blob: 4ed8a814c6cbc73214897bae499ab0949651d4d8 [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>
82#include <linux/fcntl.h>
83#include <linux/mm.h>
84#include <linux/interrupt.h>
85#include <linux/stat.h>
86#include <linux/init.h>
87#include <linux/poll.h>
88#include <linux/netfilter_ipv4.h>
89
90#include <asm/uaccess.h>
91#include <asm/system.h>
92
93#include <linux/smp_lock.h>
94#include <linux/inet.h>
95#include <linux/igmp.h>
96#include <linux/netdevice.h>
97#include <net/ip.h>
98#include <net/protocol.h>
99#include <net/arp.h>
100#include <net/route.h>
101#include <net/ip_fib.h>
Arnaldo Carvalho de Melo295f7322005-08-09 20:11:56 -0700102#include <net/inet_connection_sock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103#include <net/tcp.h>
104#include <net/udp.h>
105#include <linux/skbuff.h>
106#include <net/sock.h>
107#include <net/raw.h>
108#include <net/icmp.h>
109#include <net/ipip.h>
110#include <net/inet_common.h>
111#include <net/xfrm.h>
112#ifdef CONFIG_IP_MROUTE
113#include <linux/mroute.h>
114#endif
115
Eric Dumazetba899662005-08-26 12:05:31 -0700116DEFINE_SNMP_STAT(struct linux_mib, net_statistics) __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118extern void ip_mc_drop_socket(struct sock *sk);
119
120/* The inetsw table contains everything that inet_create needs to
121 * build a new socket.
122 */
123static struct list_head inetsw[SOCK_MAX];
124static DEFINE_SPINLOCK(inetsw_lock);
125
126/* New destruction routine */
127
128void inet_sock_destruct(struct sock *sk)
129{
130 struct inet_sock *inet = inet_sk(sk);
131
132 __skb_queue_purge(&sk->sk_receive_queue);
133 __skb_queue_purge(&sk->sk_error_queue);
134
135 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
136 printk("Attempt to release TCP socket in state %d %p\n",
137 sk->sk_state, sk);
138 return;
139 }
140 if (!sock_flag(sk, SOCK_DEAD)) {
141 printk("Attempt to release alive inet socket %p\n", sk);
142 return;
143 }
144
145 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
146 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
147 BUG_TRAP(!sk->sk_wmem_queued);
148 BUG_TRAP(!sk->sk_forward_alloc);
149
Jesper Juhla51482b2005-11-08 09:41:34 -0800150 kfree(inet->opt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151 dst_release(sk->sk_dst_cache);
Arnaldo Carvalho de Meloe6848972005-08-09 19:45:38 -0700152 sk_refcnt_debug_dec(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153}
154
155/*
156 * The routines beyond this point handle the behaviour of an AF_INET
157 * socket object. Mostly it punts to the subprotocols of IP to do
158 * the work.
159 */
160
161/*
162 * Automatically bind an unbound socket.
163 */
164
165static int inet_autobind(struct sock *sk)
166{
167 struct inet_sock *inet;
168 /* We may need to bind the socket. */
169 lock_sock(sk);
170 inet = inet_sk(sk);
171 if (!inet->num) {
172 if (sk->sk_prot->get_port(sk, 0)) {
173 release_sock(sk);
174 return -EAGAIN;
175 }
176 inet->sport = htons(inet->num);
177 }
178 release_sock(sk);
179 return 0;
180}
181
182/*
183 * Move a socket into listening state.
184 */
185int inet_listen(struct socket *sock, int backlog)
186{
187 struct sock *sk = sock->sk;
188 unsigned char old_state;
189 int err;
190
191 lock_sock(sk);
192
193 err = -EINVAL;
194 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
195 goto out;
196
197 old_state = sk->sk_state;
198 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
199 goto out;
200
201 /* Really, if the socket is already in listen state
202 * we can only allow the backlog to be adjusted.
203 */
204 if (old_state != TCP_LISTEN) {
Arnaldo Carvalho de Melo0a5578c2005-08-09 20:11:41 -0700205 err = inet_csk_listen_start(sk, TCP_SYNQ_HSIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 if (err)
207 goto out;
208 }
209 sk->sk_max_ack_backlog = backlog;
210 err = 0;
211
212out:
213 release_sock(sk);
214 return err;
215}
216
217/*
218 * Create an inet socket.
219 */
220
221static int inet_create(struct socket *sock, int protocol)
222{
223 struct sock *sk;
224 struct list_head *p;
225 struct inet_protosw *answer;
226 struct inet_sock *inet;
227 struct proto *answer_prot;
228 unsigned char answer_flags;
229 char answer_no_check;
Arnaldo Carvalho de Melobb97d312005-08-09 20:19:14 -0700230 int try_loading_module = 0;
Herbert Xu86c8f9d2005-12-02 20:43:26 -0800231 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232
233 sock->state = SS_UNCONNECTED;
234
235 /* Look for the requested type/protocol pair. */
236 answer = NULL;
Arnaldo Carvalho de Melobb97d312005-08-09 20:19:14 -0700237lookup_protocol:
Herbert Xu86c8f9d2005-12-02 20:43:26 -0800238 err = -ESOCKTNOSUPPORT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 rcu_read_lock();
240 list_for_each_rcu(p, &inetsw[sock->type]) {
241 answer = list_entry(p, struct inet_protosw, list);
242
243 /* Check the non-wild match. */
244 if (protocol == answer->protocol) {
245 if (protocol != IPPROTO_IP)
246 break;
247 } else {
248 /* Check for the two wild cases. */
249 if (IPPROTO_IP == protocol) {
250 protocol = answer->protocol;
251 break;
252 }
253 if (IPPROTO_IP == answer->protocol)
254 break;
255 }
Herbert Xu86c8f9d2005-12-02 20:43:26 -0800256 err = -EPROTONOSUPPORT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 answer = NULL;
258 }
259
Arnaldo Carvalho de Melobb97d312005-08-09 20:19:14 -0700260 if (unlikely(answer == NULL)) {
261 if (try_loading_module < 2) {
262 rcu_read_unlock();
263 /*
264 * Be more specific, e.g. net-pf-2-proto-132-type-1
265 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
266 */
267 if (++try_loading_module == 1)
268 request_module("net-pf-%d-proto-%d-type-%d",
269 PF_INET, protocol, sock->type);
270 /*
271 * Fall back to generic, e.g. net-pf-2-proto-132
272 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
273 */
274 else
275 request_module("net-pf-%d-proto-%d",
276 PF_INET, protocol);
277 goto lookup_protocol;
278 } else
279 goto out_rcu_unlock;
280 }
281
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 err = -EPERM;
283 if (answer->capability > 0 && !capable(answer->capability))
284 goto out_rcu_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285
286 sock->ops = answer->ops;
287 answer_prot = answer->prot;
288 answer_no_check = answer->no_check;
289 answer_flags = answer->flags;
290 rcu_read_unlock();
291
292 BUG_TRAP(answer_prot->slab != NULL);
293
294 err = -ENOBUFS;
295 sk = sk_alloc(PF_INET, GFP_KERNEL, answer_prot, 1);
296 if (sk == NULL)
297 goto out;
298
299 err = 0;
300 sk->sk_no_check = answer_no_check;
301 if (INET_PROTOSW_REUSE & answer_flags)
302 sk->sk_reuse = 1;
303
304 inet = inet_sk(sk);
Arnaldo Carvalho de Melod83d8462005-12-13 23:26:10 -0800305 inet->is_icsk = INET_PROTOSW_ICSK & answer_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306
307 if (SOCK_RAW == sock->type) {
308 inet->num = protocol;
309 if (IPPROTO_RAW == protocol)
310 inet->hdrincl = 1;
311 }
312
313 if (ipv4_config.no_pmtu_disc)
314 inet->pmtudisc = IP_PMTUDISC_DONT;
315 else
316 inet->pmtudisc = IP_PMTUDISC_WANT;
317
318 inet->id = 0;
319
320 sock_init_data(sock, sk);
321
322 sk->sk_destruct = inet_sock_destruct;
323 sk->sk_family = PF_INET;
324 sk->sk_protocol = protocol;
325 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
326
327 inet->uc_ttl = -1;
328 inet->mc_loop = 1;
329 inet->mc_ttl = 1;
330 inet->mc_index = 0;
331 inet->mc_list = NULL;
332
Arnaldo Carvalho de Meloe6848972005-08-09 19:45:38 -0700333 sk_refcnt_debug_inc(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334
335 if (inet->num) {
336 /* It assumes that any protocol which allows
337 * the user to assign a number at socket
338 * creation time automatically
339 * shares.
340 */
341 inet->sport = htons(inet->num);
342 /* Add to protocol hash chains. */
343 sk->sk_prot->hash(sk);
344 }
345
346 if (sk->sk_prot->init) {
347 err = sk->sk_prot->init(sk);
348 if (err)
349 sk_common_release(sk);
350 }
351out:
352 return err;
353out_rcu_unlock:
354 rcu_read_unlock();
355 goto out;
356}
357
358
359/*
360 * The peer socket should always be NULL (or else). When we call this
361 * function we are destroying the object and from then on nobody
362 * should refer to it.
363 */
364int inet_release(struct socket *sock)
365{
366 struct sock *sk = sock->sk;
367
368 if (sk) {
369 long timeout;
370
371 /* Applications forget to leave groups before exiting */
372 ip_mc_drop_socket(sk);
373
374 /* If linger is set, we don't return until the close
375 * is complete. Otherwise we return immediately. The
376 * actually closing is done the same either way.
377 *
378 * If the close is due to the process exiting, we never
379 * linger..
380 */
381 timeout = 0;
382 if (sock_flag(sk, SOCK_LINGER) &&
383 !(current->flags & PF_EXITING))
384 timeout = sk->sk_lingertime;
385 sock->sk = NULL;
386 sk->sk_prot->close(sk, timeout);
387 }
388 return 0;
389}
390
391/* It is off by default, see below. */
392int sysctl_ip_nonlocal_bind;
393
394int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
395{
396 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
397 struct sock *sk = sock->sk;
398 struct inet_sock *inet = inet_sk(sk);
399 unsigned short snum;
400 int chk_addr_ret;
401 int err;
402
403 /* If the socket has its own bind function then use it. (RAW) */
404 if (sk->sk_prot->bind) {
405 err = sk->sk_prot->bind(sk, uaddr, addr_len);
406 goto out;
407 }
408 err = -EINVAL;
409 if (addr_len < sizeof(struct sockaddr_in))
410 goto out;
411
412 chk_addr_ret = inet_addr_type(addr->sin_addr.s_addr);
413
414 /* Not specified by any standard per-se, however it breaks too
415 * many applications when removed. It is unfortunate since
416 * allowing applications to make a non-local bind solves
417 * several problems with systems using dynamic addressing.
418 * (ie. your servers still start up even if your ISDN link
419 * is temporarily down)
420 */
421 err = -EADDRNOTAVAIL;
422 if (!sysctl_ip_nonlocal_bind &&
423 !inet->freebind &&
424 addr->sin_addr.s_addr != INADDR_ANY &&
425 chk_addr_ret != RTN_LOCAL &&
426 chk_addr_ret != RTN_MULTICAST &&
427 chk_addr_ret != RTN_BROADCAST)
428 goto out;
429
430 snum = ntohs(addr->sin_port);
431 err = -EACCES;
432 if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
433 goto out;
434
435 /* We keep a pair of addresses. rcv_saddr is the one
436 * used by hash lookups, and saddr is used for transmit.
437 *
438 * In the BSD API these are the same except where it
439 * would be illegal to use them (multicast/broadcast) in
440 * which case the sending device address is used.
441 */
442 lock_sock(sk);
443
444 /* Check these errors (active socket, double bind). */
445 err = -EINVAL;
446 if (sk->sk_state != TCP_CLOSE || inet->num)
447 goto out_release_sock;
448
449 inet->rcv_saddr = inet->saddr = addr->sin_addr.s_addr;
450 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
451 inet->saddr = 0; /* Use device */
452
453 /* Make sure we are allowed to bind here. */
454 if (sk->sk_prot->get_port(sk, snum)) {
455 inet->saddr = inet->rcv_saddr = 0;
456 err = -EADDRINUSE;
457 goto out_release_sock;
458 }
459
460 if (inet->rcv_saddr)
461 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
462 if (snum)
463 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
464 inet->sport = htons(inet->num);
465 inet->daddr = 0;
466 inet->dport = 0;
467 sk_dst_reset(sk);
468 err = 0;
469out_release_sock:
470 release_sock(sk);
471out:
472 return err;
473}
474
475int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
476 int addr_len, int flags)
477{
478 struct sock *sk = sock->sk;
479
480 if (uaddr->sa_family == AF_UNSPEC)
481 return sk->sk_prot->disconnect(sk, flags);
482
483 if (!inet_sk(sk)->num && inet_autobind(sk))
484 return -EAGAIN;
485 return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
486}
487
488static long inet_wait_for_connect(struct sock *sk, long timeo)
489{
490 DEFINE_WAIT(wait);
491
492 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
493
494 /* Basic assumption: if someone sets sk->sk_err, he _must_
495 * change state of the socket from TCP_SYN_*.
496 * Connect() does not allow to get error notifications
497 * without closing the socket.
498 */
499 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
500 release_sock(sk);
501 timeo = schedule_timeout(timeo);
502 lock_sock(sk);
503 if (signal_pending(current) || !timeo)
504 break;
505 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
506 }
507 finish_wait(sk->sk_sleep, &wait);
508 return timeo;
509}
510
511/*
512 * Connect to a remote host. There is regrettably still a little
513 * TCP 'magic' in here.
514 */
515int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
516 int addr_len, int flags)
517{
518 struct sock *sk = sock->sk;
519 int err;
520 long timeo;
521
522 lock_sock(sk);
523
524 if (uaddr->sa_family == AF_UNSPEC) {
525 err = sk->sk_prot->disconnect(sk, flags);
526 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
527 goto out;
528 }
529
530 switch (sock->state) {
531 default:
532 err = -EINVAL;
533 goto out;
534 case SS_CONNECTED:
535 err = -EISCONN;
536 goto out;
537 case SS_CONNECTING:
538 err = -EALREADY;
539 /* Fall out of switch with err, set for this state */
540 break;
541 case SS_UNCONNECTED:
542 err = -EISCONN;
543 if (sk->sk_state != TCP_CLOSE)
544 goto out;
545
546 err = sk->sk_prot->connect(sk, uaddr, addr_len);
547 if (err < 0)
548 goto out;
549
550 sock->state = SS_CONNECTING;
551
552 /* Just entered SS_CONNECTING state; the only
553 * difference is that return value in non-blocking
554 * case is EINPROGRESS, rather than EALREADY.
555 */
556 err = -EINPROGRESS;
557 break;
558 }
559
560 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
561
562 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
563 /* Error code is set above */
564 if (!timeo || !inet_wait_for_connect(sk, timeo))
565 goto out;
566
567 err = sock_intr_errno(timeo);
568 if (signal_pending(current))
569 goto out;
570 }
571
572 /* Connection was closed by RST, timeout, ICMP error
573 * or another process disconnected us.
574 */
575 if (sk->sk_state == TCP_CLOSE)
576 goto sock_error;
577
578 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
579 * and error was received after socket entered established state.
580 * Hence, it is handled normally after connect() return successfully.
581 */
582
583 sock->state = SS_CONNECTED;
584 err = 0;
585out:
586 release_sock(sk);
587 return err;
588
589sock_error:
590 err = sock_error(sk) ? : -ECONNABORTED;
591 sock->state = SS_UNCONNECTED;
592 if (sk->sk_prot->disconnect(sk, flags))
593 sock->state = SS_DISCONNECTING;
594 goto out;
595}
596
597/*
598 * Accept a pending connection. The TCP layer now gives BSD semantics.
599 */
600
601int inet_accept(struct socket *sock, struct socket *newsock, int flags)
602{
603 struct sock *sk1 = sock->sk;
604 int err = -EINVAL;
605 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
606
607 if (!sk2)
608 goto do_err;
609
610 lock_sock(sk2);
611
612 BUG_TRAP((1 << sk2->sk_state) &
613 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE));
614
615 sock_graft(sk2, newsock);
616
617 newsock->state = SS_CONNECTED;
618 err = 0;
619 release_sock(sk2);
620do_err:
621 return err;
622}
623
624
625/*
626 * This does both peername and sockname.
627 */
628int inet_getname(struct socket *sock, struct sockaddr *uaddr,
629 int *uaddr_len, int peer)
630{
631 struct sock *sk = sock->sk;
632 struct inet_sock *inet = inet_sk(sk);
633 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
634
635 sin->sin_family = AF_INET;
636 if (peer) {
637 if (!inet->dport ||
638 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
639 peer == 1))
640 return -ENOTCONN;
641 sin->sin_port = inet->dport;
642 sin->sin_addr.s_addr = inet->daddr;
643 } else {
644 __u32 addr = inet->rcv_saddr;
645 if (!addr)
646 addr = inet->saddr;
647 sin->sin_port = inet->sport;
648 sin->sin_addr.s_addr = addr;
649 }
650 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
651 *uaddr_len = sizeof(*sin);
652 return 0;
653}
654
655int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
656 size_t size)
657{
658 struct sock *sk = sock->sk;
659
660 /* We may need to bind the socket. */
661 if (!inet_sk(sk)->num && inet_autobind(sk))
662 return -EAGAIN;
663
664 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
665}
666
667
668static ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
669{
670 struct sock *sk = sock->sk;
671
672 /* We may need to bind the socket. */
673 if (!inet_sk(sk)->num && inet_autobind(sk))
674 return -EAGAIN;
675
676 if (sk->sk_prot->sendpage)
677 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
678 return sock_no_sendpage(sock, page, offset, size, flags);
679}
680
681
682int inet_shutdown(struct socket *sock, int how)
683{
684 struct sock *sk = sock->sk;
685 int err = 0;
686
687 /* This should really check to make sure
688 * the socket is a TCP socket. (WHY AC...)
689 */
690 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
691 1->2 bit 2 snds.
692 2->3 */
693 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
694 return -EINVAL;
695
696 lock_sock(sk);
697 if (sock->state == SS_CONNECTING) {
698 if ((1 << sk->sk_state) &
699 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
700 sock->state = SS_DISCONNECTING;
701 else
702 sock->state = SS_CONNECTED;
703 }
704
705 switch (sk->sk_state) {
706 case TCP_CLOSE:
707 err = -ENOTCONN;
708 /* Hack to wake up other listeners, who can poll for
709 POLLHUP, even on eg. unconnected UDP sockets -- RR */
710 default:
711 sk->sk_shutdown |= how;
712 if (sk->sk_prot->shutdown)
713 sk->sk_prot->shutdown(sk, how);
714 break;
715
716 /* Remaining two branches are temporary solution for missing
717 * close() in multithreaded environment. It is _not_ a good idea,
718 * but we have no choice until close() is repaired at VFS level.
719 */
720 case TCP_LISTEN:
721 if (!(how & RCV_SHUTDOWN))
722 break;
723 /* Fall through */
724 case TCP_SYN_SENT:
725 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
726 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
727 break;
728 }
729
730 /* Wake up anyone sleeping in poll. */
731 sk->sk_state_change(sk);
732 release_sock(sk);
733 return err;
734}
735
736/*
737 * ioctl() calls you can issue on an INET socket. Most of these are
738 * device configuration and stuff and very rarely used. Some ioctls
739 * pass on to the socket itself.
740 *
741 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
742 * loads the devconfigure module does its configuring and unloads it.
743 * There's a good 20K of config code hanging around the kernel.
744 */
745
746int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
747{
748 struct sock *sk = sock->sk;
749 int err = 0;
750
751 switch (cmd) {
752 case SIOCGSTAMP:
753 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
754 break;
755 case SIOCADDRT:
756 case SIOCDELRT:
757 case SIOCRTMSG:
758 err = ip_rt_ioctl(cmd, (void __user *)arg);
759 break;
760 case SIOCDARP:
761 case SIOCGARP:
762 case SIOCSARP:
763 err = arp_ioctl(cmd, (void __user *)arg);
764 break;
765 case SIOCGIFADDR:
766 case SIOCSIFADDR:
767 case SIOCGIFBRDADDR:
768 case SIOCSIFBRDADDR:
769 case SIOCGIFNETMASK:
770 case SIOCSIFNETMASK:
771 case SIOCGIFDSTADDR:
772 case SIOCSIFDSTADDR:
773 case SIOCSIFPFLAGS:
774 case SIOCGIFPFLAGS:
775 case SIOCSIFFLAGS:
776 err = devinet_ioctl(cmd, (void __user *)arg);
777 break;
778 default:
779 if (!sk->sk_prot->ioctl ||
780 (err = sk->sk_prot->ioctl(sk, cmd, arg)) ==
781 -ENOIOCTLCMD)
782 err = dev_ioctl(cmd, (void __user *)arg);
783 break;
784 }
785 return err;
786}
787
Eric Dumazet90ddc4f2005-12-22 12:49:22 -0800788const struct proto_ops inet_stream_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 .family = PF_INET,
790 .owner = THIS_MODULE,
791 .release = inet_release,
792 .bind = inet_bind,
793 .connect = inet_stream_connect,
794 .socketpair = sock_no_socketpair,
795 .accept = inet_accept,
796 .getname = inet_getname,
797 .poll = tcp_poll,
798 .ioctl = inet_ioctl,
799 .listen = inet_listen,
800 .shutdown = inet_shutdown,
801 .setsockopt = sock_common_setsockopt,
802 .getsockopt = sock_common_getsockopt,
803 .sendmsg = inet_sendmsg,
804 .recvmsg = sock_common_recvmsg,
805 .mmap = sock_no_mmap,
806 .sendpage = tcp_sendpage
807};
808
Eric Dumazet90ddc4f2005-12-22 12:49:22 -0800809const struct proto_ops inet_dgram_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 .family = PF_INET,
811 .owner = THIS_MODULE,
812 .release = inet_release,
813 .bind = inet_bind,
814 .connect = inet_dgram_connect,
815 .socketpair = sock_no_socketpair,
816 .accept = sock_no_accept,
817 .getname = inet_getname,
818 .poll = udp_poll,
819 .ioctl = inet_ioctl,
820 .listen = sock_no_listen,
821 .shutdown = inet_shutdown,
822 .setsockopt = sock_common_setsockopt,
823 .getsockopt = sock_common_getsockopt,
824 .sendmsg = inet_sendmsg,
825 .recvmsg = sock_common_recvmsg,
826 .mmap = sock_no_mmap,
827 .sendpage = inet_sendpage,
828};
829
830/*
831 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
832 * udp_poll
833 */
Eric Dumazet90ddc4f2005-12-22 12:49:22 -0800834static const struct proto_ops inet_sockraw_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 .family = PF_INET,
836 .owner = THIS_MODULE,
837 .release = inet_release,
838 .bind = inet_bind,
839 .connect = inet_dgram_connect,
840 .socketpair = sock_no_socketpair,
841 .accept = sock_no_accept,
842 .getname = inet_getname,
843 .poll = datagram_poll,
844 .ioctl = inet_ioctl,
845 .listen = sock_no_listen,
846 .shutdown = inet_shutdown,
847 .setsockopt = sock_common_setsockopt,
848 .getsockopt = sock_common_getsockopt,
849 .sendmsg = inet_sendmsg,
850 .recvmsg = sock_common_recvmsg,
851 .mmap = sock_no_mmap,
852 .sendpage = inet_sendpage,
853};
854
855static struct net_proto_family inet_family_ops = {
856 .family = PF_INET,
857 .create = inet_create,
858 .owner = THIS_MODULE,
859};
860
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861/* Upon startup we insert all the elements in inetsw_array[] into
862 * the linked list inetsw.
863 */
864static struct inet_protosw inetsw_array[] =
865{
866 {
867 .type = SOCK_STREAM,
868 .protocol = IPPROTO_TCP,
869 .prot = &tcp_prot,
870 .ops = &inet_stream_ops,
871 .capability = -1,
872 .no_check = 0,
Arnaldo Carvalho de Melod83d8462005-12-13 23:26:10 -0800873 .flags = INET_PROTOSW_PERMANENT |
874 INET_PROTOSW_ICSK,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 },
876
877 {
878 .type = SOCK_DGRAM,
879 .protocol = IPPROTO_UDP,
880 .prot = &udp_prot,
881 .ops = &inet_dgram_ops,
882 .capability = -1,
883 .no_check = UDP_CSUM_DEFAULT,
884 .flags = INET_PROTOSW_PERMANENT,
885 },
886
887
888 {
889 .type = SOCK_RAW,
890 .protocol = IPPROTO_IP, /* wild card */
891 .prot = &raw_prot,
892 .ops = &inet_sockraw_ops,
893 .capability = CAP_NET_RAW,
894 .no_check = UDP_CSUM_DEFAULT,
895 .flags = INET_PROTOSW_REUSE,
896 }
897};
898
899#define INETSW_ARRAY_LEN (sizeof(inetsw_array) / sizeof(struct inet_protosw))
900
901void inet_register_protosw(struct inet_protosw *p)
902{
903 struct list_head *lh;
904 struct inet_protosw *answer;
905 int protocol = p->protocol;
906 struct list_head *last_perm;
907
908 spin_lock_bh(&inetsw_lock);
909
910 if (p->type >= SOCK_MAX)
911 goto out_illegal;
912
913 /* If we are trying to override a permanent protocol, bail. */
914 answer = NULL;
915 last_perm = &inetsw[p->type];
916 list_for_each(lh, &inetsw[p->type]) {
917 answer = list_entry(lh, struct inet_protosw, list);
918
919 /* Check only the non-wild match. */
920 if (INET_PROTOSW_PERMANENT & answer->flags) {
921 if (protocol == answer->protocol)
922 break;
923 last_perm = lh;
924 }
925
926 answer = NULL;
927 }
928 if (answer)
929 goto out_permanent;
930
931 /* Add the new entry after the last permanent entry if any, so that
932 * the new entry does not override a permanent entry when matched with
933 * a wild-card protocol. But it is allowed to override any existing
934 * non-permanent entry. This means that when we remove this entry, the
935 * system automatically returns to the old behavior.
936 */
937 list_add_rcu(&p->list, last_perm);
938out:
939 spin_unlock_bh(&inetsw_lock);
940
941 synchronize_net();
942
943 return;
944
945out_permanent:
946 printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
947 protocol);
948 goto out;
949
950out_illegal:
951 printk(KERN_ERR
952 "Ignoring attempt to register invalid socket type %d.\n",
953 p->type);
954 goto out;
955}
956
957void inet_unregister_protosw(struct inet_protosw *p)
958{
959 if (INET_PROTOSW_PERMANENT & p->flags) {
960 printk(KERN_ERR
961 "Attempt to unregister permanent protocol %d.\n",
962 p->protocol);
963 } else {
964 spin_lock_bh(&inetsw_lock);
965 list_del_rcu(&p->list);
966 spin_unlock_bh(&inetsw_lock);
967
968 synchronize_net();
969 }
970}
971
Arnaldo Carvalho de Melo32519f12005-08-09 19:50:02 -0700972/*
973 * Shall we try to damage output packets if routing dev changes?
974 */
975
976int sysctl_ip_dynaddr;
977
978static int inet_sk_reselect_saddr(struct sock *sk)
979{
980 struct inet_sock *inet = inet_sk(sk);
981 int err;
982 struct rtable *rt;
983 __u32 old_saddr = inet->saddr;
984 __u32 new_saddr;
985 __u32 daddr = inet->daddr;
986
987 if (inet->opt && inet->opt->srr)
988 daddr = inet->opt->faddr;
989
990 /* Query new route. */
991 err = ip_route_connect(&rt, daddr, 0,
992 RT_CONN_FLAGS(sk),
993 sk->sk_bound_dev_if,
994 sk->sk_protocol,
995 inet->sport, inet->dport, sk);
996 if (err)
997 return err;
998
999 sk_setup_caps(sk, &rt->u.dst);
1000
1001 new_saddr = rt->rt_src;
1002
1003 if (new_saddr == old_saddr)
1004 return 0;
1005
1006 if (sysctl_ip_dynaddr > 1) {
1007 printk(KERN_INFO "%s(): shifting inet->"
1008 "saddr from %d.%d.%d.%d to %d.%d.%d.%d\n",
1009 __FUNCTION__,
1010 NIPQUAD(old_saddr),
1011 NIPQUAD(new_saddr));
1012 }
1013
1014 inet->saddr = inet->rcv_saddr = new_saddr;
1015
1016 /*
1017 * XXX The only one ugly spot where we need to
1018 * XXX really change the sockets identity after
1019 * XXX it has entered the hashes. -DaveM
1020 *
1021 * Besides that, it does not check for connection
1022 * uniqueness. Wait for troubles.
1023 */
1024 __sk_prot_rehash(sk);
1025 return 0;
1026}
1027
1028int inet_sk_rebuild_header(struct sock *sk)
1029{
1030 struct inet_sock *inet = inet_sk(sk);
1031 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1032 u32 daddr;
1033 int err;
1034
1035 /* Route is OK, nothing to do. */
1036 if (rt)
1037 return 0;
1038
1039 /* Reroute. */
1040 daddr = inet->daddr;
1041 if (inet->opt && inet->opt->srr)
1042 daddr = inet->opt->faddr;
1043{
1044 struct flowi fl = {
1045 .oif = sk->sk_bound_dev_if,
1046 .nl_u = {
1047 .ip4_u = {
1048 .daddr = daddr,
1049 .saddr = inet->saddr,
1050 .tos = RT_CONN_FLAGS(sk),
1051 },
1052 },
1053 .proto = sk->sk_protocol,
1054 .uli_u = {
1055 .ports = {
1056 .sport = inet->sport,
1057 .dport = inet->dport,
1058 },
1059 },
1060 };
1061
1062 err = ip_route_output_flow(&rt, &fl, sk, 0);
1063}
1064 if (!err)
1065 sk_setup_caps(sk, &rt->u.dst);
1066 else {
1067 /* Routing failed... */
1068 sk->sk_route_caps = 0;
1069 /*
1070 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1071 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1072 */
1073 if (!sysctl_ip_dynaddr ||
1074 sk->sk_state != TCP_SYN_SENT ||
1075 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1076 (err = inet_sk_reselect_saddr(sk)) != 0)
1077 sk->sk_err_soft = -err;
1078 }
1079
1080 return err;
1081}
1082
1083EXPORT_SYMBOL(inet_sk_rebuild_header);
1084
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085#ifdef CONFIG_IP_MULTICAST
1086static struct net_protocol igmp_protocol = {
1087 .handler = igmp_rcv,
1088};
1089#endif
1090
1091static struct net_protocol tcp_protocol = {
1092 .handler = tcp_v4_rcv,
1093 .err_handler = tcp_v4_err,
1094 .no_policy = 1,
1095};
1096
1097static struct net_protocol udp_protocol = {
1098 .handler = udp_rcv,
1099 .err_handler = udp_err,
1100 .no_policy = 1,
1101};
1102
1103static struct net_protocol icmp_protocol = {
1104 .handler = icmp_rcv,
1105};
1106
1107static int __init init_ipv4_mibs(void)
1108{
1109 net_statistics[0] = alloc_percpu(struct linux_mib);
1110 net_statistics[1] = alloc_percpu(struct linux_mib);
1111 ip_statistics[0] = alloc_percpu(struct ipstats_mib);
1112 ip_statistics[1] = alloc_percpu(struct ipstats_mib);
1113 icmp_statistics[0] = alloc_percpu(struct icmp_mib);
1114 icmp_statistics[1] = alloc_percpu(struct icmp_mib);
1115 tcp_statistics[0] = alloc_percpu(struct tcp_mib);
1116 tcp_statistics[1] = alloc_percpu(struct tcp_mib);
1117 udp_statistics[0] = alloc_percpu(struct udp_mib);
1118 udp_statistics[1] = alloc_percpu(struct udp_mib);
1119 if (!
1120 (net_statistics[0] && net_statistics[1] && ip_statistics[0]
1121 && ip_statistics[1] && tcp_statistics[0] && tcp_statistics[1]
1122 && udp_statistics[0] && udp_statistics[1]))
1123 return -ENOMEM;
1124
1125 (void) tcp_mib_init();
1126
1127 return 0;
1128}
1129
1130static int ipv4_proc_init(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131
Herbert Xu30e224d2005-07-05 14:40:10 -07001132/*
1133 * IP protocol layer initialiser
1134 */
1135
1136static struct packet_type ip_packet_type = {
1137 .type = __constant_htons(ETH_P_IP),
1138 .func = ip_rcv,
1139};
1140
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141static int __init inet_init(void)
1142{
1143 struct sk_buff *dummy_skb;
1144 struct inet_protosw *q;
1145 struct list_head *r;
1146 int rc = -EINVAL;
1147
1148 if (sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb)) {
1149 printk(KERN_CRIT "%s: panic\n", __FUNCTION__);
1150 goto out;
1151 }
1152
1153 rc = proto_register(&tcp_prot, 1);
1154 if (rc)
1155 goto out;
1156
1157 rc = proto_register(&udp_prot, 1);
1158 if (rc)
1159 goto out_unregister_tcp_proto;
1160
1161 rc = proto_register(&raw_prot, 1);
1162 if (rc)
1163 goto out_unregister_udp_proto;
1164
1165 /*
1166 * Tell SOCKET that we are alive...
1167 */
1168
1169 (void)sock_register(&inet_family_ops);
1170
1171 /*
1172 * Add all the base protocols.
1173 */
1174
1175 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1176 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1177 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1178 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1179 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1180 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1181#ifdef CONFIG_IP_MULTICAST
1182 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1183 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1184#endif
1185
1186 /* Register the socket-side information for inet_create. */
1187 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1188 INIT_LIST_HEAD(r);
1189
1190 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1191 inet_register_protosw(q);
1192
1193 /*
1194 * Set the ARP module up
1195 */
1196
1197 arp_init();
1198
1199 /*
1200 * Set the IP module up
1201 */
1202
1203 ip_init();
1204
1205 tcp_v4_init(&inet_family_ops);
1206
1207 /* Setup TCP slab cache for open requests. */
1208 tcp_init();
1209
1210
1211 /*
1212 * Set the ICMP layer up
1213 */
1214
1215 icmp_init(&inet_family_ops);
1216
1217 /*
1218 * Initialise the multicast router
1219 */
1220#if defined(CONFIG_IP_MROUTE)
1221 ip_mr_init();
1222#endif
1223 /*
1224 * Initialise per-cpu ipv4 mibs
1225 */
1226
1227 if(init_ipv4_mibs())
1228 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n"); ;
1229
1230 ipv4_proc_init();
1231
1232 ipfrag_init();
1233
Herbert Xu30e224d2005-07-05 14:40:10 -07001234 dev_add_pack(&ip_packet_type);
1235
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236 rc = 0;
1237out:
1238 return rc;
1239out_unregister_tcp_proto:
1240 proto_unregister(&tcp_prot);
1241out_unregister_udp_proto:
1242 proto_unregister(&udp_prot);
1243 goto out;
1244}
1245
1246module_init(inet_init);
1247
1248/* ------------------------------------------------------------------------ */
1249
1250#ifdef CONFIG_PROC_FS
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251static int __init ipv4_proc_init(void)
1252{
1253 int rc = 0;
1254
1255 if (raw_proc_init())
1256 goto out_raw;
1257 if (tcp4_proc_init())
1258 goto out_tcp;
1259 if (udp4_proc_init())
1260 goto out_udp;
1261 if (fib_proc_init())
1262 goto out_fib;
1263 if (ip_misc_proc_init())
1264 goto out_misc;
1265out:
1266 return rc;
1267out_misc:
1268 fib_proc_exit();
1269out_fib:
1270 udp4_proc_exit();
1271out_udp:
1272 tcp4_proc_exit();
1273out_tcp:
1274 raw_proc_exit();
1275out_raw:
1276 rc = -ENOMEM;
1277 goto out;
1278}
1279
1280#else /* CONFIG_PROC_FS */
1281static int __init ipv4_proc_init(void)
1282{
1283 return 0;
1284}
1285#endif /* CONFIG_PROC_FS */
1286
1287MODULE_ALIAS_NETPROTO(PF_INET);
1288
1289EXPORT_SYMBOL(inet_accept);
1290EXPORT_SYMBOL(inet_bind);
1291EXPORT_SYMBOL(inet_dgram_connect);
1292EXPORT_SYMBOL(inet_dgram_ops);
1293EXPORT_SYMBOL(inet_getname);
1294EXPORT_SYMBOL(inet_ioctl);
1295EXPORT_SYMBOL(inet_listen);
1296EXPORT_SYMBOL(inet_register_protosw);
1297EXPORT_SYMBOL(inet_release);
1298EXPORT_SYMBOL(inet_sendmsg);
1299EXPORT_SYMBOL(inet_shutdown);
1300EXPORT_SYMBOL(inet_sock_destruct);
1301EXPORT_SYMBOL(inet_stream_connect);
1302EXPORT_SYMBOL(inet_stream_ops);
1303EXPORT_SYMBOL(inet_unregister_protosw);
1304EXPORT_SYMBOL(net_statistics);
Neil Hormancdac4e02005-06-13 15:12:33 -07001305EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);