1128 icmp_statistics[1] = alloc_percpu(struct icmp_mib);
1129 tcp_statistics[0] = alloc_percpu(struct tcp_mib);
1130 tcp_statistics[1] = alloc_percpu(struct tcp_mib);
1131 udp_statistics[0] = alloc_percpu(struct udp_mib);
1132 udp_statistics[1] = alloc_percpu(struct udp_mib);
1133 if (!
1134 (net_statistics[0] && net_statistics[1] && ip_statistics[0]
1135 && ip_statistics[1] && tcp_statistics[0] && tcp_statistics[1]
1136 && udp_statistics[0] && udp_statistics[1]))
1137 return -ENOMEM;
1138
1139 (void) tcp_mib_init();
1140
1141 return 0;
1142}
1143
1144static int ipv4_proc_init(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145
Herbert Xu30e224d2005-07-05 14:40:10 -07001146/*
1147 * IP protocol layer initialiser
1148 */
1149
1150static struct packet_type ip_packet_type = {
1151 .type = __constant_htons(ETH_P_IP),
1152 .func = ip_rcv,
1153};
1154
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155static int __init inet_init(void)
1156{
1157 struct sk_buff *dummy_skb;
1158 struct inet_protosw *q;
1159 struct list_head *r;
1160 int rc = -EINVAL;
1161
1162 if (sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb)) {
1163 printk(KERN_CRIT "%s: panic\n", __FUNCTION__);
1164 goto out;
1165 }
1166
1167 rc = proto_register(&tcp_prot, 1);
1168 if (rc)
1169 goto out;
1170
1171 rc = proto_register(&udp_prot, 1);
1172 if (rc)
1173 goto out_unregister_tcp_proto;
1174
1175 rc = proto_register(&raw_prot, 1);
1176 if (rc)
1177 goto out_unregister_udp_proto;
1178
1179 /*
1180 * Tell SOCKET that we are alive...
1181 */
1182
1183 (void)sock_register(&inet_family_ops);
1184
1185 /*
1186 * Add all the base protocols.
1187 */
1188
1189 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1190 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1191 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1192 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1193 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1194 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1195#ifdef CONFIG_IP_MULTICAST
1196 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1197 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1198#endif
1199
1200 /* Register the socket-side information for inet_create. */
1201 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1202 INIT_LIST_HEAD(r);
1203
1204 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1205 inet_register_protosw(q);
1206
1207 /*
1208 * Set the ARP module up
1209 */
1210
1211 arp_init();
1212
1213 /*
1214 * Set the IP module up
1215 */
1216
1217 ip_init();
1218
1219 tcp_v4_init(&inet_family_ops);
1220
1221 /* Setup TCP slab cache for open requests. */
1222 tcp_init();
1223
1224
1225 /*
1226 * Set the ICMP layer up
1227 */
1228
1229 icmp_init(&inet_family_ops);
1230
1231 /*
1232 * Initialise the multicast router
1233 */
1234#if defined(CONFIG_IP_MROUTE)
1235 ip_mr_init();
1236#endif
1237 /*
1238 * Initialise per-cpu ipv4 mibs
1239 */
1240
1241 if(init_ipv4_mibs())
1242 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n"); ;
1243
1244 ipv4_proc_init();
1245
1246 ipfrag_init();
1247
Herbert Xu30e224d2005-07-05 14:40:10 -07001248 dev_add_pack(&ip_packet_type);
1249
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 rc = 0;
1251out:
1252 return rc;
1253out_unregister_tcp_proto:
1254 proto_unregister(&tcp_prot);
1255out_unregister_udp_proto:
1256 proto_unregister(&udp_prot);
1257 goto out;
1258}
1259
Heiko Carstensa536e072006-04-28 15:19:17 -07001260fs_initcall(inet_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261
1262/* ------------------------------------------------------------------------ */
1263
1264#ifdef CONFIG_PROC_FS
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265static int __init ipv4_proc_init(void)
1266{
1267 int rc = 0;
1268
1269 if (raw_proc_init())
1270 goto out_raw;
1271 if (tcp4_proc_init())
1272 goto out_tcp;
1273 if (udp4_proc_init())
1274 goto out_udp;
1275 if (fib_proc_init())
1276 goto out_fib;
1277 if (ip_misc_proc_init())
1278 goto out_misc;
1279out:
1280 return rc;
1281out_misc:
1282 fib_proc_exit();
1283out_fib:
1284 udp4_proc_exit();
1285out_udp:
1286 tcp4_proc_exit();
1287out_tcp:
1288 raw_proc_exit();
1289out_raw:
1290 rc = -ENOMEM;
1291 goto out;
1292}
1293
1294#else /* CONFIG_PROC_FS */
1295static int __init ipv4_proc_init(void)
1296{
1297 return 0;
1298}
1299#endif /* CONFIG_PROC_FS */
1300
1301MODULE_ALIAS_NETPROTO(PF_INET);
1302
1303EXPORT_SYMBOL(inet_accept);
1304EXPORT_SYMBOL(inet_bind);
1305EXPORT_SYMBOL(inet_dgram_connect);
1306EXPORT_SYMBOL(inet_dgram_ops);
1307EXPORT_SYMBOL(inet_getname);
1308EXPORT_SYMBOL(inet_ioctl);
1309EXPORT_SYMBOL(inet_listen);
1310EXPORT_SYMBOL(inet_register_protosw);
1311EXPORT_SYMBOL(inet_release);
1312EXPORT_SYMBOL(inet_sendmsg);
1313EXPORT_SYMBOL(inet_shutdown);
1314EXPORT_SYMBOL(inet_sock_destruct);
1315EXPORT_SYMBOL(inet_stream_connect);
1316EXPORT_SYMBOL(inet_stream_ops);
1317EXPORT_SYMBOL(inet_unregister_protosw);
1318EXPORT_SYMBOL(net_statistics);
Neil Hormancdac4e02005-06-13 15:12:33 -07001319EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);