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>
102 #include <net/inet_connection_sock.h>
105 #include <linux/skbuff.h>
106 #include <net/sock.h>
108 #include <net/icmp.h>
109 #include <net/ipip.h>
110 #include <net/inet_common.h>
111 #include <net/xfrm.h>
112 #ifdef CONFIG_IP_MROUTE
113 #include <linux/mroute.h>
116 DEFINE_SNMP_STAT(struct linux_mib
, net_statistics
) __read_mostly
;
118 extern void ip_mc_drop_socket(struct sock
*sk
);
120 /* The inetsw table contains everything that inet_create needs to
121 * build a new socket.
123 static struct list_head inetsw
[SOCK_MAX
];
124 static DEFINE_SPINLOCK(inetsw_lock
);
126 /* New destruction routine */
128 void inet_sock_destruct(struct sock
*sk
)
130 struct inet_sock
*inet
= inet_sk(sk
);
132 __skb_queue_purge(&sk
->sk_receive_queue
);
133 __skb_queue_purge(&sk
->sk_error_queue
);
135 if (sk
->sk_type
== SOCK_STREAM
&& sk
->sk_state
!= TCP_CLOSE
) {
136 printk("Attempt to release TCP socket in state %d %p\n",
140 if (!sock_flag(sk
, SOCK_DEAD
)) {
141 printk("Attempt to release alive inet socket %p\n", sk
);
145 BUG_TRAP(!atomic_read(&sk
->sk_rmem_alloc
));
146 BUG_TRAP(!atomic_read(&sk
->sk_wmem_alloc
));
147 BUG_TRAP(!sk
->sk_wmem_queued
);
148 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
= inet_csk_listen_start(sk
, TCP_SYNQ_HSIZE
);
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
;
230 int try_loading_module
= 0;
233 sock
->state
= SS_UNCONNECTED
;
235 /* Look for the requested type/protocol pair. */
238 err
= -ESOCKTNOSUPPORT
;
240 list_for_each_rcu(p
, &inetsw
[sock
->type
]) {
241 answer
= list_entry(p
, struct inet_protosw
, list
);
243 /* Check the non-wild match. */
244 if (protocol
== answer
->protocol
) {
245 if (protocol
!= IPPROTO_IP
)
248 /* Check for the two wild cases. */
249 if (IPPROTO_IP
== protocol
) {
250 protocol
= answer
->protocol
;
253 if (IPPROTO_IP
== answer
->protocol
)
256 err
= -EPROTONOSUPPORT
;
260 if (unlikely(answer
== NULL
)) {
261 if (try_loading_module
< 2) {
264 * Be more specific, e.g. net-pf-2-proto-132-type-1
265 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
267 if (++try_loading_module
== 1)
268 request_module("net-pf-%d-proto-%d-type-%d",
269 PF_INET
, protocol
, sock
->type
);
271 * Fall back to generic, e.g. net-pf-2-proto-132
272 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
275 request_module("net-pf-%d-proto-%d",
277 goto lookup_protocol
;
283 if (answer
->capability
> 0 && !capable(answer
->capability
))
286 sock
->ops
= answer
->ops
;
287 answer_prot
= answer
->prot
;
288 answer_no_check
= answer
->no_check
;
289 answer_flags
= answer
->flags
;
292 BUG_TRAP(answer_prot
->slab
!= NULL
);
295 sk
= sk_alloc(PF_INET
, GFP_KERNEL
, answer_prot
, 1);
300 sk
->sk_no_check
= answer_no_check
;
301 if (INET_PROTOSW_REUSE
& answer_flags
)
306 if (SOCK_RAW
== sock
->type
) {
307 inet
->num
= protocol
;
308 if (IPPROTO_RAW
== protocol
)
312 if (ipv4_config
.no_pmtu_disc
)
313 inet
->pmtudisc
= IP_PMTUDISC_DONT
;
315 inet
->pmtudisc
= IP_PMTUDISC_WANT
;
319 sock_init_data(sock
, sk
);
321 sk
->sk_destruct
= inet_sock_destruct
;
322 sk
->sk_family
= PF_INET
;
323 sk
->sk_protocol
= protocol
;
324 sk
->sk_backlog_rcv
= sk
->sk_prot
->backlog_rcv
;
330 inet
->mc_list
= NULL
;
332 sk_refcnt_debug_inc(sk
);
335 /* It assumes that any protocol which allows
336 * the user to assign a number at socket
337 * creation time automatically
340 inet
->sport
= htons(inet
->num
);
341 /* Add to protocol hash chains. */
342 sk
->sk_prot
->hash(sk
);
345 if (sk
->sk_prot
->init
) {
346 err
= sk
->sk_prot
->init(sk
);
348 sk_common_release(sk
);
359 * The peer socket should always be NULL (or else). When we call this
360 * function we are destroying the object and from then on nobody
361 * should refer to it.
363 int inet_release(struct socket
*sock
)
365 struct sock
*sk
= sock
->sk
;
370 /* Applications forget to leave groups before exiting */
371 ip_mc_drop_socket(sk
);
373 /* If linger is set, we don't return until the close
374 * is complete. Otherwise we return immediately. The
375 * actually closing is done the same either way.
377 * If the close is due to the process exiting, we never
381 if (sock_flag(sk
, SOCK_LINGER
) &&
382 !(current
->flags
& PF_EXITING
))
383 timeout
= sk
->sk_lingertime
;
385 sk
->sk_prot
->close(sk
, timeout
);
390 /* It is off by default, see below. */
391 int sysctl_ip_nonlocal_bind
;
393 int inet_bind(struct socket
*sock
, struct sockaddr
*uaddr
, int addr_len
)
395 struct sockaddr_in
*addr
= (struct sockaddr_in
*)uaddr
;
396 struct sock
*sk
= sock
->sk
;
397 struct inet_sock
*inet
= inet_sk(sk
);
402 /* If the socket has its own bind function then use it. (RAW) */
403 if (sk
->sk_prot
->bind
) {
404 err
= sk
->sk_prot
->bind(sk
, uaddr
, addr_len
);
408 if (addr_len
< sizeof(struct sockaddr_in
))
411 chk_addr_ret
= inet_addr_type(addr
->sin_addr
.s_addr
);
413 /* Not specified by any standard per-se, however it breaks too
414 * many applications when removed. It is unfortunate since
415 * allowing applications to make a non-local bind solves
416 * several problems with systems using dynamic addressing.
417 * (ie. your servers still start up even if your ISDN link
418 * is temporarily down)
420 err
= -EADDRNOTAVAIL
;
421 if (!sysctl_ip_nonlocal_bind
&&
423 addr
->sin_addr
.s_addr
!= INADDR_ANY
&&
424 chk_addr_ret
!= RTN_LOCAL
&&
425 chk_addr_ret
!= RTN_MULTICAST
&&
426 chk_addr_ret
!= RTN_BROADCAST
)
429 snum
= ntohs(addr
->sin_port
);
431 if (snum
&& snum
< PROT_SOCK
&& !capable(CAP_NET_BIND_SERVICE
))
434 /* We keep a pair of addresses. rcv_saddr is the one
435 * used by hash lookups, and saddr is used for transmit.
437 * In the BSD API these are the same except where it
438 * would be illegal to use them (multicast/broadcast) in
439 * which case the sending device address is used.
443 /* Check these errors (active socket, double bind). */
445 if (sk
->sk_state
!= TCP_CLOSE
|| inet
->num
)
446 goto out_release_sock
;
448 inet
->rcv_saddr
= inet
->saddr
= addr
->sin_addr
.s_addr
;
449 if (chk_addr_ret
== RTN_MULTICAST
|| chk_addr_ret
== RTN_BROADCAST
)
450 inet
->saddr
= 0; /* Use device */
452 /* Make sure we are allowed to bind here. */
453 if (sk
->sk_prot
->get_port(sk
, snum
)) {
454 inet
->saddr
= inet
->rcv_saddr
= 0;
456 goto out_release_sock
;
460 sk
->sk_userlocks
|= SOCK_BINDADDR_LOCK
;
462 sk
->sk_userlocks
|= SOCK_BINDPORT_LOCK
;
463 inet
->sport
= htons(inet
->num
);
474 int inet_dgram_connect(struct socket
*sock
, struct sockaddr
* uaddr
,
475 int addr_len
, int flags
)
477 struct sock
*sk
= sock
->sk
;
479 if (uaddr
->sa_family
== AF_UNSPEC
)
480 return sk
->sk_prot
->disconnect(sk
, flags
);
482 if (!inet_sk(sk
)->num
&& inet_autobind(sk
))
484 return sk
->sk_prot
->connect(sk
, (struct sockaddr
*)uaddr
, addr_len
);
487 static long inet_wait_for_connect(struct sock
*sk
, long timeo
)
491 prepare_to_wait(sk
->sk_sleep
, &wait
, TASK_INTERRUPTIBLE
);
493 /* Basic assumption: if someone sets sk->sk_err, he _must_
494 * change state of the socket from TCP_SYN_*.
495 * Connect() does not allow to get error notifications
496 * without closing the socket.
498 while ((1 << sk
->sk_state
) & (TCPF_SYN_SENT
| TCPF_SYN_RECV
)) {
500 timeo
= schedule_timeout(timeo
);
502 if (signal_pending(current
) || !timeo
)
504 prepare_to_wait(sk
->sk_sleep
, &wait
, TASK_INTERRUPTIBLE
);
506 finish_wait(sk
->sk_sleep
, &wait
);
511 * Connect to a remote host. There is regrettably still a little
512 * TCP 'magic' in here.
514 int inet_stream_connect(struct socket
*sock
, struct sockaddr
*uaddr
,
515 int addr_len
, int flags
)
517 struct sock
*sk
= sock
->sk
;
523 if (uaddr
->sa_family
== AF_UNSPEC
) {
524 err
= sk
->sk_prot
->disconnect(sk
, flags
);
525 sock
->state
= err
? SS_DISCONNECTING
: SS_UNCONNECTED
;
529 switch (sock
->state
) {
538 /* Fall out of switch with err, set for this state */
542 if (sk
->sk_state
!= TCP_CLOSE
)
545 err
= sk
->sk_prot
->connect(sk
, uaddr
, addr_len
);
549 sock
->state
= SS_CONNECTING
;
551 /* Just entered SS_CONNECTING state; the only
552 * difference is that return value in non-blocking
553 * case is EINPROGRESS, rather than EALREADY.
559 timeo
= sock_sndtimeo(sk
, flags
& O_NONBLOCK
);
561 if ((1 << sk
->sk_state
) & (TCPF_SYN_SENT
| TCPF_SYN_RECV
)) {
562 /* Error code is set above */
563 if (!timeo
|| !inet_wait_for_connect(sk
, timeo
))
566 err
= sock_intr_errno(timeo
);
567 if (signal_pending(current
))
571 /* Connection was closed by RST, timeout, ICMP error
572 * or another process disconnected us.
574 if (sk
->sk_state
== TCP_CLOSE
)
577 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
578 * and error was received after socket entered established state.
579 * Hence, it is handled normally after connect() return successfully.
582 sock
->state
= SS_CONNECTED
;
589 err
= sock_error(sk
) ? : -ECONNABORTED
;
590 sock
->state
= SS_UNCONNECTED
;
591 if (sk
->sk_prot
->disconnect(sk
, flags
))
592 sock
->state
= SS_DISCONNECTING
;
597 * Accept a pending connection. The TCP layer now gives BSD semantics.
600 int inet_accept(struct socket
*sock
, struct socket
*newsock
, int flags
)
602 struct sock
*sk1
= sock
->sk
;
604 struct sock
*sk2
= sk1
->sk_prot
->accept(sk1
, flags
, &err
);
611 BUG_TRAP((1 << sk2
->sk_state
) &
612 (TCPF_ESTABLISHED
| TCPF_CLOSE_WAIT
| TCPF_CLOSE
));
614 sock_graft(sk2
, newsock
);
616 newsock
->state
= SS_CONNECTED
;
625 * This does both peername and sockname.
627 int inet_getname(struct socket
*sock
, struct sockaddr
*uaddr
,
628 int *uaddr_len
, int peer
)
630 struct sock
*sk
= sock
->sk
;
631 struct inet_sock
*inet
= inet_sk(sk
);
632 struct sockaddr_in
*sin
= (struct sockaddr_in
*)uaddr
;
634 sin
->sin_family
= AF_INET
;
637 (((1 << sk
->sk_state
) & (TCPF_CLOSE
| TCPF_SYN_SENT
)) &&
640 sin
->sin_port
= inet
->dport
;
641 sin
->sin_addr
.s_addr
= inet
->daddr
;
643 __u32 addr
= inet
->rcv_saddr
;
646 sin
->sin_port
= inet
->sport
;
647 sin
->sin_addr
.s_addr
= addr
;
649 memset(sin
->sin_zero
, 0, sizeof(sin
->sin_zero
));
650 *uaddr_len
= sizeof(*sin
);
654 int inet_sendmsg(struct kiocb
*iocb
, struct socket
*sock
, struct msghdr
*msg
,
657 struct sock
*sk
= sock
->sk
;
659 /* We may need to bind the socket. */
660 if (!inet_sk(sk
)->num
&& inet_autobind(sk
))
663 return sk
->sk_prot
->sendmsg(iocb
, sk
, msg
, size
);
667 static ssize_t
inet_sendpage(struct socket
*sock
, struct page
*page
, int offset
, size_t size
, int flags
)
669 struct sock
*sk
= sock
->sk
;
671 /* We may need to bind the socket. */
672 if (!inet_sk(sk
)->num
&& inet_autobind(sk
))
675 if (sk
->sk_prot
->sendpage
)
676 return sk
->sk_prot
->sendpage(sk
, page
, offset
, size
, flags
);
677 return sock_no_sendpage(sock
, page
, offset
, size
, flags
);
681 int inet_shutdown(struct socket
*sock
, int how
)
683 struct sock
*sk
= sock
->sk
;
686 /* This should really check to make sure
687 * the socket is a TCP socket. (WHY AC...)
689 how
++; /* maps 0->1 has the advantage of making bit 1 rcvs and
692 if ((how
& ~SHUTDOWN_MASK
) || !how
) /* MAXINT->0 */
696 if (sock
->state
== SS_CONNECTING
) {
697 if ((1 << sk
->sk_state
) &
698 (TCPF_SYN_SENT
| TCPF_SYN_RECV
| TCPF_CLOSE
))
699 sock
->state
= SS_DISCONNECTING
;
701 sock
->state
= SS_CONNECTED
;
704 switch (sk
->sk_state
) {
707 /* Hack to wake up other listeners, who can poll for
708 POLLHUP, even on eg. unconnected UDP sockets -- RR */
710 sk
->sk_shutdown
|= how
;
711 if (sk
->sk_prot
->shutdown
)
712 sk
->sk_prot
->shutdown(sk
, how
);
715 /* Remaining two branches are temporary solution for missing
716 * close() in multithreaded environment. It is _not_ a good idea,
717 * but we have no choice until close() is repaired at VFS level.
720 if (!(how
& RCV_SHUTDOWN
))
724 err
= sk
->sk_prot
->disconnect(sk
, O_NONBLOCK
);
725 sock
->state
= err
? SS_DISCONNECTING
: SS_UNCONNECTED
;
729 /* Wake up anyone sleeping in poll. */
730 sk
->sk_state_change(sk
);
736 * ioctl() calls you can issue on an INET socket. Most of these are
737 * device configuration and stuff and very rarely used. Some ioctls
738 * pass on to the socket itself.
740 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
741 * loads the devconfigure module does its configuring and unloads it.
742 * There's a good 20K of config code hanging around the kernel.
745 int inet_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
747 struct sock
*sk
= sock
->sk
;
752 err
= sock_get_timestamp(sk
, (struct timeval __user
*)arg
);
757 err
= ip_rt_ioctl(cmd
, (void __user
*)arg
);
762 err
= arp_ioctl(cmd
, (void __user
*)arg
);
775 err
= devinet_ioctl(cmd
, (void __user
*)arg
);
778 if (!sk
->sk_prot
->ioctl
||
779 (err
= sk
->sk_prot
->ioctl(sk
, cmd
, arg
)) ==
781 err
= dev_ioctl(cmd
, (void __user
*)arg
);
787 struct proto_ops inet_stream_ops
= {
789 .owner
= THIS_MODULE
,
790 .release
= inet_release
,
792 .connect
= inet_stream_connect
,
793 .socketpair
= sock_no_socketpair
,
794 .accept
= inet_accept
,
795 .getname
= inet_getname
,
798 .listen
= inet_listen
,
799 .shutdown
= inet_shutdown
,
800 .setsockopt
= sock_common_setsockopt
,
801 .getsockopt
= sock_common_getsockopt
,
802 .sendmsg
= inet_sendmsg
,
803 .recvmsg
= sock_common_recvmsg
,
804 .mmap
= sock_no_mmap
,
805 .sendpage
= tcp_sendpage
808 struct proto_ops inet_dgram_ops
= {
810 .owner
= THIS_MODULE
,
811 .release
= inet_release
,
813 .connect
= inet_dgram_connect
,
814 .socketpair
= sock_no_socketpair
,
815 .accept
= sock_no_accept
,
816 .getname
= inet_getname
,
819 .listen
= sock_no_listen
,
820 .shutdown
= inet_shutdown
,
821 .setsockopt
= sock_common_setsockopt
,
822 .getsockopt
= sock_common_getsockopt
,
823 .sendmsg
= inet_sendmsg
,
824 .recvmsg
= sock_common_recvmsg
,
825 .mmap
= sock_no_mmap
,
826 .sendpage
= inet_sendpage
,
830 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
833 static struct proto_ops inet_sockraw_ops
= {
835 .owner
= THIS_MODULE
,
836 .release
= inet_release
,
838 .connect
= inet_dgram_connect
,
839 .socketpair
= sock_no_socketpair
,
840 .accept
= sock_no_accept
,
841 .getname
= inet_getname
,
842 .poll
= datagram_poll
,
844 .listen
= sock_no_listen
,
845 .shutdown
= inet_shutdown
,
846 .setsockopt
= sock_common_setsockopt
,
847 .getsockopt
= sock_common_getsockopt
,
848 .sendmsg
= inet_sendmsg
,
849 .recvmsg
= sock_common_recvmsg
,
850 .mmap
= sock_no_mmap
,
851 .sendpage
= inet_sendpage
,
854 static struct net_proto_family inet_family_ops
= {
856 .create
= inet_create
,
857 .owner
= THIS_MODULE
,
860 /* Upon startup we insert all the elements in inetsw_array[] into
861 * the linked list inetsw.
863 static struct inet_protosw inetsw_array
[] =
867 .protocol
= IPPROTO_TCP
,
869 .ops
= &inet_stream_ops
,
872 .flags
= INET_PROTOSW_PERMANENT
,
877 .protocol
= IPPROTO_UDP
,
879 .ops
= &inet_dgram_ops
,
881 .no_check
= UDP_CSUM_DEFAULT
,
882 .flags
= INET_PROTOSW_PERMANENT
,
888 .protocol
= IPPROTO_IP
, /* wild card */
890 .ops
= &inet_sockraw_ops
,
891 .capability
= CAP_NET_RAW
,
892 .no_check
= UDP_CSUM_DEFAULT
,
893 .flags
= INET_PROTOSW_REUSE
,
897 #define INETSW_ARRAY_LEN (sizeof(inetsw_array) / sizeof(struct inet_protosw))
899 void inet_register_protosw(struct inet_protosw
*p
)
901 struct list_head
*lh
;
902 struct inet_protosw
*answer
;
903 int protocol
= p
->protocol
;
904 struct list_head
*last_perm
;
906 spin_lock_bh(&inetsw_lock
);
908 if (p
->type
>= SOCK_MAX
)
911 /* If we are trying to override a permanent protocol, bail. */
913 last_perm
= &inetsw
[p
->type
];
914 list_for_each(lh
, &inetsw
[p
->type
]) {
915 answer
= list_entry(lh
, struct inet_protosw
, list
);
917 /* Check only the non-wild match. */
918 if (INET_PROTOSW_PERMANENT
& answer
->flags
) {
919 if (protocol
== answer
->protocol
)
929 /* Add the new entry after the last permanent entry if any, so that
930 * the new entry does not override a permanent entry when matched with
931 * a wild-card protocol. But it is allowed to override any existing
932 * non-permanent entry. This means that when we remove this entry, the
933 * system automatically returns to the old behavior.
935 list_add_rcu(&p
->list
, last_perm
);
937 spin_unlock_bh(&inetsw_lock
);
944 printk(KERN_ERR
"Attempt to override permanent protocol %d.\n",
950 "Ignoring attempt to register invalid socket type %d.\n",
955 void inet_unregister_protosw(struct inet_protosw
*p
)
957 if (INET_PROTOSW_PERMANENT
& p
->flags
) {
959 "Attempt to unregister permanent protocol %d.\n",
962 spin_lock_bh(&inetsw_lock
);
963 list_del_rcu(&p
->list
);
964 spin_unlock_bh(&inetsw_lock
);
971 * Shall we try to damage output packets if routing dev changes?
974 int sysctl_ip_dynaddr
;
976 static int inet_sk_reselect_saddr(struct sock
*sk
)
978 struct inet_sock
*inet
= inet_sk(sk
);
981 __u32 old_saddr
= inet
->saddr
;
983 __u32 daddr
= inet
->daddr
;
985 if (inet
->opt
&& inet
->opt
->srr
)
986 daddr
= inet
->opt
->faddr
;
988 /* Query new route. */
989 err
= ip_route_connect(&rt
, daddr
, 0,
993 inet
->sport
, inet
->dport
, sk
);
997 sk_setup_caps(sk
, &rt
->u
.dst
);
999 new_saddr
= rt
->rt_src
;
1001 if (new_saddr
== old_saddr
)
1004 if (sysctl_ip_dynaddr
> 1) {
1005 printk(KERN_INFO
"%s(): shifting inet->"
1006 "saddr from %d.%d.%d.%d to %d.%d.%d.%d\n",
1009 NIPQUAD(new_saddr
));
1012 inet
->saddr
= inet
->rcv_saddr
= new_saddr
;
1015 * XXX The only one ugly spot where we need to
1016 * XXX really change the sockets identity after
1017 * XXX it has entered the hashes. -DaveM
1019 * Besides that, it does not check for connection
1020 * uniqueness. Wait for troubles.
1022 __sk_prot_rehash(sk
);
1026 int inet_sk_rebuild_header(struct sock
*sk
)
1028 struct inet_sock
*inet
= inet_sk(sk
);
1029 struct rtable
*rt
= (struct rtable
*)__sk_dst_check(sk
, 0);
1033 /* Route is OK, nothing to do. */
1038 daddr
= inet
->daddr
;
1039 if (inet
->opt
&& inet
->opt
->srr
)
1040 daddr
= inet
->opt
->faddr
;
1043 .oif
= sk
->sk_bound_dev_if
,
1047 .saddr
= inet
->saddr
,
1048 .tos
= RT_CONN_FLAGS(sk
),
1051 .proto
= sk
->sk_protocol
,
1054 .sport
= inet
->sport
,
1055 .dport
= inet
->dport
,
1060 err
= ip_route_output_flow(&rt
, &fl
, sk
, 0);
1063 sk_setup_caps(sk
, &rt
->u
.dst
);
1065 /* Routing failed... */
1066 sk
->sk_route_caps
= 0;
1068 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1069 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1071 if (!sysctl_ip_dynaddr
||
1072 sk
->sk_state
!= TCP_SYN_SENT
||
1073 (sk
->sk_userlocks
& SOCK_BINDADDR_LOCK
) ||
1074 (err
= inet_sk_reselect_saddr(sk
)) != 0)
1075 sk
->sk_err_soft
= -err
;
1081 EXPORT_SYMBOL(inet_sk_rebuild_header
);
1083 #ifdef CONFIG_IP_MULTICAST
1084 static struct net_protocol igmp_protocol
= {
1085 .handler
= igmp_rcv
,
1089 static struct net_protocol tcp_protocol
= {
1090 .handler
= tcp_v4_rcv
,
1091 .err_handler
= tcp_v4_err
,
1095 static struct net_protocol udp_protocol
= {
1097 .err_handler
= udp_err
,
1101 static struct net_protocol icmp_protocol
= {
1102 .handler
= icmp_rcv
,
1105 static int __init
init_ipv4_mibs(void)
1107 net_statistics
[0] = alloc_percpu(struct linux_mib
);
1108 net_statistics
[1] = alloc_percpu(struct linux_mib
);
1109 ip_statistics
[0] = alloc_percpu(struct ipstats_mib
);
1110 ip_statistics
[1] = alloc_percpu(struct ipstats_mib
);
1111 icmp_statistics
[0] = alloc_percpu(struct icmp_mib
);
1112 icmp_statistics
[1] = alloc_percpu(struct icmp_mib
);
1113 tcp_statistics
[0] = alloc_percpu(struct tcp_mib
);
1114 tcp_statistics
[1] = alloc_percpu(struct tcp_mib
);
1115 udp_statistics
[0] = alloc_percpu(struct udp_mib
);
1116 udp_statistics
[1] = alloc_percpu(struct udp_mib
);
1118 (net_statistics
[0] && net_statistics
[1] && ip_statistics
[0]
1119 && ip_statistics
[1] && tcp_statistics
[0] && tcp_statistics
[1]
1120 && udp_statistics
[0] && udp_statistics
[1]))
1123 (void) tcp_mib_init();
1128 static int ipv4_proc_init(void);
1131 * IP protocol layer initialiser
1134 static struct packet_type ip_packet_type
= {
1135 .type
= __constant_htons(ETH_P_IP
),
1139 static int __init
inet_init(void)
1141 struct sk_buff
*dummy_skb
;
1142 struct inet_protosw
*q
;
1143 struct list_head
*r
;
1146 if (sizeof(struct inet_skb_parm
) > sizeof(dummy_skb
->cb
)) {
1147 printk(KERN_CRIT
"%s: panic\n", __FUNCTION__
);
1151 rc
= proto_register(&tcp_prot
, 1);
1155 rc
= proto_register(&udp_prot
, 1);
1157 goto out_unregister_tcp_proto
;
1159 rc
= proto_register(&raw_prot
, 1);
1161 goto out_unregister_udp_proto
;
1164 * Tell SOCKET that we are alive...
1167 (void)sock_register(&inet_family_ops
);
1170 * Add all the base protocols.
1173 if (inet_add_protocol(&icmp_protocol
, IPPROTO_ICMP
) < 0)
1174 printk(KERN_CRIT
"inet_init: Cannot add ICMP protocol\n");
1175 if (inet_add_protocol(&udp_protocol
, IPPROTO_UDP
) < 0)
1176 printk(KERN_CRIT
"inet_init: Cannot add UDP protocol\n");
1177 if (inet_add_protocol(&tcp_protocol
, IPPROTO_TCP
) < 0)
1178 printk(KERN_CRIT
"inet_init: Cannot add TCP protocol\n");
1179 #ifdef CONFIG_IP_MULTICAST
1180 if (inet_add_protocol(&igmp_protocol
, IPPROTO_IGMP
) < 0)
1181 printk(KERN_CRIT
"inet_init: Cannot add IGMP protocol\n");
1184 /* Register the socket-side information for inet_create. */
1185 for (r
= &inetsw
[0]; r
< &inetsw
[SOCK_MAX
]; ++r
)
1188 for (q
= inetsw_array
; q
< &inetsw_array
[INETSW_ARRAY_LEN
]; ++q
)
1189 inet_register_protosw(q
);
1192 * Set the ARP module up
1198 * Set the IP module up
1203 tcp_v4_init(&inet_family_ops
);
1205 /* Setup TCP slab cache for open requests. */
1210 * Set the ICMP layer up
1213 icmp_init(&inet_family_ops
);
1216 * Initialise the multicast router
1218 #if defined(CONFIG_IP_MROUTE)
1222 * Initialise per-cpu ipv4 mibs
1225 if(init_ipv4_mibs())
1226 printk(KERN_CRIT
"inet_init: Cannot init ipv4 mibs\n"); ;
1232 dev_add_pack(&ip_packet_type
);
1237 out_unregister_tcp_proto
:
1238 proto_unregister(&tcp_prot
);
1239 out_unregister_udp_proto
:
1240 proto_unregister(&udp_prot
);
1244 module_init(inet_init
);
1246 /* ------------------------------------------------------------------------ */
1248 #ifdef CONFIG_PROC_FS
1249 static int __init
ipv4_proc_init(void)
1253 if (raw_proc_init())
1255 if (tcp4_proc_init())
1257 if (udp4_proc_init())
1259 if (fib_proc_init())
1261 if (ip_misc_proc_init())
1278 #else /* CONFIG_PROC_FS */
1279 static int __init
ipv4_proc_init(void)
1283 #endif /* CONFIG_PROC_FS */
1285 MODULE_ALIAS_NETPROTO(PF_INET
);
1287 EXPORT_SYMBOL(inet_accept
);
1288 EXPORT_SYMBOL(inet_bind
);
1289 EXPORT_SYMBOL(inet_dgram_connect
);
1290 EXPORT_SYMBOL(inet_dgram_ops
);
1291 EXPORT_SYMBOL(inet_getname
);
1292 EXPORT_SYMBOL(inet_ioctl
);
1293 EXPORT_SYMBOL(inet_listen
);
1294 EXPORT_SYMBOL(inet_register_protosw
);
1295 EXPORT_SYMBOL(inet_release
);
1296 EXPORT_SYMBOL(inet_sendmsg
);
1297 EXPORT_SYMBOL(inet_shutdown
);
1298 EXPORT_SYMBOL(inet_sock_destruct
);
1299 EXPORT_SYMBOL(inet_stream_connect
);
1300 EXPORT_SYMBOL(inet_stream_ops
);
1301 EXPORT_SYMBOL(inet_unregister_protosw
);
1302 EXPORT_SYMBOL(net_statistics
);
1303 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind
);