blob: 9e83d7773d8f29a79ef5cfa1d248130d78fbe138 [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>
102#include <net/tcp.h>
103#include <net/udp.h>
104#include <linux/skbuff.h>
105#include <net/sock.h>
106#include <net/raw.h>
107#include <net/icmp.h>
108#include <net/ipip.h>
109#include <net/inet_common.h>
110#include <net/xfrm.h>
111#ifdef CONFIG_IP_MROUTE
112#include <linux/mroute.h>
113#endif
114
115DEFINE_SNMP_STAT(struct linux_mib, net_statistics);
116
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117extern void ip_mc_drop_socket(struct sock *sk);
118
119/* The inetsw table contains everything that inet_create needs to
120 * build a new socket.
121 */
122static struct list_head inetsw[SOCK_MAX];
123static DEFINE_SPINLOCK(inetsw_lock);
124
125/* New destruction routine */
126
127void inet_sock_destruct(struct sock *sk)
128{
129 struct inet_sock *inet = inet_sk(sk);
130
131 __skb_queue_purge(&sk->sk_receive_queue);
132 __skb_queue_purge(&sk->sk_error_queue);
133
134 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
135 printk("Attempt to release TCP socket in state %d %p\n",
136 sk->sk_state, sk);
137 return;
138 }
139 if (!sock_flag(sk, SOCK_DEAD)) {
140 printk("Attempt to release alive inet socket %p\n", sk);
141 return;
142 }
143
144 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
145 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
146 BUG_TRAP(!sk->sk_wmem_queued);
147 BUG_TRAP(!sk->sk_forward_alloc);
148
149 if (inet->opt)
150 kfree(inet->opt);
151 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) {
205 err = tcp_listen_start(sk);
206 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;
230 int err;
231
232 sock->state = SS_UNCONNECTED;
233
234 /* Look for the requested type/protocol pair. */
235 answer = NULL;
236 rcu_read_lock();
237 list_for_each_rcu(p, &inetsw[sock->type]) {
238 answer = list_entry(p, struct inet_protosw, list);
239
240 /* Check the non-wild match. */
241 if (protocol == answer->protocol) {
242 if (protocol != IPPROTO_IP)
243 break;
244 } else {
245 /* Check for the two wild cases. */
246 if (IPPROTO_IP == protocol) {
247 protocol = answer->protocol;
248 break;
249 }
250 if (IPPROTO_IP == answer->protocol)
251 break;
252 }
253 answer = NULL;
254 }
255
256 err = -ESOCKTNOSUPPORT;
257 if (!answer)
258 goto out_rcu_unlock;
259 err = -EPERM;
260 if (answer->capability > 0 && !capable(answer->capability))
261 goto out_rcu_unlock;
262 err = -EPROTONOSUPPORT;
263 if (!protocol)
264 goto out_rcu_unlock;
265
266 sock->ops = answer->ops;
267 answer_prot = answer->prot;
268 answer_no_check = answer->no_check;
269 answer_flags = answer->flags;
270 rcu_read_unlock();
271
272 BUG_TRAP(answer_prot->slab != NULL);
273
274 err = -ENOBUFS;
275 sk = sk_alloc(PF_INET, GFP_KERNEL, answer_prot, 1);
276 if (sk == NULL)
277 goto out;
278
279 err = 0;
280 sk->sk_no_check = answer_no_check;
281 if (INET_PROTOSW_REUSE & answer_flags)
282 sk->sk_reuse = 1;
283
284 inet = inet_sk(sk);
285
286 if (SOCK_RAW == sock->type) {
287 inet->num = protocol;
288 if (IPPROTO_RAW == protocol)
289 inet->hdrincl = 1;
290 }
291
292 if (ipv4_config.no_pmtu_disc)
293 inet->pmtudisc = IP_PMTUDISC_DONT;
294 else
295 inet->pmtudisc = IP_PMTUDISC_WANT;
296
297 inet->id = 0;
298
299 sock_init_data(sock, sk);
300
301 sk->sk_destruct = inet_sock_destruct;
302 sk->sk_family = PF_INET;
303 sk->sk_protocol = protocol;
304 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
305
306 inet->uc_ttl = -1;
307 inet->mc_loop = 1;
308 inet->mc_ttl = 1;
309 inet->mc_index = 0;
310 inet->mc_list = NULL;
311
Arnaldo Carvalho de Meloe6848972005-08-09 19:45:38 -0700312 sk_refcnt_debug_inc(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313
314 if (inet->num) {
315 /* It assumes that any protocol which allows
316 * the user to assign a number at socket
317 * creation time automatically
318 * shares.
319 */
320 inet->sport = htons(inet->num);
321 /* Add to protocol hash chains. */
322 sk->sk_prot->hash(sk);
323 }
324
325 if (sk->sk_prot->init) {
326 err = sk->sk_prot->init(sk);
327 if (err)
328 sk_common_release(sk);
329 }
330out:
331 return err;
332out_rcu_unlock:
333 rcu_read_unlock();
334 goto out;
335}
336
337
338/*
339 * The peer socket should always be NULL (or else). When we call this
340 * function we are destroying the object and from then on nobody
341 * should refer to it.
342 */
343int inet_release(struct socket *sock)
344{
345 struct sock *sk = sock->sk;
346
347 if (sk) {
348 long timeout;
349
350 /* Applications forget to leave groups before exiting */
351 ip_mc_drop_socket(sk);
352
353 /* If linger is set, we don't return until the close
354 * is complete. Otherwise we return immediately. The
355 * actually closing is done the same either way.
356 *
357 * If the close is due to the process exiting, we never
358 * linger..
359 */
360 timeout = 0;
361 if (sock_flag(sk, SOCK_LINGER) &&
362 !(current->flags & PF_EXITING))
363 timeout = sk->sk_lingertime;
364 sock->sk = NULL;
365 sk->sk_prot->close(sk, timeout);
366 }
367 return 0;
368}
369
370/* It is off by default, see below. */
371int sysctl_ip_nonlocal_bind;
372
373int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
374{
375 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
376 struct sock *sk = sock->sk;
377 struct inet_sock *inet = inet_sk(sk);
378 unsigned short snum;
379 int chk_addr_ret;
380 int err;
381
382 /* If the socket has its own bind function then use it. (RAW) */
383 if (sk->sk_prot->bind) {
384 err = sk->sk_prot->bind(sk, uaddr, addr_len);
385 goto out;
386 }
387 err = -EINVAL;
388 if (addr_len < sizeof(struct sockaddr_in))
389 goto out;
390
391 chk_addr_ret = inet_addr_type(addr->sin_addr.s_addr);
392
393 /* Not specified by any standard per-se, however it breaks too
394 * many applications when removed. It is unfortunate since
395 * allowing applications to make a non-local bind solves
396 * several problems with systems using dynamic addressing.
397 * (ie. your servers still start up even if your ISDN link
398 * is temporarily down)
399 */
400 err = -EADDRNOTAVAIL;
401 if (!sysctl_ip_nonlocal_bind &&
402 !inet->freebind &&
403 addr->sin_addr.s_addr != INADDR_ANY &&
404 chk_addr_ret != RTN_LOCAL &&
405 chk_addr_ret != RTN_MULTICAST &&
406 chk_addr_ret != RTN_BROADCAST)
407 goto out;
408
409 snum = ntohs(addr->sin_port);
410 err = -EACCES;
411 if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
412 goto out;
413
414 /* We keep a pair of addresses. rcv_saddr is the one
415 * used by hash lookups, and saddr is used for transmit.
416 *
417 * In the BSD API these are the same except where it
418 * would be illegal to use them (multicast/broadcast) in
419 * which case the sending device address is used.
420 */
421 lock_sock(sk);
422
423 /* Check these errors (active socket, double bind). */
424 err = -EINVAL;
425 if (sk->sk_state != TCP_CLOSE || inet->num)
426 goto out_release_sock;
427
428 inet->rcv_saddr = inet->saddr = addr->sin_addr.s_addr;
429 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
430 inet->saddr = 0; /* Use device */
431
432 /* Make sure we are allowed to bind here. */
433 if (sk->sk_prot->get_port(sk, snum)) {
434 inet->saddr = inet->rcv_saddr = 0;
435 err = -EADDRINUSE;
436 goto out_release_sock;
437 }
438
439 if (inet->rcv_saddr)
440 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
441 if (snum)
442 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
443 inet->sport = htons(inet->num);
444 inet->daddr = 0;
445 inet->dport = 0;
446 sk_dst_reset(sk);
447 err = 0;
448out_release_sock:
449 release_sock(sk);
450out:
451 return err;
452}
453
454int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
455 int addr_len, int flags)
456{
457 struct sock *sk = sock->sk;
458
459 if (uaddr->sa_family == AF_UNSPEC)
460 return sk->sk_prot->disconnect(sk, flags);
461
462 if (!inet_sk(sk)->num && inet_autobind(sk))
463 return -EAGAIN;
464 return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
465}
466
467static long inet_wait_for_connect(struct sock *sk, long timeo)
468{
469 DEFINE_WAIT(wait);
470
471 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
472
473 /* Basic assumption: if someone sets sk->sk_err, he _must_
474 * change state of the socket from TCP_SYN_*.
475 * Connect() does not allow to get error notifications
476 * without closing the socket.
477 */
478 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
479 release_sock(sk);
480 timeo = schedule_timeout(timeo);
481 lock_sock(sk);
482 if (signal_pending(current) || !timeo)
483 break;
484 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
485 }
486 finish_wait(sk->sk_sleep, &wait);
487 return timeo;
488}
489
490/*
491 * Connect to a remote host. There is regrettably still a little
492 * TCP 'magic' in here.
493 */
494int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
495 int addr_len, int flags)
496{
497 struct sock *sk = sock->sk;
498 int err;
499 long timeo;
500
501 lock_sock(sk);
502
503 if (uaddr->sa_family == AF_UNSPEC) {
504 err = sk->sk_prot->disconnect(sk, flags);
505 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
506 goto out;
507 }
508
509 switch (sock->state) {
510 default:
511 err = -EINVAL;
512 goto out;
513 case SS_CONNECTED:
514 err = -EISCONN;
515 goto out;
516 case SS_CONNECTING:
517 err = -EALREADY;
518 /* Fall out of switch with err, set for this state */
519 break;
520 case SS_UNCONNECTED:
521 err = -EISCONN;
522 if (sk->sk_state != TCP_CLOSE)
523 goto out;
524
525 err = sk->sk_prot->connect(sk, uaddr, addr_len);
526 if (err < 0)
527 goto out;
528
529 sock->state = SS_CONNECTING;
530
531 /* Just entered SS_CONNECTING state; the only
532 * difference is that return value in non-blocking
533 * case is EINPROGRESS, rather than EALREADY.
534 */
535 err = -EINPROGRESS;
536 break;
537 }
538
539 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
540
541 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
542 /* Error code is set above */
543 if (!timeo || !inet_wait_for_connect(sk, timeo))
544 goto out;
545
546 err = sock_intr_errno(timeo);
547 if (signal_pending(current))
548 goto out;
549 }
550
551 /* Connection was closed by RST, timeout, ICMP error
552 * or another process disconnected us.
553 */
554 if (sk->sk_state == TCP_CLOSE)
555 goto sock_error;
556
557 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
558 * and error was received after socket entered established state.
559 * Hence, it is handled normally after connect() return successfully.
560 */
561
562 sock->state = SS_CONNECTED;
563 err = 0;
564out:
565 release_sock(sk);
566 return err;
567
568sock_error:
569 err = sock_error(sk) ? : -ECONNABORTED;
570 sock->state = SS_UNCONNECTED;
571 if (sk->sk_prot->disconnect(sk, flags))
572 sock->state = SS_DISCONNECTING;
573 goto out;
574}
575
576/*
577 * Accept a pending connection. The TCP layer now gives BSD semantics.
578 */
579
580int inet_accept(struct socket *sock, struct socket *newsock, int flags)
581{
582 struct sock *sk1 = sock->sk;
583 int err = -EINVAL;
584 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
585
586 if (!sk2)
587 goto do_err;
588
589 lock_sock(sk2);
590
591 BUG_TRAP((1 << sk2->sk_state) &
592 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE));
593
594 sock_graft(sk2, newsock);
595
596 newsock->state = SS_CONNECTED;
597 err = 0;
598 release_sock(sk2);
599do_err:
600 return err;
601}
602
603
604/*
605 * This does both peername and sockname.
606 */
607int inet_getname(struct socket *sock, struct sockaddr *uaddr,
608 int *uaddr_len, int peer)
609{
610 struct sock *sk = sock->sk;
611 struct inet_sock *inet = inet_sk(sk);
612 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
613
614 sin->sin_family = AF_INET;
615 if (peer) {
616 if (!inet->dport ||
617 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
618 peer == 1))
619 return -ENOTCONN;
620 sin->sin_port = inet->dport;
621 sin->sin_addr.s_addr = inet->daddr;
622 } else {
623 __u32 addr = inet->rcv_saddr;
624 if (!addr)
625 addr = inet->saddr;
626 sin->sin_port = inet->sport;
627 sin->sin_addr.s_addr = addr;
628 }
629 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
630 *uaddr_len = sizeof(*sin);
631 return 0;
632}
633
634int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
635 size_t size)
636{
637 struct sock *sk = sock->sk;
638
639 /* We may need to bind the socket. */
640 if (!inet_sk(sk)->num && inet_autobind(sk))
641 return -EAGAIN;
642
643 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
644}
645
646
647static ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
648{
649 struct sock *sk = sock->sk;
650
651 /* We may need to bind the socket. */
652 if (!inet_sk(sk)->num && inet_autobind(sk))
653 return -EAGAIN;
654
655 if (sk->sk_prot->sendpage)
656 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
657 return sock_no_sendpage(sock, page, offset, size, flags);
658}
659
660
661int inet_shutdown(struct socket *sock, int how)
662{
663 struct sock *sk = sock->sk;
664 int err = 0;
665
666 /* This should really check to make sure
667 * the socket is a TCP socket. (WHY AC...)
668 */
669 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
670 1->2 bit 2 snds.
671 2->3 */
672 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
673 return -EINVAL;
674
675 lock_sock(sk);
676 if (sock->state == SS_CONNECTING) {
677 if ((1 << sk->sk_state) &
678 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
679 sock->state = SS_DISCONNECTING;
680 else
681 sock->state = SS_CONNECTED;
682 }
683
684 switch (sk->sk_state) {
685 case TCP_CLOSE:
686 err = -ENOTCONN;
687 /* Hack to wake up other listeners, who can poll for
688 POLLHUP, even on eg. unconnected UDP sockets -- RR */
689 default:
690 sk->sk_shutdown |= how;
691 if (sk->sk_prot->shutdown)
692 sk->sk_prot->shutdown(sk, how);
693 break;
694
695 /* Remaining two branches are temporary solution for missing
696 * close() in multithreaded environment. It is _not_ a good idea,
697 * but we have no choice until close() is repaired at VFS level.
698 */
699 case TCP_LISTEN:
700 if (!(how & RCV_SHUTDOWN))
701 break;
702 /* Fall through */
703 case TCP_SYN_SENT:
704 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
705 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
706 break;
707 }
708
709 /* Wake up anyone sleeping in poll. */
710 sk->sk_state_change(sk);
711 release_sock(sk);
712 return err;
713}
714
715/*
716 * ioctl() calls you can issue on an INET socket. Most of these are
717 * device configuration and stuff and very rarely used. Some ioctls
718 * pass on to the socket itself.
719 *
720 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
721 * loads the devconfigure module does its configuring and unloads it.
722 * There's a good 20K of config code hanging around the kernel.
723 */
724
725int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
726{
727 struct sock *sk = sock->sk;
728 int err = 0;
729
730 switch (cmd) {
731 case SIOCGSTAMP:
732 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
733 break;
734 case SIOCADDRT:
735 case SIOCDELRT:
736 case SIOCRTMSG:
737 err = ip_rt_ioctl(cmd, (void __user *)arg);
738 break;
739 case SIOCDARP:
740 case SIOCGARP:
741 case SIOCSARP:
742 err = arp_ioctl(cmd, (void __user *)arg);
743 break;
744 case SIOCGIFADDR:
745 case SIOCSIFADDR:
746 case SIOCGIFBRDADDR:
747 case SIOCSIFBRDADDR:
748 case SIOCGIFNETMASK:
749 case SIOCSIFNETMASK:
750 case SIOCGIFDSTADDR:
751 case SIOCSIFDSTADDR:
752 case SIOCSIFPFLAGS:
753 case SIOCGIFPFLAGS:
754 case SIOCSIFFLAGS:
755 err = devinet_ioctl(cmd, (void __user *)arg);
756 break;
757 default:
758 if (!sk->sk_prot->ioctl ||
759 (err = sk->sk_prot->ioctl(sk, cmd, arg)) ==
760 -ENOIOCTLCMD)
761 err = dev_ioctl(cmd, (void __user *)arg);
762 break;
763 }
764 return err;
765}
766
767struct proto_ops inet_stream_ops = {
768 .family = PF_INET,
769 .owner = THIS_MODULE,
770 .release = inet_release,
771 .bind = inet_bind,
772 .connect = inet_stream_connect,
773 .socketpair = sock_no_socketpair,
774 .accept = inet_accept,
775 .getname = inet_getname,
776 .poll = tcp_poll,
777 .ioctl = inet_ioctl,
778 .listen = inet_listen,
779 .shutdown = inet_shutdown,
780 .setsockopt = sock_common_setsockopt,
781 .getsockopt = sock_common_getsockopt,
782 .sendmsg = inet_sendmsg,
783 .recvmsg = sock_common_recvmsg,
784 .mmap = sock_no_mmap,
785 .sendpage = tcp_sendpage
786};
787
788struct proto_ops inet_dgram_ops = {
789 .family = PF_INET,
790 .owner = THIS_MODULE,
791 .release = inet_release,
792 .bind = inet_bind,
793 .connect = inet_dgram_connect,
794 .socketpair = sock_no_socketpair,
795 .accept = sock_no_accept,
796 .getname = inet_getname,
797 .poll = udp_poll,
798 .ioctl = inet_ioctl,
799 .listen = sock_no_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 = inet_sendpage,
807};
808
809/*
810 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
811 * udp_poll
812 */
813static struct proto_ops inet_sockraw_ops = {
814 .family = PF_INET,
815 .owner = THIS_MODULE,
816 .release = inet_release,
817 .bind = inet_bind,
818 .connect = inet_dgram_connect,
819 .socketpair = sock_no_socketpair,
820 .accept = sock_no_accept,
821 .getname = inet_getname,
822 .poll = datagram_poll,
823 .ioctl = inet_ioctl,
824 .listen = sock_no_listen,
825 .shutdown = inet_shutdown,
826 .setsockopt = sock_common_setsockopt,
827 .getsockopt = sock_common_getsockopt,
828 .sendmsg = inet_sendmsg,
829 .recvmsg = sock_common_recvmsg,
830 .mmap = sock_no_mmap,
831 .sendpage = inet_sendpage,
832};
833
834static struct net_proto_family inet_family_ops = {
835 .family = PF_INET,
836 .create = inet_create,
837 .owner = THIS_MODULE,
838};
839
840
841extern void tcp_init(void);
842extern void tcp_v4_init(struct net_proto_family *);
843
844/* Upon startup we insert all the elements in inetsw_array[] into
845 * the linked list inetsw.
846 */
847static struct inet_protosw inetsw_array[] =
848{
849 {
850 .type = SOCK_STREAM,
851 .protocol = IPPROTO_TCP,
852 .prot = &tcp_prot,
853 .ops = &inet_stream_ops,
854 .capability = -1,
855 .no_check = 0,
856 .flags = INET_PROTOSW_PERMANENT,
857 },
858
859 {
860 .type = SOCK_DGRAM,
861 .protocol = IPPROTO_UDP,
862 .prot = &udp_prot,
863 .ops = &inet_dgram_ops,
864 .capability = -1,
865 .no_check = UDP_CSUM_DEFAULT,
866 .flags = INET_PROTOSW_PERMANENT,
867 },
868
869
870 {
871 .type = SOCK_RAW,
872 .protocol = IPPROTO_IP, /* wild card */
873 .prot = &raw_prot,
874 .ops = &inet_sockraw_ops,
875 .capability = CAP_NET_RAW,
876 .no_check = UDP_CSUM_DEFAULT,
877 .flags = INET_PROTOSW_REUSE,
878 }
879};
880
881#define INETSW_ARRAY_LEN (sizeof(inetsw_array) / sizeof(struct inet_protosw))
882
883void inet_register_protosw(struct inet_protosw *p)
884{
885 struct list_head *lh;
886 struct inet_protosw *answer;
887 int protocol = p->protocol;
888 struct list_head *last_perm;
889
890 spin_lock_bh(&inetsw_lock);
891
892 if (p->type >= SOCK_MAX)
893 goto out_illegal;
894
895 /* If we are trying to override a permanent protocol, bail. */
896 answer = NULL;
897 last_perm = &inetsw[p->type];
898 list_for_each(lh, &inetsw[p->type]) {
899 answer = list_entry(lh, struct inet_protosw, list);
900
901 /* Check only the non-wild match. */
902 if (INET_PROTOSW_PERMANENT & answer->flags) {
903 if (protocol == answer->protocol)
904 break;
905 last_perm = lh;
906 }
907
908 answer = NULL;
909 }
910 if (answer)
911 goto out_permanent;
912
913 /* Add the new entry after the last permanent entry if any, so that
914 * the new entry does not override a permanent entry when matched with
915 * a wild-card protocol. But it is allowed to override any existing
916 * non-permanent entry. This means that when we remove this entry, the
917 * system automatically returns to the old behavior.
918 */
919 list_add_rcu(&p->list, last_perm);
920out:
921 spin_unlock_bh(&inetsw_lock);
922
923 synchronize_net();
924
925 return;
926
927out_permanent:
928 printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
929 protocol);
930 goto out;
931
932out_illegal:
933 printk(KERN_ERR
934 "Ignoring attempt to register invalid socket type %d.\n",
935 p->type);
936 goto out;
937}
938
939void inet_unregister_protosw(struct inet_protosw *p)
940{
941 if (INET_PROTOSW_PERMANENT & p->flags) {
942 printk(KERN_ERR
943 "Attempt to unregister permanent protocol %d.\n",
944 p->protocol);
945 } else {
946 spin_lock_bh(&inetsw_lock);
947 list_del_rcu(&p->list);
948 spin_unlock_bh(&inetsw_lock);
949
950 synchronize_net();
951 }
952}
953
954#ifdef CONFIG_IP_MULTICAST
955static struct net_protocol igmp_protocol = {
956 .handler = igmp_rcv,
957};
958#endif
959
960static struct net_protocol tcp_protocol = {
961 .handler = tcp_v4_rcv,
962 .err_handler = tcp_v4_err,
963 .no_policy = 1,
964};
965
966static struct net_protocol udp_protocol = {
967 .handler = udp_rcv,
968 .err_handler = udp_err,
969 .no_policy = 1,
970};
971
972static struct net_protocol icmp_protocol = {
973 .handler = icmp_rcv,
974};
975
976static int __init init_ipv4_mibs(void)
977{
978 net_statistics[0] = alloc_percpu(struct linux_mib);
979 net_statistics[1] = alloc_percpu(struct linux_mib);
980 ip_statistics[0] = alloc_percpu(struct ipstats_mib);
981 ip_statistics[1] = alloc_percpu(struct ipstats_mib);
982 icmp_statistics[0] = alloc_percpu(struct icmp_mib);
983 icmp_statistics[1] = alloc_percpu(struct icmp_mib);
984 tcp_statistics[0] = alloc_percpu(struct tcp_mib);
985 tcp_statistics[1] = alloc_percpu(struct tcp_mib);
986 udp_statistics[0] = alloc_percpu(struct udp_mib);
987 udp_statistics[1] = alloc_percpu(struct udp_mib);
988 if (!
989 (net_statistics[0] && net_statistics[1] && ip_statistics[0]
990 && ip_statistics[1] && tcp_statistics[0] && tcp_statistics[1]
991 && udp_statistics[0] && udp_statistics[1]))
992 return -ENOMEM;
993
994 (void) tcp_mib_init();
995
996 return 0;
997}
998
999static int ipv4_proc_init(void);
1000extern void ipfrag_init(void);
1001
Herbert Xu30e224d2005-07-05 14:40:10 -07001002/*
1003 * IP protocol layer initialiser
1004 */
1005
1006static struct packet_type ip_packet_type = {
1007 .type = __constant_htons(ETH_P_IP),
1008 .func = ip_rcv,
1009};
1010
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011static int __init inet_init(void)
1012{
1013 struct sk_buff *dummy_skb;
1014 struct inet_protosw *q;
1015 struct list_head *r;
1016 int rc = -EINVAL;
1017
1018 if (sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb)) {
1019 printk(KERN_CRIT "%s: panic\n", __FUNCTION__);
1020 goto out;
1021 }
1022
1023 rc = proto_register(&tcp_prot, 1);
1024 if (rc)
1025 goto out;
1026
1027 rc = proto_register(&udp_prot, 1);
1028 if (rc)
1029 goto out_unregister_tcp_proto;
1030
1031 rc = proto_register(&raw_prot, 1);
1032 if (rc)
1033 goto out_unregister_udp_proto;
1034
1035 /*
1036 * Tell SOCKET that we are alive...
1037 */
1038
1039 (void)sock_register(&inet_family_ops);
1040
1041 /*
1042 * Add all the base protocols.
1043 */
1044
1045 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1046 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1047 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1048 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1049 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1050 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1051#ifdef CONFIG_IP_MULTICAST
1052 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1053 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1054#endif
1055
1056 /* Register the socket-side information for inet_create. */
1057 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1058 INIT_LIST_HEAD(r);
1059
1060 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1061 inet_register_protosw(q);
1062
1063 /*
1064 * Set the ARP module up
1065 */
1066
1067 arp_init();
1068
1069 /*
1070 * Set the IP module up
1071 */
1072
1073 ip_init();
1074
1075 tcp_v4_init(&inet_family_ops);
1076
1077 /* Setup TCP slab cache for open requests. */
1078 tcp_init();
1079
1080
1081 /*
1082 * Set the ICMP layer up
1083 */
1084
1085 icmp_init(&inet_family_ops);
1086
1087 /*
1088 * Initialise the multicast router
1089 */
1090#if defined(CONFIG_IP_MROUTE)
1091 ip_mr_init();
1092#endif
1093 /*
1094 * Initialise per-cpu ipv4 mibs
1095 */
1096
1097 if(init_ipv4_mibs())
1098 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n"); ;
1099
1100 ipv4_proc_init();
1101
1102 ipfrag_init();
1103
Herbert Xu30e224d2005-07-05 14:40:10 -07001104 dev_add_pack(&ip_packet_type);
1105
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 rc = 0;
1107out:
1108 return rc;
1109out_unregister_tcp_proto:
1110 proto_unregister(&tcp_prot);
1111out_unregister_udp_proto:
1112 proto_unregister(&udp_prot);
1113 goto out;
1114}
1115
1116module_init(inet_init);
1117
1118/* ------------------------------------------------------------------------ */
1119
1120#ifdef CONFIG_PROC_FS
1121extern int fib_proc_init(void);
1122extern void fib_proc_exit(void);
Robert Olsson19baf832005-06-21 12:43:18 -07001123#ifdef CONFIG_IP_FIB_TRIE
1124extern int fib_stat_proc_init(void);
1125extern void fib_stat_proc_exit(void);
1126#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127extern int ip_misc_proc_init(void);
1128extern int raw_proc_init(void);
1129extern void raw_proc_exit(void);
1130extern int tcp4_proc_init(void);
1131extern void tcp4_proc_exit(void);
1132extern int udp4_proc_init(void);
1133extern void udp4_proc_exit(void);
1134
1135static int __init ipv4_proc_init(void)
1136{
1137 int rc = 0;
1138
1139 if (raw_proc_init())
1140 goto out_raw;
1141 if (tcp4_proc_init())
1142 goto out_tcp;
1143 if (udp4_proc_init())
1144 goto out_udp;
1145 if (fib_proc_init())
1146 goto out_fib;
Robert Olsson19baf832005-06-21 12:43:18 -07001147#ifdef CONFIG_IP_FIB_TRIE
1148 if (fib_stat_proc_init())
1149 goto out_fib_stat;
Stephen Hemmingerc877efb2005-07-19 14:01:51 -07001150#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151 if (ip_misc_proc_init())
1152 goto out_misc;
1153out:
1154 return rc;
1155out_misc:
Robert Olsson19baf832005-06-21 12:43:18 -07001156#ifdef CONFIG_IP_FIB_TRIE
1157 fib_stat_proc_exit();
1158out_fib_stat:
1159#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 fib_proc_exit();
1161out_fib:
1162 udp4_proc_exit();
1163out_udp:
1164 tcp4_proc_exit();
1165out_tcp:
1166 raw_proc_exit();
1167out_raw:
1168 rc = -ENOMEM;
1169 goto out;
1170}
1171
1172#else /* CONFIG_PROC_FS */
1173static int __init ipv4_proc_init(void)
1174{
1175 return 0;
1176}
1177#endif /* CONFIG_PROC_FS */
1178
1179MODULE_ALIAS_NETPROTO(PF_INET);
1180
1181EXPORT_SYMBOL(inet_accept);
1182EXPORT_SYMBOL(inet_bind);
1183EXPORT_SYMBOL(inet_dgram_connect);
1184EXPORT_SYMBOL(inet_dgram_ops);
1185EXPORT_SYMBOL(inet_getname);
1186EXPORT_SYMBOL(inet_ioctl);
1187EXPORT_SYMBOL(inet_listen);
1188EXPORT_SYMBOL(inet_register_protosw);
1189EXPORT_SYMBOL(inet_release);
1190EXPORT_SYMBOL(inet_sendmsg);
1191EXPORT_SYMBOL(inet_shutdown);
1192EXPORT_SYMBOL(inet_sock_destruct);
1193EXPORT_SYMBOL(inet_stream_connect);
1194EXPORT_SYMBOL(inet_stream_ops);
1195EXPORT_SYMBOL(inet_unregister_protosw);
1196EXPORT_SYMBOL(net_statistics);
Neil Hormancdac4e02005-06-13 15:12:33 -07001197EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);