blob: e68103475cca2a76a349d270d1db72c62132b4c7 [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
Herbert Xuf4c50d92006-06-22 03:02:40 -070070#include <linux/err.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070071#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>
David S. Millerb3da2cf2007-03-23 11:40:27 -070090#include <linux/random.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070091
92#include <asm/uaccess.h>
93#include <asm/system.h>
94
Linus Torvalds1da177e2005-04-16 15:20:36 -070095#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>
Gerrit Renkerba4e58e2006-11-27 11:10:57 -0800107#include <net/udplite.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108#include <linux/skbuff.h>
109#include <net/sock.h>
110#include <net/raw.h>
111#include <net/icmp.h>
112#include <net/ipip.h>
113#include <net/inet_common.h>
114#include <net/xfrm.h>
115#ifdef CONFIG_IP_MROUTE
116#include <linux/mroute.h>
117#endif
118
Eric Dumazetba899662005-08-26 12:05:31 -0700119DEFINE_SNMP_STAT(struct linux_mib, net_statistics) __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121extern void ip_mc_drop_socket(struct sock *sk);
122
123/* The inetsw table contains everything that inet_create needs to
124 * build a new socket.
125 */
126static struct list_head inetsw[SOCK_MAX];
127static DEFINE_SPINLOCK(inetsw_lock);
128
129/* New destruction routine */
130
131void inet_sock_destruct(struct sock *sk)
132{
133 struct inet_sock *inet = inet_sk(sk);
134
135 __skb_queue_purge(&sk->sk_receive_queue);
136 __skb_queue_purge(&sk->sk_error_queue);
137
138 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
139 printk("Attempt to release TCP socket in state %d %p\n",
140 sk->sk_state, sk);
141 return;
142 }
143 if (!sock_flag(sk, SOCK_DEAD)) {
144 printk("Attempt to release alive inet socket %p\n", sk);
145 return;
146 }
147
148 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
149 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
150 BUG_TRAP(!sk->sk_wmem_queued);
151 BUG_TRAP(!sk->sk_forward_alloc);
152
Jesper Juhla51482b2005-11-08 09:41:34 -0800153 kfree(inet->opt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 dst_release(sk->sk_dst_cache);
Arnaldo Carvalho de Meloe6848972005-08-09 19:45:38 -0700155 sk_refcnt_debug_dec(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156}
157
158/*
159 * The routines beyond this point handle the behaviour of an AF_INET
160 * socket object. Mostly it punts to the subprotocols of IP to do
161 * the work.
162 */
163
164/*
165 * Automatically bind an unbound socket.
166 */
167
168static int inet_autobind(struct sock *sk)
169{
170 struct inet_sock *inet;
171 /* We may need to bind the socket. */
172 lock_sock(sk);
173 inet = inet_sk(sk);
174 if (!inet->num) {
175 if (sk->sk_prot->get_port(sk, 0)) {
176 release_sock(sk);
177 return -EAGAIN;
178 }
179 inet->sport = htons(inet->num);
180 }
181 release_sock(sk);
182 return 0;
183}
184
185/*
186 * Move a socket into listening state.
187 */
188int inet_listen(struct socket *sock, int backlog)
189{
190 struct sock *sk = sock->sk;
191 unsigned char old_state;
192 int err;
193
194 lock_sock(sk);
195
196 err = -EINVAL;
197 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
198 goto out;
199
200 old_state = sk->sk_state;
201 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
202 goto out;
203
204 /* Really, if the socket is already in listen state
205 * we can only allow the backlog to be adjusted.
206 */
207 if (old_state != TCP_LISTEN) {
Eric Dumazet72a3eff2006-11-16 02:30:37 -0800208 err = inet_csk_listen_start(sk, backlog);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 if (err)
210 goto out;
211 }
212 sk->sk_max_ack_backlog = backlog;
213 err = 0;
214
215out:
216 release_sock(sk);
217 return err;
218}
219
Eric Dumazetbe776282007-03-27 13:53:04 -0700220u32 inet_ehash_secret __read_mostly;
David S. Millerb3da2cf2007-03-23 11:40:27 -0700221EXPORT_SYMBOL(inet_ehash_secret);
222
Eric Dumazetbe776282007-03-27 13:53:04 -0700223/*
224 * inet_ehash_secret must be set exactly once
225 * Instead of using a dedicated spinlock, we (ab)use inetsw_lock
226 */
David S. Millerb3da2cf2007-03-23 11:40:27 -0700227void build_ehash_secret(void)
228{
Eric Dumazetbe776282007-03-27 13:53:04 -0700229 u32 rnd;
230 do {
231 get_random_bytes(&rnd, sizeof(rnd));
232 } while (rnd == 0);
233 spin_lock_bh(&inetsw_lock);
234 if (!inet_ehash_secret)
235 inet_ehash_secret = rnd;
236 spin_unlock_bh(&inetsw_lock);
David S. Millerb3da2cf2007-03-23 11:40:27 -0700237}
238EXPORT_SYMBOL(build_ehash_secret);
239
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240/*
241 * Create an inet socket.
242 */
243
244static int inet_create(struct socket *sock, int protocol)
245{
246 struct sock *sk;
247 struct list_head *p;
248 struct inet_protosw *answer;
249 struct inet_sock *inet;
250 struct proto *answer_prot;
251 unsigned char answer_flags;
252 char answer_no_check;
Arnaldo Carvalho de Melobb97d312005-08-09 20:19:14 -0700253 int try_loading_module = 0;
Herbert Xu86c8f9d2005-12-02 20:43:26 -0800254 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255
David S. Millerb3da2cf2007-03-23 11:40:27 -0700256 if (sock->type != SOCK_RAW &&
257 sock->type != SOCK_DGRAM &&
258 !inet_ehash_secret)
259 build_ehash_secret();
260
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 sock->state = SS_UNCONNECTED;
262
263 /* Look for the requested type/protocol pair. */
264 answer = NULL;
Arnaldo Carvalho de Melobb97d312005-08-09 20:19:14 -0700265lookup_protocol:
Herbert Xu86c8f9d2005-12-02 20:43:26 -0800266 err = -ESOCKTNOSUPPORT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 rcu_read_lock();
268 list_for_each_rcu(p, &inetsw[sock->type]) {
269 answer = list_entry(p, struct inet_protosw, list);
270
271 /* Check the non-wild match. */
272 if (protocol == answer->protocol) {
273 if (protocol != IPPROTO_IP)
274 break;
275 } else {
276 /* Check for the two wild cases. */
277 if (IPPROTO_IP == protocol) {
278 protocol = answer->protocol;
279 break;
280 }
281 if (IPPROTO_IP == answer->protocol)
282 break;
283 }
Herbert Xu86c8f9d2005-12-02 20:43:26 -0800284 err = -EPROTONOSUPPORT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 answer = NULL;
286 }
287
Arnaldo Carvalho de Melobb97d312005-08-09 20:19:14 -0700288 if (unlikely(answer == NULL)) {
289 if (try_loading_module < 2) {
290 rcu_read_unlock();
291 /*
292 * Be more specific, e.g. net-pf-2-proto-132-type-1
293 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
294 */
295 if (++try_loading_module == 1)
296 request_module("net-pf-%d-proto-%d-type-%d",
297 PF_INET, protocol, sock->type);
298 /*
299 * Fall back to generic, e.g. net-pf-2-proto-132
300 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
301 */
302 else
303 request_module("net-pf-%d-proto-%d",
304 PF_INET, protocol);
305 goto lookup_protocol;
306 } else
307 goto out_rcu_unlock;
308 }
309
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 err = -EPERM;
311 if (answer->capability > 0 && !capable(answer->capability))
312 goto out_rcu_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313
314 sock->ops = answer->ops;
315 answer_prot = answer->prot;
316 answer_no_check = answer->no_check;
317 answer_flags = answer->flags;
318 rcu_read_unlock();
319
320 BUG_TRAP(answer_prot->slab != NULL);
321
322 err = -ENOBUFS;
323 sk = sk_alloc(PF_INET, GFP_KERNEL, answer_prot, 1);
324 if (sk == NULL)
325 goto out;
326
327 err = 0;
328 sk->sk_no_check = answer_no_check;
329 if (INET_PROTOSW_REUSE & answer_flags)
330 sk->sk_reuse = 1;
331
332 inet = inet_sk(sk);
Paul Moore469de9b2007-01-09 14:37:06 -0800333 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334
335 if (SOCK_RAW == sock->type) {
336 inet->num = protocol;
337 if (IPPROTO_RAW == protocol)
338 inet->hdrincl = 1;
339 }
340
341 if (ipv4_config.no_pmtu_disc)
342 inet->pmtudisc = IP_PMTUDISC_DONT;
343 else
344 inet->pmtudisc = IP_PMTUDISC_WANT;
345
346 inet->id = 0;
347
348 sock_init_data(sock, sk);
349
350 sk->sk_destruct = inet_sock_destruct;
351 sk->sk_family = PF_INET;
352 sk->sk_protocol = protocol;
353 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
354
355 inet->uc_ttl = -1;
356 inet->mc_loop = 1;
357 inet->mc_ttl = 1;
358 inet->mc_index = 0;
359 inet->mc_list = NULL;
360
Arnaldo Carvalho de Meloe6848972005-08-09 19:45:38 -0700361 sk_refcnt_debug_inc(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362
363 if (inet->num) {
364 /* It assumes that any protocol which allows
365 * the user to assign a number at socket
366 * creation time automatically
367 * shares.
368 */
369 inet->sport = htons(inet->num);
370 /* Add to protocol hash chains. */
371 sk->sk_prot->hash(sk);
372 }
373
374 if (sk->sk_prot->init) {
375 err = sk->sk_prot->init(sk);
376 if (err)
377 sk_common_release(sk);
378 }
379out:
380 return err;
381out_rcu_unlock:
382 rcu_read_unlock();
383 goto out;
384}
385
386
387/*
388 * The peer socket should always be NULL (or else). When we call this
389 * function we are destroying the object and from then on nobody
390 * should refer to it.
391 */
392int inet_release(struct socket *sock)
393{
394 struct sock *sk = sock->sk;
395
396 if (sk) {
397 long timeout;
398
399 /* Applications forget to leave groups before exiting */
400 ip_mc_drop_socket(sk);
401
402 /* If linger is set, we don't return until the close
403 * is complete. Otherwise we return immediately. The
404 * actually closing is done the same either way.
405 *
406 * If the close is due to the process exiting, we never
407 * linger..
408 */
409 timeout = 0;
410 if (sock_flag(sk, SOCK_LINGER) &&
411 !(current->flags & PF_EXITING))
412 timeout = sk->sk_lingertime;
413 sock->sk = NULL;
414 sk->sk_prot->close(sk, timeout);
415 }
416 return 0;
417}
418
419/* It is off by default, see below. */
Brian Haleyab32ea52006-09-22 14:15:41 -0700420int sysctl_ip_nonlocal_bind __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421
422int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
423{
424 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
425 struct sock *sk = sock->sk;
426 struct inet_sock *inet = inet_sk(sk);
427 unsigned short snum;
428 int chk_addr_ret;
429 int err;
430
431 /* If the socket has its own bind function then use it. (RAW) */
432 if (sk->sk_prot->bind) {
433 err = sk->sk_prot->bind(sk, uaddr, addr_len);
434 goto out;
435 }
436 err = -EINVAL;
437 if (addr_len < sizeof(struct sockaddr_in))
438 goto out;
439
440 chk_addr_ret = inet_addr_type(addr->sin_addr.s_addr);
441
442 /* Not specified by any standard per-se, however it breaks too
443 * many applications when removed. It is unfortunate since
444 * allowing applications to make a non-local bind solves
445 * several problems with systems using dynamic addressing.
446 * (ie. your servers still start up even if your ISDN link
447 * is temporarily down)
448 */
449 err = -EADDRNOTAVAIL;
450 if (!sysctl_ip_nonlocal_bind &&
451 !inet->freebind &&
452 addr->sin_addr.s_addr != INADDR_ANY &&
453 chk_addr_ret != RTN_LOCAL &&
454 chk_addr_ret != RTN_MULTICAST &&
455 chk_addr_ret != RTN_BROADCAST)
456 goto out;
457
458 snum = ntohs(addr->sin_port);
459 err = -EACCES;
460 if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
461 goto out;
462
463 /* We keep a pair of addresses. rcv_saddr is the one
464 * used by hash lookups, and saddr is used for transmit.
465 *
466 * In the BSD API these are the same except where it
467 * would be illegal to use them (multicast/broadcast) in
468 * which case the sending device address is used.
469 */
470 lock_sock(sk);
471
472 /* Check these errors (active socket, double bind). */
473 err = -EINVAL;
474 if (sk->sk_state != TCP_CLOSE || inet->num)
475 goto out_release_sock;
476
477 inet->rcv_saddr = inet->saddr = addr->sin_addr.s_addr;
478 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
479 inet->saddr = 0; /* Use device */
480
481 /* Make sure we are allowed to bind here. */
482 if (sk->sk_prot->get_port(sk, snum)) {
483 inet->saddr = inet->rcv_saddr = 0;
484 err = -EADDRINUSE;
485 goto out_release_sock;
486 }
487
488 if (inet->rcv_saddr)
489 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
490 if (snum)
491 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
492 inet->sport = htons(inet->num);
493 inet->daddr = 0;
494 inet->dport = 0;
495 sk_dst_reset(sk);
496 err = 0;
497out_release_sock:
498 release_sock(sk);
499out:
500 return err;
501}
502
503int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
504 int addr_len, int flags)
505{
506 struct sock *sk = sock->sk;
507
508 if (uaddr->sa_family == AF_UNSPEC)
509 return sk->sk_prot->disconnect(sk, flags);
510
511 if (!inet_sk(sk)->num && inet_autobind(sk))
512 return -EAGAIN;
513 return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
514}
515
516static long inet_wait_for_connect(struct sock *sk, long timeo)
517{
518 DEFINE_WAIT(wait);
519
520 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
521
522 /* Basic assumption: if someone sets sk->sk_err, he _must_
523 * change state of the socket from TCP_SYN_*.
524 * Connect() does not allow to get error notifications
525 * without closing the socket.
526 */
527 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
528 release_sock(sk);
529 timeo = schedule_timeout(timeo);
530 lock_sock(sk);
531 if (signal_pending(current) || !timeo)
532 break;
533 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
534 }
535 finish_wait(sk->sk_sleep, &wait);
536 return timeo;
537}
538
539/*
540 * Connect to a remote host. There is regrettably still a little
541 * TCP 'magic' in here.
542 */
543int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
544 int addr_len, int flags)
545{
546 struct sock *sk = sock->sk;
547 int err;
548 long timeo;
549
550 lock_sock(sk);
551
552 if (uaddr->sa_family == AF_UNSPEC) {
553 err = sk->sk_prot->disconnect(sk, flags);
554 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
555 goto out;
556 }
557
558 switch (sock->state) {
559 default:
560 err = -EINVAL;
561 goto out;
562 case SS_CONNECTED:
563 err = -EISCONN;
564 goto out;
565 case SS_CONNECTING:
566 err = -EALREADY;
567 /* Fall out of switch with err, set for this state */
568 break;
569 case SS_UNCONNECTED:
570 err = -EISCONN;
571 if (sk->sk_state != TCP_CLOSE)
572 goto out;
573
574 err = sk->sk_prot->connect(sk, uaddr, addr_len);
575 if (err < 0)
576 goto out;
577
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900578 sock->state = SS_CONNECTING;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579
580 /* Just entered SS_CONNECTING state; the only
581 * difference is that return value in non-blocking
582 * case is EINPROGRESS, rather than EALREADY.
583 */
584 err = -EINPROGRESS;
585 break;
586 }
587
588 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
589
590 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
591 /* Error code is set above */
592 if (!timeo || !inet_wait_for_connect(sk, timeo))
593 goto out;
594
595 err = sock_intr_errno(timeo);
596 if (signal_pending(current))
597 goto out;
598 }
599
600 /* Connection was closed by RST, timeout, ICMP error
601 * or another process disconnected us.
602 */
603 if (sk->sk_state == TCP_CLOSE)
604 goto sock_error;
605
606 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
607 * and error was received after socket entered established state.
608 * Hence, it is handled normally after connect() return successfully.
609 */
610
611 sock->state = SS_CONNECTED;
612 err = 0;
613out:
614 release_sock(sk);
615 return err;
616
617sock_error:
618 err = sock_error(sk) ? : -ECONNABORTED;
619 sock->state = SS_UNCONNECTED;
620 if (sk->sk_prot->disconnect(sk, flags))
621 sock->state = SS_DISCONNECTING;
622 goto out;
623}
624
625/*
626 * Accept a pending connection. The TCP layer now gives BSD semantics.
627 */
628
629int inet_accept(struct socket *sock, struct socket *newsock, int flags)
630{
631 struct sock *sk1 = sock->sk;
632 int err = -EINVAL;
633 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
634
635 if (!sk2)
636 goto do_err;
637
638 lock_sock(sk2);
639
640 BUG_TRAP((1 << sk2->sk_state) &
641 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE));
642
643 sock_graft(sk2, newsock);
644
645 newsock->state = SS_CONNECTED;
646 err = 0;
647 release_sock(sk2);
648do_err:
649 return err;
650}
651
652
653/*
654 * This does both peername and sockname.
655 */
656int inet_getname(struct socket *sock, struct sockaddr *uaddr,
657 int *uaddr_len, int peer)
658{
659 struct sock *sk = sock->sk;
660 struct inet_sock *inet = inet_sk(sk);
661 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
662
663 sin->sin_family = AF_INET;
664 if (peer) {
665 if (!inet->dport ||
666 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
667 peer == 1))
668 return -ENOTCONN;
669 sin->sin_port = inet->dport;
670 sin->sin_addr.s_addr = inet->daddr;
671 } else {
Al Viro714e85b2006-11-14 20:51:49 -0800672 __be32 addr = inet->rcv_saddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 if (!addr)
674 addr = inet->saddr;
675 sin->sin_port = inet->sport;
676 sin->sin_addr.s_addr = addr;
677 }
678 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
679 *uaddr_len = sizeof(*sin);
680 return 0;
681}
682
683int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
684 size_t size)
685{
686 struct sock *sk = sock->sk;
687
688 /* We may need to bind the socket. */
689 if (!inet_sk(sk)->num && inet_autobind(sk))
690 return -EAGAIN;
691
692 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
693}
694
695
696static ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
697{
698 struct sock *sk = sock->sk;
699
700 /* We may need to bind the socket. */
701 if (!inet_sk(sk)->num && inet_autobind(sk))
702 return -EAGAIN;
703
704 if (sk->sk_prot->sendpage)
705 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
706 return sock_no_sendpage(sock, page, offset, size, flags);
707}
708
709
710int inet_shutdown(struct socket *sock, int how)
711{
712 struct sock *sk = sock->sk;
713 int err = 0;
714
715 /* This should really check to make sure
716 * the socket is a TCP socket. (WHY AC...)
717 */
718 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
719 1->2 bit 2 snds.
720 2->3 */
721 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
722 return -EINVAL;
723
724 lock_sock(sk);
725 if (sock->state == SS_CONNECTING) {
726 if ((1 << sk->sk_state) &
727 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
728 sock->state = SS_DISCONNECTING;
729 else
730 sock->state = SS_CONNECTED;
731 }
732
733 switch (sk->sk_state) {
734 case TCP_CLOSE:
735 err = -ENOTCONN;
736 /* Hack to wake up other listeners, who can poll for
737 POLLHUP, even on eg. unconnected UDP sockets -- RR */
738 default:
739 sk->sk_shutdown |= how;
740 if (sk->sk_prot->shutdown)
741 sk->sk_prot->shutdown(sk, how);
742 break;
743
744 /* Remaining two branches are temporary solution for missing
745 * close() in multithreaded environment. It is _not_ a good idea,
746 * but we have no choice until close() is repaired at VFS level.
747 */
748 case TCP_LISTEN:
749 if (!(how & RCV_SHUTDOWN))
750 break;
751 /* Fall through */
752 case TCP_SYN_SENT:
753 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
754 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
755 break;
756 }
757
758 /* Wake up anyone sleeping in poll. */
759 sk->sk_state_change(sk);
760 release_sock(sk);
761 return err;
762}
763
764/*
765 * ioctl() calls you can issue on an INET socket. Most of these are
766 * device configuration and stuff and very rarely used. Some ioctls
767 * pass on to the socket itself.
768 *
769 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
770 * loads the devconfigure module does its configuring and unloads it.
771 * There's a good 20K of config code hanging around the kernel.
772 */
773
774int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
775{
776 struct sock *sk = sock->sk;
777 int err = 0;
778
779 switch (cmd) {
780 case SIOCGSTAMP:
781 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
782 break;
Eric Dumazetae40eb12007-03-18 17:33:16 -0700783 case SIOCGSTAMPNS:
784 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
785 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 case SIOCADDRT:
787 case SIOCDELRT:
788 case SIOCRTMSG:
789 err = ip_rt_ioctl(cmd, (void __user *)arg);
790 break;
791 case SIOCDARP:
792 case SIOCGARP:
793 case SIOCSARP:
794 err = arp_ioctl(cmd, (void __user *)arg);
795 break;
796 case SIOCGIFADDR:
797 case SIOCSIFADDR:
798 case SIOCGIFBRDADDR:
799 case SIOCSIFBRDADDR:
800 case SIOCGIFNETMASK:
801 case SIOCSIFNETMASK:
802 case SIOCGIFDSTADDR:
803 case SIOCSIFDSTADDR:
804 case SIOCSIFPFLAGS:
805 case SIOCGIFPFLAGS:
806 case SIOCSIFFLAGS:
807 err = devinet_ioctl(cmd, (void __user *)arg);
808 break;
809 default:
Christoph Hellwigb5e5fa52006-01-03 14:18:33 -0800810 if (sk->sk_prot->ioctl)
811 err = sk->sk_prot->ioctl(sk, cmd, arg);
812 else
813 err = -ENOIOCTLCMD;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 break;
815 }
816 return err;
817}
818
Eric Dumazet90ddc4f2005-12-22 12:49:22 -0800819const struct proto_ops inet_stream_ops = {
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -0800820 .family = PF_INET,
821 .owner = THIS_MODULE,
822 .release = inet_release,
823 .bind = inet_bind,
824 .connect = inet_stream_connect,
825 .socketpair = sock_no_socketpair,
826 .accept = inet_accept,
827 .getname = inet_getname,
828 .poll = tcp_poll,
829 .ioctl = inet_ioctl,
830 .listen = inet_listen,
831 .shutdown = inet_shutdown,
832 .setsockopt = sock_common_setsockopt,
833 .getsockopt = sock_common_getsockopt,
David S. Miller3516ffb2007-08-02 19:23:56 -0700834 .sendmsg = tcp_sendmsg,
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -0800835 .recvmsg = sock_common_recvmsg,
836 .mmap = sock_no_mmap,
837 .sendpage = tcp_sendpage,
Dmitry Mishin3fdadf72006-03-20 22:45:21 -0800838#ifdef CONFIG_COMPAT
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -0800839 .compat_setsockopt = compat_sock_common_setsockopt,
840 .compat_getsockopt = compat_sock_common_getsockopt,
Dmitry Mishin3fdadf72006-03-20 22:45:21 -0800841#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842};
843
Eric Dumazet90ddc4f2005-12-22 12:49:22 -0800844const struct proto_ops inet_dgram_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 = udp_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
869/*
870 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
871 * udp_poll
872 */
Eric Dumazet90ddc4f2005-12-22 12:49:22 -0800873static const struct proto_ops inet_sockraw_ops = {
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -0800874 .family = PF_INET,
875 .owner = THIS_MODULE,
876 .release = inet_release,
877 .bind = inet_bind,
878 .connect = inet_dgram_connect,
879 .socketpair = sock_no_socketpair,
880 .accept = sock_no_accept,
881 .getname = inet_getname,
882 .poll = datagram_poll,
883 .ioctl = inet_ioctl,
884 .listen = sock_no_listen,
885 .shutdown = inet_shutdown,
886 .setsockopt = sock_common_setsockopt,
887 .getsockopt = sock_common_getsockopt,
888 .sendmsg = inet_sendmsg,
889 .recvmsg = sock_common_recvmsg,
890 .mmap = sock_no_mmap,
891 .sendpage = inet_sendpage,
Dmitry Mishin3fdadf72006-03-20 22:45:21 -0800892#ifdef CONFIG_COMPAT
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -0800893 .compat_setsockopt = compat_sock_common_setsockopt,
894 .compat_getsockopt = compat_sock_common_getsockopt,
Dmitry Mishin3fdadf72006-03-20 22:45:21 -0800895#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896};
897
898static struct net_proto_family inet_family_ops = {
899 .family = PF_INET,
900 .create = inet_create,
901 .owner = THIS_MODULE,
902};
903
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904/* Upon startup we insert all the elements in inetsw_array[] into
905 * the linked list inetsw.
906 */
907static struct inet_protosw inetsw_array[] =
908{
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900909 {
910 .type = SOCK_STREAM,
911 .protocol = IPPROTO_TCP,
912 .prot = &tcp_prot,
913 .ops = &inet_stream_ops,
914 .capability = -1,
915 .no_check = 0,
916 .flags = INET_PROTOSW_PERMANENT |
Arnaldo Carvalho de Melod83d8462005-12-13 23:26:10 -0800917 INET_PROTOSW_ICSK,
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900918 },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900920 {
921 .type = SOCK_DGRAM,
922 .protocol = IPPROTO_UDP,
923 .prot = &udp_prot,
924 .ops = &inet_dgram_ops,
925 .capability = -1,
926 .no_check = UDP_CSUM_DEFAULT,
927 .flags = INET_PROTOSW_PERMANENT,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 },
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900929
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930
931 {
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900932 .type = SOCK_RAW,
933 .protocol = IPPROTO_IP, /* wild card */
934 .prot = &raw_prot,
935 .ops = &inet_sockraw_ops,
936 .capability = CAP_NET_RAW,
937 .no_check = UDP_CSUM_DEFAULT,
938 .flags = INET_PROTOSW_REUSE,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 }
940};
941
942#define INETSW_ARRAY_LEN (sizeof(inetsw_array) / sizeof(struct inet_protosw))
943
944void inet_register_protosw(struct inet_protosw *p)
945{
946 struct list_head *lh;
947 struct inet_protosw *answer;
948 int protocol = p->protocol;
949 struct list_head *last_perm;
950
951 spin_lock_bh(&inetsw_lock);
952
953 if (p->type >= SOCK_MAX)
954 goto out_illegal;
955
956 /* If we are trying to override a permanent protocol, bail. */
957 answer = NULL;
958 last_perm = &inetsw[p->type];
959 list_for_each(lh, &inetsw[p->type]) {
960 answer = list_entry(lh, struct inet_protosw, list);
961
962 /* Check only the non-wild match. */
963 if (INET_PROTOSW_PERMANENT & answer->flags) {
964 if (protocol == answer->protocol)
965 break;
966 last_perm = lh;
967 }
968
969 answer = NULL;
970 }
971 if (answer)
972 goto out_permanent;
973
974 /* Add the new entry after the last permanent entry if any, so that
975 * the new entry does not override a permanent entry when matched with
976 * a wild-card protocol. But it is allowed to override any existing
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900977 * non-permanent entry. This means that when we remove this entry, the
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 * system automatically returns to the old behavior.
979 */
980 list_add_rcu(&p->list, last_perm);
981out:
982 spin_unlock_bh(&inetsw_lock);
983
984 synchronize_net();
985
986 return;
987
988out_permanent:
989 printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
990 protocol);
991 goto out;
992
993out_illegal:
994 printk(KERN_ERR
995 "Ignoring attempt to register invalid socket type %d.\n",
996 p->type);
997 goto out;
998}
999
1000void inet_unregister_protosw(struct inet_protosw *p)
1001{
1002 if (INET_PROTOSW_PERMANENT & p->flags) {
1003 printk(KERN_ERR
1004 "Attempt to unregister permanent protocol %d.\n",
1005 p->protocol);
1006 } else {
1007 spin_lock_bh(&inetsw_lock);
1008 list_del_rcu(&p->list);
1009 spin_unlock_bh(&inetsw_lock);
1010
1011 synchronize_net();
1012 }
1013}
1014
Arnaldo Carvalho de Melo32519f12005-08-09 19:50:02 -07001015/*
1016 * Shall we try to damage output packets if routing dev changes?
1017 */
1018
Brian Haleyab32ea52006-09-22 14:15:41 -07001019int sysctl_ip_dynaddr __read_mostly;
Arnaldo Carvalho de Melo32519f12005-08-09 19:50:02 -07001020
1021static int inet_sk_reselect_saddr(struct sock *sk)
1022{
1023 struct inet_sock *inet = inet_sk(sk);
1024 int err;
1025 struct rtable *rt;
Al Viro714e85b2006-11-14 20:51:49 -08001026 __be32 old_saddr = inet->saddr;
1027 __be32 new_saddr;
Al Virobada8ad2006-09-26 21:27:15 -07001028 __be32 daddr = inet->daddr;
Arnaldo Carvalho de Melo32519f12005-08-09 19:50:02 -07001029
1030 if (inet->opt && inet->opt->srr)
1031 daddr = inet->opt->faddr;
1032
1033 /* Query new route. */
1034 err = ip_route_connect(&rt, daddr, 0,
1035 RT_CONN_FLAGS(sk),
1036 sk->sk_bound_dev_if,
1037 sk->sk_protocol,
David S. Miller8eb90862007-02-08 02:09:21 -08001038 inet->sport, inet->dport, sk, 0);
Arnaldo Carvalho de Melo32519f12005-08-09 19:50:02 -07001039 if (err)
1040 return err;
1041
1042 sk_setup_caps(sk, &rt->u.dst);
1043
1044 new_saddr = rt->rt_src;
1045
1046 if (new_saddr == old_saddr)
1047 return 0;
1048
1049 if (sysctl_ip_dynaddr > 1) {
1050 printk(KERN_INFO "%s(): shifting inet->"
1051 "saddr from %d.%d.%d.%d to %d.%d.%d.%d\n",
1052 __FUNCTION__,
1053 NIPQUAD(old_saddr),
1054 NIPQUAD(new_saddr));
1055 }
1056
1057 inet->saddr = inet->rcv_saddr = new_saddr;
1058
1059 /*
1060 * XXX The only one ugly spot where we need to
1061 * XXX really change the sockets identity after
1062 * XXX it has entered the hashes. -DaveM
1063 *
1064 * Besides that, it does not check for connection
1065 * uniqueness. Wait for troubles.
1066 */
1067 __sk_prot_rehash(sk);
1068 return 0;
1069}
1070
1071int inet_sk_rebuild_header(struct sock *sk)
1072{
1073 struct inet_sock *inet = inet_sk(sk);
1074 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
Al Viro3ca3c682006-09-27 18:28:07 -07001075 __be32 daddr;
Arnaldo Carvalho de Melo32519f12005-08-09 19:50:02 -07001076 int err;
1077
1078 /* Route is OK, nothing to do. */
1079 if (rt)
1080 return 0;
1081
1082 /* Reroute. */
1083 daddr = inet->daddr;
1084 if (inet->opt && inet->opt->srr)
1085 daddr = inet->opt->faddr;
1086{
1087 struct flowi fl = {
1088 .oif = sk->sk_bound_dev_if,
1089 .nl_u = {
1090 .ip4_u = {
1091 .daddr = daddr,
1092 .saddr = inet->saddr,
1093 .tos = RT_CONN_FLAGS(sk),
1094 },
1095 },
1096 .proto = sk->sk_protocol,
1097 .uli_u = {
1098 .ports = {
1099 .sport = inet->sport,
1100 .dport = inet->dport,
1101 },
1102 },
1103 };
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001104
Venkat Yekkiralabeb8d132006-08-04 23:12:42 -07001105 security_sk_classify_flow(sk, &fl);
Arnaldo Carvalho de Melo32519f12005-08-09 19:50:02 -07001106 err = ip_route_output_flow(&rt, &fl, sk, 0);
1107}
1108 if (!err)
1109 sk_setup_caps(sk, &rt->u.dst);
1110 else {
1111 /* Routing failed... */
1112 sk->sk_route_caps = 0;
1113 /*
1114 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1115 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1116 */
1117 if (!sysctl_ip_dynaddr ||
1118 sk->sk_state != TCP_SYN_SENT ||
1119 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1120 (err = inet_sk_reselect_saddr(sk)) != 0)
1121 sk->sk_err_soft = -err;
1122 }
1123
1124 return err;
1125}
1126
1127EXPORT_SYMBOL(inet_sk_rebuild_header);
1128
Herbert Xua430a432006-07-08 13:34:56 -07001129static int inet_gso_send_check(struct sk_buff *skb)
1130{
1131 struct iphdr *iph;
1132 struct net_protocol *ops;
1133 int proto;
1134 int ihl;
1135 int err = -EINVAL;
1136
1137 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1138 goto out;
1139
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001140 iph = ip_hdr(skb);
Herbert Xua430a432006-07-08 13:34:56 -07001141 ihl = iph->ihl * 4;
1142 if (ihl < sizeof(*iph))
1143 goto out;
1144
1145 if (unlikely(!pskb_may_pull(skb, ihl)))
1146 goto out;
1147
Arnaldo Carvalho de Melobadff6d2007-03-13 13:06:52 -03001148 __skb_pull(skb, ihl);
1149 skb_reset_transport_header(skb);
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001150 iph = ip_hdr(skb);
Herbert Xua430a432006-07-08 13:34:56 -07001151 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1152 err = -EPROTONOSUPPORT;
1153
1154 rcu_read_lock();
1155 ops = rcu_dereference(inet_protos[proto]);
1156 if (likely(ops && ops->gso_send_check))
1157 err = ops->gso_send_check(skb);
1158 rcu_read_unlock();
1159
1160out:
1161 return err;
1162}
1163
Herbert Xu576a30e2006-06-27 13:22:38 -07001164static struct sk_buff *inet_gso_segment(struct sk_buff *skb, int features)
Herbert Xuf4c50d92006-06-22 03:02:40 -07001165{
1166 struct sk_buff *segs = ERR_PTR(-EINVAL);
1167 struct iphdr *iph;
1168 struct net_protocol *ops;
1169 int proto;
1170 int ihl;
1171 int id;
1172
Stephen Hemmingerd212f872007-06-27 00:47:37 -07001173 if (!(features & NETIF_F_V4_CSUM))
1174 features &= ~NETIF_F_SG;
1175
Herbert Xubbcf4672006-07-03 19:38:35 -07001176 if (unlikely(skb_shinfo(skb)->gso_type &
1177 ~(SKB_GSO_TCPV4 |
1178 SKB_GSO_UDP |
1179 SKB_GSO_DODGY |
1180 SKB_GSO_TCP_ECN |
1181 0)))
1182 goto out;
1183
1184 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
Herbert Xuf4c50d92006-06-22 03:02:40 -07001185 goto out;
1186
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001187 iph = ip_hdr(skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07001188 ihl = iph->ihl * 4;
1189 if (ihl < sizeof(*iph))
1190 goto out;
1191
Herbert Xubbcf4672006-07-03 19:38:35 -07001192 if (unlikely(!pskb_may_pull(skb, ihl)))
Herbert Xuf4c50d92006-06-22 03:02:40 -07001193 goto out;
1194
Arnaldo Carvalho de Melobadff6d2007-03-13 13:06:52 -03001195 __skb_pull(skb, ihl);
1196 skb_reset_transport_header(skb);
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001197 iph = ip_hdr(skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07001198 id = ntohs(iph->id);
1199 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1200 segs = ERR_PTR(-EPROTONOSUPPORT);
1201
1202 rcu_read_lock();
1203 ops = rcu_dereference(inet_protos[proto]);
Herbert Xubbcf4672006-07-03 19:38:35 -07001204 if (likely(ops && ops->gso_segment))
Herbert Xu576a30e2006-06-27 13:22:38 -07001205 segs = ops->gso_segment(skb, features);
Herbert Xuf4c50d92006-06-22 03:02:40 -07001206 rcu_read_unlock();
1207
Herbert Xu576a30e2006-06-27 13:22:38 -07001208 if (!segs || unlikely(IS_ERR(segs)))
Herbert Xuf4c50d92006-06-22 03:02:40 -07001209 goto out;
1210
1211 skb = segs;
1212 do {
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001213 iph = ip_hdr(skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07001214 iph->id = htons(id++);
1215 iph->tot_len = htons(skb->len - skb->mac_len);
1216 iph->check = 0;
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -07001217 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
Herbert Xuf4c50d92006-06-22 03:02:40 -07001218 } while ((skb = skb->next));
1219
1220out:
1221 return segs;
1222}
1223
Herbert Xu5e0f0432007-04-24 21:53:35 -07001224unsigned long snmp_fold_field(void *mib[], int offt)
1225{
1226 unsigned long res = 0;
1227 int i;
1228
1229 for_each_possible_cpu(i) {
1230 res += *(((unsigned long *) per_cpu_ptr(mib[0], i)) + offt);
1231 res += *(((unsigned long *) per_cpu_ptr(mib[1], i)) + offt);
1232 }
1233 return res;
1234}
1235EXPORT_SYMBOL_GPL(snmp_fold_field);
1236
1237int snmp_mib_init(void *ptr[2], size_t mibsize, size_t mibalign)
1238{
1239 BUG_ON(ptr == NULL);
1240 ptr[0] = __alloc_percpu(mibsize);
1241 if (!ptr[0])
1242 goto err0;
1243 ptr[1] = __alloc_percpu(mibsize);
1244 if (!ptr[1])
1245 goto err1;
1246 return 0;
1247err1:
1248 free_percpu(ptr[0]);
1249 ptr[0] = NULL;
1250err0:
1251 return -ENOMEM;
1252}
1253EXPORT_SYMBOL_GPL(snmp_mib_init);
1254
1255void snmp_mib_free(void *ptr[2])
1256{
1257 BUG_ON(ptr == NULL);
1258 free_percpu(ptr[0]);
1259 free_percpu(ptr[1]);
1260 ptr[0] = ptr[1] = NULL;
1261}
1262EXPORT_SYMBOL_GPL(snmp_mib_free);
1263
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264#ifdef CONFIG_IP_MULTICAST
1265static struct net_protocol igmp_protocol = {
1266 .handler = igmp_rcv,
1267};
1268#endif
1269
1270static struct net_protocol tcp_protocol = {
1271 .handler = tcp_v4_rcv,
1272 .err_handler = tcp_v4_err,
Herbert Xua430a432006-07-08 13:34:56 -07001273 .gso_send_check = tcp_v4_gso_send_check,
Herbert Xuf4c50d92006-06-22 03:02:40 -07001274 .gso_segment = tcp_tso_segment,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 .no_policy = 1,
1276};
1277
1278static struct net_protocol udp_protocol = {
1279 .handler = udp_rcv,
1280 .err_handler = udp_err,
1281 .no_policy = 1,
1282};
1283
1284static struct net_protocol icmp_protocol = {
1285 .handler = icmp_rcv,
1286};
1287
1288static int __init init_ipv4_mibs(void)
1289{
YOSHIFUJI Hideaki33490172007-04-20 15:57:15 -07001290 if (snmp_mib_init((void **)net_statistics,
1291 sizeof(struct linux_mib),
1292 __alignof__(struct linux_mib)) < 0)
1293 goto err_net_mib;
1294 if (snmp_mib_init((void **)ip_statistics,
1295 sizeof(struct ipstats_mib),
1296 __alignof__(struct ipstats_mib)) < 0)
1297 goto err_ip_mib;
1298 if (snmp_mib_init((void **)icmp_statistics,
1299 sizeof(struct icmp_mib),
1300 __alignof__(struct icmp_mib)) < 0)
1301 goto err_icmp_mib;
1302 if (snmp_mib_init((void **)tcp_statistics,
1303 sizeof(struct tcp_mib),
1304 __alignof__(struct tcp_mib)) < 0)
1305 goto err_tcp_mib;
1306 if (snmp_mib_init((void **)udp_statistics,
1307 sizeof(struct udp_mib),
1308 __alignof__(struct udp_mib)) < 0)
1309 goto err_udp_mib;
1310 if (snmp_mib_init((void **)udplite_statistics,
1311 sizeof(struct udp_mib),
1312 __alignof__(struct udp_mib)) < 0)
1313 goto err_udplite_mib;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314
YOSHIFUJI Hideaki33490172007-04-20 15:57:15 -07001315 tcp_mib_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316
1317 return 0;
YOSHIFUJI Hideaki33490172007-04-20 15:57:15 -07001318
1319err_udplite_mib:
1320 snmp_mib_free((void **)udp_statistics);
1321err_udp_mib:
1322 snmp_mib_free((void **)tcp_statistics);
1323err_tcp_mib:
1324 snmp_mib_free((void **)icmp_statistics);
1325err_icmp_mib:
1326 snmp_mib_free((void **)ip_statistics);
1327err_ip_mib:
1328 snmp_mib_free((void **)net_statistics);
1329err_net_mib:
1330 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331}
1332
1333static int ipv4_proc_init(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334
Herbert Xu30e224d2005-07-05 14:40:10 -07001335/*
1336 * IP protocol layer initialiser
1337 */
1338
1339static struct packet_type ip_packet_type = {
1340 .type = __constant_htons(ETH_P_IP),
1341 .func = ip_rcv,
Herbert Xua430a432006-07-08 13:34:56 -07001342 .gso_send_check = inet_gso_send_check,
Herbert Xuf4c50d92006-06-22 03:02:40 -07001343 .gso_segment = inet_gso_segment,
Herbert Xu30e224d2005-07-05 14:40:10 -07001344};
1345
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346static int __init inet_init(void)
1347{
1348 struct sk_buff *dummy_skb;
1349 struct inet_protosw *q;
1350 struct list_head *r;
1351 int rc = -EINVAL;
1352
YOSHIFUJI Hideakief047f52006-09-01 00:29:06 -07001353 BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354
1355 rc = proto_register(&tcp_prot, 1);
1356 if (rc)
1357 goto out;
1358
1359 rc = proto_register(&udp_prot, 1);
1360 if (rc)
1361 goto out_unregister_tcp_proto;
1362
1363 rc = proto_register(&raw_prot, 1);
1364 if (rc)
1365 goto out_unregister_udp_proto;
1366
1367 /*
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001368 * Tell SOCKET that we are alive...
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 */
1370
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001371 (void)sock_register(&inet_family_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372
1373 /*
1374 * Add all the base protocols.
1375 */
1376
1377 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1378 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1379 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1380 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1381 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1382 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1383#ifdef CONFIG_IP_MULTICAST
1384 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1385 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1386#endif
1387
1388 /* Register the socket-side information for inet_create. */
1389 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1390 INIT_LIST_HEAD(r);
1391
1392 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1393 inet_register_protosw(q);
1394
1395 /*
1396 * Set the ARP module up
1397 */
1398
1399 arp_init();
1400
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001401 /*
1402 * Set the IP module up
1403 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404
1405 ip_init();
1406
1407 tcp_v4_init(&inet_family_ops);
1408
1409 /* Setup TCP slab cache for open requests. */
1410 tcp_init();
1411
Gerrit Renkerba4e58e2006-11-27 11:10:57 -08001412 /* Add UDP-Lite (RFC 3828) */
1413 udplite4_register();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414
1415 /*
1416 * Set the ICMP layer up
1417 */
1418
1419 icmp_init(&inet_family_ops);
1420
1421 /*
1422 * Initialise the multicast router
1423 */
1424#if defined(CONFIG_IP_MROUTE)
1425 ip_mr_init();
1426#endif
1427 /*
1428 * Initialise per-cpu ipv4 mibs
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001429 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430
Stephen Hemminger132adf52007-03-08 20:44:43 -08001431 if (init_ipv4_mibs())
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n"); ;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001433
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 ipv4_proc_init();
1435
1436 ipfrag_init();
1437
Herbert Xu30e224d2005-07-05 14:40:10 -07001438 dev_add_pack(&ip_packet_type);
1439
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 rc = 0;
1441out:
1442 return rc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443out_unregister_udp_proto:
1444 proto_unregister(&udp_prot);
Olaf Kirch321efff2006-09-27 16:33:45 -07001445out_unregister_tcp_proto:
1446 proto_unregister(&tcp_prot);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 goto out;
1448}
1449
Heiko Carstensa536e072006-04-28 15:19:17 -07001450fs_initcall(inet_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451
1452/* ------------------------------------------------------------------------ */
1453
1454#ifdef CONFIG_PROC_FS
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455static int __init ipv4_proc_init(void)
1456{
1457 int rc = 0;
1458
1459 if (raw_proc_init())
1460 goto out_raw;
1461 if (tcp4_proc_init())
1462 goto out_tcp;
1463 if (udp4_proc_init())
1464 goto out_udp;
1465 if (fib_proc_init())
1466 goto out_fib;
1467 if (ip_misc_proc_init())
1468 goto out_misc;
1469out:
1470 return rc;
1471out_misc:
1472 fib_proc_exit();
1473out_fib:
1474 udp4_proc_exit();
1475out_udp:
1476 tcp4_proc_exit();
1477out_tcp:
1478 raw_proc_exit();
1479out_raw:
1480 rc = -ENOMEM;
1481 goto out;
1482}
1483
1484#else /* CONFIG_PROC_FS */
1485static int __init ipv4_proc_init(void)
1486{
1487 return 0;
1488}
1489#endif /* CONFIG_PROC_FS */
1490
1491MODULE_ALIAS_NETPROTO(PF_INET);
1492
1493EXPORT_SYMBOL(inet_accept);
1494EXPORT_SYMBOL(inet_bind);
1495EXPORT_SYMBOL(inet_dgram_connect);
1496EXPORT_SYMBOL(inet_dgram_ops);
1497EXPORT_SYMBOL(inet_getname);
1498EXPORT_SYMBOL(inet_ioctl);
1499EXPORT_SYMBOL(inet_listen);
1500EXPORT_SYMBOL(inet_register_protosw);
1501EXPORT_SYMBOL(inet_release);
1502EXPORT_SYMBOL(inet_sendmsg);
1503EXPORT_SYMBOL(inet_shutdown);
1504EXPORT_SYMBOL(inet_sock_destruct);
1505EXPORT_SYMBOL(inet_stream_connect);
1506EXPORT_SYMBOL(inet_stream_ops);
1507EXPORT_SYMBOL(inet_unregister_protosw);
1508EXPORT_SYMBOL(net_statistics);
Neil Hormancdac4e02005-06-13 15:12:33 -07001509EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);