blob: c75f20b49935ce0ecba149200d3ef7ef38cf47e4 [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
Eric W. Biederman1b8d7ae2007-10-08 23:24:22 -0700244static int inet_create(struct net *net, struct socket *sock, int protocol)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245{
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
Eric W. Biederman1b8d7ae2007-10-08 23:24:22 -0700256 if (net != &init_net)
257 return -EAFNOSUPPORT;
258
David S. Millerb3da2cf2007-03-23 11:40:27 -0700259 if (sock->type != SOCK_RAW &&
260 sock->type != SOCK_DGRAM &&
261 !inet_ehash_secret)
262 build_ehash_secret();
263
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 sock->state = SS_UNCONNECTED;
265
266 /* Look for the requested type/protocol pair. */
267 answer = NULL;
Arnaldo Carvalho de Melobb97d312005-08-09 20:19:14 -0700268lookup_protocol:
Herbert Xu86c8f9d2005-12-02 20:43:26 -0800269 err = -ESOCKTNOSUPPORT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 rcu_read_lock();
271 list_for_each_rcu(p, &inetsw[sock->type]) {
272 answer = list_entry(p, struct inet_protosw, list);
273
274 /* Check the non-wild match. */
275 if (protocol == answer->protocol) {
276 if (protocol != IPPROTO_IP)
277 break;
278 } else {
279 /* Check for the two wild cases. */
280 if (IPPROTO_IP == protocol) {
281 protocol = answer->protocol;
282 break;
283 }
284 if (IPPROTO_IP == answer->protocol)
285 break;
286 }
Herbert Xu86c8f9d2005-12-02 20:43:26 -0800287 err = -EPROTONOSUPPORT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 answer = NULL;
289 }
290
Arnaldo Carvalho de Melobb97d312005-08-09 20:19:14 -0700291 if (unlikely(answer == NULL)) {
292 if (try_loading_module < 2) {
293 rcu_read_unlock();
294 /*
295 * Be more specific, e.g. net-pf-2-proto-132-type-1
296 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
297 */
298 if (++try_loading_module == 1)
299 request_module("net-pf-%d-proto-%d-type-%d",
300 PF_INET, protocol, sock->type);
301 /*
302 * Fall back to generic, e.g. net-pf-2-proto-132
303 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
304 */
305 else
306 request_module("net-pf-%d-proto-%d",
307 PF_INET, protocol);
308 goto lookup_protocol;
309 } else
310 goto out_rcu_unlock;
311 }
312
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 err = -EPERM;
314 if (answer->capability > 0 && !capable(answer->capability))
315 goto out_rcu_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316
317 sock->ops = answer->ops;
318 answer_prot = answer->prot;
319 answer_no_check = answer->no_check;
320 answer_flags = answer->flags;
321 rcu_read_unlock();
322
323 BUG_TRAP(answer_prot->slab != NULL);
324
325 err = -ENOBUFS;
Pavel Emelyanov6257ff22007-11-01 00:39:31 -0700326 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 if (sk == NULL)
328 goto out;
329
330 err = 0;
331 sk->sk_no_check = answer_no_check;
332 if (INET_PROTOSW_REUSE & answer_flags)
333 sk->sk_reuse = 1;
334
335 inet = inet_sk(sk);
Paul Moore469de9b2007-01-09 14:37:06 -0800336 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337
338 if (SOCK_RAW == sock->type) {
339 inet->num = protocol;
340 if (IPPROTO_RAW == protocol)
341 inet->hdrincl = 1;
342 }
343
344 if (ipv4_config.no_pmtu_disc)
345 inet->pmtudisc = IP_PMTUDISC_DONT;
346 else
347 inet->pmtudisc = IP_PMTUDISC_WANT;
348
349 inet->id = 0;
350
351 sock_init_data(sock, sk);
352
353 sk->sk_destruct = inet_sock_destruct;
354 sk->sk_family = PF_INET;
355 sk->sk_protocol = protocol;
356 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
357
358 inet->uc_ttl = -1;
359 inet->mc_loop = 1;
360 inet->mc_ttl = 1;
361 inet->mc_index = 0;
362 inet->mc_list = NULL;
363
Arnaldo Carvalho de Meloe6848972005-08-09 19:45:38 -0700364 sk_refcnt_debug_inc(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365
366 if (inet->num) {
367 /* It assumes that any protocol which allows
368 * the user to assign a number at socket
369 * creation time automatically
370 * shares.
371 */
372 inet->sport = htons(inet->num);
373 /* Add to protocol hash chains. */
374 sk->sk_prot->hash(sk);
375 }
376
377 if (sk->sk_prot->init) {
378 err = sk->sk_prot->init(sk);
379 if (err)
380 sk_common_release(sk);
381 }
382out:
383 return err;
384out_rcu_unlock:
385 rcu_read_unlock();
386 goto out;
387}
388
389
390/*
391 * The peer socket should always be NULL (or else). When we call this
392 * function we are destroying the object and from then on nobody
393 * should refer to it.
394 */
395int inet_release(struct socket *sock)
396{
397 struct sock *sk = sock->sk;
398
399 if (sk) {
400 long timeout;
401
402 /* Applications forget to leave groups before exiting */
403 ip_mc_drop_socket(sk);
404
405 /* If linger is set, we don't return until the close
406 * is complete. Otherwise we return immediately. The
407 * actually closing is done the same either way.
408 *
409 * If the close is due to the process exiting, we never
410 * linger..
411 */
412 timeout = 0;
413 if (sock_flag(sk, SOCK_LINGER) &&
414 !(current->flags & PF_EXITING))
415 timeout = sk->sk_lingertime;
416 sock->sk = NULL;
417 sk->sk_prot->close(sk, timeout);
418 }
419 return 0;
420}
421
422/* It is off by default, see below. */
Brian Haleyab32ea52006-09-22 14:15:41 -0700423int sysctl_ip_nonlocal_bind __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424
425int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
426{
427 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
428 struct sock *sk = sock->sk;
429 struct inet_sock *inet = inet_sk(sk);
430 unsigned short snum;
431 int chk_addr_ret;
432 int err;
433
434 /* If the socket has its own bind function then use it. (RAW) */
435 if (sk->sk_prot->bind) {
436 err = sk->sk_prot->bind(sk, uaddr, addr_len);
437 goto out;
438 }
439 err = -EINVAL;
440 if (addr_len < sizeof(struct sockaddr_in))
441 goto out;
442
443 chk_addr_ret = inet_addr_type(addr->sin_addr.s_addr);
444
445 /* Not specified by any standard per-se, however it breaks too
446 * many applications when removed. It is unfortunate since
447 * allowing applications to make a non-local bind solves
448 * several problems with systems using dynamic addressing.
449 * (ie. your servers still start up even if your ISDN link
450 * is temporarily down)
451 */
452 err = -EADDRNOTAVAIL;
453 if (!sysctl_ip_nonlocal_bind &&
454 !inet->freebind &&
455 addr->sin_addr.s_addr != INADDR_ANY &&
456 chk_addr_ret != RTN_LOCAL &&
457 chk_addr_ret != RTN_MULTICAST &&
458 chk_addr_ret != RTN_BROADCAST)
459 goto out;
460
461 snum = ntohs(addr->sin_port);
462 err = -EACCES;
463 if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
464 goto out;
465
466 /* We keep a pair of addresses. rcv_saddr is the one
467 * used by hash lookups, and saddr is used for transmit.
468 *
469 * In the BSD API these are the same except where it
470 * would be illegal to use them (multicast/broadcast) in
471 * which case the sending device address is used.
472 */
473 lock_sock(sk);
474
475 /* Check these errors (active socket, double bind). */
476 err = -EINVAL;
477 if (sk->sk_state != TCP_CLOSE || inet->num)
478 goto out_release_sock;
479
480 inet->rcv_saddr = inet->saddr = addr->sin_addr.s_addr;
481 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
482 inet->saddr = 0; /* Use device */
483
484 /* Make sure we are allowed to bind here. */
485 if (sk->sk_prot->get_port(sk, snum)) {
486 inet->saddr = inet->rcv_saddr = 0;
487 err = -EADDRINUSE;
488 goto out_release_sock;
489 }
490
491 if (inet->rcv_saddr)
492 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
493 if (snum)
494 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
495 inet->sport = htons(inet->num);
496 inet->daddr = 0;
497 inet->dport = 0;
498 sk_dst_reset(sk);
499 err = 0;
500out_release_sock:
501 release_sock(sk);
502out:
503 return err;
504}
505
506int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
507 int addr_len, int flags)
508{
509 struct sock *sk = sock->sk;
510
511 if (uaddr->sa_family == AF_UNSPEC)
512 return sk->sk_prot->disconnect(sk, flags);
513
514 if (!inet_sk(sk)->num && inet_autobind(sk))
515 return -EAGAIN;
516 return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
517}
518
519static long inet_wait_for_connect(struct sock *sk, long timeo)
520{
521 DEFINE_WAIT(wait);
522
523 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
524
525 /* Basic assumption: if someone sets sk->sk_err, he _must_
526 * change state of the socket from TCP_SYN_*.
527 * Connect() does not allow to get error notifications
528 * without closing the socket.
529 */
530 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
531 release_sock(sk);
532 timeo = schedule_timeout(timeo);
533 lock_sock(sk);
534 if (signal_pending(current) || !timeo)
535 break;
536 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
537 }
538 finish_wait(sk->sk_sleep, &wait);
539 return timeo;
540}
541
542/*
543 * Connect to a remote host. There is regrettably still a little
544 * TCP 'magic' in here.
545 */
546int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
547 int addr_len, int flags)
548{
549 struct sock *sk = sock->sk;
550 int err;
551 long timeo;
552
553 lock_sock(sk);
554
555 if (uaddr->sa_family == AF_UNSPEC) {
556 err = sk->sk_prot->disconnect(sk, flags);
557 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
558 goto out;
559 }
560
561 switch (sock->state) {
562 default:
563 err = -EINVAL;
564 goto out;
565 case SS_CONNECTED:
566 err = -EISCONN;
567 goto out;
568 case SS_CONNECTING:
569 err = -EALREADY;
570 /* Fall out of switch with err, set for this state */
571 break;
572 case SS_UNCONNECTED:
573 err = -EISCONN;
574 if (sk->sk_state != TCP_CLOSE)
575 goto out;
576
577 err = sk->sk_prot->connect(sk, uaddr, addr_len);
578 if (err < 0)
579 goto out;
580
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900581 sock->state = SS_CONNECTING;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582
583 /* Just entered SS_CONNECTING state; the only
584 * difference is that return value in non-blocking
585 * case is EINPROGRESS, rather than EALREADY.
586 */
587 err = -EINPROGRESS;
588 break;
589 }
590
591 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
592
593 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
594 /* Error code is set above */
595 if (!timeo || !inet_wait_for_connect(sk, timeo))
596 goto out;
597
598 err = sock_intr_errno(timeo);
599 if (signal_pending(current))
600 goto out;
601 }
602
603 /* Connection was closed by RST, timeout, ICMP error
604 * or another process disconnected us.
605 */
606 if (sk->sk_state == TCP_CLOSE)
607 goto sock_error;
608
609 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
610 * and error was received after socket entered established state.
611 * Hence, it is handled normally after connect() return successfully.
612 */
613
614 sock->state = SS_CONNECTED;
615 err = 0;
616out:
617 release_sock(sk);
618 return err;
619
620sock_error:
621 err = sock_error(sk) ? : -ECONNABORTED;
622 sock->state = SS_UNCONNECTED;
623 if (sk->sk_prot->disconnect(sk, flags))
624 sock->state = SS_DISCONNECTING;
625 goto out;
626}
627
628/*
629 * Accept a pending connection. The TCP layer now gives BSD semantics.
630 */
631
632int inet_accept(struct socket *sock, struct socket *newsock, int flags)
633{
634 struct sock *sk1 = sock->sk;
635 int err = -EINVAL;
636 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
637
638 if (!sk2)
639 goto do_err;
640
641 lock_sock(sk2);
642
643 BUG_TRAP((1 << sk2->sk_state) &
644 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE));
645
646 sock_graft(sk2, newsock);
647
648 newsock->state = SS_CONNECTED;
649 err = 0;
650 release_sock(sk2);
651do_err:
652 return err;
653}
654
655
656/*
657 * This does both peername and sockname.
658 */
659int inet_getname(struct socket *sock, struct sockaddr *uaddr,
660 int *uaddr_len, int peer)
661{
662 struct sock *sk = sock->sk;
663 struct inet_sock *inet = inet_sk(sk);
664 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
665
666 sin->sin_family = AF_INET;
667 if (peer) {
668 if (!inet->dport ||
669 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
670 peer == 1))
671 return -ENOTCONN;
672 sin->sin_port = inet->dport;
673 sin->sin_addr.s_addr = inet->daddr;
674 } else {
Al Viro714e85b2006-11-14 20:51:49 -0800675 __be32 addr = inet->rcv_saddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 if (!addr)
677 addr = inet->saddr;
678 sin->sin_port = inet->sport;
679 sin->sin_addr.s_addr = addr;
680 }
681 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
682 *uaddr_len = sizeof(*sin);
683 return 0;
684}
685
686int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
687 size_t size)
688{
689 struct sock *sk = sock->sk;
690
691 /* We may need to bind the socket. */
692 if (!inet_sk(sk)->num && inet_autobind(sk))
693 return -EAGAIN;
694
695 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
696}
697
698
699static ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
700{
701 struct sock *sk = sock->sk;
702
703 /* We may need to bind the socket. */
704 if (!inet_sk(sk)->num && inet_autobind(sk))
705 return -EAGAIN;
706
707 if (sk->sk_prot->sendpage)
708 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
709 return sock_no_sendpage(sock, page, offset, size, flags);
710}
711
712
713int inet_shutdown(struct socket *sock, int how)
714{
715 struct sock *sk = sock->sk;
716 int err = 0;
717
718 /* This should really check to make sure
719 * the socket is a TCP socket. (WHY AC...)
720 */
721 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
722 1->2 bit 2 snds.
723 2->3 */
724 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
725 return -EINVAL;
726
727 lock_sock(sk);
728 if (sock->state == SS_CONNECTING) {
729 if ((1 << sk->sk_state) &
730 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
731 sock->state = SS_DISCONNECTING;
732 else
733 sock->state = SS_CONNECTED;
734 }
735
736 switch (sk->sk_state) {
737 case TCP_CLOSE:
738 err = -ENOTCONN;
739 /* Hack to wake up other listeners, who can poll for
740 POLLHUP, even on eg. unconnected UDP sockets -- RR */
741 default:
742 sk->sk_shutdown |= how;
743 if (sk->sk_prot->shutdown)
744 sk->sk_prot->shutdown(sk, how);
745 break;
746
747 /* Remaining two branches are temporary solution for missing
748 * close() in multithreaded environment. It is _not_ a good idea,
749 * but we have no choice until close() is repaired at VFS level.
750 */
751 case TCP_LISTEN:
752 if (!(how & RCV_SHUTDOWN))
753 break;
754 /* Fall through */
755 case TCP_SYN_SENT:
756 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
757 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
758 break;
759 }
760
761 /* Wake up anyone sleeping in poll. */
762 sk->sk_state_change(sk);
763 release_sock(sk);
764 return err;
765}
766
767/*
768 * ioctl() calls you can issue on an INET socket. Most of these are
769 * device configuration and stuff and very rarely used. Some ioctls
770 * pass on to the socket itself.
771 *
772 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
773 * loads the devconfigure module does its configuring and unloads it.
774 * There's a good 20K of config code hanging around the kernel.
775 */
776
777int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
778{
779 struct sock *sk = sock->sk;
780 int err = 0;
781
782 switch (cmd) {
783 case SIOCGSTAMP:
784 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
785 break;
Eric Dumazetae40eb12007-03-18 17:33:16 -0700786 case SIOCGSTAMPNS:
787 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
788 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 case SIOCADDRT:
790 case SIOCDELRT:
791 case SIOCRTMSG:
792 err = ip_rt_ioctl(cmd, (void __user *)arg);
793 break;
794 case SIOCDARP:
795 case SIOCGARP:
796 case SIOCSARP:
797 err = arp_ioctl(cmd, (void __user *)arg);
798 break;
799 case SIOCGIFADDR:
800 case SIOCSIFADDR:
801 case SIOCGIFBRDADDR:
802 case SIOCSIFBRDADDR:
803 case SIOCGIFNETMASK:
804 case SIOCSIFNETMASK:
805 case SIOCGIFDSTADDR:
806 case SIOCSIFDSTADDR:
807 case SIOCSIFPFLAGS:
808 case SIOCGIFPFLAGS:
809 case SIOCSIFFLAGS:
810 err = devinet_ioctl(cmd, (void __user *)arg);
811 break;
812 default:
Christoph Hellwigb5e5fa52006-01-03 14:18:33 -0800813 if (sk->sk_prot->ioctl)
814 err = sk->sk_prot->ioctl(sk, cmd, arg);
815 else
816 err = -ENOIOCTLCMD;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 break;
818 }
819 return err;
820}
821
Eric Dumazet90ddc4f2005-12-22 12:49:22 -0800822const struct proto_ops inet_stream_ops = {
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -0800823 .family = PF_INET,
824 .owner = THIS_MODULE,
825 .release = inet_release,
826 .bind = inet_bind,
827 .connect = inet_stream_connect,
828 .socketpair = sock_no_socketpair,
829 .accept = inet_accept,
830 .getname = inet_getname,
831 .poll = tcp_poll,
832 .ioctl = inet_ioctl,
833 .listen = inet_listen,
834 .shutdown = inet_shutdown,
835 .setsockopt = sock_common_setsockopt,
836 .getsockopt = sock_common_getsockopt,
David S. Miller3516ffb2007-08-02 19:23:56 -0700837 .sendmsg = tcp_sendmsg,
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -0800838 .recvmsg = sock_common_recvmsg,
839 .mmap = sock_no_mmap,
840 .sendpage = tcp_sendpage,
Jens Axboe9c55e012007-11-06 23:30:13 -0800841 .splice_read = tcp_splice_read,
Dmitry Mishin3fdadf72006-03-20 22:45:21 -0800842#ifdef CONFIG_COMPAT
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -0800843 .compat_setsockopt = compat_sock_common_setsockopt,
844 .compat_getsockopt = compat_sock_common_getsockopt,
Dmitry Mishin3fdadf72006-03-20 22:45:21 -0800845#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846};
847
Eric Dumazet90ddc4f2005-12-22 12:49:22 -0800848const struct proto_ops inet_dgram_ops = {
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -0800849 .family = PF_INET,
850 .owner = THIS_MODULE,
851 .release = inet_release,
852 .bind = inet_bind,
853 .connect = inet_dgram_connect,
854 .socketpair = sock_no_socketpair,
855 .accept = sock_no_accept,
856 .getname = inet_getname,
857 .poll = udp_poll,
858 .ioctl = inet_ioctl,
859 .listen = sock_no_listen,
860 .shutdown = inet_shutdown,
861 .setsockopt = sock_common_setsockopt,
862 .getsockopt = sock_common_getsockopt,
863 .sendmsg = inet_sendmsg,
864 .recvmsg = sock_common_recvmsg,
865 .mmap = sock_no_mmap,
866 .sendpage = inet_sendpage,
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
873/*
874 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
875 * udp_poll
876 */
Eric Dumazet90ddc4f2005-12-22 12:49:22 -0800877static const struct proto_ops inet_sockraw_ops = {
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -0800878 .family = PF_INET,
879 .owner = THIS_MODULE,
880 .release = inet_release,
881 .bind = inet_bind,
882 .connect = inet_dgram_connect,
883 .socketpair = sock_no_socketpair,
884 .accept = sock_no_accept,
885 .getname = inet_getname,
886 .poll = datagram_poll,
887 .ioctl = inet_ioctl,
888 .listen = sock_no_listen,
889 .shutdown = inet_shutdown,
890 .setsockopt = sock_common_setsockopt,
891 .getsockopt = sock_common_getsockopt,
892 .sendmsg = inet_sendmsg,
893 .recvmsg = sock_common_recvmsg,
894 .mmap = sock_no_mmap,
895 .sendpage = inet_sendpage,
Dmitry Mishin3fdadf72006-03-20 22:45:21 -0800896#ifdef CONFIG_COMPAT
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -0800897 .compat_setsockopt = compat_sock_common_setsockopt,
898 .compat_getsockopt = compat_sock_common_getsockopt,
Dmitry Mishin3fdadf72006-03-20 22:45:21 -0800899#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900};
901
902static struct net_proto_family inet_family_ops = {
903 .family = PF_INET,
904 .create = inet_create,
905 .owner = THIS_MODULE,
906};
907
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908/* Upon startup we insert all the elements in inetsw_array[] into
909 * the linked list inetsw.
910 */
911static struct inet_protosw inetsw_array[] =
912{
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900913 {
914 .type = SOCK_STREAM,
915 .protocol = IPPROTO_TCP,
916 .prot = &tcp_prot,
917 .ops = &inet_stream_ops,
918 .capability = -1,
919 .no_check = 0,
920 .flags = INET_PROTOSW_PERMANENT |
Arnaldo Carvalho de Melod83d8462005-12-13 23:26:10 -0800921 INET_PROTOSW_ICSK,
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900922 },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900924 {
925 .type = SOCK_DGRAM,
926 .protocol = IPPROTO_UDP,
927 .prot = &udp_prot,
928 .ops = &inet_dgram_ops,
929 .capability = -1,
930 .no_check = UDP_CSUM_DEFAULT,
931 .flags = INET_PROTOSW_PERMANENT,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 },
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900933
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934
935 {
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900936 .type = SOCK_RAW,
937 .protocol = IPPROTO_IP, /* wild card */
938 .prot = &raw_prot,
939 .ops = &inet_sockraw_ops,
940 .capability = CAP_NET_RAW,
941 .no_check = UDP_CSUM_DEFAULT,
942 .flags = INET_PROTOSW_REUSE,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943 }
944};
945
Denis Chengc40f6ff2007-09-16 16:39:25 -0700946#define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947
948void inet_register_protosw(struct inet_protosw *p)
949{
950 struct list_head *lh;
951 struct inet_protosw *answer;
952 int protocol = p->protocol;
953 struct list_head *last_perm;
954
955 spin_lock_bh(&inetsw_lock);
956
957 if (p->type >= SOCK_MAX)
958 goto out_illegal;
959
960 /* If we are trying to override a permanent protocol, bail. */
961 answer = NULL;
962 last_perm = &inetsw[p->type];
963 list_for_each(lh, &inetsw[p->type]) {
964 answer = list_entry(lh, struct inet_protosw, list);
965
966 /* Check only the non-wild match. */
967 if (INET_PROTOSW_PERMANENT & answer->flags) {
968 if (protocol == answer->protocol)
969 break;
970 last_perm = lh;
971 }
972
973 answer = NULL;
974 }
975 if (answer)
976 goto out_permanent;
977
978 /* Add the new entry after the last permanent entry if any, so that
979 * the new entry does not override a permanent entry when matched with
980 * a wild-card protocol. But it is allowed to override any existing
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900981 * non-permanent entry. This means that when we remove this entry, the
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 * system automatically returns to the old behavior.
983 */
984 list_add_rcu(&p->list, last_perm);
985out:
986 spin_unlock_bh(&inetsw_lock);
987
988 synchronize_net();
989
990 return;
991
992out_permanent:
993 printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
994 protocol);
995 goto out;
996
997out_illegal:
998 printk(KERN_ERR
999 "Ignoring attempt to register invalid socket type %d.\n",
1000 p->type);
1001 goto out;
1002}
1003
1004void inet_unregister_protosw(struct inet_protosw *p)
1005{
1006 if (INET_PROTOSW_PERMANENT & p->flags) {
1007 printk(KERN_ERR
1008 "Attempt to unregister permanent protocol %d.\n",
1009 p->protocol);
1010 } else {
1011 spin_lock_bh(&inetsw_lock);
1012 list_del_rcu(&p->list);
1013 spin_unlock_bh(&inetsw_lock);
1014
1015 synchronize_net();
1016 }
1017}
1018
Arnaldo Carvalho de Melo32519f12005-08-09 19:50:02 -07001019/*
1020 * Shall we try to damage output packets if routing dev changes?
1021 */
1022
Brian Haleyab32ea52006-09-22 14:15:41 -07001023int sysctl_ip_dynaddr __read_mostly;
Arnaldo Carvalho de Melo32519f12005-08-09 19:50:02 -07001024
1025static int inet_sk_reselect_saddr(struct sock *sk)
1026{
1027 struct inet_sock *inet = inet_sk(sk);
1028 int err;
1029 struct rtable *rt;
Al Viro714e85b2006-11-14 20:51:49 -08001030 __be32 old_saddr = inet->saddr;
1031 __be32 new_saddr;
Al Virobada8ad2006-09-26 21:27:15 -07001032 __be32 daddr = inet->daddr;
Arnaldo Carvalho de Melo32519f12005-08-09 19:50:02 -07001033
1034 if (inet->opt && inet->opt->srr)
1035 daddr = inet->opt->faddr;
1036
1037 /* Query new route. */
1038 err = ip_route_connect(&rt, daddr, 0,
1039 RT_CONN_FLAGS(sk),
1040 sk->sk_bound_dev_if,
1041 sk->sk_protocol,
David S. Miller8eb90862007-02-08 02:09:21 -08001042 inet->sport, inet->dport, sk, 0);
Arnaldo Carvalho de Melo32519f12005-08-09 19:50:02 -07001043 if (err)
1044 return err;
1045
1046 sk_setup_caps(sk, &rt->u.dst);
1047
1048 new_saddr = rt->rt_src;
1049
1050 if (new_saddr == old_saddr)
1051 return 0;
1052
1053 if (sysctl_ip_dynaddr > 1) {
1054 printk(KERN_INFO "%s(): shifting inet->"
1055 "saddr from %d.%d.%d.%d to %d.%d.%d.%d\n",
1056 __FUNCTION__,
1057 NIPQUAD(old_saddr),
1058 NIPQUAD(new_saddr));
1059 }
1060
1061 inet->saddr = inet->rcv_saddr = new_saddr;
1062
1063 /*
1064 * XXX The only one ugly spot where we need to
1065 * XXX really change the sockets identity after
1066 * XXX it has entered the hashes. -DaveM
1067 *
1068 * Besides that, it does not check for connection
1069 * uniqueness. Wait for troubles.
1070 */
1071 __sk_prot_rehash(sk);
1072 return 0;
1073}
1074
1075int inet_sk_rebuild_header(struct sock *sk)
1076{
1077 struct inet_sock *inet = inet_sk(sk);
1078 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
Al Viro3ca3c682006-09-27 18:28:07 -07001079 __be32 daddr;
Arnaldo Carvalho de Melo32519f12005-08-09 19:50:02 -07001080 int err;
1081
1082 /* Route is OK, nothing to do. */
1083 if (rt)
1084 return 0;
1085
1086 /* Reroute. */
1087 daddr = inet->daddr;
1088 if (inet->opt && inet->opt->srr)
1089 daddr = inet->opt->faddr;
1090{
1091 struct flowi fl = {
1092 .oif = sk->sk_bound_dev_if,
1093 .nl_u = {
1094 .ip4_u = {
1095 .daddr = daddr,
1096 .saddr = inet->saddr,
1097 .tos = RT_CONN_FLAGS(sk),
1098 },
1099 },
1100 .proto = sk->sk_protocol,
1101 .uli_u = {
1102 .ports = {
1103 .sport = inet->sport,
1104 .dport = inet->dport,
1105 },
1106 },
1107 };
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001108
Venkat Yekkiralabeb8d132006-08-04 23:12:42 -07001109 security_sk_classify_flow(sk, &fl);
Arnaldo Carvalho de Melo32519f12005-08-09 19:50:02 -07001110 err = ip_route_output_flow(&rt, &fl, sk, 0);
1111}
1112 if (!err)
1113 sk_setup_caps(sk, &rt->u.dst);
1114 else {
1115 /* Routing failed... */
1116 sk->sk_route_caps = 0;
1117 /*
1118 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1119 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1120 */
1121 if (!sysctl_ip_dynaddr ||
1122 sk->sk_state != TCP_SYN_SENT ||
1123 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1124 (err = inet_sk_reselect_saddr(sk)) != 0)
1125 sk->sk_err_soft = -err;
1126 }
1127
1128 return err;
1129}
1130
1131EXPORT_SYMBOL(inet_sk_rebuild_header);
1132
Herbert Xua430a432006-07-08 13:34:56 -07001133static int inet_gso_send_check(struct sk_buff *skb)
1134{
1135 struct iphdr *iph;
1136 struct net_protocol *ops;
1137 int proto;
1138 int ihl;
1139 int err = -EINVAL;
1140
1141 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1142 goto out;
1143
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001144 iph = ip_hdr(skb);
Herbert Xua430a432006-07-08 13:34:56 -07001145 ihl = iph->ihl * 4;
1146 if (ihl < sizeof(*iph))
1147 goto out;
1148
1149 if (unlikely(!pskb_may_pull(skb, ihl)))
1150 goto out;
1151
Arnaldo Carvalho de Melobadff6d2007-03-13 13:06:52 -03001152 __skb_pull(skb, ihl);
1153 skb_reset_transport_header(skb);
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001154 iph = ip_hdr(skb);
Herbert Xua430a432006-07-08 13:34:56 -07001155 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1156 err = -EPROTONOSUPPORT;
1157
1158 rcu_read_lock();
1159 ops = rcu_dereference(inet_protos[proto]);
1160 if (likely(ops && ops->gso_send_check))
1161 err = ops->gso_send_check(skb);
1162 rcu_read_unlock();
1163
1164out:
1165 return err;
1166}
1167
Herbert Xu576a30e2006-06-27 13:22:38 -07001168static struct sk_buff *inet_gso_segment(struct sk_buff *skb, int features)
Herbert Xuf4c50d92006-06-22 03:02:40 -07001169{
1170 struct sk_buff *segs = ERR_PTR(-EINVAL);
1171 struct iphdr *iph;
1172 struct net_protocol *ops;
1173 int proto;
1174 int ihl;
1175 int id;
1176
Stephen Hemmingerd212f872007-06-27 00:47:37 -07001177 if (!(features & NETIF_F_V4_CSUM))
1178 features &= ~NETIF_F_SG;
1179
Herbert Xubbcf4672006-07-03 19:38:35 -07001180 if (unlikely(skb_shinfo(skb)->gso_type &
1181 ~(SKB_GSO_TCPV4 |
1182 SKB_GSO_UDP |
1183 SKB_GSO_DODGY |
1184 SKB_GSO_TCP_ECN |
1185 0)))
1186 goto out;
1187
1188 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
Herbert Xuf4c50d92006-06-22 03:02:40 -07001189 goto out;
1190
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001191 iph = ip_hdr(skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07001192 ihl = iph->ihl * 4;
1193 if (ihl < sizeof(*iph))
1194 goto out;
1195
Herbert Xubbcf4672006-07-03 19:38:35 -07001196 if (unlikely(!pskb_may_pull(skb, ihl)))
Herbert Xuf4c50d92006-06-22 03:02:40 -07001197 goto out;
1198
Arnaldo Carvalho de Melobadff6d2007-03-13 13:06:52 -03001199 __skb_pull(skb, ihl);
1200 skb_reset_transport_header(skb);
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001201 iph = ip_hdr(skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07001202 id = ntohs(iph->id);
1203 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1204 segs = ERR_PTR(-EPROTONOSUPPORT);
1205
1206 rcu_read_lock();
1207 ops = rcu_dereference(inet_protos[proto]);
Herbert Xubbcf4672006-07-03 19:38:35 -07001208 if (likely(ops && ops->gso_segment))
Herbert Xu576a30e2006-06-27 13:22:38 -07001209 segs = ops->gso_segment(skb, features);
Herbert Xuf4c50d92006-06-22 03:02:40 -07001210 rcu_read_unlock();
1211
Herbert Xu576a30e2006-06-27 13:22:38 -07001212 if (!segs || unlikely(IS_ERR(segs)))
Herbert Xuf4c50d92006-06-22 03:02:40 -07001213 goto out;
1214
1215 skb = segs;
1216 do {
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001217 iph = ip_hdr(skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07001218 iph->id = htons(id++);
1219 iph->tot_len = htons(skb->len - skb->mac_len);
1220 iph->check = 0;
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -07001221 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
Herbert Xuf4c50d92006-06-22 03:02:40 -07001222 } while ((skb = skb->next));
1223
1224out:
1225 return segs;
1226}
1227
Herbert Xu5e0f0432007-04-24 21:53:35 -07001228unsigned long snmp_fold_field(void *mib[], int offt)
1229{
1230 unsigned long res = 0;
1231 int i;
1232
1233 for_each_possible_cpu(i) {
1234 res += *(((unsigned long *) per_cpu_ptr(mib[0], i)) + offt);
1235 res += *(((unsigned long *) per_cpu_ptr(mib[1], i)) + offt);
1236 }
1237 return res;
1238}
1239EXPORT_SYMBOL_GPL(snmp_fold_field);
1240
1241int snmp_mib_init(void *ptr[2], size_t mibsize, size_t mibalign)
1242{
1243 BUG_ON(ptr == NULL);
1244 ptr[0] = __alloc_percpu(mibsize);
1245 if (!ptr[0])
1246 goto err0;
1247 ptr[1] = __alloc_percpu(mibsize);
1248 if (!ptr[1])
1249 goto err1;
1250 return 0;
1251err1:
1252 free_percpu(ptr[0]);
1253 ptr[0] = NULL;
1254err0:
1255 return -ENOMEM;
1256}
1257EXPORT_SYMBOL_GPL(snmp_mib_init);
1258
1259void snmp_mib_free(void *ptr[2])
1260{
1261 BUG_ON(ptr == NULL);
1262 free_percpu(ptr[0]);
1263 free_percpu(ptr[1]);
1264 ptr[0] = ptr[1] = NULL;
1265}
1266EXPORT_SYMBOL_GPL(snmp_mib_free);
1267
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268#ifdef CONFIG_IP_MULTICAST
1269static struct net_protocol igmp_protocol = {
1270 .handler = igmp_rcv,
1271};
1272#endif
1273
1274static struct net_protocol tcp_protocol = {
1275 .handler = tcp_v4_rcv,
1276 .err_handler = tcp_v4_err,
Herbert Xua430a432006-07-08 13:34:56 -07001277 .gso_send_check = tcp_v4_gso_send_check,
Herbert Xuf4c50d92006-06-22 03:02:40 -07001278 .gso_segment = tcp_tso_segment,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 .no_policy = 1,
1280};
1281
1282static struct net_protocol udp_protocol = {
1283 .handler = udp_rcv,
1284 .err_handler = udp_err,
1285 .no_policy = 1,
1286};
1287
1288static struct net_protocol icmp_protocol = {
1289 .handler = icmp_rcv,
1290};
1291
1292static int __init init_ipv4_mibs(void)
1293{
YOSHIFUJI Hideaki33490172007-04-20 15:57:15 -07001294 if (snmp_mib_init((void **)net_statistics,
1295 sizeof(struct linux_mib),
1296 __alignof__(struct linux_mib)) < 0)
1297 goto err_net_mib;
1298 if (snmp_mib_init((void **)ip_statistics,
1299 sizeof(struct ipstats_mib),
1300 __alignof__(struct ipstats_mib)) < 0)
1301 goto err_ip_mib;
1302 if (snmp_mib_init((void **)icmp_statistics,
1303 sizeof(struct icmp_mib),
1304 __alignof__(struct icmp_mib)) < 0)
1305 goto err_icmp_mib;
David L Stevens96793b42007-09-17 09:57:33 -07001306 if (snmp_mib_init((void **)icmpmsg_statistics,
1307 sizeof(struct icmpmsg_mib),
1308 __alignof__(struct icmpmsg_mib)) < 0)
1309 goto err_icmpmsg_mib;
YOSHIFUJI Hideaki33490172007-04-20 15:57:15 -07001310 if (snmp_mib_init((void **)tcp_statistics,
1311 sizeof(struct tcp_mib),
1312 __alignof__(struct tcp_mib)) < 0)
1313 goto err_tcp_mib;
1314 if (snmp_mib_init((void **)udp_statistics,
1315 sizeof(struct udp_mib),
1316 __alignof__(struct udp_mib)) < 0)
1317 goto err_udp_mib;
1318 if (snmp_mib_init((void **)udplite_statistics,
1319 sizeof(struct udp_mib),
1320 __alignof__(struct udp_mib)) < 0)
1321 goto err_udplite_mib;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322
YOSHIFUJI Hideaki33490172007-04-20 15:57:15 -07001323 tcp_mib_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324
1325 return 0;
YOSHIFUJI Hideaki33490172007-04-20 15:57:15 -07001326
1327err_udplite_mib:
1328 snmp_mib_free((void **)udp_statistics);
1329err_udp_mib:
1330 snmp_mib_free((void **)tcp_statistics);
1331err_tcp_mib:
David L Stevens96793b42007-09-17 09:57:33 -07001332 snmp_mib_free((void **)icmpmsg_statistics);
1333err_icmpmsg_mib:
YOSHIFUJI Hideaki33490172007-04-20 15:57:15 -07001334 snmp_mib_free((void **)icmp_statistics);
1335err_icmp_mib:
1336 snmp_mib_free((void **)ip_statistics);
1337err_ip_mib:
1338 snmp_mib_free((void **)net_statistics);
1339err_net_mib:
1340 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341}
1342
1343static int ipv4_proc_init(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344
Herbert Xu30e224d2005-07-05 14:40:10 -07001345/*
1346 * IP protocol layer initialiser
1347 */
1348
1349static struct packet_type ip_packet_type = {
1350 .type = __constant_htons(ETH_P_IP),
1351 .func = ip_rcv,
Herbert Xua430a432006-07-08 13:34:56 -07001352 .gso_send_check = inet_gso_send_check,
Herbert Xuf4c50d92006-06-22 03:02:40 -07001353 .gso_segment = inet_gso_segment,
Herbert Xu30e224d2005-07-05 14:40:10 -07001354};
1355
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356static int __init inet_init(void)
1357{
1358 struct sk_buff *dummy_skb;
1359 struct inet_protosw *q;
1360 struct list_head *r;
1361 int rc = -EINVAL;
1362
YOSHIFUJI Hideakief047f52006-09-01 00:29:06 -07001363 BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364
1365 rc = proto_register(&tcp_prot, 1);
1366 if (rc)
1367 goto out;
1368
1369 rc = proto_register(&udp_prot, 1);
1370 if (rc)
1371 goto out_unregister_tcp_proto;
1372
1373 rc = proto_register(&raw_prot, 1);
1374 if (rc)
1375 goto out_unregister_udp_proto;
1376
1377 /*
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001378 * Tell SOCKET that we are alive...
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 */
1380
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001381 (void)sock_register(&inet_family_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382
1383 /*
1384 * Add all the base protocols.
1385 */
1386
1387 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1388 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1389 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1390 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1391 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1392 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1393#ifdef CONFIG_IP_MULTICAST
1394 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1395 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1396#endif
1397
1398 /* Register the socket-side information for inet_create. */
1399 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1400 INIT_LIST_HEAD(r);
1401
1402 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1403 inet_register_protosw(q);
1404
1405 /*
1406 * Set the ARP module up
1407 */
1408
1409 arp_init();
1410
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001411 /*
1412 * Set the IP module up
1413 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414
1415 ip_init();
1416
1417 tcp_v4_init(&inet_family_ops);
1418
1419 /* Setup TCP slab cache for open requests. */
1420 tcp_init();
1421
Gerrit Renkerba4e58e2006-11-27 11:10:57 -08001422 /* Add UDP-Lite (RFC 3828) */
1423 udplite4_register();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424
1425 /*
1426 * Set the ICMP layer up
1427 */
1428
1429 icmp_init(&inet_family_ops);
1430
1431 /*
1432 * Initialise the multicast router
1433 */
1434#if defined(CONFIG_IP_MROUTE)
1435 ip_mr_init();
1436#endif
1437 /*
1438 * Initialise per-cpu ipv4 mibs
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001439 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440
Stephen Hemminger132adf52007-03-08 20:44:43 -08001441 if (init_ipv4_mibs())
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n"); ;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001443
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 ipv4_proc_init();
1445
1446 ipfrag_init();
1447
Herbert Xu30e224d2005-07-05 14:40:10 -07001448 dev_add_pack(&ip_packet_type);
1449
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 rc = 0;
1451out:
1452 return rc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453out_unregister_udp_proto:
1454 proto_unregister(&udp_prot);
Olaf Kirch321efff2006-09-27 16:33:45 -07001455out_unregister_tcp_proto:
1456 proto_unregister(&tcp_prot);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 goto out;
1458}
1459
Heiko Carstensa536e072006-04-28 15:19:17 -07001460fs_initcall(inet_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461
1462/* ------------------------------------------------------------------------ */
1463
1464#ifdef CONFIG_PROC_FS
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465static int __init ipv4_proc_init(void)
1466{
1467 int rc = 0;
1468
1469 if (raw_proc_init())
1470 goto out_raw;
1471 if (tcp4_proc_init())
1472 goto out_tcp;
1473 if (udp4_proc_init())
1474 goto out_udp;
1475 if (fib_proc_init())
1476 goto out_fib;
1477 if (ip_misc_proc_init())
1478 goto out_misc;
1479out:
1480 return rc;
1481out_misc:
1482 fib_proc_exit();
1483out_fib:
1484 udp4_proc_exit();
1485out_udp:
1486 tcp4_proc_exit();
1487out_tcp:
1488 raw_proc_exit();
1489out_raw:
1490 rc = -ENOMEM;
1491 goto out;
1492}
1493
1494#else /* CONFIG_PROC_FS */
1495static int __init ipv4_proc_init(void)
1496{
1497 return 0;
1498}
1499#endif /* CONFIG_PROC_FS */
1500
1501MODULE_ALIAS_NETPROTO(PF_INET);
1502
1503EXPORT_SYMBOL(inet_accept);
1504EXPORT_SYMBOL(inet_bind);
1505EXPORT_SYMBOL(inet_dgram_connect);
1506EXPORT_SYMBOL(inet_dgram_ops);
1507EXPORT_SYMBOL(inet_getname);
1508EXPORT_SYMBOL(inet_ioctl);
1509EXPORT_SYMBOL(inet_listen);
1510EXPORT_SYMBOL(inet_register_protosw);
1511EXPORT_SYMBOL(inet_release);
1512EXPORT_SYMBOL(inet_sendmsg);
1513EXPORT_SYMBOL(inet_shutdown);
1514EXPORT_SYMBOL(inet_sock_destruct);
1515EXPORT_SYMBOL(inet_stream_connect);
1516EXPORT_SYMBOL(inet_stream_ops);
1517EXPORT_SYMBOL(inet_unregister_protosw);
1518EXPORT_SYMBOL(net_statistics);
Neil Hormancdac4e02005-06-13 15:12:33 -07001519EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);