blob: 4f539bd48718eb0d26c692866a8d3f429fcbc5a6 [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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246/*
247 * Create an inet socket.
248 */
249
Eric W. Biederman1b8d7ae2007-10-08 23:24:22 -0700250static int inet_create(struct net *net, struct socket *sock, int protocol)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251{
252 struct sock *sk;
253 struct list_head *p;
254 struct inet_protosw *answer;
255 struct inet_sock *inet;
256 struct proto *answer_prot;
257 unsigned char answer_flags;
258 char answer_no_check;
Arnaldo Carvalho de Melobb97d312005-08-09 20:19:14 -0700259 int try_loading_module = 0;
Herbert Xu86c8f9d2005-12-02 20:43:26 -0800260 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261
Eric W. Biederman1b8d7ae2007-10-08 23:24:22 -0700262 if (net != &init_net)
263 return -EAFNOSUPPORT;
264
David S. Millerb3da2cf2007-03-23 11:40:27 -0700265 if (sock->type != SOCK_RAW &&
266 sock->type != SOCK_DGRAM &&
267 !inet_ehash_secret)
268 build_ehash_secret();
269
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 sock->state = SS_UNCONNECTED;
271
272 /* Look for the requested type/protocol pair. */
273 answer = NULL;
Arnaldo Carvalho de Melobb97d312005-08-09 20:19:14 -0700274lookup_protocol:
Herbert Xu86c8f9d2005-12-02 20:43:26 -0800275 err = -ESOCKTNOSUPPORT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 rcu_read_lock();
277 list_for_each_rcu(p, &inetsw[sock->type]) {
278 answer = list_entry(p, struct inet_protosw, list);
279
280 /* Check the non-wild match. */
281 if (protocol == answer->protocol) {
282 if (protocol != IPPROTO_IP)
283 break;
284 } else {
285 /* Check for the two wild cases. */
286 if (IPPROTO_IP == protocol) {
287 protocol = answer->protocol;
288 break;
289 }
290 if (IPPROTO_IP == answer->protocol)
291 break;
292 }
Herbert Xu86c8f9d2005-12-02 20:43:26 -0800293 err = -EPROTONOSUPPORT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 answer = NULL;
295 }
296
Arnaldo Carvalho de Melobb97d312005-08-09 20:19:14 -0700297 if (unlikely(answer == NULL)) {
298 if (try_loading_module < 2) {
299 rcu_read_unlock();
300 /*
301 * Be more specific, e.g. net-pf-2-proto-132-type-1
302 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
303 */
304 if (++try_loading_module == 1)
305 request_module("net-pf-%d-proto-%d-type-%d",
306 PF_INET, protocol, sock->type);
307 /*
308 * Fall back to generic, e.g. net-pf-2-proto-132
309 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
310 */
311 else
312 request_module("net-pf-%d-proto-%d",
313 PF_INET, protocol);
314 goto lookup_protocol;
315 } else
316 goto out_rcu_unlock;
317 }
318
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 err = -EPERM;
320 if (answer->capability > 0 && !capable(answer->capability))
321 goto out_rcu_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322
323 sock->ops = answer->ops;
324 answer_prot = answer->prot;
325 answer_no_check = answer->no_check;
326 answer_flags = answer->flags;
327 rcu_read_unlock();
328
329 BUG_TRAP(answer_prot->slab != NULL);
330
331 err = -ENOBUFS;
Pavel Emelyanov6257ff22007-11-01 00:39:31 -0700332 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 if (sk == NULL)
334 goto out;
335
336 err = 0;
337 sk->sk_no_check = answer_no_check;
338 if (INET_PROTOSW_REUSE & answer_flags)
339 sk->sk_reuse = 1;
340
341 inet = inet_sk(sk);
Paul Moore469de9b2007-01-09 14:37:06 -0800342 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343
344 if (SOCK_RAW == sock->type) {
345 inet->num = protocol;
346 if (IPPROTO_RAW == protocol)
347 inet->hdrincl = 1;
348 }
349
350 if (ipv4_config.no_pmtu_disc)
351 inet->pmtudisc = IP_PMTUDISC_DONT;
352 else
353 inet->pmtudisc = IP_PMTUDISC_WANT;
354
355 inet->id = 0;
356
357 sock_init_data(sock, sk);
358
359 sk->sk_destruct = inet_sock_destruct;
360 sk->sk_family = PF_INET;
361 sk->sk_protocol = protocol;
362 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
363
364 inet->uc_ttl = -1;
365 inet->mc_loop = 1;
366 inet->mc_ttl = 1;
367 inet->mc_index = 0;
368 inet->mc_list = NULL;
369
Arnaldo Carvalho de Meloe6848972005-08-09 19:45:38 -0700370 sk_refcnt_debug_inc(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371
372 if (inet->num) {
373 /* It assumes that any protocol which allows
374 * the user to assign a number at socket
375 * creation time automatically
376 * shares.
377 */
378 inet->sport = htons(inet->num);
379 /* Add to protocol hash chains. */
380 sk->sk_prot->hash(sk);
381 }
382
383 if (sk->sk_prot->init) {
384 err = sk->sk_prot->init(sk);
385 if (err)
386 sk_common_release(sk);
387 }
388out:
389 return err;
390out_rcu_unlock:
391 rcu_read_unlock();
392 goto out;
393}
394
395
396/*
397 * The peer socket should always be NULL (or else). When we call this
398 * function we are destroying the object and from then on nobody
399 * should refer to it.
400 */
401int inet_release(struct socket *sock)
402{
403 struct sock *sk = sock->sk;
404
405 if (sk) {
406 long timeout;
407
408 /* Applications forget to leave groups before exiting */
409 ip_mc_drop_socket(sk);
410
411 /* If linger is set, we don't return until the close
412 * is complete. Otherwise we return immediately. The
413 * actually closing is done the same either way.
414 *
415 * If the close is due to the process exiting, we never
416 * linger..
417 */
418 timeout = 0;
419 if (sock_flag(sk, SOCK_LINGER) &&
420 !(current->flags & PF_EXITING))
421 timeout = sk->sk_lingertime;
422 sock->sk = NULL;
423 sk->sk_prot->close(sk, timeout);
424 }
425 return 0;
426}
427
428/* It is off by default, see below. */
Brian Haleyab32ea52006-09-22 14:15:41 -0700429int sysctl_ip_nonlocal_bind __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430
431int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
432{
433 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
434 struct sock *sk = sock->sk;
435 struct inet_sock *inet = inet_sk(sk);
436 unsigned short snum;
437 int chk_addr_ret;
438 int err;
439
440 /* If the socket has its own bind function then use it. (RAW) */
441 if (sk->sk_prot->bind) {
442 err = sk->sk_prot->bind(sk, uaddr, addr_len);
443 goto out;
444 }
445 err = -EINVAL;
446 if (addr_len < sizeof(struct sockaddr_in))
447 goto out;
448
Eric W. Biederman6b175b22008-01-10 03:25:28 -0800449 chk_addr_ret = inet_addr_type(&init_net, addr->sin_addr.s_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450
451 /* Not specified by any standard per-se, however it breaks too
452 * many applications when removed. It is unfortunate since
453 * allowing applications to make a non-local bind solves
454 * several problems with systems using dynamic addressing.
455 * (ie. your servers still start up even if your ISDN link
456 * is temporarily down)
457 */
458 err = -EADDRNOTAVAIL;
459 if (!sysctl_ip_nonlocal_bind &&
460 !inet->freebind &&
461 addr->sin_addr.s_addr != INADDR_ANY &&
462 chk_addr_ret != RTN_LOCAL &&
463 chk_addr_ret != RTN_MULTICAST &&
464 chk_addr_ret != RTN_BROADCAST)
465 goto out;
466
467 snum = ntohs(addr->sin_port);
468 err = -EACCES;
469 if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
470 goto out;
471
472 /* We keep a pair of addresses. rcv_saddr is the one
473 * used by hash lookups, and saddr is used for transmit.
474 *
475 * In the BSD API these are the same except where it
476 * would be illegal to use them (multicast/broadcast) in
477 * which case the sending device address is used.
478 */
479 lock_sock(sk);
480
481 /* Check these errors (active socket, double bind). */
482 err = -EINVAL;
483 if (sk->sk_state != TCP_CLOSE || inet->num)
484 goto out_release_sock;
485
486 inet->rcv_saddr = inet->saddr = addr->sin_addr.s_addr;
487 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
488 inet->saddr = 0; /* Use device */
489
490 /* Make sure we are allowed to bind here. */
491 if (sk->sk_prot->get_port(sk, snum)) {
492 inet->saddr = inet->rcv_saddr = 0;
493 err = -EADDRINUSE;
494 goto out_release_sock;
495 }
496
497 if (inet->rcv_saddr)
498 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
499 if (snum)
500 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
501 inet->sport = htons(inet->num);
502 inet->daddr = 0;
503 inet->dport = 0;
504 sk_dst_reset(sk);
505 err = 0;
506out_release_sock:
507 release_sock(sk);
508out:
509 return err;
510}
511
512int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
513 int addr_len, int flags)
514{
515 struct sock *sk = sock->sk;
516
517 if (uaddr->sa_family == AF_UNSPEC)
518 return sk->sk_prot->disconnect(sk, flags);
519
520 if (!inet_sk(sk)->num && inet_autobind(sk))
521 return -EAGAIN;
522 return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
523}
524
525static long inet_wait_for_connect(struct sock *sk, long timeo)
526{
527 DEFINE_WAIT(wait);
528
529 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
530
531 /* Basic assumption: if someone sets sk->sk_err, he _must_
532 * change state of the socket from TCP_SYN_*.
533 * Connect() does not allow to get error notifications
534 * without closing the socket.
535 */
536 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
537 release_sock(sk);
538 timeo = schedule_timeout(timeo);
539 lock_sock(sk);
540 if (signal_pending(current) || !timeo)
541 break;
542 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
543 }
544 finish_wait(sk->sk_sleep, &wait);
545 return timeo;
546}
547
548/*
549 * Connect to a remote host. There is regrettably still a little
550 * TCP 'magic' in here.
551 */
552int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
553 int addr_len, int flags)
554{
555 struct sock *sk = sock->sk;
556 int err;
557 long timeo;
558
559 lock_sock(sk);
560
561 if (uaddr->sa_family == AF_UNSPEC) {
562 err = sk->sk_prot->disconnect(sk, flags);
563 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
564 goto out;
565 }
566
567 switch (sock->state) {
568 default:
569 err = -EINVAL;
570 goto out;
571 case SS_CONNECTED:
572 err = -EISCONN;
573 goto out;
574 case SS_CONNECTING:
575 err = -EALREADY;
576 /* Fall out of switch with err, set for this state */
577 break;
578 case SS_UNCONNECTED:
579 err = -EISCONN;
580 if (sk->sk_state != TCP_CLOSE)
581 goto out;
582
583 err = sk->sk_prot->connect(sk, uaddr, addr_len);
584 if (err < 0)
585 goto out;
586
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900587 sock->state = SS_CONNECTING;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588
589 /* Just entered SS_CONNECTING state; the only
590 * difference is that return value in non-blocking
591 * case is EINPROGRESS, rather than EALREADY.
592 */
593 err = -EINPROGRESS;
594 break;
595 }
596
597 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
598
599 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
600 /* Error code is set above */
601 if (!timeo || !inet_wait_for_connect(sk, timeo))
602 goto out;
603
604 err = sock_intr_errno(timeo);
605 if (signal_pending(current))
606 goto out;
607 }
608
609 /* Connection was closed by RST, timeout, ICMP error
610 * or another process disconnected us.
611 */
612 if (sk->sk_state == TCP_CLOSE)
613 goto sock_error;
614
615 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
616 * and error was received after socket entered established state.
617 * Hence, it is handled normally after connect() return successfully.
618 */
619
620 sock->state = SS_CONNECTED;
621 err = 0;
622out:
623 release_sock(sk);
624 return err;
625
626sock_error:
627 err = sock_error(sk) ? : -ECONNABORTED;
628 sock->state = SS_UNCONNECTED;
629 if (sk->sk_prot->disconnect(sk, flags))
630 sock->state = SS_DISCONNECTING;
631 goto out;
632}
633
634/*
635 * Accept a pending connection. The TCP layer now gives BSD semantics.
636 */
637
638int inet_accept(struct socket *sock, struct socket *newsock, int flags)
639{
640 struct sock *sk1 = sock->sk;
641 int err = -EINVAL;
642 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
643
644 if (!sk2)
645 goto do_err;
646
647 lock_sock(sk2);
648
649 BUG_TRAP((1 << sk2->sk_state) &
650 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE));
651
652 sock_graft(sk2, newsock);
653
654 newsock->state = SS_CONNECTED;
655 err = 0;
656 release_sock(sk2);
657do_err:
658 return err;
659}
660
661
662/*
663 * This does both peername and sockname.
664 */
665int inet_getname(struct socket *sock, struct sockaddr *uaddr,
666 int *uaddr_len, int peer)
667{
668 struct sock *sk = sock->sk;
669 struct inet_sock *inet = inet_sk(sk);
670 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
671
672 sin->sin_family = AF_INET;
673 if (peer) {
674 if (!inet->dport ||
675 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
676 peer == 1))
677 return -ENOTCONN;
678 sin->sin_port = inet->dport;
679 sin->sin_addr.s_addr = inet->daddr;
680 } else {
Al Viro714e85b2006-11-14 20:51:49 -0800681 __be32 addr = inet->rcv_saddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 if (!addr)
683 addr = inet->saddr;
684 sin->sin_port = inet->sport;
685 sin->sin_addr.s_addr = addr;
686 }
687 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
688 *uaddr_len = sizeof(*sin);
689 return 0;
690}
691
692int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
693 size_t size)
694{
695 struct sock *sk = sock->sk;
696
697 /* We may need to bind the socket. */
698 if (!inet_sk(sk)->num && inet_autobind(sk))
699 return -EAGAIN;
700
701 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
702}
703
704
705static ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
706{
707 struct sock *sk = sock->sk;
708
709 /* We may need to bind the socket. */
710 if (!inet_sk(sk)->num && inet_autobind(sk))
711 return -EAGAIN;
712
713 if (sk->sk_prot->sendpage)
714 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
715 return sock_no_sendpage(sock, page, offset, size, flags);
716}
717
718
719int inet_shutdown(struct socket *sock, int how)
720{
721 struct sock *sk = sock->sk;
722 int err = 0;
723
724 /* This should really check to make sure
725 * the socket is a TCP socket. (WHY AC...)
726 */
727 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
728 1->2 bit 2 snds.
729 2->3 */
730 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
731 return -EINVAL;
732
733 lock_sock(sk);
734 if (sock->state == SS_CONNECTING) {
735 if ((1 << sk->sk_state) &
736 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
737 sock->state = SS_DISCONNECTING;
738 else
739 sock->state = SS_CONNECTED;
740 }
741
742 switch (sk->sk_state) {
743 case TCP_CLOSE:
744 err = -ENOTCONN;
745 /* Hack to wake up other listeners, who can poll for
746 POLLHUP, even on eg. unconnected UDP sockets -- RR */
747 default:
748 sk->sk_shutdown |= how;
749 if (sk->sk_prot->shutdown)
750 sk->sk_prot->shutdown(sk, how);
751 break;
752
753 /* Remaining two branches are temporary solution for missing
754 * close() in multithreaded environment. It is _not_ a good idea,
755 * but we have no choice until close() is repaired at VFS level.
756 */
757 case TCP_LISTEN:
758 if (!(how & RCV_SHUTDOWN))
759 break;
760 /* Fall through */
761 case TCP_SYN_SENT:
762 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
763 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
764 break;
765 }
766
767 /* Wake up anyone sleeping in poll. */
768 sk->sk_state_change(sk);
769 release_sock(sk);
770 return err;
771}
772
773/*
774 * ioctl() calls you can issue on an INET socket. Most of these are
775 * device configuration and stuff and very rarely used. Some ioctls
776 * pass on to the socket itself.
777 *
778 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
779 * loads the devconfigure module does its configuring and unloads it.
780 * There's a good 20K of config code hanging around the kernel.
781 */
782
783int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
784{
785 struct sock *sk = sock->sk;
786 int err = 0;
Denis V. Luneve5b13cb2008-02-28 20:51:43 -0800787 struct net *net = sk->sk_net;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788
789 switch (cmd) {
790 case SIOCGSTAMP:
791 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
792 break;
Eric Dumazetae40eb12007-03-18 17:33:16 -0700793 case SIOCGSTAMPNS:
794 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
795 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 case SIOCADDRT:
797 case SIOCDELRT:
798 case SIOCRTMSG:
Denis V. Luneve5b13cb2008-02-28 20:51:43 -0800799 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 break;
801 case SIOCDARP:
802 case SIOCGARP:
803 case SIOCSARP:
Denis V. Luneve5b13cb2008-02-28 20:51:43 -0800804 err = arp_ioctl(net, cmd, (void __user *)arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 break;
806 case SIOCGIFADDR:
807 case SIOCSIFADDR:
808 case SIOCGIFBRDADDR:
809 case SIOCSIFBRDADDR:
810 case SIOCGIFNETMASK:
811 case SIOCSIFNETMASK:
812 case SIOCGIFDSTADDR:
813 case SIOCSIFDSTADDR:
814 case SIOCSIFPFLAGS:
815 case SIOCGIFPFLAGS:
816 case SIOCSIFFLAGS:
Denis V. Luneve5b13cb2008-02-28 20:51:43 -0800817 err = devinet_ioctl(net, cmd, (void __user *)arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 break;
819 default:
Christoph Hellwigb5e5fa52006-01-03 14:18:33 -0800820 if (sk->sk_prot->ioctl)
821 err = sk->sk_prot->ioctl(sk, cmd, arg);
822 else
823 err = -ENOIOCTLCMD;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 break;
825 }
826 return err;
827}
828
Eric Dumazet90ddc4f2005-12-22 12:49:22 -0800829const struct proto_ops inet_stream_ops = {
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -0800830 .family = PF_INET,
831 .owner = THIS_MODULE,
832 .release = inet_release,
833 .bind = inet_bind,
834 .connect = inet_stream_connect,
835 .socketpair = sock_no_socketpair,
836 .accept = inet_accept,
837 .getname = inet_getname,
838 .poll = tcp_poll,
839 .ioctl = inet_ioctl,
840 .listen = inet_listen,
841 .shutdown = inet_shutdown,
842 .setsockopt = sock_common_setsockopt,
843 .getsockopt = sock_common_getsockopt,
David S. Miller3516ffb2007-08-02 19:23:56 -0700844 .sendmsg = tcp_sendmsg,
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -0800845 .recvmsg = sock_common_recvmsg,
846 .mmap = sock_no_mmap,
847 .sendpage = tcp_sendpage,
Jens Axboe9c55e012007-11-06 23:30:13 -0800848 .splice_read = tcp_splice_read,
Dmitry Mishin3fdadf72006-03-20 22:45:21 -0800849#ifdef CONFIG_COMPAT
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -0800850 .compat_setsockopt = compat_sock_common_setsockopt,
851 .compat_getsockopt = compat_sock_common_getsockopt,
Dmitry Mishin3fdadf72006-03-20 22:45:21 -0800852#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853};
854
Eric Dumazet90ddc4f2005-12-22 12:49:22 -0800855const struct proto_ops inet_dgram_ops = {
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -0800856 .family = PF_INET,
857 .owner = THIS_MODULE,
858 .release = inet_release,
859 .bind = inet_bind,
860 .connect = inet_dgram_connect,
861 .socketpair = sock_no_socketpair,
862 .accept = sock_no_accept,
863 .getname = inet_getname,
864 .poll = udp_poll,
865 .ioctl = inet_ioctl,
866 .listen = sock_no_listen,
867 .shutdown = inet_shutdown,
868 .setsockopt = sock_common_setsockopt,
869 .getsockopt = sock_common_getsockopt,
870 .sendmsg = inet_sendmsg,
871 .recvmsg = sock_common_recvmsg,
872 .mmap = sock_no_mmap,
873 .sendpage = inet_sendpage,
Dmitry Mishin3fdadf72006-03-20 22:45:21 -0800874#ifdef CONFIG_COMPAT
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -0800875 .compat_setsockopt = compat_sock_common_setsockopt,
876 .compat_getsockopt = compat_sock_common_getsockopt,
Dmitry Mishin3fdadf72006-03-20 22:45:21 -0800877#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878};
879
880/*
881 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
882 * udp_poll
883 */
Eric Dumazet90ddc4f2005-12-22 12:49:22 -0800884static const struct proto_ops inet_sockraw_ops = {
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -0800885 .family = PF_INET,
886 .owner = THIS_MODULE,
887 .release = inet_release,
888 .bind = inet_bind,
889 .connect = inet_dgram_connect,
890 .socketpair = sock_no_socketpair,
891 .accept = sock_no_accept,
892 .getname = inet_getname,
893 .poll = datagram_poll,
894 .ioctl = inet_ioctl,
895 .listen = sock_no_listen,
896 .shutdown = inet_shutdown,
897 .setsockopt = sock_common_setsockopt,
898 .getsockopt = sock_common_getsockopt,
899 .sendmsg = inet_sendmsg,
900 .recvmsg = sock_common_recvmsg,
901 .mmap = sock_no_mmap,
902 .sendpage = inet_sendpage,
Dmitry Mishin3fdadf72006-03-20 22:45:21 -0800903#ifdef CONFIG_COMPAT
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -0800904 .compat_setsockopt = compat_sock_common_setsockopt,
905 .compat_getsockopt = compat_sock_common_getsockopt,
Dmitry Mishin3fdadf72006-03-20 22:45:21 -0800906#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907};
908
909static struct net_proto_family inet_family_ops = {
910 .family = PF_INET,
911 .create = inet_create,
912 .owner = THIS_MODULE,
913};
914
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915/* Upon startup we insert all the elements in inetsw_array[] into
916 * the linked list inetsw.
917 */
918static struct inet_protosw inetsw_array[] =
919{
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900920 {
921 .type = SOCK_STREAM,
922 .protocol = IPPROTO_TCP,
923 .prot = &tcp_prot,
924 .ops = &inet_stream_ops,
925 .capability = -1,
926 .no_check = 0,
927 .flags = INET_PROTOSW_PERMANENT |
Arnaldo Carvalho de Melod83d8462005-12-13 23:26:10 -0800928 INET_PROTOSW_ICSK,
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900929 },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900931 {
932 .type = SOCK_DGRAM,
933 .protocol = IPPROTO_UDP,
934 .prot = &udp_prot,
935 .ops = &inet_dgram_ops,
936 .capability = -1,
937 .no_check = UDP_CSUM_DEFAULT,
938 .flags = INET_PROTOSW_PERMANENT,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 },
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900940
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941
942 {
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900943 .type = SOCK_RAW,
944 .protocol = IPPROTO_IP, /* wild card */
945 .prot = &raw_prot,
946 .ops = &inet_sockraw_ops,
947 .capability = CAP_NET_RAW,
948 .no_check = UDP_CSUM_DEFAULT,
949 .flags = INET_PROTOSW_REUSE,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 }
951};
952
Denis Chengc40f6ff2007-09-16 16:39:25 -0700953#define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954
955void inet_register_protosw(struct inet_protosw *p)
956{
957 struct list_head *lh;
958 struct inet_protosw *answer;
959 int protocol = p->protocol;
960 struct list_head *last_perm;
961
962 spin_lock_bh(&inetsw_lock);
963
964 if (p->type >= SOCK_MAX)
965 goto out_illegal;
966
967 /* If we are trying to override a permanent protocol, bail. */
968 answer = NULL;
969 last_perm = &inetsw[p->type];
970 list_for_each(lh, &inetsw[p->type]) {
971 answer = list_entry(lh, struct inet_protosw, list);
972
973 /* Check only the non-wild match. */
974 if (INET_PROTOSW_PERMANENT & answer->flags) {
975 if (protocol == answer->protocol)
976 break;
977 last_perm = lh;
978 }
979
980 answer = NULL;
981 }
982 if (answer)
983 goto out_permanent;
984
985 /* Add the new entry after the last permanent entry if any, so that
986 * the new entry does not override a permanent entry when matched with
987 * a wild-card protocol. But it is allowed to override any existing
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900988 * non-permanent entry. This means that when we remove this entry, the
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 * system automatically returns to the old behavior.
990 */
991 list_add_rcu(&p->list, last_perm);
992out:
993 spin_unlock_bh(&inetsw_lock);
994
995 synchronize_net();
996
997 return;
998
999out_permanent:
1000 printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
1001 protocol);
1002 goto out;
1003
1004out_illegal:
1005 printk(KERN_ERR
1006 "Ignoring attempt to register invalid socket type %d.\n",
1007 p->type);
1008 goto out;
1009}
1010
1011void inet_unregister_protosw(struct inet_protosw *p)
1012{
1013 if (INET_PROTOSW_PERMANENT & p->flags) {
1014 printk(KERN_ERR
1015 "Attempt to unregister permanent protocol %d.\n",
1016 p->protocol);
1017 } else {
1018 spin_lock_bh(&inetsw_lock);
1019 list_del_rcu(&p->list);
1020 spin_unlock_bh(&inetsw_lock);
1021
1022 synchronize_net();
1023 }
1024}
1025
Arnaldo Carvalho de Melo32519f12005-08-09 19:50:02 -07001026/*
1027 * Shall we try to damage output packets if routing dev changes?
1028 */
1029
Brian Haleyab32ea52006-09-22 14:15:41 -07001030int sysctl_ip_dynaddr __read_mostly;
Arnaldo Carvalho de Melo32519f12005-08-09 19:50:02 -07001031
1032static int inet_sk_reselect_saddr(struct sock *sk)
1033{
1034 struct inet_sock *inet = inet_sk(sk);
1035 int err;
1036 struct rtable *rt;
Al Viro714e85b2006-11-14 20:51:49 -08001037 __be32 old_saddr = inet->saddr;
1038 __be32 new_saddr;
Al Virobada8ad2006-09-26 21:27:15 -07001039 __be32 daddr = inet->daddr;
Arnaldo Carvalho de Melo32519f12005-08-09 19:50:02 -07001040
1041 if (inet->opt && inet->opt->srr)
1042 daddr = inet->opt->faddr;
1043
1044 /* Query new route. */
1045 err = ip_route_connect(&rt, daddr, 0,
1046 RT_CONN_FLAGS(sk),
1047 sk->sk_bound_dev_if,
1048 sk->sk_protocol,
David S. Miller8eb90862007-02-08 02:09:21 -08001049 inet->sport, inet->dport, sk, 0);
Arnaldo Carvalho de Melo32519f12005-08-09 19:50:02 -07001050 if (err)
1051 return err;
1052
1053 sk_setup_caps(sk, &rt->u.dst);
1054
1055 new_saddr = rt->rt_src;
1056
1057 if (new_saddr == old_saddr)
1058 return 0;
1059
1060 if (sysctl_ip_dynaddr > 1) {
1061 printk(KERN_INFO "%s(): shifting inet->"
1062 "saddr from %d.%d.%d.%d to %d.%d.%d.%d\n",
1063 __FUNCTION__,
1064 NIPQUAD(old_saddr),
1065 NIPQUAD(new_saddr));
1066 }
1067
1068 inet->saddr = inet->rcv_saddr = new_saddr;
1069
1070 /*
1071 * XXX The only one ugly spot where we need to
1072 * XXX really change the sockets identity after
1073 * XXX it has entered the hashes. -DaveM
1074 *
1075 * Besides that, it does not check for connection
1076 * uniqueness. Wait for troubles.
1077 */
1078 __sk_prot_rehash(sk);
1079 return 0;
1080}
1081
1082int inet_sk_rebuild_header(struct sock *sk)
1083{
1084 struct inet_sock *inet = inet_sk(sk);
1085 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
Al Viro3ca3c682006-09-27 18:28:07 -07001086 __be32 daddr;
Arnaldo Carvalho de Melo32519f12005-08-09 19:50:02 -07001087 int err;
1088
1089 /* Route is OK, nothing to do. */
1090 if (rt)
1091 return 0;
1092
1093 /* Reroute. */
1094 daddr = inet->daddr;
1095 if (inet->opt && inet->opt->srr)
1096 daddr = inet->opt->faddr;
1097{
1098 struct flowi fl = {
1099 .oif = sk->sk_bound_dev_if,
1100 .nl_u = {
1101 .ip4_u = {
1102 .daddr = daddr,
1103 .saddr = inet->saddr,
1104 .tos = RT_CONN_FLAGS(sk),
1105 },
1106 },
1107 .proto = sk->sk_protocol,
1108 .uli_u = {
1109 .ports = {
1110 .sport = inet->sport,
1111 .dport = inet->dport,
1112 },
1113 },
1114 };
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001115
Venkat Yekkiralabeb8d132006-08-04 23:12:42 -07001116 security_sk_classify_flow(sk, &fl);
Denis V. Lunevf1b050b2008-01-22 22:07:10 -08001117 err = ip_route_output_flow(&init_net, &rt, &fl, sk, 0);
Arnaldo Carvalho de Melo32519f12005-08-09 19:50:02 -07001118}
1119 if (!err)
1120 sk_setup_caps(sk, &rt->u.dst);
1121 else {
1122 /* Routing failed... */
1123 sk->sk_route_caps = 0;
1124 /*
1125 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1126 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1127 */
1128 if (!sysctl_ip_dynaddr ||
1129 sk->sk_state != TCP_SYN_SENT ||
1130 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1131 (err = inet_sk_reselect_saddr(sk)) != 0)
1132 sk->sk_err_soft = -err;
1133 }
1134
1135 return err;
1136}
1137
1138EXPORT_SYMBOL(inet_sk_rebuild_header);
1139
Herbert Xua430a432006-07-08 13:34:56 -07001140static int inet_gso_send_check(struct sk_buff *skb)
1141{
1142 struct iphdr *iph;
1143 struct net_protocol *ops;
1144 int proto;
1145 int ihl;
1146 int err = -EINVAL;
1147
1148 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1149 goto out;
1150
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001151 iph = ip_hdr(skb);
Herbert Xua430a432006-07-08 13:34:56 -07001152 ihl = iph->ihl * 4;
1153 if (ihl < sizeof(*iph))
1154 goto out;
1155
1156 if (unlikely(!pskb_may_pull(skb, ihl)))
1157 goto out;
1158
Arnaldo Carvalho de Melobadff6d2007-03-13 13:06:52 -03001159 __skb_pull(skb, ihl);
1160 skb_reset_transport_header(skb);
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001161 iph = ip_hdr(skb);
Herbert Xua430a432006-07-08 13:34:56 -07001162 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1163 err = -EPROTONOSUPPORT;
1164
1165 rcu_read_lock();
1166 ops = rcu_dereference(inet_protos[proto]);
1167 if (likely(ops && ops->gso_send_check))
1168 err = ops->gso_send_check(skb);
1169 rcu_read_unlock();
1170
1171out:
1172 return err;
1173}
1174
Herbert Xu576a30e2006-06-27 13:22:38 -07001175static struct sk_buff *inet_gso_segment(struct sk_buff *skb, int features)
Herbert Xuf4c50d92006-06-22 03:02:40 -07001176{
1177 struct sk_buff *segs = ERR_PTR(-EINVAL);
1178 struct iphdr *iph;
1179 struct net_protocol *ops;
1180 int proto;
1181 int ihl;
1182 int id;
1183
Stephen Hemmingerd212f872007-06-27 00:47:37 -07001184 if (!(features & NETIF_F_V4_CSUM))
1185 features &= ~NETIF_F_SG;
1186
Herbert Xubbcf4672006-07-03 19:38:35 -07001187 if (unlikely(skb_shinfo(skb)->gso_type &
1188 ~(SKB_GSO_TCPV4 |
1189 SKB_GSO_UDP |
1190 SKB_GSO_DODGY |
1191 SKB_GSO_TCP_ECN |
1192 0)))
1193 goto out;
1194
1195 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
Herbert Xuf4c50d92006-06-22 03:02:40 -07001196 goto out;
1197
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001198 iph = ip_hdr(skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07001199 ihl = iph->ihl * 4;
1200 if (ihl < sizeof(*iph))
1201 goto out;
1202
Herbert Xubbcf4672006-07-03 19:38:35 -07001203 if (unlikely(!pskb_may_pull(skb, ihl)))
Herbert Xuf4c50d92006-06-22 03:02:40 -07001204 goto out;
1205
Arnaldo Carvalho de Melobadff6d2007-03-13 13:06:52 -03001206 __skb_pull(skb, ihl);
1207 skb_reset_transport_header(skb);
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001208 iph = ip_hdr(skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07001209 id = ntohs(iph->id);
1210 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1211 segs = ERR_PTR(-EPROTONOSUPPORT);
1212
1213 rcu_read_lock();
1214 ops = rcu_dereference(inet_protos[proto]);
Herbert Xubbcf4672006-07-03 19:38:35 -07001215 if (likely(ops && ops->gso_segment))
Herbert Xu576a30e2006-06-27 13:22:38 -07001216 segs = ops->gso_segment(skb, features);
Herbert Xuf4c50d92006-06-22 03:02:40 -07001217 rcu_read_unlock();
1218
Herbert Xu576a30e2006-06-27 13:22:38 -07001219 if (!segs || unlikely(IS_ERR(segs)))
Herbert Xuf4c50d92006-06-22 03:02:40 -07001220 goto out;
1221
1222 skb = segs;
1223 do {
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001224 iph = ip_hdr(skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07001225 iph->id = htons(id++);
1226 iph->tot_len = htons(skb->len - skb->mac_len);
1227 iph->check = 0;
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -07001228 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
Herbert Xuf4c50d92006-06-22 03:02:40 -07001229 } while ((skb = skb->next));
1230
1231out:
1232 return segs;
1233}
1234
Herbert Xu5e0f0432007-04-24 21:53:35 -07001235unsigned long snmp_fold_field(void *mib[], int offt)
1236{
1237 unsigned long res = 0;
1238 int i;
1239
1240 for_each_possible_cpu(i) {
1241 res += *(((unsigned long *) per_cpu_ptr(mib[0], i)) + offt);
1242 res += *(((unsigned long *) per_cpu_ptr(mib[1], i)) + offt);
1243 }
1244 return res;
1245}
1246EXPORT_SYMBOL_GPL(snmp_fold_field);
1247
YOSHIFUJI Hideakic69bce22008-01-23 22:31:45 -08001248int snmp_mib_init(void *ptr[2], size_t mibsize)
Herbert Xu5e0f0432007-04-24 21:53:35 -07001249{
1250 BUG_ON(ptr == NULL);
1251 ptr[0] = __alloc_percpu(mibsize);
1252 if (!ptr[0])
1253 goto err0;
1254 ptr[1] = __alloc_percpu(mibsize);
1255 if (!ptr[1])
1256 goto err1;
1257 return 0;
1258err1:
1259 free_percpu(ptr[0]);
1260 ptr[0] = NULL;
1261err0:
1262 return -ENOMEM;
1263}
1264EXPORT_SYMBOL_GPL(snmp_mib_init);
1265
1266void snmp_mib_free(void *ptr[2])
1267{
1268 BUG_ON(ptr == NULL);
1269 free_percpu(ptr[0]);
1270 free_percpu(ptr[1]);
1271 ptr[0] = ptr[1] = NULL;
1272}
1273EXPORT_SYMBOL_GPL(snmp_mib_free);
1274
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275#ifdef CONFIG_IP_MULTICAST
1276static struct net_protocol igmp_protocol = {
1277 .handler = igmp_rcv,
1278};
1279#endif
1280
1281static struct net_protocol tcp_protocol = {
1282 .handler = tcp_v4_rcv,
1283 .err_handler = tcp_v4_err,
Herbert Xua430a432006-07-08 13:34:56 -07001284 .gso_send_check = tcp_v4_gso_send_check,
Herbert Xuf4c50d92006-06-22 03:02:40 -07001285 .gso_segment = tcp_tso_segment,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 .no_policy = 1,
1287};
1288
1289static struct net_protocol udp_protocol = {
1290 .handler = udp_rcv,
1291 .err_handler = udp_err,
1292 .no_policy = 1,
1293};
1294
1295static struct net_protocol icmp_protocol = {
1296 .handler = icmp_rcv,
Herbert Xu8b7817f2007-12-12 10:44:43 -08001297 .no_policy = 1,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298};
1299
1300static int __init init_ipv4_mibs(void)
1301{
YOSHIFUJI Hideaki33490172007-04-20 15:57:15 -07001302 if (snmp_mib_init((void **)net_statistics,
YOSHIFUJI Hideakic69bce22008-01-23 22:31:45 -08001303 sizeof(struct linux_mib)) < 0)
YOSHIFUJI Hideaki33490172007-04-20 15:57:15 -07001304 goto err_net_mib;
1305 if (snmp_mib_init((void **)ip_statistics,
YOSHIFUJI Hideakic69bce22008-01-23 22:31:45 -08001306 sizeof(struct ipstats_mib)) < 0)
YOSHIFUJI Hideaki33490172007-04-20 15:57:15 -07001307 goto err_ip_mib;
1308 if (snmp_mib_init((void **)icmp_statistics,
YOSHIFUJI Hideakic69bce22008-01-23 22:31:45 -08001309 sizeof(struct icmp_mib)) < 0)
YOSHIFUJI Hideaki33490172007-04-20 15:57:15 -07001310 goto err_icmp_mib;
David L Stevens96793b42007-09-17 09:57:33 -07001311 if (snmp_mib_init((void **)icmpmsg_statistics,
YOSHIFUJI Hideakic69bce22008-01-23 22:31:45 -08001312 sizeof(struct icmpmsg_mib)) < 0)
David L Stevens96793b42007-09-17 09:57:33 -07001313 goto err_icmpmsg_mib;
YOSHIFUJI Hideaki33490172007-04-20 15:57:15 -07001314 if (snmp_mib_init((void **)tcp_statistics,
YOSHIFUJI Hideakic69bce22008-01-23 22:31:45 -08001315 sizeof(struct tcp_mib)) < 0)
YOSHIFUJI Hideaki33490172007-04-20 15:57:15 -07001316 goto err_tcp_mib;
1317 if (snmp_mib_init((void **)udp_statistics,
YOSHIFUJI Hideakic69bce22008-01-23 22:31:45 -08001318 sizeof(struct udp_mib)) < 0)
YOSHIFUJI Hideaki33490172007-04-20 15:57:15 -07001319 goto err_udp_mib;
1320 if (snmp_mib_init((void **)udplite_statistics,
YOSHIFUJI Hideakic69bce22008-01-23 22:31:45 -08001321 sizeof(struct udp_mib)) < 0)
YOSHIFUJI Hideaki33490172007-04-20 15:57:15 -07001322 goto err_udplite_mib;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323
YOSHIFUJI Hideaki33490172007-04-20 15:57:15 -07001324 tcp_mib_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325
1326 return 0;
YOSHIFUJI Hideaki33490172007-04-20 15:57:15 -07001327
1328err_udplite_mib:
1329 snmp_mib_free((void **)udp_statistics);
1330err_udp_mib:
1331 snmp_mib_free((void **)tcp_statistics);
1332err_tcp_mib:
David L Stevens96793b42007-09-17 09:57:33 -07001333 snmp_mib_free((void **)icmpmsg_statistics);
1334err_icmpmsg_mib:
YOSHIFUJI Hideaki33490172007-04-20 15:57:15 -07001335 snmp_mib_free((void **)icmp_statistics);
1336err_icmp_mib:
1337 snmp_mib_free((void **)ip_statistics);
1338err_ip_mib:
1339 snmp_mib_free((void **)net_statistics);
1340err_net_mib:
1341 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342}
1343
1344static int ipv4_proc_init(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345
Herbert Xu30e224d2005-07-05 14:40:10 -07001346/*
1347 * IP protocol layer initialiser
1348 */
1349
1350static struct packet_type ip_packet_type = {
1351 .type = __constant_htons(ETH_P_IP),
1352 .func = ip_rcv,
Herbert Xua430a432006-07-08 13:34:56 -07001353 .gso_send_check = inet_gso_send_check,
Herbert Xuf4c50d92006-06-22 03:02:40 -07001354 .gso_segment = inet_gso_segment,
Herbert Xu30e224d2005-07-05 14:40:10 -07001355};
1356
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357static int __init inet_init(void)
1358{
1359 struct sk_buff *dummy_skb;
1360 struct inet_protosw *q;
1361 struct list_head *r;
1362 int rc = -EINVAL;
1363
YOSHIFUJI Hideakief047f52006-09-01 00:29:06 -07001364 BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365
1366 rc = proto_register(&tcp_prot, 1);
1367 if (rc)
1368 goto out;
1369
1370 rc = proto_register(&udp_prot, 1);
1371 if (rc)
1372 goto out_unregister_tcp_proto;
1373
1374 rc = proto_register(&raw_prot, 1);
1375 if (rc)
1376 goto out_unregister_udp_proto;
1377
1378 /*
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001379 * Tell SOCKET that we are alive...
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 */
1381
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001382 (void)sock_register(&inet_family_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383
1384 /*
1385 * Add all the base protocols.
1386 */
1387
1388 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1389 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1390 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1391 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1392 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1393 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1394#ifdef CONFIG_IP_MULTICAST
1395 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1396 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1397#endif
1398
1399 /* Register the socket-side information for inet_create. */
1400 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1401 INIT_LIST_HEAD(r);
1402
1403 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1404 inet_register_protosw(q);
1405
1406 /*
1407 * Set the ARP module up
1408 */
1409
1410 arp_init();
1411
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001412 /*
1413 * Set the IP module up
1414 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415
1416 ip_init();
1417
Denis V. Lunev9b0f9762008-02-29 11:13:15 -08001418 tcp_v4_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419
1420 /* Setup TCP slab cache for open requests. */
1421 tcp_init();
1422
Hideo Aoki95766ff2007-12-31 00:29:24 -08001423 /* Setup UDP memory threshold */
1424 udp_init();
1425
Gerrit Renkerba4e58e2006-11-27 11:10:57 -08001426 /* Add UDP-Lite (RFC 3828) */
1427 udplite4_register();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428
1429 /*
1430 * Set the ICMP layer up
1431 */
1432
Denis V. Luneva5710d62008-02-29 11:14:50 -08001433 if (icmp_init() < 0)
1434 panic("Failed to create the ICMP control socket.\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435
1436 /*
1437 * Initialise the multicast router
1438 */
1439#if defined(CONFIG_IP_MROUTE)
1440 ip_mr_init();
1441#endif
1442 /*
1443 * Initialise per-cpu ipv4 mibs
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001444 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445
Stephen Hemminger132adf52007-03-08 20:44:43 -08001446 if (init_ipv4_mibs())
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n"); ;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001448
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 ipv4_proc_init();
1450
1451 ipfrag_init();
1452
Herbert Xu30e224d2005-07-05 14:40:10 -07001453 dev_add_pack(&ip_packet_type);
1454
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 rc = 0;
1456out:
1457 return rc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458out_unregister_udp_proto:
1459 proto_unregister(&udp_prot);
Olaf Kirch321efff2006-09-27 16:33:45 -07001460out_unregister_tcp_proto:
1461 proto_unregister(&tcp_prot);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 goto out;
1463}
1464
Heiko Carstensa536e072006-04-28 15:19:17 -07001465fs_initcall(inet_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466
1467/* ------------------------------------------------------------------------ */
1468
1469#ifdef CONFIG_PROC_FS
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470static int __init ipv4_proc_init(void)
1471{
1472 int rc = 0;
1473
1474 if (raw_proc_init())
1475 goto out_raw;
1476 if (tcp4_proc_init())
1477 goto out_tcp;
1478 if (udp4_proc_init())
1479 goto out_udp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 if (ip_misc_proc_init())
1481 goto out_misc;
1482out:
1483 return rc;
1484out_misc:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 udp4_proc_exit();
1486out_udp:
1487 tcp4_proc_exit();
1488out_tcp:
1489 raw_proc_exit();
1490out_raw:
1491 rc = -ENOMEM;
1492 goto out;
1493}
1494
1495#else /* CONFIG_PROC_FS */
1496static int __init ipv4_proc_init(void)
1497{
1498 return 0;
1499}
1500#endif /* CONFIG_PROC_FS */
1501
1502MODULE_ALIAS_NETPROTO(PF_INET);
1503
1504EXPORT_SYMBOL(inet_accept);
1505EXPORT_SYMBOL(inet_bind);
1506EXPORT_SYMBOL(inet_dgram_connect);
1507EXPORT_SYMBOL(inet_dgram_ops);
1508EXPORT_SYMBOL(inet_getname);
1509EXPORT_SYMBOL(inet_ioctl);
1510EXPORT_SYMBOL(inet_listen);
1511EXPORT_SYMBOL(inet_register_protosw);
1512EXPORT_SYMBOL(inet_release);
1513EXPORT_SYMBOL(inet_sendmsg);
1514EXPORT_SYMBOL(inet_shutdown);
1515EXPORT_SYMBOL(inet_sock_destruct);
1516EXPORT_SYMBOL(inet_stream_connect);
1517EXPORT_SYMBOL(inet_stream_ops);
1518EXPORT_SYMBOL(inet_unregister_protosw);
1519EXPORT_SYMBOL(net_statistics);
Neil Hormancdac4e02005-06-13 15:12:33 -07001520EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);