blob: 7ab0bd64c9d17a73df6605d6289ba5e5dfe1321f [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
Pavel Emelyanov9ba63972007-12-05 01:38:23 -0800129struct ipv4_config ipv4_config;
130
131EXPORT_SYMBOL(ipv4_config);
132
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133/* New destruction routine */
134
135void inet_sock_destruct(struct sock *sk)
136{
137 struct inet_sock *inet = inet_sk(sk);
138
139 __skb_queue_purge(&sk->sk_receive_queue);
140 __skb_queue_purge(&sk->sk_error_queue);
141
Hideo Aoki95766ff2007-12-31 00:29:24 -0800142 sk_mem_reclaim(sk);
143
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
145 printk("Attempt to release TCP socket in state %d %p\n",
146 sk->sk_state, sk);
147 return;
148 }
149 if (!sock_flag(sk, SOCK_DEAD)) {
150 printk("Attempt to release alive inet socket %p\n", sk);
151 return;
152 }
153
154 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
155 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
156 BUG_TRAP(!sk->sk_wmem_queued);
157 BUG_TRAP(!sk->sk_forward_alloc);
158
Jesper Juhla51482b2005-11-08 09:41:34 -0800159 kfree(inet->opt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160 dst_release(sk->sk_dst_cache);
Arnaldo Carvalho de Meloe6848972005-08-09 19:45:38 -0700161 sk_refcnt_debug_dec(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162}
163
164/*
165 * The routines beyond this point handle the behaviour of an AF_INET
166 * socket object. Mostly it punts to the subprotocols of IP to do
167 * the work.
168 */
169
170/*
171 * Automatically bind an unbound socket.
172 */
173
174static int inet_autobind(struct sock *sk)
175{
176 struct inet_sock *inet;
177 /* We may need to bind the socket. */
178 lock_sock(sk);
179 inet = inet_sk(sk);
180 if (!inet->num) {
181 if (sk->sk_prot->get_port(sk, 0)) {
182 release_sock(sk);
183 return -EAGAIN;
184 }
185 inet->sport = htons(inet->num);
186 }
187 release_sock(sk);
188 return 0;
189}
190
191/*
192 * Move a socket into listening state.
193 */
194int inet_listen(struct socket *sock, int backlog)
195{
196 struct sock *sk = sock->sk;
197 unsigned char old_state;
198 int err;
199
200 lock_sock(sk);
201
202 err = -EINVAL;
203 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
204 goto out;
205
206 old_state = sk->sk_state;
207 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
208 goto out;
209
210 /* Really, if the socket is already in listen state
211 * we can only allow the backlog to be adjusted.
212 */
213 if (old_state != TCP_LISTEN) {
Eric Dumazet72a3eff2006-11-16 02:30:37 -0800214 err = inet_csk_listen_start(sk, backlog);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 if (err)
216 goto out;
217 }
218 sk->sk_max_ack_backlog = backlog;
219 err = 0;
220
221out:
222 release_sock(sk);
223 return err;
224}
225
Eric Dumazetbe776282007-03-27 13:53:04 -0700226u32 inet_ehash_secret __read_mostly;
David S. Millerb3da2cf2007-03-23 11:40:27 -0700227EXPORT_SYMBOL(inet_ehash_secret);
228
Eric Dumazetbe776282007-03-27 13:53:04 -0700229/*
230 * inet_ehash_secret must be set exactly once
231 * Instead of using a dedicated spinlock, we (ab)use inetsw_lock
232 */
David S. Millerb3da2cf2007-03-23 11:40:27 -0700233void build_ehash_secret(void)
234{
Eric Dumazetbe776282007-03-27 13:53:04 -0700235 u32 rnd;
236 do {
237 get_random_bytes(&rnd, sizeof(rnd));
238 } while (rnd == 0);
239 spin_lock_bh(&inetsw_lock);
240 if (!inet_ehash_secret)
241 inet_ehash_secret = rnd;
242 spin_unlock_bh(&inetsw_lock);
David S. Millerb3da2cf2007-03-23 11:40:27 -0700243}
244EXPORT_SYMBOL(build_ehash_secret);
245
Denis V. Lunev2342fd72008-03-24 15:33:42 -0700246static inline int inet_netns_ok(struct net *net, int protocol)
247{
248 int hash;
249 struct net_protocol *ipprot;
250
251 if (net == &init_net)
252 return 1;
253
254 hash = protocol & (MAX_INET_PROTOS - 1);
255 ipprot = rcu_dereference(inet_protos[hash]);
256
257 if (ipprot == NULL)
258 /* raw IP is OK */
259 return 1;
260 return ipprot->netns_ok;
261}
262
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263/*
264 * Create an inet socket.
265 */
266
Eric W. Biederman1b8d7ae2007-10-08 23:24:22 -0700267static int inet_create(struct net *net, struct socket *sock, int protocol)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268{
269 struct sock *sk;
270 struct list_head *p;
271 struct inet_protosw *answer;
272 struct inet_sock *inet;
273 struct proto *answer_prot;
274 unsigned char answer_flags;
275 char answer_no_check;
Arnaldo Carvalho de Melobb97d312005-08-09 20:19:14 -0700276 int try_loading_module = 0;
Herbert Xu86c8f9d2005-12-02 20:43:26 -0800277 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278
David S. Millerb3da2cf2007-03-23 11:40:27 -0700279 if (sock->type != SOCK_RAW &&
280 sock->type != SOCK_DGRAM &&
281 !inet_ehash_secret)
282 build_ehash_secret();
283
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 sock->state = SS_UNCONNECTED;
285
286 /* Look for the requested type/protocol pair. */
287 answer = NULL;
Arnaldo Carvalho de Melobb97d312005-08-09 20:19:14 -0700288lookup_protocol:
Herbert Xu86c8f9d2005-12-02 20:43:26 -0800289 err = -ESOCKTNOSUPPORT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290 rcu_read_lock();
291 list_for_each_rcu(p, &inetsw[sock->type]) {
292 answer = list_entry(p, struct inet_protosw, list);
293
294 /* Check the non-wild match. */
295 if (protocol == answer->protocol) {
296 if (protocol != IPPROTO_IP)
297 break;
298 } else {
299 /* Check for the two wild cases. */
300 if (IPPROTO_IP == protocol) {
301 protocol = answer->protocol;
302 break;
303 }
304 if (IPPROTO_IP == answer->protocol)
305 break;
306 }
Herbert Xu86c8f9d2005-12-02 20:43:26 -0800307 err = -EPROTONOSUPPORT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308 answer = NULL;
309 }
310
Arnaldo Carvalho de Melobb97d312005-08-09 20:19:14 -0700311 if (unlikely(answer == NULL)) {
312 if (try_loading_module < 2) {
313 rcu_read_unlock();
314 /*
315 * Be more specific, e.g. net-pf-2-proto-132-type-1
316 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
317 */
318 if (++try_loading_module == 1)
319 request_module("net-pf-%d-proto-%d-type-%d",
320 PF_INET, protocol, sock->type);
321 /*
322 * Fall back to generic, e.g. net-pf-2-proto-132
323 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
324 */
325 else
326 request_module("net-pf-%d-proto-%d",
327 PF_INET, protocol);
328 goto lookup_protocol;
329 } else
330 goto out_rcu_unlock;
331 }
332
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 err = -EPERM;
334 if (answer->capability > 0 && !capable(answer->capability))
335 goto out_rcu_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336
Denis V. Lunev2342fd72008-03-24 15:33:42 -0700337 err = -EAFNOSUPPORT;
338 if (!inet_netns_ok(net, protocol))
339 goto out_rcu_unlock;
340
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 sock->ops = answer->ops;
342 answer_prot = answer->prot;
343 answer_no_check = answer->no_check;
344 answer_flags = answer->flags;
345 rcu_read_unlock();
346
347 BUG_TRAP(answer_prot->slab != NULL);
348
349 err = -ENOBUFS;
Pavel Emelyanov6257ff22007-11-01 00:39:31 -0700350 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 if (sk == NULL)
352 goto out;
353
354 err = 0;
355 sk->sk_no_check = answer_no_check;
356 if (INET_PROTOSW_REUSE & answer_flags)
357 sk->sk_reuse = 1;
358
359 inet = inet_sk(sk);
Paul Moore469de9b2007-01-09 14:37:06 -0800360 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361
362 if (SOCK_RAW == sock->type) {
363 inet->num = protocol;
364 if (IPPROTO_RAW == protocol)
365 inet->hdrincl = 1;
366 }
367
368 if (ipv4_config.no_pmtu_disc)
369 inet->pmtudisc = IP_PMTUDISC_DONT;
370 else
371 inet->pmtudisc = IP_PMTUDISC_WANT;
372
373 inet->id = 0;
374
375 sock_init_data(sock, sk);
376
377 sk->sk_destruct = inet_sock_destruct;
378 sk->sk_family = PF_INET;
379 sk->sk_protocol = protocol;
380 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
381
382 inet->uc_ttl = -1;
383 inet->mc_loop = 1;
384 inet->mc_ttl = 1;
385 inet->mc_index = 0;
386 inet->mc_list = NULL;
387
Arnaldo Carvalho de Meloe6848972005-08-09 19:45:38 -0700388 sk_refcnt_debug_inc(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389
390 if (inet->num) {
391 /* It assumes that any protocol which allows
392 * the user to assign a number at socket
393 * creation time automatically
394 * shares.
395 */
396 inet->sport = htons(inet->num);
397 /* Add to protocol hash chains. */
398 sk->sk_prot->hash(sk);
399 }
400
401 if (sk->sk_prot->init) {
402 err = sk->sk_prot->init(sk);
403 if (err)
404 sk_common_release(sk);
405 }
406out:
407 return err;
408out_rcu_unlock:
409 rcu_read_unlock();
410 goto out;
411}
412
413
414/*
415 * The peer socket should always be NULL (or else). When we call this
416 * function we are destroying the object and from then on nobody
417 * should refer to it.
418 */
419int inet_release(struct socket *sock)
420{
421 struct sock *sk = sock->sk;
422
423 if (sk) {
424 long timeout;
425
426 /* Applications forget to leave groups before exiting */
427 ip_mc_drop_socket(sk);
428
429 /* If linger is set, we don't return until the close
430 * is complete. Otherwise we return immediately. The
431 * actually closing is done the same either way.
432 *
433 * If the close is due to the process exiting, we never
434 * linger..
435 */
436 timeout = 0;
437 if (sock_flag(sk, SOCK_LINGER) &&
438 !(current->flags & PF_EXITING))
439 timeout = sk->sk_lingertime;
440 sock->sk = NULL;
441 sk->sk_prot->close(sk, timeout);
442 }
443 return 0;
444}
445
446/* It is off by default, see below. */
Brian Haleyab32ea52006-09-22 14:15:41 -0700447int sysctl_ip_nonlocal_bind __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448
449int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
450{
451 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
452 struct sock *sk = sock->sk;
453 struct inet_sock *inet = inet_sk(sk);
454 unsigned short snum;
455 int chk_addr_ret;
456 int err;
457
458 /* If the socket has its own bind function then use it. (RAW) */
459 if (sk->sk_prot->bind) {
460 err = sk->sk_prot->bind(sk, uaddr, addr_len);
461 goto out;
462 }
463 err = -EINVAL;
464 if (addr_len < sizeof(struct sockaddr_in))
465 goto out;
466
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900467 chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468
469 /* Not specified by any standard per-se, however it breaks too
470 * many applications when removed. It is unfortunate since
471 * allowing applications to make a non-local bind solves
472 * several problems with systems using dynamic addressing.
473 * (ie. your servers still start up even if your ISDN link
474 * is temporarily down)
475 */
476 err = -EADDRNOTAVAIL;
477 if (!sysctl_ip_nonlocal_bind &&
478 !inet->freebind &&
Al Viroe6f1ceb2008-03-17 22:44:53 -0700479 addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 chk_addr_ret != RTN_LOCAL &&
481 chk_addr_ret != RTN_MULTICAST &&
482 chk_addr_ret != RTN_BROADCAST)
483 goto out;
484
485 snum = ntohs(addr->sin_port);
486 err = -EACCES;
487 if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
488 goto out;
489
490 /* We keep a pair of addresses. rcv_saddr is the one
491 * used by hash lookups, and saddr is used for transmit.
492 *
493 * In the BSD API these are the same except where it
494 * would be illegal to use them (multicast/broadcast) in
495 * which case the sending device address is used.
496 */
497 lock_sock(sk);
498
499 /* Check these errors (active socket, double bind). */
500 err = -EINVAL;
501 if (sk->sk_state != TCP_CLOSE || inet->num)
502 goto out_release_sock;
503
504 inet->rcv_saddr = inet->saddr = addr->sin_addr.s_addr;
505 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
506 inet->saddr = 0; /* Use device */
507
508 /* Make sure we are allowed to bind here. */
509 if (sk->sk_prot->get_port(sk, snum)) {
510 inet->saddr = inet->rcv_saddr = 0;
511 err = -EADDRINUSE;
512 goto out_release_sock;
513 }
514
515 if (inet->rcv_saddr)
516 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
517 if (snum)
518 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
519 inet->sport = htons(inet->num);
520 inet->daddr = 0;
521 inet->dport = 0;
522 sk_dst_reset(sk);
523 err = 0;
524out_release_sock:
525 release_sock(sk);
526out:
527 return err;
528}
529
530int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
531 int addr_len, int flags)
532{
533 struct sock *sk = sock->sk;
534
535 if (uaddr->sa_family == AF_UNSPEC)
536 return sk->sk_prot->disconnect(sk, flags);
537
538 if (!inet_sk(sk)->num && inet_autobind(sk))
539 return -EAGAIN;
540 return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
541}
542
543static long inet_wait_for_connect(struct sock *sk, long timeo)
544{
545 DEFINE_WAIT(wait);
546
547 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
548
549 /* Basic assumption: if someone sets sk->sk_err, he _must_
550 * change state of the socket from TCP_SYN_*.
551 * Connect() does not allow to get error notifications
552 * without closing the socket.
553 */
554 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
555 release_sock(sk);
556 timeo = schedule_timeout(timeo);
557 lock_sock(sk);
558 if (signal_pending(current) || !timeo)
559 break;
560 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
561 }
562 finish_wait(sk->sk_sleep, &wait);
563 return timeo;
564}
565
566/*
567 * Connect to a remote host. There is regrettably still a little
568 * TCP 'magic' in here.
569 */
570int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
571 int addr_len, int flags)
572{
573 struct sock *sk = sock->sk;
574 int err;
575 long timeo;
576
577 lock_sock(sk);
578
579 if (uaddr->sa_family == AF_UNSPEC) {
580 err = sk->sk_prot->disconnect(sk, flags);
581 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
582 goto out;
583 }
584
585 switch (sock->state) {
586 default:
587 err = -EINVAL;
588 goto out;
589 case SS_CONNECTED:
590 err = -EISCONN;
591 goto out;
592 case SS_CONNECTING:
593 err = -EALREADY;
594 /* Fall out of switch with err, set for this state */
595 break;
596 case SS_UNCONNECTED:
597 err = -EISCONN;
598 if (sk->sk_state != TCP_CLOSE)
599 goto out;
600
601 err = sk->sk_prot->connect(sk, uaddr, addr_len);
602 if (err < 0)
603 goto out;
604
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900605 sock->state = SS_CONNECTING;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606
607 /* Just entered SS_CONNECTING state; the only
608 * difference is that return value in non-blocking
609 * case is EINPROGRESS, rather than EALREADY.
610 */
611 err = -EINPROGRESS;
612 break;
613 }
614
615 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
616
617 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
618 /* Error code is set above */
619 if (!timeo || !inet_wait_for_connect(sk, timeo))
620 goto out;
621
622 err = sock_intr_errno(timeo);
623 if (signal_pending(current))
624 goto out;
625 }
626
627 /* Connection was closed by RST, timeout, ICMP error
628 * or another process disconnected us.
629 */
630 if (sk->sk_state == TCP_CLOSE)
631 goto sock_error;
632
633 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
634 * and error was received after socket entered established state.
635 * Hence, it is handled normally after connect() return successfully.
636 */
637
638 sock->state = SS_CONNECTED;
639 err = 0;
640out:
641 release_sock(sk);
642 return err;
643
644sock_error:
645 err = sock_error(sk) ? : -ECONNABORTED;
646 sock->state = SS_UNCONNECTED;
647 if (sk->sk_prot->disconnect(sk, flags))
648 sock->state = SS_DISCONNECTING;
649 goto out;
650}
651
652/*
653 * Accept a pending connection. The TCP layer now gives BSD semantics.
654 */
655
656int inet_accept(struct socket *sock, struct socket *newsock, int flags)
657{
658 struct sock *sk1 = sock->sk;
659 int err = -EINVAL;
660 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
661
662 if (!sk2)
663 goto do_err;
664
665 lock_sock(sk2);
666
667 BUG_TRAP((1 << sk2->sk_state) &
668 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE));
669
670 sock_graft(sk2, newsock);
671
672 newsock->state = SS_CONNECTED;
673 err = 0;
674 release_sock(sk2);
675do_err:
676 return err;
677}
678
679
680/*
681 * This does both peername and sockname.
682 */
683int inet_getname(struct socket *sock, struct sockaddr *uaddr,
684 int *uaddr_len, int peer)
685{
686 struct sock *sk = sock->sk;
687 struct inet_sock *inet = inet_sk(sk);
688 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
689
690 sin->sin_family = AF_INET;
691 if (peer) {
692 if (!inet->dport ||
693 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
694 peer == 1))
695 return -ENOTCONN;
696 sin->sin_port = inet->dport;
697 sin->sin_addr.s_addr = inet->daddr;
698 } else {
Al Viro714e85b2006-11-14 20:51:49 -0800699 __be32 addr = inet->rcv_saddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 if (!addr)
701 addr = inet->saddr;
702 sin->sin_port = inet->sport;
703 sin->sin_addr.s_addr = addr;
704 }
705 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
706 *uaddr_len = sizeof(*sin);
707 return 0;
708}
709
710int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
711 size_t size)
712{
713 struct sock *sk = sock->sk;
714
715 /* We may need to bind the socket. */
716 if (!inet_sk(sk)->num && inet_autobind(sk))
717 return -EAGAIN;
718
719 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
720}
721
722
723static ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
724{
725 struct sock *sk = sock->sk;
726
727 /* We may need to bind the socket. */
728 if (!inet_sk(sk)->num && inet_autobind(sk))
729 return -EAGAIN;
730
731 if (sk->sk_prot->sendpage)
732 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
733 return sock_no_sendpage(sock, page, offset, size, flags);
734}
735
736
737int inet_shutdown(struct socket *sock, int how)
738{
739 struct sock *sk = sock->sk;
740 int err = 0;
741
742 /* This should really check to make sure
743 * the socket is a TCP socket. (WHY AC...)
744 */
745 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
746 1->2 bit 2 snds.
747 2->3 */
748 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
749 return -EINVAL;
750
751 lock_sock(sk);
752 if (sock->state == SS_CONNECTING) {
753 if ((1 << sk->sk_state) &
754 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
755 sock->state = SS_DISCONNECTING;
756 else
757 sock->state = SS_CONNECTED;
758 }
759
760 switch (sk->sk_state) {
761 case TCP_CLOSE:
762 err = -ENOTCONN;
763 /* Hack to wake up other listeners, who can poll for
764 POLLHUP, even on eg. unconnected UDP sockets -- RR */
765 default:
766 sk->sk_shutdown |= how;
767 if (sk->sk_prot->shutdown)
768 sk->sk_prot->shutdown(sk, how);
769 break;
770
771 /* Remaining two branches are temporary solution for missing
772 * close() in multithreaded environment. It is _not_ a good idea,
773 * but we have no choice until close() is repaired at VFS level.
774 */
775 case TCP_LISTEN:
776 if (!(how & RCV_SHUTDOWN))
777 break;
778 /* Fall through */
779 case TCP_SYN_SENT:
780 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
781 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
782 break;
783 }
784
785 /* Wake up anyone sleeping in poll. */
786 sk->sk_state_change(sk);
787 release_sock(sk);
788 return err;
789}
790
791/*
792 * ioctl() calls you can issue on an INET socket. Most of these are
793 * device configuration and stuff and very rarely used. Some ioctls
794 * pass on to the socket itself.
795 *
796 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
797 * loads the devconfigure module does its configuring and unloads it.
798 * There's a good 20K of config code hanging around the kernel.
799 */
800
801int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
802{
803 struct sock *sk = sock->sk;
804 int err = 0;
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900805 struct net *net = sock_net(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806
807 switch (cmd) {
808 case SIOCGSTAMP:
809 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
810 break;
Eric Dumazetae40eb12007-03-18 17:33:16 -0700811 case SIOCGSTAMPNS:
812 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
813 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 case SIOCADDRT:
815 case SIOCDELRT:
816 case SIOCRTMSG:
Denis V. Luneve5b13cb2008-02-28 20:51:43 -0800817 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 break;
819 case SIOCDARP:
820 case SIOCGARP:
821 case SIOCSARP:
Denis V. Luneve5b13cb2008-02-28 20:51:43 -0800822 err = arp_ioctl(net, cmd, (void __user *)arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 break;
824 case SIOCGIFADDR:
825 case SIOCSIFADDR:
826 case SIOCGIFBRDADDR:
827 case SIOCSIFBRDADDR:
828 case SIOCGIFNETMASK:
829 case SIOCSIFNETMASK:
830 case SIOCGIFDSTADDR:
831 case SIOCSIFDSTADDR:
832 case SIOCSIFPFLAGS:
833 case SIOCGIFPFLAGS:
834 case SIOCSIFFLAGS:
Denis V. Luneve5b13cb2008-02-28 20:51:43 -0800835 err = devinet_ioctl(net, cmd, (void __user *)arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 break;
837 default:
Christoph Hellwigb5e5fa52006-01-03 14:18:33 -0800838 if (sk->sk_prot->ioctl)
839 err = sk->sk_prot->ioctl(sk, cmd, arg);
840 else
841 err = -ENOIOCTLCMD;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 break;
843 }
844 return err;
845}
846
Eric Dumazet90ddc4f2005-12-22 12:49:22 -0800847const struct proto_ops inet_stream_ops = {
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -0800848 .family = PF_INET,
849 .owner = THIS_MODULE,
850 .release = inet_release,
851 .bind = inet_bind,
852 .connect = inet_stream_connect,
853 .socketpair = sock_no_socketpair,
854 .accept = inet_accept,
855 .getname = inet_getname,
856 .poll = tcp_poll,
857 .ioctl = inet_ioctl,
858 .listen = inet_listen,
859 .shutdown = inet_shutdown,
860 .setsockopt = sock_common_setsockopt,
861 .getsockopt = sock_common_getsockopt,
David S. Miller3516ffb2007-08-02 19:23:56 -0700862 .sendmsg = tcp_sendmsg,
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -0800863 .recvmsg = sock_common_recvmsg,
864 .mmap = sock_no_mmap,
865 .sendpage = tcp_sendpage,
Jens Axboe9c55e012007-11-06 23:30:13 -0800866 .splice_read = tcp_splice_read,
Dmitry Mishin3fdadf72006-03-20 22:45:21 -0800867#ifdef CONFIG_COMPAT
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -0800868 .compat_setsockopt = compat_sock_common_setsockopt,
869 .compat_getsockopt = compat_sock_common_getsockopt,
Dmitry Mishin3fdadf72006-03-20 22:45:21 -0800870#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871};
872
Eric Dumazet90ddc4f2005-12-22 12:49:22 -0800873const struct proto_ops inet_dgram_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 = udp_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
898/*
899 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
900 * udp_poll
901 */
Eric Dumazet90ddc4f2005-12-22 12:49:22 -0800902static const struct proto_ops inet_sockraw_ops = {
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -0800903 .family = PF_INET,
904 .owner = THIS_MODULE,
905 .release = inet_release,
906 .bind = inet_bind,
907 .connect = inet_dgram_connect,
908 .socketpair = sock_no_socketpair,
909 .accept = sock_no_accept,
910 .getname = inet_getname,
911 .poll = datagram_poll,
912 .ioctl = inet_ioctl,
913 .listen = sock_no_listen,
914 .shutdown = inet_shutdown,
915 .setsockopt = sock_common_setsockopt,
916 .getsockopt = sock_common_getsockopt,
917 .sendmsg = inet_sendmsg,
918 .recvmsg = sock_common_recvmsg,
919 .mmap = sock_no_mmap,
920 .sendpage = inet_sendpage,
Dmitry Mishin3fdadf72006-03-20 22:45:21 -0800921#ifdef CONFIG_COMPAT
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -0800922 .compat_setsockopt = compat_sock_common_setsockopt,
923 .compat_getsockopt = compat_sock_common_getsockopt,
Dmitry Mishin3fdadf72006-03-20 22:45:21 -0800924#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925};
926
927static struct net_proto_family inet_family_ops = {
928 .family = PF_INET,
929 .create = inet_create,
930 .owner = THIS_MODULE,
931};
932
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933/* Upon startup we insert all the elements in inetsw_array[] into
934 * the linked list inetsw.
935 */
936static struct inet_protosw inetsw_array[] =
937{
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900938 {
939 .type = SOCK_STREAM,
940 .protocol = IPPROTO_TCP,
941 .prot = &tcp_prot,
942 .ops = &inet_stream_ops,
943 .capability = -1,
944 .no_check = 0,
945 .flags = INET_PROTOSW_PERMANENT |
Arnaldo Carvalho de Melod83d8462005-12-13 23:26:10 -0800946 INET_PROTOSW_ICSK,
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900947 },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900949 {
950 .type = SOCK_DGRAM,
951 .protocol = IPPROTO_UDP,
952 .prot = &udp_prot,
953 .ops = &inet_dgram_ops,
954 .capability = -1,
955 .no_check = UDP_CSUM_DEFAULT,
956 .flags = INET_PROTOSW_PERMANENT,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 },
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900958
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959
960 {
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900961 .type = SOCK_RAW,
962 .protocol = IPPROTO_IP, /* wild card */
963 .prot = &raw_prot,
964 .ops = &inet_sockraw_ops,
965 .capability = CAP_NET_RAW,
966 .no_check = UDP_CSUM_DEFAULT,
967 .flags = INET_PROTOSW_REUSE,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 }
969};
970
Denis Chengc40f6ff2007-09-16 16:39:25 -0700971#define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972
973void inet_register_protosw(struct inet_protosw *p)
974{
975 struct list_head *lh;
976 struct inet_protosw *answer;
977 int protocol = p->protocol;
978 struct list_head *last_perm;
979
980 spin_lock_bh(&inetsw_lock);
981
982 if (p->type >= SOCK_MAX)
983 goto out_illegal;
984
985 /* If we are trying to override a permanent protocol, bail. */
986 answer = NULL;
987 last_perm = &inetsw[p->type];
988 list_for_each(lh, &inetsw[p->type]) {
989 answer = list_entry(lh, struct inet_protosw, list);
990
991 /* Check only the non-wild match. */
992 if (INET_PROTOSW_PERMANENT & answer->flags) {
993 if (protocol == answer->protocol)
994 break;
995 last_perm = lh;
996 }
997
998 answer = NULL;
999 }
1000 if (answer)
1001 goto out_permanent;
1002
1003 /* Add the new entry after the last permanent entry if any, so that
1004 * the new entry does not override a permanent entry when matched with
1005 * a wild-card protocol. But it is allowed to override any existing
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001006 * non-permanent entry. This means that when we remove this entry, the
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 * system automatically returns to the old behavior.
1008 */
1009 list_add_rcu(&p->list, last_perm);
1010out:
1011 spin_unlock_bh(&inetsw_lock);
1012
1013 synchronize_net();
1014
1015 return;
1016
1017out_permanent:
1018 printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
1019 protocol);
1020 goto out;
1021
1022out_illegal:
1023 printk(KERN_ERR
1024 "Ignoring attempt to register invalid socket type %d.\n",
1025 p->type);
1026 goto out;
1027}
1028
1029void inet_unregister_protosw(struct inet_protosw *p)
1030{
1031 if (INET_PROTOSW_PERMANENT & p->flags) {
1032 printk(KERN_ERR
1033 "Attempt to unregister permanent protocol %d.\n",
1034 p->protocol);
1035 } else {
1036 spin_lock_bh(&inetsw_lock);
1037 list_del_rcu(&p->list);
1038 spin_unlock_bh(&inetsw_lock);
1039
1040 synchronize_net();
1041 }
1042}
1043
Arnaldo Carvalho de Melo32519f12005-08-09 19:50:02 -07001044/*
1045 * Shall we try to damage output packets if routing dev changes?
1046 */
1047
Brian Haleyab32ea52006-09-22 14:15:41 -07001048int sysctl_ip_dynaddr __read_mostly;
Arnaldo Carvalho de Melo32519f12005-08-09 19:50:02 -07001049
1050static int inet_sk_reselect_saddr(struct sock *sk)
1051{
1052 struct inet_sock *inet = inet_sk(sk);
1053 int err;
1054 struct rtable *rt;
Al Viro714e85b2006-11-14 20:51:49 -08001055 __be32 old_saddr = inet->saddr;
1056 __be32 new_saddr;
Al Virobada8ad2006-09-26 21:27:15 -07001057 __be32 daddr = inet->daddr;
Arnaldo Carvalho de Melo32519f12005-08-09 19:50:02 -07001058
1059 if (inet->opt && inet->opt->srr)
1060 daddr = inet->opt->faddr;
1061
1062 /* Query new route. */
1063 err = ip_route_connect(&rt, daddr, 0,
1064 RT_CONN_FLAGS(sk),
1065 sk->sk_bound_dev_if,
1066 sk->sk_protocol,
David S. Miller8eb90862007-02-08 02:09:21 -08001067 inet->sport, inet->dport, sk, 0);
Arnaldo Carvalho de Melo32519f12005-08-09 19:50:02 -07001068 if (err)
1069 return err;
1070
1071 sk_setup_caps(sk, &rt->u.dst);
1072
1073 new_saddr = rt->rt_src;
1074
1075 if (new_saddr == old_saddr)
1076 return 0;
1077
1078 if (sysctl_ip_dynaddr > 1) {
1079 printk(KERN_INFO "%s(): shifting inet->"
1080 "saddr from %d.%d.%d.%d to %d.%d.%d.%d\n",
Harvey Harrison0dc47872008-03-05 20:47:47 -08001081 __func__,
Arnaldo Carvalho de Melo32519f12005-08-09 19:50:02 -07001082 NIPQUAD(old_saddr),
1083 NIPQUAD(new_saddr));
1084 }
1085
1086 inet->saddr = inet->rcv_saddr = new_saddr;
1087
1088 /*
1089 * XXX The only one ugly spot where we need to
1090 * XXX really change the sockets identity after
1091 * XXX it has entered the hashes. -DaveM
1092 *
1093 * Besides that, it does not check for connection
1094 * uniqueness. Wait for troubles.
1095 */
1096 __sk_prot_rehash(sk);
1097 return 0;
1098}
1099
1100int inet_sk_rebuild_header(struct sock *sk)
1101{
1102 struct inet_sock *inet = inet_sk(sk);
1103 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
Al Viro3ca3c682006-09-27 18:28:07 -07001104 __be32 daddr;
Arnaldo Carvalho de Melo32519f12005-08-09 19:50:02 -07001105 int err;
1106
1107 /* Route is OK, nothing to do. */
1108 if (rt)
1109 return 0;
1110
1111 /* Reroute. */
1112 daddr = inet->daddr;
1113 if (inet->opt && inet->opt->srr)
1114 daddr = inet->opt->faddr;
1115{
1116 struct flowi fl = {
1117 .oif = sk->sk_bound_dev_if,
1118 .nl_u = {
1119 .ip4_u = {
1120 .daddr = daddr,
1121 .saddr = inet->saddr,
1122 .tos = RT_CONN_FLAGS(sk),
1123 },
1124 },
1125 .proto = sk->sk_protocol,
1126 .uli_u = {
1127 .ports = {
1128 .sport = inet->sport,
1129 .dport = inet->dport,
1130 },
1131 },
1132 };
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001133
Venkat Yekkiralabeb8d132006-08-04 23:12:42 -07001134 security_sk_classify_flow(sk, &fl);
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +09001135 err = ip_route_output_flow(sock_net(sk), &rt, &fl, sk, 0);
Arnaldo Carvalho de Melo32519f12005-08-09 19:50:02 -07001136}
1137 if (!err)
1138 sk_setup_caps(sk, &rt->u.dst);
1139 else {
1140 /* Routing failed... */
1141 sk->sk_route_caps = 0;
1142 /*
1143 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1144 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1145 */
1146 if (!sysctl_ip_dynaddr ||
1147 sk->sk_state != TCP_SYN_SENT ||
1148 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1149 (err = inet_sk_reselect_saddr(sk)) != 0)
1150 sk->sk_err_soft = -err;
1151 }
1152
1153 return err;
1154}
1155
1156EXPORT_SYMBOL(inet_sk_rebuild_header);
1157
Herbert Xua430a432006-07-08 13:34:56 -07001158static int inet_gso_send_check(struct sk_buff *skb)
1159{
1160 struct iphdr *iph;
1161 struct net_protocol *ops;
1162 int proto;
1163 int ihl;
1164 int err = -EINVAL;
1165
1166 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1167 goto out;
1168
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001169 iph = ip_hdr(skb);
Herbert Xua430a432006-07-08 13:34:56 -07001170 ihl = iph->ihl * 4;
1171 if (ihl < sizeof(*iph))
1172 goto out;
1173
1174 if (unlikely(!pskb_may_pull(skb, ihl)))
1175 goto out;
1176
Arnaldo Carvalho de Melobadff6d2007-03-13 13:06:52 -03001177 __skb_pull(skb, ihl);
1178 skb_reset_transport_header(skb);
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001179 iph = ip_hdr(skb);
Herbert Xua430a432006-07-08 13:34:56 -07001180 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1181 err = -EPROTONOSUPPORT;
1182
1183 rcu_read_lock();
1184 ops = rcu_dereference(inet_protos[proto]);
1185 if (likely(ops && ops->gso_send_check))
1186 err = ops->gso_send_check(skb);
1187 rcu_read_unlock();
1188
1189out:
1190 return err;
1191}
1192
Herbert Xu576a30e2006-06-27 13:22:38 -07001193static struct sk_buff *inet_gso_segment(struct sk_buff *skb, int features)
Herbert Xuf4c50d92006-06-22 03:02:40 -07001194{
1195 struct sk_buff *segs = ERR_PTR(-EINVAL);
1196 struct iphdr *iph;
1197 struct net_protocol *ops;
1198 int proto;
1199 int ihl;
1200 int id;
1201
Stephen Hemmingerd212f872007-06-27 00:47:37 -07001202 if (!(features & NETIF_F_V4_CSUM))
1203 features &= ~NETIF_F_SG;
1204
Herbert Xubbcf4672006-07-03 19:38:35 -07001205 if (unlikely(skb_shinfo(skb)->gso_type &
1206 ~(SKB_GSO_TCPV4 |
1207 SKB_GSO_UDP |
1208 SKB_GSO_DODGY |
1209 SKB_GSO_TCP_ECN |
1210 0)))
1211 goto out;
1212
1213 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
Herbert Xuf4c50d92006-06-22 03:02:40 -07001214 goto out;
1215
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001216 iph = ip_hdr(skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07001217 ihl = iph->ihl * 4;
1218 if (ihl < sizeof(*iph))
1219 goto out;
1220
Herbert Xubbcf4672006-07-03 19:38:35 -07001221 if (unlikely(!pskb_may_pull(skb, ihl)))
Herbert Xuf4c50d92006-06-22 03:02:40 -07001222 goto out;
1223
Arnaldo Carvalho de Melobadff6d2007-03-13 13:06:52 -03001224 __skb_pull(skb, ihl);
1225 skb_reset_transport_header(skb);
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001226 iph = ip_hdr(skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07001227 id = ntohs(iph->id);
1228 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1229 segs = ERR_PTR(-EPROTONOSUPPORT);
1230
1231 rcu_read_lock();
1232 ops = rcu_dereference(inet_protos[proto]);
Herbert Xubbcf4672006-07-03 19:38:35 -07001233 if (likely(ops && ops->gso_segment))
Herbert Xu576a30e2006-06-27 13:22:38 -07001234 segs = ops->gso_segment(skb, features);
Herbert Xuf4c50d92006-06-22 03:02:40 -07001235 rcu_read_unlock();
1236
Herbert Xu576a30e2006-06-27 13:22:38 -07001237 if (!segs || unlikely(IS_ERR(segs)))
Herbert Xuf4c50d92006-06-22 03:02:40 -07001238 goto out;
1239
1240 skb = segs;
1241 do {
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001242 iph = ip_hdr(skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07001243 iph->id = htons(id++);
1244 iph->tot_len = htons(skb->len - skb->mac_len);
1245 iph->check = 0;
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -07001246 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
Herbert Xuf4c50d92006-06-22 03:02:40 -07001247 } while ((skb = skb->next));
1248
1249out:
1250 return segs;
1251}
1252
Denis V. Lunev3d58b5f2008-04-03 14:22:32 -07001253int inet_ctl_sock_create(struct socket **sock, unsigned short family,
1254 unsigned short type, unsigned char protocol)
1255{
1256 int rc = sock_create_kern(family, type, protocol, sock);
1257
1258 if (rc == 0) {
1259 (*sock)->sk->sk_allocation = GFP_ATOMIC;
1260 inet_sk((*sock)->sk)->uc_ttl = -1;
1261 /*
1262 * Unhash it so that IP input processing does not even see it,
1263 * we do not wish this socket to see incoming packets.
1264 */
1265 (*sock)->sk->sk_prot->unhash((*sock)->sk);
1266 }
1267 return rc;
1268}
1269
1270EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1271
Herbert Xu5e0f0432007-04-24 21:53:35 -07001272unsigned long snmp_fold_field(void *mib[], int offt)
1273{
1274 unsigned long res = 0;
1275 int i;
1276
1277 for_each_possible_cpu(i) {
1278 res += *(((unsigned long *) per_cpu_ptr(mib[0], i)) + offt);
1279 res += *(((unsigned long *) per_cpu_ptr(mib[1], i)) + offt);
1280 }
1281 return res;
1282}
1283EXPORT_SYMBOL_GPL(snmp_fold_field);
1284
YOSHIFUJI Hideakic69bce22008-01-23 22:31:45 -08001285int snmp_mib_init(void *ptr[2], size_t mibsize)
Herbert Xu5e0f0432007-04-24 21:53:35 -07001286{
1287 BUG_ON(ptr == NULL);
1288 ptr[0] = __alloc_percpu(mibsize);
1289 if (!ptr[0])
1290 goto err0;
1291 ptr[1] = __alloc_percpu(mibsize);
1292 if (!ptr[1])
1293 goto err1;
1294 return 0;
1295err1:
1296 free_percpu(ptr[0]);
1297 ptr[0] = NULL;
1298err0:
1299 return -ENOMEM;
1300}
1301EXPORT_SYMBOL_GPL(snmp_mib_init);
1302
1303void snmp_mib_free(void *ptr[2])
1304{
1305 BUG_ON(ptr == NULL);
1306 free_percpu(ptr[0]);
1307 free_percpu(ptr[1]);
1308 ptr[0] = ptr[1] = NULL;
1309}
1310EXPORT_SYMBOL_GPL(snmp_mib_free);
1311
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312#ifdef CONFIG_IP_MULTICAST
1313static struct net_protocol igmp_protocol = {
1314 .handler = igmp_rcv,
1315};
1316#endif
1317
1318static struct net_protocol tcp_protocol = {
1319 .handler = tcp_v4_rcv,
1320 .err_handler = tcp_v4_err,
Herbert Xua430a432006-07-08 13:34:56 -07001321 .gso_send_check = tcp_v4_gso_send_check,
Herbert Xuf4c50d92006-06-22 03:02:40 -07001322 .gso_segment = tcp_tso_segment,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 .no_policy = 1,
Denis V. Lunev92f1fec2008-03-24 15:34:06 -07001324 .netns_ok = 1,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325};
1326
1327static struct net_protocol udp_protocol = {
1328 .handler = udp_rcv,
1329 .err_handler = udp_err,
1330 .no_policy = 1,
Denis V. Lunev92f1fec2008-03-24 15:34:06 -07001331 .netns_ok = 1,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332};
1333
1334static struct net_protocol icmp_protocol = {
1335 .handler = icmp_rcv,
Herbert Xu8b7817f2007-12-12 10:44:43 -08001336 .no_policy = 1,
Denis V. Lunev92f1fec2008-03-24 15:34:06 -07001337 .netns_ok = 1,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338};
1339
1340static int __init init_ipv4_mibs(void)
1341{
YOSHIFUJI Hideaki33490172007-04-20 15:57:15 -07001342 if (snmp_mib_init((void **)net_statistics,
YOSHIFUJI Hideakic69bce22008-01-23 22:31:45 -08001343 sizeof(struct linux_mib)) < 0)
YOSHIFUJI Hideaki33490172007-04-20 15:57:15 -07001344 goto err_net_mib;
1345 if (snmp_mib_init((void **)ip_statistics,
YOSHIFUJI Hideakic69bce22008-01-23 22:31:45 -08001346 sizeof(struct ipstats_mib)) < 0)
YOSHIFUJI Hideaki33490172007-04-20 15:57:15 -07001347 goto err_ip_mib;
1348 if (snmp_mib_init((void **)icmp_statistics,
YOSHIFUJI Hideakic69bce22008-01-23 22:31:45 -08001349 sizeof(struct icmp_mib)) < 0)
YOSHIFUJI Hideaki33490172007-04-20 15:57:15 -07001350 goto err_icmp_mib;
David L Stevens96793b42007-09-17 09:57:33 -07001351 if (snmp_mib_init((void **)icmpmsg_statistics,
YOSHIFUJI Hideakic69bce22008-01-23 22:31:45 -08001352 sizeof(struct icmpmsg_mib)) < 0)
David L Stevens96793b42007-09-17 09:57:33 -07001353 goto err_icmpmsg_mib;
YOSHIFUJI Hideaki33490172007-04-20 15:57:15 -07001354 if (snmp_mib_init((void **)tcp_statistics,
YOSHIFUJI Hideakic69bce22008-01-23 22:31:45 -08001355 sizeof(struct tcp_mib)) < 0)
YOSHIFUJI Hideaki33490172007-04-20 15:57:15 -07001356 goto err_tcp_mib;
1357 if (snmp_mib_init((void **)udp_statistics,
YOSHIFUJI Hideakic69bce22008-01-23 22:31:45 -08001358 sizeof(struct udp_mib)) < 0)
YOSHIFUJI Hideaki33490172007-04-20 15:57:15 -07001359 goto err_udp_mib;
1360 if (snmp_mib_init((void **)udplite_statistics,
YOSHIFUJI Hideakic69bce22008-01-23 22:31:45 -08001361 sizeof(struct udp_mib)) < 0)
YOSHIFUJI Hideaki33490172007-04-20 15:57:15 -07001362 goto err_udplite_mib;
David S. Millerdb8dac22008-03-06 16:22:02 -08001363
YOSHIFUJI Hideaki33490172007-04-20 15:57:15 -07001364 tcp_mib_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365
1366 return 0;
YOSHIFUJI Hideaki33490172007-04-20 15:57:15 -07001367
1368err_udplite_mib:
1369 snmp_mib_free((void **)udp_statistics);
1370err_udp_mib:
1371 snmp_mib_free((void **)tcp_statistics);
1372err_tcp_mib:
David L Stevens96793b42007-09-17 09:57:33 -07001373 snmp_mib_free((void **)icmpmsg_statistics);
1374err_icmpmsg_mib:
YOSHIFUJI Hideaki33490172007-04-20 15:57:15 -07001375 snmp_mib_free((void **)icmp_statistics);
1376err_icmp_mib:
1377 snmp_mib_free((void **)ip_statistics);
1378err_ip_mib:
1379 snmp_mib_free((void **)net_statistics);
1380err_net_mib:
1381 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382}
1383
1384static int ipv4_proc_init(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385
Herbert Xu30e224d2005-07-05 14:40:10 -07001386/*
1387 * IP protocol layer initialiser
1388 */
1389
1390static struct packet_type ip_packet_type = {
1391 .type = __constant_htons(ETH_P_IP),
1392 .func = ip_rcv,
Herbert Xua430a432006-07-08 13:34:56 -07001393 .gso_send_check = inet_gso_send_check,
Herbert Xuf4c50d92006-06-22 03:02:40 -07001394 .gso_segment = inet_gso_segment,
Herbert Xu30e224d2005-07-05 14:40:10 -07001395};
1396
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397static int __init inet_init(void)
1398{
1399 struct sk_buff *dummy_skb;
1400 struct inet_protosw *q;
1401 struct list_head *r;
1402 int rc = -EINVAL;
1403
YOSHIFUJI Hideakief047f52006-09-01 00:29:06 -07001404 BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405
1406 rc = proto_register(&tcp_prot, 1);
1407 if (rc)
1408 goto out;
1409
1410 rc = proto_register(&udp_prot, 1);
1411 if (rc)
1412 goto out_unregister_tcp_proto;
1413
1414 rc = proto_register(&raw_prot, 1);
1415 if (rc)
1416 goto out_unregister_udp_proto;
1417
1418 /*
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001419 * Tell SOCKET that we are alive...
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420 */
1421
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001422 (void)sock_register(&inet_family_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423
1424 /*
1425 * Add all the base protocols.
1426 */
1427
1428 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1429 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1430 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1431 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1432 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1433 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1434#ifdef CONFIG_IP_MULTICAST
1435 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1436 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1437#endif
1438
1439 /* Register the socket-side information for inet_create. */
1440 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1441 INIT_LIST_HEAD(r);
1442
1443 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1444 inet_register_protosw(q);
1445
1446 /*
1447 * Set the ARP module up
1448 */
1449
1450 arp_init();
1451
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001452 /*
1453 * Set the IP module up
1454 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455
1456 ip_init();
1457
Denis V. Lunev9b0f9762008-02-29 11:13:15 -08001458 tcp_v4_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459
1460 /* Setup TCP slab cache for open requests. */
1461 tcp_init();
1462
Hideo Aoki95766ff2007-12-31 00:29:24 -08001463 /* Setup UDP memory threshold */
1464 udp_init();
1465
Gerrit Renkerba4e58e2006-11-27 11:10:57 -08001466 /* Add UDP-Lite (RFC 3828) */
1467 udplite4_register();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468
1469 /*
1470 * Set the ICMP layer up
1471 */
1472
Denis V. Luneva5710d62008-02-29 11:14:50 -08001473 if (icmp_init() < 0)
1474 panic("Failed to create the ICMP control socket.\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475
1476 /*
1477 * Initialise the multicast router
1478 */
1479#if defined(CONFIG_IP_MROUTE)
1480 ip_mr_init();
1481#endif
1482 /*
1483 * Initialise per-cpu ipv4 mibs
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001484 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485
Stephen Hemminger132adf52007-03-08 20:44:43 -08001486 if (init_ipv4_mibs())
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n"); ;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001488
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 ipv4_proc_init();
1490
1491 ipfrag_init();
1492
Herbert Xu30e224d2005-07-05 14:40:10 -07001493 dev_add_pack(&ip_packet_type);
1494
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 rc = 0;
1496out:
1497 return rc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498out_unregister_udp_proto:
1499 proto_unregister(&udp_prot);
Olaf Kirch321efff2006-09-27 16:33:45 -07001500out_unregister_tcp_proto:
1501 proto_unregister(&tcp_prot);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 goto out;
1503}
1504
Heiko Carstensa536e072006-04-28 15:19:17 -07001505fs_initcall(inet_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506
1507/* ------------------------------------------------------------------------ */
1508
1509#ifdef CONFIG_PROC_FS
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510static int __init ipv4_proc_init(void)
1511{
1512 int rc = 0;
1513
1514 if (raw_proc_init())
1515 goto out_raw;
1516 if (tcp4_proc_init())
1517 goto out_tcp;
1518 if (udp4_proc_init())
1519 goto out_udp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 if (ip_misc_proc_init())
1521 goto out_misc;
1522out:
1523 return rc;
1524out_misc:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 udp4_proc_exit();
1526out_udp:
1527 tcp4_proc_exit();
1528out_tcp:
1529 raw_proc_exit();
1530out_raw:
1531 rc = -ENOMEM;
1532 goto out;
1533}
1534
1535#else /* CONFIG_PROC_FS */
1536static int __init ipv4_proc_init(void)
1537{
1538 return 0;
1539}
1540#endif /* CONFIG_PROC_FS */
1541
1542MODULE_ALIAS_NETPROTO(PF_INET);
1543
1544EXPORT_SYMBOL(inet_accept);
1545EXPORT_SYMBOL(inet_bind);
1546EXPORT_SYMBOL(inet_dgram_connect);
1547EXPORT_SYMBOL(inet_dgram_ops);
1548EXPORT_SYMBOL(inet_getname);
1549EXPORT_SYMBOL(inet_ioctl);
1550EXPORT_SYMBOL(inet_listen);
1551EXPORT_SYMBOL(inet_register_protosw);
1552EXPORT_SYMBOL(inet_release);
1553EXPORT_SYMBOL(inet_sendmsg);
1554EXPORT_SYMBOL(inet_shutdown);
1555EXPORT_SYMBOL(inet_sock_destruct);
1556EXPORT_SYMBOL(inet_stream_connect);
1557EXPORT_SYMBOL(inet_stream_ops);
1558EXPORT_SYMBOL(inet_unregister_protosw);
1559EXPORT_SYMBOL(net_statistics);
Neil Hormancdac4e02005-06-13 15:12:33 -07001560EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);