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/config.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/fcntl.h>
84 #include <linux/interrupt.h>
85 #include <linux/stat.h>
86 #include <linux/init.h>
87 #include <linux/poll.h>
88 #include <linux/netfilter_ipv4.h>
90 #include <asm/uaccess.h>
91 #include <asm/system.h>
93 #include <linux/smp_lock.h>
94 #include <linux/inet.h>
95 #include <linux/igmp.h>
96 #include <linux/netdevice.h>
98 #include <net/protocol.h>
100 #include <net/route.h>
101 #include <net/ip_fib.h>
104 #include <linux/skbuff.h>
105 #include <net/sock.h>
107 #include <net/icmp.h>
108 #include <net/ipip.h>
109 #include <net/inet_common.h>
110 #include <net/xfrm.h>
111 #ifdef CONFIG_IP_MROUTE
112 #include <linux/mroute.h>
115 DEFINE_SNMP_STAT(struct linux_mib
, net_statistics
);
117 extern void ip_mc_drop_socket(struct sock
*sk
);
119 /* The inetsw table contains everything that inet_create needs to
120 * build a new socket.
122 static struct list_head inetsw
[SOCK_MAX
];
123 static DEFINE_SPINLOCK(inetsw_lock
);
125 /* New destruction routine */
127 void inet_sock_destruct(struct sock
*sk
)
129 struct inet_sock
*inet
= inet_sk(sk
);
131 __skb_queue_purge(&sk
->sk_receive_queue
);
132 __skb_queue_purge(&sk
->sk_error_queue
);
134 if (sk
->sk_type
== SOCK_STREAM
&& sk
->sk_state
!= TCP_CLOSE
) {
135 printk("Attempt to release TCP socket in state %d %p\n",
139 if (!sock_flag(sk
, SOCK_DEAD
)) {
140 printk("Attempt to release alive inet socket %p\n", sk
);
144 BUG_TRAP(!atomic_read(&sk
->sk_rmem_alloc
));
145 BUG_TRAP(!atomic_read(&sk
->sk_wmem_alloc
));
146 BUG_TRAP(!sk
->sk_wmem_queued
);
147 BUG_TRAP(!sk
->sk_forward_alloc
);
151 dst_release(sk
->sk_dst_cache
);
152 sk_refcnt_debug_dec(sk
);
156 * The routines beyond this point handle the behaviour of an AF_INET
157 * socket object. Mostly it punts to the subprotocols of IP to do
162 * Automatically bind an unbound socket.
165 static int inet_autobind(struct sock
*sk
)
167 struct inet_sock
*inet
;
168 /* We may need to bind the socket. */
172 if (sk
->sk_prot
->get_port(sk
, 0)) {
176 inet
->sport
= htons(inet
->num
);
183 * Move a socket into listening state.
185 int inet_listen(struct socket
*sock
, int backlog
)
187 struct sock
*sk
= sock
->sk
;
188 unsigned char old_state
;
194 if (sock
->state
!= SS_UNCONNECTED
|| sock
->type
!= SOCK_STREAM
)
197 old_state
= sk
->sk_state
;
198 if (!((1 << old_state
) & (TCPF_CLOSE
| TCPF_LISTEN
)))
201 /* Really, if the socket is already in listen state
202 * we can only allow the backlog to be adjusted.
204 if (old_state
!= TCP_LISTEN
) {
205 err
= tcp_listen_start(sk
);
209 sk
->sk_max_ack_backlog
= backlog
;
218 * Create an inet socket.
221 static int inet_create(struct socket
*sock
, int protocol
)
225 struct inet_protosw
*answer
;
226 struct inet_sock
*inet
;
227 struct proto
*answer_prot
;
228 unsigned char answer_flags
;
229 char answer_no_check
;
232 sock
->state
= SS_UNCONNECTED
;
234 /* Look for the requested type/protocol pair. */
237 list_for_each_rcu(p
, &inetsw
[sock
->type
]) {
238 answer
= list_entry(p
, struct inet_protosw
, list
);
240 /* Check the non-wild match. */
241 if (protocol
== answer
->protocol
) {
242 if (protocol
!= IPPROTO_IP
)
245 /* Check for the two wild cases. */
246 if (IPPROTO_IP
== protocol
) {
247 protocol
= answer
->protocol
;
250 if (IPPROTO_IP
== answer
->protocol
)
256 err
= -ESOCKTNOSUPPORT
;
260 if (answer
->capability
> 0 && !capable(answer
->capability
))
262 err
= -EPROTONOSUPPORT
;
266 sock
->ops
= answer
->ops
;
267 answer_prot
= answer
->prot
;
268 answer_no_check
= answer
->no_check
;
269 answer_flags
= answer
->flags
;
272 BUG_TRAP(answer_prot
->slab
!= NULL
);
275 sk
= sk_alloc(PF_INET
, GFP_KERNEL
, answer_prot
, 1);
280 sk
->sk_no_check
= answer_no_check
;
281 if (INET_PROTOSW_REUSE
& answer_flags
)
286 if (SOCK_RAW
== sock
->type
) {
287 inet
->num
= protocol
;
288 if (IPPROTO_RAW
== protocol
)
292 if (ipv4_config
.no_pmtu_disc
)
293 inet
->pmtudisc
= IP_PMTUDISC_DONT
;
295 inet
->pmtudisc
= IP_PMTUDISC_WANT
;
299 sock_init_data(sock
, sk
);
301 sk
->sk_destruct
= inet_sock_destruct
;
302 sk
->sk_family
= PF_INET
;
303 sk
->sk_protocol
= protocol
;
304 sk
->sk_backlog_rcv
= sk
->sk_prot
->backlog_rcv
;
310 inet
->mc_list
= NULL
;
312 sk_refcnt_debug_inc(sk
);
315 /* It assumes that any protocol which allows
316 * the user to assign a number at socket
317 * creation time automatically
320 inet
->sport
= htons(inet
->num
);
321 /* Add to protocol hash chains. */
322 sk
->sk_prot
->hash(sk
);
325 if (sk
->sk_prot
->init
) {
326 err
= sk
->sk_prot
->init(sk
);
328 sk_common_release(sk
);
339 * The peer socket should always be NULL (or else). When we call this
340 * function we are destroying the object and from then on nobody
341 * should refer to it.
343 int inet_release(struct socket
*sock
)
345 struct sock
*sk
= sock
->sk
;
350 /* Applications forget to leave groups before exiting */
351 ip_mc_drop_socket(sk
);
353 /* If linger is set, we don't return until the close
354 * is complete. Otherwise we return immediately. The
355 * actually closing is done the same either way.
357 * If the close is due to the process exiting, we never
361 if (sock_flag(sk
, SOCK_LINGER
) &&
362 !(current
->flags
& PF_EXITING
))
363 timeout
= sk
->sk_lingertime
;
365 sk
->sk_prot
->close(sk
, timeout
);
370 /* It is off by default, see below. */
371 int sysctl_ip_nonlocal_bind
;
373 int inet_bind(struct socket
*sock
, struct sockaddr
*uaddr
, int addr_len
)
375 struct sockaddr_in
*addr
= (struct sockaddr_in
*)uaddr
;
376 struct sock
*sk
= sock
->sk
;
377 struct inet_sock
*inet
= inet_sk(sk
);
382 /* If the socket has its own bind function then use it. (RAW) */
383 if (sk
->sk_prot
->bind
) {
384 err
= sk
->sk_prot
->bind(sk
, uaddr
, addr_len
);
388 if (addr_len
< sizeof(struct sockaddr_in
))
391 chk_addr_ret
= inet_addr_type(addr
->sin_addr
.s_addr
);
393 /* Not specified by any standard per-se, however it breaks too
394 * many applications when removed. It is unfortunate since
395 * allowing applications to make a non-local bind solves
396 * several problems with systems using dynamic addressing.
397 * (ie. your servers still start up even if your ISDN link
398 * is temporarily down)
400 err
= -EADDRNOTAVAIL
;
401 if (!sysctl_ip_nonlocal_bind
&&
403 addr
->sin_addr
.s_addr
!= INADDR_ANY
&&
404 chk_addr_ret
!= RTN_LOCAL
&&
405 chk_addr_ret
!= RTN_MULTICAST
&&
406 chk_addr_ret
!= RTN_BROADCAST
)
409 snum
= ntohs(addr
->sin_port
);
411 if (snum
&& snum
< PROT_SOCK
&& !capable(CAP_NET_BIND_SERVICE
))
414 /* We keep a pair of addresses. rcv_saddr is the one
415 * used by hash lookups, and saddr is used for transmit.
417 * In the BSD API these are the same except where it
418 * would be illegal to use them (multicast/broadcast) in
419 * which case the sending device address is used.
423 /* Check these errors (active socket, double bind). */
425 if (sk
->sk_state
!= TCP_CLOSE
|| inet
->num
)
426 goto out_release_sock
;
428 inet
->rcv_saddr
= inet
->saddr
= addr
->sin_addr
.s_addr
;
429 if (chk_addr_ret
== RTN_MULTICAST
|| chk_addr_ret
== RTN_BROADCAST
)
430 inet
->saddr
= 0; /* Use device */
432 /* Make sure we are allowed to bind here. */
433 if (sk
->sk_prot
->get_port(sk
, snum
)) {
434 inet
->saddr
= inet
->rcv_saddr
= 0;
436 goto out_release_sock
;
440 sk
->sk_userlocks
|= SOCK_BINDADDR_LOCK
;
442 sk
->sk_userlocks
|= SOCK_BINDPORT_LOCK
;
443 inet
->sport
= htons(inet
->num
);
454 int inet_dgram_connect(struct socket
*sock
, struct sockaddr
* uaddr
,
455 int addr_len
, int flags
)
457 struct sock
*sk
= sock
->sk
;
459 if (uaddr
->sa_family
== AF_UNSPEC
)
460 return sk
->sk_prot
->disconnect(sk
, flags
);
462 if (!inet_sk(sk
)->num
&& inet_autobind(sk
))
464 return sk
->sk_prot
->connect(sk
, (struct sockaddr
*)uaddr
, addr_len
);
467 static long inet_wait_for_connect(struct sock
*sk
, long timeo
)
471 prepare_to_wait(sk
->sk_sleep
, &wait
, TASK_INTERRUPTIBLE
);
473 /* Basic assumption: if someone sets sk->sk_err, he _must_
474 * change state of the socket from TCP_SYN_*.
475 * Connect() does not allow to get error notifications
476 * without closing the socket.
478 while ((1 << sk
->sk_state
) & (TCPF_SYN_SENT
| TCPF_SYN_RECV
)) {
480 timeo
= schedule_timeout(timeo
);
482 if (signal_pending(current
) || !timeo
)
484 prepare_to_wait(sk
->sk_sleep
, &wait
, TASK_INTERRUPTIBLE
);
486 finish_wait(sk
->sk_sleep
, &wait
);
491 * Connect to a remote host. There is regrettably still a little
492 * TCP 'magic' in here.
494 int inet_stream_connect(struct socket
*sock
, struct sockaddr
*uaddr
,
495 int addr_len
, int flags
)
497 struct sock
*sk
= sock
->sk
;
503 if (uaddr
->sa_family
== AF_UNSPEC
) {
504 err
= sk
->sk_prot
->disconnect(sk
, flags
);
505 sock
->state
= err
? SS_DISCONNECTING
: SS_UNCONNECTED
;
509 switch (sock
->state
) {
518 /* Fall out of switch with err, set for this state */
522 if (sk
->sk_state
!= TCP_CLOSE
)
525 err
= sk
->sk_prot
->connect(sk
, uaddr
, addr_len
);
529 sock
->state
= SS_CONNECTING
;
531 /* Just entered SS_CONNECTING state; the only
532 * difference is that return value in non-blocking
533 * case is EINPROGRESS, rather than EALREADY.
539 timeo
= sock_sndtimeo(sk
, flags
& O_NONBLOCK
);
541 if ((1 << sk
->sk_state
) & (TCPF_SYN_SENT
| TCPF_SYN_RECV
)) {
542 /* Error code is set above */
543 if (!timeo
|| !inet_wait_for_connect(sk
, timeo
))
546 err
= sock_intr_errno(timeo
);
547 if (signal_pending(current
))
551 /* Connection was closed by RST, timeout, ICMP error
552 * or another process disconnected us.
554 if (sk
->sk_state
== TCP_CLOSE
)
557 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
558 * and error was received after socket entered established state.
559 * Hence, it is handled normally after connect() return successfully.
562 sock
->state
= SS_CONNECTED
;
569 err
= sock_error(sk
) ? : -ECONNABORTED
;
570 sock
->state
= SS_UNCONNECTED
;
571 if (sk
->sk_prot
->disconnect(sk
, flags
))
572 sock
->state
= SS_DISCONNECTING
;
577 * Accept a pending connection. The TCP layer now gives BSD semantics.
580 int inet_accept(struct socket
*sock
, struct socket
*newsock
, int flags
)
582 struct sock
*sk1
= sock
->sk
;
584 struct sock
*sk2
= sk1
->sk_prot
->accept(sk1
, flags
, &err
);
591 BUG_TRAP((1 << sk2
->sk_state
) &
592 (TCPF_ESTABLISHED
| TCPF_CLOSE_WAIT
| TCPF_CLOSE
));
594 sock_graft(sk2
, newsock
);
596 newsock
->state
= SS_CONNECTED
;
605 * This does both peername and sockname.
607 int inet_getname(struct socket
*sock
, struct sockaddr
*uaddr
,
608 int *uaddr_len
, int peer
)
610 struct sock
*sk
= sock
->sk
;
611 struct inet_sock
*inet
= inet_sk(sk
);
612 struct sockaddr_in
*sin
= (struct sockaddr_in
*)uaddr
;
614 sin
->sin_family
= AF_INET
;
617 (((1 << sk
->sk_state
) & (TCPF_CLOSE
| TCPF_SYN_SENT
)) &&
620 sin
->sin_port
= inet
->dport
;
621 sin
->sin_addr
.s_addr
= inet
->daddr
;
623 __u32 addr
= inet
->rcv_saddr
;
626 sin
->sin_port
= inet
->sport
;
627 sin
->sin_addr
.s_addr
= addr
;
629 memset(sin
->sin_zero
, 0, sizeof(sin
->sin_zero
));
630 *uaddr_len
= sizeof(*sin
);
634 int inet_sendmsg(struct kiocb
*iocb
, struct socket
*sock
, struct msghdr
*msg
,
637 struct sock
*sk
= sock
->sk
;
639 /* We may need to bind the socket. */
640 if (!inet_sk(sk
)->num
&& inet_autobind(sk
))
643 return sk
->sk_prot
->sendmsg(iocb
, sk
, msg
, size
);
647 static ssize_t
inet_sendpage(struct socket
*sock
, struct page
*page
, int offset
, size_t size
, int flags
)
649 struct sock
*sk
= sock
->sk
;
651 /* We may need to bind the socket. */
652 if (!inet_sk(sk
)->num
&& inet_autobind(sk
))
655 if (sk
->sk_prot
->sendpage
)
656 return sk
->sk_prot
->sendpage(sk
, page
, offset
, size
, flags
);
657 return sock_no_sendpage(sock
, page
, offset
, size
, flags
);
661 int inet_shutdown(struct socket
*sock
, int how
)
663 struct sock
*sk
= sock
->sk
;
666 /* This should really check to make sure
667 * the socket is a TCP socket. (WHY AC...)
669 how
++; /* maps 0->1 has the advantage of making bit 1 rcvs and
672 if ((how
& ~SHUTDOWN_MASK
) || !how
) /* MAXINT->0 */
676 if (sock
->state
== SS_CONNECTING
) {
677 if ((1 << sk
->sk_state
) &
678 (TCPF_SYN_SENT
| TCPF_SYN_RECV
| TCPF_CLOSE
))
679 sock
->state
= SS_DISCONNECTING
;
681 sock
->state
= SS_CONNECTED
;
684 switch (sk
->sk_state
) {
687 /* Hack to wake up other listeners, who can poll for
688 POLLHUP, even on eg. unconnected UDP sockets -- RR */
690 sk
->sk_shutdown
|= how
;
691 if (sk
->sk_prot
->shutdown
)
692 sk
->sk_prot
->shutdown(sk
, how
);
695 /* Remaining two branches are temporary solution for missing
696 * close() in multithreaded environment. It is _not_ a good idea,
697 * but we have no choice until close() is repaired at VFS level.
700 if (!(how
& RCV_SHUTDOWN
))
704 err
= sk
->sk_prot
->disconnect(sk
, O_NONBLOCK
);
705 sock
->state
= err
? SS_DISCONNECTING
: SS_UNCONNECTED
;
709 /* Wake up anyone sleeping in poll. */
710 sk
->sk_state_change(sk
);
716 * ioctl() calls you can issue on an INET socket. Most of these are
717 * device configuration and stuff and very rarely used. Some ioctls
718 * pass on to the socket itself.
720 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
721 * loads the devconfigure module does its configuring and unloads it.
722 * There's a good 20K of config code hanging around the kernel.
725 int inet_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
727 struct sock
*sk
= sock
->sk
;
732 err
= sock_get_timestamp(sk
, (struct timeval __user
*)arg
);
737 err
= ip_rt_ioctl(cmd
, (void __user
*)arg
);
742 err
= arp_ioctl(cmd
, (void __user
*)arg
);
755 err
= devinet_ioctl(cmd
, (void __user
*)arg
);
758 if (!sk
->sk_prot
->ioctl
||
759 (err
= sk
->sk_prot
->ioctl(sk
, cmd
, arg
)) ==
761 err
= dev_ioctl(cmd
, (void __user
*)arg
);
767 struct proto_ops inet_stream_ops
= {
769 .owner
= THIS_MODULE
,
770 .release
= inet_release
,
772 .connect
= inet_stream_connect
,
773 .socketpair
= sock_no_socketpair
,
774 .accept
= inet_accept
,
775 .getname
= inet_getname
,
778 .listen
= inet_listen
,
779 .shutdown
= inet_shutdown
,
780 .setsockopt
= sock_common_setsockopt
,
781 .getsockopt
= sock_common_getsockopt
,
782 .sendmsg
= inet_sendmsg
,
783 .recvmsg
= sock_common_recvmsg
,
784 .mmap
= sock_no_mmap
,
785 .sendpage
= tcp_sendpage
788 struct proto_ops inet_dgram_ops
= {
790 .owner
= THIS_MODULE
,
791 .release
= inet_release
,
793 .connect
= inet_dgram_connect
,
794 .socketpair
= sock_no_socketpair
,
795 .accept
= sock_no_accept
,
796 .getname
= inet_getname
,
799 .listen
= sock_no_listen
,
800 .shutdown
= inet_shutdown
,
801 .setsockopt
= sock_common_setsockopt
,
802 .getsockopt
= sock_common_getsockopt
,
803 .sendmsg
= inet_sendmsg
,
804 .recvmsg
= sock_common_recvmsg
,
805 .mmap
= sock_no_mmap
,
806 .sendpage
= inet_sendpage
,
810 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
813 static struct proto_ops inet_sockraw_ops
= {
815 .owner
= THIS_MODULE
,
816 .release
= inet_release
,
818 .connect
= inet_dgram_connect
,
819 .socketpair
= sock_no_socketpair
,
820 .accept
= sock_no_accept
,
821 .getname
= inet_getname
,
822 .poll
= datagram_poll
,
824 .listen
= sock_no_listen
,
825 .shutdown
= inet_shutdown
,
826 .setsockopt
= sock_common_setsockopt
,
827 .getsockopt
= sock_common_getsockopt
,
828 .sendmsg
= inet_sendmsg
,
829 .recvmsg
= sock_common_recvmsg
,
830 .mmap
= sock_no_mmap
,
831 .sendpage
= inet_sendpage
,
834 static struct net_proto_family inet_family_ops
= {
836 .create
= inet_create
,
837 .owner
= THIS_MODULE
,
841 extern void tcp_init(void);
842 extern void tcp_v4_init(struct net_proto_family
*);
844 /* Upon startup we insert all the elements in inetsw_array[] into
845 * the linked list inetsw.
847 static struct inet_protosw inetsw_array
[] =
851 .protocol
= IPPROTO_TCP
,
853 .ops
= &inet_stream_ops
,
856 .flags
= INET_PROTOSW_PERMANENT
,
861 .protocol
= IPPROTO_UDP
,
863 .ops
= &inet_dgram_ops
,
865 .no_check
= UDP_CSUM_DEFAULT
,
866 .flags
= INET_PROTOSW_PERMANENT
,
872 .protocol
= IPPROTO_IP
, /* wild card */
874 .ops
= &inet_sockraw_ops
,
875 .capability
= CAP_NET_RAW
,
876 .no_check
= UDP_CSUM_DEFAULT
,
877 .flags
= INET_PROTOSW_REUSE
,
881 #define INETSW_ARRAY_LEN (sizeof(inetsw_array) / sizeof(struct inet_protosw))
883 void inet_register_protosw(struct inet_protosw
*p
)
885 struct list_head
*lh
;
886 struct inet_protosw
*answer
;
887 int protocol
= p
->protocol
;
888 struct list_head
*last_perm
;
890 spin_lock_bh(&inetsw_lock
);
892 if (p
->type
>= SOCK_MAX
)
895 /* If we are trying to override a permanent protocol, bail. */
897 last_perm
= &inetsw
[p
->type
];
898 list_for_each(lh
, &inetsw
[p
->type
]) {
899 answer
= list_entry(lh
, struct inet_protosw
, list
);
901 /* Check only the non-wild match. */
902 if (INET_PROTOSW_PERMANENT
& answer
->flags
) {
903 if (protocol
== answer
->protocol
)
913 /* Add the new entry after the last permanent entry if any, so that
914 * the new entry does not override a permanent entry when matched with
915 * a wild-card protocol. But it is allowed to override any existing
916 * non-permanent entry. This means that when we remove this entry, the
917 * system automatically returns to the old behavior.
919 list_add_rcu(&p
->list
, last_perm
);
921 spin_unlock_bh(&inetsw_lock
);
928 printk(KERN_ERR
"Attempt to override permanent protocol %d.\n",
934 "Ignoring attempt to register invalid socket type %d.\n",
939 void inet_unregister_protosw(struct inet_protosw
*p
)
941 if (INET_PROTOSW_PERMANENT
& p
->flags
) {
943 "Attempt to unregister permanent protocol %d.\n",
946 spin_lock_bh(&inetsw_lock
);
947 list_del_rcu(&p
->list
);
948 spin_unlock_bh(&inetsw_lock
);
955 * Shall we try to damage output packets if routing dev changes?
958 int sysctl_ip_dynaddr
;
960 static int inet_sk_reselect_saddr(struct sock
*sk
)
962 struct inet_sock
*inet
= inet_sk(sk
);
965 __u32 old_saddr
= inet
->saddr
;
967 __u32 daddr
= inet
->daddr
;
969 if (inet
->opt
&& inet
->opt
->srr
)
970 daddr
= inet
->opt
->faddr
;
972 /* Query new route. */
973 err
= ip_route_connect(&rt
, daddr
, 0,
977 inet
->sport
, inet
->dport
, sk
);
981 sk_setup_caps(sk
, &rt
->u
.dst
);
983 new_saddr
= rt
->rt_src
;
985 if (new_saddr
== old_saddr
)
988 if (sysctl_ip_dynaddr
> 1) {
989 printk(KERN_INFO
"%s(): shifting inet->"
990 "saddr from %d.%d.%d.%d to %d.%d.%d.%d\n",
996 inet
->saddr
= inet
->rcv_saddr
= new_saddr
;
999 * XXX The only one ugly spot where we need to
1000 * XXX really change the sockets identity after
1001 * XXX it has entered the hashes. -DaveM
1003 * Besides that, it does not check for connection
1004 * uniqueness. Wait for troubles.
1006 __sk_prot_rehash(sk
);
1010 int inet_sk_rebuild_header(struct sock
*sk
)
1012 struct inet_sock
*inet
= inet_sk(sk
);
1013 struct rtable
*rt
= (struct rtable
*)__sk_dst_check(sk
, 0);
1017 /* Route is OK, nothing to do. */
1022 daddr
= inet
->daddr
;
1023 if (inet
->opt
&& inet
->opt
->srr
)
1024 daddr
= inet
->opt
->faddr
;
1027 .oif
= sk
->sk_bound_dev_if
,
1031 .saddr
= inet
->saddr
,
1032 .tos
= RT_CONN_FLAGS(sk
),
1035 .proto
= sk
->sk_protocol
,
1038 .sport
= inet
->sport
,
1039 .dport
= inet
->dport
,
1044 err
= ip_route_output_flow(&rt
, &fl
, sk
, 0);
1047 sk_setup_caps(sk
, &rt
->u
.dst
);
1049 /* Routing failed... */
1050 sk
->sk_route_caps
= 0;
1052 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1053 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1055 if (!sysctl_ip_dynaddr
||
1056 sk
->sk_state
!= TCP_SYN_SENT
||
1057 (sk
->sk_userlocks
& SOCK_BINDADDR_LOCK
) ||
1058 (err
= inet_sk_reselect_saddr(sk
)) != 0)
1059 sk
->sk_err_soft
= -err
;
1065 EXPORT_SYMBOL(inet_sk_rebuild_header
);
1067 #ifdef CONFIG_IP_MULTICAST
1068 static struct net_protocol igmp_protocol
= {
1069 .handler
= igmp_rcv
,
1073 static struct net_protocol tcp_protocol
= {
1074 .handler
= tcp_v4_rcv
,
1075 .err_handler
= tcp_v4_err
,
1079 static struct net_protocol udp_protocol
= {
1081 .err_handler
= udp_err
,
1085 static struct net_protocol icmp_protocol
= {
1086 .handler
= icmp_rcv
,
1089 static int __init
init_ipv4_mibs(void)
1091 net_statistics
[0] = alloc_percpu(struct linux_mib
);
1092 net_statistics
[1] = alloc_percpu(struct linux_mib
);
1093 ip_statistics
[0] = alloc_percpu(struct ipstats_mib
);
1094 ip_statistics
[1] = alloc_percpu(struct ipstats_mib
);
1095 icmp_statistics
[0] = alloc_percpu(struct icmp_mib
);
1096 icmp_statistics
[1] = alloc_percpu(struct icmp_mib
);
1097 tcp_statistics
[0] = alloc_percpu(struct tcp_mib
);
1098 tcp_statistics
[1] = alloc_percpu(struct tcp_mib
);
1099 udp_statistics
[0] = alloc_percpu(struct udp_mib
);
1100 udp_statistics
[1] = alloc_percpu(struct udp_mib
);
1102 (net_statistics
[0] && net_statistics
[1] && ip_statistics
[0]
1103 && ip_statistics
[1] && tcp_statistics
[0] && tcp_statistics
[1]
1104 && udp_statistics
[0] && udp_statistics
[1]))
1107 (void) tcp_mib_init();
1112 static int ipv4_proc_init(void);
1113 extern void ipfrag_init(void);
1116 * IP protocol layer initialiser
1119 static struct packet_type ip_packet_type
= {
1120 .type
= __constant_htons(ETH_P_IP
),
1124 static int __init
inet_init(void)
1126 struct sk_buff
*dummy_skb
;
1127 struct inet_protosw
*q
;
1128 struct list_head
*r
;
1131 if (sizeof(struct inet_skb_parm
) > sizeof(dummy_skb
->cb
)) {
1132 printk(KERN_CRIT
"%s: panic\n", __FUNCTION__
);
1136 rc
= proto_register(&tcp_prot
, 1);
1140 rc
= proto_register(&udp_prot
, 1);
1142 goto out_unregister_tcp_proto
;
1144 rc
= proto_register(&raw_prot
, 1);
1146 goto out_unregister_udp_proto
;
1149 * Tell SOCKET that we are alive...
1152 (void)sock_register(&inet_family_ops
);
1155 * Add all the base protocols.
1158 if (inet_add_protocol(&icmp_protocol
, IPPROTO_ICMP
) < 0)
1159 printk(KERN_CRIT
"inet_init: Cannot add ICMP protocol\n");
1160 if (inet_add_protocol(&udp_protocol
, IPPROTO_UDP
) < 0)
1161 printk(KERN_CRIT
"inet_init: Cannot add UDP protocol\n");
1162 if (inet_add_protocol(&tcp_protocol
, IPPROTO_TCP
) < 0)
1163 printk(KERN_CRIT
"inet_init: Cannot add TCP protocol\n");
1164 #ifdef CONFIG_IP_MULTICAST
1165 if (inet_add_protocol(&igmp_protocol
, IPPROTO_IGMP
) < 0)
1166 printk(KERN_CRIT
"inet_init: Cannot add IGMP protocol\n");
1169 /* Register the socket-side information for inet_create. */
1170 for (r
= &inetsw
[0]; r
< &inetsw
[SOCK_MAX
]; ++r
)
1173 for (q
= inetsw_array
; q
< &inetsw_array
[INETSW_ARRAY_LEN
]; ++q
)
1174 inet_register_protosw(q
);
1177 * Set the ARP module up
1183 * Set the IP module up
1188 tcp_v4_init(&inet_family_ops
);
1190 /* Setup TCP slab cache for open requests. */
1195 * Set the ICMP layer up
1198 icmp_init(&inet_family_ops
);
1201 * Initialise the multicast router
1203 #if defined(CONFIG_IP_MROUTE)
1207 * Initialise per-cpu ipv4 mibs
1210 if(init_ipv4_mibs())
1211 printk(KERN_CRIT
"inet_init: Cannot init ipv4 mibs\n"); ;
1217 dev_add_pack(&ip_packet_type
);
1222 out_unregister_tcp_proto
:
1223 proto_unregister(&tcp_prot
);
1224 out_unregister_udp_proto
:
1225 proto_unregister(&udp_prot
);
1229 module_init(inet_init
);
1231 /* ------------------------------------------------------------------------ */
1233 #ifdef CONFIG_PROC_FS
1234 extern int fib_proc_init(void);
1235 extern void fib_proc_exit(void);
1236 #ifdef CONFIG_IP_FIB_TRIE
1237 extern int fib_stat_proc_init(void);
1238 extern void fib_stat_proc_exit(void);
1240 extern int ip_misc_proc_init(void);
1241 extern int raw_proc_init(void);
1242 extern void raw_proc_exit(void);
1243 extern int tcp4_proc_init(void);
1244 extern void tcp4_proc_exit(void);
1245 extern int udp4_proc_init(void);
1246 extern void udp4_proc_exit(void);
1248 static int __init
ipv4_proc_init(void)
1252 if (raw_proc_init())
1254 if (tcp4_proc_init())
1256 if (udp4_proc_init())
1258 if (fib_proc_init())
1260 #ifdef CONFIG_IP_FIB_TRIE
1261 if (fib_stat_proc_init())
1264 if (ip_misc_proc_init())
1269 #ifdef CONFIG_IP_FIB_TRIE
1270 fib_stat_proc_exit();
1285 #else /* CONFIG_PROC_FS */
1286 static int __init
ipv4_proc_init(void)
1290 #endif /* CONFIG_PROC_FS */
1292 MODULE_ALIAS_NETPROTO(PF_INET
);
1294 EXPORT_SYMBOL(inet_accept
);
1295 EXPORT_SYMBOL(inet_bind
);
1296 EXPORT_SYMBOL(inet_dgram_connect
);
1297 EXPORT_SYMBOL(inet_dgram_ops
);
1298 EXPORT_SYMBOL(inet_getname
);
1299 EXPORT_SYMBOL(inet_ioctl
);
1300 EXPORT_SYMBOL(inet_listen
);
1301 EXPORT_SYMBOL(inet_register_protosw
);
1302 EXPORT_SYMBOL(inet_release
);
1303 EXPORT_SYMBOL(inet_sendmsg
);
1304 EXPORT_SYMBOL(inet_shutdown
);
1305 EXPORT_SYMBOL(inet_sock_destruct
);
1306 EXPORT_SYMBOL(inet_stream_connect
);
1307 EXPORT_SYMBOL(inet_stream_ops
);
1308 EXPORT_SYMBOL(inet_unregister_protosw
);
1309 EXPORT_SYMBOL(net_statistics
);
1310 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind
);