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.
6 * PF_INET protocol family socket handler.
8 * Version: $Id: af_inet.c,v 1.137 2002/02/01 22:01:03 davem Exp $
11 * 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>
15 * Changes (see also sock.c)
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
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
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
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
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.
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.
70 #include <linux/err.h>
71 #include <linux/errno.h>
72 #include <linux/types.h>
73 #include <linux/socket.h>
75 #include <linux/kernel.h>
76 #include <linux/module.h>
77 #include <linux/sched.h>
78 #include <linux/timer.h>
79 #include <linux/string.h>
80 #include <linux/sockios.h>
81 #include <linux/net.h>
82 #include <linux/capability.h>
83 #include <linux/fcntl.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>
90 #include <linux/random.h>
92 #include <asm/uaccess.h>
93 #include <asm/system.h>
95 #include <linux/inet.h>
96 #include <linux/igmp.h>
97 #include <linux/inetdevice.h>
98 #include <linux/netdevice.h>
100 #include <net/protocol.h>
102 #include <net/route.h>
103 #include <net/ip_fib.h>
104 #include <net/inet_connection_sock.h>
107 #include <net/udplite.h>
108 #include <linux/skbuff.h>
109 #include <net/sock.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>
119 DEFINE_SNMP_STAT(struct linux_mib
, net_statistics
) __read_mostly
;
121 extern void ip_mc_drop_socket(struct sock
*sk
);
123 /* The inetsw table contains everything that inet_create needs to
124 * build a new socket.
126 static struct list_head inetsw
[SOCK_MAX
];
127 static DEFINE_SPINLOCK(inetsw_lock
);
129 struct ipv4_config ipv4_config
;
131 EXPORT_SYMBOL(ipv4_config
);
133 /* New destruction routine */
135 void inet_sock_destruct(struct sock
*sk
)
137 struct inet_sock
*inet
= inet_sk(sk
);
139 __skb_queue_purge(&sk
->sk_receive_queue
);
140 __skb_queue_purge(&sk
->sk_error_queue
);
144 if (sk
->sk_type
== SOCK_STREAM
&& sk
->sk_state
!= TCP_CLOSE
) {
145 printk("Attempt to release TCP socket in state %d %p\n",
149 if (!sock_flag(sk
, SOCK_DEAD
)) {
150 printk("Attempt to release alive inet socket %p\n", sk
);
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
);
160 dst_release(sk
->sk_dst_cache
);
161 sk_refcnt_debug_dec(sk
);
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
171 * Automatically bind an unbound socket.
174 static int inet_autobind(struct sock
*sk
)
176 struct inet_sock
*inet
;
177 /* We may need to bind the socket. */
181 if (sk
->sk_prot
->get_port(sk
, 0)) {
185 inet
->sport
= htons(inet
->num
);
192 * Move a socket into listening state.
194 int inet_listen(struct socket
*sock
, int backlog
)
196 struct sock
*sk
= sock
->sk
;
197 unsigned char old_state
;
203 if (sock
->state
!= SS_UNCONNECTED
|| sock
->type
!= SOCK_STREAM
)
206 old_state
= sk
->sk_state
;
207 if (!((1 << old_state
) & (TCPF_CLOSE
| TCPF_LISTEN
)))
210 /* Really, if the socket is already in listen state
211 * we can only allow the backlog to be adjusted.
213 if (old_state
!= TCP_LISTEN
) {
214 err
= inet_csk_listen_start(sk
, backlog
);
218 sk
->sk_max_ack_backlog
= backlog
;
226 u32 inet_ehash_secret __read_mostly
;
227 EXPORT_SYMBOL(inet_ehash_secret
);
230 * inet_ehash_secret must be set exactly once
231 * Instead of using a dedicated spinlock, we (ab)use inetsw_lock
233 void build_ehash_secret(void)
237 get_random_bytes(&rnd
, sizeof(rnd
));
239 spin_lock_bh(&inetsw_lock
);
240 if (!inet_ehash_secret
)
241 inet_ehash_secret
= rnd
;
242 spin_unlock_bh(&inetsw_lock
);
244 EXPORT_SYMBOL(build_ehash_secret
);
247 * Create an inet socket.
250 static int inet_create(struct net
*net
, struct socket
*sock
, int protocol
)
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
;
259 int try_loading_module
= 0;
262 if (net
!= &init_net
)
263 return -EAFNOSUPPORT
;
265 if (sock
->type
!= SOCK_RAW
&&
266 sock
->type
!= SOCK_DGRAM
&&
268 build_ehash_secret();
270 sock
->state
= SS_UNCONNECTED
;
272 /* Look for the requested type/protocol pair. */
275 err
= -ESOCKTNOSUPPORT
;
277 list_for_each_rcu(p
, &inetsw
[sock
->type
]) {
278 answer
= list_entry(p
, struct inet_protosw
, list
);
280 /* Check the non-wild match. */
281 if (protocol
== answer
->protocol
) {
282 if (protocol
!= IPPROTO_IP
)
285 /* Check for the two wild cases. */
286 if (IPPROTO_IP
== protocol
) {
287 protocol
= answer
->protocol
;
290 if (IPPROTO_IP
== answer
->protocol
)
293 err
= -EPROTONOSUPPORT
;
297 if (unlikely(answer
== NULL
)) {
298 if (try_loading_module
< 2) {
301 * Be more specific, e.g. net-pf-2-proto-132-type-1
302 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
304 if (++try_loading_module
== 1)
305 request_module("net-pf-%d-proto-%d-type-%d",
306 PF_INET
, protocol
, sock
->type
);
308 * Fall back to generic, e.g. net-pf-2-proto-132
309 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
312 request_module("net-pf-%d-proto-%d",
314 goto lookup_protocol
;
320 if (answer
->capability
> 0 && !capable(answer
->capability
))
323 sock
->ops
= answer
->ops
;
324 answer_prot
= answer
->prot
;
325 answer_no_check
= answer
->no_check
;
326 answer_flags
= answer
->flags
;
329 BUG_TRAP(answer_prot
->slab
!= NULL
);
332 sk
= sk_alloc(net
, PF_INET
, GFP_KERNEL
, answer_prot
);
337 sk
->sk_no_check
= answer_no_check
;
338 if (INET_PROTOSW_REUSE
& answer_flags
)
342 inet
->is_icsk
= (INET_PROTOSW_ICSK
& answer_flags
) != 0;
344 if (SOCK_RAW
== sock
->type
) {
345 inet
->num
= protocol
;
346 if (IPPROTO_RAW
== protocol
)
350 if (ipv4_config
.no_pmtu_disc
)
351 inet
->pmtudisc
= IP_PMTUDISC_DONT
;
353 inet
->pmtudisc
= IP_PMTUDISC_WANT
;
357 sock_init_data(sock
, sk
);
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
;
368 inet
->mc_list
= NULL
;
370 sk_refcnt_debug_inc(sk
);
373 /* It assumes that any protocol which allows
374 * the user to assign a number at socket
375 * creation time automatically
378 inet
->sport
= htons(inet
->num
);
379 /* Add to protocol hash chains. */
380 sk
->sk_prot
->hash(sk
);
383 if (sk
->sk_prot
->init
) {
384 err
= sk
->sk_prot
->init(sk
);
386 sk_common_release(sk
);
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.
401 int inet_release(struct socket
*sock
)
403 struct sock
*sk
= sock
->sk
;
408 /* Applications forget to leave groups before exiting */
409 ip_mc_drop_socket(sk
);
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.
415 * If the close is due to the process exiting, we never
419 if (sock_flag(sk
, SOCK_LINGER
) &&
420 !(current
->flags
& PF_EXITING
))
421 timeout
= sk
->sk_lingertime
;
423 sk
->sk_prot
->close(sk
, timeout
);
428 /* It is off by default, see below. */
429 int sysctl_ip_nonlocal_bind __read_mostly
;
431 int inet_bind(struct socket
*sock
, struct sockaddr
*uaddr
, int addr_len
)
433 struct sockaddr_in
*addr
= (struct sockaddr_in
*)uaddr
;
434 struct sock
*sk
= sock
->sk
;
435 struct inet_sock
*inet
= inet_sk(sk
);
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
);
446 if (addr_len
< sizeof(struct sockaddr_in
))
449 chk_addr_ret
= inet_addr_type(&init_net
, addr
->sin_addr
.s_addr
);
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)
458 err
= -EADDRNOTAVAIL
;
459 if (!sysctl_ip_nonlocal_bind
&&
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
)
467 snum
= ntohs(addr
->sin_port
);
469 if (snum
&& snum
< PROT_SOCK
&& !capable(CAP_NET_BIND_SERVICE
))
472 /* We keep a pair of addresses. rcv_saddr is the one
473 * used by hash lookups, and saddr is used for transmit.
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.
481 /* Check these errors (active socket, double bind). */
483 if (sk
->sk_state
!= TCP_CLOSE
|| inet
->num
)
484 goto out_release_sock
;
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 */
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;
494 goto out_release_sock
;
498 sk
->sk_userlocks
|= SOCK_BINDADDR_LOCK
;
500 sk
->sk_userlocks
|= SOCK_BINDPORT_LOCK
;
501 inet
->sport
= htons(inet
->num
);
512 int inet_dgram_connect(struct socket
*sock
, struct sockaddr
* uaddr
,
513 int addr_len
, int flags
)
515 struct sock
*sk
= sock
->sk
;
517 if (uaddr
->sa_family
== AF_UNSPEC
)
518 return sk
->sk_prot
->disconnect(sk
, flags
);
520 if (!inet_sk(sk
)->num
&& inet_autobind(sk
))
522 return sk
->sk_prot
->connect(sk
, (struct sockaddr
*)uaddr
, addr_len
);
525 static long inet_wait_for_connect(struct sock
*sk
, long timeo
)
529 prepare_to_wait(sk
->sk_sleep
, &wait
, TASK_INTERRUPTIBLE
);
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.
536 while ((1 << sk
->sk_state
) & (TCPF_SYN_SENT
| TCPF_SYN_RECV
)) {
538 timeo
= schedule_timeout(timeo
);
540 if (signal_pending(current
) || !timeo
)
542 prepare_to_wait(sk
->sk_sleep
, &wait
, TASK_INTERRUPTIBLE
);
544 finish_wait(sk
->sk_sleep
, &wait
);
549 * Connect to a remote host. There is regrettably still a little
550 * TCP 'magic' in here.
552 int inet_stream_connect(struct socket
*sock
, struct sockaddr
*uaddr
,
553 int addr_len
, int flags
)
555 struct sock
*sk
= sock
->sk
;
561 if (uaddr
->sa_family
== AF_UNSPEC
) {
562 err
= sk
->sk_prot
->disconnect(sk
, flags
);
563 sock
->state
= err
? SS_DISCONNECTING
: SS_UNCONNECTED
;
567 switch (sock
->state
) {
576 /* Fall out of switch with err, set for this state */
580 if (sk
->sk_state
!= TCP_CLOSE
)
583 err
= sk
->sk_prot
->connect(sk
, uaddr
, addr_len
);
587 sock
->state
= SS_CONNECTING
;
589 /* Just entered SS_CONNECTING state; the only
590 * difference is that return value in non-blocking
591 * case is EINPROGRESS, rather than EALREADY.
597 timeo
= sock_sndtimeo(sk
, flags
& O_NONBLOCK
);
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
))
604 err
= sock_intr_errno(timeo
);
605 if (signal_pending(current
))
609 /* Connection was closed by RST, timeout, ICMP error
610 * or another process disconnected us.
612 if (sk
->sk_state
== TCP_CLOSE
)
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.
620 sock
->state
= SS_CONNECTED
;
627 err
= sock_error(sk
) ? : -ECONNABORTED
;
628 sock
->state
= SS_UNCONNECTED
;
629 if (sk
->sk_prot
->disconnect(sk
, flags
))
630 sock
->state
= SS_DISCONNECTING
;
635 * Accept a pending connection. The TCP layer now gives BSD semantics.
638 int inet_accept(struct socket
*sock
, struct socket
*newsock
, int flags
)
640 struct sock
*sk1
= sock
->sk
;
642 struct sock
*sk2
= sk1
->sk_prot
->accept(sk1
, flags
, &err
);
649 BUG_TRAP((1 << sk2
->sk_state
) &
650 (TCPF_ESTABLISHED
| TCPF_CLOSE_WAIT
| TCPF_CLOSE
));
652 sock_graft(sk2
, newsock
);
654 newsock
->state
= SS_CONNECTED
;
663 * This does both peername and sockname.
665 int inet_getname(struct socket
*sock
, struct sockaddr
*uaddr
,
666 int *uaddr_len
, int peer
)
668 struct sock
*sk
= sock
->sk
;
669 struct inet_sock
*inet
= inet_sk(sk
);
670 struct sockaddr_in
*sin
= (struct sockaddr_in
*)uaddr
;
672 sin
->sin_family
= AF_INET
;
675 (((1 << sk
->sk_state
) & (TCPF_CLOSE
| TCPF_SYN_SENT
)) &&
678 sin
->sin_port
= inet
->dport
;
679 sin
->sin_addr
.s_addr
= inet
->daddr
;
681 __be32 addr
= inet
->rcv_saddr
;
684 sin
->sin_port
= inet
->sport
;
685 sin
->sin_addr
.s_addr
= addr
;
687 memset(sin
->sin_zero
, 0, sizeof(sin
->sin_zero
));
688 *uaddr_len
= sizeof(*sin
);
692 int inet_sendmsg(struct kiocb
*iocb
, struct socket
*sock
, struct msghdr
*msg
,
695 struct sock
*sk
= sock
->sk
;
697 /* We may need to bind the socket. */
698 if (!inet_sk(sk
)->num
&& inet_autobind(sk
))
701 return sk
->sk_prot
->sendmsg(iocb
, sk
, msg
, size
);
705 static ssize_t
inet_sendpage(struct socket
*sock
, struct page
*page
, int offset
, size_t size
, int flags
)
707 struct sock
*sk
= sock
->sk
;
709 /* We may need to bind the socket. */
710 if (!inet_sk(sk
)->num
&& inet_autobind(sk
))
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
);
719 int inet_shutdown(struct socket
*sock
, int how
)
721 struct sock
*sk
= sock
->sk
;
724 /* This should really check to make sure
725 * the socket is a TCP socket. (WHY AC...)
727 how
++; /* maps 0->1 has the advantage of making bit 1 rcvs and
730 if ((how
& ~SHUTDOWN_MASK
) || !how
) /* MAXINT->0 */
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
;
739 sock
->state
= SS_CONNECTED
;
742 switch (sk
->sk_state
) {
745 /* Hack to wake up other listeners, who can poll for
746 POLLHUP, even on eg. unconnected UDP sockets -- RR */
748 sk
->sk_shutdown
|= how
;
749 if (sk
->sk_prot
->shutdown
)
750 sk
->sk_prot
->shutdown(sk
, how
);
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.
758 if (!(how
& RCV_SHUTDOWN
))
762 err
= sk
->sk_prot
->disconnect(sk
, O_NONBLOCK
);
763 sock
->state
= err
? SS_DISCONNECTING
: SS_UNCONNECTED
;
767 /* Wake up anyone sleeping in poll. */
768 sk
->sk_state_change(sk
);
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.
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.
783 int inet_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
785 struct sock
*sk
= sock
->sk
;
787 struct net
*net
= sk
->sk_net
;
791 err
= sock_get_timestamp(sk
, (struct timeval __user
*)arg
);
794 err
= sock_get_timestampns(sk
, (struct timespec __user
*)arg
);
799 err
= ip_rt_ioctl(net
, cmd
, (void __user
*)arg
);
804 err
= arp_ioctl(net
, cmd
, (void __user
*)arg
);
817 err
= devinet_ioctl(net
, cmd
, (void __user
*)arg
);
820 if (sk
->sk_prot
->ioctl
)
821 err
= sk
->sk_prot
->ioctl(sk
, cmd
, arg
);
829 const struct proto_ops inet_stream_ops
= {
831 .owner
= THIS_MODULE
,
832 .release
= inet_release
,
834 .connect
= inet_stream_connect
,
835 .socketpair
= sock_no_socketpair
,
836 .accept
= inet_accept
,
837 .getname
= inet_getname
,
840 .listen
= inet_listen
,
841 .shutdown
= inet_shutdown
,
842 .setsockopt
= sock_common_setsockopt
,
843 .getsockopt
= sock_common_getsockopt
,
844 .sendmsg
= tcp_sendmsg
,
845 .recvmsg
= sock_common_recvmsg
,
846 .mmap
= sock_no_mmap
,
847 .sendpage
= tcp_sendpage
,
848 .splice_read
= tcp_splice_read
,
850 .compat_setsockopt
= compat_sock_common_setsockopt
,
851 .compat_getsockopt
= compat_sock_common_getsockopt
,
855 const struct proto_ops inet_dgram_ops
= {
857 .owner
= THIS_MODULE
,
858 .release
= inet_release
,
860 .connect
= inet_dgram_connect
,
861 .socketpair
= sock_no_socketpair
,
862 .accept
= sock_no_accept
,
863 .getname
= inet_getname
,
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
,
875 .compat_setsockopt
= compat_sock_common_setsockopt
,
876 .compat_getsockopt
= compat_sock_common_getsockopt
,
881 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
884 static const struct proto_ops inet_sockraw_ops
= {
886 .owner
= THIS_MODULE
,
887 .release
= inet_release
,
889 .connect
= inet_dgram_connect
,
890 .socketpair
= sock_no_socketpair
,
891 .accept
= sock_no_accept
,
892 .getname
= inet_getname
,
893 .poll
= datagram_poll
,
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
,
904 .compat_setsockopt
= compat_sock_common_setsockopt
,
905 .compat_getsockopt
= compat_sock_common_getsockopt
,
909 static struct net_proto_family inet_family_ops
= {
911 .create
= inet_create
,
912 .owner
= THIS_MODULE
,
915 /* Upon startup we insert all the elements in inetsw_array[] into
916 * the linked list inetsw.
918 static struct inet_protosw inetsw_array
[] =
922 .protocol
= IPPROTO_TCP
,
924 .ops
= &inet_stream_ops
,
927 .flags
= INET_PROTOSW_PERMANENT
|
933 .protocol
= IPPROTO_UDP
,
935 .ops
= &inet_dgram_ops
,
937 .no_check
= UDP_CSUM_DEFAULT
,
938 .flags
= INET_PROTOSW_PERMANENT
,
944 .protocol
= IPPROTO_IP
, /* wild card */
946 .ops
= &inet_sockraw_ops
,
947 .capability
= CAP_NET_RAW
,
948 .no_check
= UDP_CSUM_DEFAULT
,
949 .flags
= INET_PROTOSW_REUSE
,
953 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
955 void inet_register_protosw(struct inet_protosw
*p
)
957 struct list_head
*lh
;
958 struct inet_protosw
*answer
;
959 int protocol
= p
->protocol
;
960 struct list_head
*last_perm
;
962 spin_lock_bh(&inetsw_lock
);
964 if (p
->type
>= SOCK_MAX
)
967 /* If we are trying to override a permanent protocol, bail. */
969 last_perm
= &inetsw
[p
->type
];
970 list_for_each(lh
, &inetsw
[p
->type
]) {
971 answer
= list_entry(lh
, struct inet_protosw
, list
);
973 /* Check only the non-wild match. */
974 if (INET_PROTOSW_PERMANENT
& answer
->flags
) {
975 if (protocol
== answer
->protocol
)
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
988 * non-permanent entry. This means that when we remove this entry, the
989 * system automatically returns to the old behavior.
991 list_add_rcu(&p
->list
, last_perm
);
993 spin_unlock_bh(&inetsw_lock
);
1000 printk(KERN_ERR
"Attempt to override permanent protocol %d.\n",
1006 "Ignoring attempt to register invalid socket type %d.\n",
1011 void inet_unregister_protosw(struct inet_protosw
*p
)
1013 if (INET_PROTOSW_PERMANENT
& p
->flags
) {
1015 "Attempt to unregister permanent protocol %d.\n",
1018 spin_lock_bh(&inetsw_lock
);
1019 list_del_rcu(&p
->list
);
1020 spin_unlock_bh(&inetsw_lock
);
1027 * Shall we try to damage output packets if routing dev changes?
1030 int sysctl_ip_dynaddr __read_mostly
;
1032 static int inet_sk_reselect_saddr(struct sock
*sk
)
1034 struct inet_sock
*inet
= inet_sk(sk
);
1037 __be32 old_saddr
= inet
->saddr
;
1039 __be32 daddr
= inet
->daddr
;
1041 if (inet
->opt
&& inet
->opt
->srr
)
1042 daddr
= inet
->opt
->faddr
;
1044 /* Query new route. */
1045 err
= ip_route_connect(&rt
, daddr
, 0,
1047 sk
->sk_bound_dev_if
,
1049 inet
->sport
, inet
->dport
, sk
, 0);
1053 sk_setup_caps(sk
, &rt
->u
.dst
);
1055 new_saddr
= rt
->rt_src
;
1057 if (new_saddr
== old_saddr
)
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",
1065 NIPQUAD(new_saddr
));
1068 inet
->saddr
= inet
->rcv_saddr
= new_saddr
;
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
1075 * Besides that, it does not check for connection
1076 * uniqueness. Wait for troubles.
1078 __sk_prot_rehash(sk
);
1082 int inet_sk_rebuild_header(struct sock
*sk
)
1084 struct inet_sock
*inet
= inet_sk(sk
);
1085 struct rtable
*rt
= (struct rtable
*)__sk_dst_check(sk
, 0);
1089 /* Route is OK, nothing to do. */
1094 daddr
= inet
->daddr
;
1095 if (inet
->opt
&& inet
->opt
->srr
)
1096 daddr
= inet
->opt
->faddr
;
1099 .oif
= sk
->sk_bound_dev_if
,
1103 .saddr
= inet
->saddr
,
1104 .tos
= RT_CONN_FLAGS(sk
),
1107 .proto
= sk
->sk_protocol
,
1110 .sport
= inet
->sport
,
1111 .dport
= inet
->dport
,
1116 security_sk_classify_flow(sk
, &fl
);
1117 err
= ip_route_output_flow(&init_net
, &rt
, &fl
, sk
, 0);
1120 sk_setup_caps(sk
, &rt
->u
.dst
);
1122 /* Routing failed... */
1123 sk
->sk_route_caps
= 0;
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
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
;
1138 EXPORT_SYMBOL(inet_sk_rebuild_header
);
1140 static int inet_gso_send_check(struct sk_buff
*skb
)
1143 struct net_protocol
*ops
;
1148 if (unlikely(!pskb_may_pull(skb
, sizeof(*iph
))))
1153 if (ihl
< sizeof(*iph
))
1156 if (unlikely(!pskb_may_pull(skb
, ihl
)))
1159 __skb_pull(skb
, ihl
);
1160 skb_reset_transport_header(skb
);
1162 proto
= iph
->protocol
& (MAX_INET_PROTOS
- 1);
1163 err
= -EPROTONOSUPPORT
;
1166 ops
= rcu_dereference(inet_protos
[proto
]);
1167 if (likely(ops
&& ops
->gso_send_check
))
1168 err
= ops
->gso_send_check(skb
);
1175 static struct sk_buff
*inet_gso_segment(struct sk_buff
*skb
, int features
)
1177 struct sk_buff
*segs
= ERR_PTR(-EINVAL
);
1179 struct net_protocol
*ops
;
1184 if (!(features
& NETIF_F_V4_CSUM
))
1185 features
&= ~NETIF_F_SG
;
1187 if (unlikely(skb_shinfo(skb
)->gso_type
&
1195 if (unlikely(!pskb_may_pull(skb
, sizeof(*iph
))))
1200 if (ihl
< sizeof(*iph
))
1203 if (unlikely(!pskb_may_pull(skb
, ihl
)))
1206 __skb_pull(skb
, ihl
);
1207 skb_reset_transport_header(skb
);
1209 id
= ntohs(iph
->id
);
1210 proto
= iph
->protocol
& (MAX_INET_PROTOS
- 1);
1211 segs
= ERR_PTR(-EPROTONOSUPPORT
);
1214 ops
= rcu_dereference(inet_protos
[proto
]);
1215 if (likely(ops
&& ops
->gso_segment
))
1216 segs
= ops
->gso_segment(skb
, features
);
1219 if (!segs
|| unlikely(IS_ERR(segs
)))
1225 iph
->id
= htons(id
++);
1226 iph
->tot_len
= htons(skb
->len
- skb
->mac_len
);
1228 iph
->check
= ip_fast_csum(skb_network_header(skb
), iph
->ihl
);
1229 } while ((skb
= skb
->next
));
1235 unsigned long snmp_fold_field(void *mib
[], int offt
)
1237 unsigned long res
= 0;
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
);
1246 EXPORT_SYMBOL_GPL(snmp_fold_field
);
1248 int snmp_mib_init(void *ptr
[2], size_t mibsize
)
1250 BUG_ON(ptr
== NULL
);
1251 ptr
[0] = __alloc_percpu(mibsize
);
1254 ptr
[1] = __alloc_percpu(mibsize
);
1259 free_percpu(ptr
[0]);
1264 EXPORT_SYMBOL_GPL(snmp_mib_init
);
1266 void snmp_mib_free(void *ptr
[2])
1268 BUG_ON(ptr
== NULL
);
1269 free_percpu(ptr
[0]);
1270 free_percpu(ptr
[1]);
1271 ptr
[0] = ptr
[1] = NULL
;
1273 EXPORT_SYMBOL_GPL(snmp_mib_free
);
1275 #ifdef CONFIG_IP_MULTICAST
1276 static struct net_protocol igmp_protocol
= {
1277 .handler
= igmp_rcv
,
1281 static struct net_protocol tcp_protocol
= {
1282 .handler
= tcp_v4_rcv
,
1283 .err_handler
= tcp_v4_err
,
1284 .gso_send_check
= tcp_v4_gso_send_check
,
1285 .gso_segment
= tcp_tso_segment
,
1289 static struct net_protocol udp_protocol
= {
1291 .err_handler
= udp_err
,
1295 static struct net_protocol icmp_protocol
= {
1296 .handler
= icmp_rcv
,
1300 static int __init
init_ipv4_mibs(void)
1302 if (snmp_mib_init((void **)net_statistics
,
1303 sizeof(struct linux_mib
)) < 0)
1305 if (snmp_mib_init((void **)ip_statistics
,
1306 sizeof(struct ipstats_mib
)) < 0)
1308 if (snmp_mib_init((void **)icmp_statistics
,
1309 sizeof(struct icmp_mib
)) < 0)
1311 if (snmp_mib_init((void **)icmpmsg_statistics
,
1312 sizeof(struct icmpmsg_mib
)) < 0)
1313 goto err_icmpmsg_mib
;
1314 if (snmp_mib_init((void **)tcp_statistics
,
1315 sizeof(struct tcp_mib
)) < 0)
1317 if (snmp_mib_init((void **)udp_statistics
,
1318 sizeof(struct udp_mib
)) < 0)
1320 if (snmp_mib_init((void **)udplite_statistics
,
1321 sizeof(struct udp_mib
)) < 0)
1322 goto err_udplite_mib
;
1329 snmp_mib_free((void **)udp_statistics
);
1331 snmp_mib_free((void **)tcp_statistics
);
1333 snmp_mib_free((void **)icmpmsg_statistics
);
1335 snmp_mib_free((void **)icmp_statistics
);
1337 snmp_mib_free((void **)ip_statistics
);
1339 snmp_mib_free((void **)net_statistics
);
1344 static int ipv4_proc_init(void);
1347 * IP protocol layer initialiser
1350 static struct packet_type ip_packet_type
= {
1351 .type
= __constant_htons(ETH_P_IP
),
1353 .gso_send_check
= inet_gso_send_check
,
1354 .gso_segment
= inet_gso_segment
,
1357 static int __init
inet_init(void)
1359 struct sk_buff
*dummy_skb
;
1360 struct inet_protosw
*q
;
1361 struct list_head
*r
;
1364 BUILD_BUG_ON(sizeof(struct inet_skb_parm
) > sizeof(dummy_skb
->cb
));
1366 rc
= proto_register(&tcp_prot
, 1);
1370 rc
= proto_register(&udp_prot
, 1);
1372 goto out_unregister_tcp_proto
;
1374 rc
= proto_register(&raw_prot
, 1);
1376 goto out_unregister_udp_proto
;
1379 * Tell SOCKET that we are alive...
1382 (void)sock_register(&inet_family_ops
);
1385 * Add all the base protocols.
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");
1399 /* Register the socket-side information for inet_create. */
1400 for (r
= &inetsw
[0]; r
< &inetsw
[SOCK_MAX
]; ++r
)
1403 for (q
= inetsw_array
; q
< &inetsw_array
[INETSW_ARRAY_LEN
]; ++q
)
1404 inet_register_protosw(q
);
1407 * Set the ARP module up
1413 * Set the IP module up
1418 tcp_v4_init(&inet_family_ops
);
1420 /* Setup TCP slab cache for open requests. */
1423 /* Setup UDP memory threshold */
1426 /* Add UDP-Lite (RFC 3828) */
1427 udplite4_register();
1430 * Set the ICMP layer up
1433 icmp_init(&inet_family_ops
);
1436 * Initialise the multicast router
1438 #if defined(CONFIG_IP_MROUTE)
1442 * Initialise per-cpu ipv4 mibs
1445 if (init_ipv4_mibs())
1446 printk(KERN_CRIT
"inet_init: Cannot init ipv4 mibs\n"); ;
1452 dev_add_pack(&ip_packet_type
);
1457 out_unregister_udp_proto
:
1458 proto_unregister(&udp_prot
);
1459 out_unregister_tcp_proto
:
1460 proto_unregister(&tcp_prot
);
1464 fs_initcall(inet_init
);
1466 /* ------------------------------------------------------------------------ */
1468 #ifdef CONFIG_PROC_FS
1469 static int __init
ipv4_proc_init(void)
1473 if (raw_proc_init())
1475 if (tcp4_proc_init())
1477 if (udp4_proc_init())
1479 if (ip_misc_proc_init())
1494 #else /* CONFIG_PROC_FS */
1495 static int __init
ipv4_proc_init(void)
1499 #endif /* CONFIG_PROC_FS */
1501 MODULE_ALIAS_NETPROTO(PF_INET
);
1503 EXPORT_SYMBOL(inet_accept
);
1504 EXPORT_SYMBOL(inet_bind
);
1505 EXPORT_SYMBOL(inet_dgram_connect
);
1506 EXPORT_SYMBOL(inet_dgram_ops
);
1507 EXPORT_SYMBOL(inet_getname
);
1508 EXPORT_SYMBOL(inet_ioctl
);
1509 EXPORT_SYMBOL(inet_listen
);
1510 EXPORT_SYMBOL(inet_register_protosw
);
1511 EXPORT_SYMBOL(inet_release
);
1512 EXPORT_SYMBOL(inet_sendmsg
);
1513 EXPORT_SYMBOL(inet_shutdown
);
1514 EXPORT_SYMBOL(inet_sock_destruct
);
1515 EXPORT_SYMBOL(inet_stream_connect
);
1516 EXPORT_SYMBOL(inet_stream_ops
);
1517 EXPORT_SYMBOL(inet_unregister_protosw
);
1518 EXPORT_SYMBOL(net_statistics
);
1519 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind
);